Advances in Heat Transfer - Advances in Heat Transfer Volume 34 2000 PDF

ADVANCES IN
HEAT TRANSFER
Volume 34
a
This Page Intentionally Left Blank
Advances in
HEAT
TRANSFER
Serial Editors
James P. Hartnett Thomas F. Irvine, Jr.
Energy Resources Center Department of Mechanical Engineering
University of Illinois at Chicago State University of New York at Stony Brook
Chicago, Illinois Stony Brook, New York
Serial Associate Editors
Young I. Cho George A. Greene
Department of Mechanical Engineering Department of Advanced Technology
Drexel University Brookhaven National Laboratory
Philadelphia, Pennsylvania Upton, New York
Volume 34
San Diego San Francisco New York Boston London Sydney Tokyo
This book is printed on acid-free paper.
Copyright 2001 by Academic Press
All rights reserved.
No part of this publication may be reproduced or transmitted in any form or by any means,
electronic or mechanical, including photocopy, recording, or any information storage and
retrieval system, without permission in writing from the publisher.
The appearance of code at the bottom of the rst page of a chapter in this book indicates the
Publishers consent that copies of the chapter may be made for personal or internal use of
specic clients. This consent is given on the condition, however, that the copier pay the stated
per-copy fee through the Copyright Clearance Center, Inc. (222 Rosewood Drive, Danvers,
Massachusetts 01923), for copying beyond that permitted by Sections 107 or 108 of the U. S.
Copyright Law. This consent does not extend to other kinds of copying, such as copying
for general distribution, for advertising or promotional purposes, for creating new collective
works, or for resale. Copy fees for chapters are as shown on the title pages; if no fee code
appears on the chapter title page, the copy fee is the same for current chapters.
0065-2717/01 $35.00
ACADEMIC PRESS
525 B Street, Suite 1900, San Diego, CA 92101-4495, USA
http://www.academicpress.com
Academic Press
Harcourt Place, 32 Jamestown Road, London NW1 7BY, UK
International Standard Book Number: 0-12-020034-1
International Standard Serial Number: 0065-2717
Printed in the United States of America
00 01 02 03 QW 9 8 7 6 5 4 3 2 1
CONTENTS
Contributors . . . . . . . . . . . . . . . . . . . . . . . . . . . ix
Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi
Transport Phenomena in Heterogeneous Media Based
on Volume Averaging Theory
V. S. T:vxiN :Nn I. C:11oN
I. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 1
II. Fundamentals of Hierarchical Volume Averaging
Techniques . . . . . . . . . . . . . . . . . . . . . . . . 4
A. Theoretical Verication of Central VAT Theorem and Its Consequences . 10
III. Nonlinear and Turbulent Transport in Porous Media . . . . 14
A. Laminar Flow with Constant Coefcients . . . . . . . . . . . . . 17
B. Nonlinear Fluid Medium Equations in Laminar Flow . . . . . . . . 19
C. Porous Medium Turbulent VAT Equations . . . . . . . . . . . . 21
D. Development of Turbulent Transport Models in Highly Porous Media . 26
E. Closure Theories and Approaches for Transport in Porous Media . . . 32
IV. Microscale Heat Transport Description Problems and
VAT Approach . . . . . . . . . . . . . . . . . . . . . . . 37
A. Traditional Descriptions of Microscale Heat Transport . . . . . . . . 38
B. VAT-Based Two-Temperature Conservation Equations . . . . . . . . 43
C. Subcrystalline Single Crystal Domain Wave Heat Transport Equations . 45
D. Nonlocal Electrodynamics and Heat Transport in Superstructures . . . 46
E. Photonic Crystals Band-Gap Problem: Conventional DMM-DNM and
VAT Treatment . . . . . . . . . . . . . . . . . . . . . . . 52
V. Radiative Heat Transport in Porous and Heterogeneous
Media . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
VI. Flow Resistance Experiments and VAT-Based Data
Reduction in Porous Media . . . . . . . . . . . . . . . . . 66
VII. Experimental Measurements and Analysis of Internal Heat
Transfer Coefcients in Porous Media . . . . . . . . . . . . 85
VIII. Thermal Conductivity Measurement in a Two-Phase
Medium . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
IX. VAT-Based Compact Heat Exchanger Design and
Optimization . . . . . . . . . . . . . . . . . . . . . . . . 111
A. A Short Review of Current Practice in Heat Exchanger Modeling . . . 112
v
B. New Kinds of Heat Exchanger Mathematical Models . . . . . . . . 116
C. VAT-Based Compact Heat Exchanger Modeling . . . . . . . . . . 117
D. Optimal Control Problems in Heat Exchanger Design . . . . . . . . 123
E. A VAT-Based Optimization Technique for Heat Exchangers . . . . . . 124
X. New Optimization Technique for Material Design Based
on VAT . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
XI. Concluding Remarks . . . . . . . . . . . . . . . . . . . . 129
Nomenclature . . . . . . . . . . . . . . . . . . . . . . . . 131
References . . . . . . . . . . . . . . . . . . . . . . . . . . 133
Two-Phase Flow in Microchannels
S. M. Gni::si::N :Nn S. I. Annri-Kn:iix
I. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 145
II. Characteristics of Microchannel Flow . . . . . . . . . . . . 146
III. Two-Phase Flow Regimes and Void Fraction in
Microchannels. . . . . . . . . . . . . . . . . . . . . . . . 147
A. Denition of Major Two-Phase Flow Regimes . . . . . . . . . . . 148
B. Two-Phase Flow Regimes in Microchannels . . . . . . . . . . . . 150
C. Review of Previous Experimental Studies and Their Trends . . . . . . 153
D. Flow Regime Transition Models and Correlations . . . . . . . . . 161
E. Flow Patterns in a Micro-Rod Bundle . . . . . . . . . . . . . . 166
F. Void Fraction . . . . . . . . . . . . . . . . . . . . . . . . 169
G. Two-Phase Flow in Narrow Rectangular and Annular Channels . . . . 170
H. Two-Phase Flow Caused by the Release of Dissolved Noncondensables . 178
IV. Pressure Drop . . . . . . . . . . . . . . . . . . . . . . . . 180
A. General Remarks . . . . . . . . . . . . . . . . . . . . . . 180
B. Frictional Pressure Drop in Two-Phase Flow . . . . . . . . . . . 180
C. Review of Previous Experimental Studies . . . . . . . . . . . . . 184
D. Frictional Pressure Drop in Narrow Rectangular and Annular Channels . 189
V. Forced Flow Subcooled Boiling . . . . . . . . . . . . . . . 191
A. General Remarks . . . . . . . . . . . . . . . . . . . . . . 191
B. Void Fraction Regimes in Heated Channels . . . . . . . . . . . . 192
C. Onset of Nucleate Boiling . . . . . . . . . . . . . . . . . . . 195
D. Onset of Signicant Void and Onset of Flow Instability . . . . . . . 198
E. Observations on Bubble Nucleation and Boiling . . . . . . . . . . 205
VI. Critical Heat Flux in Microchannels . . . . . . . . . . . . . . . . . 209
A. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 209
B. Experimental Data and Their Trends . . . . . . . . . . . . . . . 210
C. Effects of Pressure, Mass Flux, and Noncondensables . . . . . . . . 215
D. Empirical Correlations . . . . . . . . . . . . . . . . . . . . 216
E. Theoretical Models . . . . . . . . . . . . . . . . . . . . . . 221
VII. Critical Flow in Cracks and Slits . . . . . . . . . . . . . . . 224
A. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 224
B. Experimental Critical Flow Data . . . . . . . . . . . . . . . . 225
vi coN1rN1s
C. General Remarks on Models for Two-Phase Critical Flow in
Microchannels . . . . . . . . . . . . . . . . . . . . . . . 230
D. Integral Models . . . . . . . . . . . . . . . . . . . . . . . 232
E. Models Based on Numerical Solution of Differential Conservation
Equations . . . . . . . . . . . . . . . . . . . . . . . . . 236
VIII. Concluding Remarks . . . . . . . . . . . . . . . . . . . . 240
Nomenclature . . . . . . . . . . . . . . . . . . . . . . . . 242
References . . . . . . . . . . . . . . . . . . . . . . . . . . 244
Turbulent Flow and Convection: The Prediction of Turbulent Flow
and Convection in a Round Tube
S1i:1 W. Cnicniii
I. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 256
A. Turbulent Flow . . . . . . . . . . . . . . . . . . . . . . . 257
B. Turbulent Convection . . . . . . . . . . . . . . . . . . . . 259
II. The Quantitative Representation of Turbulent Flow . . . . . 260
A. Historical Highlights . . . . . . . . . . . . . . . . . . . . . 260
B. New Improved Formulations and Correlating Equations . . . . . . . . . 294
III. The Quantitative Representation of Fully Developed
Turbulent Convection . . . . . . . . . . . . . . . . . . . . 304
A. Essentially Exact Formulations . . . . . . . . . . . . . . . . . 305
B. Essentially Exact Numerical Solutions . . . . . . . . . . . . . . 323
C. Correlation for Nu . . . . . . . . . . . . . . . . . . . . . . 335
IV. Summary and Conclusions . . . . . . . . . . . . . . . . . . 348
A. Turbulent Flow . . . . . . . . . . . . . . . . . . . . . . . 348
B. Turbulent Convection . . . . . . . . . . . . . . . . . . . . 353
References . . . . . . . . . . . . . . . . . . . . . . . . . . 356
Progress in the Numerical Analysis of Compact Heat
Exchanger Surfaces
R. K. Sn:n, M. R. Hrix:i, B. TnoNoN, :Nn P. TocnoN
I. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 363
II. Physics of Flow and Heat Transfer of CHE Surfaces . . . . . 366
A. Interrupted Flow Passages . . . . . . . . . . . . . . . . . . . 366
B. Uninterrupted Complex Flow Passages . . . . . . . . . . . . . . 371
C. Unsteady Laminar versus Low Reynolds Number
Turbulent Flow . . . . . . . . . . . . . . . . . . . . . . . 374
III. Numerical Analysis . . . . . . . . . . . . . . . . . . . . . 375
A. Mesh Generation . . . . . . . . . . . . . . . . . . . . . . 376
B. Boundary Conditions . . . . . . . . . . . . . . . . . . . . . 376
C. Solution Algorithm and Numerical Scheme . . . . . . . . . . . . 378
vii coN1rN1s
IV. Turbulence Models . . . . . . . . . . . . . . . . . . . . . 380
A. Reynolds Averaged NavierStokes (RANS) Equations . . . . . . . . 381
B. Large Eddy Simulation (LES) . . . . . . . . . . . . . . . . . 392
C. Direct Numerical Simulation . . . . . . . . . . . . . . . . . . 395
D. Concluding Remarks on Turbulence Modeling . . . . . . . . . . . 397
V. Numerical Results of the CHE Surfaces . . . . . . . . . . . 397
A. Offset Strip Fins . . . . . . . . . . . . . . . . . . . . . . . 398
B. Louver Fins . . . . . . . . . . . . . . . . . . . . . . . . 406
C. Wavy Channels . . . . . . . . . . . . . . . . . . . . . . . 416
D. Chevron Trough Plates . . . . . . . . . . . . . . . . . . . . 425
VI. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . 432
Nomenclature . . . . . . . . . . . . . . . . . . . . . . . . 434
References . . . . . . . . . . . . . . . . . . . . . . . . . . 435
Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . 445
Subject Index . . . . . . . . . . . . . . . . . . . . . . . . . . . 461
viii coN1rN1s
CONTRIBUTORS
Numbers in parentheses indicate the pages on which the authors contributions begin
S. I. Annri-Kn:iix (145), G. W. Woodruff School of Mechanical Engineer-
ing, Georgia Institute of Technology, Atlanta, Georgia 30332.
I. C:11oN (1), Department of Mechanical and Aerospace Engineering,
University of California, Los Angeles, Los Angeles, California 90095.
S1i:1 W. Cnicniii (255), Department of Chemical Engineering, The
University of Pennsylvania, Philadelphia, Pennsylvania 19104.
S. M. Gni::si::N (145), G. W. Woodruff School of Mechanical Engineer-
ing, Georgia Institute of Technology Atlanta, Georgia 30332.
M. R. Hrix:i (363), University of Brighton, Brighton, United Kingdom.
R. K. Sn:n (363), Delphi Harrison Thermal Systems, Lockport, New York
14094.
B. TnoNoN (363), CEA-Grenoble, DTP/GRETh, Grenoble, France.
P. TocnoN (363), CEA-Grenoble, DTP/GRETh, Grenoble, France.
V. S. T:vxiN (1), Department of Mechanical and Aerospace Engineering,
University of California, Los Angeles, Los Angeles, California 90095.
ix
aa
PREFACE
For over a third of a century this serial publication, Advances in Heat
Transfer, has lled the information gap between regularly published journals
and university-level textbooks. The series presents review articles on special
topics of current interest. Each contribution starts from widely understood
principles and brings the reader up to the forefront of the topic being
addressed. The favorable response by the international scientic and engin-
eering community to the thirty-four volumes published to date is an
indication of the success of our authors in fullling this purpose.
In recent years, the editors have undertaken to publish topical volumes
dedicated to specic elds of endeavor. Several examples of such topical
volumes are Volume 22 (Bioengineering Heat Transfer), Volume 28 (Trans-
port Phenomena in Materials Processing), and Volume 29 (Heat Transfer
in Nuclear Reactor Safety). As a result of the enthusiastic response of the
readers, the editors intend to continue the practice of publishing topical
volumes as well as the traditional general volumes.
The editorial board expresses their appreciation to the contributing
authors of this volume who have maintained the high standards associated
with Advances in Heat Transfer. Lastly, the editors acknowledge the efforts
of the professional staff at Academic Press who have been responsible for
the attractive presentation of the published volumes over the years.
xi
aa
ADVANCES IN
HEAT TRANSFER
Volume 34
aa
Transport Phenomena in
Heterogeneous Media Based on
Volume Averaging Theory
V. S. TRAVKIN and I. CATTON
Department of Mechanical and Aerospace Engineering
University of California, Los Angeles
Los Angeles, California 90095
I. Introduction
Determination of ow variables and scalar transport for problems
involving heterogeneous (and porous) media is difcult, even when the
problem is subject to simplications allowing the specication of medium
periodicity or regularity. Linear or linearized models fail to intrinsically
account for transport phenomena, requiring dynamic coefcient models to
correct for shortcomings in the governing models. Allowing inhomogeneities
to adopt random or stochastic character further confounds the already
daunting task of properly identifying pertinent transport mechanisms and
predicting transport phenomena.
This problem is presently treated by procedures that are mostly heuristic
in nature because sufciently detailed descriptions are not included in the
description of the problem and consequently are not available. The ability
to describe the details, and features, of a proposed material with precision
will help reduce the need for a heuristic approach.
Some aspects of the development of the needed theory are now well
understood and have seen substantial progress in the thermal physics and
in uid mechanics sciences, particularly in porous media transport phenom-
ena. The basis for this progress is the so-called volume averaging theory
(VAT), which was rst proposed in the 1960s by Anderson and Jackson [1],
Slattery [2], Marle [3], Whitaker [4], and Zolotarev and Radushkevich [5].
ADVANCES IN HEAT TRANSFER, VOLUME 34
1
ADVANCES IN HEAT TRANSFER, VOL. 34
ISBN: 0-12-020034-1 Copyright 2001 by Academic Press. All rights of reproduction in any form reserved.
0065-2717/01 $35.00
Further advances in the use of VAT are found in the work of Slattery [6],
Kaviany [7], Gray et al. [8], and Whitaker [9, 10]). Many of the important
details and examples of application are found in books by Kheifets and
Neimark [11], Dullien [12], and Adler [13].
Publications on turbulent transport in porous media based on VAT
began to appear in 1986. Primak et al. [14], Shcherban et al. [15], and later
studies by Travkin and Catton [16, 18, 20, 21], etc., Travkin et al. [17, 19,
22], Gratton et al. [26, 27] and Catton and Travkin [28] present a
generalized development of VAT for heterogeneous media applicable to
nonlinear physical phenomena in thermal physics and uid mechanics.
In most physically realistic cases, highly complex integraldifferential
equations result. When additional terms in the two- and three-phase
statements are encountered, the level of difculty in attempting to obtain
closure and, hence, effective coefcients, increases greatly. The largest
challenge is surmounting problems associated with the consistent lack of
understanding of new, advanced equations and insufcient development of
closure theory, especially for integraldifferential equations. The ability to
accurately evaluate various kinds of medium morphology irregularities
results from the modeling methodology once a porous medium morphology
is assigned. Further, when attempting to describe transport processes in a
heterogeneous media, the correct form of the governing equations remains
an area of continuously varying methods among researchers (see some
discussion in Travkin and Catton [16, 21]).
An important feature of VAT is being able to consider specic medium
types and morphologies, lower-scale uctuations of variables, cross-effects of
different variable uctuations, interface variable uctuations effects, etc. It is
not possible to include all of these characteristics in current models using
conventional theoretical approaches. The VAT approach has the following
desirable features:
1. Effects of interfaces and grain boundaries can be included in the
modeling.
2. The effect of morphology of the different phases is incorporated. The
morphology decription is directly incorporated into the eld equa-
tions.
3. Separate and combined elds and their interactions are described
exactly. No assumptions about effective coefcients are required.
4. Effective coefcients correct mathematical descriptionthose the-
ories presently used for that purpose are only approximate descrip-
tion, and often simply wrong.
5. Correct description of experiments in heterogeneous media again, at
present the homogeneous presentation of medium properties is used
2 v. s. 1:vxiN :Nn i. c:11oN
for this purpose, and explanation of experiments is done via bulk
features. Those bulk features describe the eld as by classical homo-
geneous medium differential equations.
6. Deliberate design and optimization of materials using hierarchical
physical descriptions based on the VAT governing equations can be
used to connect properties and morphological characteristics to com-
ponent features. What is usually done is to carry out an experimental
search by adding a third or a fourth component to the piezoelectric
material, for example. This can be done in a more direct, more
observable way, and with a more correct understanding of the effects
of adding additional components and, of course, of the morphology of
the fourth component.
In this work we restrict ourselves to a brief analysis of previous work and
show that the best theoretical tool is the nonlocal description of hierarchical,
multiscaled processes resulting from application of VAT. Application of
VAT to radiative transport in a porous medium is based on our advances
in electrodynamics and microscale energy transport phenomena in two-
phase heterogeneous media. Some other governing conservation equations
for transport in porous media can be found in Travkin and Catton [21] and
the references therein.
One of the aims of this work is to outline the possibilities for a method
for optimizing transport in heterogeneous as well as porous structures that
can be used in different engineering elds. Applications range from heat and
mass exchangers and reactors in mechanical engineering design to environ-
mental engineering usage (Travkin et al. [19]). A recent application is in
urban air pollution, where optimal control of a pollutant level in a
contaminated area is determined, along with the design of an optimal
control point network for the control of constituent dispersion and remedi-
ation actions. Using second-order turbulent models, equation sets were
obtained for turbulent ltration and two-temperature or two-concentration
diffusion in nonisotropic porous media and interphase exchange and micro-
roughness. Previous work has shown that the ow resistance and heat
transfer over highly rough surfaces or in a rough channel or pipe can be
properly predicted using the technique of averaging the transport equations
over the near surface representative elementary volume (REV). Prescribing
the statistical structure of the capillary or globular porous medium mor-
phology gives the basis for transforming the integraldifferential transport
equations into differential equations with probability density functions
governing their coefcients and source terms. Several different closure
models for these terms for some uniform, nonuniform, nonisotropic, and
specically random nonisotropic highly porous layers were developed. Quite
3 voiixr :vr:ciNc 1nrox
different situations arise when describing processes occuring in irregular or
random morphology. The latest results, obtained with the help of exact
closure modeling for canonical morphologies, open a new eld of possibili-
ties for a purposeful search for optimal design of spacial heterogeneous
transport structures. A way to nd and govern momentum transport
through a capillary nonintersecting medium by altering its morphometrical
characteristics is given as validation of the process.
II. Fundamentals of Hierarchical Volume Averaging Techniques
Since the porosity in a porous medium is often anisotropic and randomly
inhomogeneous, the random porosity function can be decomposed into
additive components: the average value of m(x), in the REV and its
uctuations in various directions,
m
(x) m
(x), m`
(x), m
,
AD
AD
.
The averaged equations of turbulent ltration for a highly porous
medium are similar to those in an anisotropic porous medium. Five types
of averaging over an REV function f are dened by the following averaging
operators arranged in their order of seniority (Primak et al. [14]): average
of f over the whole REV,
f , f ,
f ,
`
m
, f
(1 m
,) f
`
, (1)
phase averages of f in each component of the medium,
f ,
,
1
AD
,
`'
`
f (t, x) dcm
, f
(2)
f ,
`
m
`
,
1
AD
`
,
`'
f (t, x) dcm
`
, f
`
(3)
and intraphase averages,
f }
1
AD
,
`'
`
f (t, x) dc (4)
f }
`
f
`
1
AD
`
,
`'
f (t, x) dc. (5)

When the interface is xed in space, the averaged functions for the rst
and second phase (as liquid and solid) within the REV and over the entire
REV fulll the conditions
f g}
f }
g}
and a
} a const (6)
for steady-state phases and
c f
ct
c f }
ct
, fg}
fg` (7)
except for the differentiation condition,
V f }
V f
1
AD
,
c
`
U
fd
f f f, f AD
, (8)
where cS
"
is the inner surface in the REV, and d s is the solid-phase,
inward-directed differential area in the REV (d s n dS). The fourth condi-
tion implies an unchanging porous medium morphology.
The three types of averaging fulll all four of the preceding conditions as
well as the following four consequences:
f}
f, f}
f f}
0 (9)
fg` }
fg` , fg` }
f g` 0 (10)
Meanwhile, f ,
and f ,
`
fulll neither the third of the conditions,
a,
a, a,
,a, (11)
nor all the consequences of the other averaging conditions. Futher, the
differential condition becomes
Vf ,
V f ,
1
AD ,
c
`
`
f ds
, (12)
in accordance with one of the major averaging theorems the theorem of
averaging the V operator (Slattery [6]; Gray et al. [8]; Whitaker [10]).
If the statistical characteristics of the REV morphology and the averaging
conditions with their consequences lead to the following special ergodic
hypothesis: the spacial averages, ( f ,
, f, and f }
`
), then this theorem
converges with increases in the averaging volume to the appropriate
probability (statistical) average of the function f of a random value with
probability density distribution p. This hypothesis is stated mathematically
as follows:
f'(x)
,
`
`
f (x, :)p
:, x} d:
lim
`'`
f f'. (13)
Quintard and Whitaker [29] expressed some concern about the connec-
tion between different scale volume averaged variables, for example,
T
1
AD ,
`'
D
T
dc (14)
T
1
AD
,
`'
D
T
dc. (15)
In a truly periodic system it is known that the steady temperature in phase
f can be written as
T
(r
) h r
(r
) T
"
, (16)
where h and T
"
are constants and T
(r
) is a periodic function of zero mean

over the f-phase. Applying the phase averaging operator ,
to this
function, one nds
T
(x),
(x),
(r
),
m,T
"
,
while Quintard and Whitaker [29] obtain (their Eq. (13))
T
(r
),
(r
),
h r
m,T
"
, (17)
meaning that
T
(r
),
0. (18)
The parameter T
(r
),
cannot always be equal to zero, because it

depends on the peculiarities of the chosen REV. In some instances, when the
REV is not the volume that contains the known number of exact function
periods, the averaged function T
(r
),
value should not be zero. If it is

assumed, however, that the REV volume AD contains the exact number of
spacial periods, then
T
(r
),
0.
Averaging the uid temperature, T
, over AD
yields the intrinsic average

T
(x)}
h r
T
"
h xh y
T
"
, (19)
because the averaging of r
(see Fig. 1) results in

r
x y
, (20)
while Quintard and Whitaker [29] obtain (their Eq. (15))
T
(r
)}
h r
T
"
(21)
They note (see p. 375), now represent r
in terms of the position vector,

x, that locates the centroid of the averaging volume, and the relative
Fic. 1. Representative elementary averaging volume with the virtual points of representa-
tion inside of the REV (Carbonell and Whitaker [31]; Quintard and Whitaker [29]).
position vector y
as indicated in Fig. 3 (see Fig. 1), or

r
x y
=r
(x, y
)}
x y
(r
, x)}
, (22)
so that Eq. (21) can be written with dependence on both x and r
,
T
(x, r
)}
(x, y
)}
h x h y
(r
, x)}
T
"
, (23)
meaning that after averaging, T
(r
) continues to be dependent on the

position of the virtual point r
, which may have changed location within

the AD
.
To do this, they introduce a so-called virtual REV allowing the
averaged value inside of the REV to be variable (see the remark on p. 354
of Quintard and Whitaker [30]: In all our previous studies of multiphase
transport phenomena, we have always assumed that averaged quantities
could be treated as constants within the averaging volume and that the
average of the spatial deviations was zero. We now wish to avoid these
assumptions. . . .), and the result is a virtual averaged variable that is not
constant within the xed volume of the REV. When Quintard and Whit-
taker derive the gradient of the average of the function (23), they use its
dependence on x for the two right-hand-side terms in (23) to obtain
VT
(x, r
)}
h h V
x
y
(r
, x)}
. (24)
Several comments about the Quintard and Whittaker treatment that need
to be considered are the following:
1. How the communication of the variables from different spaces r
at the
lower scale space and x at the upper scale space is established is not
meaningful. Their connection must be determined at the beginning of the
averaging process and their communication is very limited.
2. One should only connect a value at a point at the higher level to the
lower level REV, not only to a point within the lower level REV. When one
considers an averaged variable at any point other than the representative
point x for a particular REV, then
VT
(x)}
V(h xT
(x)}
T
"
),
and for the upper scale, the exact result is
VT
(x)}
V(h xT
"
) h. (25)
3. If a function and its gradient are periodic, then the averaged function
should be periodic. The VAT-based answer should be seen by determination
of the averaged values, which are not averaged, only the REV being used at
the lower scale.
The work by Quintard and Whittaker and the improving of understand-
ing of some basic principles of averaging has led us to state the following
lemma and then point out differences from the work of Whittaker and his
colleagues.
Lemma. If a function , representing any continuous physical eld, is
averaged over the subdomain AD
, which is the subregion occupied by phase

f ( uid phase) of the REV AD
in the heterogeneous two-phase medium (Fig.

1), and the averaged function (x)}
is assumed to have different values at

different locations x
'
within the AD
, then the averaged function (x)}
can
have discontinuities of the rst kind at the boundary cAD
of the REV AD
.
Proof. Consider the situation where the point y
`
(Fig. 1; see also Fig. 3,
p. 375, in the paper by Quintard and Whitaker [29]) is located an innitely
distance from the boundary of the REV AD
within cAD
. It represents the
intrinsic phase averaged value
`
}
of variable averaged in the REV

AD
. According to Carbonell and Whitaker [31] and Quintard and

Whitaker [29], its value can be different from
or
`
}
.
Fic. 2. Representative elementary averaging volumes with the xed points of representation.
Next, consider a point y
`
located an innitely small distance outside of
the initial REV AD
within a boundary cAD
. The point y
`
represents the
averaged value
`
}
which belongs to some REV AD

`
, as shown in Fig.
2, with its center at the point x
`
(y
`
c) (R
/2), with c being an

innitely small constant.
Following arguments of Carbonell and Whitaker [31], this point y
`
is
allowed to be in at least one more REV, AD
`
, which has its center x
`
just
shifted from the point x
an innitely small distance, as does y

`
from the
boundary cAD
.
Further, suppose that one is approaching the boundary cAD
from both
sides by points y
`
and y
`
. According to Carbonell and Whitaker [31] and
Quintard and Whitaker [29], the values
`
}
and
`
}
can be different
when cAD
is reached, which means that the averaged value }
experien-
ces (can have) a discontinuity at each and every point of the boundary
cAD
.
As long as the boundary cAD
of the REV can be arbitrarily moved,

changed or assigned, then the consequence of this change is that
can
have discontinuities at each point of a REV.
The relationships between different scale variables and their points of
representation can be found by noting the following points:
1. There is a xed relationship between the location of the point x
of the
upper scale eld and averaging within the REV AD
. In other words, for

each determined AD
there is only one x
that represents the value
(x
)}
on the upper eld level (macroscale eld) if both are mapped on

the same region (excluding close to boundary regions).
2. If there is the value
(x
`
)}
, (x
`
x
) c, then there is another

AD
`
AD
, and in it
(x
`
)}
1
AD
`
,
`'
D
(r
) dc
1
AD
,
`'
D`
(r
) dc, (26)
where
(r
) const.
A. Tnror1ic:i Vriric:1ioN or CrN1:i VAT Tnrorx
:Nn I1s CoNsrqirNcrs
When the coefcient of thermal conductivity k
is a constant value, the

uid stedy-state conduction regime is described by
k
V`(m,T
) V
_
k
AD ,
c
`
U
T
ds
AD ,
c
`
U
VT
ds 0. (27)
The full 1D Cartesian coordinates version of this equation, without any
source, for a xed solid matrix in is
c
cx _
m,
cT
cx
c
cx _
1
AD ,
c
`
U
T
ds
1
AD ,
c
`
U
cT
cx
'
ds 0, (28)
c
cx _
m,
cT
cx
MD
`
MD
`
0, (29)
where the second and third terms on the right-hand side are the so-called
morphodiffusive terms, MD
`
and MD
`
, respectively (see also, for example,
Travkin and Catton [21]),
The solid-phase equation with constant k
`
equation is of the same form,
c
cx
s,
cT
`
}
`
cx
c
cx _
1
AD ,
c
`
U
T
`
ds
1
AD ,
c
`
U
cT
`
cx
'
ds
0, (30)
which can also be written in terms of the uctuating variable,
s,
c`T
`
cx`

c
cx _
1
AD ,
c
`
U
T
`
ds
1
AD ,
c
`
U
cT
`
cx
'
ds
0. (31)
Travkin and Catton [16, 18, 20] suggested that the integral heat transfer
terms in Eqs. (28, (30), and (31) be closed in a natural way by a third (III)
kind of heat transfer law. The second integral term reects the changing
averaged surface temperature along the x coordinate. Equations (28) and
(30) can be treated using heat transfer correlations for the heat exchange
integral term (the last term). Regular dilute arrangements of pores, spherical
particles, or cylinders have been studied much more than random mor-
phologies. Using separate element or cell modeling methods (Sangani and
Acrivos [32] and Gratton et al. [26]) to nd the interface temperature eld
allows one to close the second, surface diffusion integral terms in (28), (30),
and (27).
Many forms of the energy equation are used in the analysis of transport
phenomena in porous media. The primary difference between such equa-
tions and those resulting from a more rigorous development based on VAT
are certain additional terms. The best way to evaluate the need for these
additional more complex terms is to obtain an exact mathematical solution
and compare the results with calculations using the VAT equations. This
will clearly display the need for using the more complex VAT mathematical
statements.
Consider a two-phase heterogeneous medium consisting of an isotropic
continuous (solid or uid) matrix and an isotropic discontinuous phase
(spherical particles or pores). The volume fraction of the matrix, or f-phase,
is m,m
AD
/AD, and the volume fraction of ller, or s-phase, is

m
`
1 m
AD
`
/AD, where ADAD
AD
`
is the volume of the REV.
The constant properties (phase conductivities, k
and k
`
), stationary (time-
independent) heat conduction differential equations for T
and T
`
, the local
phase temperatures, are
V q
V`T
0, V q
`
k
`
V`T
`
0,
with the fourth (IVth) kind interfacial ( f s) thermal boundary conditions
T
T
`
, ds
c
`
U
ds
q
`
c
`
U
.
Here q
VT
and q
`
k
`
VT
`
are the local heat ux vectors, cS
"
is the interfacial surface, and ds
is the unit vector outward to the s-phase.

No internal heat sources are present inside the composite sample, so the
temperature eld is determined by the boundary conditions at the external
surface of the sample. After correct formulation of these conditions, the
problem is completely stated and has a unique solution.
Two ways to realize a solution to this problem were compared (Travkin
and Kushch [33, 34]). The rst is the conventional way of replacing the
actual composite medium by an equivalent homogeneous medium with an
effective thermal conductivity coefcient, k k
(s,, k
, k
`
), assuming one
knows how to obtain or calculate it. The exact effective thermal coefcient
was obtained using direct numerical modeling (DNM) based on the math-
ematical theory of globular morphology multiphase elds developed by
Kushch (see, for example, [3538]).
Averaging the heat ux, q,, and temperature, T ,, over the REV yields
q,k
VT ,, and for the stationary case there results

V (k
VT ,) 0. (32)
The boundary conditions for this equation are formulated in the same
manner as for a homogeneous medium.
The second way is to solve the problem using the VAT two-equation,
three-term integrodifferential equations (28) and (30). To evaluate and
compare solutions to these equations with the DNM results, one needs to
know the local solution characteristics, the averaged characteristics over the
both phases in each cell, and, in this case, the additional morphodiffusive
terms.
An innite homogeneous isotropic medium containing a three-dimen-
sional (3D) array of spherical particles is chosen for analysis. The particles
are arranged so that their centers lie at the nodes of a simple cubic lattice
with period a. The temperature eld in this heterogeneous medium is caused
by a constant heat ux Q
prescribed at the sample boundaries, which,

because of the absence of heat sources, leads to the equality of averaged
internal heat ux q,Q
.
When all the particles have the same radii, the result is the triple periodic
structure used widely, beginning from Rayleighs work [39], to evaluate the
effective conductivity of particle-reinforced composites.
The composite medium model consists of the three regions shown in Fig.
3. The half-space lying above the AA plane has a volume content of the
disperse phase m
`
m
, and for the half-space below the BB plane

m
`
m
"
. To dene the problem, let m
m
"
. The third part is the
composite layer between the plane boundaries AA and BB containing N
double periodic lattices of spheres (screens) with changing diameters.
Solutions to the VAT equations (28) and (30) for a composite with
varying volume content of disperse phase with accurate DNM closure of the
micro model VAT integrodifferential terms were obtained implicitly, mean-
ing that each term was calculated independently using the results of DNM
calculations.
For the one-dimensional case, Eq. (32) becomes
c
cz
k
cT
cz
0, (z
z z
`
), m
`
m
`
(z),
where k
(m
`
) is the effective conductivity coefcient.
Fic. 3. Model of two-phase medium with variable volume fraction of disperse phase.
The normalized solution of the both models (VAT and DNM) for the case
of linearly changing porosity m
`
m''
`
z(m'`'
`
m''
`
), where m
`
(z
) m''
`
,
m
`
(z
`
) m'`'
`
, z
0, z
`
1, between AA and BB and with effective
conductivity coefcients of k
0, 0.2, 1, 10, and 10,000, are presented in

Fig. 4. There is practically no difference (less than 10`) between the
solutions, and what there is is probably because of numerical error accumu-
lation (Travkin and Kushch [34]).
Lines 15 represent solutions of the one-term equation, respectively,
whereas the points (circles, triangles, etc.) represent the solutions of the VAT
equations with accurate DNM closure of the micro model VAT integrodif-
ferential terms MD
`
and MD
`
for the composite with varying volume
content of disperse phase. Here the number of screens is nine, corresponding
to a relatively small particle phase concentration gradient.
The coincidence of the results of the exact calculation of the two-equation,
three-term conductive-diffusion transport VAT model (28) and (30) with the
exact DNM solution and with the one-temperature effective coefcient
model for heterogeneous media with nonconstant spatial morphology
clearly demonstrates the need for using all the terms in the VAT equations.
The need for the morphodiffusive terms in the energy equation is further
demonstrated by noting that their magnitudes are all of the same order.
Conrmation of the fact that there is no difference in solutions between
the correct one-term, one-temperature effective diffusivity equation and the
Fic. 4. Comparison of VAT three-term equation particle temperature (symbols) with the
exact analytical based on the effective conductance coefcient obtained by exact DNM (solid
lines).
three-terms, two-temperature VAT equations does not mean that it is better
to take for modeling and analysis the effective diffusivity one-term, one-
temperature equation (see Subsection VI, E and arguments in Sections VII
and VIII). Among other issues one needs to analyze goals of modeling and
to understand that the good solution of the effective diffusivity one-term
one-temperature equation as it was found and described in the preceding
statements means nothing less than the ground of the exact solution of the
VAT problem. Also, it is important that for the exact (or accurate) solution
of conventional diffusivity equation, the effective coefcient needs to be
found, and this means in turn that nding the solution of the two-eld
problem is imperative and consequently appears to be the major problem.
Meanwhile, this is the problem that was posed just at the beginning as the
original one.
III. Nonlinear and Turbulent Transport in Porous Media
To a great extent, the analysis of porous media linear transport phenom-
ena are given in the numerous studies by Whitaker and coauthors; see, for
example, [10, 30, 31, 4046], as well as by studies by Gray and coauthors
[8, 4750]. Our present work is mostly devoted to the description of other
physical elds, along with development of their physical and mathematical
models. Still, the connection to linear and partially linear problem state-
ments needs to be outlined.
The linear Stokes equations are
VV 0,
0 Vp jV`V j
g, (33)
and although the Stokes equation is adequate for many problems, linear as
well as nonlinear processes will result in different equations and modeling
features.
The general averaged form of the transport equations will be developed
for permeable interface boundaries between the phases. Two forms of the
right-hand-side Laplacian term will be considered. First, one can have two
forms of the diffusive ux in gradient form that can be written
jVV ,
jVV ,
j
AD ,
c
`
U
V ds (34)
or
jVV ,
jm,VV
j
AD ,
c
`
U
V ds. (35)
It was pointed out rst by Whitaker [42, 43] that these forms allow greater
versatility in addressing particular problems. Using the two averaged forms
of the velocity gradient, (34) and (35), one can obtain two averaged versions
of the diffusion term in Eq. (33), namely,
jV(VV ),
jV (Vm,V ) jV
_
1
AD ,
c
`
U
Vds
j
AD ,
c
`
U
VV ds,
(36)
where the production term VV ds is a tensorial variable, and the version
with uctuations in the second integral term
jV(VV ),
jV (m,VV ) jV
_
1
AD ,
c
`
U
V ds
j
AD ,
c
`
U
VV ds,
(37)
Using these two forms of the momentum viscous diffusion term, one can
write two versions of the averaged Stokes equations. The rst version is
VV ,
1
AD ,
c
`
U
U
'
ds 0, U
'
V (38)
and
0 Vp,
1
AD ,
c
`
U
pds jV (Vm,V )
jV
_
1
AD ,
c
`
U
V ds
j
AD ,
c
`
U
VV ds m,j
g , (39)
and the second version is found by using the following relation for the
pressure gradient:
Vp,
1
AD ,
c
`
U
pds m,Vp`
1
AD ,
c
`
U
p` ds. (40)
Using the averaging rules developed by Primak et al. [14], Shcherban et
al. [15] and Travkin and Catton [16, 18] facilitated the development of the
momentum equation. By combining equations (37) and (40), one is able to
write the momentum transport equations in the second form with velocity
uctuations
VV ,
V }
Vm,
1
AD ,
c
`
U
U
'
ds 0, (41)
obtained using
V V }
V
1
AD ,
c
`
U
V ds V }
Vm,
1
AD ,
c
`
U
U
'
ds, (42)
and the momentum equation
0 m,Vp`
1
AD ,
c
`
U
p` ds jV (m,VV )
jV
_
1
AD ,
c
`
U
V ds
j
AD ,
c
`
U
VV ds m,j
g . (43)
The third version of these equations is almost never used but can be found
in [21].
A. L:xiN: Fiov vi1n CoNs1:N1 CorrricirN1s
The transport equations for a uid phase with linear diffusive terms are
cU
'
cx
'
0 (44)
cU
'
ct
U
'
cU
'
cx
'
1
j
cp
cx
'
v
c
cx
'

cU
'
cx
'
(45)
cu
ct
U
'
cu
cx
'
D
c
cx
'

cu
cx
'
S
"
D
. (46)
Here u represents any scalar eld (for example, concentration C) that might
be transported into either of the porous medium phases, and the last terms
on the right-hand side of (45) and (46) are source terms. In the solid phase,
the diffusion equation is
cu
`
ct
D
`
c
cx
'

cu
`
cx
'
S
"
Q
. (47)
The averaged convective operator term in divergence form becomes, after
phase averaging,
c
cx
'
(U
'
U
'
)
V(U
'
U
'
),
VU
'
U
'
,
1
AD ,
c
`
U
U
'
U
'
ds
V[m,U
'
U
'
m,u`
'
u`
'
}
]
1
AD ,
c
`
U
U
'
U
'
ds. (48)
Decomposition of the rst term on the right-hand side of (48) yields
uctuation types of terms that need to be treated in some way.
The nondivergent version of the averaged convective term in the momen-
tum equation is
c
cx
'
(U
'
U
'
)
m,U
'
VU
'
U
'
VU
'
,
Vu`
'
u`
'
,
1
AD ,
c
`
U
U
'
U
'
ds
m,U
'
c
cx
'
U
'
U
'
1
AD ,
c
`
U
U
'
ds
Vu`
'
u`
'
,
1
AD ,
c
`
U
U
'
U
'
ds. (49)
The divergent and nondivergent forms of the averaged convective term in
the diffusion equation are
V(CU
'
),
VCU
'
,
1
AD ,
c
`
U
CU
'
ds
V[m,CU
'
m,c` u`
'
}
]
1
AD ,
c
`
U
CU
'
ds
m,U
'
c
cx
'
CC
1
AD ,
c
`
U
U
'
ds Vc` u`
'
,
1
AD ,
c
`
U
CU
'
ds.
(50)
Other averaged versions of this term can be obtained using impermeable
interface conditions (see also Whitaker [42] and Plumb and Whitaker [44]).
For constant diffusion coefcient D, the averaged diffusion term becomes
V (DVC),
DV V(m,C ) DV
_
1
AD ,
c
`
U
Cds
D
AD ,
c
`
U
VC ds,
(51)
or
V (DVC),
DV (m,VC ) DV
_
1
AD ,
c
`
U
c` ds
D
AD ,
c
`
U
VC ds,
(52)
or
D (DVC),
Dm,V`C DV
_
1
AD ,
c
`
U
c`ds
D
AD ,
c
`
U
Vc` ds.
(53)
Other forms of Eq. (52), using the averaging operator for constant
diffusion coefcient, constant porosity, and absence of interface surface
permeability and transmittivity, can be found in works by Whitaker [42]
and Plumb and Whitaker [44], as well as by Levec and Carbonell [46].
A similar derivation can be carried out for the momentum equation to
treat cases where Stokes ow is invalid. Two versions of the momentum
equation will result. The equation without the uctuation terms is
j
m,
cV
ct
m,V VV V
1
AD ,
c
`
U
V ds Vv`v`,
1
AD ,
c
`
U
V V ds
V(m,p` )
1
AD ,
c
`
U
pds jV V(m,V )
jV
_
1
AD ,
c
`
U
V ds
j
AD ,
c
`
U
VV ds m,j
g . (54)
with the uctuation diffusion terms it becomes
j
m,
cV
ct
m,V VV V
1
AD ,
c
`
U
V ds Vv`v`,
1
AD ,
c
`
U
V V ds
m,Vp`
1
AD ,
c
`
U
p` ds jV (m,VV )
jV
_
1
AD ,
c
`
U
v`ds
j
AD ,
c
`
U
VV ds m,j
g . (55)
The steady-state momentum transport equations for systems with imper-
meable interfaces can readily be derived from Eq. (54) and (55). They are
j
(m,V VV Vv`v`,
) V(m,p` )
1
AD ,
c
`
U
pds jV V(m,V )
j
AD ,
c
`
U
VV ds m,j
g , (56)
or
j
(m,V VV Vv`v`,
) m,Vp` )
1
AD ,
c
`
U
p` ds jV V(m,V )
j
AD ,
c
`
U
VV ds m,j
g . (57)
B. NoNiiNr: Fiiin Mrniix Eqi:1ioNs iN L:xiN: Fiov
To properly account for Newtonian uid ow phenomena within a
porous medium in a general way, modeling should begin with the Navier
Stokes equations for variable uid properties,
j

cV
ct
V VV
Vp V [j(VV (VV )*)] j
g (58)
j j(V, C
'
, T ),
rather than the constant viscosity NavierStokes equations. The following
form of the momentum equation will be used in further developments:
j
cV
ct
V VV
Vp V (2jS) j
g (59)
j j(V, C
'
, T ).
The negative stress tensor o
''
in this equation is
N
''
o
''
2j(VV )` 2jS, (60)
and the symmetric tensor S is the deformation tensor
S(VV )`
1
2
(VV (VV )*), (61)
with (VV )* being the transposed diad VV.
The homogeneous phase diffusion equations are
cu
ct
U
'
cu
cx
'
c
cx
'
(z
(x, u
, V )
cu
cx
'
S
"
D
(62)
and
cu
`
ct

c
cx
'

z
`
cu
`
cx
'
S
"
Q
. (63)
Here u
and z are scalar elds and nonlinear diffusion coefcients for

these elds. The averaging procedures for transport equation convective
terms were established earlier. The averaged nonlinear diffusion term yields
V (DVC),
V (m,DVC ) V
_
D
1
AD ,
c
`
U
c` ds
V (DVc`,
)
1
AD ,
c
`
U
DVC ds. (64)
The other version of the diffusive terms with the full value of concentration
on the interface surface is
V (DVC),
V (DV(m,C )) V
_
D
1
AD ,
c
`
U
Cds
V (DVc` ,
)
1
AD ,
c
`
U
DVC ds. (65)
General forms of the nonlinear transport equations can be derived for
impermeable and permeable interface surfaces. The averaged momentum
diffusion term is
c
cx
'
(2jS )
V (2jS ),
V (2jS ,
)
1
AD ,
c
`
U
2jS ds
V 2(m,j`S m,j` S}
)
2
AD ,
c
`
U
jS ds. (66)
The general nonlinear averaged momentum equation for a porous medium
is
j
m,
cV
ct
m,V VV V
1
AD ,
c
`
U
V ds Vv` v` ,
1
AD ,
c
`
U
V V ds
V(m,p` )
1
AD ,
c
`
U
pds V 2(m,j` S m,j` S}
2
AD ,
c
`
U
jS ds m,j
g . (67)
The steady-state momentum transport equations for systems with imper-
meable interfaces follows from Eq. (67),
j
(m,V VV Vv` v` ,
)
V(m,p` )
1
AD ,
c
`
U
pds V 2(m,j` S m,j` S}
j
AD ,
c
`
U
jS ds m,j
g . (68)
The averaged nonlinear mass transport equation in porous medium
follows
m,
cC
ct
m,U
'
VC
AD ,
c
`
U
U
'
ds Vc`
u`
'
,
1
AD ,
c
`
U
C
U
'
ds
V (D V(m,C )) V
_
D
1
AD ,
c
`
U
Cds
V (DVc`,
)
1
AD ,
c
`
U
DVC ds m,S
'
D
. (69)
A few simpler transport equations that can be readily used while main-
taining fundamental relationships in heterogeneous medium transport are
given by Travkin and Catton [21].
C. Poois Mrniix TiniirN1 VAT Eqi:1ioNs
Turbulent transport processes in highly structured or porous media are
of great importance because of the large variety of heat- and mass-exchange
equipment used in modern technology. These include heterogeneous media
for heat exchangers and grain layers, packed columns, and reactors. In all
cases there occurs a jet or stalled ow of uids in channels or around the
obstacles. There are, however, few theoretical developments for ow and
heat exchange in channels of complex conguration or when owing around
nonhomogeneous bodies with randomly varied parameters. The advanced
forms of laminar transport equations in porous media were developed in a
paper by Crapiste et al. [41]. For turbulent transport in heterogeneous
media, there are few modeling approaches and their theoretical basis and
nal modeling equations differ.
The lack of a sound theoretical basis affects the development of math-
ematical models for turbulent transport in the complex geometrical environ-
ments found in nuclear reactors subchannels where rod-bundle geometries
are considered to be formed by subchannels. Processes in each subchannel
are calculated separately (see Teyssedou et al. [51]). The equations used in
this work has often been obtained from two-phase transport modeling
equations [52] with heterogeneity of spacial phase distributions neglected in
the bulk. Three-dimensional two-uid ow equations were obtained by Ishii
[52] using a statistical averaging method. In his development, he essentially
neglected nonlinear phenomena and took the ux forms of the diffusive
terms to avoid averaging of the second power differential operators. Ishii
and Mishima [53] averaged a two-uid momentum equation of the form
c:
'
j
'
v
'
ct
V (:
'
j
'
v
'
v
'
) :
'
Vp
'
V :
'
(t' t'
'
)
:
'
j
'
g v
''
I
'
M
''
V:
'
t
'
, (70)
where :
'
is the local void fraction, t
'
is the mean interfacial shear stress, t'
'
is the turbulent stress for the kth phase, t' is the averaged viscous stress for
the kth phase, I
'
is the mass generation, and M
''
is the generalized
interfacial drag. Using the area average in the second time averaging
procedure, Ishii and Mishima [53] introduced a distribution of parameters
to take into consideration the nonlinearity of convective term averaging.
This approach cannot strictly take into account the stochastic character of
various kinds of spatial phase distributions. The equations used by Lahey
and Lopez de Bertodano [54] and Lopez de Bertodano et al. [55] are very
similar, with the momentum equation being
:
'
j
'
D
'
u
''
Dt
:
'
Vp
'
V :
'
[j
'
Vu
''
j
'
(u'
''
u'
''
)]
:
'
j
'
g M
''
M
"'
t
'
V:
'
( p
'
G
p
'
)V:
'
. (71)
Here the index i denotes interfacial phenomena and M
"'
is the volumetric
wall force on phase k. Additional terms in Eq. (70) and (71) are usually
based on separate micro modeling efforts and experimental data.
One of the more detailed derivations of the two-phase ow governing
equations by Lahey and Drew [56] is based on a volume averaging
methodology. Among the problems was that the authors developed their
own volume averaging technique without consideration of theoretical ad-
vancements developed by Whitaker and colleagues [10, 42] and Gray et al.
[8] for laminar and half-linear transport equations. The most important
weaknesses are the lack of nonlinear terms (apart from the convection
terms) that naturally arise and the nonexistence of interphase uctuations.
Zhang and Prosperetti [57] derived averaged equations for the motion of
equal-sized rigid spheres suspended in a potential ow using an equation for
the probability distribution. They used the small particle dilute limit
approximation to close the momentum equations. After approximate
resolution of the continuous phase uctuation tensor M
'
, the vector
A
'
(x, t), and the uctuating particle volume ux tensor, M
'
, they recog-
nized that (p. 199) Closure of the system requires an expression for the
uctuating particle volume ux tensor M
'
. . . . This missing information
cannot be supplied internally by the theory without a specication of the
initial conditions imposed on the particle probability distribution. They
also considered the case of nite volume fractions for the linear problem
where the problem equations were formulated for inviscid and unconvec-
tional media. The development by Zhang and Prosperetti [57] is a good
example of the correct application of ensemble averaging. The equations
they derive compare exactly with those derived from rigorous volume
averaging theory (VAT) [24].
Transport phenomena in tube bundles of nuclear reactors and heat
exchangers can be modeled by treating them as porous media [58]. The
two-dimensional momentum equations for a constant porosity distribution
usually have the form [59]
cU
cx

cV
cy
0 (72)
cU`
cx

cUV
cy

1
j
cP
cx
v
VÙA
`
VU (73)
cUV
cx

cV `
cy

1
j
cP
cy
v
V`V A
'
VV, n 0, (74)
where the physical quantities are written as averaged values and the solid
phase effects are included in two coefcients of bulk resistance, A
`
and A
'
,
and an effective eddy viscosity, v
, that is not equal to the turbulent eddy

viscosity. These kinds of equations were not designed to deal with non-
linearities induced by the physics of the problem and the medium variable
porosity or to take into account local inhomogeneities.
Some of the more interesting applications of turbulent transport in
heterogeneous media are to agrometeorology, urban planning, and air
pollution. The rst signicant papers on momentum and pollutant diffusion
in urban environment treated as a two-phase medium were those by Popov
[60, 61]. In these investigations, an urban porosity function was dened
based on statistical averaging of a characteristic function p(x, y, z) for the
surface roughness that is equal to zero inside of buildings and other
structures and equal to unity in an outdoor space. The turbulent diffusion
equation for an urban roughness porous medium after ensemble averaging
is
cm(x
'
)C
,
ct

c
cx
'
(m(x
'
)V
'
,C
,)
c
cx
'
(v
'
)'(c
)',
c
cx
'
v'
'
c'
,
c
cx
'

D
cC
,
cx
'

, n 1, 2, 3, 4 . . . ,
(75)
where , means porous volume ensemble averaging, and m(x
'
) is porosity.
Closure of the two morphological terms, the rst and the second terms on
the right-hand side, was obtained using a Boussinesque analogy,
c
cx
'
(v
'
)'(c
)',
c
cx
'
v'
'
c'
,K
''
cC
,
cx
'
. (76)
A descriptive analysis of the deviation variables (v
'
)', (c
)' and the effective

diffusion coefcient K
''
was not given. In many studies of meteorology and
agronomy, the only modeling of the increase in the volume drag resistance
is by addition of a nonlinear term as done by Yamada [62],
cU
ct
f
'
V
1
j
cP
cx

c
cz
(u'w') (1 m
`
)c
S(z)U U (77)
cV
ct
f
'
U
1
j
cP
cy

c
cz
(v'w') (1 m
`
)c
S(z)V V ,
where (1 m
`
) is the fraction of the earth surface occupied by forest, m
`
is
the area porosity due to a tree volume, and f
'
is a Coriolis parameter.
The averaging technique used by Raupach and Shaw [63] to obtain a
turbulent transport equation for a two-phase medium of agro- and forest
cultures is a plain surface 2D averaging procedure where the averaged
function is dened by
f ,
1
AD
,
`'
ND
f dc, (78)
with AD
being the area within the volume AD
occupied by air. Raupach

et al. [64] and Coppin et al. [65] assumed that the dispersive covariances
were unimportant,
u "
'
u "
'
,
, (79)
where u "
'
is a uctuation value within the canopy and u "
'
u'
'
. The
contribution of these covariances was found by Raupach et al. [64] to be
small in the region just above the canopy from experiments with a regular
rough morphology. This nding has been explained by Scherban et al. [15],
Primak et al. [14], and Travkin and Catton [16, 20] for regular porous
(roughness) morphology. Covariances are, however, the result of irregular
or random two-phase media. When the surface averaging used by Raupach
and Shaw [64] is used instead of volume averaging, especially in the case of
nonisotropic media, the neglect of one of the dimensions in the averaging
process results in an incorrect value. This result should be called a 2D
averaging procedure, particularly when 3D averaging procedures are re-
placed by 2D for nonisotropic urban rough layer (URL) when developing
averaged transport equations.
Raupach et al. [6466] later introduced a true volume averaging pro-
cedure within an air volume AD
that yielded the averaged equation for

momentum conservation
cU
'
ct
U
'
c
cx
'
(U
'
)
1
j
c
cx
'
P
c
cx
'
u'
'
u'
'
}
vAU
'
v
AD
,
c
`
U
c
cx
'
U
'
ds
c
cx
'
u "
'
u "
'
}
1
j
AD
,
c
`
U
Pds, i, j 1 3, (80)
where cS
"
is interfacial area. Development of this equation is based on
intrinsic averaged values of }
or U
'
, whereas averages of vector eld
variables over the entire REV are more correct (Kheifets and Neimark
[11]). Raupach et al. [64] next simplied all the closure requirements by
developing a bulk overall drag coefcient. The second, third, and fth terms
on the right-hand side of Eq. (80) are represented by a common drag
resistance term. For a stationary fully developed boundary layer, they write
c
cz _
u'w'}
u "w "}
v
cU
cz
1
2
C
`, (81)
where C
is an element drag coefcient and S
is an element area
densityfrontal area per unit volume.
A wide range of ow regimes is reported in papers by Fand et al. [67]
and Dybbs and Edwards [68]. The latter work revealed that there were four
regimes for regular spherical packing, and that only when the Reynolds
number based on pore diameter, Re
'"
, exceeded 350 could the ow regime
be considered to be turbulent ow. The Fand et al. [67] investigation of a
randomnly packed porous medium made up of single size spheres showed
that the fully developed turbulent regime occurs when Re
'
120, where Re
is particle Reynolds number.

Volume averaging procedures were used by Masuoka and Takatsu [69]
to derive their volume-averaged turbulent transport equations. As in numer-
ous other studies of multiphase transport, the major difculties of averaging
the terms on the right-hand side were overcome by using assumed closure
models for the stress components. As a result, the averaged turbulent
momentum equation, for example, has conventional additional resistance
terms such as the averaged momentum equation developed by Vafai and
Tien [70] for laminar regime transport in porous medium. A major
assumption is the linearity of the uctuation terms obtained, for example,
by neglect of additional terms in the momentum equation.
A meaningful experimental study by Howle et al. [71] conrmed the
importance of the role of randomness in the enhancement of transport
processes. The results show the very distinct patterns of ow and heat
transfer for two cases of regular and nonuniform 2D structured nonorthog-
onal porous media. Their experimental results clearly demonstrate the
inuence of nonuniformity of the porous structure on the enhancement of
heat transfer.
D. DrvrioixrN1 or TiniirN1 T:Nsio1 Monris
iN Hicnix Poois Mrni:
Fluid ow in a porous layer or medium can be characterized by several
modes. Let us single out from among them the three modes found in a
highly porous media. The rst is ow around isolated ostacle elements, or
inside an isolated pore. The second is interaction of traces or a hyperturbu-
lent mode. The third is uid ow between obstacles or inside a blocked
interconnected swarm of channels (ltration mode). The models developed
by Scherban et al. [15], Primak et al. [14], and Travkin and Catton [1621]
are primarily for nonlinear laminar ltration and hyperturbulent modes in
nonlinear transport.
Specic features of ows in the channels of ltered media include the
following:
1. Increased drag due to microroughness on the channel boundary
surfaces
2. Gravity effects
3. Free convection effects
4. The effects of secondary ows of the second kind and curved stream-
lines
5. Large-scale vortex effects
6. The anisotropic nature of turbulent transfer and resulting anisotropy
of turbulent viscosity
It is well known that in spacial boundary ows, an important role is
played by the gradients of normal Reynolds stresses and that this is the case
for ows in porous medium channels as well. As a rule, ow symmetry is
not observed in these channels. Therefore, in channel turbulence models, the
shear components of the Reynolds stress tensors have a decisive effect on
the ow characteristics. At present, however, turbulence models that are less
than second-order can not be successfully employed for simulating such
ows (Rodi [72], Lumley [73], and Shvab and Bezprozvannykh [74]).
Derivation of the equations of turbulent ow and diffusion in a highly
porous medium during the ltration mode is based on the theory of
averaging of the turbulent transfer equation in the liquid phase and the
transfer equations in the solid phase of a heterogeneous medium (Primak et
al. [14] and Scherban et al. [15]) over a specied REV.
The initial turbulent transport equation set for the rst level of the
hierarchy, microelement, or pore, was taken to be of the form (see, for
example, Rodi [72] and Patel et al. [75])
cU
'
ct
U
'
cU
'
cx
'
1
j
cp
cx
'
c
cx
'

v
cU
'
cx
'
u'
'
u'
'
S
'
G
(82)
cu
ct
U
'
cu
cx
'
c
cx
'

D
cu
cx
'
u'
'
'
S
"
D
(83)
cU
'
cx
'
0. (84)
Here u
and its uctuation represent any scalar eld that might be

transported into either of the porous medium phases, and the last terms on
the right-hand side of (82) and (83) are source terms.
Next we introduce free stream turbulence into the hierarchy Let us
represent the turbulent values as
UU u' U
'
U
'
u'
'
u'
'
(85)
U U
u
`
,
where the index k stands for the turbulent components independent of
inhomogeneities of dimensions and properties of the multitude of porous
medium channels (pores), and r stands for contributions due to the porous
medium inhomogeneity. Being independent of the dimensions and proper-
ties of the inhomogeneities of the porous medium congurations, sections,
and boundary surfaces does not mean that the distribution of values of U
'
and u'
'
are altogether independent of the distance to the wall, pressure
distribution, etc. Thus, the values U
'
or u'
'
stand for the values generally
accepted in the turbulence theory, that is, when a plane surface is referred
to, these values are those of a classical turbulent boundary layer. When a
round-section channel is involved, and even if the cross-section of this
channel is not round, but without disturbing nonhomogeneities in the
section, then the characteristics of this regular sections (and ow) may be
considered to be those that could be marked with index k. Hence, if a
channel in a porous medium can be approximately by superposition of
smooth regular (of regular shape) channels, it is possible to give such a ow
its characteristics and designated them with the index k, which stands for
the basic (canonical) values of the turbulent quantities.
Triple decomposition techniques have been used in papers by Brereton
and Kodal [76] and Bisset et al. [77], among others. The latter utilized
triple decomposition, conditional averaging, and double averaging to ana-
lyze the structure of large-scale organized motion over the rough plate.
It should be noted that there are problems where U
'
and u'
'
can be found
from known theoretical or experimental expressions (correlations) where the
denitions of U
'
and u'
'
are equivalent to the solution of an independent
problem (for example, turbulent ow in a curved channel). The same thing
can be said about ow around a separate obstacle located on a plain surface.
In this case one can write
U U
U
'
U
'
, u' u'
'
. (86)
The term u'
'
u
`
appears if the ow is through a nonuniform array of
obstacles. If all the obstacles are the same and ordered, then u
`
can be taken
equal to 0. Naturally, the term u'
'
in this particular case does not equal the
uctuation vector u'
'`
over a smooth, plain surface.
The following hypothesis about the additive components is developed to
correct the foregoing deciencies:
U U
u
`
U
'
U
'
}
u'
'
, u' u'
'
U
'
U
'
U
'
U
'
, u
`
u'
'
(u'
'
) 0, u
`
}
u'
'
}
0. (87)
It should be noted that solutions to the equations for the turbulent
characteristics may be inuenced by external parameters of the problem,
namely, by the coefcients and boundary conditions, which themselves can
carry information about porous medium morphological features. The adop-
tion of a hypothesis about the additive components of functions represen-
ting turbulent ltration facilitates the problem of averaging the equations
for the Reynolds stresses and covariations of uctuations (ows) in pores
over the REV.
After averaging the basic initial set of turbulent transport equations over
the REV and using the averaging formalism developed in the works by
Primak et al. [14], Shcherban et al. [15], and Primak and Travkin [78], one
obtains equations for mass conservation,
c
cx
'
U
'
,
1
AD ,
c
`
U
U
'
ds 0, (88)
for turbulent ltration (with molecular viscosity terms neglected for
simplicity),
m,
cU
'
ct

c
cx
'
(m,U
'
U
'
)
1
j
c
cx
'
(m,p
`
)
c
cx
'
u'
'
u'
'
,
c
cx
'
u
`
'
u
`
'
,
1
j
AD ,
c
`
U
pds
1
AD ,
c
`
U
U
'
U
'
ds
1
AD ,
c
`
U
u'
'
u'
'
ds m,S
'
G
, i, j 13, (89)
and for scalar diffusion (with molecular diffusivity terms neglected),
m,
cu
ct

c
cx
'
(m,U
'
u
c
cx
'
u'
'
'
c
cx
'
u
`
'
1
AD ,
c
`
U
U
'
u
ds
1
AD ,
c
`
U
u'
'
'
ds m,S
"
D
, i 13. (90)
Many details and possible variants of the preceding equations with
tensorial terms are found in Primak et al. [14], Scherban et al. [15], and
Travkin and Catton [16, 21]. Using an approximation to K-theory in an
elementary channel (pore), the equation for turbulent diffusion of nth species
takes the following more complex form after being averaged:
m,
cC
ct
V ,
VC
Vv
`
'
c
`
V (K
'
V(m,C
)) V
_
K
'
1
AD ,
c
`
U
C
ds
V (k
'
Vc
`
)
1
AD ,
c
`
U
K
'
VC
ds
AD ,
c
`
U
U
'
ds
1
AD ,
c
`
U
C
U
'
ds m,S
,
n 1, 2, 3, 4 . . . . (91)
In the more general case, the momentum ux integrals on the right-hand
sides of Eq. (89) through (91) do not equal zero, since there could be
penetration through the phase transition boundary changing the boundary
conditions in the microelement to allow for heat and mass exchange through
the interface surface as the values of velocity, concentrations, and tempera-
ture at cS
"
do not equal zero (see also Crapiste et al. [41]). The rst term
on the right-hand side of Eq. (91) is the divergence of the REV averaged
product of velocity uctuations and admixture concentration caused by
random morphological properties of the medium being penetrated and is
responsible for morphoconvectional dispersion of admixture in this particu-
lar porous medium. The third term on the right-hand side of Eq. (91) can
be associated with the notion of morphodiffusive dispersion of a substance
or heat in a randomly nonhomogeneous medium. The term with S
may also
reect, specically, the impact of microroughness from the previous level of
the simulation hierarchy. The importance of accounting for this roughness
has been demonstrated by many studies. The remaining step is to account
for the microroughness characteristics of the previous level.
One-dimensional mathematical statements will be used in what follows
for simplicity. Admission of specic types of medium irregularity or random-
ness requires that complicated additional expressions be included in the
generalized governing equations. Treatment of these additional terms be-
comes a crucial step once the governing averaged equations are written. An
attempt to implement some basic departures from a porous medium with
strictly regular morphology descriptions into a method for evaluation of
some of the less tractable, additional terms is explained next.
The 1D momentum equation with terms representing a detailed descrip-
tion of the medium morphology is depicted as
c
cx
m,(K
"
v)
cU
cx
c
cx
K
"
cu
`
cx
c
cx
(u
`
u
`
,
)
m, U
cU
cx
'
1
AD ,
c
`
U
(K
"
v)
cU
cx
'
ds
1
j
AD ,
c
`
U
pds
1
j
c
cx
(m,p
`
)
m,U
cU
cx
'
u`
*''
S
"
(x)
1
j
AD ,
c
`
U
p ds
1
j
c
cx
(m,p
`
), (92)
where K
"
is the turbulent eddy viscosity, and u`
*''
is the square friction
velocity at the upper boundary of surface roughness layer h
'
averaged over
interface surface S
"
.
General statements for energy transport in a porous medium require
two-temperature treatments. Travkin et al. [19, 26] showed that the proper
form for the turbulent heat transfer equation in the uid phase using
one-equation K-theory closure with primarily 1D convective heat transfer is
c
m,U
cT
cx

c
cx _
m,(K
'
k
)
cT
cx
c
cx
K
'
cT
cx
c
cx
(m,T
u
`
}
)
c
cx _
(K
'
k
)
AD ,
c
`
U
T
ds
1
AD ,
c
`
U
(K
'
k
)
cT
cx
'
ds, (93)
whereas in the neighboring solid phase, the corresponding equation is
c
cx
(1 m,)K
`'
}
`
cT
`
}
`
cx
c
cx
K
`'
cT
`
cx
`
c
cx _
K
`'
}
`
AD ,
c
`
U
T
`
ds
1
AD ,
c
`
U
K
`'
cT
`
cx
'
ds
. (94)
The generalized longitudinal 1D mass transport equation in the uid
phase, including description of potential morphouctuation inuence, for a
medium morphology with only 1D uctuations is written
c
cx _
m,(K
'
D
)
cC
cx
c
cx
K
'
cC
cx
c
cx
(m,c
`
u
`
}
)
c
cx _
(K
'
D
)
AD ,
c
`
U
c
`
ds
1
AD ,
c
`
U
(K
'
D
)
cC
cx
'
ds m,S
'
m,U
cC
cx
, (95)
whereas the corresponding nonlinear equation for the solid phase is
c
cx
(1 m,)D
`
}
`
cC
`
}
`
cx
c
cx
D
`
cc`
`
cx
`
c
cx _
D
`
}
`
AD ,
c
`
U
C
`
ds
1
AD ,
c
`
U
D
`
cC
`
cx
'
ds
. (96)
E. Ciosir Tnroirs :Nn Aiio:cnrs ro T:Nsio1
iN Poois Mrni:
Closure theories for transport equations in heterogeneous media have
been the primary measure of advancement and for measuring success in
research on transport in porous media. It is believed that the only way to
achieve substantial gains is to maintain the connection between porous
medium morphology and the rigorous formulation of mathematical equa-
tions for transport. There are only two well-known types of porous media
morphologies for which researchers have had major successes. But even for
these morphologies, namely straight parallel pores and equal-size spherical
inclusions, not enough evidence is available to state that the closure
problems for them are closed.
One of the few existing studies of closure for VAT type equations is by
Hsu and Cheng [79, 80]. They used a one-temperature averaged equation
[Equation (40a) in Hsu and Cheng [80]) without the morphodiffusive term
V [(k
k
`
)T (Vm,)] V [(k
`
k
)T (Vm,)].
The reasoning often applied to the morphoconvective term closure
problem in averaged scalar and momentum transport equations is that the
terms needing closure may be negligible. The basis for this reasoning is (see
Kheifets and Neimark [11])
c` VCd
'"
, and jDVC, so Vc` j,
DVC`,
d
'"
l
,
where l is the characteristic length associated with averaging volume (see,
for example, the work of Lehner [81] and others) and d
'"
the mean diameter
of pores in a REV. It is not obvious that the length scale, d
'"
, taken for the
approximation of c` follows from use of l as a scale for the second derivative.
Furthermore, assuming that the variable to be averaged over the REV
changes very slowly over the REV does not mean that it changes very slowly
in the neighborhood of the primary REV.
Various closure attempts for heterogeneous medium transport equations
resulted in various nal equations. One needs to know what these equations
are all about. Treatment of the one-dimensional heat conduction equation
with a stochastic function for the thermal diffusivity in a paper by Fox and
Barakat [82] yielded a spatially fourth-order partial differential equation to
be solved. Gelhar et al. [83], after having eliminated the second-order terms
in the species conservation equation for a stochastic media, were able to
develop an interesting procedure for deriving a mean concentration trans-
port equation. The equation form includes an innite series of derivatives
on the right-hand side of the equation. Analysis of this equation allows the
derivation of the nal form of the mass transport equation,
cC *
ct
U
cC *
cx
(A a
'
)U
c`C *
cx`
B
c`C *
ctcx`
BU
c`C *
cx`
,
where the most important term is the second term on the right-hand side.
In the derivation of this equation, the stochastic character of the existing
assigned elds of velocity, concentration, and dispersion coefcients were
assumed.
A simple form of the advective diffusion equation with constant diffusion
coefcients was developed without sorption effects by Tang et al. [84]:
cm,C
ct
m,V
'
VCDV (m,VC).
They transformed the equation with the help of ensemble averaging into a
stochastic transport equation,
cm,C *
ct
m,u` *
'
VC * DV (m,VC *) m,j
''
c`C *
cx
'
cx
'
,
where the tensor of the ensemble dispersion coefcient is a correlation
function denoted by
j
''
1
2
u'
'
u'
'
}*
u }*u}*
x u }*,
with u }* being the ensemble averaged velocity. The additional term,
reecting the inuence of the stochastic or inhomogeneous nature of the
spatial velocity and concentration uctuations in the ensemble averaged
stochastic equation developed by Tang et al. [84], has the dispersivity
coefcient fully dependent on the velocity uctuations. As can be seen by
this equation, the effect of concentration uctuations was eliminated.
Torquato and coauthors (see, for example, Torquato et al. [85], Miller
and Torquato [86], Kim and Torquato [87]) have been developing means
to characterize the various mathematical dependencies of a composite
medium microstructure in a statistically homogeneous medium. Some of the
quantities considered by Torquato are useful in obtaining resolution to
certain closure problems for VAT developed mathematical models of
globular morphologies. In particular, the different near-neighbor distance
distribution density functions deserve special mention (Lu and Torquato
[88], Torquato et al. [85]).
Carbonell and Whitaker [89] combined the methods of volume averaging
and the morphology approach to specify the dispersion tensor for the
problem of convective diffusion for cases where there is no reaction or
adsorption on the solid phase surface,
D
cC
cn
0, x - cS
"
,
and considered a constant diffusion coefcient and constant porosity m,,
which greatly simplies the closure problems. They expressed the spatial
deviation function as
c` f( r ) VC ,
where f is a vector function of position in the uid phase. Averaged
equations of convective diffusion are the same as the convective heat transfer
equation given by Levec and Carbonell [46] with the exclusion of ux
surface integral term. The closure technique used in their paper is analogous
to a turbulence theory scheme, helping them to derive the closure equation
for the spatial deviation function in the form of a partial differential
equation,
V (V V )V f DV` f , n Vf n , x - cS
"
,
One should note that the spatial deviation functions dened for a periodic
medium are periodic themselves.
Nozad et al. [40] suggested that the same closure scheme be used to
represent the uctuation terms T
and T
`
for a one-temperature model by
using
T
f VT , , T
`
g VT ,
for a transient heat conduction problem with constant coefcients in a
two-phase system (stationary). Partial differential equations for f , g , , and
are found. They obtained excellent predictions of the effective thermal
conductivity for conductivity ratios k k
`
/k
100.
Carbonell [90] attempted to obtain an averaged convective-diffusion
equation for a straight tube morphological model and obtained an equation
with three different concentration variables. This demonstrates that the
averaging procedures, taken too literally, can result in incorrect expressions
or conclusions.
A common form of the averaged governing equations for closure of
multiphase laminar transport in porous media was obtained by Crapiste et
al. [41]. They developed a closure approach that led to a complex integro-
differential equation for the spatial deviations of a substance in the void or
uid phase volume of the macro REV. This means that solving the
boundary value problem for spatial concentration uctuations, for example,
requires that one obtain a solution to second-order partial differential or
coupled integro differential equations in a real complex geometric volume
within the porous medium.
For a heterogeneous porous medium, this means that the coupled
integrodifferential equation sets for the averaged spatial deviation variables
must be solved for at least two scales. For averaged variables the scales are
the external scale or L domain, and for the spatial deviations it is the volume
of the uid phase considered at the local (pore) scale. This presents a great
challenge and has not yet been resolved by a real mathematical statement.
To close the reaction-diffusion problem Crapiste et al. [41] made a series
of assumptions: (1) the diffusion coefcient D and the rst-order reaction
rate coefcient k
'
are constant; (2) diffusion is linear in the solid part of the
porous medium, (3) the spatial concentration uctuation is linearly depend-
ent on the gradient of the intrinsic averaged concentration and the averaged
concentration itself, (4) the intrinsic averaged concentration and solid
surface averaged concentration are equal, (5) the restriction
k
'
d
D
1
should be satised; and (6) spatial uctuations of the intrinsic concentration
and the surface concentration uctuations are equal. The fourth and sixth
assumptions are equivalent to an equality of surface and intrinsic concen-
trations, which means that the adsorption mechanisms are taken to be
volumetric phenomena.
In our previous efforts we have obtained some results for both morphol-
ogies and demonstrated the strength of morphological closure procedures.
A model of turbulent ow and two-temperature heat transfer in a highly
porous medium was evaluated numerically for a layer of regular packed
particles (Travkin and Catton [16, 20]; Gratton et al. [26, 27]) with heat
exchange from the side surfaces. Nonlinear two-temperature heat and
momentum turbulent transport equations were developed on the basis of
VAT, requiring the evaluation of transport coefcient models. This ap-
proach required that the coefcients in the equations, as well as the form of
the equations themselves, be consistent to accurately model the processes
and morphology of the porous medium. The integral terms in the equations
were dropped or transformed in a rigorous fashion consistent with physical
arguments regarding the porous medium structure, ow and heat transfer
regimes (Travkin and Catton [20]; Travkin et al. [17]). The form of the
Darcy term as well as the quadratic term was shown to depend directly on
the assumed version of the convective and diffusion terms. More impor-
tantly, both diffusion (Brinkman) and drag resistance terms in the nal
forms of the ow equations were proven to be directly connected. These
relations follow naturally from the closure process. The resulting necessity
for transport coefcient models for forced, single phase uid convection led
to their development for nonuniformly and randomly structured highly
porous media.
A regular morphology structure was used to determine the characteristic
morphology functions, (porosity m,, and specic surface S
"
) that were
used in the equations in the form of analytically calculable functions. A rst
approximation for the coefcients, for example, drag resistance or heat
transfer, was obtained from experimentally determined coefcient correla-
tions. Existing models for variable morphology functions such as porosity
and specic surface were used by Travkin and Catton [20] and Gratton et
al. [27] to obtain comparisons with other work in a relatively high Reynolds
number range.
All the coefcient models they used were strictly connected to assumed
(or admitted) porous medium morphology models, meaning that the coef-
cient values are determined in a manner consistent with the selected
geometry. Comparison of modeling results was sometimes difcult because
other models utilized mathematical treatments or models that do not allow
a complete description of the medium morphology; see Travkin and Catton
[16]. Closures were developed for capillary and globular medium morphol-
ogy models (Travkin and Catton [16, 17, 20]; Gratton et al. [26, 27]). It was
shown that the approach taken to close the integral resistance terms in the
momentum equation for a regular structure allows the second-order terms
for the laminar and turbulent regimes to naturally occur. These terms were
taken to be analogous to the Darcy or Forchheimer terms for different ow
velocities. Numerical evaluations of the models show distinct differences in
the overall drag coefcent among the straight capillary and globular models
for both the regular and simple cubic morphologies.
IV. Microscale Heat Transport Description Problems and VAT Approach
Study of energy transport at different scales in a heterogeneous media or
system emphasizes the importance of transport phenomena at subcrystalline
and atomic scales. Among many works addressing subcrystalline transport
phenomena (see Fushinobu et al. [91]; Caceres and Wio [92]; Tzou et al.
[93]; Majumdar [94]; Peterson [95], etc.), the governing energy transport
equations, whether they are of differential type or integrodifferential, are for
homogeneous or homogenized matter. This idealization signicantly reduc-
es the value of the physical description that results. VAT shows great
promise as a tool for development of models for this type of phenomena
because it becomes possible to include the inherent nonlinearity and
heterogeneity found at the subcrystalline level and reect the impact at the
upper levels or scales.
A heuristic approach suggested by Tzou [96] lumps all the atomic and
subcrystalline scale phenomena into the delayed response in time in the
macroscopic formulation. This approach was proposed by author to close
the existing gap in knowledge and to help engineers develop applications.
Unfortunately, the coupling between the characteristics of the subscale
phenomena and delayed response time is lost. There is an ongoing search
for the transport equations describing many-body systems that exhibit
highly nonequilibrium behavior, including non-Markovian diffusion. The
more exact the description of a physical phenomenon provided by a
mathematical model, the more possibilities there are for innovative improve-
ments in the function of a particular material or device. Our contribution to
the effort is an extensive analysis of existing approaches to the development
of theories for the subcrystalline and atomic scale levels. We have also made
progress in the development of VAT-based tools applicable at the atomic
and nanoscale level for description of transport of heat, mass, and charge in
SiC and superconductive ceramics.
At the subcrystalline scale, we will consider energy transport using a VAT
description for effects of crystal defects and impurities on phononphonon
scattering, which has a substantial impact on thermal conductivity. At the
crystal scale, the importance of thermal resistance (different models) due to
various mechanisms lattice unharmonic resistance and crystal boundary
defects will be treated. Including these phenomena shows that they have
a major impact on the transport characteristics in critical applications such
as optical ceramics and superconductive ceramics.
A. T:ni1ioN:i Drscii1ioNs or Micosc:ir Hr:1 T:Nsio1
Kaganov et al. [97] rst developed a theory to describe energy exchange
between electrons and the lattice of a solid for arbitrary temperatures using
earlier advances in this eld by Ginzburg and Shabanskii [98] and by
Akhiezer and Pomeranchuk [99]. In their work, they assumed that the
electron gas was in an equilibrium state. After a brief summary of this early
work, an analysis leading to a method for estimating the relaxation
processes between the electron uid temperature T
and the phonon tem-

perature T
'
will be presented.
The heat balance equation for the electron temperature is
c
(T )
cT
ct
U Q, (97)
where Q is the heat source, c
(T ) is the electronic specic heat,

c
(T )
2
`
k
"
n
k
"
T
c
"
,
and
c
"
(3n
/8)``(2)`/(2m*).
U is the heat exchange term,
U
2`
3
m*c`
`
n
(T
T
'
)
t(T
'
)
(T
T
'
)
T
'
, T
'
T
'
; (T
T
'
) T
'
(98)
U
`
6
m*c`
`
n
t(T
'
)
(T
T
'
)
T
'
, T
'
T
'
; (T
T
'
) T
'
, (99)
where m* is the effective electronic mass, c
`
is the sound velocity, n
is the
number of electrons per unit volume, t(T
'
) is the time to traverse a mean
free path of electrons under the condition that the lattice temperature
coincides with the electron temperature and is equal respectively to T
'
.
When the lattice temperature is assumed to be much less than the
temperature of the electrons (an assumption later found to be weak), then
U
2`
15
m*c`
`
n
t(T
)
, (T
T
'
; T
'
T
);
`
6
m*c`
`
n
t(T
)
, (T
T
'
; T
'
T
).
. (100)
Kaganov et al. [97] used an equation for elastic lattice vibration of the form
cÙ
'
cr`
c`
`
AU
'
U
j
Vo(r Vt). (101)
This also allowed them to develop the heat exchange term (here U
'
is the
displacement vector). In this equation, j M/d` is the density of lattice, M
is the mass of the lattice atom, V is the lattice volume, and U is the
interaction constant of the electron with the lattice that appears in the
expression for the time to travel the mean free path.
It was nearly 20 years before needs in different physical elds (namely,
intense short-timespan energy heating in laser applications) brought atten-
tion to this phenomenon and to use it for further technological advances.
Anisimov et al. [100, 101] introduced a simplied two-uid temperature
model for heat transport in solids,
C
(T
)
cT
ct
,AT
(T
T
'
) f (r, t) (102)
C
'
cT
'
ct
:
(T
T
'
), :
`
6
m*c`
`
n
t(T
)
. (103)
Further development of the idea of a two-eld two-temperature model for
energy transport in metals by Qiu and Tien [102104] used this model.
They modied the energy exchange rate coefcent (heat transfer) model in
a way that uses the coefcient of conductivity K
in the following formula

instead of time between collisions t(T
):
U G
"(n
c
`
k
"
)`
K
. (104)
Tzou et al. [93] used the two-uid model with two equations for the
electronphonon transport in metals based on previous works by Anisimov
et al. [101], Fujimoto et al. [105], Elsayed-Ali [106], and others. The
equation for diffusion in an electron gas is a parabolic heat conduction
equation with an exchange term
C
cT
ct
V (K
VT
) G
(T
T
'
), (105)
with phonon transport (phononelectron interactions) for the metal lattice
(just simplied equation) being described by
C
'
cT
'
ct
G
(T
T
'
), (106)
where K
is the thermal conductivity of the electron gas. Using the

WiedermannFranz law for the electronphonon interaction, Qiu and Tien
[102, 103] show that the coupling factor G
can be approximated by
G
"(n
c
`
k
"
)`
K
, (107)
where c
`
, the speed of sound in solid, is
c
`
k
"
2
(6`n
)`T
'
, (108)
T
'
is the Debye temperature, is Plancks constant, and n
and n
are the
electronic and atomic volumetric number densities. Assuming constant
thermal properties, the two equations can be combined, yielding a one-
temperature equation
c`T
'
cx`

:
C`
'
c`T
'
cx`ct
1
:
'
cT
'
ct

1
C`
'
c`T
'
ct`
, (109)
where the thermal diffusivity of electron gas :
, equivalent thermal diffusiv-

ity :
'
, and thermal wave speed C
'
are dened by
:
, :
'
C
'
, C`
'
G
C
C
'
. (110)
Tzou [96] later proposed a unied two-uid model to derive the general
hyperbolic equation with two relaxation times t
'
and t
,
V`T t
'
c
ct
(V`T )
1
:
cT
ct

t
:
c`T
ct`
, (111)
which he argues is the same equation derived from two-step models in
metals. A more complex two-temperatures model was obtained by Gladkov
[107] using parabolic equations
cT
ct
V
cT
cx
,
c`T
cx`
:
`
(T
T
`
) (112)
and
cT
`
ct
,
`
c`T
`
cx`
:
`
(T
T
`
), (113)
It can be seen from his work that the coefcients of heat transfer :
`
and
:
`
are not equal. After combining the two equations into one, an equation
for a mobile (liquid) medium results:
1
:
`
:
`
cT
ct

1
:
`
c`T
ct`

V
:
`
c`T
ctcx
:
`
V
:
`
cT
cx
,
`
V
:
`
c`T
cx`

,
,
`
:
`
c"T
cx"
,
c`T
cx`
. (114)
There are other works (see, for example, Joseph and Preziosi [108]) treating
the two-uid heat transport and obtaining the same kind of hyperbolic
equation.
1. Equation of Phonon Radiative Transfer
Majumdar [94] suggested an equation for phonon radiative transfer
(EPRT). In three dimensions the equation is
cL
ct
(V
"
VI
)
I"
(T (x)) I
t(c, T )
, (115)
where I
is the directional-spectral phonon intensity, V

"
is the phonon
propagation speed, and I"
(T (x)) is the equilibrium intensity corresponding

to a blackbody intensity at temperatures below the Debye temperature. To
make matters more complex, it should be noted that as stressed by Peterson
[95], However, fundamental differences exist between phonon and photon
behavior in the regime where scattering and collisional processes are
important . . . . Even in perfect crystals, the so-called unklapp processes that
are responsible for nite thermal conductivity do not obey momentum
conservation.
2. Hyperbolic Heat Conduction Equations
The work by Vernotte, Cattaneo, Morse, and Feshbach that led to the
hyperbolic heat conduction equation was primarily heuristic in nature
(without a rst principle physical basis). The nal form is often presented as
a telegraph equation (see Joseph and Preziosi [108]),
c`T
ct`

1
t
cT
ct

k
(t)
V`T, (116)
or
t
c`T
ct`

cT
ct

1
V (KVT ), (117)
for nonconstant thermal conductivity K; here is the heat capacity.
Majumdar et al. [109] produced microphotographs of thermal images
that show the grain structure, visible in the topographical image, and notes
that the grain boundaries appear hotter than within the grain. It is at
present not clear why this occurs . . . . The hot electrons collide with the
lattice and transfer energy by the emission of phonons. The governing
equations for a nonmagnetic medium they use are conservation of electrons,
cn
ct
V (nV
) 0; (118)
conservation of electron momentum,
cV
ct
(V
V)V
e
m*
E
k
"
m*n
V(nT
t
"
; (119)
where the last term is the collision and scattering term analogous to the
Darcy term in porous media ow; conservation of electron energy,
cW
ct
V (W
) e(nV
E) k
"
V (nV
)
V (k
VT
(W
(3/2)k
"
T
)
t

; (120)
conservation of lattice optical phonon energy,
C
cT
ct

(W
(3/2)k
"
T
)
t
(T
)
t

; (121)
and conservation of acoustical energy,
C
cT
ct
V (k
VT
(T
)
t

. (122)
The last four equations have terms, the last term on the right-hand side, that
qualitatively reect the collision and scattering rates in each process. Here
t
"
is the electron momentum relaxation time, t
is the electron optical

relaxation time, t
is the optical acoustical relaxation time, and k

"
is
Boltzmanns constant. In those equations assumed a scalar effective mass for
the electrons m*.
The electric eld is determined using the Gauss law equation written in
terms of electric potential (EVu),
V (cVu) e(N
n p) eN
C
(123)
N
C
(N
n p),
where c is the dielectric constant of Si, N
is the n-doping concentration, N
is the p-doping concentration, and p is the hole number density.

B. VAT-B:srn Tvo-Trxir:1ir CoNsrv:1ioN Eqi:1ioNs
Conservation equations derived using VAT enable one to capture all of
the physics associated with transport of heat at the micro scale with more
rigor than any other method. VAT allows one to avoid the ad hoc
assumptions that are often required to close an equation set. The resulting
equations will have sufcient generality for one to begin to optimize material
design from the nanoscale upward. The theoretical development is briey
outlined in what follows.
The nonlinear paraboic VAT-based heat conduction equation in one of
the phases of the superstructure (where superstructure is to be determined
as the micro- or nanoscale materials organized morphology along with its
local characteristics) is
s
,(jc
cT
ct
V [s
,K
VT
] V [s
,K
VT
]
V
_
K
AD ,
c
`
`
T
ds
1
AD ,
c
`
`
K
cT
cx
'
ds
,S
'
`
}
.
(124)
For constant thermal conductivity, the averaged equation for heat transfer
in the rst phase can be written
s
,(jc
cT
ct
k
V`(s
,T
) k
V
_
1
AD ,
c
`
`
T
ds
AD ,
c
`
`
VT
ds
,S
'
`
}
. (125)
These VAT equations (124) and (125), written for the two phases, will be
seen to yield the same pair of parabolic equations derived by researchers
such as Gladkov [107], but with quite different arguments. Closure to Eq.
(125) is needed for the second and third terms on the right-hand side. The
steps to closure are
1
AD ,
c
`
`
k
cT
cx
'
ds
1
AD ,
c
`
`
k
cT
cn
`
ds n
`
1
AD ,
c
`
`
q
`
ds
`
:`
`
S
`
(T }
`
T }
), (126)
with the heat transfer coefcient, :`
`
(cS
`
), dened in phase 2. This closure
procedure is appropriate for description of uidsolid medium heat ex-
change and might be considered as the analog to solidsolid heat exchange
found in many works. A more precise integration of the heat ux over the
interface surface, cS
`
, yields exact closure for that term in governing
equations for both neighboring phases.
Industry needs to lead one to attempt to estimate, or simulate by
numerical calculation or other methods, the effective transport properties of
heterogenous material. Among the many diverse methods used to do this,
VAT presents itself as an effective tool for evaluating and bringing together
different methods and is useful in providing a basis for comparative
validation of techniques. To demonstrate the value of a VAT-based process,
the effective thermal conductivity will be determined within the VAT
framework. The averagd energy equation in phase 1 of a medium is
k
V`(s
,T
) k
V
_
1
AD ,
c
`
`
T
ds
V [q
].
The right-hand-side (diffusive-like) ux is different from that convention-
ally found,
q
[k
VT
] k
V(s
,T
)
k
AD ,
c
`
`
T
ds
, (127)
where
k
_
k
V(s
,T
)
k
AD ,
c
`
`
T
ds
(VT
). (128)
After these transformations, the heat transfer equation for phase 1 becomes
s
,(jc
cT
ct
V [k
VT
] :`
`
S
`
(T }
`
T }
) s
,S
'
`
}.
(129)
This is the same type of heat transport equation routinely used in two-uid
models. The equation for heat transport in the second phase (if any) would
be the same, and one can easily obtain the hyperbolic type two-uid
temperature model.
A similar VAT-based equation can be obtained for the heat transfer in
phase 1 when the heat conductivity coefcient is a function of the tempera-
ture or other scalar eld (nonlinear) (Eq. (124)), but the effective conductiv-
ity will have an additional term reecting the mean surface temperature over
the interface surface inside of the REV,
K
_
K
V(s
,T
) s
,K
VT
AD ,
c
`
`
T
ds
(VT
).
(130)
Equation (124) simplies to
s
,(jc
cT
ct
V [K
VT
] :`
`
S
`
(T }
`
T }
) s
,S
'
`
}
.
(131)
The third term on the right-hand side of (124) plays a different role when
the interface between two phases is only a mathematical surface without
thickness neglecting the transport within the surface means there is no need
to consider this medium separately. When this is the case, this term can be
equal for the both phases, simplifying the closure problem. The problem
becomes signicantly more complicated when transport within the interface
must be accounted for.
C. Sincxs1:iiiNr SiNcir Cxs1:i Dox:iN W:vr Hr:1
T:Nsio1 Eqi:1ioNs
Some features of energy transport, including electrodynamics, that are
above the scale of close capture quantum phenomena are considered next.
Limiting the scope of the problem allows us to concentrate on the descrip-
tion of heat transport phenomena in the medium above the quantum scale
where there are at least the three substantially different physical and spatial
scales to consider. Within this scope, the heat transport equation in a single
grain (crystalline) can be written in the form
t
c`T
ct`

cT
ct

1
V (KVT
)
1
S
'
E
. (132)
Comparing this equation with the equation developed by Tzou [96] with
two relaxation times, t
'
and t
,
t
c`T
ct`

cT
ct
:V`T t
'
:
c
ct
(V`T ) :S
'
E
, (133)
and the parabolic equation obtained by Gladkov [107] for the model with
two temperatures for constant coefcients,
1
:
`
:
`
cT
ct

1
:
`
c`T
ct`

V
:
`
c`T
ctcx
:
`
V
:
`
cT
cx
,
`
V
:
`
c`T
cx`

,
,
`
:
`
c"T
cx"
,
c`T
cx`
, (134)
one can see that all belong to the family of VAT two-temperature conduc-
tion problems with nonconstant effective coefcients for the charged carriers,
cT
'
ct
a
'
V [K
'
VT
'
] b
'
(T }
'
T }
'
) S
'
A
, (135)
and for phonon temperature transport,
cT
'
ct
a
'
V [K
'
VT
'
] b
'
(T
'
T
'
) S
'
J
. (136)
This pair of equations is the wave transport equations shown in previous
sections.
Our current interest, however, is not to justify past assumptions made to
develop appropriate scale level energy transport equations, but to develop
mathematical models for heat transport and electrodynamics in multiscale
microelectronics superstructures.
D. NoNioc:i Eirc1onxN:xics :Nn Hr:1 T:Nsio1 iN
Siirs1ic1irs
Many microscale heterogeneous heat transport equations and some of the
solutions provided elsewhere (see, for example, [110, 111, 112, 113, 109])
required substantial analysis, and many need improvement. Goodson [113],
for example, directly addresses the need to model nonhomogeneous medium
(diamond CVD layer) thermal transport by accounting for the presence of
grains. The PeierlsBoltzmann equation for phonon transport was used
along with information on grain structure. In the present work, the goal is
to give some insight to situations (and those are substantial in number)
where the medium cannot be considered as homogeneous even at the
microscale level. For these circumstances, the governing eld equations
should be based on conservation equations for a heterogeneous medium, for
example, the VAT governing equations.
The VAT governing equations for heterostructures will be found starting
from a set of governing equations for a solid-state electron plasma uid.
Phase averaging of the electron conservation equation (118) yields
cn
ct
"
V (nV
),
"
0 (137)
where ,
"
means averaging over the major phase of the material. The VAT
nal form for this equation is
cn,
"
ct
V nV
,
"
1
AD ,
c
`
KQ
nV
ds 0, (138)
or
cn,
"
ct
V [s
"
,n` V
m,n` V
}
"
]
1
AD ,
c
`
`
nV
ds
"
0, (139)
where cS
"`
is the interface (real or imaginary) of phases and scatterers.
It will be assumed that only immobile scatterers produce phase separ-
ation. This is not an essential restriction and is only taken to simplify the
appearance of the equations and streamline the development. We recognize
that defects and other scattering objects where processes are also occuring,
such as nonmajor phases, occur, but we are not interested in them at this
time because their volumetric fractions are very small and their importance
is decreased by scattering of the elds in a major phase.
The electron uid momentum transport equation can be written in two
forms, and the form inuences the nal appearance of the VAT equations.
The rst is
cV
ct
"
(V
V)V
,
"
e
m*
E,
"
k
"
m*
1
n
V(nT
"
t
"
.
^ ^
(140)
Using the transformation
1
n
V(nT
"
VT
1
n
Vn
"
VT
V(ln n),
"
VT
V(Z
),
"
, Z
lnn, (141)
it can be written as
cV
ct
"
(V
V)V
,
"
e
m*
E,
"
k
"
m*
VT
V(Z
),
"
t
"
,
^ ^
(142)
where the brackets
]
^
]
^
dene the problem uncertainty in the treatment of
this relaxation term. Strictly speaking, this term should not be in this form
and may not exist.
The same equation written in conservative form is
cnV
ct
"
V (nV
),
"
e
m*
nE,
"
k
"
m*
V(nT
),
"
nV
t
"

,
^ ^
(143)
Using
(V
V)V
,
"
V (nV
),
"
V
(V (nV
)),
"
V nV
,
"

1
AD ,
c
`
KQ
(nV
) ds
"
s
"
,V
V (nV
)}
"
V
(V (nV
))
()
,
"
}, (144)
Eq. (142) can be written in the VAT form as
cV
,
"
ct
V nV
,
"
1
AD ,
c
`
KQ
(nV
) ds
"
s
"
,V
V (nV
)}
"
V
(V (nV
))
()
,
"
}
e
m*
E,
"
k
"
m*
VT
,
"

1
AD ,
c
`
KQ
T
ds
"
k
"
m*
s
"
,T
VZ
}
"
1
AD
"
,
c
`
KQ
Z
ds
"
(V(Z
))
()
,
"
,
(145)
where the last term on the right-hand side of (142), the scattering and
collision reection term, has been replaced by a number of terms, each
reecting interface-specic phenomena, including scattering and collision.
Some manipulation of the convection terms of the conservative form of the
momentum equation has been done to combine the forms of the equations
of mass and momentum.
The second conservative form of the momentum equation is derived in
the form
n,
"
cV
ct
(nV
,
"
V)V
c
ct
n` V
,
"
V nV
,
"
V

1
AD ,
c
`
KQ
nV
ds
1
AD ,
c
`
KQ
(nV
) ds
"
e
m*
nE,
"
k
"
m*
V[s
"
,n`T
s
"
,n` T
}
"
]
1
AD ,
c
`
KQ
nT
ds
"
,
(146)
where a number of the integral terms are scattering and collision terms.
There are other possible forms of the left-hand side of the momentum
equation VAT equations that will not be pursued at this time.
The homogeneous volume averaged electron gas energy equation for a
heterogeneous polycrystal becomes
cW
ct
"
V (W
),
"
enV
E,
"
k
"
V (nV
),
"
V (k
VT
),
"
^
(W
(3/2)k
"
T
)
t

^
, (147)
or
cW
,
"
ct
V W
,
"
1
AD ,
c
`
KQ
(W
) ds
"
enV
E,
"
k
"
V (nV
),
"
k
"
AD ,
c
`
KQ
(nV
) ds
"
V [k
}
"
V(s
"
,T
}
"
)] V [s
,k
VT
}
"
]
V
_
k
}
"
AD ,
c
`
KQ
T
ds
"
1
AD ,
c
`
KQ
k
cT
cx
'
ds
"
. (148)
The integral terms again reect scattering and collision that appear as a
result of the heterogeneous medium transport description.
The equation for longitudinal phonon temperature is
cT
ct
"
cW
ct
'
cW
''
ct
'
, (149)
or
cT
ct
"
k
"
AD ,
c
`
KQ
(nV
) ds
"
1
AD ,
c
`
KQ
(W
) ds
"
V
_
k
}
"
AD ,
c
`
KQ
T
ds
"
1
AD ,
c
`
KQ
k
cT
cx
'
ds
"
V
_
k
}
"
AD ,
c
`
KQ
T
ds
"
1
AD ,
c
`
KQ
k
cT
cx
'
ds
"
. (150)
The equation of acoustical phonon energy is
cT
ct
"
V (k
VT
),
"

^
C
(T
)
t

^
(151)
or
cT
ct
"
V [k
}
"
V(s
"
,T
}
"
)] V [s
,k
VT
}
"
]
V
_
k
}
"
AD ,
c
`
KQ
T
ds
"
1
AD ,
c
`
KQ
k
cT
cx
'
ds
"
.
(152)
Describing phonon scattering and collision is an unsolved problem and as
noted by Peterson [95], The complexity of this aspect of the problem
precludes the relatively simple solution used in simulating rareed gas
ows.
Another kind of single phase equation for momentum transport of
electronic uid results for magnetized materials:
cV
ct
"
(V
V)V
,
"
e
m*
EV
B,
"
k
"
m*
1
n
V(nT
"
^
V
t
"
^
. (153)
The VAT form of this equation is
cV
,
"
ct
V nV
,
"
1
AD ,
c
`
KQ
(nV
) ds
"
s
"
,V
V (nV
)}
"
V
(V (nV
))
()
,
"
}
e
m*
(E,
"
V
B,
"
)
k
"
m*
VT
,
"
1
AD ,
c
`
KQ
T
ds
"
k
"
m*
s
"
,T

Vz
}
"
1
AD
"
,
c
`
KQ
Z
ds
"
(V(Z
))
()
,
"
.
(154)
The Maxwell equations for electromagnetic elds used to develop the
VAT Maxwell equations for electromagnetic elds are
V (c
"
E
"
) j
"
, V (j
"
H
"
) 0 (155)
VE
"
cB
"
ct
(156)
VH
"
j
"
c
ct
(D
"
), (157)
with constitutive relationships
B
"
j
"
H
"
, D
"
c
"
E
"
, j
"
o
"
E
"
. (158)
A full description of the derivation of the VAT nonlocal electrodynamics
governing equations is given by Travkin et al. [114, 115] with only a limited
number shown here.
For the electric eld, the Maxwell equations, after averaging over phase
(m) using ,
"
, become
V [s
"
,c`
"
E
"
] V [s
"
,c`
"
E
"
}
"
]
1
AD ,
c
`
`
(c
"
E
"
) ds
"
j
"
,
"
(159)
V(s
"
,E
"
)
1
AD ,
c
`
KQ
ds
"
E
"

c
ct
j
"
H
"
,
. (160)
The phase averaged magnetic eld equations are
V (s
"
,j`
"
H
"
) V [s
"
,j`
"
H
"
}
"
]
1
AD ,
c
`
KQ
(j
"
H
"
) ds
"
0, (161)
and
V(s
"
,H
"
)
1
AD ,
c
`
KQ
ds
"
H
"
c
ct
c
"
E
"
,
"
[s
"
,o`
"
E
"
s
"
,o`
"
E
"
}
]. (162)
These equations and some of their variations, such as the electric eld
wave equations
VÈ
"
j
"
o
"
cE
"
ct
j
"
c
"
cÈ
"
ct`
V
j
"
c
"
, (163)
which becomes
V`(s
"
,E
"
) V
_
1
AD ,
c
`
KQ
E
"
ds
"
1
AD ,
c
`
KQ
VE
"
ds
"
j
"
o
"
cE,
"
ct
j
"
c
"
cÈ,
"
ct`

1
c
"
V(s
"
,j`
"
)
1
c
"
AD ,
c
`
KQ
j
"
ds
"
,
(164)
are the basis for modeling of electric and magnetic elds at the microscale
level in heterostructures.
The primary advantage of the VAT-based heterogenous media elec-
trodynamics equations is the inclusion of terms reecting phenomena on the
interface surface cS
"`
that can be used to precisely incorporate multiple
morphological effects occuring at interfaces.
E. Pno1oNic Cxs1:is B:Nn-G:i Ponirx:
CoNvrN1ioN:i DMM-DNM :Nn VAT Tr:1xrN1
One of the possible applications of VAT electrodynamics is the formula-
tion of models describing electromagnetic waves in a dielectric medium of
materials considered to be photonic crystals [118, 116, 117, 119, 126, 120].
The problem of photonic band-gap in composite materials has received
great attention since 1987 [118, 116] because of its exciting promises. The
most interesting applications appear in the purposeful design of materials
exhibiting selective, at least in some wave bands, propagation of electromag-
netic energy [120].
Figotin and Kuchment [122] were the rst who theoretically demon-
strated the existence of band-gaps in certain morphologies. Unfortunately,
this problem as presently formulated is based on the homogeneous Maxwell
equations. The most common way to treat such problems has been to seek
a solution by doing numerical experiments over more or less the exact
morphology of interest, a method called detailed micromodeling (DMM),
which is often done using direct numerical modeling (DNM) (for example,
see [124]). As a result, questions arise about differences between DMM-
DNM and heterogeneous media modeling (HMM), which is the modeling
of an averaged medium to determine its properties. How the averaging for
HMM is accomplished is often not clear or not done at all.
So, why cannot DMM be self-sufcient in the description of heterogen-
eous medium transport phenomena? The answers can be primarily under-
stood by analyzing, among others [23], the following issues:
1. A basic principal mismatch occurs at the boundaries, causing bound-
ary condition problems. This means that for DMM and for the bulk
(averaged characteristics) material elds, the boundary conditions are prin-
cipally different.
2. The DMM solution must be matched to a corresponding HMM to
make it meaningful at the upper scales. This can only be done for regular
morphologies. Discrete continuum gap closure or mismatching will occur
with DMM-DNM, precluding generalization to the next or higher levels in
the hierarchy.
3. The spatial scaling of heterogeneous problems with the chosen REV
(for DMM) is needed to address large or small deviations in elements
considered that are governed by different underlying physics. When spatial
heterogeneities of the characteristics or morphology are evolving along the
coordinates, DMM cannot be used.
4. Numerical experiments provided by DMM-DNM need to be trans-
lated to a form that implies that the overall spatially averaged bulk
characteristics model random morphologies. It is not clear what kind of
equations are to be used as the governing equations, nor what variables
should be compared. In the case of the local porosity theory [128, 129], for
example, the results of using real porous medium digitized images for
morphological analysis to calculate the effective dielectric constant assumes
that the HMM equations are applicable.
5. Interpretation of the results of DMM-DNM is always a problem. If
results are presented for a heterogeneous continuum, then the previous
point applies. If the results are being used as a solution for some discrete
problem, then the question is how to relate that solution to the continuum
problem of interest or even to a slightly different problem. If the results
obtained are t into a statistical model, then the phenomena are being
subjected to a statistical averaging procedure that is in most cases only
correct for independent events.
6. The most sought-after characteristics in heterogeneous media trans-
port studies are the effective transport coefcients that can only be correctly
evaluated from
j ,o*Vu,o
`
Vu,(o
o
`
)
1
AD ,
`'
`
Vudc,
using the DMM-DNM exact solutions for a small fraction of the problems
of interest. The issue is that problems of interest having inhomogeneous,
nonlinear coefcients and, in many transient problems, two-eld DMM-
DNM exact solutions are not enough to nd the effective coefcients.
Fractal methods are sometimes used to describe multiscale phenomena.
The use of fractals is not relevant to most of the morphologies of interest
and the fractal phenomenon description is generally too morphological,
lacking many of the needed physical features. For example, descriptions of
both phases, of the phase interchange, etc., are need to represent the physical
phenomenon.
For the simplest case of a superlattice or multilayer medium there can be
many difculties. When the boundaries are not evenly located, crossing the
regular boundary cells of the medium, then the problem must be solved
again and again. If the coefcients are space dependent, because of the layers
or grain boundaries, they will inuence scattering. Grain boundaries are not
perfect and are not just mathematical surfaces without thickness or physical
properties. They cannot be treated as mathematical surfaces without any
properties. Imperfections in the internal spacial structures must be treated
as domain morphologies are not perfect at any spacial level.
The insufciency of a homogeneous wave propagation description of a
heterogeneous medium was addressed in [125] from a pure mathematical
point of view by searching for another type of governing operator that could
better explain the behavior of the frequency spectrum eigenmodes via
heuristic arguments. The general band-gap formulation should be treated
using the HMM statements developed from the analysis of the VAT
equations. A straightforward description of one of the band-gap problems
is given next.
Representing electromagnetic eld components with time-harmonic com-
ponents,
E(x, y, z)e
ict
, H(x, y, z)e
ict
, i (1) , (165)
The equations describing a dielectric medium becomes
V (cE) 0, V H0, j 1 (166)
VEicH, VHiccE, (167)
where c is the complex dielectric constant dened by c c i(o/c), and
o0(x), c c(x), c c(x, c).
Taking the curl of the both sides of the vector equations,
V
1
c
VH
V(icE), V(VE) V(icH), (168)

yields
V
1
c
VH
ic(icH) c`H (169)

V(VE) ic(iccE) ccÈ. (170)
This is the set of equations usually used when problems of photonic
band-gap materials are under investigation; see the study by Figotin and
Kuchment [123], p. 1564. These equations can be transformed to
AE
`
(x) c`c(x)E
`
(x) (171)
for E-polarized elds and
V
1
c(x)
VH
`
(x)
c`H
`
(x) (172)
for H-polarized elds.
The further treatment by Figotin and Kuchment [123] reduces the
mathematics to two equations,
V
1
c(x)
Vf
(x)
z f
(x) (173)
1
c(x)
Af
(x) z f
(x), (174)
where f
is the H
`
or E
`
polarization determined components of electric or
magnetic elds.
These equations state the eigenvalue problem characterizing the spectrum
of electromagnetic wave propagation in a dielectric two-phase medium,
which is supposed to describe the photonic materials band-gap problem of
EM propagation (see equations on p. 1568 in Figotin and Kuchment [123])
There are no spatial morphological terms or functions involved in the
description, just the permittivity, which is supposed to be a space-dependent
function with changes at the interface boundary.
When these equations are phase averaged to represent the macroscale
characteristics of wave propagation in a two-phase dielectric medium, the
equations become
V (`
V(m
, f
)) V
_
`
1
AD ,
c
`
`
f
ds
V (` V f
)
1
AD ,
c
`
`
Vf
ds
zm
, f
(175)
(x)
1
c
V`(m
, f
) V
_
1
AD ,
c
`
`
f
ds
1
AD ,
c
`
`
V f
ds
z[m
,c`
,c`
]. (176)
The three additional terms appear along with the porosity (or volume
fraction) function m
, as a factor on the right-hand side of each of the

equations. When the dielectric permittivity function is homogeneous in each
of the two phases, then the VAT photonic band-gap equations can be
reduced to one equation in each phase and written in a simpler form,
V`(m
, f
) V
_
1
AD ,
c
`
`
f
ds
1
AD ,
c
`
`
Vf
ds
zc
, f
(177)
and
V`(m
`
, f
) V
_
1
AD ,
c
`
`
f
ds
`
1
AD ,
c
`
`
V f
ds
`
zc
`
m
`
,f
.
(178)
The equations are almost the same as equations for heat or charge
conductance in a heterogeneous medium. The similarity of the equations
means that the analysis of the simplest band-gap problem should also be
very similar.
Using DMM-DNM, Pereverzev and Umtsev [121] found that exact
micromodel solutions among others features can have medium. . . internal
generation that might be well characterized by the impact of the additional
terms in the VAT Maxwell equations in both phases and in the combined
electric eld and effective coefcient equations; see Sections V and VIII. The
exact closure and direct numerical modeling derived by Travkin and
Kushch [33, 34] demonstrated how important and inuential the additional
VAT morphoterms can be (Section I). These terms do not explicitly appear
in either the microscale basic mathematical statements or in microscale eld
solutions. The terms appear and become very important when averaged
bulk characteristics are being modeled and calculated.
V. Radiative Heat Transport in Porous and Heterogeneous Media
Radiation transport problems in porous (and heterogeneous) media,
including work by Tien [130], Siegel and Howell [131], Hendricks and
Howell [132], Kumar et al. [133], Singh and Kaviany [134], Tien and
Drolen [135], and Lee et al. [139], are primarily based on governing
equations resulting from the assumption of a homogeneous medium. This
implicitly implies that specic problem features due to heterogeneities can
be decribed using different methods for evaluation of the interim transport
coefcients, as, for example, done by Al-Nimr and Arpaci [136], Kumar and
Tien [137], Lee [138], Lee et al. [139], and Dombrovsky [140]. Although
this kind of approach is legitimate, it presents no fundamental understand-
ing of the processes because the governing equations suffer from the initial
assumption that strictly describes only homogeneous media. Further, it is
difcult to represent hierarchical physical systems behavior with such
models as will be touched on later.
Review papers like that of Reiss [141] describe the progress in the eld
of dispersed media radiative transfer. The few works on heterogeneous
radiative or electromagnetic transport (see Dombrovsky [140], Adzerikho
et al. [142], van de Hulst [143], Bohren and Huffman [144], Lorrain and
Corson [145], Lindell et al. [146], and Lakhtakia et al. [147]) approach the
study of transport in disperse media with the emphasis on known scattering
techniques and their improvements.
The area of neutron transport and radiative transport in heterogeneous
medium being developed by Pomraning [148151] and Malvagi and
Pomraning [152] treats linear transport in a two-phase (two materials)
medium with stochastic coefcients. This approach is the same as that which
has been used to treat thermal and electrical conductivity in heterogeneous
media, and to this point it has not been brought to a high enough level to
include variable properties, their nonlinearities, and cross-eld (electrical
and thermal or magnetic) phenomena.
Research by Lee et al. [139] on attenuation of electromagnetic and
radiation elds in brous media has shown a high extinction rate for
infrared radiation. The problem is treated as a scattering problem for a
single two-layer cylinder by Farone and Querfeld [153], Samaddar [154],
and Bohren and Huffman [144]. The process of radiative heat transport in
porous media is very similar to propagation of electromagnetic waves in
porous media and will also be evaluated. These two very close elds seem
not to have been considered as a coherent area. Complicated problems of
propagation of electromagnetic waves through the ber gratings have been
primarily the subject of electrodynamics. The most notable work in this area
is that of Pereverzev and Umtsev [121], Figotin and Kuchment [122, 125],
Figotin and Godin [124], Botten et al. [155], and McPhedran et al. [156,
157]. No effort seems to have been made to translate results obtained for
polarized electromagnetic radiation to the area of heat radiative transfer.
Detailed micromodeling (DMM) of electromagnetic wave scattering has
been based on single particles or specic arrangements of particulate media.
Direct numerical modeling (DNM) of the problem seems enables one to do
a full analysis of the elds involved. As already discussed, the analysis of the
results of a DNM is limited in the performance of a scaling analysis, which
is the goal in most situations. Performing DNM without a proper scaling
theory is like performing experiments, often very challenging and expensive;
without proper data analysis, it yields a certain amount of detailed eld
results, but not the needed bulk or mean media physical characteristics.
Most recent work on radiative transport is based on linearized radiative
transfer equations for porous media. We rst review this work to set the
stage for the development that follows. This radiative transport related work
extends our results in the theoretical advancement of uid mechanics, heat
transport, and electrodynamics in heterogeneous media (Travkin et al. [19];
Catton and Travkin [28, 158]; Travkin and Catton [20, 159163]; Travkin
et al. [114, 115]) and provides a means for formulation of radiative
transport problem in porous media using the heterogeneous VAT approach
and electrodynamics language. Based on our previous work, a theoretical
description of radiative transport in porous media is developed along with
the Maxwell equations for a heterogeneous medium.
1. L inear Radiative Transfer Equations in Porous Media
The equation for radiative transport in a homogeneous medium can be
written in the general form
1
c
cI
`
ct
V (DI
`
) [
(r)
`
(r)]I
`
(r)I
`'
(T )
1
4

`
(r)
,
""
p(D' D)I
`
(r, D')dD' (179)
I
`
I
`
(r, D, t),
with
(r) the absorption and

`
(r) scattering coefcients, and for steady
state, using the identity
V (DI
`
) D VI
`
,
in the form
D VI
`
[
(r)
`
(r)]I
`
(r)I
`'
(T )
1
4

`
(r)
,
""
p(D' D)I
`
(r, D') dD'.
(180)
In terms of a spectral source function S
`
(s), the equation can be written in
a particularly simple form,
1
[
`
D VI
`
I
`
S
`
(s), (181)
where the extinction coefcient (total cross sectionPomraning [150, 151]
is
[
`
(r)
`
(r).
Linear particle (neutron, for example) transport in heterogeneous medium
is assumed by Malvagi and Pomraning [152] and Pomraning [151] to be
decribed by
D V[(r) S(r, D)
1
4 ,
""
`
(r, D' D)(r, D') dD', (182)
where the quantities [(r),
`
(r), and S(r, D) are taken to be two-states
discrete random variables. By assuming this, one needs to treat the porous
(heterogeneous) medium as a binary medium that has two magnitudes for
each of the random variables, and a particle encounters alternating segments
of medium with those magnitudes while traversing the medium. When [,
`
,
and S are assumed to be random variables, Eq. (182) is treated as an
ensemble-averaged equation (see Malvagi and Pomraning [152] and Pom-
raning and Su [164])
D V( p
'
'
})p
'
[
'
}
'
}p
'
S
'
}
`'
}
4
p
'
'
p
'
`
'
z
'
p
'
`
'
z
'
, i1, 2, ji
(183)
,
""
(r, D')dD',
where
'
} is the conditional ensemble averaged function at some point r
that is in phase i, and `
'
and `
'
are the interface ensemble-averaged uxes.
The solution to this equation is also supposed to be ensemble-averaged. The
overall averaging over the both phases is given by
(r, D), p
} p
`
`
}, (184)
where p
and p
`
are the probabilities of point r being in medium i 1 or 2,
and
'
} is the conditional ensemble averaged value of , when r is in
medium i.
Ensemble averaging in this representation is obtained by averaging of
medium features, including coefcients, along a straight line the D direc-
tionor by nonlocal 1D line averaging in terms of the physical elds
considered. Most of this kind of work is related to the Markovian statistics
by alternating along the line of two phases of the medium (Pomraning [148,
151]).
The ensemble averaging procedure suggested in (183) signies that the
two last terms in the averaged equation reect the nite correlation length
(interconnection) in a single nonlinear term [(r). This kind of averaging
results in very simple closure statements derived using hierarchical volume
averaging theory procedures, as shown later. A major problem in using
ensemble averaging techniques is that the processes and phenomena going
at each separate site within separate elements of the heterogeneous medium
cannot be resolved completely with the purely statistical approach of
ensemble averaging.
To make an ensemble averaging method workable, researchers always
need to formulate the nal problem or solution in terms of spacially specic
statements or in terms of the original spatial volume averaging theory
(VAT). Examples of this are numerous; see the review by Buyevich and
Theofanous [165].
2. Nonlocal Volume Averaged Radiative Transfer Equations
The basis for the development in this eld will be the volume averaging
theory. We will present some aspects of VAT that are now becoming well
understood and have seen substantial progress in thermal physics and in
uid mechanics. The need for a method that enables one to develop general,
physically based models of a group of physical objects (for example,
molecules, atoms, crystals, phases) that can be substantiated by data
(statistical or analytical) is clear. In modern physics it is usually accom-
plished using statistical data and theoretical methods. One of the major
drawbacks of this widely used approach is that it does not give a researcher
the capability to relate the spatial and morphological parameters of a group
of objects to the phenomena of interest when it is described at the upper
level of the hierarchy. Often the equations obtained by these methods differ
from one another even when describing the same physical phenomena.
The drawbacks of existing methods do not arise when the VAT math-
ematical approach is used. At the present time, there is an extensive
literature and many books on linear, homogeneous, and layered system
electromagnetic and acoustic wave propagation (Adzerikho et al. [142];
Bohren and Huffman [144]; Dombrovsky [140]; Lindell et al. [146];
Lakhtakia et al. [147]; Lorrain and Corson [145]; Siegel and Howell [131];
van de Hulst [143]). It is surprising that these phenomena are often
described by almost identical mathematical statements and governing equa-
tions for both heterogeneous and homogeneous media.
Major developments in the use of VAT, showing the potential for
application to eletrophysical and acoustics phenomena in heterogeneous
media, are found in Travkin and Catton [21], Travkin et al. [159, 114, 115],
and with experimental applications to ferromagnetism in Ryvkina et al.
[160, 162] and Ponomarenko et al. [161].
It has been demonstrated during the past 20 years of VAT-based
modeling in the thermal physics and uid mechanics area (see Slattery [6];
Whitaker [10]; Kaviany [7]; Gray et al. [8]) that the potential of the
approach is enormous. Substantial success has also been achieved in
analyzing the more narrow phenomena of electromagnetic wave propaga-
tion in porous media.
We consider here radiative transfer in porous media using a hierarchical
approach to describe physical phenomena in a heterogeneous medium. The
physical features of lowest scale of the medium are considered and their
averaged characteristics are obtained using special mathematical instru-
ments for describing hierarchical processes, namely VAT. At the next higher
level of the hierarchy, physical phenomena have the physical medium
pointwise characteristics resulting from averaged lower scale characteristics.
The same kind of operators and averaging theorems used in preceding
sections are applied to the following development, involving the rot oper-
ator, in which averaging will result because of the following averaging
theorems:
Vf,
Vf ,
1
AD ,
c
`
`
ds
f (185)
Vf }
Vf }
1
AD ,
c
`
`
ds
f. (186)
Rigorous application to linear and nonlinear electrodynamics and electro-
static problems is described in Travkin et al. [114, 115].
The phase averaging the equation for linear local thermal equilibrium
radiative transfer,
D VI
`
[
`
(r)I
`
(r)I
`'
(T )
1
4

`
(r)
,
""
p(D' D)I
`
(r, D') dD', (187)
in phase 1 yields the VAT radiative equation (VARE)
D
_
V(I
`
,
1
AD ,
c
`
`
I
`
ds
[
`
I
`
,
(r)I
`'
(T ),
`
4

[
`
I
`
,
, i 1 (188)
,
""
p(D' D)I
`
(r, D') dD',
when it is assumed that
`'
is a constant, as done by Malvagi and
Pomraning [152], Pomraning and Su [164], and others.
The additional terms appearing in the VARE in some instances are
similar, but in others they have a different interpretation in the ensemble
averaged equation (183). For example, the term
[
`
I
`
,
(189)
in (188) is the result of uctuations correlation inside of medium 1 in the
REV, but it is described by
p
'
`
'
z
'
p
'
`
'
z
'
(190)
in Malvagi and Pomraning [152], as it is an exchange of energy term
between the two phases across the interface surface area cS
`
. Because
ensemble averaging methodologies in Malvagi and Pomraning [152] do not
treat nonlinear terms very well and incorrectly average differential operators
such as V, terms do not appear in equation (183) that reect the interface
ux exchange. In VARE, Eq. (188), the interface exchange term naturally
appears as a result of averaging the V operator,
D
1
AD ,
c
`
`
I
`
ds
. (191)
When the coefcients in the radiative transfer equation are dependent
functions, more linearized terms are observed in the corresponding VARE,
D VI
`
,
[
`
I
`
,
I
`'
(T ),
I
`'
,
1
4
(`
`
`
`
`
`
)
D
1
AD ,
c
`
`
I
`
ds
[
`
I
`
,
, i 1, (192)
while continuing to treat the emissivity as via the Plancks function. This
equation should be accompanied by the VAT heat transfer equations in
both porous medium phases (see, for example, Travkin and Catton [21]).
The heat transport within solid phase 2, combining conductive and
possible radiative transfer, is described by
s
`
,(jc
)
`
cT
`
ct
k
`
V`(s
`
,T
`
) k
`
V
_
1
AD ,
c
`
`
T
`
ds
`
k
`
AD ,
c
`
`
VT
`
ds
`
V q',
`
1
AD ,
c
`
`
q' ds
`
. (193)
The third and fth terms on the r.h.s. model the heat exchange rate between
the phases. In an optically thick medium, for example, the radiation ux
term written in terms of the total blackbody radiation intensity is
V q',
`
V
4
3[
V(I
'
)
`
V
4
3[
V
nòT "

`
, (194)
where [ is the total extinction coefcient. An energy equation similar to Eq.
(193) needs to be written for the uid-lled volume, phase 1 of the porous
medium. The radiation ux term would be much more complex because of
the spectral characteristics of radiation in a uid.
Closure is needed for the second, third, and fth terms in Eq. (193) on the
r.h.s. For convective heat exchange, the last term can be written
k
`
AD ,
c
`
`
cT
cx
'
ds
`
:`
`
S
`
(T }
T }
`
) (195)
by noting that
1
AD ,
c
`
`
k
`
cT
cx
'
ds
`
1
AD ,
c
`
`
k
`
cT
cn
ds n
1
AD ,
c
`
`
q
ds
:`
`
S
`
(T }
T }
`
). (196)
This type of closure procedure is appropriate for description of uidsolid
media heat exchange and has been considered by many as an analog for
solidsolid heat exchange. A more strict and precise integration of the heat
ux over the interfce surface, using the IVth kind of boundary conditions,
gives the exact closure for the term in the governing equations for the
neighboring phase. This would be an adequate solution for the portion of
heat exchange by conduction to and from the uid phase, a conjugate
problem.
The radiative energy exchange across the interface surface is difcult to
formulate because of its spectral characteristics and the boundary conditions
that must be satised. When the uid phase is assumed to be optically thin,
an approximate closure expression results,
1
AD ,
c
`
`
q' ds
`
1
AD ,
c
`
`
o(T "
`
T "
`
)
1
c
`
1
c
`
1
ds
`
5
o((T `
`
)" (T `
`
)")S
`
1
c
`
1
c
`
1

, (197)
using an interpretation of the averaged surface temperatures on opposite
sides of the interface developed by Malvagi and Pomraning [152]. Another
approximation is justiable for an optically thick uid phase. It uses the
specic blackbody surface radiation intensity I
'`
nòT "
`
to close the
integral energy exchange term as follows:
1
AD ,
c
`
`
q' ds
`
5
1
AD ,
c
`
`
(nòT "
`
)ds
`
5c'
`
o(T `
`
)"S
`
. (198)
Here, c'
`
is the total radiative hemispherical emissivity from phase 2 to
phase 1 in the REV.
The closure of Eqs. (183) is accomplished by assuming equality (Malvagi
and Pomraning [152]; Pomraning and Su [164]) between the interface
surface and ensemble (1D in this case) averaged functions,
`
'

'
}, (199)
as was done in heat and mass transfer porous medium problems; see, for
example, Crapiste et al. [41].
3. Radiation Transport in Heterogeneous Media Using Harmonic
Field Equations
Representing the electromagnetic eld components with time-harmonic
components results in
V (c
E) j, V (j
"
H) 0 (200)
VEicj
"
H, VHicc
E. (201)
Here, as outlined earlier, c
is the complex dielectric function c
i(o
/c), and c
(x), o
(x), j
"
j
"
(x, c), c
(x, c). In many

contemporary applications the spatial dependency of these functions is
neglected. Electrophysical coefcients often need to be treated as nonlinear.
For example, the dielectric function can depend on E and c
(x, E). The

wave formulation of the Maxwell equations with constant phase coefcients
for the magnetic eld is
V`Hj
"
o
cH
ct
j
"
c
c`H
ct`
0, (202)
whereas the electric eld wave equation is almost the same,
VÈj
"
o
cE
ct
j
"
c
cÈ
ct`
V
j
c
. (203)
Another form of the equation for E appears in Cartesian coordinates
when electromagnetic elds are time-harmonic functions:
VÈkÈ0, (204)
Here, the inhomogeneous function k` c`j
"
c
is the wave number

squared. This equation is often applicable to linear acoustics phenomena.
This category of equations can be transformed to a form legitimate for
application to heterogeneous media problems.
The time-harmonic forms of equations for rot of electromagnetic elds are
V(m
,E
)
1
AD ,
c
`
`
ds
ic[m
,j`
"
H
,j`
"
H
]
(205)
V(m
,H
)
1
AD ,
c
`
`
ds
ic[m
,c
`
,c
`
].
(206)
The magnetic eld wave form equation with constant coefcients, when
averaged over phase 1, transforms to
V`(m
,H
) V
_
1
AD ,
c
`
`
H
ds
1
AD ,
c
`
`
VH
ds
j
"
o
cH,
ct
j
"
c
c`H,
ct`
, (207)
and the electric eld wave equation (203) becomes
V`(m
,E
) V
_
1
AD ,
c
`
`
E
ds
1
AD ,
c
`
`
VE ds
j
"
o
cE,
ct
j
"
c
cÈ,
ct`

1
c
V(m
,j`
)
1
c
AD ,
c
`
`
j
ds
.
(208)
An analogous form of the averaged equation is obtained for the time-
harmonic electrical eld:
V`(m
,E
) V
_
1
AD ,
c
`
`
E
ds
1
AD ,
c
`
`
VE
ds
,kÈ
0.
(209)
It is the naturally appearing feature of the heterogeneous medium elec-
trodynamics equations as the terms reecting phenomena on the interface
surface cS
`
, and that fact is to be used to incorporate morphologically
precise polarization phenomena as well as tunneling into heterogeneous
electrodynamics, as is being done in uid mechanics and heat transport
(Travkin and Catton [21]; Catton and Travkin [28]).
Using the orthogonal locally calculated directional elds E
'
and E
'
of
averaged electrical eld E
, one can seek the Stokes parameters I, Q, U,

and V,
I E
'
E *
'
,
'
E
'
E *
'
,
'
(210)
QE
'
E *
'
,
'
E
'
E *
'
,
'
(211)
URe[2E
'
E *
'
,
'
] (212)
V Im[2E
'
E *
'
,
'
], (213)
which characterize the intensity of polarized radiation in a porous medium.
We will not here construct the general forms of equations for effective
coefcients, as this will be done in a succeeding section for the case of
temperature elds; still, the same questions of multiple versions, applicability
of current methods, and variance in interpretation are the present agenda.
VAT-based models were developed recently while addressing the prob-
lems of modeling of electrodynamic properties of a liquid-impregnated
porous ferrite medium (Ponomarenko et al. [161]), coupled electrostatic-
diffusion processes in composites (Travkin et al. [159]), and to analyze heat
conductivity experimental data in high-T
superconductors (Travkin and

Catton [166]). Powders of ferrites with NFMR frequency in the microwave
range were used as the porous magnetic medium in Ponomarenko et al.
[161]. The search for tunable levels of reection and absorption of elec-
tromagnetic waves was conducted using a few morphologies that were
arbitrarily chosen. Thus, the need for closer consideration of experiment and
models presenting the data using VAT heterogeneous description tools for
both became obvious.
VI. Flow Resistance Experiments and VAT-Based
Data Reduction in Porous Media
It is well known that existing measurements of transport coefcients in
porous (and heterogeneous) media must be used with care. As long as a
complete description of an experiment is provided and the data analysis is
carried out using correct mathematical formulations (models), the relation-
ship between the experiment and its analysis is maintained in a comsistent,
general, and useful way. Unfortunately, this is not always the case, because
heuristic equations and models are often the basis for coefcient matching
and model tuning when heterogeneous medium experimental data is re-
duced to correlations.
The various approaches, and even disarray, in the eld can be contributed
to a lack of understanding of the general theoretical basis for transport
phenomena in porous and heterogeneous media. As long as the correlations
used for momentum transport comparison are generated from empirical
Darcy and ReynoldsForchheimer expressions, or effective heat and electri-
cal conductivity and permittivity derived from homogeneous models, prob-
lems in heterogeneous media experimental validation and comparison will
persist.
Modeling based on volume averaging theory will be shown to provide a
basis for consistency to experimental procedures and to data reduction
processes by a series of analyses and examples. Many of the common
correlations, and their weaknesses, are examined using a unied scaling
procedure that allows them to be compared to one another. For example,
momentum resistance and internal heat transfer dependencies are analyzed
and compared. VAT-based analysis is shown to reveal the inuence of
morphological characteristics of the medium; to suggest scaling parameters
that allow a wide variety of different porous medium morphologies to be
normalized, often eliminating the need for further experimental efforts; and
to clarify the relationships between differing experimental congurations.
The origin, and insufciency, of electrical conductivity and momentum
transport cross-correlation approaches based on analogies using math-
ematical models without examining the physical foundation of the phenom-
ena will be described and explained.
1. Experimental Assessment of Flow Resistance in Porous Medium
A one-term ow resistance model for porous medium experimental data
analysis often used is
dp
`
dx
f
S
"
m,
j
u
`
`
2
, (214)
where f is some coefcient of hydraulic resistance. On the other hand, most
two-term models used for ow resistance experimental data reduction have
rst-order and second-order velocity terms, the DarcyForchheimer ow
resistance models. These models were obtained primarily for direct compari-
son with established empirical and semiempirical Darcy and DarcyFor-
chheimer type ow resistance data. Thus, the momentum equation for
laminar as well as the high (turbulent) ow regime often used is the model
by Ergun [167],
dp
`
dx
j
k
'
m, u
`
j
Am,ù
`
`. (215)
Similarly, the model given by Vafai and Kim [168] for the middle part of
a porous layer is
dp
`
dx
j
k
'
m,u
`
j
m,`
F
k`
'
u
`
`, (216)
and the Poulikakos and Renken [169] equation for the turbulent regime is
dp
`
dx
j
k
'
u
`
j
Au
`
`. (217)
Analysis of a simple idealized morphology where solutions are known will
show that the Darcy and DarcyForchheimer or Ergun type model corre-
lations are not matched consistently for any regime. Further, they are also
without theoretical foundation. Thus, problems arise when studies to
improve the description of transport use combined models for ow resis-
tance and momentum transport in a porous medium because the analysis
does not start with the correct theoretical basis. Further, which of the three
equations just listed should one use?
A model of ideal parallel tube morphology yields the following Darcy
friction coefcient (see, for example, Schlichting [170]):
f
'
8t
"
(j
`)
, t
"
d
"
Ap
4L
, u`
*
t
"
j
f
'
U
`
8
(218)
Ap
L

dp
`
dx
4
d
"
j
u`
*
f
'
d
"
j
`
2
. (219)
The morphology function S
"
/m, for a straight equal-diameter tube mor-
phology is
S
"
cS
"
AD

2R
pAy
, m,
R`
pAy
,
S
"
m,
4
d
"
(220)
and an exact expression for the Darcy friction factor is
Ap
L
f
'
j
d
"
U
`
2
, f
'
2d
"
j
`
Ap
L
. (221)
The Fanning friction factor for this specic morphology is (using (220))
Ap
L

f
'
4
4
d
"
`
2

f
'
4
S
"
m,
j
`
2
(222)
f
d
"
2j
`
Ap
L
, (223)
and a relationship to the Darcy friction coefcient is (Travkin and Catton
[16, 20])
f
f
'
4
. (224)
The friction coefcient c
for smooth tubes often calculated using the

Nikuradze and Blasius formulas [170] is the same as the Fanning friction
factor.
A model representing a porous medium with slit morphology was treated
in conformity with the denition
t
"
Ap
L
h c'
`
2
, c'
2t
"
j
2u`
*
U
2h
j
`
Ap
L
, u`
*
h
j
Ap
L
.
(225)
The morphology ratio S
"
/m, for a porous medium morphology model of
straight equal slits is found as follows:
S
"
(2L Ay)
( pL Ay)

2
p
, m,
(HL Ay)
( pL Ay)

H
p
(226)
S
"
m,
2
H
1
h
, d
"
4h, (227)
yielding the Fanning friction factor,
dp
`
dx
Ap
L
f

1
h
j
`
2
f

S
"
m,
j
`
2
(228)
f
H
j
`
Ap
L
(229)
As one can easily see, these ow resistance models are written with the
second power of bulk velocity variable. The convergency of the VAT-based
ow resistance transport models to these classical constructions was dem-
onstrated on several occasions by Travkin and Catton [16, 20, 21, 23] and
Travkin et al. [25].
Exact ow resistance results obtained on the basis of VAT governing
equations by Travkin and Catton [16, 26, 23] for the random pore diameter
distribution for almost the same morphology as was used by Achdou and
Avellaneda [171] demonstrated the wide departure from the Darcy-law-
based treatments. That was shown even for the morphology where a single
pore exists with diameter different from the all others. Meanwhile, by
consistently using the VAT-based procedures (Travkin and Catton [23]),
one can easily develop the needed variable, nonlinear permeability coef-
cient for Darcy dependency,
k
'
_
c

S
"
m,`
U
2v
, (230)
where c
is derived for this particular morphology using exact analyti-

cal (in the laminar regime) or well-established correlations for the Fanning
friction factor in tubes.
2. Momentum Resistance in 1D Membrane and Porous Layer Transport
The steady-state VAT-based governing equations for laminar transport in
porous media (Travkin and Catton [21]) are
m,U
cU
cx

c
cx
u` u` ,
1
j
c
cx
(m, p}
1
j
AD ,
c
`
U
pds v
c
cx
cm,U
cx
v
AD ,
c
`
U
cU
cx
'
ds (231)
and
c
m,U
cT
cx
k
c
cx _
cm,T
cx
c
c
cx
(m,T
u` }
c
cx _
k
AD ,
c
`
U
T
ds
1
AD ,
c
`
U
k
cT
cx
'
ds (232)
c
cx
cs,T
`
}
`
cx
c
cx _
1
AD ,
c
`
U
T
`
ds
1
AD ,
c
`
U
cT
`
cx
'
ds
0. (233)
The momentum equation for turbulent ow of an incompressible uid in
porous media based on K-theory can be written in the form (Gratton et al.
[26], Travkin and Catton [20])
m,
cU
ct
U
cU
cx
1
AD ,
c
`
U
(K
"
v)
cU
cx
'
ds
c
cx _
m,(K
"
v)
cU
cx
c
cx _
m,
K
"
cu
`
cx
c
cx
(m,u
`
u
`
}
_
1
j
AD ,
c
`
U
pds
1
j
c
cx
(m,p
`
). (234)
By comparing these equations with conventional mathematical models and
experimental correlations, one can easily see the differences.
The one-dimensional momentum equation for a homogeneous, regular
porous medium is
c
cx
p}
1
m,AD ,
c
`
U
p`ds
j
m,AD ,
c
`
U
VV ds. (235)
Closure of the ow resistance terms in the simplied VAT equation can be
obtained following procedures developed by Travkin and Catton [16, 17].
The skin friction term is
j
AD ,
c
`
U
'
cU
cx
'
ds
j
AD ,
c
`
U
t
"'
ds
1
2
c
'
(x)S
"'
(x)[j
U `(x)],
(236)
with
t
"'
j
cU
cx
'
, u
''
`
c
'
U `(x),
and closure of the form drag resistance integral term using a form drag
coefcient, c
, is
1
AD ,
c
`
U
pds
1
2
c
S
"
(x)[j
U `(x)]. (237)
For these equations, the specic surface has two parts. The rst part, S
"'
, is
S
"'
(x)
1
AD ,
c
`
U
'
ds,
1
m
, (238)
where cS
"'
is the laminar subregion of the interface surface element cS
"
,
and
S
"
(x)
1
AD ,
c
`
UN
ds
S
AD
,
1
m
, (239)
where cS
"
is the cross ow projected area of the surface of the solid phase
inside the REV. Substitution into the one-dimensional momentum equation
yields
c
cx
p}
(c
'
(x)S
"'
(x) c
S
"
(x))
j
U `(x)
2m,

p`
m,
(Vm,). (240)
When the porosity is constant, the ow is laminar and S
"'
S
"
, the
equation becomes
dp`
dx
S
"
S
"
S
"
m,
j
U `
2
c
(U
, M
'
)
S
"
m,
j
U `
2
, (241)
where c
is the friction factor and c
the form drag, S

"
is the cross ow
form drag specic surface, and M
'
is a set of porous medium morphological
parameters or descriptive functions (see Travkin and Catton [16, 20]). The
drag terms can be combined for simplicity into a single total drag coefcient
to model the ow resistance terms in the general simplied momentum VAT
equation
c
(U
, M
'
)
S
"
S
"
. (242)
Correlations for drag resistance can be evaluated for a homogeneous
porous medium from experimental relationships for pressure drop. For
example, the equation often used for packed beds is
dp
`
dx
f

S
"
m,
j
`
2
. (243)
The complete VAT version of this equation is
c
cx
(m,p` )
1
AD ,
c
`
U
pds j
u`
''
S
"
(x)
m,j
U
cU
cx
j
c
cx _
cm,U
cx
c
cx
[j
m,u` u` }
]. (244)
If the porosity function is constant (a frequent assumption), the left-hand
side of Eq. (244) reduces to
dp`
dx
f

S
"
m,
j
U `
2
. (245)
Setting Eq. (245) equal to zero recovers equation (243). As a result, data
correlation using Eq. (243) incorporates the right-hand side of Eq. (244)
implicitly into the correlation. Friction factor data presented in this way
detracts from objectivity. The correlation can be written to reect all the
right-hand terms from Eq. (244),
d(m,p` )
dx

S
"
S
"
F
F
`
F
`
(S
"
(x))
j
U `
2
, (246)
where F
, . . . , F
`
are deduced from the following relationship:
(F
F
`
F
`
)
S
"
(x)
j
U `
2
m,j
U
cU
cx
j
c
cx
[m,u` u` }
]
j
c
cx _
cm,U
cx
. (247)
In the middle part of a porous medium sample, one can assume that the
porosity and ow regime are constant and steady state and then neglect all
terms on the right-hand side of (244). In reality, a large number of
experiments are being carried out under conditions where inputoutput
zones are present and can add signicantly to the value of the friction
coefcient because of the inputoutput pressure losses. If one wants to
separate the effects of inputoutput pressure loss from the viscous friction
and drag resistance components inside the porous medium, then taking into
account the terms shown in Eq. (247) is essential. There are correlations that
reect a dependence on sample thickness as a result of this oversight. An
even more complex situation arises when the ow and temperature inside
the medium are transient, such as one might nd in a regenerator, and very
inhomogeneous in space because of sharp gradients. The inhomogeneity in
space and time precludes neglecting the four right-hand terms in Eq. (244).
The inhomogeneous terms on the right-hand side of (247) may be
analyzed by scaling. Some of these terms are easily interpreted. For example,
the rst term on the right-hand side is the convective term
S
"
(x)
j
`
2
F
m,j
cU
cx
, (248)
and its importance can be strongly dependent on the thickness of the porous
specimen. This is why many studies report an obvious correlation with
specimen thickness. The remaining terms are the morphoconvective term
S
"
(x)
j
`
2
F
`
c
cx
(j
m,u
`
u
`
}
) (249)
and the momentum diffusion term
S
"
(x)
j
`
2
F
`
j
c
cx _
cm,U
cx
. (250)
The complete momentum equation written in a proper form for experi-
mental data reduction is
d(m,p` )
dx

S
"
S
"
F
F
`
F
`
(S
"
(x))
j
U `
2
(c
R
`
)(S
"
(x))
j
U `
2
, (251)
where
c
S
"
S
"
(252)
and
R
`
F
F
`
F
`
. (253)
The features of an experiment needed to treat terms such as F
, F
`
, F
`
are
discussed later.
The momentum resistance coefcient for a heterogeneous porous medium
can be written in the form
f
'
c
R
`
. (254)
This is the variable usually determined in most of porous medium ow
resistance experiments. Nevertheless, if this correlation value taken from an
experiment is later substituted into a modeling equation (with variable
porosity) of the form
m,U
cU
cx

1
j
c
cx
(m, p}
) v
c
cx
cm,U
cx
c
'
(x)
S
"
(x)
m,
j
U `
2
(255)
or
m,U
cU
cx

1
j
c
cx
(m, p}
) v
c
cx
cm,U
cx
j
k
'
m,U j
m,`
F
k`
'
U `, (256)
as is done by many, then the uctuation term c[m,u` u` }
]/cx is
neglected and the equation
2m,U
cU
cx

1
j
c
cx
(m, p}
) 2v
c
cx
cm,U
cx
1
j
AD ,
c
`
pds
v
AD ,
c
`
U
cU
cx
'
ds
c
cx
[m,u` u` }
] (257)
is being used as the problems model instead of
m,U
cU
cx

1
j
c
cx
(m, p}
) v
c
cx
cm,U
cx
1
j
AD ,
c
`
U
pds
v
AD ,
c
`
U
cU
cx
'
ds
c
cx
[m,u` u` }
] (258)
because the model used the coefcient c
'
(x) determined from
c
'
(x)
S
"
(x)
m(x),
j
U `
2
(c
(x) R
`
(x))
S
"
(x)
m(x),
j
U `
2
S
"
S
"
F
F
`
F
`
S
"
(x)
m(x),
j
U `
2
1
j
AD ,
c
`
U
pds
v
AD ,
c
`
U
cU
cx
'
ds
c
cx
[m,u` u` }
]
v
c
cx
cm,U
cx
m,U
cU
cx
, (259)
instead of using the coefcient c
(x) determined from

c
(x)
S
"
(x)
m(x),
j
U `
2

S
"
S
"
F
`
S
"
(x)
m(x),
j
U `
2
1
j
AD ,
c
`
U
pds
v
AD ,
c
`
U
cU
cx
'
ds
c
cx
[m,u` u` }
].
(260)
The terms needed for experimental data reduction model should include
all ve active terms,
d(m,p` )
dx

S
"
S
"
F
F
`
F
`
(S
"
(x))
j
U `
2
(c
R
`
)(S
"
(x))
j
U `
2
, (261)
with
f
'
c
R
`
. (262)
The general 1D VAT laminar regime constant viscosity momentum equa-
tion has six terms,
m,U
cU
cx

c
cx
u` u` ,
1
j
c
cx
(m, p}
1
j
AD ,
c
`
U
pds v
c
cx
cm,U
cx
v
AD ,
c
`
U
cU
cx
'
ds. (263)
For simplicity, Eq. (263) is written in the following shorthand notation:
UC
UMC
UP
UMP
UD
UMF
. (264)
The two right-hand integral terms reect the morphology-induced ow
resistance of the medium. Three ow resistance models are needed to
properly tie everything together.
a. Flow Resistance Model 1 The rst ow resistance model is for the
internal frictional and form drag resistance:
c
(U
, M
'
, x)
S
"
(x)U `(x)
2
(c
S
"
(x) c
'
(x)S
"'
(x))
U `(x)
2
1
j
AD ,
c
`
U
pds
v
AD ,
c
`
U
cU
cx
'
ds.
(265)
b. Flow Resistance Model 2 The second ow resistance model reects the
addition of the uid uctuation term UMC
:
c
`
(U
, M
'
, x)
S
"
(x)U `(x)
2
S
"
S
"
F
`
S
"
U `
2
(c
S
"
(x) c
'
(x)S
"'
(x))
U `(x)
2

S
"'
(x)
U `
2
F
`
1
j
AD ,
c
`
pds
v
AD ,
c
`
U
cU
cx
'
ds
c
cx
u` u` ,
.
(266)
c. Flow Resistance Model 3 The third ow resistance model reects all of
the terms responsible for momentum resistance in a porous medium:
c
`
(U
, M
'
, x)
S
"
(x)U `(x)
2
S
"
(x)
U `
2
R
`
c
(U
, M
'
)
S
"
U `
2
m,U
cU
cx
v
c
cx
cm,U
cx
1
j
AD ,
c
`
U
pds
v
AD ,
c
`
U
cU
cx
'
ds
c
cx
u` u` ,
, (267)
where
s
"
(x)
U `
2
R
`
(F
F
`
F
`
)
S
"
(x)
U `
2
m,U
cU
cx

c
cx
[m,u` u` }
] v
c
cx _
cm,U
cx
. (268)
Using the notation developed earlier for the terms in the momentum
equation (264) leads to a form for each of the ow resistance models that
properly reects their completeness,
c
`
(U
, M
'
, x) (UMP
UMF
S
"
(x)U `(x)
2
(269)
c
`
(U
, M
'
, x) (UMP
UMF
UMC
S
"
(x)U `(x)
2
(270)
c
(U
, M
'
, x)
(UC
UD
UMP
UMF
UMC
S
"
(x)U `(x)
2
. (271)
Each of the different forms will yield a correlation of a given set of data.
The problem is that the effects of the different characteristics that are
manifested in the terms in the equations are lost from consideration. If
predictive tools are to be developed, consideration must be given to the
impact of the details that the terms reect.
3. Scaling in Pressure L oss Experiments and Data Analysis
Direct use of any Ergun type friction factor in a Fanning or Darcy friction
factor correlation is incorrect. Ergun [167] suggested two types of friction
factors, one of which is the so-called kinetic energy friction factor f
''
, which
differs from the Fanning friction factor by a factor of three for the same
medium:
f
d
"
2j
u
`
`
AP
L
f
''
3
. (272)
For the same reason, direct implementation of the correlations given by
Kays and London [172] should be treated with care. For example, the
correlations for friction factor (Fanning) given by Kays and London for
ow through an innite randomly stacked, woven-screen matrix uses surface
porosity p, and specic surface :[1/m] to dene a hydraulic radius r
"
,
r
"
p
:
m,
`
S
"
.
Here the specic surface S
"
is dened as the interface surface divided by the
volume of the REV. Unfortunately, the surface porosity m,
`
and volume
porosity m, are not of the same value and even if they were, the expression
differs from that found earlier by a factor of 2.
Bird et al. [173] used the ratio of the volume available for ow to the
cross section available for ow in their derivation of hydraulic radius r
"'
.
This assumption led them to the formula
r
"'
m,d
6(1 m,)
. (273)
It would be double this value if a consistent denition were used for all
systems,
d
"
4m,
S
"
4m,
a
'
(1 m,)

2m,
3(1 m,)
d
4r
"'
, (274)
where a
'
is the particle specic surface (the total particle surface area
divided by the volume of the particle), and
S
"
a
'
(1 m,). (275)
The expression given by (274) is justied when an equal or mean particle
diameter is
d
6
a
'
,
which is the exact equation for spherical particles and is often used as
substitution for granular media particles. The value of hydraulic radius
given by Bird et al. [173], (273), was chosen by Chhabra [174] and was used
in determining the specic friction factor in capillary media.
Media of globular morphologies can be described in terms of S
"
, m,,
and d
and can generally be considered to be spherical particles with

S
"
6(1 m,)
d
, d
"
2
3
m,
(1 m,)
d
. (276)
This expression has the same dependency on equivalent pore diameter as
found for a one-diameter capillary morphology, leading naturally to
S
"
6(1 m,)
d
6(1 m,)
3
2
(1 m,)
m,
d
"
4m,
d
"
. (277)
This observation leads to dening a simple universal porous medium
scale,
d
"
d
'
4m,
S
"
, (278)
that meets the needs of both major morphologies, capillary and globular. A
large amount of data exists that demonstrates the insufciencies of the
Ergun drag resistance correlation (287). Because it was developed for a
specic morphology, a globular granular medium, application of the
Ergun correlation to a medium with arbitrary relationships between poros-
ity m,, specic surface S
"
, and pore (particle) diameter d
"
can lead to large
errors.
The particle diameter d
is often used as a length scale when reducing

experimental data. Chhabra [174], for example, writes the friction factor
f
''
m,`j
u
`
`
Ap
L
, (279)
This friction factor can be related to the friction factor f
'
, given by Eq.
(6.4-1) of Bird et al. [173], to the Fanning friction fator f
, and to the Ergun

kinetic energy friction factor f
''
as follows:
f
''
2 f
'
f
''

1 m,
m,`
f
1 m,
m,`
. (280)
These models all use different length scales, leading to large uncertainties
and confusion when a correlation must be selected for a particular applica-
tion. Little attention is paid to these differences, often requiring new
experimental data for a new medium conguration.
Only a few of the many issues important to modeling of pressure loss in
porous media are addressed here. As it is known, the two-term quadratic
ReynoldsForchheimer pressure loss equation is
AP
L
:jU
m,[j
`m,`; :
1
k
'
. (281)
By comparison with the simplied VAT (SVAT) momentum equation for
constant morphological characteristics and ow eld properties and only
the resistance coefcient c
,
AP
L
c

S
"
m,
j
`
2
, (282)
a set of transfer relationships can be found to transform Ergun-type
correlations and the SVAT expression. The transfer formula (Travkin and
Catton [21]) is
c
_
:j
j
[m,
2m,`
S
"

, (283)
where
: 150
1 m,)`
d`
m,`
, [1.75
(1 m,)
d
m,`
, (284)
or
c
A
Re
'
B, A
8:m,`
S`
"
, B2[
m,`
S
"
, (285)
where
Re
'
4U
m,
vS
"
.
The Ergun energy friction factor relation can be written in terms of the
VAT-based formulae (Travkin and Catton [21]) as
Ap
L
f
'

S
"
m,
j
`
2
. (286)
If the Ergun correlation is written using common notation, it becomes
Ap
L

150
(1 m,)`
d`
m,`
jm,U
1.75
(1 m,)
d
m,`
j
m,Ù
`, (287)
and if it can be further transformed to the (SVAT) Fanning friction factor,
then
f
'
A*
Re
B*
, A*
50(1 m,)
m,
, B*

3.5
6
0.583, (288)
where the particle Reynolds number is
Re
(U
)/v, (289)
and
f
'
A*
'"
Re
'
B*
'"
, with A*
'"
100
3
33.33, and B*
'"
B*
0.583,
(290)
where
Re
'
d
"
v

2
3
m,
(1 m,)
U
v
, and Re
'

3(1 m,)
2m,
Re
v
.
(291)
The common scaling length just derived will allow a great deal of data to
be brought to a common basis and allow greater condence in predictions.
4. Simulation Procedures
A large amount of data exists that demonstrates the inadequacies of the
Ergun drag resistance correlation (287). This is because the Ergun correla-
tion is used with arbitrary relationships between porosity m,, specic
surface S
"
, and pore (particle) diameter d
"
when it was originally developed
for granular media. How unsatisfactory it can be is shown in Fig. 5.
Fic. 5. Fanning friction factor f
(bulk ow resistance in SVAT for different medium

morphologies, materials, and scales used), reduced based on VAT scale transformations in
experiments by 1, Gortyshov et al. [175]; 2, Kays and London [172]; 3, Laminar, intermediate,
and turbulent laws in tube; 4, Gortyshov et al. [176]; 5, Beavers and Sparrow [177]; 6, SiC
foam (UCLA, 1997); 7, Ergun [167]; 8, Souto and Moyne [181]; 9, Macdonald et al. [180]; 10,
Travkin and Catton [23].
With specically assigned morphology characteristics (primarily S
"
), the
Ergun drag resistance correlation will be much closer to correlations by
Beavers and Sparrow [177] and Gortyshov et al. [176], as shown in Fig. 5.
A similar behavior was seen between the Ergun drag resistance correlation
and the drag resistance correlation by Gortyshov et al. [175].
Several other correlations are compared in Fig. 5. Gortyshov et al. [175]
experimentally derived correlations for the ReynoldsForchheimer momen-
tum equation in the form
: 6.61 10`(d
"
)"`m,'"``', (292)
[5.16 10`(d
"
)"`m,''', (293)
where hydraulic diameter d
"
(mm) is
d
"
d
"
[m]
0.001[m]
. (294)
These correlations have to be used in (285) and are for highly porous
(m,0.870.97) foamy metallic media. A Darcy type of friction factor
obtained by Gortyshov et al. [176] for very low conductivity porous
porcelain with high porosity is
f
'
(Re
"
)
40
Re
"
(1 2.5 10`m,``Re
"
), m,0.83 0.92, (295)
where
Re
"
d
"
m,
v
.
To transform this correlation, the Reynolds number must be transformed
and the result divided by 4 to yield the Fanning friction factor,
f
(Re
'
)
1
4
40
Re
'
m,
(1 2.5 10`m,``Re
'
m,)
, (296)
with
Re
'
5Re
"
/m, (297)
The correlation derived by Beavers and Sparrow [177] seems to be of
little value in the original form,
F
'`
(R
"
)
1
R
"
0.074, (298)
because the Reynolds number,
R
"
m,k
'
v
, (299)
contains the permeability of the medium and is usually not known. Noting
that, as pointed out by Beavers and Sparrow [177] the viscous resistance
coefcient : 1/k
'
, where k
'
is the Darcy permeability, and using the
transformation
F
'`
1
R
"
[k
'
, (300)
where
k
'
1
:
, R
"
m,
v:
(301)
Re
'
R
"
4
S
"
k
'
R
"
:
4
S
"
, R
"
Re
'

S
"
4:
, (302)
yields
F
'`
m,k
'
v
[k
'
or
F
'`
(R
"
)
1
:j
`m,`

AP
Ax
, (303)
and when compared to
f
(Re
'
)
2m,
j
`S
"
AP
Ax
, (304)
one obtains
f
(R
"
)
1
:m,`
1
j
`

AP
Ax
:m,`(2m,)
S
"

F
'`
(R
"
)
2:m,`
S
"

. (305)
This means that the Fanning friction factor, f
, can be assessed from the

friction factor suggested by Ward [178] and Beavers and Sparrow [177],
f
'`
, from
f
(R
"
) F
'`
(R
"
)
2:m,`
S
"

. (306)
To accomplish the transformation of F
'`
to f
, the permeability k
'
or the
viscous coefcient of resistance : porosity m, and specic surface S
"
must
be known. Estimates of f
were obtained from measured values of F

'`
for
FOAMETAL (Beavers and Sparrow sample Type C) using
k
'
19.01 10" [cm`] 19.01 10` [m`]
:
1
k
'
0.0526 10`
_
1
m`
(307)
and Eqs (299), (298) or (300), and (306) to transform the Beavers and
Sparrow [177] experimental data correlation to the Fanning friction factor
correlation. With
F
'`
(R
"
)
1
R
"
0.074 and R
"
Re
'

S
"
4:
F
'`
(Re
'
)
1
Re
'

4:
S
"

0.074, (308)
then
f
(Re
'
)
1
Re
'

4:
S
"

0.074
2:m,`
S
"

. (309)
Kurshin [179] has analyzed a vast amount of data using a consistent
procedure he developed to embrace all three ow regimes in porous media.
To carry out the procedure, the following parameters must be known:
(a) The viscous resistant coefcient :
, evaluated for laminar ow in a

pipe from the following:
AP
L
:
jm,U
, :
1
k
'
, U

d`
32j
AP
Ax
. (310)
(b) A characteristic length d
evaluated by equating the preceding ex-

pressions:
d
32
:
m,
`
32k
'
m,
`
. (311)
(This is only justied for straight parallel capillary morphology where
d
"
d
.)
(c) Critical numbers Re
'
and Re
'`
to distinguish the viscous, transi-
tional, and turbulent ltration regimes.
(d) Dimensionless viscous :
`
and inertial resistance [
`
coefcients in the
turbulent regime. Unfortunately, Kurshin [179] did not present any data for
foam materials and the porous metals he evaluated have low porosity in the
range m,0.5.
Now one can say that by reformulating existing experimental correlations
to the SVAT 1D form,
AP
L
f
(Re
'
)
S
"
m,
j
`
2
, (312)
the Fanning friction factor correlations can be easily compared with one
another as they have a common consistent basis. A number of correlations
were transformed and are in Fig. 5. The reason for the spread in the results
is thought to be inadequate accounting for details of the medium.
Analysis of Macdonald et al. [180] reformulated with the help of the
foregoing developed procedures gives the corrected Ergun-like type of
correlation
f
`
40
Re
'
0.6. (313)
Meanwhile, Souto and Moyne [181], using the DMM-DNM solutions,
came to the number of resistance curves that are separate for each morphol-
ogy. One of them for rectangular rods in VAT terms appears as
f
``
1
3
f
''
54.3
Re
'
, Re
'
0. (314)
VII. Experimental Measurements and Analysis of Internal
Heat Transfer Coefcients in Porous Media
A VAT-based approach applied to heat transfer in a porous medium
allows one to analyze and measure effective internal heat transfer coefcients
in a porous medium. As noted by Viskanta [182], Convective heat and
mass transfer in consolidated porous materials has received practically no
theoretical research attention. This is partially due to the complexity which
arises as a result of physical and chemical heterogeneity that is difcult to
characterize with the limited amount of data that can be obtained through
experiments. Viskanta [182, 183] generalized the data he analyzed for
internal heat transfer coefcient porous ceramic media using a correlation
of the form
Nu
'
2.0 a Re'Pr`, (315)
by assuming that the limiting Nusselt number should be 2.0 when the Re
decreases to zero. This assumption is only justied for unconsolidated sparse
spherical particle morphologies and is suspect for other porous medium
morphologies, especially consolidated media. For this reason, some re-
searches neglect this articial low Re limit and correlate their ndings
without it. The VAT approach is applied to heat transfer in porous media
to develop a more consistent correlation.
1. Experimental Assessment and Modeling of Heat Exchange
in Porous Media
The correct form of the steady-state heat transfer equation in the uid
phase of a porous media with primarily convective 1D averaged heat
transfer is
c
m,U
cT
cx

k
AD ,
c
`
U
cT
cx
'
ds k
c`m,T
cx`
c
c
cx
(m,u` T
)
c
cx _
k
AD ,
c
`
U
T
ds
. (316)
Equation (316) can be rewritten as
:`
'
S
"
(T }
`
T }
c
cx _
cm,T
cx
c
c
cx
(m,u` T }
c
cx _
k
AD ,
c
`
U
T ds
m,U
cT
cx
, (317)
where
k
AD ,
c
`
U
cT
cx
'
ds :`
'
S
"
(T }
`
T }
).
The right-hand side of Eq. (316) can also be written in the form
:`
'
S
"
(T }
`
T }
)
c
cx _
K
cT
cx
:`
'
S
"
(T }
`
T }
)
c
cx
[q
`
],
(318)
where the right-hand side (diffusive-like) ux contains more terms than
are conventionally considered:
q
`
_
K
cT
cx
m,k
cT
cx
c
m,u` T
AD ,
c
`
U
T
ds
. (319)
The corresponding equation for the solid phase is
c
cx
cs,T
`
}
`
cx
c
cx _
1
AD ,
c
`
U
T
`
ds
1
AD ,
c
`
U
cT
`
cx
'
ds
0. (320)
The three terms are written in the following shorthand form:
T
`
D
T
`
MD
T
`
ME
0. (321)
Equation (320) can also be written
0
1
k
`
:`
'
S
"
(T }
T }
`
)
c
cx
k
`
cT }
`
cx
:`
'
S
"
(T }
T }
`
)
c
cx
[q
``
]. (322)
Using the closure term for interface heat ux found earlier (they are
equal),
:`
'
S
"
(T }
T }
`
)
k
`
AD ,
c
`
U
cT
`
cx
'
ds
.
Equation (322) has a term that is usually overlooked (the second term on
the right):
q
``
K
`
cT }
`
cx
cs,T
`
cx

1
AD ,
c
`
U
T
`
ds
. (323)
Three heat transfer coefcient models are needed to properly tie every-
thing together. The rst model incorporates only the heat transfer coefcient
between the phases.
a. Model 1 of Heat Transfer Coefcient in Porous Media: Conventional
Modeling If it is assumed that the porous medium heat transfer coefcient
is dened by
:`
'
AD ,
c
`
U
cT
cx
'
ds
[S
"
(T }
`
T }
)], (324)
then the heat transfer equation becomes
c
m,U
cT
cx
k
c
cx _
cm,T
cx
:`
'
S
"
(T }
`
T }
), (325)
and when the porosity is constant, the equation becomes
c
U
cT
cx
k
c
cx _
cT
cx
:`
'
S
"
(T }
`
T }
)/m,. (326)
Most work uses an equation of this type. The experiments carried out will
reect the use of Eq. (326), and the data reduction will lead to a correlation
for :`
'
S
"
that is only valid for the particular medium used in the experi-
ment. There will be no generality in the results. By redening :`
'
, further
medium characteristics can be incorporated into the correlation. The second
model incorporates velocity and temperature uctuations.
b. Model 2 of Heat Transfer Coefcient in Porous Media: With Nonlinear
Fluctuations If we dene the heat transfer coefcient in a way that includes
the uctuations,
:`
'`
AD ,
c
`
U
cT
cx
'
ds c
c
cx
(m,u` T
))/[S
"
(T
`
}
`
T
)],
(327)
the second heat transfer model in porous media is almost the same as the
rst,
c
m,U
cT
cx
k
c
cx _
cm,T
cx
:`
'`
S
"
(T
`
}
`
T
). (328)
The third model is obtained by using the complete energy equation for the
uid phase. This is again done by redenition of the heat transfer coefcient.
c. Model 3 of Heat Transfer Coefcient in Porous Media: Full Equation
Energy Equation
:`
'`
AD ,
c
`
U
cT
cx
'
ds c
c
cx
(m,u` T
)
c
cx _
k
AD ,
c
`
U
T
ds
S
"
(T }
`
T }
(329)
The energy equation is again very similar:
c
m,U
cT
cx
k
c
cx _
cm,T
cx
:`
'`
S
"
(T
`
}
`
T
). (330)
Each of the models reects the data obtained for a given medium. Only the
coefcient :`
'`
, however, allows for a complete representation of the par-
ameters that reect the characteristics of the medium. In attempts by some
researchers to improve the modeling, a more complete equation is used
along with the more conventional denitions of the heat transfer coefcient.
The relative inaccuracy of substitution of coefcient into the correct mathe-
matical model,
c
m,U
cT
cx
c
c
cx
(m,u` T }
)
k
c
cx _
cm,T
cx

c
cx _
k
AD ,
c
`
U
T
ds
:`
'
S
"
(T }
`
T }
), (331)
can easily be seen by comparison with the denition of :`
'`
. The additional
terms are already a part of the coefcient, and double accounting has
occurred. The seriousness of such a mistake depends on the problem.
To summarize, the heat transfer coefcients and their respectively uid
heat transport equations can be written in terms of the notation given by
Eq. (321),
:`
'
(T
ME
)/[S
"
(T
`
}
`
T
)], (332)
AD ,
c
`
U
cT
cx
'
ds
[S
"
(T }
`
T }
)],
:`
'`
(T
ME
MC
)/[S
"
(T
`
}
`
T
)], (333)
AD ,
c
`
U
cT
cx
'
ds c
c
cx
(m,u` T
[S
"
(T
`
}
`
T
)],
:`
'`
(T
ME
MC
MD
)/[S
"
(T }
`
T }
)], (334)
AD ,
c
`
U
cT
cx
'
ds c
c
cx
(m,u` T
)
c
cx _
k
AD ,
c
`
U
T
ds
S
"
(T }
`
T }
)
.
Substitution of either of the preceding effective coefcients into the
equation
c
m,U
cT
cx
k
c
cx _
cm,T
cx
c
c
cx
(m,T
u`
'
}
c
cx _
k
AD ,
c
`
U
T
ds
1
AD ,
c
`
U
k
cT
cx
'
ds, (335)
T
MC
MC
ME
,
would result having different models for experimental data reduction and
even for experimental setup.
2. Simulation Procedures
Kar and Dybbs [184] developed several correlations for the internal heat
transfer in different porous media. Their model for assessment of internal
surface heat transfer coefcient is based on the formula (constructed slightly
differently than done by Kar and Dybbs [184] but with all the features)
:`
'^'
S
''
(c
`
T
`
c
)
S
"
AD
(T }
`
T
)
, (336)
which accounts for the heat exchange when T
`
and T
are the tempera-

tures of uid exiting and entering the control volume, which is taken to be
equal to AD
, through cross ow surface area S

''
[m`] with mass ow rate
Mj
S
''
[kg/s]. This denition of heat transfer coefcient corresponds
to the continuum mathematical model of heat exchange in the porous
medium formulated as
m,( jc
U
'
VT
:`
'^'
S
"
(T }
`
T
), (337)
instead of the correct equation,
m,( jc
U
'
VT
(jc
V T
u`
'
,
VV(m,T
)
k
V
_
1
AD ,
c
`
U
T
ds
AD ,
c
`
U
VT
ds. (338)
The last term can be modeled using the heat transfer coefcient given by
k
AD ,
c
`
U
cT
cx
'
ds
:`
'
S
"
(T }
`
T }
), (339)
which results from the closure relationship
1
AD ,
c
`
U
k
cT
cx
'
ds
1
AD ,
c
`
U
k
cT
cn
`
ds n
`
1
AD ,
c
`
U
q
`
ds
`
:`
'
S
"
(T }
`
T }
). (340)
Kar and Dybbs measured the temperatures T
`
and T
and treated them as

if they were the mean (averaged) temperatures. As a result, they measured
yet another heat transfer coefcient, :`
'"
, that is dened by
:`
'"
S
"
(T }
`
T
) :`
'^'
S
"
(T }
`
T
)
(jc
V T
u`
'
,
VV(m,T
)
k
V
_
1
AD ,
c
`
U
T
ds
AD ,
c
`
U
VT
ds. (341)
The second and third terms in Eq. (341) are usually negligible. When they
are, the measured heat transfer coefcient reduces to the second heat transfer
coefcient in porous medium :`
'`
,
:`
'`
S
"
(T }
`
T
) :`
'^'
S
"
(T }
`
T
)
(jc
VT
u`
'
,
AD ,
c
`
U
VT
ds. (342)
Fic. 6. Internal effective heat transfer coefcient in porous media, reduced based on VAT
scale transformations in experiments by 1, Kar and Dybbs [184] for laminar regime; 2,
Rajkumar [185]; 3, Achenbach [186]; 4, Younis and Viskanta [187]; 5, Galitseysky and
Moshaev [189]; 6, Kokorev et al. [190]; 7, Gortyshov et al. [175]; 8, Kays and London [172];
9, Heat Exchangers Design Handbook [191].
This is probably why the correlation developed by Kar and Dybbs [184] is
located low among the second group of correlations in Fig. 6, where a
number of correlations are presented after being rescaled using VAT. If the
measured coefcient is :`
'
, the result will be even lower than :`
'`
.
As the number of terms that can be estimated increases, the value of the
coefcient decreases. This is probably the case with the rst group of
correlations shown in Fig. 6. A large amount of the data analyzed by
Viskanta [182, 183] was used to deduce consistent correlations for compari-
son of internal porous media heat transfer characteristics. The same scaling
VAT approach used for ow resistance in porous media is used for heat
transfer.
One of the correlations developed by Kar and Dybbs [184], correlation
(11) on p. 86, is for laminar ow in sintered powder metal specimens. It is
Nu
"
h
`
d
"
z
0.004 Re``
"
Pr`, (343)
where both Nu and Re are based on the mean pore diameter. If a single
hydraulic diameter d
"
is
d
"
5d
'
4m,
S
"
, (344)
then
Re
"
Re
'
4U
m,
vS
"
(345)
Nu
'
(Re
'
)
h
`
d
'
z
5Nu
"
(Re
"
, m,, S
"
) 0.004 Re``
'
Pr`. (346)
This correlation is shown in Fig. 6. The correlation developed by Rajkumar
[185] for hollow ceramic spheres is
Nu
h
`
d
1.1
Re
Pr
d
L
"`"
, (347)
with d
2.53.5 [10`m], 18 Re
980, m,0.380.39, Pr 0.71,

and
Re
v
.
The particle Reynolds number Re
can be rewritten using

Re
Re
'
3(1 m,)
2m,
. (348)
Nu
needs to be transformed to Nu
'
by relating the particle diameter d
to
the hydraulic diameter. The result is
Nu
'
5
h
`
d
"
z
Nu
"
2m,
3(1 m,)
Nu
(Re
)
2m,
3(1 m,)
Nu
(x),
(349)
where
x
3(1 m,)
2m,
Re
'
.
Then
Nu
'
2m,
3(1 m,)
Nu
(x, Pr, d
, L )
2m,
3(1 m,)
Nu

3(1 m,)
2m,
Re
'
, Pr, d
, L
. (350)
Achenbach [186] developed the correlation
Nu
"
(1.18 Re"``
"
)"
_
0.23
Re
"
m,
"``
"
"
, (351)
for Pr 0.71, m,0.387, and 1 (Re
"
/m,) 7.710`. The Reynolds
number used by Achenbach is based on hydraulics and
Re
"
5Re
'
m,,
and his denition of Nu
"
is
Nu
'
(Re
'
) 5Nu
"
(Re
'
m,). (352)
A correlation developed for cellular consolidated ceramics by Younis and
Viskanta [187, 188] is
Nu
'"
h
'
d`
"
z
0.0098 0.11
d
"
L
Re`
"
Pr`, (353)
where m,0.830.87. The correlation yields an increasing Nu
'"
when the
test specimen thickness is decreased. This is a clear inuence of inow and
outow boundaries on heat transfer. Transforming from a volumetric
Nusselt number Nu
'
to a conventional surcial value Nu yields
Nu
'
Nu
'"
(Re
'
m,)
4m,
. (354)
Viskanta [183] presents a correlation from a study of low porosity media,
0.167m,0.354, by Galitseysky and Moshaev [189]:
Nu
'"
Am,`(1 m,)`` Re
"
Pr. (355)
The coefcient, A given by Viscanta [183] is
A
37.2
d
"
L
0.59
(m,(1 m,)`)"`, (356)

for 0.15 d
"
/L 0.23, 10 Re
"
530, Pr 0.71. The volumetric Nusselt
number is transformed to the surcial Nusselt number with Eq. (354).
A semiempirical theory was used by Kokorev et al. [190] to develop a
correlation between resistance coefcient and heat transfer coefcient for
extensive ow regimes in porous media that only contains one empirical
(apparently universal for the turbulent regime) constant. On the basis of this
relationship, the concept of uctuation speed scale of movement is used to
obtain an expression for the heat transfer coefcient from the Darcy friction
factor, f
'
4 f
4c
:
Nu
h
`
d
[0.14(4c
Re`
"
)" Pr`]. (357)
Transforming their expression to the general form of the media Nusselt
number yields
Nu
'
2m,
3(1 m,)
Nu
(Re
'
m,). (358)
The heat transfer coefcient given in the Heat Exchanger Design Hand-
book [191] is based on a single sphere heat transfer coefcient for the porous
medium,
h
`
( f
'
Nu
`
), Nu
`
2 (Nu`
'
Nu`
'
)`, (359)
where
Nu
'
0.664Re`
Pr`
Nu
'
(0.037Re"`
Pr)
(1 2.443Re"
(Pr`` 1))
,
for 1 Re
10`, 0.6 Pr 10`, and the form coefcient for 0.26

m,1.0 is
f
'
1 1.5(1 m,).
Transformation of the Nusselt number yields
Nu
'
2m,
3(1 m,)
Nu
(Re
). (360)
Nu
'
values at low Reynolds number are unrealistic, leading to the
conclusion that the transition type expression used to treat both laminar
and turbulent ows is probably not adequate for heat transfer in porous
media.
Gortyshov et al. [175] developed a correlation for the internal heat
transfer coefcient for a highly porous metallic cellular (foamy) medium
with porosity in the range 0.87 m,0.97,
Nu
'"
h
'
d`
"
z
0.606Pe"`'
"
m,``, (361)
where
Pe
"
Re
"
Pr
U
m,d
"
a
, (362)
d
"
is in millimeters (see (294)), and Nu
'"
is the volumetric internal heat
transfer coefcient assessed using
h
'
q
'
h
`
q
`
cS
"
D
`
S`
"
. (363)
Also,
Nu
'
(Re
'
)
Nu
'"
(m, Re
'
)
4m,
. (364)
The correlation given by Kays and London [172] is
StPr'``' 1.4Re""`
'
, (365)
which is transformed by
NuPr'``'
Re
'
Pr
1.4Re""`
'
, =Nu
'
1.4Re"``
'
Pr`. (366)
Some useful observations can be made by comparing the heat transfer
relationships shown in Fig. 6. One of the most signicant observations is
that the large differences between the correlations by Kar and Dybbs [184],
Younis and Viskanta [187, 188], Rajkumar [185], and others cannot be
explained if one does not take into account the specic details of the
medium and the experimental data treatment. Given this, the remarkable
agreement, almost coincidence, of the correlations by Kays and London
[172], Achenbach [186], and Kokorev et al. [190] should be noted. These
correlations were developed using different techniques and basic ap-
proaches. The correlation given in the Heat Exchangers Design Handbook
[191] reects careful adjustment in the low Reynolds number range. The
correlation is not based on a specic type of medium (for example, a
globular morphology with a specic globular diameter). Rather, it was
developed to summarize heat transfer coefcient data in packed beds for a
wide range of Reynolds numbers using an assigned globular diameter. As a
result, it is not solidly based on physics, and a simple transformation from,
particle to pore scale does not work properly.
VIII. Thermal Conductivity Measurement in a Two-Phase Medium
A majority of thermal conduction experiments are based on a constant
heat ux through the experimental specimen and measurement of interface
temperatures. Data reduction (see, for example, Uher [192]) is accomplished
using
K
QL
AAT
, (367)
where Q is the electrical power from heater dissipated through the specimen,
L is the distance used to measure the temperature difference, and A is the
uniform cross-sectional area of the sample.
1. Traditional L ocal and Piecewise Distributed Coefcient Heat
Conductivity Problem Formulations
In DMM-DNM as, for example, for a dielectric medium, the equation
usually used is
V (k(r)VT (r)) 0, r - D, (368)
where the conductivity coefcient function k is
k(r) k
,''(r) k
`
,'`'(r), (369)
and ,''' is the characteristic function of phase i 1 ^2 (see, for example,
Cheng and Torquato [193]). Interface boundary conditions assumed for
these equalities are
T
(r) T
`
(r), r - cS
`
(370)
k
(n VT
(r)) k
`
(n VT
`
(r)), r - cS
`
. (371)
2. Effective Coefcients Modeling
To begin, we choose the conductivity problem and rst will be treating
the example of constant phase conductivity coefcient conventional equa-
tions (368) for the heterogenous medium.
As shown elsewhere (see, for example, Travkin and Catton [21]), this
mathematical statement is incorrect when the equation is applied to the
volume containing both phases, even when coefcient k(r) is taken as a
random scalar or tensorial function. The reason for this is incorrect
averaging over the medium, which has discontinuities.
Conventional theories of treatment of this problem do not specify the
meaning of the eld T, assuming that it is the local variable, or T T (r),
where at the point r the point value of potential T exists.
Next, the analysis shows that the coefcient k k(r), as long as in each
separate lower scale level point r there exists the local k with the value of
either phase 1 or phase 2, and in each of the phases the value of k
'
is
constant.
In the DMM-DNM approaches the mathematical statement usually deals
with the local elds, and as soon as the boundary conditions are taken in
some way, the problem became formulated correctly and can be solved
exactly, as in work by Cheng and Torquato [193].
Difculties arise when the result of this solution needs to be interpreted
and this is in the majority of problem statements in heterogeneous media,
in terms of nonlocal elds, but averaged in some way. The averaging
procedure usually is stated as being done either by stochastic or by spatial,
volumetric integration. Almost all of these averaging developments are done
incorrectly because of a disregard of averaging theorems for differential
operators in a heterogeneous medium. More analysis of this matter is given
in work by Travkin et al. [115].
Further, a more complicated situation arises when the intention is to
formulate and nd effective transport coefcients in a heterogeneous me-
dium. Let us consider the conductivity problem in a two-phase medium.
According to most accepted mathematical statements this problem is given
as (368)(371).
3. Conventional Formulation of the Effective Conductivity Problem in a
Two-Phase Medium
One of the methods of closure of mathematical models of diffusion
processes in a heterogeneous medium is the quasihomogeneous method
(Travkin and Catton [21]). In this case, the transfer process is modeled as
an ideal continuum with homogeneous effective transport characteristics
instead of the real heterogeneous characteristics of a porous medium. This
method of closure of the diffusive terms in the heat and mass diffusion
equations results in certain limitations: (a) the two-phase medium compo-
nents are without uctuations of the type T, c` in each of the phases; and (b)
the transfer coefcients being constant in each of the phases (Khoroshun
[194, 195]) results in reducing them to additional algebraic equations. These
equations relate the unknown averaged diffusion ows in each of the phases
in the form
j,
j,
`
k*
VT ,, (372)
when for constant (effective) coefcients it is
k
VT ,
k`
VT ,
`
k*
VT ,, (373)
and also
VT , VT ,
VT ,
`
, (374)
so it might be written as
(k
) j ,
(k`
) j ,
`
VT ,. (375)
Here k
, k`
are the transfer coefcient tensors in each of the phases, and

k*
is the effective conductivity coefcient. Thus, at least in this case, the

problem of closure has been reduced to nding k*
.
Applying the closure relation, for example,
k
VT ,
k`
VT ,
`
, (376)
yields the effective stagnant coefcient
k*
2k
k`
(k
k`
)
, (377)
which represents the lower bound of the effective stagnant conductivity for
a two-phase material from the known boundaries of HashinShtrikman
(see, for example, [196], Kudinov and Moizhes [197]) for equal volume
fraction of phases. Other closure equations for calculating the stagnant
effective conductivity are found in work by Hadley [198] and by Kudinov
and Moizhes [197]. The quasi homogeneous approach has several defects:
(a) The basis for the quasi-homogeneous equations is in question, (b) the
local uctuation values, as well as inhomogeneity and dispersivity of the
medium, are neglected, and (c) the interdependence of the correlated
coefcients and arbitrary adjustment to t data signicantly reduce the
generality of the results.
4. VAT-Based Considerations for Heterogeneous Media Heat Conductivity
Experimental Data Reduction
Let us consider the data reduction procedure of the heterogeneous
material thermal conductivity experiment.
a. Constant Heat Conductivity Coefcient We treat the example of the
constant coefcient heat transfer equation for a heterogeneous medium and
show the problem in terms of conventional experimental bulk data reduc-
tion procedures and pertinent modeling equations.
Consider an experiment on determining the thermal coefcient of phase 1
(for example) in composite (or in material that is considered as being a pure
substance, but really is composite) material.
The heat transport for material phase 1 is described by
s
,(jc
cT
ct
k
V`(s
,T
) k
V
_
1
AD ,
c
`
`
T
ds
AD ,
c
`
`
VT
ds
,
which needs the closure of the second and the third r.h.s. terms. The latter is
k
AD ,
c
`
`
cT
cx
'
ds
:`
`
S
`
(T }
`
T }
), (378)
where the closure procedure is quite applicable to description of the
uidsolid medium heat exchange and might be considered as the analogs
for the case of solidsolid heat exchange, as done in many papers. The more
strict and precise integration of the heat ux over the interface surface gives
the exact closure for that term in governing equations for both neighboring
phases.
Also considering the two terms on the r.h.s., having them as diffusion bulk
terms means that
k
V`(s
,T
) k
V
_
1
AD ,
c
`
`
T
ds
V [q
],
where the diffusive-like ux q
contains some more terms than are

conventionally considered,
q
V(s
,T
) k
V(s
,T
)
k
AD ,
c
`
`
T
ds
, (379)
where the heat ux in phase 1 is determined through the averaged tempera-
ture T
.
So, the effective (not homogeneous) conductivity coefcient in phase 1 is
k
_
V(s
,T
)
1
AD ,
c
`
`
T
ds
(V(s
,T
))
k
_
1
1
AD ,
c
`
`
T
ds
(V(s
,T
))
. (380)
There is a difference between this introduced coefcient k
and that
traditionally determined through the ux in phase 1, which is
q
[k
VT
] k

V(s
,T
)
1
AD ,
c
`
`
T
ds
. (381)
Arising in this situation is the effective conductivity coefcient determina-
tion
k
_
V(s
,T
)
1
AD ,
c
`
`
T
ds
VT
, (382)
which is a different variable indeed and which is still the one that is not the
traditional effective heterogeneous medium heat conductivity coefcient
(determined in all phases),
q [k
VT ,] k
[VT ,
VT ,
`
]
k
V(s
,T
s
`
,T
`
) k
VT ,. (383)
After those transformations the heat transfer equation in phase 1 becomes
s
,( jc
cT
ct
V [k
V(s
,T
)] :`
`
S
`
(T }
`
T }
). (384)
Repeating all of this for the steady-state heat conductivity equation
V`(s
,T
) V
_
1
AD ,
c
`
`
T
ds
1
AD ,
c
`
`
VT
ds
0, (385)
one obtains
k
_
V(s
,T
)
1
AD ,
c
`
`
T
ds
(V(s
,T
)) (386)
for the equation
V [k
V(s
,T
)]
:`
`
k
S
`
(T }
`
T }
) 0, (387)
where k
does not even depend explicitly on the phase heat conductivity

coefcient k
(if the latter is taken as a constant value). Generally speaking,

it depends on k
implicitly through the boundary conditions and the

conditions at the interface surface cS
`
.
Of course, the situation changes if the heat exchange term (last term in
(385)) is taken into account as the input correlation factor for conventional
bulk effective heat conductivity coefcient k
in the equation
V [k
V(s
,T
)] 0. (388)
The main reason why in the present problem treatment the interphase
heat exchange term is separated from the other two terms in the r.h.s. of Eq.
(385) is that this logistics gives clarity in analysis and modeling of interface
transport processes, which is not present in conventional composite medium
modeling.
Also, in the more complete and challenging physics of interface transport
modeling as in the third phase, this third interphase exchange term, along
with the second term, is an issue tightly connected to the closure problem
and to the models of interface surface transport.
b. Nonlinear Heat Conductivity of a Pure Phase Material Meanwhile, for
materials such as high-temperature superconductors (HTSC), a constant
heat conductivity coefcient is not a justiable choice, as the usual analysis
of approaches has shown above. That means complications in treating the
equation with a nonlinear heat conductivity coefcient in phase 1,
s
,(jc
cT
ct
V [K
V(s
,T
)] V [s
,K
VT
]
V
_
K
AD ,
c
`
`
T
ds
1
AD ,
c
`
`
K
cT
cx
'
ds
,S
'
`
}
,
(389)
where the effective conductivity model has two additional terms, one of
which reects the mean surface temperature over the interface surface inside
of the REV, and the other of which results from nonlinearity of the elds
inside subvolume AD
,
K
_
K
V(s
,T
) s
,K
VT
AD ,
c
`
`
T
ds
(V(s
,T
)), (390)
which when inserted in the heat transport equation gives
s
,(jc
cT
ct
V [K
V(s
,T
)] :`
`
S
`
(T }
`
T }
) s
,S
'
`
}
. (391)
Meanwhile, when an experimentalist evaluates his or her experimental
data using the equation
(jc
cT
ct
V [k
VT
] (393)
with the calculation shown earlier of the thermal conductivity coefcient
using experimental data, he or she makes two mistakes:
1. He is confusing the materials clear homogeneous conductivity coef-
cient k
(which is the subject of his experiment) with the effective coefcient

k
of the same phase in a composite which is just another variable.

2. Doing data reduction as for the modeling equation
(jc
cT
ct
V [k
`
VT
] (392)
meaning that
k
5k
, (394)
and seriously believing that he measures the real homogeneous k
he seeks,
he drops out (but in reality he takes implicitly into account) the term
reecting the exchange rate,
:`
`
S
`
(T }
`
T }
), (395)
in the composite material, which is experiencing at least two temperatures
and usually a great inuence of the internal exchange rate (see work by
Travkin and Kushch [33, 34] and Travkin et al. [21]). In this way, an
experimentalist makes a second mistake due to miscalculation of the
inuence of this additional termyet the conductivity coefcient
k
`
evaluated from experiment is not the value it is considered to be
k
`
5/ k
.
When the experimentalists goal is the measurement, not of a bulk
effective coefcient of a material, but of the pure materials conductivity
coefcient, considerations regarding the issues of homogeneity and experi-
mental data modeling are of primary interest.
The standard denition of the effective (macroscopic) conductivity tensor
is determined from
j ,k*
''
VT ,, (396)
in which it is assumed that
j ,j ,
j ,
`
k
VT ,
k
`
VT ,
`
k*
''
VT ,k*
''
VT ,
k*
''
[VT ,
VT ,
`
] k*
''
VT ,
k*
''
VT ,
`
, (397)
so, for the usually assumed interface cS
`
physics, the effective coefcient is
determined to be
k*
''
VT , [k
VT ,
k
`
VT ,
`
]
k
V(m
,T
) k
`
V(m
`
,T
`
) (k
k
`
)
1
AD ,
c
`
`
T
ds
(398)
or
k*
''
_
k
V(m
,T
) k
`
V(m
`
,T
`
) (k
k
`
)
1
AD ,
c
`
`
T
ds
VT ,,
(399)
or
k*
''
_
k
V(m
,T
) k
`
V(m
`
,T
`
) (k
`
k
)
1
AD ,
c
`
`
T
`
ds
`
[VT ,
VT ,
`
]
,
(400)
which involves knowledge of three different functions, T
, T
`
, T
cS
`
, in the
volume D. This formula for the steady-state effective conductivity can be
shown to be equal to the known expression
k*
''
VT ,k
`
VT ,(k
k
`
)
1
AD ,
`'
`
VT dc
k
`
VT ,(k
k
`
)VT ,
. (401)
It is worth noting here that the known formulae for the effective heat
conductivity (or dielectric permittivity) of the layered medium
k*
_
'
m
'
,k
'
, i 1, 2, (402)
for a eld applied parallel to the interface of layers, and
k*

_
_
'
m
'
,
k
'

(403)
when the heat ux is perpendicular to the interface, are easily derived from
the general expression (399) using assumptions that intraphase elds are
equal, T
T
`
, that interface boundary conditions are valid for averaged
elds, and that adjoining surface interface temperatures are close in magni-
tude. The same assumptions are effectual when conventional volume aver-
aging techniques are applied toward the derivation of formulae (402) and
(403).
5. Bulk Heat Conductivity Coefcients of a Composite Material
The problem becomes no easier in the case when the effective conductivity
coefcient is meant to serve for the whole composite material. Combining
both temperature equations (if only two phases are present) for the simplest
case of constant coefcients,
s
,(jc
cT
ct
k
V`(s
,T
) k
V
_
1
AD ,
c
`
`
T
ds
AD ,
c
`
`
VT
ds
s
`
,(jc
)
`
cT
`
ct
k
`
V`(s
`
,T
`
)k
`
V
_
1
AD ,
c
`
`
T
`
ds
`
k
`
AD ,
c
`
`
VT
`
ds
`
,
into one equation by adding one to another, we obtain
s
,(jc
cT
ct
s
`
,(jc
)
`
cT
`
ct
V (k
V(s
,T
) k
`
V(s
`
,T
`
))
V
_
k
AD ,
c
`
`
T
ds
k
`
AD ,
c
`
`
T
`
ds
`
AD ,
c
`
`
VT
ds
k
`
AD ,
c
`
`
VT
`
ds
`
,
(404)
keeping in mind that the two-phase averaged temperature is
T ,s
,T
s
`
,T
`
. (405)
One can write down the mixture temperature equation when summation
of the equations gives (when taking into account the boundary condition of
temperature uxes equality at the interface surface, (k
VT
) (k
`
VT
`
))
s
,(jc
cT
ct
s
`
,(jc
)
`
cT
`
ct
V (k
V(s
,T
) k
`
V(s
`
,T
`
)) (k
k
`
)V
_
1
AD ,
c
`
`
T
ds
,
(406)
or, written in terms of thermal diffusivities a
and a
`
,
cT ,
ct
V [a
V(s
,T
) a
`
V(s
`
,T
`
)] (a
a
`
)V
_
1
AD ,
c
`
`
T
ds

1
a
`
a
k
`
_
1
AD ,
c
`
`
VT
ds
, a
'
k
'
(jc
)
'
, i 1, 2, (407)
which has the three different temperatures T
, T
`
, and T
(cS
`
) (here
T ,s
,T
s
`
,T
`
).
And, assuming only a local thermal equilibrium,
T ,s
,T
s
`
,T
`
T * T
T
`
, (408)
the mixed temperature equation becomes two-temperature T *, T
(cS
`
)
dependable with simplied left hand part of the equation
(s
,(jc
s
`
,(jc
)
`
)
cT *
ct
V [(k
V(s
,T *) k
`
V(s
`
,T *))]
(k
k
`
)V
_
1
AD ,
c
`
`
T
ds
. (409)
With the two different temperatures, the effective coefcient of conductiv-
ity is equal to
k*
[(k
V(s
,T *) k
`
V(s
`
,T *))]
(k
k
`
)
_
1
AD ,
c
`
`
T
ds
(VT *,). (410)

This formula coincides with the effective coefcient of conductivity for the
steady-state effective conductivity in the medium and can be shown to be
equal to the known expression
k*
VT ,k
`
VT ,(k
k
`
)
1
AD ,
`'
`
VT dc. (411)
From this formula an important conclusion can be drawn: that the most
sought-after characteristics in heterogeneous media transport, which are the
effective transport coefcients, can be correctly determined using the con-
ventional denition for the effective conductivity for example, for the
steady-state problem
j ,k*
VT , k
`
VT ,(k
k
`
)
1
AD ,
`'
`
VT dc, (412)
but only in a fraction of problems, while employing the DMM-DNM exact
solution. The issue is that in a majority of problems, such as for in-
homogeneous, nonlinear coefcients and in many transient problems, hav-
ing the two-eld DMM-DNM exact solution is not enough to nd effective
coefcients. As shown earlier, only the requirement of thermal equilibrium
warrants the equality of steady-state and transient effective conductivities in
a two-phase medium.
The second form of the same equation with the surface integral of the
uctuation temperature in phase 1 is
(s
,( jc
s
`
,(jc
)
`
)
cT *
ct
V [(k
,k
`
s
`
,)V(T *)]
(k
k
`
)V
_
1
AD ,
c
`
`
T
ds
, (413)
still having the phase 1 temperature uctuation variable in one of the terms.
The following equality arises while comparing the two last equations (409)
and (413):
[(k
V(s
,T *) k
`
V(s
`
,T *))] (k
k
`
)
_
1
AD ,
c
`
`
T
ds
[(k
, k
`
s
`
,)V(T *)] (k
k
`
)
_
1
AD ,
c
`
`
T
ds
. (414)
As can be seen, the transient effective diffusivity coefcent a
in the VAT
nonequilibrium two-temperature equation (407) can be derived through the
equality
a
VT , a
V(s
,T
) a
`
V(s
`
,T
`
) (a
a
`
)
_
1
AD ,
c
`
`
T
ds
V''

1
a
`
a
k
`
_
1
AD ,
c
`
`
VT
ds
(415)
or
a
VT , a
V(s
,T
) a
`
V(s
`
,T
`
) (a
a
`
)
_
1
AD ,
c
`
`
T
ds
A,
(416)
where V'' is the inverse operator V (V''( f )) f such that if
V Aa

1
a
`
a
k
`
_
1
AD ,
c
`
`
VT
ds
, (417)
then
AV''

1
a
`
a
k
`
_
1
AD ,
c
`
`
VT
ds
. (418)
From the preceding expression, the transient effective nonequilibrium
coefcient in a two-phase medium can be determined as
k
(s
,(jc
s
`
,(jc
)
`
), (419)
which looks rather inconvenient for analytical or experimental assessment
or numerical calculation. The solution of this problem, which includes as an
imperative part the nding of the effective bulk composite material heat
conductivity (diffusivity), coefcient, is equal to the solution of the exact
two-phase problem. We see that the two-temperature DMM-DNM is not
enough for the convenient construction of the effective coefcient of conduc-
tivity. As we can compare the expressions for transient coefcient (419) and
thermal equilibrium coefcient (410) they are of great difference in denition
and in calculation. And it does not matter which kind of mathematical
statement is used for the problemthe two separate heat transfer equations
or the VAT statement the problem complexity is the same. Only by using
the VAT equations is the correct estimation of the transient effective
coefcients on the upper scale available.
If we adopt the idea that phase temperature variables in each of the
subvolumes AD
and AD
`
can be presented as sums of the overall tempera-
ture and local uctuations (Nozad et al. [40]),
T
T ,T
V
, T
`
T ,T
V
`
, (420)
which means an introduction of the two new variables T
V
and T
V
`
, then the
equation for the composite averaged temperature follows (Nozad et al. [40])
in the form
(s
,(jc
s
`
,(jc
)
`
)
cT ,
ct
V
,k
_
VT ,
1
AD
,
c
`
`
T
ds
s
`
,k
`
_
VT ,
1
AD
`
,
c
`
`
T
`
ds
`
,(jc
cT
V
ct
s
`
,(jc
)
`
cT
V
`
ct
V (s
,k
VT
V
s
`
,k
`
VT
V
`
)
(421)
which has ve variable temperatures. If the assumptions and constraints
given in Nozad et al. [40] are all satised, then the nal equation with only
three different temperatures resumes:
(s
,(jc
s
`
,(jc
)
`
)
cT ,
ct
V
,k
_
VT ,
1
AD
,
c
`
`
T
ds
s
`
,k
`
_
VT ,
1
AD
`
,
c
`
`
T
`
ds
`
.
(422)
This means that the neglect of the global deviation T
V
, T
V
`
terms still does
not remove the requirement of a two-temperature solution.
a. Effective Conductivity Coefcients in a Porous Medium When Phase One
Is a Fluid In phase 1 the VAT equation is written for the laminar regime.
In the work by Kuwahara and Nakayama [199] is given the DMM-DNM
solution of the 2D problem of uniformly located quadratic rods with equal
spacing in both directions. Studies were undertaken of both the For-
chheimer and post-Forchheimer ow regimes.
This work is a good example of how DMM-DNM goals cannot be
accomplished, even if the solution on the microlevel is obtained completely,
if the proper VAT scaling procedures basics are not applied.
The one structural unit periodic cell in the mediumwas taken for
DMM-DNM.
Equations were taken with constant coefcients, and in phase 1 the VAT
equation was written for the laminar regime as
m,(jc
cT
ct
m,(jc
U
'
VT
(jc
V T
u`
'
,
VV (m,T
)
k
V
_
1
AD ,
c
`
U
T
ds
AD ,
c
`
U
VT
ds. (423)
Adding this equation to the VAT solid-phase (second phase) two-
temperature equation gives
m,(jc
cT
ct
s
`
,(jc
)
`
cT
`
ct
m,(jc
U
'
VT
(jc
V T
u`
'
,
V (k
V(m,T
) k
`
V(s
`
,T
`
))
V
_
k
AD ,
c
`
`
T
ds
k
`
AD ,
c
`
`
T
`
ds
`
AD ,
c
`
`
VT
ds
k
`
AD ,
c
`
`
VT
`
ds
`
, (424)
which reduces because of interface ux equality to
m,(jc
cT
ct
s
`
,(jc
)
`
cT
`
ct
m,(jc
U
'
VT
V (k
V(m,T
) k
`
V(s
`
,T
`
)) (jc
V T
u`
'
,
(k
k
`
)V
_
1
AD ,
c
`
`
T
ds
, (425)
which has two averaged temperatures T
and T
`
, interface surface integrated
temperature T
(cS
`
), and two elds of uctuations T
(x) and u`
'
(x),
assuming that the velocity eld is also computed and known.
We now write the effective conductivity coefcients for (425) and for the
one-temperature equation when temperature equilibrium is assumed.
In the rst case, for the weighted temperature,
T ", (m,(jc
s
`
,(jc
)
`
T
`
)/w
'
(426)
w
'
m,(jc
s
`
,(jc
)
`
const, (427)
the equation can be written as
w
'
cT ",
ct
m,(jc
U
'
VT
V (k
V(m,T
) k
`
V(s
`
,T
`
)) (jc
V T
u`
'
,
(k
k
`
)V
_
1
AD ,
c
`
`
T
ds
, (428)
where three temperatures are unknown, T ",, T
, and T
`
, plus the interface
surface temperature integral T
(cS
`
) and uctuation elds T
(x) and u`
'
(x).
The effective coefcient of conductivity can be looked for is
k
VT ",(k
V(m,T
) k
`
V(s
`
,T
`
)) (jc
u`
'
,
(k
k
`
)
_
1
AD ,
c
`
`
T
ds
. (429)
In order to avoid the complicated problems with effective conductivity
coefcient denition in a multitemperature environment, Kuwahara and
Kakayama [199], while performing DMM-DNM for the problem of
laminar regime transport in a porous medium, decided to justify the local
thermal equilibrium condition
T ,m,T
s
`
,T
`
T * T
T
`
,
which greatly changes the one effective temperature equation. This equation
becomes simpler with only one unknown temperature T * and variable eld
T
and is written as
(m,(jc
s
`
,(jc
)
`
)
cT *
ct
m,(jc
U
'
VT *
V (k
V(m,T *) k
`
V(s
`
,T *)) (jc
VT
u`
'
,
(k
k
`
)V
_
1
AD ,
c
`
`
T
ds
, (430)
as the variable temperature and velocity uctuation elds T
and u`
'
should
be known, although this is a problem. As long as the denition of the
effective conductivity coefcient is
k*
VT *, k
V(m,T *) k
`
V(s
`
,T *) (jc
u`
'
,
(k
k
`
)
_
1
AD ,
c
`
`
T
ds
, (431)
then the effective conductivity can be calculated subject to known T *, T
,
T
, and u`
'
. At the same time, the important issue is that in DMM-DNM the
assumption of thermal equilibrium has no sense at all as long as the
problem have been already calculated as the two-temperature problem.
To further perform the correct estimation or calculation of effective
characteristics, one needs to know what are those characteristics in terms of
denition and mathematical description or model?
This is the one more place where the DMM-DNM as it is performed now
is in trouble if it does not comply with the same hierarchical theory
derivations and conclusions as the VAT (see also the studies by Travkin et
al. [115] and Travkin and Catton [114, 21]).
As shown earlier, only the requirement of thermal equilibrium warrants
the equality of steady-state and transient effective conductivities in a
two-phase medium.
Consequently, if taken correctly, the two-temperature model will intro-
duce more trouble in treatment and even interpretation of the needed bulk,
averaged temperature (as long as this problem is already known to exist and
is treated in nonlinear and temperature-dependent situations) and the
corresponding effective conductivity coefcient (or coefcients).
1. Thus, comparing the two effective conductivity coefcients (429) and
(431), one can assess the difference in the second term form and
consequently, the value of computed coefcients.
Comparing the expressions for one equilibrium temperature and one
effective weighted temperature, as well as for their effective conductiv-
ity coefcients, one can also observe the great imbalance and inequal-
ity in their denitions and computations.
2. Summarizing application of DMM-DNM approach by Kuwahara and
Nakayama [199], it can be said that it is questionable procedure to
make an assumption of equilibrium temperatures when the problem
was stated and computed as via DNM for two temperatures.
3. In the calculation of the effective coefcients of conductivity stag-
nant thermal conductivity k
; tortuosity molecular diffusion k

''
; and
thermal dispersion k
'`
Kuwahara and Nakayama [199] used a
questionable procedure for calculation of the two last coefcients.
They used one-cell (REV) computation for surface and uctuation tem-
peratures for periodical morphology of the medium, and at the same time
they used the innite REV denition for the effective temperature gradient
for their calculation (assigned in the problem); see the expressions for
calculation of these coefcients, (21)(24) on p. 413. That action means the
mixture of two different scale variables in one expression for effective
characteristics which is incorrect by denition. If this is used consciously,
the fact should be stated on that matter explicitly, because it alters the
results.
IX. VAT-Based Compact Heat Exchanger Design and Optimization
At the present time, compact heat exchanger (CHE) design is based
primarily on utilization of known standard heat exchanger calculation
procedures (see, for example, Kays and London [172]). Typical analysis of
a heat exchanger design depends on the simple heat balance equations that
are widely used in the process equipment industry. Analytically based
models are composed for a properly constructed set of formulas for a given
spatial design of heat transfer elements that allow, most of the existing heat
transfer mechanisms to be accounted for.
Analogies between heat transfer and friction have been shown by Church-
ill [200] and by Churchill and Chan [201] to be inadequate for describing
many of the HE congurations of interest. This has been suspected for some
time and will seriously affect the use of the j-factor in HE modeling and
design.
Modeling of a specic heat exchanger geometry by Tsay and Weinbaum
[202] provides a useful preliminary step and a potential benchmark test
case. Though the study only considered hydrodynamic effects and restricted
itself to consideration of regular media and the creeping ow regime, the
effects of morphology-characteristic variation upon momentum transport
phenomena were explored. The authors show that the overall bed drag
coefcient in the creep ow regime increases dramatically as the inner-
cylinder spacing approaches the order of the channel half-height.
Analysis of processes in regular and randomly organized heterogeneous
media and CHE can be performed in different ways. Some CHE structures
have the characteristics of a porous medium and can be studied by
application of the developments of porous media modeling. In this work, a
theoretical basis for employing heat and momentum transport equations
obtained from volume averaging theory (VAT) is developed for modeling
and design of heat exchangers. Using different ow regime transport models,
equation sets are obtained for momentum transport and two- and three-
temperature transfer in nonisotropic heterogeneous CHE media with ac-
counting for interphase exchange and microroughness.
The development of new optimization problems based on the VAT-
formulated CHE models using a dual optimization approach is suggested.
Dual optimization is the optimization of the morphological parameters
(size, morphology of working spaces) and the thermophysical properties
(characteristics) of the working solid and liquid materials to maximize heat
transfer while minimizing pressure loss. This allows heat exchanger
modeling and possible optimization to be based on theoretically correct eld
equations rather than the usual balance equations. The problems of shape
optimization traditionally have been addressed in HE design on the basis of
general statements that include heat and momentum equations along with
their boundary conditions stated on the assigned known volumes and
surfaces; see, for example, Bejan and Morega [203].
A. A Sno1 Rrvirv or CirN1 P:c1icr
iN Hr:1 Excn:Ncr MonriiNc
Analysis of heat exchanger designs, as described by Butterworth [204],
depends on the heat balance equations that are widely used in the heat
design industry. The general form of the thermal design equation for heat
exchangers (see, for example, Figs. 79) can be written (Butterworth [204])
dQ: dAAT,
where Q is the heat rate, and A is the transfer surface area. As outlined by
Martin [205], the coupled differential equations for a cross ow tube heat
exchanger (Fig. 7) modeling are (for simplicity only one row is considered)
dG
G'
`
dG
`
d
`
G
G
`
,
where G
, G
`
, and G'
`
are dimensionless rst and second uid temperatures
Fic. 7. Three-phase tube heat exchanger unconsolidated morphology.
Fic. 8. CHE morphology with separated subchannels for each of the uids.
and the second temperature being averaged over the tubes row width. As
follows from these equations, all information about a given heat exchangers
peculiarities and design specics is included in the dimensionless coordinates
:A
(Mc
)
'
z
'
L
'
, i 1, 2,
where : is the overall heat transfer coefcient and M is the mass ow rate.
Second-order ordinary differential equations are developed for HE as well
(see, for example, Paffenbarger [206]).
Fic. 9. Compact heat exchanger (CHE) with contracted-tube layer morphology for one of
the uids.
Webb in a book [207], and in his invited talk at the 10th International
Heat Transfer Conference [208], distinguishes four basic approaches to
predicting the heat transfer j-factor and the Fanning friction factor f for
heat exchanger design. They are (1) power-law correlations; (2) asymptotic
correlations; (3) analytically based models; and (4) numerical solutions.
Analytically based models are properly constructed set of formulas for a
given spatial construction of heat transfer elements that allows most of the
existing heat transfer mechanisms to be accounted for. Many examples are
given in publications by Webb [207, 208], Bergles [209], and other
researchers.
The major differences between the measured characteristics of air-cooled
heat exchangers with aluminum or copper nned tubes with large height,
small thickness, and narrow-pitch ns, and high-temperature waste heat
recovery exchangers with steel nned tubes with rather low height and
thickness and wide-pitched ns, are given in a paper by Fukagawa et al.
[210] Despite the fact that morphology of the heat exchange medium is
essentially the same, the correlations predicting heat transfer and pressure
drop values do not work for both HE types altogether. For this particular
heat exchange morphology, a wide-ranging experiment program is needed
for different ratios of the morphology parameters. There is, at present, no
general approach for describing the dependencies of heat transfer effective-
ness or frictional losses for a reasonably wide range of morphological
properties and their ratios.
The eld of compact heat exchangers has received special attention during
the past several years. A wide variety of plate n heat exchangers (PFHE)
has been developed for applications in heat recovery systems, seawater
evaporators, condensers for heat pumps, etc. It is proposed that a theoretical
Fic. 10. Initial optimization scheme for benchmark tube heat exchanger morphology.
basis for employing heat and momentum transport equations obtained with
volume averaging theory be developed for the design of heat exchangers.
An assumption of the equilibrium streams is common in HE design (see,
for example, Butterworth [204]). Almost all commercial design software
assume plug ow with occasional simple corrections to reect deviations
from the plug ow. CFD has applications in simplied situations, when the
geometry of the channels or heat transfer surfaces can be described fairly.
Butterworth [204] further noted that the space outside tubes in heat
exchangers presents an enormously complicated geometry and modeling
these exchangers fully, even with simplied turbulence models just men-
tioned is still impracticable. We do not agree with this view and propose
to use techniques developed as part of our work to show that practical
modeling methods exist.
During the past few years considerable attention has been given to the
problem of active control of uid ows. This interest is motivated by a
number of potential applications in areas such as control of ow separation,
combustion, uidstructure interaction, and supermaneuverable aircraft. In
this direction, Burns et al. [211, 212] developed several computational
algorithms for active control design for the Burgers equation, a simple model
for convectiondiffusion phenomena such as shock waves and trafc ows.
Generally, the optimal control problems with partial differential equa-
tions (PDE), to which VAT-based HE models convert, can have detailed
solutions of the linear quadratic regulator problem, including conditions for
the convergence of modal approximation schemes. However, for more
general optimal control problems involving PDEs, the main approach has
been to use some method for constructing a particular nite-dimensional
approximating optimal control problem and then to solve this problem by
some method or other (Teo and Wu [213]).
It seems that no attention has been given to the optimal control systems
governed by the partial integrodifferential equations like volume averaging
theory equations for HE design.
B. Nrv KiNns or Hr:1 Excn:Ncr M:1nrx:1ic:i Monris
Our earlier work has shown that ow resistance and heat transfer in HEs
and CHEs can be treated as highly porous structures and that their behavior
can be properly predicted by averaging the transport equations over a
representative elementary volume (REV) in the region neighboring the surface.
The averaging of processes in regular and randomly organized heterogeneous
media and in HE can be performed in different ways. Travkin and Catton [21,
28] discussed alternate forms for the mass, momentum, and heat transport
equations recently presented by various researchers. The alternate forms of the
transport equations are often quite different. The differences among the
transport equation forms advocated by the numerous authors demonstrate the
fact that research on the basic form of the governing equations of transport
processes in heterogeneous media is still an evolving eld of study. Derivation
of the equations of ow and heat transport for a highly porous medium during
the ltration mode is based on the theory of averaging by certain REV of the
transfer equation in the liquid phase and transfer equations in the solid phase
of the heterogeneous medium (see, for example, Whitaker [42, 10] for laminar
regime developments, and Shcherban et al. [15], Primak et al. [14], and
Travkin and Catton [16, 21, 23] for turbulent ltration).
These models account for the medium morphology characteristics. Using
second-order turbulent models, equation sets are obtained for turbulent
ltration and two-temperature diffusion in nonisotropic porous media with
interphase exchange and micro-roughness. The equations differ from those
found in the literature. They were developed using an advanced averaging
technique, a hierarchical modeling methodology, and fully turbulent models
with Reynolds stresses and uxes in every pore space.
Independent treatment of turbulent energy transport in the uid phase
and energy transport in the solid phase, connected through the specic
surface (the soliduid interface in the REV), allows for more accurate
modeling of the heat transfer mechanisms between rough surfaces or porous
insert of HE and the uid phases.
C. VAT-B:srn Coxi:c1 Hr:1 Excn:Ncr MonriiNc
For a pin n (PFHE), with cross-ow morphology, the governing
equations can be written in the following form:
Momentum equation for the rst uid:
c
cx
m
,(K
"
v
)
cU
cx
c
cx
K
"
cu
`
cx
c
cx
(u
`
u
`
)
m
, U
cx

1
AD ,
c
`
U`
(K
"
v
)
cU
cx
'
ds
1
j
AD ,
c
`
U`
p
ds
1
j
c
cx
(m
,p
`
). (432)
momentum equation for the second uid:
c
cz
m
`
,(K
"
v
`
)
cW
`
cz
c
cz
K
"`
cw
`
`
cz
c
cz
(w
`
`
w
`
`
,
)
m
`
, W
`
cW
`
cz

1
AD ,
c
`
U
(K
"`
v
`
)
cW
`
cx
'
ds
1
j
`
AD ,
c
`
U
p
`
ds
1
j
`
c
cz
(m
`
,p
`
`
). (433)
Energy equation for the rst uid:
c
,U
cT
cx

c
cx _
m
,(K
'
k
)
cT
cx
c
cz _
m
,(K
'
k
)
cT
cz
c
cx
K
'
cT
cx
c
cz
K
'
cT
cz
c
cx
(m
,T
u
`
c
cx _
(K
'
k
)
AD ,
c
`
U`
T
ds
c
cz _
(K
'
k
)
AD ,
c
`
U`
T
ds
1
AD ,
c
`
U`
(K
'
k
)
cT
cx
'
ds. (434)
Energy equation for the solid phase:
c
cx
s,K
`'
}
`
cT
`
}
`
cx
c
cx
K
`'
cT
`
cx
`
c
cz
s,K
`'
}
`
cT
`
}
`
cx
c
cz
K
`'
cT
`
cx
`
c
cx _
K
`'
}
`
AD ,
c
`
U`
T
`
ds
c
cz _
K
`'
}
`
AD ,
c
`
U`
T
`
ds
1
AD ,
c
`
U`
K
`'
cT
`
cx
'
ds
0. (435)
Energy equation for the second uid:
c
`
j
`
m
`
,W
`
cT
`
cz

c
cx _
m
`
,(K
'`
k
`
)
cT
`
cx
c
cz _
m
`
,(K
'`
k
`
)
cT
`
cz
c
cx
K
'`
cT
`
cx
c
cz
K
'`
cT
`
cz
c
`
j
`
c
cz
(m
`
,T
`
w
`
`
}
c
cx _
(K
'`
k
`
)
AD ,
c
`
U
T
`
ds
c
cz _
(K
'`
k
`
)
AD ,
c
`
U
T
`
ds
1
AD ,
c
`
U
(K
'`
k
`
)
cT
`
cx
'
ds. (436)
The volumes for averaging in equations are AD, AD
, AD
`
, AD
`
.
A majority of the additional terms in these equations can be treated using
closure procedures developed in previous work (see, for example, Travkin
and Catton [16, 19]), for selected n geometries and solid matrices of a HE.
Our generic interest, however, is in the theoretical applications of the VAT
governing equations and possible advantages gained by introduction of
irregular or random morphology into heat exchange volumes and surfaces.
Cocurrent parallel ow matrix type CHE morphology can be described
using the next VAT-based set of governing equation.
Momentum equation for the rst uid:
m
,U
cU
cx

1
AD ,
c
`
U`
(K
"
v
)
cU
cx
'
ds
1
j
AD ,
c
`
U`
p
ds
1
j
c
cx
(m
,p
`
)
c
cx
(m
,(K
"
v
)
cU
cz
c
cx
K
"
cu
`
cx
c
cx
( u
`
u
`
). (438)
Momentum equation for the second uid:
m
`
,U
`
cU
`
cx

1
AD ,
c
`
U
(K
"`
v
`
)
cU
`
cx
'
ds
1
j
`
AD ,
c
`
U
p
`
ds
1
j
`
c
cx
(m
`
,p
`
`
)
c
cx
(m
`
,(K
"`
v
`
)
cU
`
cx
c
cx
K
"`
cu
`
`
cx
c
cx
( u
`
`
u
`
`
,
). (437)
The corresponding energy equations are like those given earlier. A simple
example typies the general morphology of cocurrent and countercurrent
CHEs when widths of the channels are different and the heat transfer
enhancing devices are to be determined by shape optimization. For this
purpose, consider two conjugate at channels of different heights that are
both lled with unknown (or assigned) heat transfer elements or porous
media. A set of governing equations for each of the channels were developed
by Travkin and Catton ([16, 20]).
A model of the momentum equation for a horizontally homogeneous
stream under steady conditions has the form
c
cz
m,(K
"'
v
'
)
cU
'
cz
c
cz
K
"'
cu
`
'
cz
c
cz
(u
`
'
w
`
'
,
1
AD ,
cS
"''
K
"'
cU
'
cx
'
dS
1
AD ,
cS
"''
v
'
cU
'
cx
'
dS
1
j
'
AD ,
c
`
UH
p
'
dS
1
j
'
cp ,
'
cx
. (439)
This equation can be further simplied for turbulent ow in a layer with a
porous lling or insert that has regular morphology,
c
cz
m(z),(K
"'
v
'
)
cU
'
(z)
cz
U
'`'
(U
'
, cS
"
, K
"'
) U
'`'
(U
'
, cS
"
, v
'
)
U
'`'"
( p
'
, cS
"
)
1
j
'
c(m(z),p
`
'
)
cx
, (440)
where the three morphology-based terms are dened by
U
'`'
(U
'
, cS
"
, K
"'
)
1
AD ,
cS
"''
K
"'
cU
'
cx
'
dS (441)
U
'`'
(U
'
, cS
"
, v
'
)
1
AD ,
cS
"''
v
'
cU
'
cx
'
dS (442)
U
'`'"
(p
'
, cS
"
)
1
j
'
AD ,
c
`
UH
p
'
dS. (443)
It is obvious that the result is controlled by three morphology terms.
The equation for the mean turbulent uctuation energy b(z) is written in
the following simple form, which includes the effect of obstacles in the ow
and temperature stratication across the layer, the z direction:
K
"'
(z)
cU
'
cz
`
d
dz
K
"'
o
'
v
'
db
'
(z)
dz
f
'
(c
)S
"'
(z)
m,
U
`
'
g
T
o
'
_
K
"'
cT
'
cz
2v
db`
'
(z)
dz
`
C
b`
'
(z)
K
"'
. (444)
Here, f
(c
) is approximately the friction factor for constant and nearly

constant morphology functions, and the mean eddy viscosity is given by
K
"'
(z) C"
l(z)b`
'
(z), (445)
where l(z) is the turbulent scale function dened by the assumed porous
medium structure.
Similarly, the equation of turbulent heat transfer in the homogeneous
porous medium uid phase is
c
'
j
'
m
'
,U
'
(z)
cT
'
(x, z)
cx

c
cz
m
'
,(K
''
k
'
)
cT
'
(x, z)
cz
T
'`''
(T
'
, cS
"
, K
''
) T
'`''
(T
'
, cS
"
, k
'
)
1
AD ,
cS
"''
k
cT
'
cx
'
dS, (446)
with two morphology terms that control the solution being
T
'`''
(T
'
, cS
"
, K
''
)
1
AD ,
cS
"''
K
''
cT
'
cx
'
dS (447)
T
'`''
(T
'
, cS
"
, k
'
)
1
AD ,
cS
"''
k
'
cT
'
cx
'
dS. (448)
In the solid phase of CHE, the energy equation is
c
cz
(1 m,)K
`'
(z)
cT
`
(x, z)
cz
T
``'
(T
`
, cS
"
, K
`'
) 0, (449)
with the one control term
T
``'
(T
`
, cS
"
, K
`'
)
1
AD ,
c
`
U`
K
`'
cT
`
cx
'
dS,
where
dS
dS.
If we apply the closure procedures described earlier, the equation of
motion becomes
c
cz
m(z),K
"'
(U
, b, l )
cU
'
(z)
cz
1
2
[c
'
(z, U
'
)S
"'
(z) c`
(z, U
'
)S
"'
(z) c
(z, U
'
)S
"'
(z)]U
`
'
1
j
dp
'
,
dx
c
S
"
U
`
'
2

1
j
dp
'
,
dx
, (450)
where
K
"'
K
"'
v
'
,
and the lumped ow resistance coefcient c
is the complex morphology

dependent function. The energy equation in the j th uid phase is
c
'
j
'
m,U
'
(z)
cT
'
(x, z)
cx

c
cz
(m(z),K
''
(z)
cT
'
(x, z)
cz
:`
'
(z)S
"
(z)(T
`
(x, z) T
'
(x, z)), (451)
with (x, z) - AD
, and the energy equation in the solid phase

c
cz
(1 m(z),K
`'
(z)
cT
`
(x, z)
cz
:`
'
(z)S
"
(z)(T
`
(x, z) T
'
(x, z))(x, z) - AD, (452)
with
P
''
1 : K
''
K
"'
c
'
j
'
k
'
, (453)
where index j determines the uid phase number j 1, 2 in conjugate
channels 1 and 2.
In Eqs. (444), (445), (450), and (452), the coefcient functions and specic
surface functions must be determined by assuming real or invented mor-
phological models of the porous structure. The pressure gradient term in Eq.
(450) is modeled as a constant value in the layer, or simulated by the local
value of the right-hand side of the experimental correlations. The boundary
conditions for these equations are
z 0 : U
'
0,
cb
'
cz
0
K
"
v, Q
"
K
''
cT
'
cz
Q
"
K
`'
cT
`
cz
(454)
z
h
'
:
cU
'
cz
0,
cb
'
cz
0
cT
'
cz
0,
cT
`
cz
0, (455)
where h
'
is the half channel width. The control terms in the preceding
equations depend on temperature and velocity distributions as well as on
morphological characteristics of the media.
Comparing the three latest equation (450)(452) with the equations
derived by Paffenbarger [206] for practically the same structural design of
HE, one will nd numerous discrepancies. For example, the energy balance
equations in Paffenbargers [206] work have energy conservation terms that
do not match each other.
The VAT-based general transport equations for a single phase uid in an
HE medium have more integral and differential terms than the homogenized
or classical continuum mechanics equations. Various descriptions of the
porous medium structural morphology determines the importance of these
terms and the range of application of the closure schemes. Prescribing
regular, assigned, or statistical structure to the capillary or globular HE
medium morphology gives the basis for transforming the integrodifferential
transport equations into differential equations with probability density
functions governing their stochastic coefcients and source terms. Several
different closure models for these terms for some uniform, nonuniform,
nonisotropic, and specically random nonisotropic highly porous layers
were developed in work by Travkin and Catton [16, 17, 23], etc. The natural
way to close the integral terms in the transfer equations is to attempt to nd
the integrals over the interphase surface, or over outlined areas of this
surface. Closure models allow one to nd connections between experimental
correlations for bulk processes and the simulation representation and then
incorporate them into numerical procedures.
D. Oi1ix:i CoN1oi Ponirxs iN Hr:1 Excn:Ncr DrsicN
A variety of the optimization problems that can be formulated in the area
of heterogeneous medium transport involve differential equations modeling
the physics of the process. Many of them have a fairly complicated form.
The contemporary literature on optimal control deals with problems that
are mathematically similar but consider much simpler formulations of the
optimization problem with constraints in the form of differential equations.
Linear optimal control systems governed by parabolic partial differential
equations (PPDEs) are relatively well studied. The CHE modeling equa-
tions resulting from the VAT-based analysis are also PPDEs, but they are
nonlinear and have additional integral and integrodifferential terms. The
models presented and the resulting differential equations contain additional
integral and integrodifferential terms not studied in the literature.
The performance of a heat exchanger depends on the design criteria for
optimizing the liquid ow velocity, dimensions of the heat exchanger, the
heat transfer area between hot side and cold side, etc. Thermal optimization
of an HE requires selection of many features for example, both the
optimum n spacing and optimum n thickness, each determined to
maximize total heat dissipation for a given added mass or prole area. These
criteria set the optimal conditions for HE operation. Theoretically, the
optimal dimensions of an HE require a large number of tiny tubelets with
diameters tending to zero with increasing number of tubes. This leads to a
very ne dispersion problem with porous mediumlike behavior. Extremely
compact micro heat exchangers with platen cross ow have already been
built. However, the optimization problems involving such designs are more
complex than traditional designs and require new simulation techniques.
E. A VAT-B:srn Oi1ixiz:1ioN TrcnNiqir ro Hr:1 Excn:Ncrs
A variety of optimal control problems that can be formulated in the area
the physics of the process. Some of them have a fairly complicated form.
Meanwhile, the contemporary literature on optimal control considers too
simple formulations of the optimization problems with constraints in form
of differential equations.
Optimal control systems governed by parabolic partial differential equa-
tions have been studied intensively. For example, Ahmed and Teo [214] give
a survey on main results in this eld. Questions concerning necessary
conditions for optimality and existence of optimal controls for these
problems have been investigated in work by Ahmed and Teo (215217] and
Fleming [218]. Moreover, a few results by Teo et al. (1980) on the
computational methods of nding optimal controls are also available in the
literature (Teo and Wu [213]). However, turbulent transport equations in
highly porous media were proposed by Travkin et al. [19] for optimization
problems and developed in more detail in Section IV with additional
morphlogical as well as integral and integrodifferential terms. Recent
literature studies show optimal control problems involving PPDE either in
general form or in divergence form and propose computational methods
such as variational technique and gradient method (see, for example, Ahmed
and Teo [214]). These studies seems to be helpful for solving various
optimization problems involving integrodifferential transport equations
considered by Travkin et al. [19]. However, complete research has to be
done for this class of equations, including analysis of necessary conditions
and existence of optimal control, as well as developing computational
methods for solving various optimal control problems.
Optimal control for some classes of integrodifferential equations has also
been considered in recent years. Da Prato and Ichikawa [219] studied the
quadratic control problems for integrodifferential equations of parabolic
type. A state-space representation of the system is obtained by choosing an
appropriate product space. By using the standard method based on the
Riccati equation, a unique optimal control over a nite horizon and under
a stabilizability condition is obtained and the quadratic problem over an
innite horizon is solved. Butkovski [220] was the rst to discuss the
optimal control problems for distributed parameter systems. The maximum
principle as a set of necessary conditions for optimal control of distributed
parameter systems has been studied by many authors.
Since it is well known that the maximum principle may be false for
distributed parameter systems (see Balakrishnan [221]), there are many
papers that give some conditions ensuring that the maximum principle
remains true (see, for example, Ahmed and Teo [214]; Balakrishnan [221];
Curtain and Pritchard [222]). We note that the references just mentioned
discuss the cases for distributed parameter systems or functional differential
systems with no end constraints and/or with the control domain being
convex; thus, they do not include Pontryagins original result on maximum
principle as a special case.
Fattorini [223] also proposed an existence theory and formulated maxi-
mum principle for relaxed innite-dimensional optimal control problems.
He considered relaxed optimal control problems described by semilinear
systems ODE and used relaxed controls whose values are nitely additive
probability measures. Under suitable conditions, relaxed trajectories co-
incide with those obtained from differential inclusions. The existence the-
orems for relaxed controls were obtained; they are applied to distributed
parameter systems described by semilinear parabolic and wave equations, as
well as a version of Pontryagins maximum principle for relaxed optimal
control problems.
Optimal control problems involving equations such as (432)(438) have
control terms with the structures
V(m, f
(x)V f
`
(x)}
)
V(m, f
`
(x) f
`
(x)}
)
V
(x)
,
c
`
U
f
`
(x) ds
,
c
`
U
[
`
(x)V( f
"
(x, f
`
(x)))] ds, (456)
with controls f
, f
`
, f
`
, f
"
. Such statements of the control problem are hardly
seen in the contemporary literature on optimal control distributed-par-
ameter systems (see, for example, Ahmed and Teo [214]). The existence of
optimal controls for equations much simpler than those here were developed
only very recently; see Fattorini [223]. Thus, for linear heat- and mass-
diffusion problems with impulse control that is a function of magnitude or
spatial locations of the impulses, Anita [224] obtained a formulation of
maximum principles for both optimal problems. Ahmed and Xiang [225]
proved the existence of optimal controls for clear nonlinear evolution
equations on Banach spaces with the control term in the equations being
represented as an additivemultiplicative term B(t)u(t).
Reduction of hererogeneous terms in the corresponding momentum
equation by an overall representation of diffusive and diffusionlike terms
yields
k
"
cA
cx

_
m,(K
"
v)
cA
cx

K
"
ca
`
cx
a
`
a
`
,
. (457)
Here, the velocity and uctuating viscosity coefcient variables are taken in
a form suitable for both laminar and turbulent ow regimes. For problems
with a constant bulk viscosity coefcient (K
"
constant), the second term
in this relation vanishes and the whole problem essentially becomes one of
evaluating the inuence of dispersion by irregularities of the soil medium on
the momentum. Thermal dispersion effects realized through the second
derivative terms and relaxation terms and, for example, in the uid phase
with constant thermal characteristics heat transport dispersion can be
expressed as
K
'
cT
cx
_
m,(K
'
k
)
cT
cx
K
'
cT
cx
m,T
u
`
}
(K
'
k
)
AD ,
c
`
U
T
ds
, (458)
where the rst and last terms resemble the effective thermal conductivity
coefcient for each phase, using constant coefcients, found in the work by
Nozad et al. [40]. By allowing the control terms to be added to the bulk
transport coefcients, another variation of a mathematical statement for
optimal control can be found.
As far as optimal control problems with PDE dynamics are concerned,
one can nd a detailed solution of the linear quadratic regulator problem,
including conditions for the convergence of modal approximation schemes.
However, for more general optimal control problems involving PDE, the
main approach has been to use some method for constructing a particular
nite-dimensional approximating optimal control problem and then to
solve this problem. The relationship between the solutions and stationary
points of the approximating optimal control problem and those of the
original optimal control problem is not established in these papers.
For the models and differential equations describing HEs to be useful, the
additional integral and integrodifferential terms need to be addressed in a
systematic way. VAT has the unique ability to enable the combination of
direct general physical and mathematical problem statement analysis with
the convenience of the segmented analysis usually employed in HE design.
A segmented approach is a method where overall physical processes or
groups of phenomena are divided into selected subprocesses or phenomena
that are interconnected to others by an adopted chain or set of depend-
encies. A few of the obvious steps that need to be taken are the following:
1. Model what increases the heat transfer rate
2. Model what decreases of ow resistance (pressure drop)
3. Combine the transport (thermal/mass transfer) analysis and structural
analysis (spatial) and design
4. Find the minimum volume (the combination of parameters yielding a
minimum weight HE)
5. Include nonlinear conditions and nonlinear physical characteristics
into analysis and design procedures
The power and convenience of this method is clear, but its credibility is
greatly undermined by variability and freedom of choice in selection of
subportions of the whole system or process. The greatest weakness is that
the whole process of phenomena described by a voluntarily assigned set of
rules for the description of each segment is sometimes done without serious
consideration of the implications of such segmentation. Strict physical
analysis and consideration of the consequences of segmentation is not
possible without a strict formulation of the problem that the VAT-based
modeling supplies. Structural optimization of a plate HE, for example, using
the VAT approach might consist of the following steps: (1) optimization of
the number of plates, plate spacing and n spacing; (2) optimization of the
n shape; (3) simultaneous optimization of multiple mathematical state-
ments. This approach also allows consideration and description of hydraul-
ically and thermally developing processes by representing them through the
distributed partial differential systems.
X. New Optimization Technique for Material Design Based on VAT
the physics of the process. Many of them have a fairly complicated form,
and the contemporary literature on optimal control considers much simpler
formulations of the optimization problems with constraints in form of
differential equations.
When the diffusion equations are written in nonlocal VAT form, there are
additional terms appearing in the mathematical statements. These terms can
be considered to be morphology controls involving differential and integral
operators. The nonlinear diffusion equation written without source terms
has three control terms,
s
,
cC
ct
V (D
Vs
,C
) V
_
D
1
AD ,
cS
"
C
ds
V (D
Vc`
)
1
AD ,
cS
"
D
VC
ds
V (D
Vs
,C
) F
'
(C
, D
, M
) F
'`
(c`
, D
, M
)
F
'`
(C
, D
, M
), (459)
where the morphology characteristics set M
contains many parameters, c
,
such as phase fraction s
, and specic surface area cS

`
,
M
(s
,, cS
"
, c
`
, c
`
, . . . ).
The equation for an electrostatic electrical eld in a particulate medium
(polycrystalline medium) is
V [s
,c`
] V c`
1
AD ,
c
`
`
(c
) ds
j,
,
which becomes
V [s
,c`
] F
'
(c`
, E
, M
) F
'`
(c
, E
, M
) j,
. (460)
Additional equations are
V(s
,E
)
1
AD ,
c
`
`
ds
0
V(s
,E
) F
'`
(E
, M
) 0. (461)
A temperature control equation for the solid phase with the two morphol-
ogy control terms can be written
cT
"
ct
a
"
c`T
"
cz`
T
"`'
(T
"
, cS
`
, t, z) T
"`'
(T
"
, cS
"
, t, z), (462)
where
T
"`'
a
"
c
cz _
1
AD
"
,
c
`
`
T
"
ds
, T
"`'
a
"
AD
"
,
c
`
`
cT
"
cx
'
ds
, (463)
and in the void phase
cT
`
}
`
ct
a
`
c`T
`
}
`
cz`
T
``'
(T
`
, t, z) T
``'
(T
`
, t, z)
T
``'
(T
`
, cS
`
, t, z) a
`
c
cz _
1
AD
`
,
c
`
`
T
`
ds
`
(464)
T
``'
(T
`
, cS
`
, t, z)
a
`
AD
`
,
c
`
`
VT
`
ds
`
. (465)
These terms are not equal and their calculation or estimation presents a
challenge. However, these are the real driving forces that will differentiate
the behavior of one composite from another. Their application will lead to
a direct connection between design goals and morphological solutions.
XI. Concluding Remarks
Determination of the effective parameters in model equations are usually
based on a medium morphology model and there are dozens of associated
quasi-homogeneous and quasi-stochastic methods that claim to accomplish
this. In most cases, quasi-homogeneous and quasi-stochastic methods have
no well treated solutions and, most important, they are not sufcient for
description of the physical process features in heterogeneous media, especial-
ly when treating a multiscale processes.
The hierarchical approach applied to radiative transfer in a porous
medium and to the electrodynamics governing equations (Maxwells equa-
tions) in a heterogeneous medium yielded new volume averaged radiative
transfer equations VAREs. These equations have additional terms reect-
ing the inuence of interfaces and inhomogeneities on radiation intensity in
a porous medium and, when solved, will allow one to relate the lower scale
parameters to the upper scale material behavior. The general nature of this
result makes it applicable to any subatomic particle transport, including
neutron transport, as well as radiative transport in the heterogeneous media
eld. Direct closure based on theoretical and numerical developments that
have been developed for thermal, momentum, and mass transport processes
in a specic random porous and composite medium established a basis for
closure modeling in problems in radiative and electromagnetic phenomena.
In this work, transport models and equation sets were obtained for a
number of different circumstances with a well substantiated mathematical
theory called volume averaging theory (VAT) that included linear, non-
linear, laminar, and turbulent hierarchical transport in nonisotropic hetero-
geneous media, accounting for modeling level, interphase exchange, and
microroughness. Models were developed, for example, for porous media
using an advanced averaging technique, a hierarchical modeling methodol-
ogy, and fully turbulent models with Reynolds stresses and uxes. It is worth
noting that nonlocal mathematical modeling is very different from hom-
ogenization modeling. The new integrodifferential transport statements in
heterogeneous media and application of these nonclassical types of equa-
tions is the current issue. The theory allows one to take into consideration
characteristics of multicomponent multiphase composites with perfect as
well as imperfect morphologies and interphases. The transport equations
obtained using VAT involved additional terms that quantify the inuence of
the medium morphology. Various descriptions of the porous medium
structural morphology determine the importance of these terms and the
range of application of closure schemes.
Many mathematical models currently in use have not received a critical
review because there was nothing to review them against. The more
common models were compared with the more rigorous VAT-based models
and found decient in many respects. This does not mean they do not serve
a useful purpose. Rather, they are incomplete and suffer from lack of
generality.
VAT-based modeling is very powerful, allowing random morphology
uctuations to be incorporated into the VAT-based transport equations by
means of randomly varying morphoconvective and morphodiffusive terms.
Closure of some of the resulting morphouctuation in the governing
transport equations has been outlined, resulting in some well-developed
closure expressions for the VAT-based transport equations in porous media.
Some of them exploit the properties of available solutions to transport
problems for individual morphological elements, and others are based on
the natural morphological data of porous media.
Statistical and numerical techniques were applied to classical irregular
morphologies to treat the morphodiffusive and morphoconvective terms
along with integral terms. The challenging problem in irregular and random
morphologies is to produce an analytical or numerical evaluation of the
deviations in scalar or vector elds. In previous work, the authors have
presented a few exact closures for predetermined regular and random
porous medium morphologies. The questions related to effective coefcient
dependencies, boundary conditions, and porous medium experiment analy-
sis are discussed.
Analysis of heat exchanger designs depends on the heat balance equations
that are widely used in the heat design industry. A theoretical basis for
employing heat and momentum transport equations obtained with volume
averaging theory was developed for modeling and design of heat exchangers.
This application of VAT results in a correct set of mathematical equations
for heat exchanger modeling and optimization through implementation of
general eld equations rather than the usual balance equations. The
performance of a heat exchanger depends on the design criteria for optimiz-
ing the liquid ow velocity, dimensions of the heat exchanger, the heat
transfer area between the hot side and cold side, etc. However, the optimiz-
ation problems involving such designs are more complex than for tradi-
tional designs and require new optimal control simulation techniques.
the physics of the process. Many of them have a fairly complicated form,
and the contemporary literature on optimal control considers much simpler
formulations of the optimization problems with constraints in the form of
differential equations. Linear optimal control systems governed by parabolic
partial differential equations (PDEs) are relatively well studied in the
literature. The modeling CHE equations resulting from VAT-based analysis
are also PDEs, but they are nonlinear and have additional integral and
integrodifferential terms.
It is well known that some matrix composites (often porous) represent the
promise for design of a series of materials with highly desirable characteris-
tics such as high temperature accommodation and enhanced toughness.
Their performance is very dependent on the volume fraction of the consti-
tuent materials, reinforcement interface and matrix morphologies, and
consolidation. Scale characteristics (nanostructural composites) give the
abnormal physical properties, such as magnetic, and mechanical transport
and state a great challenge in formulating the hierarchical models contain-
ing the design objectives.
The importance of the physical processes taking place in a heterogeneous
multiscalemultiphasecomposite medium creates the need for the develop-
ment of new tools to characterize such media. It leads to the development
of new approaches to describing these processes. One of them (VAT) has
great advantages and is the subject of this review.
Acknowledgments
This work was partly sponsored by the Department of Energy, Ofce of Basic Energy
Sciences, through the grant DE-FG03-89ER14033 A002.
Nomenclature
a thermal diffusivity [m`/s] c`
mean skin friction coefcient

c
mean drag resistance coefcient over the turbulent area of

in the REV [-] cS
"
[-]
c
mean form resistance coefcient

in the REV [-]
c
`"
drag resistance coefcient upon
single sphere [-]
c
'
mean skin friction coefcient
over the laminar region inside
of the REV [-]
c
specic heat [J/(kg K)]

C
constant coefcient in
Kolmogorov turbulent
exchange coefcient correlation
[-]
d
'"
character pore size in the cross
section [m]
d
'
diameter of ith pore [m]
d
particle diameter [m]

ds interphase differential area in
porous medium [m`]
D
molecular diffusion coefcient

[m`/s]; also tube or pore
diameter [m]
D
"
at channel hydraulic diameter
[m]
D
`
diffusion coefcient in solid
[m`/s]
cS
"
internal surface in the REV
[m`]
f f }
averaged over AD
value f
intrinsic averaged variable
f ,
value f, averaged over AD
din
an REVphase averaged
variable
f morphouctuation value of f
in a D
g gravitational constant [1/m`]

H width of the channel [m]
h averaged heat transfer
coefcient over cS
"
[W/(m`/K)]; half-width of the
channel [m]
h
'
pore scale microroughness
layer thickness [m]
cS
"
internal surface in the REV
[m`]
k
uid thermal conductivity [W/

(mK)]
k
`
solid phase thermal
conductivity [W/(mK)]
K permeability [m`]
K
'
turbulent kinetic energy
exchange coefcient [m`/s]
K
'
turbulent diffusion coefcient
[m`/s]
K
"
turbulent eddy viscosity [m`/s]
K
`'
effective thermal conductivity
of solid phase [W/(mK)]
K
'
turbulent eddy thermal
conductivity [W/(mK)]
l turbulence mixing length [m]
L scale [m]
m, averaged porosity [-]
m
`
surface porosity [-]
n number of pores [-]
n
'
number of pores with diameter
of type i [-]
Nu
'

h
`
d
"
z
, interface surface
Nusselt number [-]
p pressure [Pa]; or pitch in
regular porous 2D and 3D
medium [m]; or phase function
[-]
Pe
"
Re
"
Pr, Darcy velocity pore
scale Peclet number [-]
Pe
Re
'
Pr, particle radius Peclet
number [-]
Pr
v
a
, Prandtl number [-]

Q
"
outward heat ux [W/m`]
Re
'"
Reynolds number of pore
hydraulic diameter [-]
Re
"

m,u
`
d
"
v
, Darcy velocity
Reynolds number of pore
hydraulic diameter [-]
Re

u
`
d
v
, particle Reynolds
number [-]
Re
'

u
`
d
'
v
, Reynolds number of
general scale pore hydraulic
diameter [-]
S
''
total cross-sectional area
available to ow [m`]
S
"
specic surface of a porous
medium cS
"
/AD [1/m]
S
"
S
/AD [1/m]
S
S
'
cross ow projected area
of obstacles [m`]
T temperature [K]
T
characteristic temperature for

given temperature range [K]
T
`
solid phase temperature [K]
T
"
wall temperature [K]
T
"
reference temperature [K]
U, u velocity in x direction [m/s]
u`
''
square friction velocity at the
upper boundary of HR
averaged over surface cS
"
[m`/s`]
V velocity [m/s]
V
'
u
`
m, Darcy velocity [m/s]
W velocity in z direction [m/s]
Sinscii1s
e effective
f uid phase
i component of turbulent vector
variable; or species or pore type
k component of turbulent
variable that designates
turbulent microeffects on a
pore level
L laminar
m scale value or medium
r roughness
s solid phase
T turbulent
w wall
Siirscii1s
value in uid phase averaged
over the REV
value in solid phase averaged
over the REV
mean turbulent quantity
' turbulent uctuation value
* equilibrium values at the
assigned surface or complex
conjugate variable
Grrx Lr11rs
:`
'
averaged heat transfer
coefcient over cS
"
[W/(m`K)]
AD representative elementary
volume (REV) [m`]
AD
pore volume in a REV [m`]

AD
`
solid phase volume in a REV
[m`]
c
, c
"
electric permittivity [Fr/m]
j dynamic viscosity [kg/(ms)] or
[Pas]
j
"
magnetic permeability [H/m]
v kinematic viscosity [m`/s]; also
v, frequency [Hz]
j density [kg/m`]; also j, electric
charge density [C/m`]
o
medium specic electric

conductivity [A/V/m]
u electric scalar potential [V]
particle intensity per unit
energy (frequency)
} ensemble-averaged value of
`
'
interface ensemble-averaged
value of , with phase j being to
the left
c angular frequency [rad/s]
, magnetic susceptibility [-]
absorption coefcient [1/m]
``
`
scattering coefcient [1/m]
References
1. Anderson, T. B., and Jackson, R. (1967). A uid mechanical description of uidized beds.
Int. Eng. Chem. Fundam. 6, 527538.
2. Slattery, J. C. (1967). Flow of viscoelastic uids through porous media. AIChE J. 13,
10661071.
3. Marle, C. M. (1967). Ecoulements monophasiques en milieu poreux. Rev. Inst. Francais
du Petrole 22, 14711509.
4. Whitaker, S. (1967). Diffusion and dispersion in porous media. AIChE J. 13, 420427.
5. Zolotarev, P. P., and Radushkevich, L. V. (1968). The equations for dynamic sorption in
an undeformed porous medium. Doklady Physical Chemistry 182, 643646.
6. Slattery, J. C. (1980). Momentum, Energy and Mass Transfer in Continua. Krieger, Malabar.
7. Kaviany, M. (1995). Principles of Heat Transfer in Porous Media, 2nd ed. Springer, New
York.
8. Gray, W. G., Leijnse, A., Kolar, R. L., and Blain, C. A. (1993). Mathematical Tools for
Changing Spatial Scales in the Analysis of Physical Systems. CRC Press, Boca Raton, FL.
9. Whitaker, S. (1977). Simultaneous heat, mass and momentum transfer in porous media:
a theory of drying. Advances in Heat Transfer 13, 119203.
10. Whitaker, S. (1997). Volume averaging of transport equations. Chapter 1 in Fluid
Transport in Porous Media (J. P. DuPlessis, ed.). Computational Mechanics Publications,
Southampton.
11. Kheifets, L. I., and Neimark, A. V. (1982). Multiphase Processes in Porous Media. Nadra,
Moscow.
12. Dullien, F. A. L. (1979). Porous Media Fluid Transport and Pore Structure. Academic
Press, New York.
13. Adler, P. M. (1992), Porous Media: Geometry and Transport. ButterworthHeinemann,
Stoneham.
14. Primak, A. V., Shcherban, A. N., and Travkin, V. S. (1986). Turbulent transfer in urban
agglomerations on the basis of experimental statistical models of roughness layer
morphological properties. In Transactions World Meteorological Organization Conference
on Air Pollution Modelling and its Application, 2, pp. 259266. WMO, Geneva.
15. Shcherban, A. N., Primak, A. V., and Travkin, V. S. (1986). Mathematical models of ow
and mass transfer in urban roughness layer. Problemy Kontrolya i Zaschita Atmosfery ot
Zagryazneniya 12, 310 (in Russian).
16. Travkin, V. S., and Catton, I. (1992). Models of turbulent thermal diffusivisty and transfer
coefcients for a regular packed bed of spheres. In Fundamentals of Heat Transfer in
Porous Media (M. Kaviany, ed.), ASME HTD-193, pp. 1523.
17. Travkin, V. S., Catton, I., and Gratton, L. (1993). Single phase turbulent transport in
prescribed non-isotropic and stochastic porous media. In Heat Transfer in Porous Media,
ASME HTD-240, pp. 4348.
18. Travkin, V. S., and Catton, I. (1994). Turbulent transport of momentum, heat and mass
in a two level highly porous media. In Heat Transfer 1994, Proc. Tenth Intern. Heat
Transfer Conf. (G. F. Hewitt, ed.) 5, pp. 399404. Chameleon Press, London.
19. Travkin, V. S., Gratton, L., and Catton, I. (1994). A morphological approach for
two-phase porous medium-transport and optimum design applications in energy engine-
ering. In Proc. 12th Symp. Energy Engin. Sciences, Argonne National Laboratory, Conf.
-9404137, pp. 4855.
20. Travkin, V. S., and Catton, I. (1995). A two-temperature model for turbulent ow and
heat transfer in a porous layer. J. Fluids Eng. 117, 181188.
21. Travkin, V. S., and Catton, I. (1998). Porous media transport descriptions nonlocal,
linear and nonlinear against effective thermal/uid properties. Adv. Colloid Interf. Sci.
76-77, 389443.
22. Travkin, V. S., Hu, K., and Catton, I. (1999). Turbulent kinetic energy and dissipation rate
equation models for momentum transport in porous media. In Proc. 3rd ASME/JSME
Fluids Engineering Conf. FEDSM99-7275, ASME, San Francisco.
23. Travkin, V. S., and Catton, I. (1999). Nonlinear effects in multiple regime transport of
momentum in longitudinal capillary porous medium morphology. To appear in J. Porous
Media.
24. Travkin, V. S., and Catton, I. (1999). Critique of theoretical models of transport
phenomena in heterogeneous media (invited). Presentation at the 3rd ASME/JSME
Fluids Engineering Conf. FEDSM99-7922, July 1823, 1999, San Francisco.
25. Travkin, V. S., Catton, I., and Hu, K. (1998). Channel ow in porous media in the limit
as porosity approaches unity. In Proc. ASME-HTD-361-1, pp. 277284.
26. Gratton, L., Travkin, V. S., and Catton, I. (1995). The impct of morphology irregularity
on bulk ow and two-temperature heat transport in highly porous media. In Proc.
ASME/JSME Thermal Eng. Joint Conf. 3, pp. 339346.
27. Gratton, L., Travkin, V. S., and Catton, I. (1996). The inuence of morphology upon
two-temperature statements for convective transport in porous media. J. Enhanced Heat
Transfer 3, 129145.
28. Catton, I., and Travkin, V. S. (1996). Turbulent ow and heat transfer in high permeability
porous media. In Proc. Intern. Conf. on Porous Media and T heir Applic. Science, Engineer.
and Ind. (K. Vafai and P. N. Shivakumar, eds.), pp. 333391. Engin. Found. & Inst. Ind.
Math. Sc., New York.
29. Quintard, M., and Whitaker, S. (1993). One and two-equation models for transient
diffusion processes in two-phase systems. Advances in Heat Transfer 23, 369465.
30. Quintard, M., and Whitaker, S. (1990). Two-phase ow in heterogeneous porous media I:
The inuence of large spatial and temporal gradients. Transport in Porous Media 5,
341379.
31. Carbonell, R. G., and Whitaker, S. (1984). Heat and mass transport in porous media. In
Fundamentals of Transport Phenomena in Porous Media (J. Bear and M. Y. Corapcioglu,
eds.), pp. 121198. Martinus Nijhof, Boston.
32. Sangani, A. S., and Acrivos, A. (1982). Slow ow through a periodic array of spheres. Int.
J. Multiphase Flow 8, 343360.
33. Travkin, V. S., and Kushch, V. I. (1999a). Averaging theorem theoretical closure and
verication. Submitted.
34. Travkin, V. S., and Kushch, V. I. (1999b). Two-temperature volume averaging equations
exact closure for globular morphology. Submitted.
35. Kushch, V. I. (1991). Heat conduction in a regular composite with transversely isotropic
matrix. Doklady AN Ukr. SSR 1, 2327 (in Russian).
36. Kushch, V. I. (1994). Thermal conductivity of composite material reinforced by period-
ically distributed spheroidal particles. Eng. Phys. Journal 66, 497504 (in Russian).
37. Kushch, V. I. (1996). Elastic equilibrium of a medium containing nite number of aligned
spheroidal inclusions. Int. J. Solids Structures 33, 11751189.
38. Kushch, V. I. (1997). Conductivity of a periodic particle composite with transversely
isotropic phases. Proc. R. Soc. L ond. A 453, 6576.
39. Rayleigh, R. S. (1892). On the inuence of obstacles arranged in rectangular order upon
the properties of a medium. Phil. Mag. 34, 481489.
40. Nozad, I., Carbonell, R. G., and Whitaker, S. (1985). Heat conduction in multi-
phase systems I: Theory and experiment for two-phase systems. Chem. Eng. Sci. 40,
843855.
41. Crapiste, G. H., Rotstein, E., and Whitaker, S. (1986). A general closure scheme for the
method of volume averaging. Chem. Eng. Sci. 41, 227235.
42. Whitaker, S. (1986). Flow in porous media I: A theoretical derivation of Darcys law.
Transport in Porous Media 1, 325.
43. Whitaker, S. (1986). Flow in porous media II: The governing equations for immiscible,
two-phase ow. Transport in Porous Media, 1, 105125.
44. Plumb, O. A., and Whitaker, S. (1990a). Diffusion, adsorption and dispersion in porous
media: small-scale averaging and local volume averaging. In Dynamics of Fluids in
Hierarchical Porous Media (J. H. Cushman, ed.), pp. 97148. Academic Press,
New York.
45. Plumb, O. A., and Whitaker, S. (1990b). Diffusion, adsorption and dispersion in hetero-
geneous porous media: The method of large-scale averaging. In Dynamics of Fluids in
Hierarchical Porous Media (J. H. Cushman, ed.), pp. 149176. Academic Press, New York.
46. Levec, J., and Carbonell, R. G. (1985). Longitudinal and lateral thermal dispersion in
packed beds. Parts I & II. AIChE J. 31, 581602.
47. Gray, W. G. (1975). A derivation of the equations for multiphse transport. Chem. Eng. Sci.
30, 229233.
48. Gray, W. G., and Lee, P. C. Y. (1977). On the theorems for local volume averaging of
multiphase systems. Int. J. Multiphase Flow 3, 333340.
49. Abriola, L. M., and Gray, W. G. (1985). On the explicit incorporation of surface effects
into the multiphase mixture balance laws. Int. J. Multiphase Flow 11, 837852.
50. Gray, W. G., and Hassanizadeh, S. M. (1989). Averaging theorems and averaged
equations for transport of interface properties in multiphase systems. Int. J. Multiphase
Flow 15, 8195.
51. Teyssedou, A., Tapucu, A., and Camarero, R. (1992). Blocked ow subchannel simulation
comparison with single-phase ow data. J. Fluids Eng. 114, 205213.
52. Ishii, M., (1975). T hermo-uid Dynamic Theory of Two-Phase Flow. Eyrolles, Paris.
53. Ishii, M., and Mishima, K. (1984). Two-uid model and hydrodynamic constitutive
relations. Nucl. Eng. Design 82, 107126.
54. Lahey, T., Jr., and Lopez de Bertodano, M. (1991). The prediction of phase distribution
using two-uid models. In Proc. ASME/JSME T hermal Engineering Conf. 2, pp. 192200.
55. Lopez de Bertodano, M., Lee, S-J., Lahey, R. T. Jr., and Drew, D. A. (1990). The
prediction of two-phase turbulence and phase distribution phenomena using a Reynolds
stress model. J. Fluids Eng. 112, 107113.
56. Lahey, R. T., Jr., and Drew, D. A. (1988). The three-dimensional time and volume
averaged conservation equations of two-phase ow. In Advances in Nuclear Science and
Technology (T. Lewins and M. Becker, eds.), 20, pp. 169.
57. Zhang, D. Z., and Prosperetti, A. (1994). Averaged equations for inviscid disperse
two-phase ow. J. Fluid Mech. 267, 185219.
58. Khan, E. U., Rohsenow, W. M., Sonin, A. A., and Todreas, N. E. (1975). A porous body
model for predicting temperature distribution in wire-wrapped rod assemblies operating
in combined forced and free convection. Nucl. Eng. Design 35, 199211.
59. Subbotin, V. I., Kashcheev, V. M., Nomolov, E. V., and Yurev, Yu.S. (1979). Computer
Problem Solving in Nuclear Reactor T hermophysics. Atomizdat, Moscow (in Russian).
60. Popov, A. M. (1974). On peculiarities of atmospheric diffusion over inhomogeneous
surface. Izv. AN SSSR, AOPh. 10, 13091312 (in Russian).
61. Popov, A. M. (1975). Atmospheric boundary layer simulation within the roughness layer.
Izv. AN SSSR, AOPh. 11, 574581 (in Russian).
62. Yamada, T. (1982). A numerical model study of turbulent airow in and above a forest
canopy. J. Meteorol. Soc. Jap. 60, 439454.
63. Raupach, M. R., and Shaw, R. H. (1982). Averaging procedures for ow within vegetation
canopies. Boundary-L ayer Meteorol. 22, 7990.
64. Raupach, M. R., Coppin, P. A., and Legg, B. J. (1986). Experiments on scalar dispersion
within a model plant canopy. Part I: the turbulence structure. Boundary-L ayer Meteorol.
35, 2152.
65. Coppin, P. A., Raupach, M. R., and Legg, B. J. (1986). Experiments on scalar dispersion
within a model plant canopy. Part II: an elevated plane source. Boundary-L ayer Meteorol.
35, 167191.
66. Legg, B. J., Raupach, M. R., and Coppin, P. A. (1986). Experiments on scalar dispersion
within a model plant canopy. Part III: an elevated line source. Boundary-L ayer Meteorol.
35, 277302.
67. Fand, R. M., Kim, B. Y. K., Lam, A. C. C., and Phan, R. T. (1987). Resistance to the ow
of uids through simple and complex porous media whose matrices are composed of
randomly packed spheres. J. Fluids Eng. 109, 268274.
68. Dybbs, A., Edwards, R. V. (1982). A new look at porous media uid mechanics Darcy
to turbulent. In Proc. NATO Advanced Study Institution on Mechanics of Fluids in Porous
Media, NATO ASI Series E 82, pp. 201256.
69. Masuoka, T., and Takatsu, Y. (1996). Turbulence model for ow through porous media.
Int. J. Heat Mass Transfer 39, 28032809.
70. Vafai, K., and Tien, C.-L. (1981). Boundary and inertia effects on ow and heat transfer
in porous media. Int. J. Heat and Mass Transfer 24, 195203.
71. Howle, L., Behringer, R. P., and Georgiadis, J. G. (1992). Pattern formation near the onset
of convection for uid in a porous medium. Private communication.
72. Rodi, W. (1980). Turbulence models for environmental problems. In Prediction Methods
for Turbulent Flows (W. Kollmann, ed.), pp. 259350. Hemisphere Publishing Corpor-
ation, New York.
73. Lumley, J. L. (1978). Computational modelling of turbulent ows. Adv. Appl. Mechan. 18,
123176.
74. Shvab, V. A., and Bezprozvannykh, V. A. (1984). On turbulent ow simulation in
rectilinear channels of noncircular cross-section. In Metody Aerodinamiki i Teplomas-
soobmena v Tekhnologicheskikh Protsessakh, pp. 325. Izdatelstvo TGU, Tomsk (in
Russian).
75. Patel, V. C., Rodi, W., and Scheurer, G. (1985). Turbulence models for near-wall and low
Reynolds number ows: a review. AIAA J. 23, 13081319.
76. Brereton, G. J., and Kodal, A. (1992). A frequency-domain ltering technique for triple
decomposition of unsteady turbulent ow. J. Fluids Eng. 114, 4551.
77. Bisset, D. K., Antonia, R. A., and Raupach, M. R. (1991). Topology and transport
properties of large-scale organized motion in a slightly heated rough wall boundary layer.
Phys. Fluids A 3, 22202228.
78. Primak, A. V., and Travkin, V. S. (1989). Simulation of turbulent transfer of meteoele-
ments and pollutants under conditions of articial anthropogenic action in urban
roughness layer as in sorbing and biporous two-phase medium. In Intern. Sym. Intercon-
nec. Problems Hydrological Cycle and Atmospheric Proc. under Conditions Anthropogenic
Inuences, Trans., Schopron.
79. Hsu, C. T., and Cheng, P. (1988). Closure schemes of the macroscopic energy equation
for convective heat transfer in porous media. Int. Comm. Heat Mass Transfer 15, 689703.
80. Hsu, C. T., and Cheng, P. (1990). Thermal dispersion in a porous medium. Int. J. Heat
Mass Transfer 33, 15871597.
81. Lehner, F. K. (1979). On the validity of Ficks law for transient diffusion through a porous
medium. Chem. Eng. Sci. 34, 821825.
82. Fox, R. F., and Barakat, R. (1976). Heat conduction in a random medium. J. Stat. Phys.
18, 171178.
83. Gelhar, L. W., Gutjahr, A. L., and Naff, R. L. (1979). Stochastic analysis of macrodisper-
sion in a stratied aquifer. Water Resources Res. 15, 13871389.
84. Tang, D. H., Schwartz, F. W., and Smith, L. (1982). Stochastic modeling of mass transport
in a random velocity eld. Water Resources Res. 18, 231244.
85. Torquato, S., Lu, B., and Rubenstein, J. (1990). Nearest-neighbor distribution functions in
many-body systems. Phys. Rev. A 41, 20592075.
86. Miller, C. A., and Torquato, S. (1990). Effective conductivity of hard-sphere dispersions.
J. Appl. Phys. 68, 54865493.
87. Kim, I. C., and Torquato, S. (1992). Diffusion of nite-sized Brownian particles in porous
media. J. Chem. Phys. 96, 14981503.
88. Lu, B., and Torquato, S. (1992). Nearest-surface distribution functions for polydispersed
particle systems. Phys. Rev. A 45, 55305544.
89. Carbonell, R. G., and Whitaker, S. (1983). Dispersion in pulsed systems II. Theoretical
developments for passive dispersion in porous media. Chem. Eng. Sci. 38, 17951802.
90. Carbonell, R. G. (1979). Effect of pore distribution and ow segregation on dispersion in
porous media. Chem. Eng. Sci. 34, 10311039.
91. Fushinobu, K., Majumdar, A., and Hijikata, K. (1995). Heat generation and transport in
submicron semiconductor devices. J. Heat Trans. 117, 2531.
92. Caceres, M. O., and Wio, H. S. (1987). Non-Markovian diffusion-like equation for
transport processes with anisotropic scattering. Physica A 142, 563578.
93. Tzou, D. Y. O zisik, M. N., and Chiffelle, R. J. (1994). The lattice temperature in the
microscopic two-step model. J. Heat Trans. 116, 10341038.
94. Majumdar, A. (1993). Microscale heat conduction in dielectric thin lms. J. Heat Trans.
115, 716.
95. Peterson, R. B. (1994). Direct simulation of phonon-mediated heat transfer in a debye
crystal. J. Heat Trans. 116, 815822.
96. Tzou, D. Y. (1995). A unied eld approach for heat conduction from macro- to
micro-scales. J. Heat Trans. 117, 816.
97. Kaganov, M. I., Lifshitz, I. M., and Tanatarov, L. V., (1957). Relaxation between electrons
and the crystalline lattice. Sov. Phys JETP 4, 173178.
98. Ginzburg, V. L., and Shabanskii, V. P. (1955). Electron kinetic temperature in metals and
anomalous electron emission. Dokl. Akad. Nauk SSSR 100, 445448.
99. Akhiezer, A. I., and Pomeranchuk, I. Ia. (1944). On the thermal equilibrium between spins
and crystal lattice. J. Phys. VIII(4), pp. 206215.
100. Anisimov, S. I., Imas, Ya. A., Romanov, G. S., and Yu. V. Khodyko (1970). Effect of
High-Intensity Radiation on Metals. Nauka, Moscow (in Russian).
101. Anisimov, S. I., Kapeliovich, B. L., and Perelman, T. L., (1974). Electron emission from
metal surfaces exposed to ultrashort laser pulses. Sov. Phys. JETP 39, 375377.
102. Qiu, T. Q., and Tien, C. L., (1992). Short-pulse laser heating on metals. Int. J. Heat Mass
Trans. 35, 719726.
103. Qiu, T. Q., and Tien, C. L. (1993). Heat transfer mechanisms during short-pulse laser
heating of metals. J. Heat Transf. 115, 835841.
104. Qiu, T. Q., and Tien, C. L. (1993). Size effects on nonequilibrium laser heating of metal
lms. J. Heat Transf. 115, 842847.
105. Fujimoto, J. G., Liu, J. M., and Ippen, E. P. (1984). Femtosecond laser interaction with
metallic tungsten and non-equilibrium electron and lattice temperature. Phys. Rev. L ett.
53, 18371840.
106. Elsayed-Ali, H. E. (1991). Femtosecond thermoreectivity and thermotransmissivity of
polycrystalline and single-crystalline gold lms. Phys. Rev. B 43, 44884491.
107. Gladkov, S. O. (1997). Physics of Porous Structures. Nauka, Moscow (in Russian).
108. Joseph, D. D., and Preziosi, L. (1989). Heat waves. Rev. Mod. Phys. 61, 4173.
109. Majumdar, A., Lai, J., Luo, K., and Shi, Z. (1995). Thermal imaging and modeling of
sub-micrometer silicon devices. In Proc. Symposium on Thermal Science and Engin. in
Honor of Chancellor Chang-L in Tien, pp. 137144.
110. Chen, G., and Tien, C. L. (1994). Thermally induced optical nonlinearity during transient
heating of thin lms. J. Heat Transf. 116, 311316.
111. Chen, G. (1997). Size and interface effects on thermal conductivity of superlattices and
periodic thin-lm structures. J. Heat Transf. 119, 220229.
112. Goodson, K. E., and Flik, M. I. (1993). Electron and phonon thermal conduction in
epitaxial high-T
superconducting lms. J. Heat Transf. 115, 1725.

113. Goodson, K. E. (1996). Thermal conduction in nonhomogeneous CVD diamond layers in
electronic microstructures. J. Heat Transf. 118, 279286.
114. Travkin, V. S., Catton, I., and Ponomarenko, A. T. (1999). Governing equations for
electrodynamics in heterogeneous meda. Submitted.
115. Travkin, V. S., Ponomarenko, A. T., and Ryvkina, N. G. (1999). Non-local formulation
of electrostatic problems in heterogeneous two-phase media. Submitted.
116. Yablonovitch, E. (1987). Inhibited spontaneous emission in solid-state physics and
electronics. Phys. Rev. L ett. 58, 20592062.
117. Yablonovitch, E., and Gmitter, T. J. (1989). Photonic band structure: the face-centered-
cubic case. Phys. Rev. L ett. 63, 19501953.
118. John, S. (1987). Strong localization of photons in certain disordered dielectric superlat-
tices. Phys. Rev. L ett. 58, 24862489.
119. John, S., and Rangarajan, R. (1988). Optimal structures for classical wave localization: an
alternative to the IoffeRegel criterion. Phys. Rev. B 38, 1010110104.
120. Cox, S. J., and Dobson, D. C. (1998). Maximizing band gaps in two-dimensional photonic
crystals. At the IVth conf. Mathematical and Numerical Aspects of Wave Propagation.
SIAM, Denver, private communication.
121. Pereverzev, S. I., and Umtsev, P. Y. (1994). Effective permittivity and permeability of a
bers grating. Electromagnetics 14, 137151.
122. Figotin, A., and Kuchment, P. (1996). Band-gap structure of spectra of periodic dielectric
and acoustic media. I. Scalar model. SIAM J. Appl. Math. 56, 6888.
123. Figotin, A., and Kuchment, P. (1996). Band-gap structure of spectra of periodic dielectric
and acoustic media. II. Two-dimensional photonic crystals. SIAM J. Appl. Math. 56,
15811620.
124. Figotin, A., and Godin, Yu. A. (1997). The computation of spectra of some 2D photonic
crystals. J. Comp. Phys. 136, 585598.
125. Figotin, A., and Kuchment, P. (1998). Spectral properties of classical waves in high-
contrast periodic media. SIAM J. Appl. Math. 58, 683702.
126. Nicorovici, N. A., McPhedran, R. C., and Botten, L. C. (1995). Photonic band gaps:
non-commuting limits and the acoustic band. Phys. Rev. L ett. 75, 15071510.
127. Nicorovici, N. A., McPhedran, R. C., and Botten, L. C. (1995). Photonic band gaps for
arrays of perfectly conducting cylinders. Phys. Rev. E 52, 11351145.
128. Hilfer, R. (1992). Local porosity theory for ow in porous media. Phys. Rev. B 45,
71157124.
129. Hilfer, R. (1993). Local porosity theory for electrical and hydrodynamical transport
through porous media. Physica A 194, 406412.
130. Tien, C.-L. (1988). Thermal radiation in packed and uidized beds. ASME J. Heat Transf.
110, 12301242.
131. Siegel, R., and Howell, J. R. (1992). Thermal Radiation Heat Transfer, 3rd ed. Hemisphere,
Washington.
132. Hendricks, T. J., and Howell, J. R. (1994). Absorption/scattering coefcients and scattering
phase functions in reticulated porous ceramics. In Radiation Heat Transfer: Current
Research (Y. Bayazitoglu, et al., eds.), ASME HTD-276.
133. Kumar, S., Majumdar, A., and Tien, C.-L. (1990). The differential-discrete ordinate
method for solution of the equation of radiative transfer. ASME J. Heat Transf. 112,
424429.
134. Singh, B. P., and Kaviany, M. (1994). Effect of particle conductivity on radiative heat
transfer in packed beds. Int. J. Heat Mass Transf. 37, 25792583.
135. Tien, C.-L., and Drolen, B. L. (1987). Thermal radiation in particulate media with
dependent and independent scattering. In Annual Review of Numerical Fluid Mechanics
and Heat Transfer, (T. C. Chawla, ed.) 1, pp. 132.
136. Al-Nimr, M. A., and Arpaci, V. S. (1992). Radiative properties of interacting particles. J.
Heat Transf. 114, 950957.
137. Kumar, S., and Tien, C.-L. (1990). Dependent scattering and absorption of radiation by
small particles. ASME J. Heat Transf. 112, 178185.
138. Lee, S. C. (1990). Scattering phase function for brous media. Int. J. Heat Mass Transf.
33, 21832190.
139. Lee, S. C., White, S., and Grzesik, J. (1994). Effective radiative properties of brous
composites containing spherical particles. J. T hermoph. Heat Transf. 8, 400405.
140. Dombrovsky, L. A. (1996). Radiation Heat Transfer in Disperse Systems. Bergell House
Inc. Publ., New York.
141. Reiss, H. (1990). Radiative transfer in nontransparent dispersed media. High Temp.High
Press. 22, 481522.
142. Adzerikho, K. S., Nogotov, E. F., and Tromov, V. P. (1990). Radiative Heat Transfer in
Two-Phase Media. CRC Press, Boca Raton, FL.
143. van de Hulst, H. C. (1981). L ight Scattering by Small Particles. Dover, New York.
144. Bohren, C. F., and Huffman, D. R. (1983). Absorption and Scattering of L ight by Small
Particles. Wiley Interscience, New York.
145. Lorrain, P., and Corson, D. R. (1970). Electromagnetic Fields and Waves, 2nd ed., pp.
422551. Freeman and Co., New York.
146. Lindell, I. V., Sihvola, A. H., Tretyakov, S. A., and Viitanen, A. J. (1994). Electromagnetic
Waves in Chiral and Bi-Isotropic Media. Artech House, Norwood, MA.
147. Lakhtakia, A., Varadan, V. K., and Varadan, V. V. (1989). Time-Harmonic Electromag-
netic Fields in Chiral Media. L ecture Notes in Physics 335. Springer-Verlag, Berlin.
148. Pomraning, G. C. (1991). A model for interface intensities in stochastic particle transport.
J. Quant. Spectrosc. Radiat. Transf. 46, 221236.
149. Pomraning, G. C. (1991b). L inear Kinetic Theory and Particle Transport in Stochastic
Mixtures. World Scientic, Singapore.
150. Pomraning, G. C. (1996). The variance in stochastic transport problems with Markovian
mixing. J. Quant. Spectrosc. Radiat. Transf. 56, 629646.
151. Pomraning, G. C. (1997). Renewal analysis for higher moments in stochastic transport. J.
Quant. Spectrosc. Radiat. Transfer 57, 295307.
152. Malvagi, F., and Pomraning, G. C. (1992). A comparison of models for particle transport
through stochastic mixtures. Nucl. Sci. Eng. 111, 215228.
153. Farone, W. A., and Querfeld, C. W. (1966). Electromagnetic scattering from radially
inhomogeneous innite cylinders at oblique incidence. J. Opt. Soc. Am. 56, 476480.
154. Samaddar, S. N. (1970). Scattering of plane electromagnetic waves by radially in-
homogeneous innite cylinders. Nuovo Cimento 66B, 3351.
155. Botten, L. C., McPhedran, R. C., Nicorovici, N. A., and Movchan, A. B. (1998). Off-axis
diffraction by perfectly conducting capacitive grids: Modal formulation and verication.
J. Electromagn. Waves Applic. 12, 847882.
156. McPhedran, R. C., Dawes, D. H., Botten, L. C., and Nicorovici, N. A. (1996). On-axis
diffraction by perfectly conducting capacitive grids. J. Electromagn. Waves Applic. 10,
10831109.
157. McPhedran, R. C., Nicorovici, N. A., and Botten, L. C. (1997). The TEM mode and
homogenization of doubly periodic structures. J. Electromagn. Waves Applic. 11, 9811012.
158. Catton, I., and Travkin, V. S. (1997). Homogeneous and non-local heterogeneous
transport phenomena with VAT application analysis. In Proc. 15th Symposium on Energy
Engin. Sciences, Argonne National Laboratory, Conf. 9705121, pp. 4855.
159. Travkin, V. S., Catton, I., Ponomarenko, A. T., and Tchmutin, I. A. (1998). A hierarchical
description of diffusion and electrostatic transport in solid and porous composites and the
development of an optimization procedure. In ACerS PCR & BSD Conf. Proc., p. 20.
160. Ryvkina, N. G., Ponomarenko, A. T., Tchmutin, I. A., and Travkin, V. S. (1998). Electrical
and magnetic properties of liquid dielectric impregnated porous ferrite media. In Proc.
XIVth International Conference on Gyromagnetic Electronics and Electrodynamics, Micro-
wave Ferrites, ICMF98, Section Spin-Electronics, 2, pp. 236249.
161.Ponomarenko, A. T., Ryvkina, N. G., Kazantseva, N. E., Tchmutin, I. A., Shevchenko, V.
G., Catton, I., and Travkin, V. S. (1999). Modeling of electrodynamic properties control
in liquid impregnated porous ferrite media. In Proc. SPIE Smart Structures and Materials
1999, Mathematics and Control in Smart Structures (V. V. Varadan, ed.), 3667, pp.
785796.
162. Ryvkina, N. G., Ponomarenko, A. T., Travkin, V. S., Tchmutin, I. A., and Shevchenko, V.
G. (1999). Liquid-impregnated porous media: structure, physical processes, electrical
properties. Materials, Technologies, Tools 4, 2741 (in Russian).
163. V. S. Travkin, I. Catton, A. T. Ponomarenko, and S. A. Gridnev (1999). Multiscale
non-local interactions of acoustical and optical elds in heterogeneous materials. Possi-
bilities for design of new materials. In Advances in Acousto-Optics 99, pp. 3132. SIOF,
Florence.
164. Pomraning, G. C., and Su, B. (1994). A closure for stochastic transport equations. In
Reactor Physics and Reactor Computations, Proc. Int. Conf. Reactor Physics & Reactor
Computations (Y. Rohen and E. Elias, eds.), pp. 672679. Negev Press, Tel-Aviv.
165. Buyevich, Y. A., and Theofanous, T. G. (1997). Ensemble averaging technique in the
mechanics of suspensions. ASME FED 243, pp. 4160.
166. Travkin, V. S., and Catton, I. (1998). Thermal transport in HT superconductors based on
hierarchical non-local description. In ACerS PCR & BSD Conf. Proc., p. 49.
167. Ergun, S. (1952). Fluid ow through packed columns. Chem. Eng. Prog. 48, 8994.
168. Vafai, K., and Kim, S. J. (1989). Forced convection in a channel lled with a porous
medium: An exact solution. J. Heat Transf. 111, 11031106.
169. Poulikakos, D., and Renken, K. (1987). Forced convection in a channel lled with porous
medium, including the effects of ow inertia, variable porosity, and Brinkman friction. J.
Heat Transf. 109, 880888.
170. Schlichting, H. (1968). Boundary L ayer Theory, 6th ed. McGraw-Hill, New York.
171. Achdou, Y., and Avellaneda, M. (1992). Inuence of pore roughness and pore-size
dispersion in estimating the permeability of a porous medium from electrical measure-
ments. Phys. Fluids A 4, 26512673.
172. Kays, W. M., and London, A. L. (1984). Compact Heat Exchangers, 3rd ed. McGraw-Hill,
New York.
173. Bird, R. B., Stewart, W. E., and Lightfoot, E. N. (1960). Transport Phenomena. Wiley, New
York.
174. Chhabra, R. P. (1993). Bubbles, Drops, and Particles in Non-Newtonian Fluids. CRC Press,
Boca Raton, FL.
175. Gortyshov, Yu, F., Muravev, G. B., and Nadyrov, I. N. (1987). Experimental study of ow
and heat exchange in highly porous structures. Eng. Phys. J. 53, 357361 (in Russian).
176. Gortyshov, Yu. F., Nadyrov, I. N., Ashikhmin, S. R., and Kunevich, A. P. (1991). Heat
transfer in the ow of a single-phase and boiling coolant in a channel with a porous insert.
Eng. Phys. J. 60, 252258 (in Russian).
177. Beavers, G. S., and Sparrow, E. M. (1969). Non-Darcy ow through brous porous media.
J. Appl. Mech. 36, 711714.
178. Ward, J. C. (1964). Turbulent ow in porous media. J. Hydraulics Division, Proc. ASCE
90, 112.
179. Kurshin, A. P. (1985). Gas ow hydraulic resistance in porous medium. Uchenie Zapiski
TsAGI 14, 7383 (in Russian).
180. Macdonald, I. F., El-Sayed, M. S., Mow, K., and Dullien, F. A. L. (1979). Flow through
porous media the Ergun equation revisited. Ind. Eng. Chem. Fund. 18(3), 199208.
181. Souto, H. P. A., and Moyne, C., (1997). Dispersion in two-dimensional periodic media.
Part I. Hydrodynamics. Phys. Fluids 9(8), 22432252.
182. Viskanta, R. (1995). Modeling of transport phenomena in porous media using a two-
energy equation model. In Proc. ASME/JSME T hermal Eng. Joint Conf. 3, pp. 1122.
183. Viskanta, R. (1995). Convective heat transfer in consolidated porous materials: a perspec-
tive. In Proc. Symposium on Thermal Science and Engineering in Honour of Chancellor
Chang-L in Tien, pp. 4350.
184. Kar, K. K., and Dybbs, A. (1982). Internal heat transfer coefcients of porous metals. In
Heat Transfer in Porous Media (J. V. Beck and L. S. Yao, eds.), 22, pp. f8191. ASME,
New York.
185. Rajkumar, M. (1993). Theoretical and experimental studies of heat transfer in transpired
porous ceramics. M.S.M.E. Thesis, Purdue University, West Lafayette, IN.
186. Achenbach, E. (1995). Heat and ow characteristics in packed beds. Exp. T herm. Fluid
Sci. 10, 1721.
187. Younis, L. B., and Viskanta, R. (1993). Experimental determination of volumetric heat
transfer coefcient between stream of air and ceramic foam. Intern. J. Heat Mass Transf.
36, 14251434.
188. Younis, L. B., and Viskanta, R. (1993). Convective heat transfer between an air stream
and reticulated ceramic. In Multiphase Transport in Porous Media 1993, (R. R. Eaton, M.
Kaviany, M. P. Sharima, K. S. Udell, and K. Vafai, eds.), 173, pp. 109116. ASME, New
York.
189. Galitseysky, B. M., and Moshaev, A. P. (1993). Heat transfer and hydraulic resistance in
porous systems. In Experimental Heat Transfer, Fluid Mechanics and T hermodynamics:
1993 (M. D. Kelleher, K. R. Sreehivasan, R. K. Shah, and Y. Toshi, eds.), pp. 15691576.
Elsevier Science Publishers, New York.
190. Kokorev, V. I., Subbotin, V. I., Fedoseev, V. N., Kharitonov, V. V., and Voskoboinikov,
V. V. (1987) Relationship between hydraulic resistance and heat transfer in porous media.
High Temp. 25, 8287.
191. Heat Exchanger Design Handbook (Spalding, B. D., Taborek, J., Armstrong, R. C. et al.,
contribs.), 1, 2 (1983). Hemisphere Publishing Corporation, New York.
192. Uher, C. (1990). Thermal conductivity of high-T
superconductors. J. Supercond. 3, 337389.

193. Cheng, H., and Torquato, S. (1997). Electric-eld uctuations in random dielectric
composites. Phys. Rev. B 56, 80608068.
194. Khoroshun, L. P. (1976). Theory of thermal conductivity of two-phase solid bodies. Sov.
Appl. Mech. 12, 657663.
195. Khoroshun, L. P. (1978). Methods of random function theory in problems on macroscopic
properties of micrononhomogeneous media. Sov. Appl. Mech. 14, 113124.
196. Beran, M. J. (1974). Application of statistical theories for the determination of thermal,
electrical, and magnetic properties of heterogeneous materials. In Mechanics of Composite
Materials (G. P. Sendeckyj, ed.), 2, pp. 209249. Academic Press, New York.
197. Kudinov, V. A., and Moizhes, B. Ya. (1979). Effective conductivity of nonuniform medium.
Iteration series and variation estimations of herring method. J. Tech. Phys. 49, 15951603.
198. Hadley, G. R. (1986). Thermal conductivity of packed metal powders. Int. J. Heat Mass
Transf. 29, 909920.
199. Kuwahara, F., and Nakayama, A. (1998). Numerical modelling of non-Darcy convective
ow in a porous medium. In Proc. 10th. Intern. Heat Transfer Conf., Industrial Sessions
Papers (Hewitt, G. F., ed.), 4, pp. 411416. Brighton.
200. Churchill, S. W. (1997). Critique of the classical algebraic analogies between heat, mass,
and momentum transfer. Ind. Eng. Chem. Res. 36, 38663878.
201. Churchill, S. W., and Chan, C. (1995). Theoretically based correlating equations for the
local characteristics of fully turbulent ow in round tubes and between parallel plates. Ind.
Eng. Chem. Res. 34, 13321341.
202. Tsay, R., and Weinbaum S. (1991). Viscous ow in a channel with periodic cross-bridging
bres: exact solutions and Brinkman approximation. J. Fluid Mech. 226, 125148.
203. Bejan, A., and Morega, A. M. (1993). Optimal arrays of pin ns and plate ns in laminar
forced convection. J. Heat Transf. 115, 7581.
204. Butterworth, D. (1994). Developments in the computer design of heat exchangers. In Proc.
10th Intern. Heat Transfer Conf., Industrial Sessions Papers (Hewitt, G. F., ed.), 1, pp.
433444. Brighton.
205. Martin, H. (1992). Heat Exchangers. Hemisphere Publishing Co., Washington.
206. Paffenbarger, J. (1990). General computer analysis of multistream, plate-n heat ex-
changers. In Compact Heat Exchangers (R. K. Shah, A. D. Kraus, and D. Metzger, eds.),
pp. 727746. Hemisphere Publishing Co., New York.
207. Webb, R. L. (1994). Principles of Enhanced Heat Transfer. Wiley Interscience, New York.
208. Webb, R. L. (1994). Advances in modeling enhanced heat transfer surfaces. In Proc. 10th Int.
Heat Transfer Conf., Industrial Sessions Papers (Hewitt, G. F., ed.), 1, pp. 445459. Brighton.
209. Bergles, A. E. (1988). Some perspectives on enhanced heat transfer: second generation heat
transfer technology. J. Heat Transf. 110, 10821096.
210. Fukagawa, M., Matsuo, T., Kanzaka, M., Motai, T., and Iwabuchi, M. (1994). Heat
transfer and pressure drop of nned tube banks with staggered arrangements in forced
convection. In Proc. 10th Int. Heat Transfer Conf., Industrial Sessions Papers (Berryman,
R. J., ed.), pp. 183188. Brighton.
211. Burns, J. A., Ito, K., and Kang, S. (1991). Unbounded observation and boundary control
problems for Burgers equation. In Proc. 30th IEEE Conference on Decision and Control,
pp. 26872692. IEEE, New York.
212. Burns, J. A., and Kang, S. (1991). A control problem for Burgers equation with bounded
input/output. In ICASE Report 90-45, 1990 NASA Langley Research Center, Nonlinear
Dynamics 2, pp. 235262. NASA, Hampton.
213. Teo, K. L., and Wu, Z. S. (1984). Computational Methods for Optimizing Distributed
Systems. Academic Press, New York.
214. Ahmed, N. U., and Teo, K. L. (1981). Optimal Control of Distributed Parameters Systems.
North-Holland, Amsterdam.
215. Ahmed, N. U., and Teo, K. L. (1974). An existence theorem on optimal control of partially
observable diffusion. SIAM J. Control 12, 351355.
216. Ahmed, N. U., and Teo, K. L. (1975). Optimal control of stochastic Ito differential
equation. Int. J. Systems Sci. 6, 749754.
217. Ahmed, N. U., and Teo, K. L. (1975b). Necessary conditions for optimality of a cauchy
problem for parabolic partial differential systems. SIAM J. Control 13, 981993.
218. Fleming, W. H. (1978). Optimal control of partially observable diffusions. SIAM J.
Control 6, 194213.
219. Da Prato, G., and Ichikawa, A. (1993). Optimal control for integrodifferential equations
of parabolic type. SIAM J. Control Optimization 31, 11671182.
220. Butkovski, A. G. (1961). Maximum principle of optimal control for distributed parameter
systems. Automat. Telemekh. 22, 12881301 (in Russian).
221. Balakrishnan, A. V. (1976). Applied Functonal Analysis. Springer-Verlag, New York.
222. Curtain, R. F., and Pritchard, A. J. (1981). Innite dimensional linear systems theory.
L ecture Notes in Control and Information Sciences 8. Springer-Verlag, New York.
223. Fattorini, H. O. (1994). Existence theory and the maximum principle for relaxed
innite-dimensional optimal control problems. SIAM J. Control and Optimization 32,
311331.
224. Anita, S. (1994). Optimal control of parameter distributed systems with impulses. Appl.
Math. Optim. 29, 93107.
225. Ahmed, N. U., and Xiang, X. (1994). Optimal control of innite-dimensional uncertain
systems. J. Optimiz. T heory Appl. 80, 261273.
Two-Phase Flow in Microchannels
S. M. GHIAASIAAN and S. I. ABDEL-KHALIK
G. W. Woodruff School of Mechanical Engineering
Georgia Institute of Technology
Atlanta, Georgia 30332
I. Introduction
The application of miniature thermal and mechanical systems is rapidly
increasing in various branches of industry. Recent technological advances
have led to extremely ne spatial and temporal thermal load resolutions,
requiring the analysis of microscale heat transfer phenomena where the
system characteristic dimension can be smaller than the mean free path of
the heat carrying particles [1].
In this article the recently published research dealing with gasliquid
two-phase ow in microchannels is reviewed. Only microchannels with
hydraulic diameters of the order of 0.1 to 1 mm and with long length-to-
hydraulic diameter ratios are considered. In these systems the channel
characteristic dimension is of the same order of magnitude, or smaller than,
the neutral interfacial wavelengths predicted by the Taylor stability analysis.
The review is also restricted to situations where the uid inertia is signicant
in comparison with surface tension. Such microchannels and ow conditions
are encountered in miniature heat exchangers, research nuclear reactors,
biotechnology systems, the cooling of high-power electronic systems, the
cooling of the plasma-facing components in fusion reactors, and the heat
rejection systems of spacecraft, to name a few. The ow through cracks and
slits when such cracks develop in vessels and piping systems containing
high-pressure liquids is another application of two-phase ow in microchan-
nels of interest here. Two-phase ow in capillaries with complex geometry
(porous media) has been reviewed in the recent past [24] and will not be
addressed.
145
0065-2717/01 $35.00
II. Characteristics of Microchannel Flow
For steady-state and developed two-phase ow in a smooth pipe, and
assuming incompressible phases, the variables that can affect the hy-
drodynamics of gasliquid two-phase ow are j
'
, j
'
, j
'
, j
'
, o, D, g, 0
, ,
U
'`
, and U
'`
. Since the minimum number of reference dimensions in
hydrodynamics is three (time, mass, and length), according to the Buckin-
ham theorem eight independent dimensionless parameters can be dened
that in general affect the hydrodynamics of gasliquid two-phase ow. The
following three dimensionless parameters are particularly important for the
characterization of microchannels:
Eo
AjgD`
o
(1)
We
'`
U`
'`
Dj
'
o
(2)
We
'`
U`
'`
Dj
'
o
. (3)
The remainder of the dimensionless parameters can be represented as Aj/j
'
,
j
'
/j
'
, 0
, , and the phasic supercial Reynolds numbers:

Re
'`
U
'`
D/v
'
(4)
Re
'`
U
'`
D/v
'
. (5)
Note that the phasic Froude numbers Fr
'`
U`
'`
/gD and Fr
'`
U`
'`
/gD,
and the capillary number Ca j
'
U
'`
/o, can all be derived by combining
two or more of these dimensionless parameters. When Eo, We
'`
, and We
'`
are all less than 1, gravitation and inertia are both insignicant in compari-
son with surface tension. Re
'`
1, furthermore, implies small inertia com-
pared with viscous forces. Flow elds in capillaries where surface tension
and viscosity dominate buoyancy and inertia have important applications
and have been extensively studied in the past [58]. In microchannels of
interest here, however, Eo 1, at least one of the Weber numbers is
typically of the order of 1 to 10`, and Re
'`
1. Thus, although surface
tension mostly dominates buoyancy, inertia can be signicant. Similar
conditions apply to two-phase ow in microgravity, resulting in important
and useful similarities between the two categories of systems with respect to
hydrodynamics of two-phase ow. In both types of systems the predomin-
ance of the surface tension force on buoyancy leads to the insensitivity of
two-phase hydrodynamics to channel orientation, and in nonseparated
two-phase ow patterns it leads to the suppression of velocity difference
146 s. x. cni::si::N :Nn s. i. :nnri-xn:iix
between the two phases in the absence of signicant acceleration.
Suo and Grifth [9] derived the following criterion for negligible buoy-
ancy effect in two-phase pipe ow:
j
'
gD`
o
0.88/3. (6)
Based on an analysis of stratied to nonstratied ow regime transition,
Brauner and Moalem-Maron [10] derived the following criterion for the
predominance of surface tension on buoyancy:
Eo (2)`. (7)
Experiments with water and air owing in pipes indicate that the transition
to the surface tensiondominated regime (where ow patterns are not
affected by channel orientation) occurs in the 1 D2 mm range [11, 12].
Equation (6) agrees well with the latter observations. The channel hydraulic
diameters considered here thus cover the aforementioned critical range.
Another important characteristic of microchannels of interest here is that
for them,
D
O(z), (8)
where
z
,
o
gAj
. (9)
The Laplace length scale, z, represents the order of magnitude of the
wavelength of the interfacial waves in Taylor instability, and the latter
instability type is known to govern important hydrodynamic processes such
as bubble and droplet breakup. Equation (8) evidently implies that some
Taylor instability-driven processes may be entirely irrelevant to microchan-
nels. The criterion of Suo and Grifth, Eq. (6), can approximately be recast
as
D0.3z. (10)
III. Two-Phase Flow Regimes and Void Fraction in Microchannels
The gas and liquid phases in a two-phase ow system can exist in various
distinct morphological congurations. Flow regimes represent the major
morphological congurations of the phases and are among the most
important characteristics of two-phase ow systems, since they strongly
inuence all the hydrodynamic and transport processes, such as pressure
147 1vo-in:sr riov iN xicocn:NNris
drop, heat and mass transfer, and ow stability. A methodology for
predicting the two-phase ow regimes is thus required for the design of
two-phase ow systems and for the specication of appropriate closure
relations for two-phase conservation equations.
Flow regimes and conditions leading to their establishment have been
extensively investigated in the past several decades. Methods for predicting
the ow regimes are often based on the ow regime map concept, according
to which the empirically determined ranges of occurrence of all major ow
patterns are specied on a two-dimensional map, with the two coordinates
representing some appropriate hydrodynamic parameters [13, 14]. How-
ever, since the gasliquid hydrodynamics are affected by a large number of
independent dimensionless parameters, the two-dimensional ow regime
maps are often in disagreement with respect to the parameters they use as
coordinates and their ranges of applicability are limited to the ranges of
their databases. More recently, semianalytical methods, where the ow
regime transition processes are mechanistically or semianalytically modeled,
have been proposed [1517] and have undergone extension and improve-
ment [1820]. The existing ow regime maps, as well as the aforementioned
semianalytical models, however, generally do poorly when compared with
experimental two-phase ow regime data representing microchannels.
A. DrriNi1ioN or M:o Tvo-Pn:sr Fiov Rrcixrs
Experiments indicate that ow regimes, which are morphologically simi-
lar to the major two-phase ow regimes common in large channels, occur
in microchannels as well. Therefore, a brief review of the major ow regimes
observed in large channels is provided in this section. Detailed explanations
of these ow regimes and their characteristics can be found in various
textbooks and monographs [13, 14] and in more recent review articles [18].
Figure 1 schematically depicts the major ow regimes in common
gasNewtonian liquid two-phase ow systems in large vertical channels.
Bubbly ow occurs at low gas and liquid ow rates and is characterized by
bubbles distorted-spherical in shape and moving upward in zigzag fashion.
The slug ow regime is characterized by bullet-shaped Taylor bubbles that
have diameters close to the diameter of the channel and are separated from
the channel wall by a thin liquid lm, with lengths that can widely vary and
may reach more than 15 times the channel diameter. The Taylor bubbles
are separated by liquid slugs that often contain small entrained bubbles. The
churn ow is established following the disruption of the Taylor bubbles due
to high gas ow rates and is characterized by chaotic oscillations and
churning. In the annular ow pattern a thin liquid lm, which can be
smooth or wavy depending on the gas velocity, ows on the wall, while the
Fic. 1. Major ow patterns in a large vertical pipe.
gas ows through the channel core. The gas often contains dispersed
droplets. In the dispersed-bubbly ow regime, which is established at very
high liquid ow rates, small spherical bubbles with little or no interaction
with each other are mixed with the liquid. Unlike the common bubbly ow
regime, in which the bubble size is controlled by Taylor instability and
aerodynamic forces, the size of the bubbles in the dispersed bubbly ow
regime is dictated by turbulence in the liquid.
The commonly observed two-phase ow patterns associated with the ow
of a gas and a Newtonian liquid in a horizontal large channel are depicted
in Fig. 2. Bubbly ow occurs at high liquid and low gas ow rates and is
followed by plug (elongated bubble), slug, and annular/dispersed ow
patterns as the gas ow rate is increased. The stratied-smooth ow regime
occurs at low liquid and low gas ow rates and is followed by stratied wavy
and annular/dispersed ow regimes as the gas ow rate is increased. The
dispersed-bubbly regime occurs at very high liquid ow rates, and its
characteristics are similar to those of the dispersed-bubbly ow regime in
vertical channels. The plug and slug ow patterns are often referred to
collectively as the intermittent ow pattern, since the distinction between
them is not always clear or important.
The ow patterns in Figs. 1 and 2 only display the major ow regimes
that are easily discernible visually and with simple photographic techniques
and are commonly addressed in ow regime maps and transition models.
Fic. 2. Major ow patterns in a large horizontal pipe.
Numerous subtle variations within some of the ow patterns can be
recognized using more sophisticated techniques, however [21].
B. Tvo-Pn:sr Fiov Rrcixrs iN Micocn:NNris
Early studies dealing with two-phase ow in microchannels were mostly
concerned with surface tension-driven ows [58]. Two-phase ow regimes
in microchannels under conditions where inertia is signicant have been
experimentally investigated by Suo and Grifth [9], Oya [22], Barnea et al.
[23], Damianides and Westwater [11], Barajas and Panton [24], Fukano
and Kariyasaki [25], Mishima and Hibiki [26], and Triplett et al. [12].
Two-phase ow patterns in narrow, rectangular channels, some simulating
slits and cracks, have also been reported in [2733]. Narrow et al. [34] and
Ekberg et al. [35] studied the two phase ow regimes in a micro-rod bundle
Fic. 3. Cross-sectional geometry of the test sections of Triplett et al. [12].
and in narrow annuli, respectively. Two-phase ow and transport phenom-
ena in the slug (bubble train) regime in microchannels have also been
investigated [36, 37].
The commonly observed ow patterns in microchannels are depicted here
using the photographs provided by Triplett et al. [12]. The major ow
regimes shown in these pictures are in agreement with the observation of
most of the other investigators, although, as will be shown later, some ow
patterns have been given different names by different authors. Triplett et al.
[12] conducted experiments using air and water at room temperature, in
horizontal, transparent circular test sections with 1.09 and 1.45 mm diam-
eter, and in microchannels with semitriangular (triangular with one corner
smoothed) cross sections with 1.09 and 1.49mm hydraulic diameters (see
Fig. 3). They identied the ow regimes using high-speed videocameras
recording ow details near the centers of the test sections. Figure 4 displays
typical photographs of the ow patterns identied in their 1.09-mm-
diameter circular test section. The overall ow pattern morphologies ob-
served with the other test sections used by Triplett et al. [12] were similar
to the pictures in Fig. 4.
Fic. 4. Representative photographs of ow patterns in the 1.1-mm-diameter test section of
Triplett et al. [12]. (With permission from [12].)
Bubbly ow (Fig. 4a) was characterized by distinct and distorted (non-
spherical) bubbles, typically considerably smaller in diameter than the
channel. The slug ow (Fig. 4b) is characterized by elongated cylindrical
bubbles. This ow pattern has been referred to as slug by some investigators
[9, 26] and plug by others [11, 24]. Unlike plug ow in larger channels
where the elongated gas bubbles typically occupy only part of the channel
cross section (Fig. 2), the bubbles in slug ow in microchannels appear to
occupy most of the channel cross section [12, 26].
Figures 4c and 4d display the churn ow pattern in the experiments of
Triplett et al. [12], who assumed two processes to characterize churn ow.
In one process, the elongated bubbles associated with the slug ow pattern
become unstable as the gas ow rate is increased and their trailing ends are
disrupted into dispersed bubbles (Fig. 4c). This ow pattern has been
referred to as pseudo-slug [9], churn [26], and frothy-slug by Zhao and
Rezkallah [38] in their microgravity experiments. The second process that
characterizes churn ow is the occurrence of churning waves that period-
ically disrupt an otherwise apparently wavy-annular ow pattern (Fig. 4d).
This ow pattern is referred to as frothy slug-annular by Zhao and
Rezkallah [38]. At relatively low liquid supercial velocities, increasing the
mixture volumetric ux leads to the merging of long bubbles that charac-
terize slug ow, and to the development of the slugannular ow regime
represented by Fig. 4e. In this ow pattern long segments of the channel
support an essentially wavy-annular ow and are interrupted by large-
amplitude solitary waves that do not grow sufciently to block the ow
path. With further increase in the gas supercial velocity, these large
amplitude solitary waves disappear and the annular ow pattern represen-
ted by Fig. 4f is established.
C. Rrvirv or Prviois ExirixrN1:i S1inirs :Nn Tnri TrNns
The important studies of microchannel two-phase ow that have ad-
dressed parameter ranges of interest here are reviewed, and their experimen-
tal results are compared, in this section. Table I is a summary of the
experimental investigations reviewed here.
1. General Trends
The study by Suo and Grifth [9] is among the earliest experimental
investigations. They could observe slugbubbly, slug, and annular ow
patterns. They observed no stratication, attributed its absence to the
predominane of surface tension over buoyancy, and proposed the criterion
in Eq. (6). Oya [22, 39] was concerned with ow patterns and pressure drop
TABLE I
Sixx:x or ExirixrN1:i D:1: ro Micocn:NNri :Nn N:ov P:ss:cr Tvo-Pn:sr Fiov Rrcixrs
U
'`
U
'`
Author Orientation Channel characteristics Fluids (m/s) (m/s)
WaterN
`
,
heptaneN
`
,
Suo and Grifth [9] Horizontal Circular, D1.0 and 1.4 mm
heptaneHe
Not given Not given
Barnea et al. [23] Horizontal and Glass, circular, D412.3 mm Waterair 0.0460 0.00210
vertical
Triplett et al. [12] Horizontal Pyrex, circular, D1.1 and 1.45mm; Waterair 0.0280 0.028.0
semitriangular (Fig. 3), D
1.1 and
1.49 mm
Damianides and Westwater [11] Horizontal Pyrex, circular, D15 mm; stack of ns Waterair 0.03100 0.0810
Fukano and Kariyasaki [25] Horizontal Circular, D1, 2.4, 4.9, 9 and 26 mm Waterair 0.130 0.022
and vertical
Mishima and Hibiki [26] Vertical Pyrex and aluminum, D1.054.08 mm Waterair 0.150 0.022
Barajas and Panton [24] Horizontal Pyrex, polyethylene, polyurethane, Waterair 0.1100 0.0032
uoropolymer resin
1
5
4
Narrow et al. [34] Horizontal Glass, seven-rod bundle, D
1.46 mm Waterair 0.0240 0035.0

Lowry and Kawaji [27] Vertical Rectangular, W8 cm, L 8 cm, S 0.5, Waterair 0.118 0.18
1, 2 mm
Wambsganss et al. [28] Horizontal Rectangular, W19.05 mm, L 1.14 m, Waterair 0.0530 0.22
S 3.18 mm
Ali and Kawaji [29] Horizontal/ Rectangular, W80 mm, L 240 mm, Waterair 0.1516 0.27.0
Vertical S 1.465 mm
Ali et al. [30] Horizontal/ Rectangular, W80 mm, L 240 mm, Waterair 0.1516 0.156.0
Vertical S 0.778 and 1.465mm
Mishima et al. [31] Vertical Rectangular, W40 mm, L 1.5 m, S 1.07, Waterair 0.0210 0.110
2.45, 5.0 mm
Wilmarth and Ishii [32] Horizontal/ Rectangular, L 630 mm; W15 mm and Waterair 0.028 0.074.0
Vertical S 1 mm; W20mm and S 2mm
Fourar and Bories [33] Horizontal Rectangular glass slit, W0.5 m, L 2 m, Waterair 0.010 0.0051
S 1 mm; brick slit, W14 cm, L 28 cm,
S 0.18, 0.4, 0.54 mm
Ekberg et al. [35] Horizontal Glass annuli; D
'
6.6 mm, D
8.6 mm; and Waterair 0.0257 0.16.1

D
'
33.2 mm and D
35.2 mm; L 35 cm
for both annuli
1
5
5
Fic. 5. Experimental ow regime maps for airwater ow in microchannels (Triplett et al.
[12]) and a micro-rod bundle (Narrow et al. [34]).
resulting from the conuence of airfossil liquid fuel in short vertical tubes
with 2, 3, and 6 mm diameters, and L /D ratios of 20 to 25, and could identify
nine distinct ow patterns. Because of the predominance of entrance effects,
however, Oyas data may not be representative of fully developed ow
patterns.
In the study by Barnea et al. [23], airwater ow regimes were compared
with the ow regime transition models of Taitel and Dukler [15] for
horizontal ow and Taitel et al. [12] for vertical ow, with some minor
modications. The latter models predicted their data well. Since the tube
diameters were relatively large, however, their data clearly show the effects
of gravity and test section orientation.
The two-phase ow regime data of Triplett et al. [12] are shown in Fig.
5. Regime transition lines representing a micro-rod bundle (Narrow et al.
[34]) are also depicted and are discussed in Subsection E of this part. The
ow patterns representing the four test sections of Triplett et al. are similar,
and none of the test sections supported stratied ow. The depicted ow
patterns indicate the predominance of intermittent (slug, churn, and slug
annular) ow patterns that together occupy most of the maps.
The two-phase ow regimes representing the ow of airwater mixture in
glass tubes with D1 mm to 2.4 mm reported by several authors are
Fic. 6. Comparison among airwater ow regime maps obtained in glass tubes with
D51 mm. (Symbols represent the test section (a) in Fig. 3 [12]).
depicted in Fig. 6. For the airwaterPyrex system 0
34 [40], implying
a partially wetting liquid. In Fig. 6, the ow pattern names in capital and
bold letters represent those reported by Damianides and Westwater [11]
and Fukano and Kariyasaki [25], respectively; the lowercase letters are
from Mishima and Hibiki [26]; and the symbols represent the data of
Triplett et al. [12].
Damianides and Westwater [11] were concerned with two-phase ow
patterns in compact heat exchangers. The ow patterns in their 1 and 2 mm
diameter tubes, which are of interest here, included dispersed bubbly,
bubbly, plug, slug, pseudoslug, dispersed-droplet, and annular. As noted, the
ow pattern identied as churn by Triplett et al. (Figs. 4c and 4d) appears
to coincide with the ow pattern identied as dispersed by Damianides and
Westwater. Furthermore, the slug and slugannular regimes in Tripletts
experiments (Figs. 4e and 4f ) coincide with the plug and slug ow regimes
in Damianides and Westwater, respectively. These differences are evidently
associated with subjective identication and naming of ow patterns, and
the two experimental sets are otherwise in good overall agreement.
Fukano and Kariyasaki [25] reported no signicant effect of channel
orientation on the ow patterns for channel diameters 4.9 mm and smaller.
Fukano and Kariyasaki identied only three ow patterns: bubbly, inter-
mittent, and annular. They compared the ranges of occurrence of the ow
regimes with the ow regime map of Mandhane et al. [41], with poor
agreement. However, their ow transition lines agreed with the transition
lines of Barnea et al. [23] for the latter authors 4 mm diameter tube tests.
The ow regime transition lines of Fukano and Kariyasaki representing the
data obtained with their 1 mm and 2.4 mm-diameter test sections are
depicted in Fig. 6. Their data are evidently in disagreement with the data of
Triplett et al. [12], and Damianides and Westwater [11], except for the
intermittent-to-bubbly ow transition line, where all three data sets are in
good agreement.
In the investigation by Mishima and Hibiki [26], except for void fraction
measurements, which were carried out in aluminum test sections, all
experiments were performed in Pyrex test sections, and ow regimes were
identied using a high-speed camera. The identied ow regimes were
bubbly, slug, churn, annular, and annular-mist. Mishima and Hibiki com-
pared their data representing 2.05 and 4.08 mm diameter test sections with
the ow regime transition models of Mishima and Ishii [42] with very good
agreement and argued that the latter ow regime transition models should
be applicable to capillary tubes as well. The ow transition lines of Mishima
and Hibiki [26] are displayed in Fig. 6 for the data obtained with their
2.05mm diameter test section, and are noted to disagree with the data of
other investigators. Mishima and Hibiki have indicated that the ow
patterns in their 1.05 mm diameter test section were similar to the patterns
for their 2.05 mm diameter test section.
2. Effect of Surface Wettability
The experimetal studies just discussed all utilized materials that represen-
ted partially wetting (90) conditions. In view of the signicance of
surface tension, however, the surface wettability can evidently affect the
two-phase ow hydrodynamics in microchannels. Barajas and Panton [24]
conducted experiments with air and water, using four different channel
materials. These included Pyrex (34), polyethylene (61), and
polyurethane (74) as partially wetting; and the FEP uoropolymer
resin (106) as a partially nonwetting combination. Figure 7 displays a
summary of their ow regime maps, where the data of Triplett et al. [12]
representing their 1.09mm diameter circular test section are also included
for comparison. The data of Barajas and Panton [24] representing their
Pyrex test section agreed well with the experimental ow regime of Damian-
Fic. 7. The effect of surface wettability on the airwater ow regimes. (Symbols represent
test section (a) in Fig. 3 [12]; ow regime names are from Barajas and Panton [24]).
ides and Westwater [11] representing the latter authors 1 mm and 2 mm
diameter test sections (which were also made of Pyrex), with the excep-
tion of the wavy stratied ow pattern, which did not occur in the 1 mm
test section of Damianides and Westwater. With the other partially wetting
test sections, polyethylene and polyurethane, the ow regimes and their
ranges of occurrence were similar to those obtained with Pyrex, with the
difference that with polyethylene and polyurethane the wavy ow pattern
was now replaced with a ow regime characterized by a single rivulet. A
small multirivulet region also occurred on the ow regime map representing
the polyurethane test section. The ow regimes observed with the partially
nonwetting channel FEP uoropolymer were signicantly different,
however, and compared with the partially wetting tubes, the ranges of
occurrence of the rivulet and multirivulet ow patterns were signicantly
wider.
3. Flow Regimes in Microgravity
As mentioned earlier in Part II, dimensional analysis indicates that
two-phase ow in common large channels in microgravity has important
similarities with two-phase ow in terrestrial microchannels, since in both
Fic. 8. Comparison of microchannel ow regime data of Triplett et al. for their 1-mm
diameter circular test section (Fig. 3a) with the microgravity data of Zhao and Rezkallah [43]
and Bousman et al. [45].
systems the surface tension predominates buoyancy, while inertia can be
signicant.
Experiments conducted aboard aircraft ying parabolic trajectories that
can maintain microgravity (_0.02g) for periods up to 22 seconds have been
reported by several research groups. Based on the published data, empirical
correlations for ow regime transitions applicable over wide ranges of uid
properties have been proposed by Rezkallah [43] and Jayawardena et al.
[44]. Zhao and Rezkallah [38] performed experiments in 9.52mm and
12.7mm diameter tubes, where the regimes associated with waterair
two-phase ow were identied using video cameras. Four major ow
patterns were identied: bubbly, slug, frothy slugannular, and annular. The
bubbly, slug, and annular ow regimes were morphologically similar to
those described before (see Figs. 4a, 4b, and 4f ). The frothy slugannular
regime as described by Zhao and Rezkallah, however, is similar to the ow
pattern depicted in Fig. 4d and has been called the pseudo-slug by Suo and
Grifth [9], and churn by others [12, 26]. Figure 8 compares the ow
regimes of Triplett et al. [12] representing their test section (a) (see Fig. 3),
with the experimental ow regime transition lines of Zhao and Rezkallah
[38].
Bousman et al. [45] utilized test sections with 12.7 mm and 25.4 mm
diameters, and studied the two-phase ow regime, void fraction, and liquid
lm thickness in the annular regime, using airwater, airwater glycerin
(j
'
6 cP, o63 dyn/cm), and airwater Zonyl FPS (a surfactant; re-
sulting in j
'
1 cP, o21 dyn/cm) mixtures. Bousman et al. identied four
major ow regimes similar to those reported by Zhao and Rezkallah [38].
Their experimental ow regime transition lines are also depicted in Fig. 8
and are noted to be in relatively poor agreement with the data of Triplett
et al. [12] with respect to the bubblyslug ow regime transition, and in
good agreement with respect to the transition to annular ow.
D. Fiov Rrcixr T:Nsi1ioN Monris :Nn Cori:1ioNs
As noted earlier, theoretical arguments and experimental evidence indi-
cate that ow patterns in microchannels should be insensitive to gravi-
tational eld, and therefore to channel orientation. Criteria for calculating
the maximum channel size for which gravity is inconsequential have been
proposed in [9, 10] (see Eqs. (68)). Equation (6) [9], which provides a
criterion for the dominance of surface tension over buoyancy to the extent
that for channel diameters smaller than the provided limit the two-phase
ow patterns are not affected by the channel orientation, agrees with the
experimental data of [11, 12].
The ow regime map of Mandhane et al. [41], a widely used empirical
ow regime map for common horizontal channels, was compared with
microchannel data by some authors with poor agreement [12, 25].
Barnea et al. [23] invetigated the two-phase ow regimes in small
channels (4 mmD12.3 mm) and extended the methodology of Taitel
and Dukler [15] for horizontal ow. In the original model of Taitel and
Dukler [15], transition from stratied to intermittent ow regimes is
assumed to take place when small but nite amplitude disturbances that
occur on the liquid surface grow. The position of the liquid surface (i.e., the
liquid depth in the channel) is predicted from the solution of one-dimen-
sional (1D) gas and liquid momentum conservation equations assuming
steady-state and fully developed stratied ow [15]. Barnea et al. [23]
argued that in very small channels the predominance of surface tension on
gravitational force, and not the interfacial wave instability, is responsible for
regime transition from stratied to intermittent. Based on a simple model,
Barnea et al. proposed that the latter ow regime transition occurs in very
small channels when the liquid depth in the channel, found from the solution
of 1D momentum equations for steady-state and fully developed stratied
ow, satises the following equation:
Dh
'
_
o
j
'
(1 /4)
`
. (11)
Barnea et al. [23] also argued that when D is smaller than the right-hand
side of the preceding equation, the aforementioned ow regime transition
occurs when
h
'
4
D. (12)
The modied Taitel and Dukler [15] models for horizontal ow and the
ow regime transition models of Taitel et al. well predicted the aforemen-
tioned data of Barnea et al. [23]. When applied to smaller channels,
however, the aforementioned semianalytical ow regime transition models
appear to do poorly [11, 12], with the exception of the model of Taitel and
Dukler [15] for the establishment of dispersed bubbly ow.
For the transition to dispersed bubbly ow, Taitel and Dukler [15]
derived the following relation based on a mechanistic model according to
which in the latter regime spherical bubbles have diameters within the size
range of inertial eddies predicted by Kolmogorovs theory of locally iso-
tropic turbulence, and their diameters are controlled by interation with the
latter eddies:
U
'`
U
'`
4.0
D""`"(o/j
'
)""`"
v""``
'
_
g(j
'
j
'
)
j
'

"""'
. (13)
This relation is valid as long as : 0.52, the latter approximately
representing the upper limit of void fraction for the existence of spherical
bubbles. In large horizontal channels, in addition to Eq. (13), another
criterion should be met according to which turbulence dominates over
buoyancy so that bubbles do not gather near the channel top [15, 20]. The
latter criterion is evidently redundant in microchannels where buoyancy
effect is insignicant. Although the basic assumptions for the development
of Eq. (13) are usually not met in microchannels [12], it appears to predict
well the transition line representing the development of bubbly ow [12, 25].
Based on the drift ow model, and arguing that the void fraction is a
suitable parameter for correlating ow regime transitions, Mishima and
Ishii [42] derived expressions for ow regime transition in vertical, upward
ow. The major ow regimes considered were bubbly, slug, churn, and
annular. The transition from bubbly to slug ow was assumed to occur
when a void fraction of 0.3 is reached, and led to
U
'`
3.33
:
1
U
'`
0.76
C

ogAj
j
'
2
"
. (14)
The slug-to-churn transition was assumed to occur when the liquid slugs
become unstable because of the wake effect caused by the Taylor bubbles.
For this transition, Mishima and Ishii derived
: 1 0.813
(C
1)(U
'`
U
'`
) 0.35gDAj/j
'
U
'`
U
'`
0.75
,
gDAj
j
'

gD`j
'
Aj
j`
'

`
(15)
where : is predicted using the following expression provided by the drift ux
model [46, 47]:
:
U
'`
C
(U
'`
U
'`
) V
'
. (16)
Mishima and Ishii [42] assumed that the transition from churn to annular
ow occurred either because of ow reversal in the liquid lms separating
large bubbles from the wall, or because of the disruption of liquid slugs. The
rst mechanism is applicable when
D
,
o
gAj
N""
'
[(1 0.11C
)/C
]`
, (17)
where
N
'
j
'
_
oj
' ,
o
gAj
`
. (18)
When the rst mechanism applies, the churn-to-annular transition occurs
when
U
'`
(: 0.11)
,
gDAj
j
'
. (19)
(Note than : used in the preceding equation must be larger than the
right-hand side of Eq. (15)). The second mechanism causes the regime
transition when
U
'`
ogAj
j
'

"
N"`
'
. (20)
The drift ux parameters C
"
in the preceding expressions should be found
from [48]
C
"
1.20.2j
'
/j
'
for round tubes (21)
C
"
1.350.35j
'
/j
'
for rectangular ducts. (22)
Mishima and Hibiki [26] compared their experimental data representing
two-phase ow in tubes with 1.05mm to 4.08mm diameters with the
aforementioned ow regime transition models of Mishima and Ishii [42],
apparently with good agreement, despite the fact that the empirical drift ux
model parameters (namely C
"
and V
'
) of the latter authors may be
inappropriate for microchannels. The representation of V
'
in the derivation
of the foregoing transition models based on expressions valid for larger
channels is evidently in disagreement with experimental data associated
with microchannels, where velocity slip and the buoyancy effect are both
negligible.
Zhao and Rezkallah [38] and Rezkallah [43], in correlating their micro-
gravity ow regime data, argued that when the buoyancy force is negligible
compared with surface tension while inertia is signicant, the phasic Weber
numbers are the most appropriate dimensionless parameters for the corre-
lation of ow regime transitions. Based on their experimental data, Zhao
and Rezkallah [38] correlated the bubbly-to-slug transition assuming equal
phasic velocities, and that this transition occurs at a void fraction of
: 0.18, and derived U
'`
4.56U
'`
accordingly.
Zhao and Rezkallah empirically correlated the ow regime transitions
from slug to frothy slugannular (equivalent to the slugannular ow
pattern in Triplett et al. [12]; see Fig. 4e) according to We
'`
1 and from
frothy slugannular to annular ow according to We
'`
20.
The preceding correlations well predicted the experimental data of Zhao
and Rezkallah [38]. When data from several other sources were also
considered, the aforementioned We
'`
constant correlations were found
inadequate [43]. However, correlations in the form We
'
constant could
well predict the entire data [43], where
We
'
j
'
U`
'
D/o, (23)
with U
'
representing the gas phase velocity. The latter We
'
constant
correlations are inconvenient for application, however, since the void
fraction is needed for the calculation of U
'
. Rezkallah [43], however,
showed that the data from several microgravity sources, which covered a
relatively wide range of uid properties, could be well correlated in a
two-dimensional map using We
'`
and We
'`
as coordinates, provided that
the entire ow regime map is divided into three regions: the surface tension
region including bubbly and slug ow regimes; the intermediate (transi-
tional) region including the frothy slugannular regime; and the inertial
region representing annular ow.
The experimental data of Triplett et al. [12] representing all four of their
test sections, and the experimental ow regime transition lines of Damian-
ides and Westwater [11] representing their 1 mm diameter test section, are
depicted in a ow regime map with We
'`
and We
'`
as coordinates in Fig. 9.
Fic. 9. Comparisn of microchannel data representing D1 mm with the microgravity ow
regime transition lines of Rezkallah [43]. Capitals: data from (11); Lowercase: data from [43].
As noted, the depicted data agree with the empirical transition lines of
Rezkallah [43] relatively well with respect to the ow regime transitions
among surface tension (bubbly or slug), transitional, and inertia (annular)
regions. The transitional region in Fig. 9 includes the churn (Figs. 4c and
4d) and slugannular (Fig. 4e), also referred to as pseudoslug [9] and frothy
slugannular [38] ow patterns. The data of Fukano and Kariyasaki [25]
are not shown, since the latter authors dened only three ow patterns
(bubbly, intermittent, and annular).
Bousman, McQuillen, and Witte [45], in correlating their microgravity
ow regime data, argued for the use of the void fraction as the criterion for
ow regime transition. The transition from bubbly to slug ow was assumed
to occur when : 0.4, and transition to annular ow was assumed to take
place when : 0.700.75, with the exact value depending on the liquid type.
They thus divided the entire ow regime map into three regions: bubbly,
transitional (slug and slug/annular), and annular. They calculated the void
fractions used for the derivation of the aforementioned criteria using the
drift ux model [46, 47], Eq. (16), assuming V
'
0 because of the absence
of velocity drift in nonseparated ow patterns in microgravity. Based on
their measurement of void fractions in their smaller (12.7 mm diameter) test
section, they derived C
"
1.21 and assumed that the same value was
applicable to their larger (25.4 mm diameter) test section. The : 0.4
criterion for transition from bubbly to slug ow patterns is in disagreement
with the aforementioned : 0.18 criterion suggested by Zhao and Rezkal-
lah [38]. It is also in disagreement with the 0.25 to 0.3 value in large
terrestrial channel ow data [17, 42].
Jayawarden et al. [44] considered the aforementioned data of Zhao and
Rezkallah [38] and Bousman et al. [45], as well as data from several other
sources covering uids with a wide range of properties. Similar to Bousman
et al. [45], they divided the entire ow regime map into three zones: bubbly,
slug/slugannular, and annular. They argued that the phasic supercial
Weber and Reynolds numbers were the primary dimensionless parameters
that should be used for correlating regime transitions in microgravity. They
demonstrated that the transition from bubbly to slug/slugannular for the
entire data considered could be correlated as
Re
'`
Re
'`
K
Su`` (24)
where
Re
'`
U
'`
D/v
'
(25)
Re
'`
U
'`
D/v
'
(26)
Su
Re`
'`
We
'`
oDj
'
j`
'
. (27)
Equation (24) was recommended for the range 10" Su 10`. For transi-
tion to the annular ow regime, Jayawarden et al. [44] derived
Re
'`
Re
'`
K
`
Su`` (28)
for Su 10', and
Re
'`
K
`
Su` (29)
for Su 10', where K
`
4,641.6, and K
`
2 10". Unfortunately, the
available microchannel two-phase ow data represent very narrow ranges
of the Su parameter, and meaningful comparison between the available data
and the aforementioned ow regime transition correlations of Jayawarden
et al. [44] is not feasible at this time.
E. Fiov P:11rNs iN : Mico-Ron BiNnir
Micro-tube bundles have potential applications in miniature heat ex-
changers. Experimental data dealing with two-phase ow in micro-rod and
-tube bundles are scarce, however.
Fic. 10. Cross-section of the micro-rod bundle test section of Narrow et al. [34]. (With
permission from [34].)
Narrow et al. [34] experimentally investigated the airwater two-phase
ow patterns and pressure drop in a horizontal, 23 cm long glass micro-rod
bundle that included seven rods congured as in Fig 10. Their test section was
entirely transparent, had an average hydraulic diameter of 1.29 mm, and
included several short components specically designed to reduce the
entrance and exit effects. They used a high-speed digital video camera near the
test section center to directly view subchannels 11 and 12 in Fig. 10, and
subchannels 5 and 6, which were visible through the transparent rod in front
of them. The owregime map of Narrow et al. is depicted in Fig. 5. Froth ow
was characterized by the absence of a discernible interfacial geometry. In the
stratiedintermittent ow pattern the upper subchannels in the test section
were in plug or slug ow patterns, while some of the bottom subchannels
carried single-phase liquid. In the annularintermittent ow pattern (a ow
pattern not reported in the past) the inner subchannels supported an
intermittent (slug or plug) or froth ow pattern, while the ow regime in the
peripheral subchannels was predominantly annular. In the annularwavy
ow pattern, liquid lms owed on all rods and on the test section wall.
The ow regime maps representing single channels and the micro-rod
bundle depicted in Fig. 5 are in fair agreement with respect to annular and
churn or froth ow patterns. The range of occurrence of the slugannular
Fic. 11. Empiricial ow rgime transition lines for the micro-rod bundle data of Narrow et
al. [34]. (With permission from [34].)
ow pattern in single channels coincides with the annularintermittent ow
pattern in the bundle. Furthermore, the plug/slug and the stratied-inter-
mittent ow patterns in the micro-rod bundle are replaced everywhere with
the slug ow pattern in single channels. The occurrence of the stratied
intermittent regime in the micro-rod bundle, which implies sensitivity to
buoyancy and rod bundle orientation, however, is a crucial difference in
comparison with single microchannels.
Horizontal rod bundles are used in CANDU nuclear reactors. The con-
ditions leading to bundle-wide or partial straticationinthe latter rod bundles
may lead to dryout and critical heat ux and have been experimentally studied
in the past [4951]. The micro-rod bundle ow regime map of Narrow et al.
did not agree with the available ow regime maps for large horizontal rod
bundles. Bundle-wide stratication, which can readily occur in large horizon-
tal rod bundles [4951], was not observed by Narrow et al. [34].
Narrow et al. [34] developed an empirical ow regime map, using void
fraction and mass ux as the coordinates (Fig. 11), where the void fraction
was predicted everywhere using the homogeneous ow assumption whereby
the forthcoming Eq. (32) with S1 was applied. For : 0.25, transition
occurred at a mass ux of G1000 kg/m`s. At higher void fractions, the
transition could be represented by the following equation:
ln(G) 4.7: 8.1. (30)
F. Voin F:c1ioN
Void fractions in microchannels have been measured by Kariyasaki et al.
[52], Mishima and Hibiki [26], Bao et al. [53], and Triplett et al. [54].
Fukano and Kariyasaki [25] and Mishima and Hibiki [26] also attempted
to measure and correlate the velocity of large bubbles.
Equation (16), which is a result of the drift ux model [46, 47], has long
been utilized for the correlation of void fraction in two-phase channel ow.
In the latter equation the two-phase distribution coefcient, C
"
, is the
cross-sectional average of the product of total volumetric ux and void
fraction, divided by the product of the averages of the two, while the gas
drift velocity, V
'
, is a weighted mean drift velocity of the gas phase with
respect to the mixture. The parameters C
"
and V
'
represent the global and
local interfacial slip effects, respectively, and are often determined empiri-
cally. Widely used correlations for C
"
and V
'
, developed based on experi-
mental data for common large channels, have also been used for correlating
microchannel data (Mishima et al. [31]). Correlations applied in this way
include Eq. (22). An empirical correlation based on the drift ux model
formulation has been developed by Chexal and co-workers [55], which is
based on an extensive database, includes a large number of empirically
adjusted parameters, and can address channels with various congurations.
The data base for this correlation does not include microchannels of interest
here, however.
Mishima and Hibiki [26] correlated their void fraction data for upward
ow in vertical channels, as well as the data of Kariyasaki et al. [52], using
the drift ux model [46], Eq (16), with V
'
0 for bubbly and slug ow
regimes. The distribution coefcient C
"
, however, was found to be a function
of channel diameter and was correlated according to [52]
C
"
1.2 0.510e"'"`'. (31)
where D
is in millimeters.
When the slip ratio, dened as SU
'
/U
'
, is known, the void fraction
can be calculated using the fundamental void-quality relation in one-
dimensional two-phase ow:
:
x
x S(j
'
/j
'
)(1 x)
. (32)
In homogeneous two-phase ow, S 1. A correlation for the slip ratio,
proposed by the CISE group (Premoli et al. [56]) and recommended by
Hewitt [57], can be represented as
S 1 B
_
y
1 yB
`
yB
`
`
(33)
where
y U
'`
/U
'`
(34)
B
1.578Re""
'"
(j
'
/j
'
)"`` (35)
B
`
0.0273We
'"
Re"`
'"
(j
'
/j
'
)""` (36)
Re
'"
GD/j
'
(37)
We
'"
G`D/(oj
'
). (38)
Bao et al. [53] measured pressure drop and void fraction in tubes with
0.74mm to 3.07 mm diameters, using air and water mixed with various
concentrations of glycerin. The void fractions were measured using two
solenoid valves located near the two ends of the test sections, which could
be closed simultaneously. Bao et al [53] compared their void fraction data
with predictions of several correlations, all taken from literature dealing
with commonly used large channels, and based on the results they recom-
mended the aforementioned empirical correlations for the slip ratio S
proposed by the CISE group (Premoli et al. [56]).
Butterworth [58] has shown that the void fraction correlations of
Lockhart and Martinelli [59] and several other investigators can be
represented in the generic form
1 :
:
A
_
1 x
x
(j
'
/j
'
)(j
'
/j
'
)' (39)
where A0.28, p 0.64, q 0.36, r 0.07 for Lockhart and Martinelli
[59]. Bao et al. [53] found good agreement between their measured void
fractions and the predictions of the Lockhart and Martinelli [59] correla-
tion. Triplett et al. [54] compared their void fraction data, estimated from
photographs taken from their circular test sections, with predictions of the
preceding correlations of Lockhart and Martinelli as presented by Butter-
worth [58], the aforementioned correlation due to the CISE group [56, 57],
and the correlation of Chexal et al. [55]. Figure 12 depicts a typical
comparison between the data of Triplett et al. and the preceding correla-
tions. These comparisons indicated that with the exception of the annular
ow regime where all the tested correlations overpredicted the data, the
homogeneous model provided the best agreement with experiment.
G. Tvo-Pn:sr Fiov iN N:ov Rrc1:Ncii: :Nn ANNii: Cn:NNris
Two-phase ow in rectangular channels with o O(1) mm occurs in the
coolant channels of research reactors, and during critical ow through
cracks that may occur in vessels containing pressurized uids. Investigations
have been reported by [2733]. The recent experimental investigations are
sumarized in Table I.
Experiments in vertical narrow channels, using air and water, with
consistent overall results with respect to the two-phase ow regime maps,
have been reported by Kawaji and co-workers [27, 29, 30], Mishima et al.
[31], and Wilmarth and Ishii [32]. Minor differences with respect to the
description and identication of the ow patterns exist among these inves-
tigators, however.
Lowry and Kawaji [27] used strobe ash photography and could identify
bubbly, slug, churn, and annular ow regimes. In the bubbly ow the
bubbles were small and near-spherical. The slug ow was characterized by
large irregular and attened bubbles, while the curn ow pattern contained
large irregularly shaped, as well as small, bubbles. Their ow transition lines
for the establishment of dispersed bubbly and annular ow patterns for the
test sections with o 1 and 2 mm disagreed with the models of Taitel et al.
[17]. Ali and Kawaji [29] and Ali et al. [30] performed an extensive
experimental study using room-temperature and near-atmospheric air and
water in rectangular narrow channels with six different congurations:
vertical, cocurrent up and down ow; 45 inclined, cocurrent up and
downow; horizontal ow between horizontal plates; and horizontal ow
between vertical plates. Their observed ow regimes and ow regime maps,
except for the last conguration, were similar and are displayed in Fig. 13.
The rivulet ow pattern occurred at very low liquid supercial velocities and
was relatively sensitive to the orientation of their test section. The ow
regimes for horizontal ow between vertical plates included bubbly, inter-
mittent, and stratiedwavy, and the ow regime maps for both gap sizes
(o 0.778 mm and 1.465 mm) were similar to the ow regime maps ob-
served in large pipes.
Mishima et al. [31] identied four major ow regimes in their experi-
ments: bubbly ow, characterized by crushed or pancake-shaped bubbles;
slug ow, represented by crushed slug (elongated) bubbles; churn ow, in
which the noses of the elongated bubbles were unstable and noticeably
disturbed; and annular ow. Figure 14 depicts the ow regimes in their
1.07mm gap test section. The ow regime map for their 2.4 mm-gap test
section was similar except for the presence of a small churn region in the
latter. With a gap of o 5.0 mm, however, the ow regime transition
boundaries were displaced in comparison with Fig. 14. The predictions of
the following correlation for the slugannular ow regime transition, due to
Jones and Zuber [60], are also shown in Fig. 14:
U
'`
1 :
'
:
'
U
'`
V
'
. (40)
Fic. 12. Comparison of some void fraction correlations with the measured data of Triplett
et al. [54] for the test section (a) depicted in Fig. 3: (a) homogeneous ow model; (b) Chexal
et al. [55]; (c) LockhartMartinelliButterworth, Eq. (39) [58]; (d) CISE [56]. (With
where :
'
0.8 represents the void fraction for the slugannular transition.
The correlation of Jones and Zuger [60] was based on airwater experi-
ments in a vertical retangular channel with o 5 mm.
The drift ux model, Eq. (16), with C
"
found from the aforementioned
correlation of Ishii [48], Eq. (22), and V
'
obtained from the forthcoming
Eq. (41) [60], could well predict all the void fraction data of Mishima et al.
[31] for o 1.07 and 2.45 mm, except for the annular ow regime, for which
the data and correlation deviated signicantly:
V
'
(0.23 0.13o/W)(AjgW /o). (41)
Wilmarth and Ishii [32] studied the two-phase ow regimes in vertical up
Fic. 12. (continued).
ow and horizontal ow between vertical plates. Their ow regimes for
vertical up ow are compared with the aforementioned data of Mishima et
al. in Fig. 14, where bubbly and cap-bubbly ow patterns have been
combined in the depicted bubbly ow regime zone. In comparing their
vertical ow data with models, Wilmarth and Ishii noted relatively good
agreement for the ow regime transition from bubbly to slug, with the ow
regime transition models of Mishima and Ishii [42], Eq. (14), and Taitel et
al. [17].
Fic. 14. Flow patterns in the experiments of Mishima et al. [31] and Wilmarth and Ishii
[32] in test sections with 1 mm gap. (With permission from [32].)
Fic. 13. Flow patterns in the experiments of Ali et al. [30]. (With permission from [30].)
For horizontal channels (i.e., ow between two horizontal parallel plates),
several experimental studies have been published. Differences with respect
to the ow regime description and identication among various authors can
be noted, however. The experimental ow regimes of Ali et al. [30] were
shown in Fig. 13. For the horizontal ow conguration, Wilmarth and Ishii
[32] could identify stratied, plug, slug, dispersed bubbly, and wavy
Fic. 15. Flow regimes in airwater experiments in horizontal rectangular channels. (With
annular ow patterns. In their experiments in similarly congured narrow
channels, as mentioned before, Ali et al. [30] identied bubbly, intermittent,
and stratiedwavy ow regimes only. The two ow regime maps are
compared in Fig. 15. The wavyannular ow pattern occurred at the low
U
'`
and high U
'`
range, which appears to be outside the range of the
experiments of Ali et al. [30]. The two sets of data are qualitatively in
agreement with respect to the bubblyplug/slug transition.
In the study by Wambsganss et al. [28], airwater tests were conducted
in transparent, horizontal rectangular test sections with two congurations,
one with the 3.18 mm side oriented vertically (i.e., ow between two
horizontal plates), and the other with the 19.05 mm side oriented vertically
(ow between two vertical plates). Their ow regime transition lines,
furthermore, disagreed with several ow regime maps that are based on
large channel data, including the ow regime map of Mandhane et al. [41].
Using an image processing technique, Wilmarth and Ishii [61] measured
the void fraction and interfacial concentrations in their airwater experi-
ments using vertical rectangular channels with o 1 and 2 mm, and
calculated the drift ux parameters C
"
and V
'
. The V
'
values have large
uncertainties. For the smaller gap, they found C
"
0.811 for bubbly ow,
indicating that bubbles moved with a smaller velocity than liquid; and
C
"
1 for the slug and churn ow patterns. For the larger channel they
obtained C
"
0.41 for bubbly ow and C
"
1 and 1.2 for slug and
churn-turbulent ow regimes, respectively.
Fic. 16. Flow regimes of Fourar and Bories [33] for airwater ow between horizontal at
plates with o 1 mm. (With permission from [33].)
Fourar and Bories [33] conducted experiments in glass and brick slits,
addressing low-liquid supercial velocities. Figure 16 depicts their experi-
mental ow regime map. Bubbly ow was characterized by small, isolated
bubbles, whereas in the ngering bubbly ow large, at, and unstable
bubbles were visible. The complex ow pattern was chaotic, without an
apparent structure (i.e., similar to froth ow). In the annular regime the
liquid was reported to have owed in the form of unstable lms on the walls
and may refer to the rivulet ow pattern identied by Ali et al. [30]. The
lms were replaced by entrained droplets at very low liquid ow rates.
Fourar and Bories [33] also measured the average void fraction in their
test section by careful measurement of its water contents. In all ow regimes
excluding annular, their test section void fraction closely agreed with the
correlation
: 1
_
X
1 X
`
, (42)
with the Martinelli factor, X, to be estimated from
X
_
j
'
U
'`
j
'
U
'`
`
. (43)*
Recently, Ekberg et al. [35] conducted experiments using two horizontal
Fic. 17. Flow regimes in narrow horizontal annuli. Regime names in capital and small
letters are for the small and large test sections of Ekberg et al. [34], respectively. Regime names
in bold letters are from Osamusali and Chang [63]. (With permission from [35].)
glass annuli with 1.02 mm spacing and studied the two-phase ow regimes,
void fraction, and pressure drop. The two-phase ow patterns in vertical
and horizontal large annular channels had earlier been studied by Kelessidis
and Dukler [62] and Osamusali and Chang [63], respectively. Osamusali
and Chang carried out experiments in three annuli, all with D
"
4.08 cm,
and with D
'
/D
"
0.375, 0.5, and 0.625 (o 4.75, 6.35 and 11.75 mm,
respectively), and noted that the ow patterns and their transition lines were
relatively insensitive to D
'
/D
"
. The experimental ow regime transition lines
of Ekberg et al. [35] are displayed in Fig. 17, where they are compared with
the experimental results of Osamusali and Chung [63]. These transition
lines disagree with the ow regime map of Mandhane et al. [41]. Stratied
ow occurred in the experiments of Ekberg et al. [35].
Ekberg et al. [35] compared their measured void fractions with the
predictions of the homogeneous mixture model, the correlation of Lockhart
and Martinelli [59] as presented by Butterworth [58], Eq. (39), the
correlations of Premoli et al. [56, 57], Eqs. (33)(38), and the drift ux
model, Eq. (16), with C
"
1.25 and V
'
0, following the results of Ali et
al. [30] for narrow channels. The LockhartMartinelliButterworth cor-
relation best agreed with their data.
H. Tvo-Pn:sr Fiov C:isrn nx 1nr Rrir:sr
or Dissoivrn NONCONDENSABLES
Liquid forced convection in microchannels is an effective cooling mech-
anism for systems with very high volumetric heating such as accelerator
targets, high-power resistive magnets, and high-power microchips and
electronic devices. The widely used correlations for turbulent friction factor
and convection heat transfer in large channels have been shown to be
inadequate for microchannels by some investigators, indicating that the
turbulence characteristics of microchannels may be different from those of
large channels [6466].
An interesting issue related to the forced ow of liquids in microchannels
is the potential effect of noncondensables dissolved in the liquid. Dissolved
noncondensables typically have a negligible effect in experiments with large
channels where they undergo little desorption. In microchannels, however,
because of the typically large axial pressure drops, signicant desorption of
noncondensables is possible. Such desorption will lead to the development
of a two-phase mixture, will increase the convection heat transfer coefcient,
and may be at least partially responsible for the experimental data of some
investigators, which suggest that the widely used correlations representing
heat and mass transfer in large channels underpredict the heat and mass
transfer in microchannels rather signicantly [65, 66].
Adams et al. [67, 68] recently studied the effect of dissolved noncondens-
ables on the hydrodynamic and heat transfer processes in a microchannel.
A theoretical model [67] indicated that the release of the dissolved noncon-
densables can have a relatively signicant effect on the channel hy-
drodynamics. In [68], experiments were performed in a channel 0.76 mm in
diameter with a 16 cm heated length, subject to water forced convection. The
range of experimental parameters were as follows: wall heat ux, 0.5 to
2.5MW/m`; liquid mean velocity at inlet, 2.07 to 8.53 m/s; channel exit
pressure, 5.9 bar. The water remained subcooled in all the experiments.
Typical results, depicting the Nusselt numbers at their test section exit
obtained with pure (degassed) water, and with water initially saturated with
air at a pressure equal to the test section exit pressure, are displayed in Fig.
18. The Nusselt numbers, dened as Nu hD/k
'
, were calculated assuming
single-phase liquid ow properties. The apparent dependence of the en-
hancement in Nu due to the noncondensables (air) on Re and heat ux in
fact represents the dependence of the results on the pressure drop and the
liquid temperature rise in the test section. Higher pressure drop and higher
liquid temperature rise in the test section both lead to increased desorption
of dissolved noncondensables from the liquid, and therefore to the enhance-
Fic. 18. The enhancement in the local convective heat transfer due to the presence of
dissolved air [68]. (With permission from [68]).
ment of the local heat transfer coefcient. The presence of dissolved air in
water could increase the measured Nu by as much as 17%.
An upper limit of voidage development resulting from the release of
dissolved air from water can be obtained by assuming (a) homogeneous-
equilibrium two-phase ow; (b) the release of dissolved air is accompanied
by evaporation such that the gasvapor mixture is everywhere saturated
with vapor; and (c) the liquid and the gasvapor mixture are everywhere at
equilibrium with respect to the concentration of the noncondensable, and
the latter equilibrium can be represented by Henrys law. Using these
assumptions, Adams et al. [68] showed that void fractions up to several
percent were possible, implying the occurrence of the bubbly ow regime in
their test section. For forced convection heat transfer in developed bubbly
ow, the Nu augmentation factor resulting from the presence of the gas
phase, which increases the ow velocity, can be shown to be of the order of
(1 :), with n representing the power of Re in the appropriate single-
phase forced convection heat transfer correlation [69]. The augmentation in
Nu in the data of Adams et al. [68] was signicantly higher, however,
indicating that the observed heat transfer enhancement should be primarily
due to the ow eld disturbance caused by the formation and release of
microbubbles on the channel walls.
IV. Pressure Drop
A. GrNr:i Rrx:xs
Pressure drop is among the most essential design parameters for piping
systems. However, frictional pressure drop in microchannels, for both single
and two-phase ow, is not well understood. Turbulence characteristics in
microchannels are known to be different from those in large channels. Thus,
although the predictions of theoretial solutions for laminar ow closely
agree with measured friction factors in microchannels [70, 71], most of the
recent experimental investigations indicate that the well-proven correlations
for turbulent ow in large channels fail to correctly predict frictional
pressure drop in circular and rectangular microchannels [7174]. Some
investigators, on the other hand, have reported good agreement between
their data and turbulent friction factor correlations for smooth pipes [26],
and for pipes with carefully measured roughness characteristics [70].
Measurement uncertainties and uncertainties associated with roughness
evidently contribute to the disagreement among published experimental
data, while the lack of adequate understanding of turbulence in microchan-
nels is believed to be the main reason for disagreement between existing data
and the commonly used correlations for large channels.
B. Fic1ioN:i Prssir Doi iN Tvo-Pn:sr Fiov
The existing methods and correlations applied in microchannel pressure
drop are mostly similar to those applied to large channels. A brief review of
the principles of two-phase frictional pressure drop modeling, and the
predictive methods that have been applied to microchannels, is provided in
this section.
The simplest method for calculating the two-phase frictional pressure
drop is to assume homogeneous ow and apply an appropriate single-phase
turbulent friction factor correlation using the homogeneous two-phase
mixture properties everywhere. Thus,
cP
cz
''
f
''
G`
2j
"
D
, (44)
where
j
"
x
j
'
1 x
j
'

. (45)
Now, applying the Blasius correlation for turbulent friction factor, for
example, one can write
f
''
0.316Re"``
''
(46)
where
Re
''
GD/j
''
. (47)
One of the most popular correlations for homogeneous mixture viscosity is
that of McAdams [75]:
j
''
x
j
'
1 x
j
'

. (48)
In the preceding equations x represents the ow quality. In dealing with
a single-component two-phase ow (i.e., the ow of a liquid and its own
vapor), and further assuming thermal equilibrium between the two phases
(the homogeneous equilibrium mixture model, HEM), xx
, where
x
(h h
)/h
. (49)
The homogeneous ow assumption has limited applicability, however,
and the foregoing method for calculating two-phase frictional pressure drop
is inaccurate in most applications.
The two-phase multiplier method is the most common technique for
correlating two-phase frictional pressure drop in channels [59], according
to which
cP
cz
''
u`
''

cP
cz
''
u`
''

cP
cz
''
(50)
or
cP
cz
''
u`
'

cP
cz
'
u`
'

cP
cz
'
, (51)
where
cP
cz
''
and
cP
cz
'
represent single-phase frictional pressure gradients in the channel when pure
liquid at mass uxes G and G(1 x), respectively, ows in the channel. The
terms
cP
cz
'"
and
cP
cz
'
are dened similarly when pure gas at mass uxes G and Gx, respectively,
ows in the channel. Equations (50) and (51) evidently provide four ways
for representing two-phase pressure drop, by correlating any of u
'"
, u
'"
,
u
'
, or u
'
. Based on the two-phase multiplier concept, many correlations
have been proposed in the past. These correlations are usually applicable
without restriction to all ow regimes.
Martinelli and co-workers were the rst to correlate u`
'
and u`
'
, graphi-
cally in terms of the Martinelli factor X, dened as
X`
(cP/cz)
'
(cP/cz)
'
. (52)
Lockhart and Martinellis graphical representation of u
'
have been utilized
for the development of algebraic correlations by some investigators [76, 77].
A widely used correlation, suggested by Chisholm and Laird [78], is
u`
'
1 C/X1/X`, (53)
where C may have values between 10 and 20. An alternative expression for
this correlation is [79]
u`
'
1 CXX`. (54)
In the foregoing equations the constant C depends on whether the gas and
liquid phases, when owing alone, are laminar (viscous) or turbulent, and
its recommended values are as follows: C20 for turbulentturbulent ow;
C12 for viscous liquid and turbulent gas; C10 for turbulent liquid and
viscous gas; and C5 for viscousviscous ow [78]. A correlation repre-
senting u`
''
as a function of ow quality, x, AP
''
/AP
''
, and n with n
representing the power of Re in the single-phase friction factor correlation,
has also been proposed by Chisholm [80, 14]. Other correlations that
account for the effect of phasic properties and the two-phase mixture mass
ux have been described in [14].
A widely used correlation, proposed by Friedel [81], is based on a vast
database covering an extensive parameter range. For horizontal channels,
the Friedel correlation is
u`
''
A3.21x"``(1x)"``"
j
'
j
'
""
j
'
j
'
""
1
j
'
j
'
"`
Fr"""`"
''
We""``
''
,
(55)
where
A(1 x)` x`j
'
f
''
(j
'
f
''
) (56)
Fr
''
G`
gDj`
''
(57)
We
''
G`D
oj
''
. (58)
According to Friedel, the single-phase friction factors should be calculated
from
f '
''
0.25[0.86859lnRe
'
/(1.964 lnRe
'
3.8215)}]` (59)
when Re
'
1055, where
Re
'
DG/j
'
. (60)
When Re
'
1055, the appropriate laminar Fanning friction factor relation
is used.
The following correlation, derived by Beattie and Whalley [82], is
convenient to use because of its simplicity. In this method, homogeneous
mixture ow is assumed and Eqs. (44)(47) are applied. The mixture
viscosity, however, is dened as
j
''
:
"
j
'
j
'
(1 :
"
)(1 2.5:
"
) (61)
where :
"
, the homogeneous void fraction, is found from Eq. (32) with S 1.
Beattie and Whalley recommend that the single-phase friction factor be
calculated, for all values of Re
''
, from the Colebrook [83] correlation,
1
f '
''
3.48 4 log
"
_
2
c
D
9.35
Re
''
f '
''
(62)
where f ' f /4 represents the Fanning friction factor.
The preceding correlations, as mentioned earlier, do not explicitly account
for ow patterns and have been developed to cover various ow patterns
covered by their databases. The fractional pressure drop, like many other
important hydrodynamic phenomena, depends on ow pattern, however.
Flow regimedependent models have been derived for the stratied [84]
and annular [85, 86] ow regimes in the past, because of the relatively
simple morphology of the latter ow patterns. These models, in addition to
correlating the wall friction, also account for the gasliquid interfacial
friction.
C. Rrvirv or Prviois ExirixrN1:i S1inirs
The occurrence of ashing two-phase ow in refrigerant restrictors was
the impetus for early studies of single- and two-phase pressure drop in long
microchannels [8790]. Two-phase pressure drop in these studies was
generally modeled using the homogeneous ow model. Table II is a
summary of the more recent experimental investigations that are reviewed
here.
Koizumi and Yokohama [91] modeled their experimental data using Eqs.
(44)(48), where the liquid and vapor phases were assumed to be at
equilibrium everywhere, and the two-phase viscosity was calculated from
j
''
j
"
v
'
based on the argument that the ashing two-phase ow in their
simulated refrigerant restrictor was predominantly bubbly. Although their
calculations were in good agreement with their total measured pressure
drops, they noted that the preceding model was in fact signicantly in error
since it did not account for the evaporation delay in their experiments.
Further investigation into the ashing and two-phase ow processes as
associated with refrigerant restrictors was conducted more recently by Lin
et al. [92]. They measured pressure drop with single-phase liquid ow, and
noted that their data could be well predicted using the Churchill [93]
correlation,
f 8
_
8
Re
`
1
(A B)``
`
, (63)
where
A
2.457 ln
_
1
7
Re
""
0.27
c
D
'
(64)
B
37530
Re
'
(65)
Based on an argument similar to that of Koizumi and Yokohama [91], Lin
et al. [92] applied the homogeneous-equilibrium mixture model (Eqs. (44),
(45), and (47)) for two-phase pressure drop calculations. For calculating the
two-phase friction factor, f
''
, they used the aforementioned Churchill
correlation (Eq. (63)) by replacing Re with Re
''
, and over a quality range
of 0 x0.25 they empirically correlated the two-phase mixture viscosity
according to
j
''
j
'
j
'
j
'
x(j
'
j
'
)
, (66)
with n 1.4 providing the best agreement between model and data.
TABLE II
Sixx:x or ExirixrN1:i D:1: ro Micocn:NNri :Nn N:ov P:ss:cr Tvo-Pn:sr Fiov Prssir Doi
Author Channel characteristics Fluid(s) Flow Range
Koizumi and Yokohama [91] Adiabatic circular channels, D1, 1.5 mm R-12 Re
'
4.310"6.410"
Lin et al. [92] Adiabatic copper tubes, R-12 G1.4410`5.0910` kg/m`s
D0.66 mm (c 2 jm), 1.17 mm (c 3.5 jm) P
'
6.313.2 bar; AT
'""'"
017K
Ungar and Crowley [94] Adiabatic circular tubes, D1.463.15 mm Ammonia Re
'
700; 450 Re
'
1.110", 0.09x0.98
Bao et al. [53] Glass and copper tubes, D0.741.9 mm Waterair, 15Re
'
210`; 0.05Re
'
410"
wateraqueous,
glycerin solutions
Bowers and Mudawar Heated copper channels (subcooled boiling) R-113 U
'`
*7.7 m/s
[95] D0.51, 2.54 mm
Fukano and Kariyasaki [25] Circular channels, D126mm Waterair 0.02*U
'`
*2 m/s; 0.1*U
'`
*30 m/s
Mishima and Hibiki [26] Pyrex and aluminum circular channels, Waterair 0.02*U
'`
*2 m/s; 0.1*U
'`
*50 m/s
D1.054.08 mm
Triplett et al. [12] Pyrex circular channels, D1.1, 1.45 mm; Waterair 0.02*U
'`
*8 m/s; 0.02*U
'`
*80 m/s
semitriangular channels, D
1.1, 1.49 mm
Narrow et al. [34] Glass seven-rod bundle, D
1.46 mm Waterair 0.03*U

'`
*5 m/s; 0.02*U
'`
*40 m/s
Lowry and Kawaji [27] Rectangular, W8 cm, L 8 cm, S 0.5, 1, 2 mm Waterair 0.1*U
'`
*8 m/s; 0.1*U
'`
*18 m/s
Ali and Kawaji [29] Rectangular, W80 mm, L 240mm, S 1.465 mm Waterair 0.15*U
'`
*16 m/s; 0.2*U
'`
*7.0 m/s
Ali et al. [30] Rectangular, W80 mm, L 240 mm, S 0.778 Waterair 0.15*U
'`
*16 m/s; 0.15*U
'`
*6.0 m/s
and 1.465mm
Mishima et al. [31] Rectangular, W40 mm, L 1.5 m, Waterair 0.1*U
'`
*10 m/s; 0.02*U
'`
*10 m/s;
S 1.07, 2.45, 5.0 mm 0.5*x*100
Fourar and Bories [33] Rectangular glass slit, W0.5 m, L 1 m, Waterair 0.005*U
'`
*1 m/s; 0.0*U
'`
*10 m/s
S 1 mm; brick slit, W14 cm, L 28 cm 0.1*x*40
S 0.18, 0.4, 0.54 mm
Yan and Lin [96, 97] Circular, D2 mm; 28 parallel pipes with R-134a Re
'
20012,000; mean quality 0.10.95
condensation and evaporation
Ekberg et al. [35] Glass annuli; D
6.6mm, D
8.63mm; and Waterair 0.1*U

'`
*6.1 m/s; 0.02*U
'`
*57 m/s
D
'
33.15 mm, D
35.2 mm; L 35 cm
1
8
5
Ungar and Cornwell [94] conducted experiments at high quality
(0.09 x 0.98); their data were therefore predominantly in the annular-
dispersed two-phase ow regime. They calculated their experimental two-
phase friction factors using channel-average properties, neglecting the effect
of axial variations in vapor properties and quality, and compared them with
predictions of several correlations. The homogeneous-equilibrium mixture
model (Eqs. (44)(47)), when applied with McAdams correlation for
mixture viscosity, Eq. (48), provided the best agreement with data. An
empirical correlation for vertical annular ow, due to Asali et al. [85], also
predicted their data well.
Bao et al. [53] performed an extensive experimental study using air and
aqueous glycerin solutions with various concentrations and calculated the
experimental friction factors using channel-average properties. They com-
pared their data with the correlations of Lockhart and Martinelli [59],
Chisholm [80], Friedel [81], and Beattie and Whalley [82], apparently
without consideration for the effect of channel roughness. With the excep-
tion of the correlation of Lockhart and Martinelli [59], which did relatively
well for Re
'
1000, all the tested correlations failed to predict the data well.
By implementing the forthcoming simple modication into the correlation
of Beattie and Whalley [82], Eq. (61), the latter correlation predicted all
their data well. The correlation of Beattie and Whalley is based on the
application of the homogeneous ow model, Eqs. (44), (45), and (47), and
the ColebrookWhite correlation (Eq. (62)) for the friction factor over the
entire two-phase Reynolds number range. Bao et al. [53] modied the
correlation of Beattie and Whalley [82] simply by using f '
''
16/Re
''
in
the Re
''
1000 range.
Bowers and Mudawar [95] studied high heat ux boiling in mini
(D2.54 mm) and micro (D0.5 mm) channels. They applied the homo-
geneous-equilibrum model, Eqs. (44), (45), and (49), assuming f
''
0.02
The homogeneous-equlibrium model could well predict the total experimen-
tal pressure drops. Because of the signicance of the acceleration pressure
drop in most of their tests, the accuracy of the homogeneous-equilibrium
model for the calculation of the frictional pressure drop in their experiments
cannot be directly assessed. The good agreement between model-predicted
and measured total pressure drops, nevertheless, may indicate that the
homogeneous-equilibrium model in its entirety is adequate for similar
applications.
The experiments of Fukano and Kariyasaki [25] were described in
Subsection B of Section III (see Table I). They compared their two-phase
pressure drop data with the correlation of Chisholm [78, 79] (Eq. (53) or
(54)), indicating a large discrepancy. The discrepancy was particularly
signicant in the intermittent (plug and slug) ow patterns. Fukano and
Kariyasaki [25] indicated that the pressure loss associated with the expan-
sion of the liquid as it ows from the lm region surrounding elongated
bubbles into the liquid slugs is a signicant component of the total frictional
pressure loss.
Mishima and Hibiki [26] reported that with single-phase liquids, the
Blasius correlation well predicted their turbulent friction factor data, where-
as in the laminar range their data agreed with the HagenPoussuille
relation within 2%. For calculating the two-phase frictional pressure drop,
they neglected the acceleration along their test sections, and chose to modify
the Chisholm correlation [78, 79], Eqs. (53) or (54), by empirically correlat-
ing the constant C according to
C21(1 e"`"') (67)
where the channel diameter, D, is in millimeters.
Triplet et al. [12] measured the frictional pressure drop for airwater ow
in circular and semitriangular microchannels with D
-1.1 to 1.49mm (see

Fig. 3). They compared their measured frictional pressure drops with the
predictions of the homogeneous mixture method, Eqs. (44)(48), and the
correlation of Friedel [81], Eqs. (55)(60). They noted that because of the
signicant axial variation of pressure in microchannels, the gas density
cannot be assumed constant. They applied a one-dimensional model, based
on the numerical solution of one-dimensional mass and momentum conser-
vation equations for the calculation of pressure drops, using the aforemen-
tioned correlations for two-phase wall friction. Overall, the homogeneous
mixture model better predicted the data. Both correlations did poorly when
applied to the annular ow regime data, however.
Yan and Lin recently measured the two-phase pressure drop and heat
transfer associated with evaporation [96] and condensation [97] of Refrig-
erant 134a in a horizontal, 28-tube bundle consisting of tubes with 2 mm
inner diameter. In calculating the frictional pressure drops for the tubes they
needed to estimate the pressure losses at inlet and exit to their tube bundle,
and the deceleration pressure change associated with the condensing two-
phase ow. Following Yang and Webb [98], who investigated the two-
phase pressure drop associated with the adiabatic two-phase ow in
extruded aluminum tubes, Yan and Lin based the aforementioned estimates
on the test section average quality and an average void fraction. The average
void fraction was calculated using the following slip ratio correlation
originally derived by Zivi [99] based on the assumption of minimum
entropy generation in steady-state, annular two-phase ow:
S(j
'
/j
'
)`. (68)
Yan and Lin calculated the entrance and exit pressure losses using common-
ly applied contraction and expansion models They correlated the experi-
mental frictional pressure drops obtained in this way using a method
originally suggested by Akers et al. [100], and recently applied by Yang and
Webb [98], according to which
AP
4 f
''
L
D
G`
'
2j
'
, (69)
where (Akers et al. [100]):
G
'
G[(1 x
"
) x
"
(j
'
/j
'
)"`], (70)
where x
"
is the average channel quality. For condensation, Yan and Lin
[97] obtained
f
''
498.3Re"`"
'
, (71)
where
Re
'
G
'
D/j
'
. (72)
Using their experimentally measured frictional pressure drops, Yan and
Lin calculated the experimental friction factors in their evaporating two-
phase ow tests based on the homogeneous mixture assumption, with the
homogeneous density j
"
dened based on channel average quality and void
friction:
AP
4 f
''
L
D
G`
2j
"
(73)
They, however, correlated the friction factors in terms of the aforementioned
equivalent Reynolds number, dened in Eq. (72), according to
f
''
0.11Re"
. (74)
Recently Narrow et al. [34] investigated the hydrodynamic processes
associated with airwater two-phase ow in a seven-rod micro-rod bundle.
Their experiments were described in Section III, E. They measured the
pressure drop in their experiments and compared their data with the
predictions of the homogeneous mixture model (Eqs. (44)(48)) and the
correlation of Friedel [81], Eqs. (55)(58) for two-phase frictional pressure
drop. Neither correlation could satisfactorily predict the data over the entire
ow regime map. The correlation of Friedel could predict most of the data
typically within a factor of 2, except for the data representing very low
supercial velocities. The homogeneous mixture model, on the other hand,
consistently underpredicted the frictional pressure drop for all ow patterns,
ecept for the annular/intermittent and plug/slug ow patterns.
D. Fic1ioN:i Prssir Doi iN N:ov Rrc1:Ncii:
:Nn ANNii: Cn:NNris
For laminar, single-phase ow, the measured friction factors in rectangu-
lar channels agree well with theory [101], For turbulent ow, the friction
factor in rectangular channels is known to depend on the Reynolds number
based on hydraulic diameter, as well as on the channel aspect ratio. Jones
[102] derived a simple method that allows for the application of smooth
pipe laminar and turbulent friction factor correlations to rectangular chan-
nels. Accordingly, a laminar equivalent diameter, D
''
, is dened as
D
''
u*D
, (75)
where the shape factor u* is a function of the aspect ratio, W/o, and is
formulated using the theoretical solution for friction factor in rectangular
channels, such that for laminar ow
f 64/Re
, (76)
where the modied Reynolds number Re* is dened according to
Re
GD
''
j
. (77)
The function u* can be found from the analytical solution of Cornish [103].
The following simple correlation, however, agrees with the aforementioned
analytical solution within 2% [102]:
u*
2
3
11
24
o
W
2
o
W
. (78)
Using Eqs. (75)(78), the turbulent smooth pipe ow correlations can be
applied to rectangular channels.
Kawaji et al. [27, 29, 30], Mishima et al. [31], and Fourar and Bories [33]
have reported two-phase pressure drop data dealing with narrow rectangu-
lar channels (see Table II).
Lowry and Kawaji [27] indicated that the wall roughness was only 1.5 jm
in their test section and that the Blasius correlation for fully turbulent
single-phase liquid ow did extremely well for their data. This result is
evidently in disagreement with the aforementioned well-known effects of the
aspect ratio on the turbulent friction factor in narrow channels. Lowry and
Kawaji compared their two-phase pressure drop data with the predictions
of the correlation of Lockhart and Martinelli [59] and indicated that the
correlation did not account for the clear dependence of the data on mass
ux. Their experimental two-phase multiplier, furthermore, was a weak
function of U
'`
and o, and a strong function of U
'`
. The experiments of Ali
et al. [30], described earlier in Section III, G, were performed in narrow
rectangular channels with o 0.778 mm and 1.465 mm, with the following
six orientations: vertical upward and downward ow; 45-inclined upward
and downward ow; horizontal ow between horizontal plates; and hori-
zontal ow between rectangular plates. With the exception of the last ow
conguration, the effect of orientation on pressure drop was quite small, and
the correlation of Chisholm and Laird [78], Eq. (53), agreed well with their
data using C values between 10 and 20, depending on mass ux. In these
comparisons, for the calculation of the single-phase frictional pressure
gradients, Ali et al. [30] used the following expressions for the DArcy
friction factors, which they derived by curve tting their own experimental
data. For laminar ow (Re 2300), f 95/Re for o 0.778mm, and
f 94/Re for o1.465 mm. For turbulent ow (Re3500),
f 0.339Re"`` for o 0.778 mm, and f 0.338Re"`` for o 1.465 mm.
For the transition 2300 Re 3500 range, they applied a linear interpola-
tion on a logarithmic scale. For horizontal ow between vertical plates, the
effect of mass ux on the two-phase multiplier was strong, and xed values
of C could not correlate the data. For the stratied ow regime in the latter
conguration, Ali et al. [30] derived the following expression, based on a
simple separated ow model that can be applied when the two phases are
both either laminar or turbulent:
`
'
[1 X`'"`']`" (79)
Here, m is the power of Re in the appropriate single-phase friction factor.
This expression well predicted the turbulentturbulent data of Ali et al.
The experiments of Mishima et al. [31] were described in Section III, G.
For single-phase ow pressure drop, their results were in agreement with the
recommendations of Jones [102]. They also noted good agreement between
their data and a correlation proposed by Sadatomi et al. [104]. For
two-phase frictional pressure drop, Mishima et al. chose the correlation of
Chisholm and Laird, Eq. (53), for modication and correlated their data
according to
C21 tanh (0.199D
) 21[1 1056 exp(0.331D
)], (80)
where D
is in millimeters.
Fourar and Bories [33] noted that, except for the annular ow regime,
the frictional pressure drops could be well correlated using
f
96
Re
"
, (81)
where the mixture Reynolds number, Re
"
, is obtained from
Re
"
2Sj
"
(U
'`
U
'`
)
j
"
(82)
j
"
:j
'
(1 :)j
'
(83)
j
"
j
'
U
'`
U
'`
U
'`
. (84)
Annular ow appeared to occur at Re
"
4000 apparently corresponding to
the establishment of turbulent mixture ow.
John et al. [105] performed an extensive series of experiments dealing
with critical ow in cracks and slits. The work of John et al. [105] is
discussed in Section VII. In view of the important effect of friction on critical
mass ux, John et al. measured and correlated the single-phase friction
factors in their test sections. They could correlate their data according to
f
3.39 log
"
D
2c
0.866
`
(85)
where c represents the surface roughness.
Ekberg et al. [35] measured pressure drops associated with airwater
two-phase ow in two horizontal annuli with o 2 and 3 mm (see Table II).
When compared with predictions of a one-dimensional model based on the
homogeneous ow assumption, Friedels correlation [81] for two-phase
frictional pressure drop, Eqs. (55)(60), predicted the experimental data
better than the homogeneous ow wall friction model.
V. Forced Flow Subcooled Boiling
A. GrNr:i Rrx:xs
Cooling by the ow of a highly subcooled liquid, and subcooled boiling,
are the heat transfer regimes of choice in numerous applications, because of
the extremely high heat uxes they can sustain at relatively low heated
surface temperatures. The extensive past studies have been reviewed in a
number of textbooks and monographs, including [14, 106108]. The work
by Tuckerman and Peasa [109], and that pursued by many other investiga-
tors, has shown that the inclusion of networks of microchannels cooled by
subcooled liquids in circuit boards, in particular, can provide effective
cooling for extremely high thermal loads. A good review of the past research
dealing with single-phase ow heat transfer in microchannels can be found
in [110]. An extensive summary is provided by Duncan and Peterson [111].
These experimental studies have also shown that the hydrodynamic and
heat transfer characteristics of microchannels are different from those of the
commonly applied large channels. These differences, the causes of which are
not fully understood, include the Reynolds number range for laminar-to-
turbulent ow pattern transition, and the wall friction factor and forced
convection heat transfer in the transition and fully turbulent regimes.
Despite its evident signicance, forced-ow boiling in microchannels with
De 0.1 to 1 mm has attracted little investigation in the past, although heat
transfer in small channels representative of modern compact heat ex-
changers (with D
values typically in the few-millimeters range) has been

investigated by several investigators recently [112115]. The limited avail-
able data for microchannels, nevertheless, indicate major differences between
small and commonly used large channels, with respect to the basic bubble
ebullition phenomenology. The experimental data of Wambsganss et al.
[114] and Tran et al. [115], for example, indicate that, unlike in large
channels, in small channels used in compact heat exchangers the nucleation
process, and not the forced convection process, is the dominant boiling heat
transfer mechanism at high ow qualities.
In the forthcoming sections, the subcooled boiling phenomena, for which
recent investigations have led to reasonably consistent results, are discussed.
Table III is a summary of the recently published experimental studies
dealing with subcooled boiling phenomena in microchannels.
B. Voin F:c1ioN Rrcixrs iN Hr:1rn Cn:NNris
The voidage development in heated channels with subcooled inlet condi-
tions has been studied extensively in the past, and is described in textbooks
[107, 116]. Figure 19 is a schematic of the axial void fraction distribution
along a uniformly-heated channels with subcooled liquid inlet conditions.
This schematic is consistent with experimental observations with water at
high pressures in commonly used large channels [117]. The ow eld
upstream of point A is single-phase liquid, and bubbles attached to the wall
can be seen at and beyond point A, referred to as the onset of nucleate
boiling (ONB) point. Bubbles remain predominantly attached to the wall
upstream of point B. Beyond the latter point bubbles can be seen detached
from the wall. Beyond point C, referred to as the onset of signicant void
(OSV), or the point of net vapor generation (NVG), the detached bubbles
can survive condensation and a rapid increase in the gradient of the void
fraction curve is observed.
Recent low-pressure experiments with water by Bibeau and Salcudean
[118120], also carried out in test sections representative of common large
channels, have shown that the phenomenology implied in the schematic of
TABLE III
Sixx:x or RrcrN1 ExirixrN1:i D:1: Dr:iiNc vi1n Sincooirn BoiiiNc iN Sx:ii :Nn Micocn:NNris
Author Channel characteristics Fluid Parameter range Phenomena studied
Inasaka et al. [130] D1, 3 mm; L 1, 3, 5, 10 cm; vertical Water G7000, 13,000, 20,000kg/m`s; ONB, OSV, OFI
T
'
20,60C; P
`''
1 bar
Vandervort et al. D0.32.6mm; L 2.566 mm; vertical Water G840042,700 kg/m`s; ONB
[131] P
`''
122 bar
Kennedy et al. D1.17, 1.45 mm, L 22 cm; Water G8004500 kg/m`s; ONB, OFI
[132] L
"
16 cm; horizontal P
`''
3.4410.34 bar
Roach et al. D1.17, 1.45 mm, circular; Water G220790 kg/m`s; OFI
[135] D
"
1.13 mm, semitriangular; P
`''
2.49.33 bar
L 22 cm; L
"
16 cm; horizontal
Blasick et al. Annuli with r
'
6.4 mm and Water G851,428 kg/m`s; OFI
[139] o 0.7241.0 mm; L
"
17.419.7 cm; P
`''
3.4410.34 bar
horizontal
Peng and Wang Rectangular, width0.20.8 mm, Water, U
''
0.22.1 m/s, AT
`''
6590C Liquid single-phase and
[153] depth0.7 mm; L 4.5 cm; methanol for water; U
''
0.21.5 m/s, subcooled boiling heat
horizontal AT
`''
4550C for methanol; transfer
P
`''
1 bar
Peng and Wang Rectangular, width0.6 mm; Water U
''"
1.54.0 m/s; T
''
3060C; Liquid single-phase and
[110] depth0.7 mm; L 6.0 cm P
`''
1 bar subcooled boiling heat
transfer
Hosaka et al. D0.5, 1, 3 mm; L /D50, vertical R-113 G930032,000 kg/m`s; Subcooled boiling heat
[155] AT
`''
5080C; P1124 bar transfer and CHF
1
9
3
Fic. 19. Void fraction variation along a uniformly heated channel.
Fig. 19 is not accurate, at least for low-pressure subcooled boiling of water,
where the region of attached voidage (the region beween points A and B in
Fig. 19, where bubbles are presumed to grow and collapse while attached to
the wall) is essentially nonexistent, and bubbles that are detached always
slide on the wall before being injected into the liquid core. The phenomenol-
ogy depicted in Fig. 19, nevertheless, has been the basis of successful
correlations [121, 122] and analytical models [123125] in the past.
Figure 20 schematically depicts the pressure drop-ow rate characteristics
(the demand curve) of a heated channel subject to a constant heat load
(constant heat ux for a uniformly heated channel). The demand curve can
be used for the analysis of static instabilities [126, 127]. When the channel
is part of a forced or natural circulatory loop, the segment of the heated
channel demand curve with negative slope (between points OFI and S in
Fig. 20) can be unstable, and the onset of ow instability (OFI) point is
dened as the relative minimum point on the demand curve. The occurrence
of OFI is due to the increase in the channel pressure drop which results from
Fic. 20. Pressure drop-mass ow rate characteristic curve of a uniformly heated channel.
subcooled boiling voidage, and in experiments with steady heat ux (or
steady mass ow rate), OFI is known to occur at a ow rate slightly lower
(or a heat ux slightly higher) than the ow rate (or the heat ux) that leads
to OSV. The OSV point can thus be considered as a conservative estimate
of OFI.
C. ONsr1 or Nicir:1r BoiiiNc
The onset of nucleate boiling (ONB), or boiling incipience, has been
modeled by several authors. A good compilation of the existing correlations
can be found in Marsh and Mudawar [128]. Most of the models are based
on the assumption that at boiling incipience stationary and stable bubbles,
attached to wall crevices, exist, and that the steady-state liquid temperature
prole is tangent to the temperature prole predicted by the Clapeyron
bubble superheat prole. According to the model of Bergles and Rohsenow
[129], one of the most widely used models of this kind, bubbles attached to
the wall at the ONB point are hemispherical, and the heat ux and wall
temperature at ONB are related to each other according to
q"
"
k
'
T
"
T
'
(r
"
)
r
"
(86)
q" h
''
(T
"
T
'
), (87)
where T
'
(r
"
) is the local liquid temperature at a distance r
"
away from the
heated surface. The aforementioned tangency criterion, furthermore, re-
quires that
T
'
(r
"
) T
"
(88)
cT
'
cy
''"
cT
"
cr
"
, (89)
where T
"
, the bubble temperature, accounts for the superheat required for
mechanical equilibrium between the bubble and its surroundings based on
Clapeyrons relation:
T
"
T
`'
T
"
T
`'
R
Mh
ln
1
2o
r
"
P
(90)
An empirical curve t to the predictions of the above equations for water in
the 1 bar P136 bar, was derived by Bergles and Rohsenow [129] as
q"
"'`"
5.30P`'[1.8(T
"
T
`'
)
'`"
], n 2.41/P""``" (91)
where q"
"'`"
is in W/m`, P is in kPa, and temperatures are in K.
The preceding correlation of Bergles and Rohsenow has been compared
with microchannel data with water as the working uid by Inasaka et al.
[130], Vandervort et al. [131], and Kennedy et al. [132].
Inasaka et al. [130] performed experiments in heated tubes with D1
and 3 mm, with G7000 to 20,000 kg/m`s. The correlation of Bergles and
Rohsenow, Eq. (91), predicted the data of Inasaka et al. relatively well.
Signicant scatter can be noted in their comparison results, however, with
maximum dicrepancies of about _50% in the prediction of q"
"'`"
.
Vandervort et al. [131] carried out an experimental investigation of
subcooled boiling phenomena in microchannels with diameters in the
D0.3 to 2.6 mm range, subject to high heat uxes, with water as the
working uid. They reported that the correlation of Bergles and Rohsenow
predicted their ONB data well. They also applied the model of Bergles and
Rohsenow (Eqs. (87)(91)), for the estimation of the released bubble
diameters, and showed that the released bubbles were typically only a few
micrometers in diameter. Using the estimated bubble sizes, they calculated
the order of magnitude of various forces that act on the bubbles, and
Fic. 21. Comparison between the ONB data of Kennedy [132] and the correlation of
Bergles and Rohsenow [129]. (D1.17 mm). (With permission from [132].)
showed that the thermocapillary (Marangoni) force and the lift force
resulting from the ambient liquid velocity radient are signicant forces that
must be considered in modeling of bubble ebullition phenomena in micro-
bubbles. A more detailed discussion of these forces is presented in the
forthcoming Subsection E of this section.
Kennedy et al. [132] studied the ONB and OFI phenomena in heated
microchannels with D1.17 and 1.45 mm, using water as the working uid.
Figure 21 displays the comparison between the ONB data for their 1.17 mm
test section and the correlation of Bergles and Rohsenow. The correlation
of Bergles and Rohsenow systematically overpredicted the experimental
data, typically by a factor of 2. The correlation, however, agreed reasonably
well (with a slight systematic overprediction) with the data of Kennedy et
al. representing their 1.45 mm diameter test section.
Inasaka et al. [130] and Kennedy et al. [132] utilized the pressure drop
characteristic curves of their test sections (similar to Fig. 20) for specifying
the ONB conditions. Inasaka et al. identied the heat ux that led to the
occurrence of ONB at the exit of their test sections, for given (constant) inlet
temperature and mass ux, as the minimum point on the AP/AP
''
versus
q"
"
curve, with AP representing the measured pressure drop in the experi-
ment with heated wall, and AP
''
representing the channel pressure drop
with adiabatic single-phase liquid ow only.
Kennedy et al. [132] also identied the ONB point on each pressure drop
versus mass ow rate (demand) curve (Fig. 20) by comparing the experimen-
tal demand curves with pressure dropow rate characteristics representing
single-phase liquid ow. The ONB occurs at the point where the slopes of
the two curves deviate. They could also recognize an easily audible whistle-
like sound from their test section, before the onset of ow instability (OFI)
occurred, which was evidently due to the appearance of vapor bubbles at
the test section exit and could be attributed to the occurrence of ONB.
Kennedy et al. [132] compared the conditions leading to ONB predicted by
the aforementioned method, with the conditions where the whistlelike sound
was heard, and noted good agreement between the two methods.
Several other correlations for ONB have been proposed in the past. A
good review of these correlations can be found in Marsh and Mudawar
[128]. Most of the correlations, however, are based on data with water only.
Yin and Abdelmessih [133] and Hino and Ueda [134] have proposed
correlations that are based on data with Freon 11 and Freon 113, respec-
tively. With the exception of the correlation of Bergles and Rohsenow [129],
however, these correlations have not been systematically compared with
microchannel experimental data.
D. ONsr1 or SicNiric:N1 Voin :Nn ONsr1 or Fiov INs1:niii1x
The onset of signicant void (OSV) point is usually identied in experi-
ments with commonly applied large channels by measuring the void fraction
prole along the channel, and dening the OSV as the point downstream of
which the slope of the void fraction prole is signicantly high (see Fig. 19).
Since OSV occurs only slightly before the onset of ow instability (OFI),
however, the conditions leading to OFI can be used for estimating the OSV
conditions. The latter approach has been used by Inasaka et al. [130],
Kennedy et al. [132], and Roach et al. [135] in their experiments with
microchannels.
The OSV phenomenon has been studied extensively in the past, and
several empirical correlations and mechanistic models have been proposed
for its prediction [121125]. Saha and Zuber [121] have proposed the
following widely used correlation, which has been successful in predicting a
wide range of experimental data dealing with commonly applied channels:
St 455/Pe for Pe 70,000 (92)
St 0.0065 for Pe 70,000, (93)
Fic. 22. Comparison between the OFI data of Kennedy et al. [132] on the correlation of
Saha and Zuber [121] for OSV. (With permission from [132].)
where
St
q"
"
j
'
U
'
C
''
(T
`'
T
'
)
(94)
Pe
GDC
''
k
'
(95)
Equation (92) represents OSV in the thermally controlled regime, where,
based on the experimental observations [117], bubbles generated on the wall
roll next to the wall, and are ejected into the bulk ow at the point where Eq.
(92) is satised. Equation (93), on the other hand, represents OSV in the
hydrodynamically controlled regime, where bubbles attached to the wall act
as surface roughness, and when the roughness height reaches a characteristic
height the bubbles are detached because of the hydrodynamic effects.
Inasaka et al. [130] compared their OFI data (dened similar to Fig. 20)
with the foregoing correlation of Saha and Zuber. For their 3 mm diameter
test section the correlation well predicted the data. For their 1 mm diameter
test section the correlation of Saha and Zuber agreed with the data
reasonably well for G7000 kg/m`s (corresponding to the thermally con-
trolled Pe 4.810").
A similar comparison, between the OFI data and the correlation of Saha
and Zuber for OSV [121], was carried out by Kennedy et al. [132]. Figure
22 depicts the results of Kennedy et al. As noted, consistent with the results
of Inasaka et al. [130], within the experimental parameter range, the
correlation of Saha and Zuber agrees with the data reasonably well in the
20,000Pe range. The apparent large uncertainty bands in the experiments
represent only _2% uncertainty in heat ux, and are a result of the integral
nature of the experiments of Kennedy et al. [132].
Successful analytical models for OSV, based on the bubble detachment
mechanism, have been proposed by Levy [123], Staub [124], and Rogers et
al. [125, 136]. These models, which are very similar in their basic approach,
assume that OSV occurs when the largest bubbles that can be thermally
sustained in the steady-state thermal boundary layer that forms on the
heated surface are detached from the heated surface by the hydrodynamic
and buoyancy forces parallel to the heated surface. The models of Levy
[123] and Staub [124] assume a high liquid mass ux and neglect the effect
of buoyancy force on bubble departure, and are known to do well when
applied to high-pressure data for water. The model of Rogers et al. [125]
considers relatively low mass ux and pressure conditions, takes account of
the buoyancy effect, explicitly accounts for the effect of advancing and
receding contact angles (surface wettability), and is based on a model for
bubble detachment due to Al-Hayes and Winterton [137]. Rogers and Li
[136] recently modied the aforementioned model of Rogers et al. [125],
thereby extending its parameter range of applicability.
Recent experimental studies by Bibeau and Salcudean [118120] have
cast doubt on the bubble detachment phenomenon as the process respons-
ible for OSV. Careful visual observations in the latter studies have shown
that bubble ejection normal to the wall, and not bubble detachment and
motion parallel to the wall, is responsible for OSV. The aforementioned
detachment models [123125, 136] do not address bubble ejection at all.
Notwithstanding, these analytical models have been relatively successful in
predicting experimental data. The model of Levy, briey described next, has
been applied to microchannel data by Inasaka et al. [130].
According to Levys model [123], in a fully turbulent subcooled ow eld
in a heated channel bubble departure occurs when the drag force on the
bubble (itself obtained from wall frictional stress) becomes equal to the
resistive surface tension force, leading to
y
"
y
"
U*
v
'
C
oD
j
'
j
'
, (96)
where y
"
is the distance from the heated wall to the tip of the bubble, and
U* t
"
/j
'
. (97)
The coefcient C0.015 is an empirically adjusted parameter, and t
"
is
obtained from
t
"
f
''
4
G`
2j
'
. (98)
The friction factor, f
''
, is found from
f
''
0.022
1
_
2x10"
c
D
10'
Re
''
, (99)
where c/D
10" is assumed to represent the surface roughness caused by

the presence of bubbles at the departure.
Levy [123] assumed that bubbles are at saturation temperature with
respect to the local ambient pressure, and can be sustained if the local liquid
temperature, T
'
( y
"
), is at least at saturation. The liquid temperature
distribution furthermore, was assumed to follow the fully developed, steady-
state turbulent boundary layer temperature prole [138], according to
which
T
"
T
'
( y) Qf ( y, Pr
'
), (100)
where yyU*/v
'
is the dimensionless distance from the wall, and
Q
q"
"
j
'
C
''
U*
(101)
f ( y, Pr
'
)
Pr
'
y, Oy5
5
Pr
'
ln
_
1 Pr
'
y
5
1
, 5 y30.
5Pr
'
ln[1 5Pr
'
] 0.5 ln(y/30)} 30 y
(102)
In fully developed and steady-state,
T
"
T
'
q"
"
/h
''
. (103)
Utilizing Eqs. (100) and (102) and requiring that T
'
( y*
"
) T
`'
at OSV, one
gets [123]
(T
`'
T
'
)
'`'
q"
"'`'
h
''
Qf ( y
"
, Pr
'
). (104)
Accordingly, based on the model of Levy [123], Eqs. (97) and (104) provide
the relationship between q"
"'`'
and the bulk liquid subcooling at OSV.
Inasaka et al. [130] compared the predictions of the model of Levy [123]
for OSV with their OFI data. (See Table III for the characteristics of their
experiments.) The comparison results are depicted in Fig. 23. As noted, the
Fic. 23. Comparison between the OFI experimental data of Inasaka et al. [130] and the
predictions of the OSV model of Levy [123].
model agrees with the data associated with the larger test section of Inasaka
et al. (D 3 mm) reasonably well, and systematically underpredicts q"
"'`'
for their smaller (D 1 mm) test section.
The model of Levy [123], as well as the aforementioned models of Staub
[124] and Rogers et al. [125, 136], all assume that at OSV the bubble
temperature, and liquid temperature at the bubble tip, must be equal to T
`'
,
the saturation temperature corresponding to the local ambient pressure.
This assumption evidently neglects the bubble superheat resulting from
surface tension and is appropriate for commonly used large channels where
the predicted size of the bubbles is typically large enough to render the effect
of surface tension on bubble temperature negligibly small. In microchannels,
however, the bubbles are small (see the discussion in the forthcoming
Subsection E).
Fic. 24. Cross-sectional geometries of the test sections of Roach et al. [135]. (With
The model of Levy [123] can be corrected for the effect of surface tension
on bubble temperature by requiring that
T
"
T
'
( y
"
) T
`'
AT
(105)
where, assuming that the bubble diameter is approximately equal to Y
"
, and
using Clapeyrons relation,
AT
2oT
`'
Y
"
2
hfg

1
j
`
1
j
'
. (106)
Results of the modied Levy model are also depicted in Fig. 23 and are
noted to agree better with the entire data of Inasaka et al. [130].
An extensive experimental study of the OFI phenomenon in microchan-
nels cooled with water was recently carried out at the Georgia Institute of
Technology, for the purpose of generating the data bases needed for the
design of the proposed Accelerator Production of Tritium (APT) system
[132, 135, 139]. A summary of the parameter ranges of these experiments is
included in Table III. The OFI data of Kennedy et al. are compared with
the correlation of Saha and Zuber [121] for OSV in Fig. 22.
The experimental data of Roach et al. [135] deal with OFI at very low
ow rates, in channels with the cross-sectional geometries displayed in Fig.
24. Test sections (a) and (b) were uniformly heated circular channels, and
Fic. 25. OFI equilibrium qualities in the experiments of Roach et al. [135]. (With per-
mission from [135].)
test sections (d) and (e) were meant to represent the ow channels in a
micro-rod bundle with triangular array. Test section (d) was uniformly
heated over its entire surface, while the test section (e) was heated over the
surfaces of the surrounding rods. Roach et al. also examined the effect of
dissolved noncondensables on OFI by performing similar experiments with
fully degassed water and with water saturated with air with respect to the
test section inlet temperature and exit pressure. The bulk of the data
indicated that OFI occurred when the coolant at channel exit had a positive
equilibrium quality, indicating that, unlike in large channels and microchan-
nels subject to high heat uxes and high coolant ow rates, subcooled
voidage was insignicant in these experiments. Figure 25 displays typical
data. These results show that the commonly used models for OFI, which
emphasize subcooled voidage, or use the onset of signicant void (OSV) as
an indicator for the eminence of OFI, may be inapplicable for microchannels
under low ow conditions. In comparison with tests with degassed water,
the total channel pressure drops in tests with air-saturated water were
consistently and rather signicantly larger, indicating strong desorption of
the noncondensables, which contributed to channel voidage and therefore
increased the total channel pressure drop. The impact of the noncondens-
ables on the conditions leading to OFI was small, however. With all
parameters including heat ux unchanged, the mass uxes leading to OFI
in air-saturated water experiments were different than in degassed water
experiments, typically by a few percent.
Blasick et al. [139] investigated the OFI phenomenon in uniformly heated
horizontal annuli, using six different test sections, all with an inner radius of
6.4mm, and gap widths in the 0.721.00 mm range. Among the parametric
effects they examined was the impact of the inner-to-outer surface heat ux
ratio (varied in the 0 range), which was found to be negligible. Kennedy
et al. [132], Roach et al. [135], and Blasick et al. [139] developed simple
and purely empirical correlations for their OFI data by comparing the ow
and boundary conditions that lead to OFI with those leading to saturation
at the exit of their test sections.
E. Onsrv:1ioNs oN Binnir Nicir:1ioN :Nn BoiiiNc
Heterogeneous bubble nucleation and ebullition phenomena in common-
ly applied large channels, as the basis of nucleate boiling heat transfer
mechanism, have been qualitatively well understood for decades [106108].
The bubble formation and release period from wall crevices is generally
divided into waiting and growth periods. The departure of a bubble from a
wall crevice disrupts the local thermal boundary layer, and the waiting
period represents the time during which a fresh thermal boundary layer
capable of initiating bubble growth on the crevice forms. The bubble growth
during the growth period is primarily due to the evaporation of a liquid
microlayer that separates the bubble from the heated surface, and the bubble
is detached from the solid surface when the buoyancy and hydrodynamic
forces that attempt to displace the bubble overcome the resistive forces,
mainly the surface tension force. Models based on the aforementioned
phenomenology have been published, among others, by [140, 141]. The
bubble ebullition process in reality is highly stochastic, however, and
accordingly semiempirical correlations have been proposed for the nucle-
ation site size number and distribution [142, 143], and bubble maximum
size and frequency [144146]. The applicability of the aforementioned
models and correlations to microchannels is questionable, however.
In extremely small channels very high wall temperatures are required for
the generation of bubbles. Lin et al. [147], for example, could produce
bubbles in water, methanol, and FC 43 liquids in 75 jm deep microchannels
by raising the channel wall temperature to the proximity of the liquid critical
temperature. In microchannels of interest to this article, furthermore, the
velocity and temperature gradients near the wall can be extremely large,
leading to the formation and release of extremely small microbubbles.
Because of the occurrence of very large temperature and velocity gradients
and the small bubble size, furthermore, forces such as the thermocapillary
(Marangoni) force and the lift force arising from the velocity gradient
become important [131]. These forces are generally neglected in the
modeling of bubble ablution phenomena in large channels.
Vandervort et al. [131] investigated the heat transfer associated with the
ow of highly subcooled water at high velocity in microchannels with
0.32.5 mm diameter. At the relatively low mass ux of G5000 kg/m`s,
with a wall heat ux that was about 70% of the heat ux that would lead
to CHF at the test section exit, the ow eld at the test section exit was
foggy, indicating the presence of large numbers of micro bubbles too small
to be discernible individually. The occurrence of fogging required higher
heat uxes as the mass ux was increased, and no fogging was visible at
G25,000 kg/m`s. Vandervort et al. [131] analytically estimated the size of
the bubbles released from the wall crevices, and the magnitude of forces
acting on them, for the following typical test conditions: D1.07 mm,
L /D25, P1.2 MPa, G25,000 kg/m`s, and AT
`'
100C.
The diameter of the released bubble, as predicted by the model of Levy
[123] in the latter authors analysis of the onset of signicant void (OSV),
was only 2.7 jm. The estimated magnitudes of other forces acting on such a
bubble, while it is still attached to the heated surface, are depicted in Fig.
26, where F
`
, F
, and F
'
represent the forces due to surface tension, drag,
and buoyancy, respectively. The forces F
''
and F
'"
are due to the generated
vapor thrust and the inertia of the liquid set in motion by the growing
bubble, respectively. All the latter forces are generally accounted for (and
some are neglected because of their relatively small magnitudes) in bubble
ebullition analysis for common large channels. The Marangoini force F
"
,
which is small for large bubbles and is therefore usually not considered in
bubble ebullition models for large channels, can be estimated from [148]
F
"
D`
"
2
co
cT
cT
cy
. (107)
Vandervort et al. [131] also estimated the magnitudes of forces that act on
the aforementioned bubble, once it is detached from the solid surface, as
depicted in Fig. 27, where F
'
is the lift force that results from the local liquid
velocity gradient and can be obtained from [149, 150]
F
'
C
D`
"
6
j
'
(U
'
U
'
)
cU
'
cr
. (108)
Fic. 26. Magnitudes of various forces acting on microbubble formed at ONB conditions in
a microchannel with D1.07 mm [131]. (With permission from [131].)
Based on airwater experimental data in a 57.1 mm diameter test section,
Wang et al. [150] correlated the coefcient C in the preceding expression as
C0.01
0.490
cot
_
log 9.3168
0.1963
, (109)
where
e
D
"
U
'
U
'

dU
'
dr
D
"
D
1
Re
"
U
'
U
`
`
, (110)
where D
"
and D are the bubble and channel diameters, respectively, and
Re
"
D
"
U
'
U
'
/v
'
(111)
U
`
1.18(og/j
'
)"``. (112)
The force F
"
is a near-wall force that opposes the contact between the
bubble and the wall and arises because of the hydrodynamic resistance
associated with the drainage of the liquid lm between the bubble and the
surface when the bubble approaches the surface. A similar force opposes the
coalescence of bubbles, and bubbleparticle coalescence, in otation [151].
Fic. 27. Magnitudes of various forces acting on a microbubble detached from the wall in a
microchannel with D1.07 mm [131]. (With permission from [131].)
Based on a two-dimensional analysis, Antal et al. [152] derived
F
"
D`
"
6
2j
'
(U
'
U
`
)`
D
"

C
"
C
"`

D
"
2y
, (113)
where:
C
"
0.1040.06(U
'
U
'
) (114)
C
"`
0.147, (115)
where y is the distance from the wall.
Observations consistent with those reported by Vandervort et al., have
also been reported by Peng and Wang [110, 153]. The latter authors have
studied the forced convective boiling and bubble nucleation associated with
the ow of subcooled deionized water and methanol in microchannels with
rectangular cross sections, 0.20.8 mm wide, and 0.7 mm deep, with near-
atmospheric test section exit pressure. The microchannels were heated on
one side through a metallic cover and were covered by a transparent cover
on the other side. The experimental boiling curves of Peng and Wang [110]
indicated essentially no partial boiling in their microchannels, and in the
portions of the boiling curves that indicated fully developed boiling the
effects of liquid velocity and subcooling were small. No visible bubbles
occurred in the heated channel, however, even under conditions clearly
representing fully developed boiling, and instead a string of bubbles could
be seen immediately beyond the exit of each test section. Peng et al. [154]
have hypothesized that true boiling and bubble formation are possible if the
microchannel is large enough to provide an evaporating space; otherwise
a ctitious boiling heat transfer regime is encountered where the well-
known fully developed boiling heat transfer characteristics (e.g., lack of
sensitivity of the heat transfer coefcient to the bulk liquid velocity and
subcooling) occur without visible bubbles. Hosaka et al. [155] have argued
that careful experiments are needed to determine whether passage dimen-
sions and length sales peculiar for each uid affect the boiling phenomena
in microchannels.
Evidently, experiments aimed at careful elucidation of the bubble ebul-
lition and other phenomena associated with boiling in microchannels are
needed.
VI. Critical Heat Flux in Microchannels
A. IN1onic1ioN
Forced convection subcooled boiling in small channels is among the most
efcient known engineering methods for heat removal and is the cooling
mechanism of choice for ultrahigh heat ux (HHF) applications, such as the
cooling of fusion reactor rst walls and plasma limiters where heat uxes as
high as 60 MW/m` may need to be handled. Critical heat ux represents the
upper limit for the safe operation of cooling systems that depend on boiling
heat transfer, and adequate knowledge of its magnitude is thus indispensable
for the design and operation of such systems.
Critical heat ux has been investigated extensively for several decades.
The majority of the investigations in the past three decades have dealt with
the safety of the cooling systems of nuclear reactors, however. Some recent
reviews include [106, 156159.] The complexity of the CHF process, the
lack of adequate understanding of the phenomenology leading to CHF, and
the urgent need for predictive methods have led to more than 500 empirical
correlations in the past [157]. The available experimental data are extensive
and cover a wide parameter range. Most of the data, nevertheless, deal with
water only, and until recently data have been scarce for certain parameter
ranges, such as low-ow, low-pressure CHF in small channels.
Some CHF experimental investigations in the past have included channels
with D
"
*0 (1 mm) [95, 131, 155, 160175], and microchannel CHF data
have been included in the databases of some of the widely used CHF
correlations [176178], apparently without consideration of the important
differences between micro- and large channels. Systematic investigation of
CHF and other boiling/two-phase ow processes in microchannels, how-
ever, have been performed only recently. Most of the reent studies were
concerned with the aforementioned cooling systems of fusion reactors,
where channels with D13 mm and with large L /D ratios carry highly
subcooled water with high mass uxes and are subjected to large heat uxes
[155, 160173]. A few experimental investigations have also addressed CHF
in microchannels under low mass ux and low wall heat ux conditions [95,
174].
In the forthcoming sections, the recently published data, models, and
correlations relevant to CHF in microchannels are reviewed. In Subsections
B and C the existing CHF data obtained with channels with D
"
*0 (1 mm)
and their important trends are discussed. The empirical correlations that
have been recently applied to CHF in microchannels are discussed in
Section D. In Section E the relevant theoretical models are discussed.
B. ExirixrN1:i D:1: :Nn Tnri TrNns
Table IV provides a summary of recent experimental investigations and
includes some older microchannel data previously reviewed by Boyd [156]
and utilized for model validation by Celata et al. [175]. Among the
investigations listed, only the data of Bowers and Mudawar [95] and Roach
et al. [174] were obtained in horizontal heated channels, and all other
experiments dealt with ow in vertical channels The depicted list does not
include experiments where enhancement techniques such as internal ns and
swirl ows were utilized.
Boyd [157] carried out a detailed assessment of the important parametric
trends based on the data associated with subcooled ow CHF available in
1983, and identied parameter ranges in need of further experimental
research. With respect to the fusion reactor applications, Boyd [156, 157]
recommended experiments with large L /D. The scarcity of data at low
pressure is also evident in Table IV. Experiments with large L /D and at low
pressure were subsequently performed by Boyd in channels with D3 and
10.2mm diameters [167, 168]. The CHF experimental investigations in
TABLE IV
Sixx:x or ExirixrN1:i D:1: Dr:iiNc vi1n CHF iN Sx:ii Cn:NNris
Critical heat
Pressure Mass ux ux
Source Channel characteristics Fluid (MPa) (Mg/m`s) Inlet conditions (MW/m`)
Ornatskiy [160] D0.5 mm, L 14 cm, vertical Water 1.03.2 2090 T
'
1.5154C 41.9224.5
Ornatskiy and Kichigan D2 mm, L 56 mm, vertical Water 1.02.5 5.030.0 T
'
2.7 204.5C 6.464.6
[161]
Ornatsky and Vinyarskiy D0.42.0 mm, L 11.256 mm, vertical Water 1.13.2 10.090.0 T
'
6.7155.6C 27.9227.9
[162]
Loomsmore and Skinner D0.62.4 mm, L 6.3150 mm, vertical Water 0.10.7 3.025.0 T
'
3.2130.9C 6.744.8
[163]
Daleas and Bergles D1.22.4 mm, L /D14.926, vertical Water 0.2 1.523.0 0.313.1
[164]'
Subbotin et al. D1.63 mm, L 180 mm, vertical Helium 0.10.2 0.080.32 x
'
0.25
[165]
Katto and Yokoya D1 mm, L /D25200, vertical Liquid He 0.199 1110` h
h
'
3.5 to
[166] 7.0 kJ/kg
Boyd [167] D3 mm, L /D96.9, horizontal Water 0.77 at exit 4.640.6 T
'
20C 6.2541.58
Nariai et al. [169, 170] D1, 2, 3 mm; L 1.0100 mm, Water 0.1 6.720.9 T
'
15.464C 4.670
vertical
Inasaka and Nariai [171] D3 mm, L 100 mm, vertical Water 0.31.1 4.330 T
'
2578C 7.344.5
Hosaka et al. [155] D0.5, 1, 3 mm; L /D50, vertical R-113 1.12.4 9.332.0 AT
`''
5080C
Celata et al. [172] D2.5 mm, L 100 mm, vertical Water 0.62.6 10.140.0 T
'
29.870.5C 12.160.6
Vandervort et al. D0.32.6 mm, L 2.566 mm, vertical Water 0.12.3 8.442.7 T
'
6.484.9C 18.7123.8
[131, 173]
Bowers and Mudawar D0.51, 2.54 mm, L 10 mm, horizontal R-113 0.138 at inlet 0.0310.15 for AT
`''
1032C
[95] D2.54 mm;
0.120.48 for
D0.5 mm
Roach et al [174] D1.17, 1.45 mm, circular; D
"
1.13 mm, Water 0.3441.043 0.251.0 T4972.5C 0.863.7
semitriangular; L 160 mm; horizontal at exit
From Celata, Cumo, and Mariani [175].
'From Boyd [156].
2
1
1
[169, 170], as noted, are primarily focused on low pressure and high mass
ux. The experiments of Boyd [167] all represented CHF under high heat
ux and subcooled bulk liquid conditions; the local subcooling at their test
section exit varied in the 3074C range. The CHF varied linearly with G
in Boyds experiments and was correlated accordingly [167]. Boyd [168]
examined the effect of L /D on CHF in a 1.0 cm-diameter channel.
The experimental data of Nariai et al. [169, 170] include subcooled as
well as saturated (two-phase) CHF data. The dependence of CHF on
channel diameter was found to vary with the local quality. When CHF
occurred in subcooled bulk liquid, CHF monotonically increased as D
decreased. With CHF occurring under x 0 conditions, however, the trend
was reversed and CHF decreased with decreasing D. The aforementioned
trend, i.e., increasing CHF in subcooled forced ow as D is decreased, had
been noted earlier by Bergles [179], who suggested that three mechanisms,
all of which deal with the vapor bubbles as they grow and are released from
wall crevices, lead to increasing CHF as D becomes smaller. As D is
decreased, (a) the vapor bubble terminal diameter (the diameter of bubbles
detaching from the wall) decreases as a result of larger liquid velocity
gradient; (b) the bubble velocity relative to the liquid is increased; and (c)
condensation at the tip of bubbles is stronger due to the large temperature
gradient in the liquid. Nariai et al. [169, 170] thus explained the aforemen-
tioned trend of increasing CHF with decreasing D in subcooled liquids by
arguing that smaller bubbles imply a thinner bubble layer and a smaller
void fraction, and lead to a higher CHF. A recent systematic assessment of
the effect of channel diameter on CHF in subcooled ow by Celata et al.
[180], based on experimental data from several sources, has conrmed the
aforementioned mechanism. Hosaka et al. [155], in their experiments with
R-113, observed a similar trend and attributed the increase in CHF
associated with decreasing D to the decreasing bubble terminal size.
The trends of the available data, however, indicate that a threshold
diameter exists beyond which the effect of channel diameter on subcooled
ow CHF is negligible. Figure 28 depicts the results of Vandervort et al.
[173]. Below a threshold diameter (about 2 mm for the depicted data), CHF
in subcooled ow increases with decreasing D, whereas for larger diameters
the inuence of variations in D on CHF is small. CHF is more sensitive to
D at higher values of G. Similar trends have been noted by some other
investigators [180]. The magnitude of the aforementioned threshold diam-
eter, which is likely to depend on geometric as well as thermalhydraulic
parameters, may not be specied with precision at the present time because
of the limited available data.
It should be mentioned that the foregoing trend (i.e., increasing CHF with
decreasing D) applies when CHF occurs in subcooled bulk ow. An
Fic. 28. Parametric dependencies in the CHF data of Vandervort et al. [131]. (With
opposite trend was reported by Nariai et al. [169, 170] for CHF occurring
when x
0.
The experiments of Celata et al. [172] covered the intermediate and low
pressure range of 0.62.6 MPa, and were all carried out in the relatively
large 2.5 mm diameter test section. They are, however, part of a database
utilized by Celata et al. [175, 181] for the validation of various models and
correlations, as well as the identication of some important trends in the
CHF data.
Vandevort et al. [173] systematically examined the effects of inlet subcool-
ing, channel diameter, pressure, and length-to-diameter ratio, dissolved
noncondensable gas, and heated wall material on CHF of subcooled water
ow. Their data, along with data from several other sources, were used for
parametric trend identication by Celata [181]. Vandervort et al. [173]
noted the frequent occurrence of premature burnout in their tests, which
they dened as any thermal failure not directly attributable to CHF or other
obvious failure mechanisms. Premature failure occurred following boiling
Fic. 29. Equilibrium quality at CHF in the experiments of Bowers and Mudawar [95].
incipience, which was typically accompanied by a boiling song in the test
section. Each test section that failed had some region experiencing incipient
boiling. The majority of the premature failures occurred in two types of
tubing (1.9 and 2.4 mm diameter stainless steel capillary tubing) and
otherwise showed no discernible dependence on the primary variables, P, G,
and subcooling. Although the cause of, and the conditions leading to, these
premature failures could not be identied with certainty, the evidence
indicated that they resulted from some thermalhydraulic phenomenon
subsequent to incipient boiling. The development of a metastable super-
heated liquid because of the scarcity of wall crevices, which can lead to
sudden and explosive boiling, was mentioned as a possible cause, and it was
argued that channel wall roughness may thus be a stabilizing factor that
reduces the possibility of premature burnup.
The experiments of Bowers and Mudawar [95] and Roach et al. [174]
addressed CHF under very low mass ux conditions. Bowers and Mudawar
[95] compared the characteristics of a mini (D2.54 mm) and a micro
(D0.51 mm) channel. CHF occurred when the channel exit equilibrum
quality was quite high, typically at x
)0.5 for the larger channel and

x
)1 for the smaller channel. At very low mass uxes, furthermore,

superheated vapor exited from the test section and the CHF results were
insensitive to the inlet subcooling. Figure 29 displays the test section exit
qualities measured by Bowers and Mudawar, where the Weber number is
dened as We G`L
"
/(oj
), and L
"
is the heated length. For the smaller
(micro) test section (D 0.51 mm) the equilibrium quality at CHF was
evidently quite high and could approach 1.5. These high equilibrium
qualities imply that a metastable superheated liquid ow occurred upstream
of the CHF point in the channel. The potential occurrence of metastable
superheated liquid also seems to be consistent with the observations of Peng
and Wang [110, 153] in their experiments dealing with boiling in micro-
channels. The latter authors studied boiling heat transfer in a channel with a
0.6mm0.7 mm rectangular channel, and did not observe visible bubbles in
their test section even under conditions that implied fully developed boiling.
Instead, strings of bubbles could be seen at the exit of their test section.
The experiments of Roach et al. [174] dealt with CHF in subcooled water
at low mass uxes in heated microchannels. The results were consistent with
the aforementioned observations of Bowers and Mudawar [95]. CHF
occurred when x
)0.36 at the exit of their test sections, suggesting the

occurrence of dryout; and x
)1 was noted in many of their tests,

suggesting the potential occurrence of metastable superheated liquid ow.
C. Errrc1s or Prssir, M:ss Fiix, :Nn NoNcoNnrNs:nirs
CHF is affected by more than 20 parameters, which include subcooling,
pressure, channel diameter, length, surface conditions and orientation, heat
ux distribution, dissolved noncondensables, and various thermophysical
properties [157]. Although the available microchannel data are limited and
do not allow for a systematic assessment of all dependencies, the existing
database associated with CHF in subcooled water ow is sufcient for the
identication of some important trends applicable to high mass ux CHF,
where CHF occurs under high local subcooling conditions. A useful system-
atic study of various parametric effects associated with CHF in microchan-
nels was performed by Vandervort et al. [173]. Utilitizing the experimental
data of several authors, Celata [181] assessed several important parametric
dependencies.
The dependence of CHF on pressure is in general monotonic. At pressures
well below the critical pressure, P*
''
, CHF is expected to increase with
increasing pressure [156]. The available data relevant to subcooled CHF in
microchannels (which virtually all represent PP*
''
) indicate that CHF is
insensitive to pressure [173, 181]. A slight decreasing trend in CHF with
respect to increasing pressure has been noted by Vandervort et al. [173] and
Hosaka et al. [155], however.
CHF monotonically inceases with increasing mass ux; it increases
monotonically, and approximately linearly, with increasing local subcooling.
Figure 30 depicts the effect of subcooling on CHF in the experiments of
Vandervort et al. [131, 173].
Fic. 30. The effect of exit subcooling on CHF in the experiments of Vandervort et al. [173].
(With permission from [173].)
Vandervort et al. [173] and Roach et al. [174] attempted to measure the
effect of dissolved air in subcooled water on CHF. The impact of the release
of the dissolved noncondensables on liquid forced-convection in microchan-
nels were discussed in Section III, H, and the impact of dissolved air on
critical (choked) ow in cracks and slits is discussed in Section VII, D. The
experimental results of both Vandervort et al. [173] and Roach et al. [174]
indicated a negligibly small effect of dissolved air on CHF. Since the
solubility of air in water is very low, and in view of the fact that considerable
evaporation due to boiling occurs in CHF, the insignicant contribution of
dissolved air to CHF is expected. It should be noted, however, that for other
uidnoncondensable pairs for which the solubility of the noncondensable
in the liquid is high, the impact of the dissolved noncondensable may not
be negligible.
D. Exiiic:i Cori:1ioNs
Most of the more than 500 models and correlations proposed in the past
for CHF are applicable over limited parameter ranges and are often in
disagreement with one another. Good reviews can be found in [157159].
In this section, only empirical correlations that have recently been applied
to, and have been successful in predicting, some microchannel CHF data are
discussed.
The following simple empirical correlation for subcooled ow CHF was
proposed by Tong [182]:
q"
'"'
h
C
G""j"'
D"'
. (116)
Tong [182] correlated the coefcient C in terms of the local equilibrium
quality, x
, according to
C1.76 7.433x
12.222x`
. (117)
Nairai et al. [169, 170] applied the correlation of Tong [182] to their
experimental data and noted that in order to achieve agreement they needed
to correlate the parameter C separately for low and high heat ux condi-
tions. More recently, Celata et al. [175] noted that the data used by Nariai
et al. [169, 170] were limited to relatively low heat ux and low pressure,
and their modication of the correlation of Tong was inadequate. Celata et
al. [175] correlated the parameter C according to
C(0.216 4.7410`P)+ (118)
+0.825 0.987x
, for 0.1 x
0 (119)
+1 for x
0.1 (120)
+1/(2 30x
) for x
0, (121)
where P must be in MPa in Eq. (118). This modied Tong correlation
evidently should apply to saturated exit conditions as well. Celata et al.
[175] indicated that the preceding correlation could predict 98.1% of their
compiled data points (which covered 0.1 P9.4 MPa, 0.3 D
25.4 mm, 0.1 L 0.61 m, 2 G90 Mg/m`s, and 90 AT
`'
230 K)
within _50%. The agreement of the correlation with the microchannel data
included in the database of Celata et al., furthermore, appeared to be
satisfactory. Celata et al. [172, 175] also compared their compiled database
with the predictions of several other empirical correlations, generally with
poor agreement in comparison with the aforementioned modied-Tong
correlation.
Hall and Mudawar [183] have recently assessed the validity of previously
published CHF experimental data and have compiled a qualied CHF
database (referred to as the PU-BTPFL CHF Database). This database
includes experiments representing 0.3 D45 mm, 10 G2484 kg/m`s,
and 2.25 x
1.0, in vertical, upow tubes, with x
representing the
local equilibrium quality at the CHF (i.e., the end of the heated segment of
test sections) point. They compared 25 widely referenced correlations
dealing with CHF in vertical, upow channels with their database and
showed that the empirical correlations of Caira et al. [184] provided the
most accurate predictions. The correlation of Bowring [178] also showed
relatively good agreement with data. Although the PU-BTPFL Database
and the aforementioned correlations generally address vertical channels with
upow, the data included in the database that represent small channels
(D*1 mm) may represent horizontal microchannels as well because of the
small inuence of channel orientation with respect to gravity on two-phase
ow in such microchannels.
The correlation of Bowring [178] can be expressed as
q"
'"'
ADGh
/4
C
(122)
where q"
'"'
is in W/m`, and
A
2.317(h
DG/4)F
1 0.0143F
`
D`G
(123)
C
0.077F
`
DG
1 0.347F
"

G
1356
(124)
n 2.0 0.5P
^
(125)
P
"
0.145P. (126)
P in Eq. (126) is in MPa, and
F
P`""`
^
exp[20.891(1P
^
)] 0.917}/1.917 (127)
F
`
1.309F
/P`'
^
exp[2.444(1P
^
)] 0.309} (128)
F
`
P`"``
^
exp[16.658(1P
^
)] 0.667}/1.667 (129)
F
"
F
`
P'""
^
. (130)
The preceding correlation could predict the low and high mass velocity data
of the PU-BTPFL database with mean absolute errors of 21.9 and 53.5%,
respectively. The correlation of Bowring systematically underpredicted the
data of Roach et al. [174], on the average by 36%.
The correlation of Caira et al. [184] can be represented as
q"
'"'
[0.25(h
h)
'''
]'
1 DL'
`
(131)
where
y
D'
`
G'
(132)
0
y
"
D'
G'
(133)
Dy
`
D'
`
G'
. (134)
All parameters are in SI units, and
y
"
10,829.55
y
0.0547
y
`
0.713
y` 0.978
y
"
0.188
y
`
0.486
y
'
0.462
y
`
0.188
y
`
1.2
y
"
0.36
y
"
0.911.
The preceding correlation agreed with the low and high mass velocity data
in the PU-BTPFL Database with mean absolute errors of 16.5 and 22.6%,
respectively. Cairas correlation agreed with the data of Roach et al. [174],
with an average overprediction of the data by only 18%.
Vandervort et al. [173] developed a statistical correlation based on their
subcooled ow water CHF data. Their correlation, however, includes more
than 20 constants.
The experimental data of Bowers and Mudawar [95], as noted in the
previous section, represented low-ow CHF, where CHF at high equilib-
rium qualities occurred. Noting the insensitivity of their data to inlet
subcooling, Bowers and Mudawar developed the following empirical correl-
ation:
q
'"'
G
"
0.16
G`L
oj
""
(L /D)"`". (135)
This correlation is remarkable for the implied recognition of the importance
of surface tension. The correlation, however, has not been validated
against data from other sources and is unphysical in its monotonic depend-
ence on D.
Shah [185] has developed an empirical CHF correlation based on a vast
pool of data representing upow in vertical channels with parameter ranges
0.315D37.5 mm, 1.3 L /D940, 4 G29041 kg/m`s, 0.0014
Pr 0.96, and inlet qualities covering 2.6 to 1.0. Shahs database includes
a wide variety of uids, and his empirical correlation is dimensionless and
utilizes all important thermophysical properties. Shahs correlation has two
versions: the upstream condition correlation (UCC), and the local condition
correlation (LCC). The UCC vesion can be expressed as
q"
'"'
/G
"
0.124(D/L
'
)"`"(10"/Y )(1 x
''
) (136)
where x
''
and L
'
are the effective inlet equilibrium quality and effective tube
length, respectively, and n is an empirical exponent. When x
'
0, L
'
is the
axial distance from the channel inlet and x
''
x
'
; when x
'
0 L
'
is equal
to the boiling length (i.e., the axial distance from the point where equilib-
rium quality is equal to zero) and x
''
0. Distinction is made between
helium and other uids. For helium n (D/L
'
)"``, and for other uids
n (D/L
'
)"`", for Y 10' (137)
n
0.12
(1 x
''
)"`
, for Y 10'. (138)
The parameter Y (Shahs correlating parameter) is dened as
Y
GDC
'
k
'
(j`
'
gD/G`)""(j
'
/j
'
)"'. (139)
The LCC correlation of Shah can be expressed as
q"
'"'
/Gh
F
'
F
`
Bo
, (140)
where F
'
, the entrance effect factor, is the smaller of 1 and [1.54 0.032(L
'
/
D)], with L
'
representing the axial distance from entrance. Parameters Bo
and F
`
are functions of the local quality, reduced pressure, P*
'
, and the
parameter Y. Shah recommends that the UCC correlation be used when
Y10' or L
'
160/P"
'
; otherwise, the correlation version predicting a
lower q"
'"'
should be chosen. Hosaka et al. [155] compared the predictions
of Shahs correlation with their data. On the average, the correlation
overpredicted the data only slightly. The correlation, however, has not been
adequately compared with other recent microchannel CHF data. Katto
[159] has indicated that the strong dependence of the parameter Y in Shahs
CHF correlation on g for high mass ux forced ow may be physically
questionable.
E. Tnror1ic:i Monris
Theoretical modeling of CHF has undergone great advances in the recent
past. Most of the developed models, however, may be inapplicable to
microchannels.
Weisman [186] has summarized the status of theoretical models for CHF
and indicated that the phenomenology of CHF depends on the two-phase
ow regime. In annular (high quality) ow, lm dryout leads to CHF. In
the slug/plug ow regime, CHF appears to occur if the time period
associated with the passage of a gas plug is long enough to allow for the
complete evaporation of the liquid lm that is left behind on the wall
following passage of a liquid slug. In highly subcooled ow, CHF is
triggered by the thermal and hydrodynamic processes adjacent to the heated
wall. Two modeling approaches have been pursued for CHF in highly
subcooled ow. Weisman and co-workers [187, 188] suggested that the
coalescence of microbubbles that form on the wall and the occurrence of a
critical void fraction in the bubble layer lead to CHF.
For subcooled or low-quality CHF, based on careful ow visualization,
Lee and Mudawar [189] proposed that CHF in the aforementioned regime
occurs when the liquid sublayer that separated vapor blankets or slugs from
the wall is disrupted. They developed a mechanistic model accordingly.
More recent ow visualization studies further support the latter liquid
sublayer dryout model [190], and models based on this mechanism, and
following the essential elements of the model by Lee and Mudawar [189],
have recently been compared with data including small channel data by
Katto [191, 192] and Celata et al. [193]. The outline of the model, as
elaborated by Katto [191], is now presented.
Figure 31 is a schematic of the ow eld [191], where vapor slugs, formed
as a result of the coalescence of smaller bubbles, are separated from the
heated surface by a liquid sublayer. The vapor blankets are assumed to
remain thin because of condensation, and their velocity is assumed to be
closely related to the local ambient uid velocity. CHF is assumed to occur
when the residence time of a vapor slug over the liquid sublayer is
sufciently long to allow for its complete evaporation and breakdown.
Katto assumed that the local ow quality can be obtained from the quality
prole t of Ahmad [194], according to which
x
x
x
`'
exp
x
`'
1
1 x
`'
exp
x
`'
1
for x
`
x
, (141)
where the equilibrium quality at the point of onset of signicant void, x
`'
,
Fic. 31. Schematic of ow eld near the CHF conditions with high subcooling and high
mass ux. (With permission from [191].)
is obtained from the empirical correlation of Saha and Zuber [121]. The
sublayer initial lm thickness at the front end of the vapor slug is found
based on an empirical correlation due to Haramura and Katto [195];
o
''"
1.70510`
j
`
j
'
""
1
j
`
j
'
o
j
`

j
`
h
q"
'

`
, (142)
where the boiling heat ux, q"
'
, is obtained by assuming that the wall heat
ux is the summation of convective and boiling terms, thereby
q"
'
q"
"
h
''
(T
"
T
'
), (143)
where h
''
is obtained from the well-known correlation of Dittus and
Boelter, assuming purely liquid ow, and T
"
T
'
is obtained from
T
"
T
'
1)(T
`'
T
'
) q"
"
/h
''
(144)
230(q"
"
/Gh
)"`. (145)
CHF is assumed to occur when
q"
"
j
'
o
''"
h
/t
'`
, (146)
where t
'`
, the residence time of the vapor slug over the lm underneath it,
can be found from
t
'`
L
"
U
"
2o(j
'
j
`
)
j
'
j
`
U`
"
. (147)
Here, the length of the vapor slug has been assumed to be equal to the
neutral wave length according to the Helmholtz stability theory. The
velocity of the vapor slug relative to the lm is assumed to follow
U
"
U
'
, (148)
where U
'
is the velocity at the distance o
''"
from the wall, as predicted by
the Karmans universal turbulent boundary layer velocity prole, based on
a wall frictional shear stress found from
t
"
( f /4)G`/2j
"
. (149)
The homogeneous density, j
"
, is found using the ow quality obtained from
Eq. (45), and the DArcy friction factor f is found from the Prandtl
Karman correlation:
1/f 2.0 log
"
(Re f ) 0.8 (150)
Re GD/[j
`
: j
'
(1 :)(1 2.5:)]. (151)
Katto empirically correlated the parameter in Eq. (148) in terms of
Re(j
`
/j
'
) and : [191, 192].
The model of Katto [191, 192] just described has been shown to predict
experimental data well for channels with D1 mm, for a variety of uids
and a pressure range of 0.1 to 2.0 MPa. More recently, Celata et al. [193]
pointed out that Kattos model is unable to calculate the CHF when the
void fraction is larger than 70%. Celata et al. further modied the aforemen-
tioned phenomenological model of Lee and Mudawar [189] and Katto
[191, 192]. They assumed that the thickness of the vapor blanket (slug) is
equal to the bubble departure diameter, and the vapor blanket is always
surrounded with saturated liquid. Celata et al. [193] obtained the vapor
blanket velocity, U
"
, from the balance between drag and buoyancy forces,
assuming a vertical, upow conguration. Other essential model elements
were similar to [189, 191]. Celata et al. [193] compared their model with
experimental data covering the following range of parameters, with good
agreement between model and data: 0.2 D25.4 mm, 25 AT
'`'
255 K, 10` G910" kg/m`s, and 0.1 P8.4 MPa. The applica-
tion of the aforementioned force balance on vapor blankets in the model of
Celata, however, implies dependence on channel orientation, which may not
apply to microchannels.
CHF under high-quality conditions is primarily due to the lm dryout
phenomenon in the annular ow regime. The lm dryout models are based
on the solution of two-phase mass, momentum, and energy conservation
equations, whereby the ow rate of the liquid lm on the heated wall and
the lm thickness are calculated. CHF is assumed to occur when the liquid
lm is depleted [159]. Annular-dispersed ow is the ow regime that occurs
during lm dryout CHF in commonly used large channels, where entrained
liquid droplets are mixed with the vapor phase and the two-phase ow is
accompanied by continuous entrainment of new droplets from the liquid
lm and deposition of droplets on the lm. The variation of the lm
thickness is evidently affected by evaporation as well as droplet entrainment
and deposition. Dryout CHF models utilize empirical correlations for
droplet entrainment and deposition [196, 197]. A model by Sugawara et al.
[198] also accounts for the inhibition of droplet deposition due to the
counter ow of vapor resulting from lm evaporation.
Film dryout models have not been systematically applied to microchan-
nels, and the current models [196, 198] employ constitutive relations
associated with interfacial transfer processes and droplet entrainment and
deposition that may not be applicable to microchannels. The recent low-
ow and high-quality CHF data [95, 174] indeed suggest that the dryout
phenomenology in microchannels may be signicantly different than in
larger channels.
VII. Critical Flow in Cracks and Slits
A. IN1onic1ioN
Critical or choked ow represents the maximum discharge rate of a uid
through an opening connecting a pressurized vessel to a low-pressure
environment. When choking happens, the ow conditions downstream of a
location where critical conditions occur do not affect the ow rate, implying
that the hydrodynamic signals originating downstream are unable to pass
through the critical location.
Critical ow of a compressible uid can be well predicted by assuming
that the one-dimensional uid velocity at the critical location is equal to the
local isentropic speed of sound, and knowledge of uid stagnation proper-
ties is sufcient for calculating the conditions at the critical cross-section.
Critical two-phase ow is considerably more complicated, however, because
of the development of thermal and mechanical nonequilibria between the
phases. The uid stagnation properties are thus insufcient for uniquely
determining the conditions at the critical cross-section in two-phase ow.
Critical ow can be used for ow control and plays an important role in a
number of hypothetical nuclear reactor accidents. Critical two-phase ow
has been extensively studied in the past. Good reviews can be found in [106,
199, 200].
The critical ow of initially highly subcooled liquids through narrow
cracks and slits is of great interest in relation to the safety of nuclear and
chemical reactors. When cracks occur in high-pressure piping systems, they
often support critical ow and, in accordance with the leak-before-break
concept, their detection and correct characterization are necessary for
prediction and prevention of major leaks. Extensive research effort has been
devoted to the critical ow in cracks and slits in the past 15 years. Most of
the studies have focused on critical ow in slits or simulated cracks with
simple and regular cross-sections. Photomicrographs of typical intergranu-
lar stress corrosion-induced cracks in type 304 stainless steel [201], however,
show that such cracks often have highly tortuous and irregular ow
passages.
Cracks often have hydraulic diameters in the D
1 mm range and have

large length-to-diameter ratios; their nucleation, boiling heat transfer, and
two-phase ow characteristics are different from those of large channels.
Consequently, the correlations and models that have been developed in the
past for critical ow in commonly applied large channels do not apply to
cracks without modication.
In the forthcoming discussion, previous experimental studies dealing with
critical ow of initially subcooled liquids in cracks are reviewed in Section
B. The theoretical models are discussed in Sections CE.
B. ExirixrN1:i Ci1ic:i Fiov D:1:
Table V is a summary of the recent experimental data. Collier et al. [202,
203] performed two sets of experiments. In Phase I of their experiments,
they measured critical ow in articial slits. The geometric and test
parameters associated with their Phase I experiments are summarized in
Table V.
The test sections were produced by splitting anges and machining crack
faces in the center of each half, then reassembling the two ange halves. The
simulated crack surfaces were roughened by shot blasting them, and the
crack width, o, was adjusted with spacer blocks placed between the two
ange halves. They measured the temperatures and pressures at three
locations along their simulated cracks. In phase II of their experiments,
Collier et al. [202, 203] used a stainless steel pipe with 32.4 cm outside
TABLE V
Sixx:x or ExirixrN1:i D:1: ro Ci1ic:i Fiov iN C:cxs :Nn Micocn:NNris
Maximum
inlet Inlet
pressure subcooling Length Roughness
Author Fluid (MPa) (K) Cross-section (mm) (jm)
Collier et al. [202, 203] Water 11.5 33118 Rectangular, W57.2 mm; S 0.21.12 mm 63.5 0.310.2;
simulated crack
Collier et al. [202, 203] Water 11.5 072 Rectangular, W 0.7427.9 mm; 20 Real crack
S0.01830.247 mm
Amos and Schrock [204] Water 16.2 065 Rectangular, W20.4 mm; S 0.1270.381 mm 6075 Smooth
Kefer et al. [205] Water 16.0 060 Rectangular, W19108 mm; 1033 2040;
S0.0970.325 mm simulated and real
cracks
John et al. [105] Water 14 260 Rectangular, W 80 mm; S 0.20.6 mm 46 2150; real cracks
with 240
Nabarayashi et al. Water 7 30 Real cracks in pipe wall, Dp 114.3, 216.3 mm, 8.6, 12.7 mm 5.412.1
[206, 207] W60160 mm, S0.5 mm;
rectangular cracks, W 30, 60 mm, 10, 20, 36 mm 3.612
S 0.071 mm
Ghiaasiaan et al. [208] Water 7.24 34258 Circular, D0.78 mm 0.78 mm Not measured
Cited in John et al. [105].
2
2
6
diameter that contained a girth butt weld at midlength with a full circum-
ferential stress-corrosion crack near the welds. The initial circumferential
cracks were deep about 90% of the wall thickness, and through cracks of
varying lengths were obtained by machining away the pipe surface in the
vicinity of the crack.
Amos and Schrock [204] used two stainless steel blocks, with their
surfaces ground at, to provide ow channels with smooth walls. Various
slit widths were created by adjusting the distance between the stainless steel
blocks. They measured axial pressure variations using several pressure taps.
The experimental data of Kefer et al. [205] are referred to by John et al.
[105]. John et al. [105] used two blocks of stainless steel to produce slits
with adjustable distances between the steel blocks in steps of 0.1 mm. Using
several pressure taps they measured pressure distribution along their
simulated cracks. To examine the effect of slit surface roughness, they varied
the surface roughness of the steel blocks by shot blasting them with sand
and steel grit. For tests with a real crack, they used blocks of real reactor
pipe steel (20 Mn Mo In 55), with a crack that was produced by cyclic
bending. They measured the width of the cracks at inlet and exit of their test
section after mounting them. In some cases the slit width at exit was slightly
larger than at its inlet.
Nabarayashi et al. [206] and Matsumoto et al. [207] were interested in
fatigue cracks in carbon steel and stainles steel pipes and performed two sets
of experiments. In the fundamental tests [206], critical ow of saturated
water and steam in articial slits with W 30 and 60 mm, L 10, 20, and
36 mm, S0.071 mm was studied. They also conducted experiments in
through-wall cracks [207], initiated by electric discharge machining and
popagated by bending. They performed a careful measurement of surface
roughnesses in the articial and fatigue-induced cracks. The ranges of
average roughnesses are given in Table V. The maximum surface rough-
nesses varied in the 2055 jm range for their articial cracks and in the
1660 jm range in the fatigue-induced cracks. The surface roughnesses,
furthermore, did not vary noticeably after tests. They empirically correlated
their data for the effect of bending undulation on the crack pressure loss.
The experimental investigation of Ghiaasiaan et al [208] was concerned
with critical ow of highly subcooled water through a very short capillary.
Some important trends in critical ow data are now described. The
experimental data indicate the occurrence of metastable liquid ow near the
entrance of cracks, due to delayed nucleation (pressure undershoot). Figure
32 is an example pressure prole [204] that shows the effect of delayed
ashing. The solid line represents the liquid single-phase pressure drop, the
dashed line represents the measured pressure prole, and the horizontal line
is the saturation pressure corresponding to the inlet temperature. Flashing
Fic. 32. Pressure prole in a critical ow experiment [204]. (With permission from [204].)
occurs when the slope of the dashed line abruptly deviates from the slope
of the solid line. This type of delay in ashing is also common in large
channels and is typically 12 K for water [209]. Amos and Schrock [204]
noted that the degree of subcooling before the inception of ashing in-
creased with increasing the stagnation subcooling. An empirical correlation
Fic. 33. Some parametric trends in the experiments of Amos and Schrock [204]. (With
for pressure undershoot has been developed by Alamgir and Lienhard [210]
and has been applied by some investigators in mechanistic modeling of
critical ow in large channels. Amos and Schrock [204] noted that the
majority of depressurization rates in their experiments were outside the
range of the latter correlation.
With respect to the important trends, the available critical ow experi-
mental data indicate that G
''
(a) increases with increasing inlet subcooling;
(b) increases with increasing stagnation pressure; (c) decreases with increas-
ing L /D; and (d) is very sensitive to frictional pressure drop and consequent-
ly decreases with increasing the channel surface roughness. Figures 33 and
34 depict typical parametric trends in the experiments of Amos and Schrock
[204] and John et al. [105], respectively. The depicted model predictions are
discussed later.
In Fig. 35 typical data of Ghiaasiaan et al. [208] are depicted and are
compared with some relevant data points from Collier and Norris [202].
The gure demonstrates that critical ow in short capillaries may be
Fic. 34. Some parametric trends in the experiments of John et al. [105]. (With permission
from [105].)
signicantly different from that in cracks and slits with large L /D. The
measured G
''
in the experiments of [208] are evidently much larger than
those measured by Collier and Norris [202] under similar conditions and
demonstrate the signicance of channel length. The data of [208], further-
more, are relatively insensitive to the inlet subcooling. The latter data,
however, represent extremely high inlet subcooling, and their apparent
insensitivity to small variations in inlet temperature may not be representa-
tive of crack critical ow in their typical applications.
C. GrNr:i Rrx:xs oN Monris ro Tvo-Pn:sr Critical Fiov
iN Micocn:NNris
Many semianalytical and mechanistic models have been developed for
two-phase critical ow in common large passages. These models can be
Fic. 35. Comparison of the data of Ghiaasiaan et al. [208] with the data of Collier and
Norris [202], the predictions of the correlations of the RETRAN-03 code [219], and the
correlation of Leung and Grolmes [220]. (With permission from [208].)
divided into two broad groups: integral models and models based on
numerical solution of differential conservation equations. The two groups of
models follow similar principles: They all search for conditions that lead to
the maximum possible mass ux without violating the rst and second laws
of thermodynamics. In the integral models, the conservation equations are
analytically integrated along the ow passage in order to derive a closed-
form solution for the mass ux. The resulting closed-form solutions, of
course, may need iterative numerical solution. Major simplifying assump-
tions are evidently needed to make closed-form solutions possible. Reviews
of the most widely used integral models can be found in [106, 200].
Mechanistic critical ow models based on the numerical solution of the
one-dimensional differential conservation equation can rigorously account
for the thermal and mechanical nonequilibria and the system integral effects
on local ow properties. In these models, the 1D conservation equations are
numerically solved and, by iteratively varying the passage discharge rate, the
ow rate leading to critical conditions (an innitely large pressure gradient,
a vanishing determinant of the coefcient matrix for the system of one-
dimensional quasi-linear conservation equations, etc.) is specied. Early
contributions to this method include [211, 212], and a discussion of
mathematical bases can be found in [213]. This technique is particularly
appropriate for cracks and slits with large L /D where the 1D ow assump-
tion is adequate. Models based on direct numerical solution of differential
conservation equations for cracks and slits have been developed by Amos
and Schrock [204], Schrock and co-workers [201, 214, 215], and Feburie et
al. [216]. More recent contributions include [217, 218], where the effect of
noncondensables was modeled.
D. IN1rc:i Monris
Integral models dealing with two-phase critical ow in large passages are
many and are reviewed among others in [106, 200]. These models are
mostly based on assumptions and/or simple models that allow for the
calculation of the velocity slip and the magnitude of thermodynamic
nonequilibrium at the critical cross-section. Most of the widely used
methods for the prediction of two-phase critical ow neglect the frictional
pressure losses, which are often small in commonly applied large passages.
These models overpredict the critical ow rates in small passages and cracks
where such pressure losses are signicant.
The isentropic homogeneous-equilibrium model is the simplest among the
models that neglect irreversible losses and leads to
G
''
[2j`
''
(h
h
''
)]` (152)
where
x
''

''
(153)
h
''
(h
xh
)
''
(154)
v
''
j
''
(v
xv
)
''
. (155)
The critical mass ux is obtained by applying (dG/dP)
''
0, which leads to
2[h
h(s
, P
''
)]
cv
cP
''
v`(s
, P
''
) 0. (156)
Note that (cv/cP) also corresponds to the thermodynamic state specied
with (s
, P
''
). Equation (156) along with Eqs. (153)(155) are thus iterative-
ly solved for P
''
, as well as other thermodynamic properties at the critical
cross-section, whereupon G
''
is obtained from Eq. (152).
Since the aforementioned iterative solution is cumbersome, simpler corre-
lations for water have been developed [219, 220]. The correlation in [219]
is a simple polynomial-type curve t to the predictions of Eqs. (152)(156)
for water, and the correlation can be found in [200]. Utilizing an approxi-
mate equation of state for water, Leung and Grolmes [220] derived the
following correlation, based on the isentropic homogeneous-equilibrium
ow assumption for water:
G
''
2w
w1
1
_
1
1
p
`

2w1
2w
`
p
`
c
`
. (157)
Here,
p
`
P
`'
(T
)/P
'
(158)
c
C
P
`'
(T
)
v
`
, (159)
where T
is the subcooled liquid stagnation temperature and P

'
is the
stagnation pressure at channel inlet when the effect of channel entrance
irreversible pressure loss is accounted for:
P
'
P
K
'

A
''
A
'
` G`
''
2j
. (160)
Predictions of the aforementioned approximate correlations are com-
pared with typical experimental data of Ghiaasiaan et al. [208] in Fig. 35.
The ow passage in the latter data is very short (L /D1) and their
frictional losses are negligible. Model predictions are of course relatively
sensitive to the entrance pressure loss coefcient, K
'
. For sharp-edged
sudden contractions with very large contraction ratio, K
'
0.5 [221], and
with smooth or conical entrances K
'
will be lower. The isentropic homo-
geneous-equilibrium model well predicted the latter data, with standard
deviations of about 10% with K
'
0.5 and about 13% for K
'
0.3.
The isentropic homogeneous-equilibrium model is not appropriate for
application to cracks and slits with larger L /D, however, and overpredicts
the data because of the signicance of frictional pressure losses in cracks and
slits.
The critical ow of initially saturated liquid of a liquidvapor mixture in
a channel, with frictional pressure loss, has been modeled by Moody [222].
This model assumes thermal equilibrium between the vapor and liquid
phases and accounts for velocity slip by assuming a slip ratio everywhere
equal to the expression derived by Zivi [99], Eq. (68), for minimum entropy
production in a steady-state liquidvapor ow. In an earlier, and widely
used, model for critical, isentropic ow, Moody derived the same expression
for the slip ratio at the channel throat. The model of Moody [222] is based
on the integration of the one-dimensional two-phase momentum conserva-
tion equation along a constant-area channel,
dP/dz f
''
L /Du`
''
G`
2j
'
d
dz
[xU
'
(1 x)U
'
]G}, (161)
where f
''
is an average DArcy channel friction factor for liquid ow. The
one-dimensional energy conservation also gives
h
h
'
h
1
2
G`
_
(1 x)`
j`
'
(1 :)`

x`
j`
'
:`
. (162)
Note that : and x are related by the slip ratio, according to Eq. (32). Since
the uid mixture is saturated everywhere, the right-hand side of Eq. (161)
can be expanded by replacing d/dz with d/dP(dP/dz), obtaining d/dP of all
properties using appropriate thermodynamics data, and factoring out (dP/
dz) in all the terms that include it. Equation (162) is also utilized in the
manipulation of Equation (161), noting that dh
/dz 0. The nal result can

be represented as
,
'
`
'
I(P, h
, G)dP f
''
L
D
, (163)
where P
is the pressure at inlet to the channel (i.e., the point beyond which
irreversible pressure losses occur), P
`
is the pressure at the location of the
throat, and I is a function that involves the pressure, slip ratio, and
thermodynamic properties and their partial derivatives. The critical mass
ux is obtained by iteratively changing P
`
and solving Eq. (163) for G, until
dG/dP
`
0, at which point P
''
P
`
and G
''
G, Moody [222] suggested
the following expression for the two-phase multiplier:
u`
''
1 x
1 :
`
. (164)
Nabarayashi et al. [206, 207] compared their measured ow rates with
predictions of the aforementioned model of Moody [222], and noted that
Moodys model well predicted their data when pressure losses due to the
entrance effect and wall friction, and ow area reduction due to the wall
roughness, were correctly accounted for. The aforementioned model of
Moody [222] well predicted the measured ow rates in their fatigue-induced
cracks as well, once the entrance and frictional pressure losses, ow area
blockage by wall roughness, and pressure losses due to bending undulation
were all included in the calculations.
Based on the experimental data of Collier et al. [202, 203], a homogeneous-
nonequilibrium model was developed at Battelle-Columbus Laboratory,
which is briey described by Abdollahian et al. [223]. The model was based
on the widely used homogeneous, nonequilibrium model of Henry [224]. In
the model, which was coded into the computer program LEAK, the ow
was assumed to remain single-phase liquid between the entrance and a point
located at a distance of z 12D downstream from the entrance; the
acceleration and frictional pressure drops between the channel inlet and the
critical cross-section were obtained using channel averaged properties and
assuming homogeneous ow everywhere; the expansion of the vapor phase
was assumed to be isentropic; and quality was assumed to vary linearly
between z 12D and the throat.
Abdollahian et al. [223] and Chexal et al. [225] modied and improved
the LEAK model and developed the LEAK 01 model by using an isenthal-
pic ow assumption for pressure drop calculations, assuming a linear ow
area variation along the channel, and improving the two-phase frictional
pressure drop calculation. Noting that for subcooled liquid inlet conditions
the models based on the homogeneous-equilibrium ow in cracks appear to
generally predict saturated liquid conditions at the critical cross-section.
Abdollahian et al. [223] also proposed the following simple expression for
the critical mass ux in long racks and slits:
G
''
2[P
P
`'
(T
)]
v
"
(1 f L /D
"
) K
'
v
`
. (165)
Here, v
"
is the average homogeneous two-phase mixture specic volume:
v
"
v
x v
. (166)
The average properties, including v
and v
, are calculated at
P [P
P
`'
(T
)]/2. (167)
The average quality is dened as
x [h
(P )]/h
. (168)
Abdollahian et al. used the unrealistically high K
'
2.7, however, and
obtained f from Karmans correlation for rough-walled channels. The two
models of Abdollahian et al. (i.e., LEAK-01 and the aforementioned simpli-
ed model) well predicted the data of Collier et al. [201, 203], but were
shown by John et al. [105] to do relatively poorly when applied to the latter
authors data, as well as the data of Amos and Schrock [204].
E. Monris B:srn oN Nixric:i Soii1ioN or DirrrrN1i:i
CoNsrv:1ioN Eqi:1ioNs
Most of the models dealing with two-phase critical ow in large channels
include constitutive relations that may not apply to cracks and slits. The
models of [209, 217, 226, 227] all apply the two-uid technique and include
interfacial transfer models of doubtful applicability to microchannels and
cracks where interfacial slip is likely to be suppressed by surface tension.
Equal phasic velocities are assumed in [228, 229]. The latter references,
however, use models for bubble nucleation and relaxation-type delayed
evaporation that may not apply to microchannels.
Several models that specically address cracks and slits have been
developed and published. These models are now briey reviewed. Amos and
Schrock [204] and Lee and Schrock [214] modeled critical two-phase ow
in cracks assuming a homogeneous (equal phasic velocities) nonequilibrium
ow, by solving numerically the one-dimensional mixture mass, momentum,
and energy conservation equations. They used empirically adjusted closure
relations, however, to obtain agreement with experimental data. Flashing
was assumed to occur when a pressure undershoot (ie., local pressure lower
than the saturated pressure corresponding to the local pressure) is obtained,
given by
AP
SAP
'
, (169)
where AP
'
is the pressure undershoot associated with bubble nucleation,
as correlated empirically by Alamgir and Lienhard [210]:
AP
'
C
_
0.252o`(T /T *
''
)``'(1 14l"`)"`
(k
`
T *
''
)"`(1 j
'
/j
'
)
(170)
In this equation, k
`
is Boltzmanns constant, and T *
''
is the thermodynamic
critical temperature. Except for l, the depressurization rate, which must be
in Matm/s, the remaining dimensions are consistent. The correction par-
ameter, C, was empirically correlated by Amos and Schrock in terms of local
velocity [204], and as a function of Reynolds and Jacobs numbers by Lee
and Schrock [214], to obtain agreement between model predictions and
data. The evaporation of the metastable liquid phase (the relaxation of the
liquid phase superheat) was assumed to follow an exponential form, with a
time constant that was also empirically adjusted to t their data. Critical
ow was assumed to occur when the mixture velocity at the critical
cross-section was equal to the local velocity of sound, and the latter was
represented by an expression by Kroeger [230] representing zero interfacial
slip. Lee and Schrock [214] compared the predictions of their model with
data from several sources. Most of the data represented critical ow in large
channels, however.
The real intergranular cracks have complicated and tortuous congur-
ations. Schrock and Revankar [215] argued that for critical ow in real
intergranular corrosion cracks, the homogeneous-equilibrium model is ap-
propriate, since signicant ow separation and thermal nonequilibrium are
unlikely. They developed the fast-running homogeneous-equilibrium code
SOURCE, in which the mixture mass, momentum, and energy conservation
equations were numerically solved using the nite-difference technique, and
the discharge rate was iteratively varied until the mixture velocity at the
critical cross-section was equal to the velocity of sound in homogeneous-
equilibrium two-phase ow. They assumed an equivalent friction factor that,
in addition to wall friction, accounts for the irreversible pressure losses
associated with ow disturbances and tortuosity. Shrock et al. [215]
compared the predictions of SOURCE with 61 out of the 82 BCL data that
were found to be acceptable. An optimum friction factor was found that best
tted the data associated with each test section, and thereby they developed
a methodology for the prediction of the equivalent friction factor for real
cracks. Since SOURCE predictions showed a systematic dependence on
inlet subcooling, furthermore, to obtain agreement between model and data,
Schrock and Revankar also developed an inlet subcooling correction factor
that must be multiplied by the mass ux predicted by the code in order to
calculate the correct mass ux.
Feburie et al. [216] developed a nonequilibrium model assuming the
existence of three phases: saturated vapor, saturated liquid, and metastable
superheated liquid. The pressure undershoot leading to the initiation of
evaporation was modeled according to [231]
P
P
`'
(T
'`
)
k
, (171)
where T
'`
is the temperature of the superheated liquid and k
0.95 to
0.97. Following the initiation of evaporation, the three phases are assumed
to move at the same velocity (homogeneous ow). Their one-dimensional
steady-state conservation equations for the mixture mass, momentum, and
energy are
d
dz
(Au/v
"
) 0 (172)
dP
dz
j
"
U
dU
dz

p
"
t
"
A
(173)
GA
ds
"
dz
GAC
''
_
ln
T
`'
T
'`
T
'`
T
`'
1
dy
dz
_
p
"'`
q"
"'`
T
'`
p
"'`
q"
"'`
p
"'
q"
"'
T
`'

U
_
p
"'`
t
"'`
T
'`
p
"'`
t
"'`
p
"'
t
"'
T
`'

, (174)
where subscripts L M, L S, and G represent the metastable superheated
liquid, saturated liquid, and saturated vapor, respectively, and (1 y)
represents the metastable liquid mass fraction in the mixture. The last two
terms on the right-hand side of Eq. (174) are the entropy sources due to wall
heat transfer and friction. The parameter p
"'`
represents the wall-super-
heated liquid perimeter through which the heat ux q"
"'`
is transferred, and
p
"'`
and t
"'`
represent the wall-superheated liquid perimeter and its
associated shear stress, respectively. For simplicity, T
'`
constant is as-
sumed once bubble nucleation starts. Other parameters are dened similar-
ly. The mixture entropy is dened as
s
"
(1 y)s
'`
xys
(1 x)ys
'`
, (175)
where xy and (1 x)y are the mass fractions of saturated vapor and
saturated liquid in the mixture, respectively. Other mixture properties,
including v
"
, are dened similarly. Two additional equations are provided
by the equation of state,
v
"
v
"
(P, s
"
, y), (176)
and a closure relation in the form [232]
dy
dz
k(1 y)
_
P
`'
(T
'`
) P
P
`''
P
`'
(T
'`
)
"
, (177)
where k is a constant assumed to depend on geometry according to
k k
`
p
"
A
. (178)
Using the Moby-Dick experimental data, k
`
0.02 m. Equations (172)
(174), (176) (upon differentiation), and (177) constitute ve coupled equa-
tions, which can be represented in the form of a quasi-linear set of coupled
ordinary differential equations:
A
dY
dz
B (179)
Y(P, U, v
"
, s
"
, y)'. (180)
Note that if v
"
and s
"
are known, x and T
'`
can be calculated. These
equations can be numerically integrated, and by iteration the critical mass
ow rate, which leads to the vanishing of the determinant of the coefcient
matrix A, can be specied. For wall friction and heat transfer, Feburie et al.
used various correlations.
Feburie et al. compared preditions of their model for G
''
with 70 data
points from John et al. [105], with agreement within _12%, and with 14
data points from Amos and Schrock [204], with agreement within about
_5%. Sensitivity analysis indicated that the model predictions were rela-
tively sensitive to the magnitude of the constant k
in Eq. (171), the wall

friction, and entrance pressure loss, and were insensitive to the magnitude
of the constant k
`
.
Geng and Ghiaasiaan [218] recently developed a homogeneous-equilib-
rium model for the critical ow of an initially subcooled liquid through
cracks and slits, where the effect of a dissolved noncondensable in the inlet
subcooled liquid was accounted for. They assumed that: (a) the solubility of
the noncondensable in the liquid is low; (b) the liquid and vapor gas phases
are everywhere at thermodynamic equilibrium and at equilibrium with
respect to the concentration of the noncondensable; and (c) the vapor
noncondensable gas mixture is everywhere saturated with respect to vapor.
For a channel with uniform cross-sectional area that is inclined with respect
to the horizontal plane by the angle 0
, the mixture mass, momentum, and

energy conservations are
d
dz
(j
"
U) 0 (181)
j
"
U
dU
dz

dP
dz
j
"
g sin 0

p
"
t
"
A
(182)
d
dz
U
_
j
'
(1 :)h
:j
1
2
j
"
U`
j
"
Ug sin 0
p
"
q"
"
/A,
(183)
where, assuming that the noncondensable is an ideal gas,
j
"
j
'
(1 :)
_
j
PP
`'
(T )
(R/M
)T
:, (184)
where T is the local mixture temperature.
Utilizing Henrys law [233] for the equilibrium between the gas and liquid
phases with respect to the concentration of the noncondensable, Geng and
Ghiaasiaan derived
d
dz
U
_
j
(1 :)M
'
j
'
:
M
'
(P/C
"
) [1 (Mg/M
)]M
'
0 (185)
d
dz
M
'
PP
`'
(T )
C
"

0. (186)
Geng and Ghiaasiaan expanded Eqs. (181)(183), (185), and (186) using
thermodynamic relation, and cast the aforementioned equations in the form
represented by Eq. (179), with
Y (U, P, :, h
, M
'
)'. (187)
Geng and Ghiaasiaan applied the correlation of John et al. [105], Eq.
(85), for single-phase wall friction and applied the homogeneous mixture
model, along with McAdams correlation (Eqs. (44)(48) for two-phase
pressure drop. They numerically integrated the foregoing differential con-
servation equations and, by iteratively varying the channel mass ow rate,
specied G
''
as the mass ux leading to det A0 at the critical cross-
section. Geng and Ghiaasiaan compared the predictions of their model with
the experimental data of Amos and Schrock [204] and John et al. [105]
with satisfactory agreement. To examine the effect of noncondensables, they
chose the data of Amos and Schrock as the basis for parametric calculations.
Figure 36 represents typical results, where model predictions with pure
water and water saturated with nitrogen at inlet are both presented. The
results indicate that desorption of nitrogen (or air) initially dissolved in
subcooled water can reduce the critical mass ux by several percent.
VIII. Concluding Remarks
The recent developments related to gasliquid two-phase ow, forced-
ow subcooled boiling, and the critical (choked) ow of initially subcooled
liquids, in channels with hydraulic diameters of the order of 0.1 to 1 mm,
were reviewed in this article. The hydrodynamic phenomena reviewed
Fic. 36. Comparison between the predictions of the model of Geng and Ghiaasiaan [218]
and the experimental data of Amos and Schrock [204] and the effect of noncondensable gas
on model predictions. (With permission from [218].)
included the two-phase ow regimes, void fraction, and frictional pressure
drop in narrow rectangular and annular passages, a micro-rod bundle, and
microchannels under conditions where surface tension and inertial forces are
both signicant. The boiling phenomena reviewed included the onset of
nucleate boiling (ONB), onset of signicant void (OSV), onset of ow
instability (OFI), and critical heat ux. The critical (choked) ow of initially
subcooled liquids in capillaries, cracks, and slits was also addressed.
The observed major two-phase ow regimes in microchannels are mor-
phologically similar to the ow regimes in large channels. However, they can
be insensitive to channel orientation and are inuenced by surface wettabil-
ity. The commonly applied predictive methods for the ow regime transi-
tions in large channels overall fail to predict the microchannel data well.
Some empirical correlations that have been developed based on the micro-
gravity experiments, however, appear to agree with microchannel data
satisfactorily. The bulk of the existing microchannel data have been ob-
tained with air and water, and more experimental data examining the effects
of uid properties, in particular the surface tension, surface wettability, and
liquid viscosity, are needed. The existing predictive methods for void
fraction and two-phase frictional pressure drop are also generally inad-
equate for microchannels, in particular for the annular ow regime.
With respect to boiling, the fundamental bubble ebullition phenomena in
microchannels are likely to be different from the qualitatively well-under-
stood bubble phenomena associated with boiling in large channels, because
of the occurrence of very large velocity and temperature gradients in the
former. Bubble ebullition in microchannels needs to be investigated.
The critical heat ux data associated with moderate and high mass uxes
in microchannels are predicted satisfactorily by some of the recent and
widely used empirical correlations, because of the presence of relevant data
in the data bases of these correlations. The available data associated with
low-ow critical heat ux in microchannels, however, are unlike similar data
in large channels and suggest the occurrence of metastable superheated
liquid prior to dryout. Investigations aimed at the elucidation of the
hydrodyamic and evaporation phenomena associated with annular ow
regime in microchannels are thus needed.
The critical (choked) ow of an initially subcooled liquid in a capillary,
crack, or slit is strongly inuenced by pressure drop and can be predicted
using models based on the homogeneous ow assumption.
Nomenclature
A coefcient matrix
A ow area; dimensionless
coefcient
B column vector
C constant
Ca capillary number
C
"
two-phase distribution
coefcient
Cp specic heat
D diameter
D
hydraulic diameter
D
'
, D
inner, outer diameters

Eo Eotvos number
F force term
F
'
, F
, forces acting on a bubble (Figs.

26 and 27)
F
'"
, F
"
, F
`
,
F
''
, F
"
f DArcy friction factor
f ' Fanning friction factor
Fr Froude number
G mass ux
g gravitational constant
h convection heat transfer
coefcient
h
'
liquid depth
h specic enthalpy
h
specic heat of vaporization

k thermal conductivity
k
`
Boltzmanns constant
K
'
entrance loss coefcient
L length
M molar mass
M mass fraction
N
viscosity number
P pressure
AP
pressure undershoot
p
"
, p
"
heated and wetted perimeters
Pe Peclet number
Pr Prandtl number
P*
''
thermodynamic critical
pressure
P*
'
reduced pressure
Q dimensionless heat ux
q"
"
wall heat ux
R universal gas constant
r radius
Re Reynolds number
S slip ratio; gap distance
s specic enthalpy
s
specic enthalpy of
vaporization
St Stanton number
T temperature
t time
T *
''
thermodynamic critical
temperature
U velocity
U* friction velocity
U
`
rise velocity
v specic volume
V
'
gas drift velocity
W width
We Weber number
X Martinelli parameter
x quality
Y Shahs correlation parameter
y distance from wall; the
combined mass fraction of
vapor and saturated liquid
z axial coordinate
Grrx Sxxnois
: Void fraction
[ Volumetric quality
o Gap distance; lm thickness
I Function in Moodys critical
ow model
c Roughness
p Pressure ratio
0
function dened in Eq. (133);

angle with respect to the
horizontal plane
dimensionless coefcient
z Laplace length scale
j dynamic viscosity
j density
o surface tension
t shear stress
v kinematic viscosity
parameter dened in Eq. (110)
u two-phase multiplier
u* shape factor dened in Eq. (78)
function dened in Eq. (132);
contact angle
+ function dened in Eqs.
(119)(121)
"
function dened in Eq. (145)
D function dened in Eq. (134)
c function dened in Eq. (159)
Sinscii1s
B bubble, vapor blanket
b boiling
cr critical (choked)
E effective
eq equilibrium
eqv equivalent
f frictional; saturated liquid
FC forced convection
G gas
g saturated vapor
GS supercial gas
G0 all-gas
H heated
h homogeneous
i inlet
L liquid
L M metastable liquid
L S supercial liquid; saturated
liquid
L 0 all liquid
m average
mod modied
n noncondensable
p pipe
sat saturation
T P two-phase
v vapor
w wall
0 stagnation
o lm edge
Siirscii1s
mean
Annrvi:1ioNs
CHF critical heat ux
HHF high heat ux
NVG net vapor generation
OFI onset of ow instability
ONB onset of nucleate boiling
OSV onset of signicant void
References
1. Tien, C. L., Qin, T. Q., and Norris, P. M. (1994). Microscale thermal phenomena in
contemporary technology, T hermal Science and Technology 2, 111.
2. Vafai, K., and So zen, M. (1990). A comparative analysis of multiphase transport in porous
media. Ann. Rev. Heat Transfer 3, 145162.
3. Dhir, V. K. (1994) Boiling and two-phase ow in porous media. Ann. Rev. Heat Transfer
5, 303350.
4. Wang, C. Y., and Cheng, P. (1997) Multiphase ow and heat transfer in porous media.
Advances in Heat Transfer 30, 93196.
5. Fairbrother, F., and Stubbs, A. E. (1935). Studies in electro-endosmosis, J. Chem. Soc.,
Part 1, 527529.
6. Marchessault, R. N., and Mason, S. G. (1960). Flow of entrapped bubbles through a
capillary. Ind. Eng. Chem. 52, 7984.
7. Taylor, G. I. (1961). Deposition of a viscous uid on the wall of a tube. J. Fluid Mech.
10, 161165.
8. Bretherton, F. B. (1961). The motion of long bubbles in tubes. J. Fluid Mech. 10, Part 2,
166188.
9. Suo, M., and Grifth, P. (1964). Two-phase ow in capillary tubes, J. Basic Eng. 86,
576582.
10. Brauner, N., and Moalem-Maron, D. (1992). Identication of the range of small
diameter conduits, regarding two-phase ow pattern transitions. Int. Comm.. Heat Mass
Transfer 19, 2939.
11. Damianides, C. A., and Westwater, J. W. (1988), Two-phase ow patterns in a compact
heat exchanger and in small tubes. Proc. UK National Heat Transfer Cont., 2nd, 1988, pp.
12571268.
12. Triplett, K. A., Ghiaasiaan, S. M., Abdel-Khalik, S. I., and Sadowski, D. L. (1999).
Gasliquid two-phase ow in microchannels. Part I: Two-phase ow patterns. Int. J.
Multiphase Flow 25, 377394.
13. Govier, F. W., and Aziz, K. (1972) T he Flow of Complex Mixtures in Pipes. Robert E.
Krieger, Malabar, FL.
14. Collier, J. G., and Thome, J. R. (1994) Convective Boiling and Condensation. Clarendon
Press, Oxford, England.
15. Taitel, Y., and Dukler, A. E. (1976). A model for predicting ow regime transitions in
horizontal and near horizontal gasliquid ow. AIChE J. 22, 4755.
16. Taitel, Y., Lee, N., and Dukler, A. E. (1978). Transient gasliquid ow in horizontal pipes:
modeling the ow pattern transitions. AIChE J. 24, 920924.
17. Taitel, Y. Bornea, D., and Dukler, A. E. (1980). Modeling ow pattern transitions for
steady upward gasliquid ow in vertical tubes. AIChE J. 26, 345354.
18. Dukler, A. E., and Taitel, Y. (1986). Flow pattern transitions in gasliquid systems:
measurement and modeling. In Multiphase Science and Technology (G. F. Hewitt, J. M.
Delhaye, and N. Zuber, eds.), Vol. 2, pp. 194. Hemisphere, Washington, D.C.
19. Barnea, D. (1987). A unied model for predicting ow-pattern transitions for the whole
range of pipe inclinations. Int. J. Multiphase Flow 13, 112.
20. Taitel, Y. (1990). Flow pattern transition in two-phase ow. Proc. Int. Heat Transfer Cont.,
9th, 1990, pp. 237254. Hemisphere, New York.
21. Spedding, P. L., and Spence, D. R. (1993) Flow regimes in two-phase gasliquid ow. Int.
J. Multiphase Flow, 19, 245280.
22. Oya, T. (1971). Upward liquid ow in small tube into which air streams. (First report,
experimental apparatus and ow patterns). Bull. JSME 14, 13201329.
23. Barnea, D. Lulinkski, Y., and Taitel, Y. (1983). Flow in small diameter pipes. Can. J.
Chem. Eng. 61, 617620.
24. Barajas, A. M., and Panton, R. L. (1993). The effect of contact angle on two-phase ow
in capillary tubes, Int. J. Multiphase Flow 19, 337346.
25. Fukano, T., and Kariyasaki, A. (1993). Characteristics of gasliquid two-phase ow in a
capillary. Nucl. Eng. Design 141, 5968.
26. Mishima, K., and Hibiki, T. (1996). Some characteristics of airwater two-phase ow in
small diameter vertical tubes. Int. J. Multiphase Flow 22, 703712.
27. Lowry, B., and Kawaji, M. (1988). Adiabatic vertical two-phase ow in narrow ow
channels. AIChE Symp. Ser. 48, 133139.
28. Wambsganss, M. W., Jendrzejczyk, J. A., and France, D. M. (1991). Two-phase owpatterns
and transitions in a small, horizontal, rectangular channel. Int. J. Multiphase Flow7, 327342.
29. Ali, M. I., and Kawaji, M. (1991) The effect of ow channel orientation on two-phase ow
in a narrow passage between at plates. Proc. ASME/JSME T hermal Eng. Conf., 1991,
Vol. 2, pp. 183190.
30. Ali, M. I., Sadatomi, M., and Kawaji, M. (1993). Adiabatic two-phase ow in narrow
channels between two at plates. Can. J. Chem. Eng. 71, 657666.
31. Mishima, K., Hibiki, T., and Nishihara, H. (1993). Some characteristics of gasliquid ow
in narrow rectangular ducts. Int. J. Multiphase Flow 19, 115124.
32. Wilmarth, T., and Ishii, M. (1994). Two-phase ow regimes in narrow rectangular vertical
and horizontal channels. Int. J. Multiphase Flow 37, 17491758.
33. Fourar, M., and Bories, S. (1995). Experimental study of airwater two-phase ow
through a fracture (narrow channel). Int. J. Multiphase Flow 21, 621637.
34. Narrow, T. L., Ghiaasiaan, S. M., Abdel-Khalik, S. I., and Sadowski, D. L. (1999).
Gasliquid two-phase ow patterns and pressure drop in a horizontal micro-rod bundle,
Parts I and II Int. J. Multiphase Flow 25, 377410.
35. Ekberg, N. P., Ghiaasiaan, S. M., Abdel-Khalik, S. I., Yoda, M., and Jeter, S. M. (1999).
Gasliquid two-phase ow in narrow horizontal annuli. Nucl. Eng. Design, 192, 5980.
36. Thulasidas, M. A., Abraham, M. A., and Cerro, R. L. (1995). Bubble-train ow in
capillaries of circular and square cross-section. Chem. Eng. Sci., 50, 183199.
37. Thulasidas, T. C., Abraham, M. A., and Cerro, R. I. (1999). Dispersion during bubble train
ow in capillaries. Chem. Eng. Sci. 54, 6176.
38. Zhao, L., and Rezkallah, K. S. (1993). Gasliquid ow patterns at microgravity condi-
tions. Int. J. Multiphase Flow 19, 751763.
39. Oya, T. (1971). Upward liquid ow in small tube into which air streams (second report.,
presure drop at the conuence). Bull. JSME, 14, 13301339.
40. Smedley, G. (1990). Preliminary drop-tower experiments on liquid-interface geometry in
partially lled containers at zero gravity. Exp. Fluids 8, 312318.
41. Mandhane, J. M., Gregory, G. A., and Aziz, K. (1974). A ow pattern map for gasliquid
ow in horizontal pipes. Int. J. Multiphase Flow, 1, 537553.
42. Mishima, K., and Ishii, M. (1984). Flow regime transition criteria for two-phase ow in
vertical tubes. Int. J. Heat Mass Transfer 27, 723737.
43. Rezkallah, K. S. (1996). Weber number based ow-pattern maps for liquidgas ows at
microgravity. Int. J. Multiphase Flow 22, 12651270.
44. Jayawardena, S. S., Balakotaiah, V., and Witte, L. C. (1997). Flow pattern transition maps
for microgravity two-phase ows. AIChE J. 43, 61371640.
45. Bousman, W. S., McQuillen, J. B., and Witte, L. C. (1996). Gasliquid ow patterns in
microgravity: Effects of tube diameter, liquid viscosity, and surface tension. Int. J.
Multiphase Flow 22, 10351053.
46. Zuber, N., and Findlay, J. (1965). Average volumetric concentration in two-phase ow
systems. J. Heat Transfer 87, 453468.
47. Wallis, G. B. (1969). One-Dimensional Two-Phase Flow. McGraw-Hill, New York.
48. Ishii, M. (1977). One-dimensional drift-ux model and constitutive equations for relative
motion between phases in various ow regimes. Argonne National Laboratory Report
ANL-77-47, Argonne, IL.
49. Krishna, V. S., and Kowalski, J. E. (1984). Stratied-slug ow transition in a horizontal
pipe containing a rod bundle. AIChE Symp. Ser., No. 236, 80, 282289.
50. Aly, M. M. (1981). Flow regime boundaries for an interior subchannel of a horizontal 37
element bundle. Can. J. Chem. Eng. 59, 158163.
51. Osamusali, S. E., Groeneveld, D. C., and Cheng, S. C. (1992). Two-phase ow regimes and
onset of ow instability in horizontal 37-rod bundles. Heat Technol 10, 4674.
52. Kariyasaki, A., Fukano, T., Ousaka, A., and Kagawa, M. (1992). Isothermal airwater
two-phase up- and downward ows in vertical capillary tube (1st report, ow pattern and
void fraction). Trans. JSME (Ser. B.) 58, 26842690.
53. Bao, Z.-Y., Bosnich, M. G., and Haynes, B. S. (1994). Estimation of void fraction and
pressure drop for two-phase ow in ne passages. Trans. Inst. Chem. Eng. 72A, 625532.
54. Triplett, K. A., Ghiaasiaan, S. M., Abdel-Khalik, S. I., LeMouel, A., and McCord, B. N.
(1999). Gasliquid two-phase ow in microchannels. Part II: Void fraction and pressure
drop. Int. J. Multiphase Flow 25, 395410.
55. Chexal, B., Merilo, M., Maulbetsch, M., Horowitz, J., Harrison, J., Westacott, J., Peterson,
C., Kastner, W., and Schmidt, H. (1997). Void Fraction Technology for Design and Analysis.
Electric Power Research Institute, Palo Alto, CA.
56. Premoli, A., Francesco, D., and Prina, A. (1971). A dimensionless correlation for
determining the density of two-phase mixtures. L o Termotecnica 25, 1726.
57. Hewitt, G. F. (1983). Gas-liquid ow. In Heat Exchanger Design Handbook, Vol. 2, pp.
229238. Hemisphere, Washington, D.C.
58. Buterworth, D. (1975). A comparison of some void-fraction relationships for co-current
gasliquid ow. Int. J. Multiphase Flow 1, 845850.
59. Lockhart, R. W., and Martinelli, R. C. (1949). Proposed correlations of data for isothermal
two-phae, two-component ow in a pipe. Chem. Eng. Prog. 45, 3949.
60. Jones, O. C., Jr., and Zuber, N. (1979). Slugannular transition with particular reference
to narrow rectangular ducts. In Two-Phase Momentum, Heat and Mass Transfer in
Chemical Process and Energy Engineering Systems (F. Durst, G. V. Tsiklauri and N. Afgan,
eds.), Vol. 1, pp. 345355. Hemisphere, Washington, D.C.
61. Wilmarth, T., and Ishii, M. (1997). Interfacial area concentration and void fraction of
two-phase ow in narrow rectangular vertical channels. J. Fluids Eng. 119, 916922.
62. Kelessidis, V. C., and Dukler, A. E. (1989). Modeling ow pattern transitions for upward
gasliquid ow in vertical concentric and eccentric annuli. Int. J. Multiphase Flow 15,
173191.
63. Osamusali, S. I. and Chang, J. S. (1988). Two-phase ow regime transition in a horizontal
pipe and annulus ow under gasliquid two-phase ow. ASME FED, Vol. 72, pp. 6369.
ASME, New York.
64. Yu, D., Warrington, R., Barron, R., and Ameal, T. (1995). An experimental and theoretical
investigation of uid ow and heat transfer in microtubes. Proc. ASME/JSME T hermal
Eng. Conf., 1995, Vol. 1, pp. 523530.
65. Choi, S. B., Barron, R. F. and Warrington, R. O. (1990). Fluid ow and heat transfer in
microtubes. Proc. ASME 1991 Winter Annual Meeting, DSC, Vol. 32, pp. 123134.
ASME, New York.
66. Adams, T. M., Abdel-Khalik, S. I., Jeter, S. M., and Qureshi, Z. H. (1997). An experimental
investigation of single-phase forced convection in microchannels. Int. J. Heat Mass
Transfer 41, 851857.
67. Adams, T. M., Ghiaasiaan, S. M., and Abdel-Khalik, S. I. (1999). Effect of dissolved
noncondensables on hydrodyamics of microchannels subject to liquid forced convection.
J. Enhanced Heat Transfer 6, 395403.
68. Adams, T. M., Ghiaasiaan, S. M., and Abdel-Khalik, S. I. (1999). Enhancement of liquid
forced convection heat transfer in microchannels due to the release of dissolved noncon-
densables. Int. J. Heat Mass Transfer 42, 35633573.
69. Michiyoshi, I. (1978). Two-phase two-component heat transfer. Proc. Int. Heat Transfer
Conf., 6th, 1978, Vol. 6, pp. 219233.
70. Mikol, E. P. (1963). Adiabatic single and two-phase ow in small bore tubes. ASHRAE
J. 5, 7586.
71. Olson, C. O., and Sunden, B. (1994). Pressure drop characteristics of small-sized tubes.
ASME Paper No. 94-WA/HT-1.
72. Acosta, R. E., Buller, R. H., and Tobias, C. W. (1985). Transport processes in narrow
(capillary) channels. AIChE J. 81, 473482.
73. Tong, W., Bergles, A. E., and Jensen, M. K. (1997). Pressure drop with highly subcooled
ow boiling in small-diameter tubes. Exp. T hermal Fluid Sci. 15, 202212.
74. Stanley, R. S., Barron, F. F., and Ameel, T. A. (1997). Two-phase ow in microchannels.
In ASME Microelectromechanical Systems, DSC-Vol. 62/HTD-Vol. 354, pp. 143152.
ASME, New York.
75. McAdams, W. H. (1954). Heat Transmission, 3rd ed. McGraw-Hill, New York.
76. Soliman, M., Schuster, J. R., and Berenson, P. J. (1968). A general heat transfer correlation
for annular ow condensation. J. Heat Transfer 90, 267276.
77. Troniewski, L., and Ulbrich, R. (1984). Two-phase gasliquid ow in rectangular
channels. Chem. Eng. Sci., 39, 751765.
78. Chisholm, D., and Laird, A. D. K. (1958). Two-phase ow in rough tubes. Trans. ASME
80, 276283.
79. Chisholm, D. (1967). A theoretical basis for the LockhartMartinelli correlation for
two-phase ow. Int. J. Heat Mass Transfer 10, 17671778.
80. Chisholm, D. A. (1973). Pressure gradients due to friction during the ow of evaporat-
ing two-phase mixture in smooth tubes and channels. Int. J. Heat Mass Transfer 16,
347358.
81. Friedel, L. (1979). Improved pressure drop correlations for horizontal and vertical
two-phase pipe ow. 3R International 18, 485492.
82. Beattie, D. R. H., and Whalley, P. B. (1982). A simple two-phase frictional pressure drop
calculation method. Int. J. Multiphase Flow 8, 8387.
83. Colebrook, C. R. (1939). Turbulent ow in pipes with particular reference to the transition
region beween the smooth and rough pipe laws. J. Inst. Civil Eng. 11, 133156.
84. Taitel, Y., and Dukler, A. E. (1976). A theoretical approach to the LockhartMartinelli
correlation for stratied ow. Int. J. Multiphase Flow 2, 591595.
85. Asali, J. C., Hanratty, T. J., and Andreussi, P. (1985). Interfacial drag and lm height for
vertical annular ow. AIChE, J. 31, 895902.
86. Yao, G., and Ghiaasiaan, S. M. (1996). Wall friction in annular-dispersed two-phase ow.
Nucl. Eng. Design 163, 149161.
87. Mikol, E. P. (1963). Adiabatic single and two-phase ow in small bore tubes. ASHRAE
J. 5, 7586.
88. Marcy, G. P. (1949). Pressure drop with change of phase in a capillary tube. Refrig. Eng.
57, 53.
89. Bolstad, M. M., and Jordan, R. C. (1948). Theory and use of the capillary tube expansion
device. Refrig. Eng. 56, 519.
90. Hopkins, N. E. (1950). Rating the restrictor tube. Refrig. Eng. 58, 1087.
91. Koizumi, H., and Yokohama, K. (1980). Characteristics of refrigerant ow in a capillary
tube. ASHRAE Trans., Part 2 86, 1927.
92. Lin, S., Kwok, C. C. K., Li, R.-Y., Chen, Z.-H., and Chen, Z.-Y. (1991). Local frictional
pressure drop during vaporization of R-12 through capillary tubes. Int. J. Multiphase
Flow 17, 95102.
93. Churchill, S. W. (1977). Frictional equation spans all uid ow regimes. Chem. Eng. 84,
9192.
94. Ungar, K. E., and Cornwell, J. D. (1992). Two-phase presure drop of ammonia in small
diameter horizontal tubes. Paper presented at AIAA 17th Aerospace Ground Testing
Conf., Nashville, TN, July 68, 1992.
95. Bowers, M. B., and Mudawar, I. (1994). High ux boiling in low ow rate, low pressure
drop mini-channel and micro-channel heat sinks. Int. J. Heat Mass Transfer 37, 321332.
96. Yan, Y.-Y., and Lin, T.-F. (1999). Evaporation heat transfer and pressure drop to
refrigerant R-134a in a small pipe. Int. J. Heat Mass Transfer 41, 41834194.
97. Yan, Y.-Y., and Lin. T.-F. (1999). Condensation heat transfer and pressure drop of
refrigerant R-134a in a small pipe. Int. J. Heat Mass Transfer 42, 697708.
98. Yang, C.-Y., and Webb, R. L. (1996). Friction pressure drop of R-12 in small hydraulic
diameter extruded aluminum tubes with and without micro-ns. Int. J. Heat Mass
Transfer 39, 801809.
99. Zivi, S. M. (1964). Estimation of steady state steam void-fraction by means of principle of
minimum entropy production. ASME Trans., Series C. 86, 237252.
100. Akers, W. W., Deans, H. A., and Crosser, O. K. (1959). Condensation heat transfer within
horizontal tube. Chem. Eng. Prog. Symp. Ser. 55, 171176.
101. Cornish, R. J. (1928). Flow in a pipe of rectangular cross-section. Proc. Roy. Soc. A 120,
691700.
102. Jones, O. C., Jr. (1976). An improvement in the calculation of turbulent friction in
rectangular ducts. J Fluid Eng. 98, 173181.
103. Conish, R. J. (1928). Flow in a pipe of rectangular cross-sections. Proc. Roy. Soc. A 120,
691700.
104. Sadatomi, Y., Sato, Y., and Saruwatari, S. (1982). Two-phase ow in vertical noncircular
channels. Int. J. Multiphase Flow 8, 641655.
105. John, H., Reimann, J., Westphal, F., and Friedel, L. (1988). Critical two-phase ow
through rough slits. Int. J. Multiphase Flow 14, 155174.
106. Hsu, Y. Y., and Graham, R. W. (1986). Transport Processes in Boiling and Two-Phase
Systems. American Nuclear Society, LaGrange Park, IL.
107. Carey, V. P. (1992). L iquidVapor Phase-Change Phenomena. Hemisphere, Washington,
D.C.
108. Stralen, S. V., and Cole, R. (1997). Boiling Phenomena, 2 vols. Hemisphere, Washington,
D.C.
109. Tuckermann, D. B., and Peasa, R. F. (1981). High performance heat sinking for VLSI.
IEEE Electr. Dev. L ett. EDL-2, 126129.
110. Peng, X. F., and Wang, B. X. (1994). Liquid ow and heat transfer in microchannels
with/without phase change. Heat Transfer 1994, Proc. Int. Heat Transfer Conf., 10th, Vol.
5, pp. 159177.
111. Duncan, A. B., and Peterson, G. P. (1994). Review of microscale heat transfer. Appl. Mech.
Rev. 47, 397428.
112. Lazarek, G. M., and Black, H. S. (1982). Evaporative heat transfer, pressure drop and
critical heat ux in a small vertical tube with R-113. Int. J. Heat Mass Transfer 25, 945960.
113. Tran, T. N., Wambsganss, M. W., France, D. M., and Jendrzejczyk, J. A. (1993). Boiling
heat transfer in a small, horizontal, rectangular channels. AIChE Symp. Ser. 89, 253261.
114. Wambsganss, M W., France, D. M., Jendrzejczyk, J. A., and Tran, T. N. (1993). Boiling
heat transfer in a horizontal small-diameter tube. J. Heat Transfer 115, 963972.
115. Tran, T. N., Wambsganss, M. W., and France, D. M (1996). Small circular- and
rectangular-channel boiling with two refrigerants. Int. J. Multiphase Flow 22, 485498.
116. Lahey, T. R., Jr., and Moody, F. J. (1993). T he Thermal-Hydraulics of Boiling Water
Nuclear Reactors, 2nd ed. American Nuclear Society, LaGrange Park, IL.
117. Dix, G. E. (1971). Vapor void fractions for forced convection with subcooled boiling at
low ow rates. Ph.D. Thesis, Univ. California, Berkeley. Also, General Electric Report
NEDO-10491.
118. Bibeau, E. L., and Salcudean, M. (1990). The effect of ow direction on void growth at
very low velocities and low pressure. Int. Comm. Heat Mass Transfer 17, 1925.
119. Bibeau, E. L., and Salcudean, M. (1994). A study of bubble ebullition in forced convective
subcooled nucleate boiling at low pressure. Int. J. Heat Mass Transfer 37, 22452259.
120. Bibeau, E. L., and Salcudean, M. (1994). Subcooled void growth mechanisms and
prediction at low pressure and low velocity. Int. J. Multiphase Flow 20, 837863.
121. Saha, P., and Zuber, N. (1974). Point of net vapor generation and vapor void fraction in
subcooled boiling. Poc. Int. Heat Transfer Conf., 5th, Vol. 4, pp. 175179.
122. Unal, H. C. (1975). Determination of the initial point of net vapor generation in ow
boiling systems. Int. J. Heat Mass Transfer 18, 10951099.
123. Levy, S. (1967). Forced convection subcooled boiling: prediction of vapor volumetric
fraction. Int. J. Heat Mass Transfer 10, 951965.
124. Staub, F. W. (1968). The void fraction in subcooled boiling: prediction of the initial point
of net vapor generation. J. Heat Transfer 90, 151157.
125. Rogers, T. J., Salcudean, M., Abdullah, Z. McLeond, D., and Poirier, D. (1987). The onset
of signicant void in up-ow boiling of water at low pressure and velocities. Int. J. Heat
Mass Transfer 30, 22472260.
126. Boure , J. A. Bergles, A. E., and Tong, L. S. (1973). Review of two-phase ow instability.
Nucl. Eng. Design 25, 165192.
127. Yadigaroglu, G. (1981). Two-phase ow instabilities and propagation phenomena. In
T hermohydraulics of Two-Phase Systems for Industrial Design and Nuclear Engineering (M.
Delhaye, M. Giot, and L M. Riethermuller, eds.), pp. 353403. Hemisphere, Washington,
D.C.
128. Marsh, W. J., and Mudawar, I. (1989). Predicting the onset of nucleate boiling in wavy
free-falling turbulent liquid lms. Int. J. Heat Mass Transfer 32, 361378.
129. Bergles, A. E., and Rohsenow, W. M. (1964) The determination of forced convection
surface boiling heat transfer. Int. J. Heat Mass Transfer 86, 365372.
130. Inasaka, F., Nariai, H., and Shimura, T. (1989). Pressure drops in subcooled boiling in
narrow tubes. Heat Transfer: Japanese Research 18, 7082.
131. Vandervort, C. L., Bergles, A. E., and Jensen, M. K. (1992). Heat transfer mechanisms in
very high heat ux subcooled boiling. ASME, Fundamentals of Subcooled Flow Boiling,
HTD-Vol. 217, pp. 19.
132. Kennedy, J. E., Roach, G. M., Jr., Dowling, M. F., Abdel-Khalik, S. I., Ghiaasiaan, S. M.,
Jeter, S. M., and Qureshi, Z. H. (2000). The onset of ow instability in uniformly heated
horizontal microchannels. J. Heat Transfer 122, 118125.
133. Yin. S. T., and Abdelmessih, A. H. (1974). Prediction of incipient ow boiling from a
uniformly heated surface. AIChE Symp. Ser. 164, 236243.
134. Hino, R., and Ueda, T. (1985). Studies on heat transfer and ow characteristics in
subcooled ow boiling Part I. Boiling characteristics. Int. J. Heat Mass Transfer 11,
269281.
135. Roch, G. M., Jr., Abdel-Khalik, S. I., Ghiaasiaan, S. M., and Jeter, S. M. (1999). Low-ow
onset of ow instability in heated microchannels. Nucl. Sci Eng. 133, 106117.
136. Rogers, J. T., and Li, J.-H. (1992). Prediction of the onset of signicant void in ow boiling
of water. ASME, Fundamentals of Subcooled Flow Boiling, HTD-Vol. 217, 4152.
137. Al-Hayes, R. A. M., and Winterton, R. H. S. (1981). Bubble diameter on detachment in
owing liquids. Int. J. Heat Mass Transfer 24, 223230.
138. Martinelli, R. C. (1947). Heat transfer to molten metals. Trans. ASME 69, 947951.
139. Blasick, A. M., Dowling, M. F., Abdel-Khalik, S. I. Ghiaasiaan, S. M., and Jeter, S. M.
(2000). Onset of ow instability in uniformly-heated thin horizontal annuli. Proc. 8th Int.
Conf. Nucl. Eng. (ICONE-8), April 26, Baltimore, MD.
140. Zijl, W., Ramakers, F. J. M., and Van Stralen, S. J. D. (1979). Global numerical solutions
of growth and departure of a vapor bubble at a horizontal superheated wall in a pure
liquid and a binary mixture. Int. J. Heat Mass Transfer 22, 401420.
141. Lee, R. C., and Nydahl, J. E. (1989). Numerical calculation of bubble growth in nucleate
boiling from inception through departure. J Heat Transfer 111, 474479.
142. Kocamustafaogullari, G., and Ishii, M. (1983). Interfacial area and nucleation site density
in boiling systems. Int. J. Heat Mass Transfer 26, 13771387.
143. Yang, S. R., and Kim, P. H. (1988). A mathematical model of the pool boiling nucleation
site density in terms of the surface characteristics, Int. J. Heat Mass Transfer 31,
11271135.
144. Unal, H. C. (1976). Maximum bubble diameter, maximum bubble-growth time and
bubble-growth rate during the subcooled nucleate ow boiling of water up to 17.7
MN/m`. Int. J. Heat Mass Transfer 19, 643649.
145. Shin, T. S. and Jones, O. C. (1993). Nucleation and ashing in nozzles 1: A distributed
nucleation model. Int. J. Multiphase Flow 19, 943964.
146. Klausner, J. F., Mei, R. and Zeng, L. Z. (1997). Predicting stochastic features of vapor
bubble detachment in ow boiling. Int. J. Heat Mass Transfer 40, 35473552.
147. Lin, L., Udell, K. S. and Pisano, A. P. (1993). Vapor bubble formation on a micro heater
in conned and unconned micro channels. ASME, Heat Transfer on the Microscale,
HTD Vol. 253, 8593.
148. Lahey, R. T., Jr., and Drew, D. A. (1988). The three-dimensional time and volume
averaged conservation equations of two-phase ow. In Advances in Nuclear Science and
Technology (J. Lewis and M. Becker, eds.), Vol. 20. pp. 169, Plenum Press, New York.
149. Drew, D. A., and Lahey, R. T., Jr. (1987). The virtual mass and lift force on a sphere in
rotating and straining inviscid ow. Int. J. Multiphase Flow 13, 113121.
150. Wang, S. K., Lee, S. J. Jones, O. C., Jr., and Lahey, R. T., Jr. (1987). 3-D turbulence
structure and phase distribution measurements in bubbly two-phase ows. Int. J Multi-
phase Flow 13, 327343.
151. Shultze, H. D. (1984). Physico-chemical Elementary Processes in Flotation, pp. 123129.
Elsevier, Amsterdam.
152. Antal, S. P., Lahey, R. T., Jr., and Flaherty, J. E. (1991). Analysis of phase distribution in
fully developed laminar bubbly two-phase ow. Int. J. Multiphase Flow 17, 635652.
153. Peng, X. F., and Wang, B.-X. (1993). Forced convection and ow boiling heat transfer for
liquid owing through microchannels. Int. J. Heat Mass Transfer 36, 34213427.
154. Peng, X. F., Hu, H. Y., and Wang, B.-X. (1994). Boiling nucleation during liquid ow in
microchannels. Int. J. Heat Mass Transfer 41, 101106.
155. Hosaka, S., Hirata, M., and Kasagi, N. (1990). Forced convective subcooled boiling heat
transfer and CHF in small diameter tubes. Proc., Int. Heat Transfer Conf., 9th, Vol. 2, pp.
129134.
156. Boyd, R. D. (1985). Subcooled ow boiling critical heat ux (CHF) and its application to
fusion energy components. Part I. A review of fundamentals of CHF and related data
base. Fusion Technol. 7, 130.
157. Boyd, R. D. (1985). Subcooled ow boiling critical heat ux (CHF) and its application to
fusion energy components. Part II. A review of microconvective, experimental, and
correlational aspects. Fusion Technol. 7, 3152.
158. Groeneveld, D. C., and Snoek, C. W. (1986). A comprehensive examination of heat
transfer correlations suitable for reactor safety analysis. In Multiphase Science and
Technology (G. F. Hewitt, J. M. Delhaye, and N. Zuber, Eds.), Vol. 2, pp. 181274.
Hemisphere, WA, D.C.
159. Katto, Y. (1994). Critical heat ux. Int. J. Multiphase Flow 20, suppl., 5390.
160. Ornatskiy, A. P. (1960). The inuence of length and tube diameter on critical heat ux for
water with forced convection and subcooling. Teploenergetika 4, 6769.
161. Ornatskiy, A. P., and Kichigan, A. M. (1962). Critical thermal loads during the boiling of
subcooled water in small diameter tubes. Teploenergetika 6, 7579.
162. Ornatskiy, A. P., and Vinyarskiy, L. S. (1964). Heat transfer crisis in a forced ow of under
heated water in small bore tubes. Teplozika Vysokikh Temperatur 3, 444451.
163. Loomsmore, C. S., and Skinner, B. C. (1965). Subcooled critical heat ux for water in
round tube. S. M. Thesis, MIT, Cambridge, MA.
164. Daleas, R. S., and Bergles, A. E. (1965). Effect of upstream compressibility on subcooled
critical heat ux. Proc. ASME/AIChE Conf. on Heat Transfer, 65-HT-67.
165. Subbotin, V. I., Deev, V. I., and Arkhipov, V. V. (1982). Critical heat ux in ow boiling
of helium. Proc. Int. Heat Transfer Conf., 7th, Vol. 4, pp. 357361.
166. Katto, Y., and Yokoya, S. (1984). Critical heat ux of liquid helium (I) in forced
convective boiling. Int. J. Multiphase Flow 10, 401413.
167. Boyd, R. D. (1988). Subcooled water ow boiling experiments under uniform high heat
ux conditions. Fusion Technol. 13, 131142.
168. Boyd, R. D. (1990). Subcooled water ow boiling transition and the L/D Effect on CHF
for a horizontal uniformly heated tube. Fusion Technol. 18, 317324.
169. Nariai, H., Inasaka, F., and Shimuara, T. (1987). Critical heat ux of subcooled ow
boiling in narrow tube. Proc. ASME/JSME T hermal Energy Joint Conf., 1987, Vol. 5, pp.
455462.
170. Nariai, H., Inasaka, F., and Uehara, K. (1989). Critical heat ux in narrow tubes with
uniform heating. Heating Transfer: Japanese Research, 18, 2130.
171. Inasaka, F., and Nariai, H. (1993). Critical heat ux of subcooled ow boiling with water
for high heat ux application. SPE, High Heat Flux Eng. II, Vol. 1997, pp. 328339.
172. Celata, G. P., Cumo, M., and Mariani, A. (1993). Burnout in highly subcooled water ow
boiling in small diameter tubes. Int. J. Heat Mass Transfer 36, 12691285.
173. Vandervort, C. L., Bergles, A. E., and Jensen, M. K. (1994). An experimental study of
critical heat ux in very high heat ux subcooled boiling. Int. J. Heat Mass Transfer 37
(suppl. 1), 161173.
174. Roach, G. M., Jr., Abdel-Khalik, S. I., Ghiaasiaan, S. M., and Jeter, S. M. (1999). Low-ow
critical heat ux in heated microchannels. Nucl. Sci. Eng. 131, 411425.
175. Celata, G. P., Cumo, M., and Mariani, A. (1994). Assessment of correlations and models
for the prediction of CHF in water subcooled ow boiling. Int. J. Heat Mass Transfer 37,
237255.
176. McBeth, R. V., and Thompson, B. (1964). Boiling water heat transfer burnout in uniformly
heated round tubes: A compilation of world data with accurate correlations. UKAEA
Report AEEW-R356, Winfrith, England.
177. McBeth, R. V. (196566). An appraisal of forced convection burnout data. Proc. Inst.
Mech. Eng., 180, 4748.
178. Bowring, R. W. (1972). A simple but accurate round tube, uniform heat ux, dryout
correlation over the pressure range 0.717 MN/m` (1002500 psia). UKAEA Report
AEEW-R789, Winfrith, England.
179. Bergles, A. E. (1962). Subcooled burnout in tubes of small diameter. ASME Paper
63-WA-182.
180. Celata, G. P., Cumo, M., Mariani, A., Nariai, H., and Inasaka, F. (1993). Inuence of
channel diameter on subcooled ow boiling burnout at high heat uxes. Int. J. Heat Mass
Transfer 36, 34073409.
181. Celata, G. P. (1993). Recent achievements in the thermal-hydraulics of high heat ux
components in fusion reactors. Exp. Therm. Fluid Sci. 7, 263278.
182. Tong, L. S. (1969). Boundary layer analysis of the ow boiling crisis. Int. J. Heat Mass
Transfer 11, 12081211.
183. Hall, D. D., and Mudawar, I. (1997). Evaluation of subcooled CHF correlations using the
PU-BTPFL CHF database for vertical upow of water in a uniformly heated round tube.
Nucl. Technol. 117, 234246.
184. Caira, M., Caruso, G., and Naviglio, A. (1995). A correlation to predict CHF in subcooled
ow boiling. Int. Comm. Heat Mass Transfer 22, 3545.
185. Shah, M. M. (1987). Improved general correlation for critical heat ux during upow in
uniformly heated vertical tubes. Int. J. Heat Fluid Flow 8, 326335.
186. Weisman, J. (1992). The current status of theoretically based approaches to the prediction
of the critical heat ux in ow boiling. Nucl. Technol 99, 1121.
187. Weisman, J., and Pei, B. S. (1983). Prediction of critical heat ux in ow boiling at low
qualities. Int. J. Heat Mass Transfer 26, 14631477.
188. Weisman, J., and Ileslamlou, S. (1988). A phenomenological model for prediction of
critical heat ux under highly subcoole conditions. Fusion Technol. 13, 654659.
189. Lee, C. H., and Mudawar, I. (1988). A phenomenological model for prediction of critical
heat ux under highly subcooled conditions. Fusion Technol. 13, 654659.
190. Galloway, J. E., and Mudawar, I. (1993). CHF mechanism in ow boiling from a short
heated wall I. Examination of near-wall conditions with the aid of photomicrography
and high-speed video imaging. Int. J. Heat Mass Transfer 36, 25112526.
191. Katto, Y. (1990). Prediction of critical heat ux of subcooled ow boiling in round tubes.
192. Katto, Y. (1992). A prediction model of subcooled ow boiling CHF for pressure in the
range 0.120.9 MPa. Int. J. Heat Mass Transfer 35, 11151123.
193. Celata, G. P., Cumo, M., Mariani, A., Simoncini, M. and Zummo, G. (1994). Rationaliz-
ation of existing mechanistic models for the prediction of water subcooled ow boiling
critical heat ux. Int. J. Heat Mass Transfer 37, Suppl. 1, 347360.
194. Ahmad, S. Y. (1970). Axial distribution of bulk temperature and void fraction in a heated
channel with inlet subcooling. Int. J. Heat Mass Transfer 92, 595609.
195. Haramura, Y., and Katto, Y. (1983). A new hydrodynamic model of critical heat ux,
applicable widely to both pool and forced convection boiling on submerged bodies in
saturated liquids. Int. J. Heat Mass Transfer 26, 389399.
196. Hewitt, G. F., and Govan, A. H. (1990). Phenomena and prediction in annular two-phase
ow. ASME Advances in GasL iquid Flows, FED Vol. 99, pp. 4156. ASME, New York.
197. Sugawara, S. (1990). Droplet deposition and entrainment modeling based on the three-
uid model. Nucl. Eng. Design 122, 6784.
198. Sugawara, S. (1990). Analytical prediction of CHF by FIDAS code based on three-uid
and lm dryout model. J. Nucl. Sci. Technol. 27, 1229.
199. Abdollahian, D., Healzer, J. Janssen, E., and Amos, C. (1982). Critical ow data review
and analysis. Electric Power Research Institute Report EPRI NP-2192, Palo Alto, CA.
200. Elias, E., and Lelluche, G. S. (1994). Two-phase critical ow. Int. J. Multiphase Flow 20,
Suppl. 91168.
201. Schrock, V. E., Revankar, S. T., and Lee, S. Y. (1988). Critical ow through pipe cracks.
In Particulate Phenomena and Multiphase Transport (N. T. Veziraglu, ed.). Hemisphere,
New York.
202. Collier, R. P., and Norris, D. M. (1983). Two-phase ow experiments through intergranu-
lar stress corrosion cracks. Proc. CSNI Specialist Meeting on L eak-Before Break in
Nuclear Reactor Piping, U.S. Nucl. Reg. Comm. Report NUREG/CP-d005, pp. 273299.
203. Collier, R. P. Stuben, F. B., Mayeld, M. E., Pope, D. B., and Scott, P. M. (1984).
Two-phase ow through intergranular stress corrosion cracks. Electric Power Research
Institute Report EPRI-NP-3540-LD, Palo Alto, CA.
204. Amos, C. N., and Schrock, V. E. (1984). Two-phase critical ow in slits. Nucl. Sci. Eng.
88, 261274.
205. Kefer, V., Kastner, W., and Kra tzer, W. (1986). Leckraten bei unterkritischen Rohrleitung
srissen. Jahrestagung, Kerntechnik, Aachen, Germany.
206. Nabarayashi, T., Ishiyama, T., Fujii, M., Matsumoto, K., Harimizu, Y., and Tanaka, Y.
(1989). Study on coolant leak rates through pipe cracks: Part I Fundamental tests.
Proc. ASME Pressure Vessels and Piping Conf., JSME Co-sponsorship, ASME PVP Vol.
165, pp. 121127. ASME, New York.
207. Matsumoto, K., Nakamura, S., Gotoh, N., Nabarayashi, T., Tanaka, Y. and Horimizu, Y.
(1989). Study on coolant leak rates through pipe cracks: Part 2 Pipe test. Proc. ASME
Pressure Vessels and Piping Conf., JSME Co-sponsorship, ASME PVP Vol. 165, pp.
113120. ASME, New York.
208. Ghiaasiaan, S. M., Muller, J. R., Sadowski, D. L., and Abdel-Khalik, S. I. (1997). Critical ow
of initially highly subcooled water through a short capillary. Nucl. Sci. Eng. 126, 229238.
209. Richter, H. J. (1983). Separated two-phase ow model: Application to critical two-phase
ow. Int. J. Multiphase Flow 9, 511530.
210. Alamgir, M. D., and Lienhard, J. H. (1981). Correlation of pressure undershoot during
hot-water depressurization. J. Heat Transfer 103, 5255.
211. Giot, M., and Fritz, A. (1972). Two-phase two- and one-component critical ow with the
variable slip model. Prog. Heat Transfer 6, 651670.
212. Ardron, K. H. (1978). A two-uid model for critical vaporliquid ow. Int. J. Multiphase
Flow 4, 323327.
213. Boure , J. A. (1997). The critical ow phenomena with reference to two-phase ow and
nuclear reactor systems. Proc. ASME Symp. T hermal-Hydraulic Aspects of Nuclear
Reactor Safety, pp. 195216. ASME, New York.
214. Lee, S. Y., and Schrock, V. E. (1988). Homogeneous non-equilibrium critical ow model
for liquid stagnation states. Proc. National Heat Transfer Conf., 7th, HTD Vol. 96, pp.
215. Schrock, V. E., Revankar, S. T., and Lee, S. Y. (1988). Critical ow through pipe cracks.
In Particulate Phenomena and Multiphase Transport (N. T. Veziroglu, ed.), Vol. 1, pp.
317. Hemisphere, Washington, D.C.
216. Feburie, V., Giot, M. Granger, S., and Seynhaever, J. M. (1993). A model for choked ow
through cracks with inlet subcooling. Int. J. Multiphase Flow 19, 541562.
217. Ghiaasiaan, S. M., and Geng, H. (1997). Mechanistic non-equilibrium modeling of critical
ashing ow of subcooled liquids containing dissolved noncondensables. Num. Heat
Transfer: B 32, 435457.
218. Geng, H., and Ghiaasiaan, S. M. (1998). Mechanistic modeling of critical ow of initially
subcooled liquid containing dissolved noncondensables through cracks and slits based on
the homogeneous-equilibrium mixture method. Nucl. Sci. Eng. 129, 294304.
219. McFadden, J. H., Paulsen, M. P., Gose, G. C., Peterson, McClure, J. A., Jensen, P. J., and
Westacott, J. L. (1992). RETRAN-03. A program for transient thermal-hydraulic analysis
of complex uid systems. Electric Power Research Institute Report EPRI NP-7450, Vol.
1, Palo Alto, CA.
220. Leung, J. C., and Grolmes, M. A. (1988). A generalized correlation for ashing choked
ow of initially subcooled liquid. AIChE J. 34, 688691.
221. Idelchik, I. E. (1994). Handbook of Hydraulic Resistances, 3rd ed., CRC Press, London.
222. Moody, F. J. (1966). Maximum two-phase vessel blowdown from pipes. J. Heat Transfer
88, 285295.
223. Abdolahian, D., Chexal, B., and Norris, D. M. (1983). Prediction of leak rates through
intergranular stress corrosion cracks. Proc. CSNI L eak-Before-Break Conf., Nucl. Reg.
Comm. Report NUREG/CP-00151, pp. 300326.
224. Henry, R. E. (1970). The two-phase critical discharge of initially saturated or subcooled
liquid. Nucl. Sci. Eng. 41, 336.
225. Chexal, B., Abdollahian, D., and Norris, D. (1984). Analytical prediction of single-phase
and two-phase ow through cracks in pipes and tubes. AIChE Symp. Ser. 80(236), pp.
1923.
226. Schwellnus, C. F., and Shoukri, M. (1991). A two-uid model for non-equilibrium
two-phase critical discharge. Can. J. Chem. Eng. 69, 187197.
227. Dagan, R., Elias, E., Wacholder, E., and Olek, S. (1993). A two-uid model for critical
ashing ows in pipes. Int. J. Multiphase Flow 19, 1525.
228. Blinkov, V. N., Jones, O. C., Jr., and Nigmatulin, B. I. (1993). Nucleation and ashing in
nozzles 2. Comparison with experiments using a ve-equation model for vapor void
development. Int. J. Multiphase Flow 19, 965986.
229. Downar-Zapolski, Z., Bilicki, Z., Bolle, L., and Franco, J. (1996). The non-equilibrium
relaxation model for one-dimensional ashing liquid ow. Int. J. Multiphase Flow 22,
473483.
230. Kroeger, P. G. (1978). Application of a non-equilibrium drift-ux model to two-phase
blowdown experiments. Paper presented at OECD/NEA Specialists Meeting on Transi-
ent Two-Phase Flow, Toronto, Canada, August 1998.
231. Lackme, C. (1979). Incompleteness of the ashing of supersaturated liquid and sonic
ejection of the produced phases. Intl. J. Multiphase Flow 5, 131141.
232. Hardy, Ph., and Mali, P. (1983). Validation and development of a model describing
subcooled critical ow through long tubes. Energie Primaire 18, 523.
233. Garrels, R. M., and Christ, C. L. (1965). Solutions, Minerals and Equilibria. Harper and
Row, New York.
Turbulent Flow and Convection:
The Prediction of Turbulent Flow and Convection
in a Round Tube
STUART W. CHURCHILL
Department of Chemical Engineering
The University of Pennsylvania
Philadelphia, Pennsylvania 19104
The quantitative prediction of turbulent ow and convection in chan-
nels extends at least from Boussinesq, who in 1877 proposed the eddy-
viscosity model, to Papavassiliou and Hanratty, who in 1997 computed the
rate of transport of molecular species by the turbulent uctuations using
Lagrangian direct numerical simulation. The history and current state of the
art of such predictions are examined herein. It is concluded that the eddy-
diffusivity, mixing-length, and c models should now be completely aban-
doned in favor of generalized correlating equations for the turbulent shear
stress and the turbulent heat ux density based on semitheoretical asymp-
totic expressions, even though those for the latter quantity are yet uncertain
and incomplete. Correlating equations for the time-averaged velocity dis-
tribution, the friction factor, the time-averaged temperature distribution,
and the heat transfer coefcient may serve as conveniences, but such
expressions are unessential since these four quantities may be determined
numerically with comparable accuracy from simple, single integrals of the
two more elementary quantities. Such direct numerical evaluations reveal
that all of the classical algebraic analogies between heat and momentum
transfer are in signicant error functionally as well as numerically, in large
part because of inaccurate representations of the radial variation of the total
heat ux density. In the interests of simplicity and clarity, this presentation
is primarily limited to fully developed heat transfer in a uniformly heated
round tube, but the methodologies and formulations may readily be adapted
255
0065-2717/01 $35.00
or extended for other one-dimensional ows and other thermal boundary
conditions.
I. Introduction
Progress in engineering and science occurs by discarding old concepts and
correlations in favor of new or improved ones. Turbulent ow and convec-
tion have been somewhat resistant to this process of renewal; many early
concepts and correlations remain enshrined in our current textbooks and
handbooks and in the software for design calculations even though they
long ago became obsolete in terms of accuracy or were shown to be
untenable in a theoretical sense. We should of course recognize and honor
the pioneers of our eld and preserve their contributions in a historical
context, but at the same time be alert and aggressive about identifying and
incorporating experimental and theoretical improvements.
Turbulent forced convection is analogous to turbulent ow in many
respects, but is far more complex and demanding analytically, experiment-
ally, and in practice. Under many circumstances, ow may be studied
independently from heat transfer. On the other hand, the detailed analysis
of forced convection requires a quantitative description of the details of the
ow. Also, turbulent convection invokes the Prandtl number as a parameter
as well as several other complexities that are not encountered in the
description of the ow. For these reasons, ow is examined rst and
convection thereafter. That division and order is observed in this Introduc-
tion as well as in the presentation as a whole.
In the interests of simplicity and clarity, the detailed descriptions that
follow are limited in the main to fully developed turbulent ow inside a
straight smooth round tube and to fully developed forced convection from
a uniformly or isothermally heated wall. However, the adaptation or
extension of these methodologies and formulations for other one-dimen-
sional ows is examined briey and shown to be rather straightforward.
Attention is also limited to single-phase Newtonian uids with invariant
physical properties, including the specic density. The latter restriction
excludes natural convection. Physical property variations are of course often
signicant in magnitude and may greatly inuence the ow and convection.
However, since the variations of the viscosity and thermal conductivity are
different for every uid and that of the density for every liquid, accounting
for these effects would eliminate almost all of the generality that is a primary
feature of the new and old developments described herein. At the present
time the best compensation for this oversimplication is to incorporate
empirical corrections in the nal idealized results for invariant physical
256 s1i:1 v. cnicniii
Fic. 1. Self-portrait of Leonardo da Vinci observing turbulent vortices behind a disturbance
in a river. (from Richter [2], Plate XXV).
properties. As computational tools and techniques advance, it may become
feasible to incorporate such variations in direct numerical simulations of the
time-dependent differential equations of conservation or numerical integra-
tions of their time-averaged counterparts.
For convenience, a single standardized notation is utilized throughout
rather than necessarily those that were originally employed in particular
contributions. All symbols are dened when they appear rst or in a new
context.
A. TiniirN1 Fiov
Lamb [1], in his classical treatise Hydrodynamics, rst published in 1879
but periodically revised by himself through 1932, a span of over half a
century, begins his rather brief treatment of turbulent motion in even the
latest of those editions with the statement, It remains to call attention to
the chief outstanding difculty of our subject. He then proceeds to explain
the reasons for that difculty, noting that the motion becomes wildly
irregular and the tube appears to be lled with interlacing and constantly
varying streams, crossing and recrossing the pipe.
257 1iniirN1 riov :Nn coNvrc1ioN
Leonardo da Vinci in 1515, at age 63, sketched in his notebook, as shown
in Fig. 1 (Plate XXV from Richter [2]), a self-portrait with a river owing
past obstructions. As indicated by the accompanying text, he was interested
in the pattern of ow in a scientic as well as an artistic sense. His
description has been translated (Richter [2], p. 200) as Observe the motion
of the surface of the water which resembles that of hair, and has two
motions, of which one goes on with the ow of the surface, the other forms
the lines of the eddies; thus the water forms eddying whirlpools one part of
which are due to the impetus of the principal current and the other to the
incidental motion and return ow. Even such a universal genius and
perceptive observer as Leonardo may sometimes err; he sketched symmet-
rical pairs of vortices rather than the alternating ones that actually occur.
The later renowned physicists and applied mathematicians who have
written on the subject of turbulent ow include Subrahmanyan Chandrasek-
har [3], Albert Einstein [4], Werner Heisenberg [5], Pyotr Kapitsa [6], Lev
Landau [7], Hendrik Lorentz [8], Isaac Newton [9], Lord Rayleigh [10],
Arnold Sommerfeld [11], George Uhlenbeck [12], Richard von Mises [13],
C. R. von Weizsa cker [14], and Yakob Zeldovich [15]. It is intimidating
and humbling for anyone who undertakes the study of turbulent ow to
realize that even these great scientists made few signicant contributions to
this subject outside the special topic of stability.
The focus herein on shear ows and in particular on fully developed ow
in a round tube avoids the necessity of reviewing the statistical develop-
ments that have found applicability primarily in the idealized domains of
isotropic and homogeneous tubulence. Another quotation is appropriate in
this regard. Schlichting [16], in the Authors Preface of the rst German
edition of Boundary L ayer T heory, wrote (in translation) No account of the
statistical theories of turbulence has been included because they have not
attained any practical signicance for engineers. In a later edition he writes
begrudgingly and deantly in apparent response to criticisms of that
statement that This (the statistical theory of turbulence) admittedly has
contributed to our understanding of turbulent ows but it has not yet
acquired any importance to engineers. Although statistical representations
of turbulence have recently been utilized in the development of approximate
expressions for the kinetic energy of turbulence, , and the rate of dissipation
of the energy of turbulence, c, in connection with the c and related
models, the details of that usage are outside the chain of development herein
and only merit this brief mention.
The general partial-differential equations for the conservation of mass and
momentum in time-dependent form are generally presumed to describe
turbulent ow insofar as the uid may be treated as a continuum. Because
of their complexity, and in particular their nonlinearity, they resisted
numerical as well as analytical methods of solution until roughly the last
decade. Accordingly, as an alternative approach, Sir Osborne Reynolds [17]
in 1895 reduced these expressions to manageable proportions and a tract-
able form by space-averaging. This great simplication has inspired a
century-long development of semitheoretical models and approximate sol-
utions for the reduced equations. In particular, Ludwig Prandtl and his
associates and contemporaries developed a very useful structure for the
prediction and correlation of turbulent shear ows by postulating mechan-
istic models for the unknown term(s) in the time-averaged (rather than
space-averaged) equations of conservation in differential form. This pro-
cedure proved to be so successful in an applied sense that only sporadic and
limited improvements were made over the ensuing half-century. Finally, in
the 1980s and 1990s, a signicant breakout occurred in the form of
essentially exact solutions of the general time-dependent equations of
conservation by direct numerical simulation (DNS). Although the results
obtained by this technique are yet very restricted in scope, both intrinsically
and because of their computational demands, this development has invigor-
ated the elds of turbulent ow and convection and has played a key role
in the development of the new formulations that constitute the principal
contribution of this article.
Before the presentation of these improved formulations, a retrospective
assessment of the historical development and validity of the quantitative
descriptions of turbulent shear ow in the classical and current literature is
provided. This survey is limited to representative contributions with long-
lasting consequences, both positive and negative, rather than being exhaus-
tive, since the primary objective is to provide a framework and perspective
for the new improved expressions. This portion of the presentation is
organized chronologically on the mean, but also topically, in part in the
interests of clarity, continuity, and the avoidance of repetition.
B. TiniirN1 CoNvrc1ioN
Although the path of development of a structure for the correlation and
prediction of convection in turbulent ow might have been expected to
follow the path of development for the ow itself because of the similar
structure of the equation of conservation for energy to those for momentum,
this is not found to be the case. Turbulent convection is much more
complicated because of (1) the coupling of the equations for the conserva-
tions of energy and momentum, (2) the possibility of different boundary
conditions, and (3) the appearance of additional parameters.
Reynolds [18] in 1874, and thus 21 years prior to his great contribution
to the development of a simplied structure for turbulent ow by means of
space averaging, derived a simple algebraic equation, now known as the
Reynolds analogy, by postulating equal mass rates of transport of energy and
momentum from the bulk of the uid stream all the way to a conning
surface by the turbulent eddies. This expression, which is free of any explicit
empiricism and independent of geometry, is only of rst-order accuracy at
best, but it lives on after 125 years by virtue of its implicit or explicit
presence within many predictive correlations for heat and mass transfer.
An alternative approach to the prediction of the rate of heat transfer
involves the adaptation of the mechanistic models of Prandtl and others to
represent the primary unknown term (the turbulent heat ux density) in the
time-averaged differential equation for the conservation of energy. Such
modeling has, however, evolved more slowly and less successfully for
convection than for ow because of the inherently more complex behavior
mentioned earlier. Direct numerical simulation has also been utilized for
convection, but even more severe limitations are encountered than for ow
and, as a consequence, less success has been achieved.
In contrast with ow, the historical development of a structure for
turbulent convection is reviewed retrospectively in the light of the new and
improved formulations that prompted this article. Finally, some representa-
tive numerical results obtained by means of these new formulations are
presented and generalized.
II. The Quantitative Representation of Turbulent Flow
Structures for the prediction and correlation of the characteristics of
turbulent ow that are important in the subsequent prediction and correla-
tion of rates of heat transfer are rst examined from a historical point of
view. New and improved formulations are then described and compared
with experimental data and numerical predictions. The derivations and nal
expressions are presented herein primarily in the context of a round tube,
even though they may have been formulated originally for ow between
parallel plates or for unconned ow along a at plate. The adaptation or
extension of these new expressions for other geometries is nally examined
briey.
A. His1oic:i Hicniicn1s
1. The Exact Structure
The partial differential equations for the conservation of momentum in
the ow of a single-phase Newtonian uid with invariant viscosity and
density may be expressed in cylindrical coordinates as follows:
Radial component:
j
cu
'
ct
u
'
cu
'
cr

u
r
cu
'
c0
u`
r
u
cu
'
cz
cp
cr
j
_
c
cr
1
r
c
cr
(ru
'
)
1
r`
cù
'
c0
`

2
r`
cu
c0
cù
'
cz`
jg
'
(1)
Angular component:
j
cu
ct
u
'
cu
cr

u
r
cu
c0
u
'
u
r
u
cu
cz
1
r
cp
c0
j
_
c
cr
1
r
c
cr
(ru
1
r`
cù
c0
`

2
r`
cu
'
c0
cù
cz`
jg
(2)
Axial component:
j
cu
ct
u
'
cu
cr

u
r
cu
c0
cu
cz
cp
cz
j
_
1
r
c
cr
r
cu
cr
1
r`
cù
c0
`

cù
cz`
jg
. (3)
Here, t represents time, r, z and 0
the radial, axial and angular coordinates,

g
'
, g
, and g
the corresponding components of the gravitational vector, u

'
,
u
, and u
those of the instantaneous velocity vector, p the instantaneous

thermodynamic pressure, j the specic density, and j the dynamic viscosity.
Equations (1)(3) are generally known as the NavierStokes equations in
recognition of their derivation by Navier [19] in 1822 and their renement
by Stokes [20] in 1845. After repeated attempts to derive or justify these
expressions on the basis of statistical mechanics, Uhlenbeck [12] noted that
Quantitatively, some of the predictions from these equations surely deviate
from experiment, but the very remarkable fact remains that qualitatively the
NavierStokes equations always describe physical phenomena sensibly. . . .
The mathematical reason for this virtue of the NavierStokes equations is
completely mysterious to me.
The greatest advance ever in the analysis of turbulent ow was made in
1895 by Reynolds [17], who not only conceived of the practical advantages
of space averaging but also derived in detail the required mathematical
procedures for this averaging. He then applied this process to Eqs. (1)(3).
Time averaging, which is generally presumed to be equivalent to space
averaging, followed by specialization for steady fully developed ow, reduces
these equations to
cP
cr

1
r
d
dr
(jru'
'
u'
'
)
j(u'
u'
)
r
0 (4)
1
r
d
dr
(ru'
'
u'
)
u'
'
u'
r
0 (5)
and
cP
cz

1
r
d
dr
jr
du
dr
(jru'
'
u'
0, (6)
where here, as contrasted with Eqs. (1)(3), u
denotes the time-averaged

velocity, and P the time-averaged dynamic pressure that arises from changes
in velocity only; u'
'
, u'
and u'
the instantaneous uctuations in velocity

about the time-averaged values; and the superbars the time-averaged values
of products of these uctuations. The validity of the equations obtained by
space or time averaging has been questioned, but no specic failures have
been demonstrated for conditions such that treatment of the uid as a
continuum is a valid approximation. Barenblatt and Goldenfeld [21] have
questioned the concept of full development for turbulent shear ows (the
attainment of a velocity eld and pressure gradient independent from z), but
such behavior is certainly attained for all practical purposes (see, for
example, Abbrecht and Churchill [22]), and such a postulate has resulted
in many apparently valid asymptotic predictions.
Equations (4)(6), even if exact, are, in contrast with Eqs. (1)(3), an
incomplete description of the uid motion. The terms u'
'
u'
'
, u'
u'
, u'
'
u'
, and
u'
'
u'
, which are known quite appropriately as the Reynolds stresses, repre-

sent the information lost by time averaging since they are indeterminate
from these equations alone. Most of the modeling of turbulent ow has
involved the postulate of empirical expressions for these time-averaged
products of the uctuating components of the velocity. On the other hand,
the structural gain from time averaging is quite evident, not only from the
relative simplicity of Eqs. (4)(6) as compared to Eqs. (1)(3), but also from
their susceptibility to analytical and formal integration with respect to r to
obtain
PP
"
ju'
'
u'
'
j
,
'
(u'
u'
u'
'
u'
'
)
dr
r
(7)
u'
'
u'
0 (8)
and
r
2
cP
cz
j
du
dr
ju'
'
u'
. (9)
Equation (7) expresses the radial variation in pressure wholly in terms of
the uctuations in u
and u
'
; Eq. (8) indicates that the Coriolis force is zero
at all radii. Here P
"
is the pressure on the wall of the pipe at r a, where
a is the radius of the pipe. Since u'
'
u'
'
and u'
u'
are not functions of z in fully

developed ow, it follows that cP/cz is also independent of z and hence a
constant that may be expressed as dP/dz. From a force balance over a
central cylindrical segment of the uid it may be shown that
t
r
2
dP
dz
, (10)
where t is the total shear stress in the z-direction imposed on the outer uid
at any radius by this inner segment. It follows that at r a,
t
"
a
2
dP
dz
. (11)
Here t
"
is the shear stress imposed on the wall in the z-direction. From the
ratio of Eqs. (10) and (11),
t
t
"
r
a
. (12)
Now for convenience and simplicity, letting u
u, u'
'
v', u'
u', and
a r y allows Eq. (9) to be reexpressed as
t
"

1
y
a
j
du
dy
ju'v' . (13)
From Eq. (13), which is the starting point for all subsequent modeling herein
for ow in a round tube, it is evident that the contribution of the turbulent
uctuations to the time-averaged velocity is wholly represented by u'v'.
Similarly, it is evident from Eq. (7) that the radial variation of the dynamic
pressure is wholly a consequence of u'
'
u'
'
and u'
u'
.
2. Dimensional, Asymptotic, and Speculative Analyses
The most useful technique for the development of functional relationships
for the characteristics of turbulent ow has proven to be a combination of
dimensional, asymptotic, and speculative analyses. Here speculation refers to
a tentative postulate whose consequences are ultimately to be tested with
experimental or exact theoretical results. It is unfortunate that the uncer-
tainty implied by this terminology has often discouraged its formal usage.
The techniques and results of dimensional, asymptotic, and speculative
analysis have evolved independently in many different contexts and there is
no general agreement on the priority of the various contributions. The
attributions herein are conceded to be somewhat arbitrary.
Fourier [23] in 1822 established the fundamental basis for dimensional
analysis by noting that all added and equated terms in a complete relation-
ship between the variables must have the same net dimensions. Rayleigh [24,
25] in 1892 illustrated the expression of functional relationships in terms of
dimensionless groups, and in 1915 proposed a general mechanistic process
for the determination of an appropriate minimal set of dimensionless
groupings to describe the behavior dened by a listing of dependent
variables, independent variables, physical properties, and any other relevant
parameters.
Asymptotic dimensional analysis, as used herein, refers to the reduction of
such a listing, and hence of the number of dimensionless groupings, for
limiting conditions or locations or times. Speculative dimensional analysis, as
dened by Churchill [26], refers to the tentative elimination of individual
variables or parameters and thereby reduction of the number of dimension-
less groupings without necessarily any justication or rationale in advance.
This latter procedure may be characterized by the question, What if . . .?
Insofar as the chosen set of variables and parameters is sufcient and
self-consistent, the results obtained by ordinary and speculative dimensional
analyses are exact. The results from an asymptotic dimensional analysis may
additionally depend on arbitrary constraints. Since the original choice of a
set of variables, whether from a mathematical model or a heuristic listing, is
always a possible source of error, Churchill [27] has proposed that all
processes of dimensional analysis be considered to be speculative and
thereby tentative until conrmed by experimental data or exact theoretical
results.
Since neither analytical nor numerical solutions of the general time-
dependent equations of conservation for conditions resulting in turbulent
shear ow have been accomplished until very recently, these several pro-
cesses of dimensional analysis have proven to be invaluable in terms of
suggesting forms for the efcient correlation of experimental data. Such
forms may be expected to serve the same role for the currently emerging
exact but discrete numerical results.
The rst application of dimensional analysis for turbulent ow was by
Reynolds [28, 17], who in 1883 determined the conditions for the onset of
turbulence in ow through a long pipe and then in 1895 surmised that the
transition was characterized by the dimensionless grouping Du
"
j/j, now
known as the Reynolds number. Here D2a is the diameter of the pipe and
u
"
the space- and time-mean velocity. This result and its direct counterparts
for the friction factor and the time-mean velocity distribution are all that
may be inferred from the variables of Eq. (13) by simple dimensional
analysis. On the other hand, Prandtl and his associates and contempories
utilized asymptotic and speculative dimensional analysis with great insight
and success to choose forms of these types for the correlation of experimen-
tal data starting from either Eqs. (1)(3) or (13). Some such analyses follow.
It may be speculated on purely physical grounds, inferred from Eqs.
(1)(3), or inferred from Eq. (13) that the local time-mean velocity in steady,
fully developed turbulent ow in a round tube with a radius a may be
expressed in general as
u y, t
"
, a, j, j}, (14)
where the notation x} designates an unknown function of any indepen-
dent variable x. [The inference of Eq. (14) from Eq. (13) implies that u'v' is
a function of the same variables as u.] It follows from the application of
ordinary dimensional analysis to Eq. (14) that one possible set of dimen-
sionless groupings is
u
j
t
"
y(t
"
j)`
j
,
y
a
. (15)
Prandtl [29] introduced the notation uu(j/t
"
)` and yy(jt
"
)`/j,
which is still used almost universally today, to reexpress Eq. (15) as
u
y,
y
a
'
y,
y
a
. (16)
Equation (13) may be reexpressed in this notation as
1
y
a
du
dy
(u'v'), (17)
where, as may be inferred, (u'v')ju'v'/t
"
.
Prandtl next speculated that near the wall u might be essentially indepen-
dent of a, thereby reducing Eq. (16) to
uy}, (18)
which is now known as the universal law of the wall. The limitation of Eq.
(18) to ya explains the terminology law of the wall, and the depend-
ence only on y suggests its possible applicability to all geometries, and
thereby the term universal.
Very, very near the wall, the contribution of the turbulent uctuations in
the velocity to the local shear stress might be expected to be negligibly small
relative to the viscous stress, permitting Eq. (17) to be integrated to obtain
uy
(y)`
2a
y. (19)
The limiting form of Eq. (19) may be noted to conform to Eq. (18), and both
the general and the limiting form to apply to laminar ow.
The corresponding speculation that near the centerline the viscous stress
might be negligible with respect to that due to the turbulent uctuations in
the velocity implies independence of du/dy from j. From the same process
of dimensional analysis for du/dy as carried out for u in Eqs. (14)(18), it
may be inferred that
du
d(y/a)

y
a
'
d(y/a})
d(y/a)
, (20)
where and designate arbitrary functions of y/a. Formal integration of
Eq. (20) from u u
'
, the velocity at the centerline at y a, leads to
u
'
u1}
y
a
y
a
. (21)
The term u
'
u, which characterizes the behavior near the centerline in
the same general sense that u does near the wall, is called the velocity defect
(or deciency), while Eq. (21) is called the law of the center.
Millikan [30], with great imagination and insight, speculated that Eqs.
(18) and (21) might have some region of overlap, far from the wall and far
from the centerline, where both were applicable, at least as an approxi-
mation. Accordingly, he reexpressed Eq. (18) in terms of the velocity defect,
that is, as
u
'
ua} y}, (22)
and noted that the only functional expression for the velocity defect
satisfying both Eqs. (21) and (22) is
u
'
uB ln
a
y
, (23)
where B is an arbitrary dimensionless coefcient. The necessary counterpart
for the velocity itself is
uAB lny}, (24)
where A is an arbitrary dimensionless constant. Although Eqs. (23) and (24)
conform to both the law of the center and the law of the wall, they would
be expected to have only a narrow identical region of validity far from both
the centerline and the wall. Von Ka rma n [31] postulated that despite this
restriction, Eq. (23) might provide an adequate approximation for the entire
cross-section insofar as integration to determine u
"
is concerned. The result
of such an integration is
u
'
u
"

3B
2
, (25)
which may be combined with Eq. (24), as specialized for ya, to obtain
2
f
`
u
"
A
3B
2
B lna}. (26)
Here f 2t
"
/ju`
"
is the Fanning friction factor. Equation (26) may of course
also be derived directly by integrating u from Eq. (24) over the cross-
section. The derivation of Eq. (26) by von Ka rma n is one of the most fateful
in the history of turbulent ow, in that it has remained to this day the most
common correlating equation for the friction factor, with its fundamental
shortcomings compensated for and disguised by differing and varying values
of A(3B/2) and B.
For a pipe with a roughness e, the same type of analysis that led to Eq.
(16) results in
u
y,
y
a
,
e
a
(27)
or the equivalent. The speculation that the velocity is independent of the
radius and dependent primarily on the roughness rather than on the
viscosity leads to the following modied law of the wall:
u
y
e
. (28)
Equation (21) remains applicable for the region near the centerline for
roughened as well as smooth pipe. The equivalent of the speculation of
Millikan results again in Eq. (23) for the velocity defect in the possible
region of overlap but the following different expression for the velocity
distribution itself in that region:
uCB ln
y
e
. (29)
Here, C is a dimensionless arbitrary constant and B is implied to have the
same value as in Eqs. (23) and (24). Integration of Eq. (29) over the
cross-section results in
2
f
`
u
"
C
3B
2
B ln
a
e
, (30)
but Eq. (25) remains applicable. It may be inferred that the effect of
roughness is simply to decrease u for a given value of y by the quantity
Blne} AC and to decrease u
"
by the same amount for a given value
of a. Here ee(t
"
j)`/j in conformity to the denition of y.
Murphree [32] and several others used a variety of methods of asym-
ptotic expansion to derive the following relationship for the time average of
the product of the uctuating components of the velocity and thereby the
turbulent shear stress very near a wall:
ju'v' :'y` ['y" . . . . (31)
Here, :',[', . . . are arbitrary dimensional coefcients. Equation (31) may be
reexpressed in terms of the previously dened dimensionless variables as
(u'v') :(y)` [(y)" . . . , (32)
where : and [ are dimensionless coefcients. Substitution of (u'v') from Eq.
(32) in Eq. (17), followed by integration from u0 at y0, leads to a
corresponding expression for u, namely,
uy
:
4
(y)"
[
5
( y)`
(y)`
2a
. (33)
Equation (33) without the term in (y)`/2a, which is negligible for typical
values of a, has also been derived directly by asymptotic expansion.
The recognition on physical grounds that the fraction of the shear stress
due to turbulence, namely ju'v'/t, is necessarily nite, positive, and less
than unity at the centerline requires, by virtue of Eq. (17), that
u
'
uE
1
y
a
`
, (34)
and therefore that
(u'v')
1
2E
a
1
y
a
, (35)
where E is an arbitrary dimensionless coefcient. Equations (34) and (35)
are the counterparts for the region near the centerline of Eqs. (33) and (32),
respectively, for the region near the wall.
The range of validity, if any, of each of the foregoing speculative
expressions, namely Eqs. (18)(35), is subsequently evaluated on the basis
of experimental data and direct numerical simulations.
3. Empirical Models
An alternative and supplementary approach to dimensional, speculative,
and asymptotic analyses is the postulate of mechanistic empirical models for
the turbulent shear stress and thereby for the prediction of the local
time-mean velocity distribution and its space mean.
a. The Eddy Viscosity Boussinesq [33] in 1877, and thus before the
identication by Reynolds in 1895 of the relationship between the turbulent
shear stress and the uctuating components of the velocity, proposed by
analogy to Newtons law for the viscous shear stress the following expression
for the total shear stress in a shear ow:
t (j j
'
)
du
dy
. (36)
Here j
'
is the eddy viscosity, an empirical quantity that is a function of local
conditions rather than a physical property such as j. This expression may
be recognized as equivalent to the following differential model for the
principal Reynolds stress:
ju'v' j
'
du
dy
. (37)
Equation (17), with the turbulent shear stress represented by Eq. (37), may
be rewritten as
1
y
a
1
j
'
j
du
dy
. (38)
b. The Mixing Length Another historically important model for the tur-
bulent shear stress was proposed by Prandtl [34] in 1925 on the basis of a
postulated analogy between the chaotic motion of the eddies and that of the
molecules of a gas. This model may be expressed as
ju'v' jl`
du
dy
du
dy
(39)
where l is a mixing length for eddies corresponding to the mean free path of
molecules as dened by the kinetic theory of gases. Although this analogy,
as noted by Bird et al. [35, p. 160], has little physical justication, the
mixing-length model has generally been accorded more respect by analysts
than the eddy viscosity model, apparently because of its mechanistic
rationale, however questionable that may be.
Von Ka rma n [31] speculated on dimensionless grounds that near the
wall, l might be proportional to the distance from the wall; that is, he
proposed the expression
l ky, (40)
where k is a dimensionless factor that is now generally called the von
Karman constant. Prandtl [29] substituted l from Eq. (40) in Eq. (39) and
in turn the resulting expression for ju'v' in Eq. (13) to obtain
t
"

1
y
a
j
du
dy
k`y`j
du
dy
`
, (41)
which may be reexpressed in the canonical dimensionless form as follows:
1
y
a
du
dy
k`(y)`
du
dy
`
. (42)
Prandtl [34], starting from Eq. (42), neglected the variation in the total
shear stress with y/a, neglected the viscous shear stress, took the square
root of the resulting expression, and integrated indenitely to derive
uA
1
k
lny}. (43)
Equation (43) is seen to be equivalent to Eq. (24) with B1/k. Because of
the two idealizations made in the reduction of Eq. (42), the resulting
expression would be expected to be invalid near the wall where the viscous
shear stress is controlling and near the centerline where the variation of the
total shear stress is important. Even within the remaining region, Eq. (43) is
subject to the two postulates represented by Eqs. (39) and (40). The
existence of a region of overlap, which was postulated by Millikan in
deriving Eq. (24), may be inferred to be equivalent to these two empirical
postulates of Prandtl. It is worthy of note that despite the postulate of
negligible viscous shear in its derivation, Eq. (43) incorporates, when
rewritten in terms of dimensional variables, a dependence on the viscosity
insofar as A is a constant independent of the Reynolds number and hence
of the viscosity. The subsequently demonstrated success of Eq. (43) and (24)
with empirical values for A and B1/k in representing experimental data
is a testament to the insight and ingenuity of both Prandtl and Millikan in
following two different and tortuous paths in their derivations.
Analytical solutions of Eq. (42) in closed form are actually possible if one
or the other of the simplications made by Prandtl in reducing Eq. (42) in
order to derive Eq. (43) is avoided. Furthermore, a solution in integral form
may be derived without making either simplication. For example, if the
viscous shear stress is taken into account, the resulting quadratic equation
in du/dy may be solved and then integrated from u0 at y0 to
obtain
u
1 [1 (2ky)`]`
2ky

1
k
ln2ky[1 ky)`]`}. (44)
Because of the imposition of the boundary condition at the wall, this
expression, which was apparently rst derived by Rotta [36], is free of the
arbitrary constant A of Eq. (43) and provides a smooth if erroneous
transition from the limiting form of Eq. (19) for y0 to Eq. (43), with an
effective value of A(1/k)(ln4k} 1) as y. The correct limiting
behavior for y0 and the smooth transition to Eq. (43) are a consequence
of accounting for the viscous shear stress. The failure of the predicted
transitional behavior to conform functionally to Eq. (33) is clearly attribu-
table to the shortcomings of Eqs. (39) and (40), but the reason for the
prediction of highly erroneous (negative) values for the equivalent of A at
large values of y for a representative value of k is more difcult to assign.
Conversely, accounting for the linear variation of the total shear stress
with y but neglecting the viscous shear stress permits derivation of the
following solution for the so-reduced form of Eq. (42) by a process similar
to that used to obtain Eqs. (43) and (44):
u
1
k _
2
1
y
a
`
2
1
y
"
a
`
ln
_
1
1
y
a
`
_
1
1
y
"
a
`
_
1
1
y
a
`
_
1
1
y
"
a
`
. (45)
Here u0 at yy
"
was invoked as an arbitrary boundary condition.
The choice of y
"
expAk} results in matching the predictions of Eqs.
(45) and (43) at that location. Equation (45) shares the limitation of
applicability of Eq. (43) to the turbulent core near the wall and is of interest
only as a measure of the effect of neglecting the variation of the total shear
stress in that regime. For larger values of y/a it is in serious error because
of its incorporation of Eq. (40).
Taking into account both the viscous shear stress and the variation of the
total shear stress, that is, starting from Eq. (42), solving this quadratic
equation in du/dy, and integrating formally from u0 at y0 results
in the following integral expression:
u2
,
'
"
1
y
a
dy
1
_
1
1
y
a
(2ky)`
`
. (46)
Equation (46) coincides with Eq. (44) for ya and represents an
improvement on Eq. (45) for larger values of y at the expense of numerical
integration, but fails for ya owing to the inapplicability of Eq. (40) for
that regime. Comparison of the predictions of Eqs. (44)(46) with Eq. (43)
for representative values of A, k, and y
"
conrms the good judgment of
Prandtl in making the simplications leading to Eq. (43) since Eq. (40),
which all of these improved expressions incorporate, is valid even as an
approximation only in the turbulent core near the wall.
Prandtl [34], and in more detail in [37], speculated that near the
centerline the mixing length might be nearly invariant, i.e.,
l 5l
, (47)
where l
is the limiting value for ya. He then substituted l
for l in
Eq. (39) and the resulting expression for ju'v' in Eq. (13), neglected the
viscous term, and integrated from u u
'
at y a to obtain, in dimensionless
form,
u
'
u
2
3
a
l

1
y
a
``
. (48)
Equation (48) correctly predicts du/dy 0 at ya, but has a different
power dependence on 1 (y/a) than does Eq. (34).
In order to improve upon Eqs. (40) and (45) and thereby on Eqs. (43) and
(48), von Ka rma n [31] postulated that
l k*
du/dy
dù/dy`
(49)
where k* is an arbitrary dimensionless coefcient similar to k. He once
explained, in response to an oral inquiry from the author of this article, that
Eq. (49) was chosen because it was the simplest dimensionally correct
expression for l involving only derivatives of the velocity. Substituting l
from Eq. (49) in Eq. (39) and following the same procedure as used to obtain
Eq. (48), but with two integrations and the equivocal boundary condition
du/dy at y 0, results in
j
'
j
1
k* _
1
y
a
`
ln
1
y
a
`
. (50)
If the variation of the total shear stress is neglected as well, the procedure
used to derive Eq. (50), except for an indenite limit for the second
integration, leads to Eq. (43) with k replaced by k*, which suggests but does
not prove their identity in general. As ya, Eq. (50) may be approxim-
ated by
u
'
u
1
2k*
1
y
a
, (51)
which not only has a different functional dependence on 1 (y/a) than
Eq. (34) but in addition fails to predict du/dy 0 at ya.
Van Driest [38] attempted to improve upon Eq. (40), the mixing length
model of Prandtl for the region near the wall, by including a term for
viscous damping similar to the one that holds for the laminar motion of a
uid subjected to the harmonic oscillation of a plate. That is, he let
l ky(1 expy}) (52)
where is an empirical dimensionless coefcient whose numerical value is
usually taken to be 1/26, in rough correspondence to the furthest limit of
the buffer layer from the wall. Introducing l from Eq. (52) in Eq. (39) and
in turn ju'v' in Eq. (13), neglecting the variation in the total shear stress,
solving the resulting quadratic equation for du/dy, and integrating
formally from u0 at y0 results in
u2
,
'
"
dy
1 (1 [2ky(1 expy}]`)`
. (53)
For y0, Eq. (53) reduces to Eq. (29), but unfortunately with : 0 and
[k``/5. For large values of y it reduces to Eq. (44), and thereby has
the merits and shortcomings already noted for that expression. It may be
inferred from Eq. (46) that the variation of the total shear stress, which van
Driest neglected, may be taken into account to obtain
u2
,
'
"
1
y
a
dy
1
_
1
1
y
a
(2ky)`(1 expy})`]`
. (54)
Just as noted with respect to Eq. (46), the result is only a slight improvement
on Eq. (53), since Eq. (52) is not applicable in the region where the terms in
1 (y/a) have a signicant role.
c. Other Models Kolmogorov [39] and Prandtl [40] independently con-
jectured on dimensional grounds (local similarity) that
j
'
c
`l*j, (55)
where here is the kinetic energy of the turbulence and l* is an unknown
length scale. Batchelor [41] subsequently conjectured that
l* c
`
``/c, (56)
where c is the rate of dissipation of turbulence. Combination of Eqs. (53) and
(56) results in
j
'
c
c
`
j`/c. (57)
Launder and Spalding [42] proposed calculating and c numerically from
differential transport equations formulated as moments of that for momen-
tum, such as Eq. (9), and then in turn calculating j
'
from Eq. (57).
Unfortunately, the approximate expressions for the terms in these moments
of the momentum balance that have been suggested by various investigators
are somewhat arbitrary, and in any event introduce a number of empirical
coefcients in addition to c
c
`
of Eq. (57). Although some success has been
achieved with the c model in the few simple ows for which an extensive
set of data exists from experimental measurements and/or direct numerical
simulations and therefore for which the model is not needed, the predictions
for more complex ows have been disappointing in accuracy or are
precluded by singularities in j
'
and/or l. The cu'v' or Reynolds-stress
model, which adds a transport equation for u'v' similar to those for and
c, appears to be free of singularities even in a concentric circular annulus
(see Hanjalic and Launder [43]) but is essentially a correlative rather than
a predictive model for the important region near the wall. The large eddy
simulation (LES) method starts from the time-dependent equations of
conservation but introduces arbitrary terms such as those of the c and
Reynolds-stress models as well as utilizing the c model or additional
empirical terms for the region near the wall. For an illustration of the
applicability of this model, again for an annulus, see Satake and Kawamura
[44].
4. T he Experimental Data of Nikuradse
Nikuradse [4547] in 1930, 1932, and 1933 obtained extensive and
precise sets of experimental data for the time-mean velocity distribution in
the turbulent core and for the axial pressure gradient for the fully developed
ow of water in smooth round tubes for 600 Re 3.2410' and in round
tubes with a uniform articial roughness e corresponding to 15(a/a)507
for 600 Re 10'. Furthermore, he presented his data in tabular as well
as graphical forms, thereby making it readily available in full numerical
detail to subsequent investigators. For more than 60 years these data have
been generally accepted as the primary standard for the development of
models and correlating equations, for the evaluation of arbitrary constants
therein, and for evaluation of the data of subsequent investigators.
However, Miller [48] in 1949 identied an apparent discrepancy in the
tabulated values of y in Table 3 of Nikuradse [46]. By means of an inquiry
addressed to Prandtl, he learned that Nikuradse had added 7.0 to each value
of y (the dimensionless distance of each point of measurement from the
wall) in order to force the measured values of u in his smooth pipes to
approach the limiting form of Eq. (19) as y0. This discovery by Miller
may readily be conrmed by comparing the values of uy} plotted in Fig.
15 of Nikuradse [45] with those in Fig. 24 of Nikuradse [46]. This
adjustment has an insignicant effect for the large values of y but
precludes the use of the tabulated values for the small values. Robertson et
al. [49] in 1968 conjectured that Nikuradse [47] might also have adjusted
his experimental values for the velocity distribution near the centerline of
the artically roughened pipes in order to force comformity of the values of
u
'
u
"
to 3B/2 3.75, and Lynn [50] in 1959 discreetly noted the
extraordinarily low scatter in the experimental values used by Nikuradse
[46] to infer (incorrectly) that the eddy viscosity approaches zero at the
centerline. However, on the whole the measurements by Nikuradse of the
velocity distribution in the turbulent core as well as those of the axial
pressure gradient for both smooth and rough pipe have stood the test of
time, and these adjustments, except possibly those implied by Lynn, have
not had any serious consequences in either uid mechanics or heat transfer.
Nikuradse [46] used his experimental data rst of all to test the law of the
wall, Eq. (18). He found conformity for the smooth pipes for all ows, all
diameters, and all locations, including even the region near the centerline
where it might not have been expected to hold. Next, he found that Eq. (24)
with A5.5 and B2.5 represented these values well for all y50, again
even for the region near the centerline. He also found Eq. (29) with C8.5
and B2.5 to be successful for representation of the measured velocity
distribution for the articially roughened pipes at the larger values of a.
However, his experimental values for the friction factor in the smooth pipes
were found to be represented better by
2
f
`
2.00 2.46 lna} (58)
than by Eq. (26) with A5.5 and B2.5, which yields A(3B/2) 1.75
and the same value of the coefcient B as for the velocity distribution. The
discrepancy in the constant (2.0 as compared to 1.75) may be attributed to
Fic. 2. Experimental velocity distribution in fully developed turbulent ow of water in a
127-mm Plexiglas tube (RRe). (Reprinted with permission from Lindgren and Chao [51],
Figure 1. Copyright 1969 American Institute of Physics.)
the neglect of the boundary layer near the wall, as represented in the limit
by Eq. (33), and of the wake, as represented in the limit by Eq. (34). Both
of these deviations from Eq. (24) are well illustrated by the much later data
of Lindgren and Chao [51] in Fig. 2. On the other hand, the discrepancy in
the coefcient (2.46 as compared to 2.50) is unacceptable on theoretical
grounds. The overly simplied and incongruent expressions for the velocity
distribution and the friction factor that appear unexplained in most of our
current textbooks and handbooks are a legacy of the failure of Nikuradse
to obtain sufciently precise and accurate values for the velocity distribution
near the wall and near the centerline and to develop correlating equations
encompassing these regions. Nikuradse is, of course, not responsible for the
failure of subsequent investigators and writers to explain and provide a
rational correction for these anomalies.
A similar discrepancy exists for the correlating equations of Nikuradse
[47] for articially roughened pipe, for which he correlated his experimental
data for the axial pressure gradient for asymptotically large values of the
Reynolds number with the expression
2
f
`
4.92 2.46 ln
a
e
. (59)
It may be noted that Eq. (30) with C8.5 and B2.5 predicts a value of
4.75 rather than 4.92 for the constant and a value of 2.5 rather than 2.46 for
the coefcient.
For the reasons just cited, the experimental measurements of the velocity
distribution by Nikuradse do not provide a test of Eqs. (19) and (33) or
allow evaluation of the coefcients of the latter. Although he apparently did
not recognize the existence of a wake, his experimental values of u
'
u
conform crudely to Eq. (34) even for y/a as low as 0.2. On the whole, they
suggest a value of E between 6.7 and 7.5. The failure of the values very near
the centerline to conform to this relationship may be due to the adjust-
ments implied by Lynn [50] as well as to the very small differences in the
measured velocities at closely adjacent locations in that region.
Nikuradse [45] determined the values of the mixing length plotted in Fig.
3 from the slope of plots of the velocity distribution. The values of the
mixing length thus determined appear to be independent of the Reynolds
number and of the roughness ratio for sufciently large values of the
Reynolds number, implying a great generality for the relationship between
l/a and y/a. Nikuradse [46] subsequently proposed representation of all of
these values by the empirical expression
l
a
0.14 0.08
1
y
a
`
0.06
1
y
a
"
, (60)
which he attributed to Prandtl and interpreted as an interpolation for-
mula between Eq. (40) with k 0.4 for y0 and Eq. (47) with l
0.14
for ya. This value of l
results in a net numerical coefcient of 4.76 in

Eq. (48). From these same slopes of the velocity distribution he determined
the values of the eddy viscosity plotted in Fig. 4 and concluded erroneously
that it approaches zero at the centerline.
The experimental data of Nikuradse for fully developed ow in a round
tube and his own correlations for u, j
'
, l, and f based on these data have
been described and analyzed here in some detail because they have had
great inuence on the predictions and correlations for convective heat
transfer. In addition to the caveats noted earlier, some subsequent investi-
gators have questioned the numerical values of the constants and coefcients
determined by Nikuradse. In particular, Hinze [52] and other have asserted
that the constant A and the coefcient B1/k of Eq. (24) are Reynolds-
Fic. 3. Experimental mixing lengths in smooth and articially roughened round tubes.
(From Nikuradse [45], Figures 9 and 12.)
number dependent. Such uncertainties and variations have not been ex-
plored herein since none of the numerical values determined from the data
of Nikuradse appear in the nal expressions for either ow or heat transfer.
5. Power-L aw Models
Power-law models for the velocity distribution and the friction factor
might not have merited attention herein had not a recent attempt been made
Fic. 4. Experimental eddy viscosities in smooth tubes. (From Nikuradse [46], Figure 27.)
to resuscitate them. Furthermore, they might logically have been included
in Section II, A, 3. The deferral to this point is because the experimental data
of Nikuradse, as described in Section II, A, 4, are essential to their
interpretation and evaluation.
Blasius [53] in 1913 plotted the available experimental data for the
friction factor for round tubes, which then extended only up to Re 10`,
versus Re in logarithmic coordinates and found that a satisfactory represen-
tation could be achieved with a straight line equivalent to
f
0.0791
Re"
. (61)
C. Freeman [54] in 1941, in a Foreword to a compilation of the extensive
set of experimental data obtained in 1892 by his father, J. R. Freeman, but
not published until 49 years later, speculated that had Blasius had access to
these values, which extend up to Re 9 10`, he might have developed a
more general correlating equation and thereby changed the course of history
in applied uid mechanics. This assertion not only was justied when it was
written, but has proven prophetic.
Prandtl [55] in 1921, and therefore prior to his development of the
mixing-length model, recognized from dimensional considerations that Eq.
(61) implies
u
u
'
y
a
`
, (62)
which by virtue of the numerical coefcient of 0.0791 may also be expressed
as
u8.562(y)`. (63)
Nikuradse [4547] tested Eq. (62) with his experimental data for the
velocity distribution and found that it provided a good representation only
for the turbulent core, only for smooth pipes, and only for Re 10`.
Accordingly, he generalized Eq. (62) as
u
u
'
y
a
, (64)
which corresponds to
u[( y). (65)
Here : is an arbitrary dimensionless exponent and [ is an arbitrary
dimensionless coefcient. From integration of Eq. (65) over the cross-section
it follows that
[
(1 :)(1 2:)u
"
2:(a)
. (66)
He determined numerical values of : and [ as functions of Re and e/a from
his experimental velocity distributions, but abandoned this mode of corre-
lation as inferior to Eqs. (24), (26), (29), and (30).
Nunner [56] in 1956 somewhat revived the power-law model by discover-
ing that the empirical relationship
: 2 f ` (67)
provides a good approximation for both smooth and roughened pipes.
However, a separate correlation is required for the friction factor as a
function of the Reynolds number and roughness ratio.
Thirty-seven years later, Barenblatt [54], apparently unaware of the work
of Nunner, rationalized the form of the power law for the velocity distribu-
tion using scaling arguments, and proposed, on the basis of the data of
Nikuradse [46] for smooth pipe, empirical expressions for : and [ in Eq.
(65) as functions of Re. These expressions are not reproduced herein, since
that for : is inferior to Eq. (67) and that for [ is equivalent but inferior to
most other correlating equations for the friction factor.
Equation (65) with : from Eq. (67) and [ from Eq. (66), and with u
"
from
Eq. (58) or (59), whichever one is appropriate, is slightly superior to Eq. (24)
for 30 y0.1a. However, it is seriously in error for larger as well as
smaller values of y. These errors might have been anticipated from the
predictions by Eq. (65) of an unbounded velocity gradient at the wall and a
nite velocity gradient at the centerline.
Equation (66) may be reexpressed as
f
(1 :)(1 2:)
[:`2''`'`Re
`''
, (68)
which implies that a xed-power dependence of the velocity on the distance
from the wall over the entire cross-section is required to obtain a power-law
dependence of the friction factor on the Reynolds number. It may therefore
be inferred from the previously cited failures of the power-law model for the
velocity near the wall and near the centerline, and more importantly from
the observed dependence of : on Re, that a pure power-law model for the
friction factor cannot have any real range of validity with respect to Re. The
semilogarithmic dependence of the square root of the reciprocal of the
friction factor on the Reynolds number has already been noted to be subject
to a related but numerically less severe defect.
Churchill [58] has recently compared the predictions of the power-law
models for the velocity distribution and the friction factor with experimental
data and other correlating equations. These comparisons support the
preceding qualitative conclusions.
The failure of the power-law models might have been anticipated on the
basis of dimensional analysis. Rayleigh [25] used a power-series expansion
as a mechanical means of identifying a minimal set of dimensionless groups
from a listing of the variables, physical properties, and parameters. He fully
recognized, as demonstrated by his own illustrations of this technique, that
the derivation of the rst term in the expansion in the form of a product of
arbitrary powers of the independent dimensionless groups did not imply
either powers or products for the unknown functional relationship. Unfor-
tunately, such misinterpretations plague us to this day. Indeed, relationships
in the form of powers other than the unity ordinarily occur only in
asymptotic expressions, such as the limiting form of Eq. (32), or in special
cases, such as with the friction factor (but not the velocity distribution) for
fully developed laminar ow in a round tube.
6. T he Analogy of MacL eod
Before examining some important recent work it is convenient if not
essential to describe a little-known conjecture that suggests a means of
obtaining congruence of the turbulent shear stress and the velocity distribu-
tion for round tubes with their counterparts for ow between parallel plates
of innite extent. Rothfus and Monrad [59] showed that complete congru-
ence of the velocity prole in fully developed laminar ow in a round tube
with that between parallel plates may be achieved by specifying t
"^
t
"'
and a b, where the subscripts R and P designate round tubes and parallel
plates, respectively, and b is the half-spacing of the plates. This requirement
is excessive; a sufcient condition for u
^
y, a} u
'
y, b} is simply
that ab. MacLeod [60] subsequently speculated that this latter
relationship might also hold for fully developed turbulent ow. His specu-
lation is beautifully conrmed in Fig. 5, in which the experimental values of
Whan and Rothfus [61] for u
'
in ow between parallel plates are compared
with a curve representing the corresponding experimental values of Senecal
and Rothfus [62] for a round tube. The velocities at the central plane and
the centerline were chosen for this comparison because as extreme values
they provide the most severe test of the analogy. It is evident that the
analogy does not hold for the regime of transition from fully developed
laminar to fully developed turbulent ow. This discrepancy was to be
anticipated since the onset of transitional ow has long been known to
occur at differing values of a and b. It may be inferred from Eqs. (17),
(38), and (39) that the analogy of MacLeod applies directly to (u'v'), j
'
/j,
and l insofar as it is valid for u.
The special importance of the analogy of MacLeod is that it provides a
formal justication for the use of experimental data as well as values from
direct numerical simulations for u and (u'v') in ow between parallel
Fic. 5. Experimental conrmation of the analogy of MacLeod for central velocities. (From
Whan and Rothfus [61], Figure 3.)
plates in the development of correlating equations for round tubes. A critical
assessment of the analogy of MacLeod as applied to (u'v') would appear
to be of crucial importance in both ow and heat transfer, but that requires
values of greater precision and reliability for both geometries than are
currently available from either experimental measurements or direct numeri-
cal simulations.
The results from direct numerical simulations for round tubes are current-
ly less extensive and reliable than those for parallel plates because of the
computational complexities associated with curvature. On the other hand,
the experimental measurements for ow in round tubes are more extensive
and reliable than those for parallel plates because of the difculty of aligning
and supporting plates of sufcient extent and small enough spacing to
minimize side-wall effects. Entrance effects also appear to be more serious.
The law of the wall of Prandtl [Eq. (18)] may be noted to be a special
case of the analogy of MacLeod for the region near the wall, but, on the
other hand, it is presumed to be applicable for all shear ows, not just those
for round tubes and parallel plates.
7. T he Colebrook Equation for the Friction Factor
in Naturally Rough Piping
The contribution of Colebrook [63] to the prediction of turbulent ow in
piping is perhaps second only to that of Nikuradse in practical importance.
Although the results of his work appear implicitly in almost every plot of
the friction factor in our handbooks and textbooks, he is seldom cited as the
primary source.
Nikuradse [47] chose uniform articial roughness for his experimental
investigation for the obvious reasons of reproducibility of the measurements
of the ow and of simple quantitative characterization of the roughness.
Such measurements would not be expected to be representative for the
naturally occurring roughness of commercial piping, which is characterized
by the highly variable and perhaps chaotic amplitude and spacing asso-
ciated with particular materials of construction (such as glass and concrete)
and different methods of manufacture (such as extruding and casting), as
well as with aging, corrosion, erosion, fouling, and different methods of
linkage (such as welding and threading).
The measurements of Colebrook for a variety of natural materials and
conditions revealed that the pressure drop not only depends on the
magnitude of the roughness but, even more importantly, has a completely
different functional dependence on the Reynolds number: The friction factor
decreases to an asymptotic value, as contrasted with an increase to an
asymptotic value for uniform articial roughness and an unending decrease
for smooth piping. In order to represent this behavior, he arbitrarily dened
and designated by e
'
a nominal roughness for each material and condition
that would result in the same asymptotic value of f
'
for the friction factor
for very large Reynolds numbers as the value of the uniform articial
roughness of Nikuradse. Thus, on the basis of Eq. (50),
e
'
a
exp
4.92 (2/ f
'
)`
2.46
. (69)
This ingenious concept of correlation avoids the necessity and difculty of
measuring and characterizing the roughness statistically, and instead focuses
directly on the behavior of primary interest, namely the shear stress on the
wall.
Colebrook also found that his measured pressure drops at less than
asymptotically large values of the Reynolds number could be represented
closely by
2
f
`
2.46 ln
e
'
7.39a
1
2.25a
. (70)
He justied the form of Eq. (70) simply by asserting without any rational-
ization that the two terms in the argument of the logarithm must be
additive. Churchill [64] subsequently reinterpreted Eq. (70) in terms of the
canonical correlating equation of Churchill and Usagi [65], namely,
yx} y
"
x} y
x}, (71)
where y
"
x} and y
x} are asymptotic values or expressions for small and

large values of x, respectively, and n is an arbitrary exponent. By trial and
error, exp(2/ f )`/2.46} was found to be a better choice for yx} in Eq. (71)
than (2/ f )` or f . Then, from Eq. (58) rearranged as
exp
(2/ f )`
2.46
2.255a, (72)
y
"
x} 2.255a, and from Eq. (59) rearranged as
exp
(2/ f )`
2.46
7.389
a
e
'
, (73 )
y
x} 7.389a/e
'
. A value of n 1 was chosen on the basis of the
experimental data of Colebrook. The result of this procedure may be
expressed as
2
f
`
2.002.46 ln
a
1 0.304(e
'
/a)a
, (74)
which is exactly equivalent to Eq. (70). Most of the graphical representa-
tions of the friction factor in the turbulent regime in the current literature
are simply plots of Eq. (74) or its near equivalent, most often in the form of
f or f ` versus Re a(8/ f )` or Re f ` 8à with e
'
/a or e
'
as a
parameter, accompanied by a table of values of e
'
for various materials and
conditions. Most of the values of e
'
that appear in the standard tabulations
were determined many decades ago by Colebrook and his contemporaries
and may not be representative of modern materials and modern methods of
manufacture and joining. A redetermination and recompilation of values of
e
'
would appear to be worthwhile.
The plot in Fig. 6 of the experimental data of Nikuradse and Colebrook
for uniformly and naturally roughened pipe, respectively, in the form of
2
f
`
2.46 ln
2a
e
as a function of
(e)`
e`(t
"
/j)
j`

e
a
`
(a)`
Fic. 6. The transitional behavior for uniformly and naturally roughened round tubes: ()
Data of Colebrook [63], natural roughness; all other points are from Nikuradse [47], uniform
roughness.
demonstrates the fundamentally different paths of transition for uniformly
roughened and naturally roughened pipe from ow in a smooth pipe, as
represented by the linear oblique asymptote, to ow controlled wholly by
roughness, as represented by the horizontal asymptote.
On the basis of Eqs. (24), (26), (29), and (30), the velocity distribution in
the turbulent core near the wall of naturally rough pipe may be predicted
speculatively by dividing y in the argument of the logarithm by
1 0.304(e
'
/a)a.
8. Experimental and Computed Values of u'v' Near the Wall
Difculty has been encountered in the past in determining : in Eq. (32)
or (33), since measurements of very small values of u'v' or u very near the
wall are required. However, the uncertainty in : has been greatly reduced in
the past decade by virtue of direct numerical simulations. This computational
method, which was pioneered by Orszag and Kells [66], is essentially free
from empiricism except for the choice of a wave form, but it is sensitive to
the number of grid points used in all three coordinate directions. The
implementation of direct numerical simulations for channels is yet limited
by computational demands, with only a few exceptions, to ow between
parallel plates of unlimited extent and even then to values of b just above
the minimal value of about 145 for fully developed turbulence. The com-
puted values of u'v' by Kim et al. [67], Lyons et al. [68], and Rutledge and
Sleicher [69] are in fair agreement with one another and with the best
experimental measurements, such as those of Eckelmann [70], for the
intrinsically important region very near the wall where the behavior is
presumed to be independent of or at least negligibly dependent on b. This
combination of computational and experimental results for parallel plates
conrms beyond question the form of Eq. (32) and indicates a value
5710" for :.
9. Experimental Values of u'v' and u Near the Centerline
The coefcient E of Eqs. (34) and (35) is the analog of : for the region
near the wall. It may in principle be determined from Eq. (34) by virtue of
experimental values of u, from Eq. (35) by virtue of experimental or
computed values of (u'v'), or from the derivative of Eq. (34) by virtue of
experimental values of du/dy. Unfortunately, all three of these methods
require experimental values of greater accuracy and precision than are
currently available. Even the values of (u'v') computed by direct numer-
ical simulation are marginal in this respect.
It follows from Eqs. (38) and (34) that for ya
j
'
ja
1
a _
dy
du
1
y
a
1
1
2E
1
a
5
1
2E
. (75)
Equation (75) predicts that near the centerline, j
'
/ja approaches an
asymptotic value independent of y and essentially independent of a. Such
behavior has been conrmed (see, for example, Figure 5 of Churchill and
Chan [71]). Groenhof [72] examined and compared experimental determi-
nations of j
'
/ja by ve sets of investigators for round tubes and one
investigator for parallel plates. These values range from 0.062 to 0.08,
corresponding to values of E from 8.06 to 6.25.
10. T he Experimental Data of Zagarola
Most of the correlating equations mentioned earlier, including the empiri-
cal constants, are based on the experimental data of Nikuradse [46, 47]
despite their age and indicated limitations. Recently a new, comprehensive
set of experimental data has been obtained for fully developed turbulent
ow in a round tube that challenges the dominant role of the measurements
of Nikuradse. Zagarola [73] in 1996, using modern instrumentation and
carefully controlled conditions, measured the time-averaged velocity and
axial pressure gradient in air owing through a 129-mm tube with a highly
polished surface. His ows extended from Re 3.5510" to 3.52610` and
thus to higher values than those of Nikuradse, but not to as low ones.
Zagarola conceded that his own measurements of the time-mean velocity
were excessively high for y/a 0.0155 for all Re. This requires discarding all
of his values in the viscous sublayer (0y10) and all but a few in the
buffer layer (10y30). Also, the slight displacement of many of the
maximum measured values of the velocity from the centerline suggests that
their accuracy is marginal for purposes of differential analysis in that region.
Despite great effort to attain an aerodynamically smooth surface, the
directly measured roughness ratio of e/a 2.410' is, according to Eq.
(74), of sufcient magnitude to have a signicant effect on both the friction
factor and the velocity distribution at the higher values of Re. The effective
roughness ratio, e
'
/a, was concluded by Churchill [58] to have the some-
what lower value of 710`, which, however, is still aerodynamically
signicant for the largest values of Re studied by Zagarola.
In spite of these caveats, the tabulated values of Zagarola for the velocity
and the friction factor represent a very signicant contribution to uid
mechanics and may be considered to supplant the tabulated experimental
data of Nikuradse within the range of conditions for which they overlap,
namely 3.15810" Re 3.2410'. They justify renement or replace-
ment of most of the algebraic and graphical correlations for the velocity
distribution and the friction factor in the current literature.
Zagarola found, as illustrated in Figs. 7 and 8, a signicant variation of
the coefcient k 1/B of Eqs. (23), (24), and (26) with Re, but concluded
that a value of 0.436 provided an adequate representation for the
semilogarithmic regimes in all three instances. The constant A of Eq. (24),
as determined on the basis of k 0.436, was also found to vary somewhat
with Re, as illustrated in Fig. 9, but a value of 6.13 was concluded by
Zagarola to provide an adequate representation for all values of y and Re
for which this equation is applicable. The plots of u
'
u versus lna/y},
such as illustrated in Fig. 10 for 3.110" Re 2.510`, which he used
as one method of evaluating k, indicated to him that the following
expression was preferable to Eq. (23) for 0.01 y/a 0.1:
u
'
u
1
0.436
ln
a
y
1.51. (76)
Fic. 7. Variation with Re of the coefcient k 1/B in Eqs. (24) and (43) for
50 y0.1a. (From Zagarola [73], Figure 4.38.)
Fic. 8. Variation of the coefcient k 1/B in Eq. (23) for u
'
and Eq. (26) for u
"
for various
sets of increasing values of Re. (From Zagarola [73], Figure 4.30.)
Fic. 9. Variation with Re of the constant A of Eqs. (24) and (43) for 50 y0.1a with
the coefcient k 1/B xed at 0.436 and free. (From Zagarola [73], Figure 4.40.)
Fic. 10. Determination of the deviation in the velocity at the centerline due to the wake.
(From Zagarola [73], Figure 4.53.)
Fic. 11. Comparison of semilogarithmic and power-law representations of the velocity
distribution for 3110` Re 4.410'. (From Zagarola [73], Figure 4.44.)
Although Zagarola did not present a correlating equation for the region of
the wake, Eq. (24) with A6.13 and B1/0.436 may be combined with
Eq. (76) to obtain the following expression for the velocity at the centerline
itself:
u
'
7.64
1
0.436
lna}. (77)
Zagarola also tested Eq. (65) and found, as illustrated in Fig. 11 for
3.110" Re 4.410' and 001 y/a 0.1, that the expression
u8.7(y)"``, (78)
where 8.7 and 0.137 are purely empirical, represents the measured velocities
better for 30 y500 but much more poorly for y500 than
u6.13
1
0.436
lny}. (79).
Both representations are seen to fail for y50. Similar representations
and misrepresentations were found to be provided by these two expressions
for other ranges of Re.
Zagarola represented his experimental data for the friction factor with the
following expression:
2
f
`
3.30
138.5
(a)""

1
0.436
lna}. (80)
He proposed the term 138.5/(a)"" as a correction for the deviation of the
velocity distribution in the boundary layer from the semilogarithmic regime
on the basis of a subsequently described expression of Spalding [74] for the
velocity distribution for all y0.1a. Zagarola actually proposed several
different leading coefcients for Eq. (80). The value of 3.30 was chosen here
for consistency with Eqs. (76) and (79).
11. Overall Correlating Equations for the Velocity Distribution
With only a few exceptions, to be noted here, the correlating equations of
the past, as well as those of Zagarola, are for a single regime, primarily the
turbulent core near the wall, although such expressions have often been
implied to be applicable for the entire turbulent core. Equations (44)(46),
as well as Eqs. (53) and (54), purport to encompass the boundary layer, but
the rst three do not include the higher-order terms of Eq. (33) and the latter
two imply erroneously that : 0. Despite the presence of a in several of
these expressions, none of them approaches Eq. (34) as ya.
Churchill and Choi [75] combined the limiting form of Eq. (19) for
y0 and Eq. (24) with A5.5 and B2.5 in the form of Eq. (71) and
chose a value of 2 for n on the basis of the experimental data of Abbrecht
and Churchill [22] to obtain
u
y
_
1
y
2.5 ln9.025y}
`
`
. (81)
Here 5.5 2.5lny} is expressed as 2.5 ln9.025y} for compactness and
to emphasize the presence of a singularity at y1/9.025 0.1108. This
singularity may be avoided without a signicant effect on the predictions of
u for any value of y by simply adding unity to the argument of the
logarithm to obtain
u
y
_
1
y
2.5 ln1 9.025y}
`
`
. (82)
Fic. 12. Representation of the velocity distribution in a smooth round tube by Eq. (82).
(From Churchill and Choi [75], Figure 2.)
Equation (82) may be seen in Fig. 12 to represent the data upon which it is
based very well for all but the largest and smallest values of y. The
deviations for y2 are probably due to experimental error but those for
the largest y are denitely a consequence of failing to account for the wake.
The previously mentioned expression of Spalding [74] for the viscous
sublayer, buffer layer, and turbulent core near the wall is
yu0.1108
_
e
0.4u
1 (0.4u)
(0.4u)`
2

(0.4u)`
3
. (83)
Equation (83) approaches Eq. (24) with A5.5 and B2.5 for large values
of y, just as does Eq. (82), but approaches Eq. (33) as y0, albeit with
a somewhat low value of 4.1310" for :. It is thereby superior to Eq. (82)
functionally but is not necessarily more accurate numerically. The inverse
form of Eq. (83) is inconvenient functionally for a specied value of y, but
numerical values of u may then readily be obtained by iteration. As
indicated by the absence of a, Eq. (83) also fails to account for the wake.
Even earlier, Reichardt [76] proposed the following expression for the
entire cross-section, including the region of the wake:
u7.8
_
1 e
y/11
y
11
e
0.33y
2.5 ln
3a(1 0.4y)(2a y)
2(3a 4ay2(y)`)
. (84 )
Equation (84) conforms to Eq. (34) with E7.5 for ya and to Eq.
(24) with A5.5 and B2.5 for intermediate values of y. Reichardt
apparently intended it to conform to the rst two terms on the right-hand
side of Eq. (33) for y0, but it fails in that respect since the coefcients
of (y)` and (y)` that result from the expansion of the logarithmic and
exponential terms in series are not quite zero. Furthermore, the correspond-
ing coefcient of the term in (y)" has the very excessive value of
1.54810` as compared to 710"/41.7510" from the direct nu-
merical simulations. (Values of 18.738 and 0.5071 in place of 11 and 0.33,
respectively, would eliminate the terms in (y)` and (y)`, but would
decrease the coefcient of (y)" only slightly to 1.37110` and therefore
insufciently.) Despite these discrepancies, the numerical predictions of Eq.
(84) do not differ greatly from those of Eqs. (82) and (83) for y0.1a
while it is more accurate functionally as well as numerically for
0.1aya.
B. Nrv Ixiovrn Foxii:1ioNs :Nn Cori:1iNc Eqi:1ioNs
The models just described all appear to have a defect or shortcoming and
the correlating equations to be limited in scope or generality. The objective
of the work described in this section has been to develop formulations and
correlating equations that avoid these defects and limitations.
1. Model-Free Formulations
The mixing-length model, as expressed by Eq. (39), was proposed by
Prandtl to facilitate the prediction of the turbulent shear stress in the
time-averaged equation for the conservation of momentum on the premise
that algebraic or differential correlating equations for the mixing length,
such as Eqs. (40), (47), and (49), would be simpler or more general than a
correlating equation for the shear stress itself. The eddy diffusivity model of
Boussinesq, as represented by Eq. (37), may be interpreted to have an
equivalent objective.
In contradistinction, Churchill and Chan [71, 77, 78] and Churchill [79]
investigated the direct use of the turbulent shear stress itself as a variable for
integration and correlation, thereby avoiding the need for heuristic variables
such as the mixing length and the eddy viscosity altogether. The conse-
quences are generally favorable and in some respects quite surprising. They
started from Eq. (17), which is expressed in terms of (u'v')ju'v'/t
"
,
namely the local turbulent shear stress as a fraction of the shear stress on
the wall. The negative sign was used in this denition since u'v' is negative
over the entire radius of a round tube. Churchill [80] subsequently proposed
as slightly advantageous the use of a new, alternative dimensionless quantity
(u'v')ju'v'/t, which may be recognized as the local fraction of the
total shear stress due to turbulence. Equation (13) then becomes
1
y
a
[1 (u'v')]
du/dy
. (85)
From physical considerations (u'v') must be positive, less than unity, and
greater than zero at all locations within the uid. This latter characteristic
gives (u'v') a signicant advantage in terms of correlation over (u'v'),
which is zero at the centerline.
Eliminating du/dy between Eqs. (38) and (85) reveals that
j
'
j

(u'v')
1 (u'v')
(86)
The eddy viscosity is thus seen to be related algebraically to (u'v'), which
is a physically well-dened and unambiguous quantity, and thereby to be
independent of its heuristic diffusional origin. (Boussinesq was either very
intuitive or just lucky.) It further follows from Eq. (85) and the indicated
behavior of (u'v') that j
'
/j is also nite and positive at all locations
within the uid, including the centerline.
It similarly follows from Eqs. (39) and (85) that
(l)`
(u'v')
1
y
a
[1 (u'v')]`
(87)
The mixing length is thus also independent of its mechanistic and heuristic
origin but is unbounded at the centerline. How did such an anomaly, which
has a counterpart in all geometries and which refutes the very concept of a
mixing length for practical purposes, escape attention for more than 70
years? One reason is the uncritical extension of respect for Prandtl to all
details of his work. A second reason is the false mindset created by Fig. 3.
A third is the requirement of even more precise data for the velocity
distribution near the centerline than is available even today, although in
retrospect the singular behavior of the mixing length at the centerline is
apparent, at least qualitatively, from most sets of data.
Although the anomalous behavior of the mixing length was apparently
rst recognized by Churchill [80] as a consequence of his derivation of Eq.
(85) and therefore because of his introduction of (u'v') as a variable, it
could have been identied much earlier merely by the substitution of
du/dy from Eq. (34), which goes back at least to Reichardt [76] in 1951,
in the combination of Eqs. (17) and (39) or the equivalent. Inference of this
singularity from Eq. (17) requires consideration of Eq. (35). In any event,
the continued use of the mixing length does not appear to have any
justication under any circumstance.
Now reconsider Eq. (85), which represents the momentum balance in a
round tube in terms of (u'v'), which, as noted, is well behaved and
constrained between zero and unity for all values of y. Formal integration
results in the following expressions for the time-averagedvelocity distribution:
u
,
'
"

1
y
a
[1 (u'v')]dy
y
(y)`
2a

,
'
"

1
y
a
(u'v')dy (88)
These may be expressed more compactly as
u
a
2 ,
^
[1 (u'v')]dR`
a
2
(1 R`)
,
^
(u'v')dR` (89)
The leftmost forms of Eqs. (88) and (89) are more convenient for numerical
integration because the rightmost ones involve small differences of large
numbers, but the latter have the advantage of demonstrating that the effect
of the turbulence is simply to provide a deduction from the well-known
expressions for purely laminar ow at the same value at a.
Equation (89) may in turn be integrated formally over the cross-sectional
area to obtain the following expression for the space- or mixed-mean
velocity and thereby the friction factor:
2
f
`
u
"

a
2 ,
"
_,
[1 (u'v')]dR`
dR`. (90 )
Equation (90) may be reduced by integration by parts to obtain
2
f
`
u
"

a
4 ,
"
[1 (u'v')]dR"
a
4

a
4 ,
"
(u'v')dR",
(91)
which involves only a single integral. The rightmost form of Eq. (91) reveals
that the effect of the turbulence on the mixed-mean velocity is also simply
a deduction from the well-known expression (Poiseuilles law) for purely
laminar ow. This deductibility of an integral term from the expressions for
the time-mean velocity distribution and the mixed-mean velocity in purely
laminar ow may seem to be obvious in retrospect, but such a structure is
not so evident in the analogs of Eqs. (89) and (91) in terms of the eddy
viscosity and the mixing length, and does not appear ever to have been
mentioned in the literature. Also, although it is evident in retrospect that the
double integral of the analogs of Eq. (90) in terms of the eddy viscosity and
the mixing length may be reduced to a single integral by means of
integration by parts, that simplication was apparently never recognized or
implemented because the more complex forms obscure this possibility. It
may be noted that (u'v') has no advantage over (u'v') in this respect, that
is, both the deductibility of the effects of turbulence and the possibility of
integration by parts are quite evident when starting from Eq. (17) rather
than Eq. (85).
Equations (88)(91) are exact insofar as Eq. (13) is valid, but some
empiricism is necessarily invoked in the required correlating equation for
(u'v'). Before turning to such expressions several partial precedents for
Eqs. (89) and (91) should be acknowledged. Kampe de Fe riet [81] derived
the equivalent of Eq. (89) and the analog of Eq. (91) for parallel plates in
terms of (u'v') but did not implement these expressions; while Bird et al.
[35, p. 175], note that the use of a correlating equation for (u'v') rather
than one for the eddy viscosity or the mixing length might lead to a simpler
integration for the velocity distribution.
2. Correlating Equations for the L ocal Turbulent Shear Stress
Despite the advantages of the dimensionless turbulent shear stress in
predicting the velocity distribution and the mixed-mean velocity, as de-
scribed in the immediately preceding paragraphs, the general failure to
recognize those advantages has resulted in a dearth of correlating equations.
One exception is due to Pai [82], who was inspired by the aforementioned
formulations of Kampe de Fe riet to represent the experimental data of
Nikuradse [46] for the velocity distribution at Re 3.2410' (his highest
rate of ow) by a polynomial in R and to substitute the derivative of that
expression in Eq. (17) to obtain
(u'v')0.9835R(1 R`"). (92 )
Equation (92) is reasonably accurate numerically and functionally for
ya(R 0), but fails badly for both small and intermediate values of
y. Integration of either Eq. (92) or the velocity distribution from which it
was derived leads to an obviously invalid expression for the friction factor.
Churchill and Chan [71] constructed a more comprehensive and general
expression for (u'v') that may be reexpressed for simplicity in terms of
(u'v') as follows:
(u'v')
_
0.7
y
10
`
``
exp
2.5
y
2.5
a
1
4y
a
``
``
.
(93 )
The construction of Eq. (93) will be described in detail since almost all
subsequent expressions herein for the velocity distribution, the friction
factor, and the heat transfer coefcient are based on this expression with
only slight numerical modications.
Equation (93) has the form of Eq. (71) with n 8/7,
(u'v')
"
0.7
y
10
`
(94 )
and
(u'v')
1
2.5
y

2.5
a
1
4y
a
. (95)
Equation (94) has the limiting form of Eq. (32) with the previously discussed
value of 710" for :. Equation (95), on the other hand, is based on the
following expression of Churchill [83] for the velocity distribution across
the entire turbulent core:
u5.5 2.5 lny}
15
4
y
a
`
10
3
y
a
`
. (96 )
The terms in ( y/a)` and (y/a)` were added to the correlating equation
of Nikuradse [46] to encompass the wake. The coefcients 15/4 and (10/3)
were chosen to force du/dy0 and u
'
u7.5(1 y/a)` as
ya. The coefcient of 7.5 is based on Eq. (84) of Reichardt [76] and
therefore indirectly on the experimental data of Reichardt himself for
parallel plates as well as that of Nikuradse for round tubes. Substituting for
du/dy in Eq. (85) from the derivative of Eq. (96) and then simplifying
algebraically results in Eq. (95). The absolute value sign and the approxim-
ation of 1 (2.5/y) by exp2.5/y} in Eq. (93) are merely mathematical
contrivances to avoid singularities in ranges of y in which these terms have
an otherwise negligible role. Equation (95) was constructed from the
correlating equation for the velocity because the supporting data are more
Fic. 13. Representation of experimental data and directly simulated values of (u'v') at
small values of b. (From Churchill [80], Figure 1.)
extensive and reliable than those for the turbulent shear stress itself.
However, the value of 8/7 for the arbitrary combining exponent is based
on the experimental data of Wei and Willmarth [84] for u'v' in ow between
parallel plates, which appear to be the most accurate ones over a wide range
of values of both y and b. Hence, the validity of the analogy of MacLeod
is implied in the value of this exponent as well as in the value of 710"
for :.
Equation (93) is compared with experimental data and values determined
by direct numerical simulations for small values of y and b in Fig. 13 and
with the experimental data of Wei and Willmarth for moderate and large
values of y and b in Fig. 14. The agreement appears to be within the
bands of uncertainty of the experimental and computed values. The small
oscillations in the curves in Fig. 13 are an artifact of the structure of Eq.
(93) rather than an error in plotting.
Because the nominal lower and upper limits of y30 and y0.1a,
respectively, for Eq. (24) with A5.5 and B2.5, coincide at a300, a
semilogarithmic regime presumably does not exist for any lesser value of a.
Equation (93), which implies the existence of a semilogarithmic regime for
Fic. 14. Representation of experimental data of Wei and Willmarth for (u'v') by Eq. (93).
(From Churchill and Chan [71], Figure 3.)
the velocity for all values of a, is therefore of questionable functionality for
intermediate values of y for a300 despite the reasonable representa-
tion in Figures 13 and 14.
The recent experimental data of Zagarola [73] for the time-averaged
velocity distribution as discussed in Section II, A, 10 suggest updating Eq.
(96) as
u6.13
1
0.436
lny} 6.824
y
a
`
5.314
y
a
`
. (97)
Equation (97) may be seen to be in agreement with Eq. (79) for ya
and with Eq. (77) for ya. For ya, Eq. (97) leads to Eq. (34) with
E10.264. Substituting for du/dy in Eq. (85) from the derivative of Eq.
(97), and then simplifying, results in the following updated version of Eq.
(95):
(u'v')
1
1
0.436y
1
0.436a
1
6.95y
a
. (98 )
Combination of Eq. (98) with Eq. (94), again with a combining exponent of
8/7 and the same mathematical contrivances, results in the nal correlat-
ing equation for (u'v'), namely
(u'v')
_
0.7
y
10
"
``
exp
1
0.436y
1
0.436a
1
6.95y
a
``
``
. (99)
As may be inferred from the detailed description of the formulation of Eq.
(93), the form of each of the three principal terms of Eq. (99) is speculative
and the values of the three numerical coefcients are subject to some
uncertainty. The most uncertain elements of Eqs. (93) and (99) are, however,
the form of Eq. (71) and the value of 8/7 for the combining exponent. On
the other hand, a virtue of correlating equations with this form is the
numerical insensitivity of their predictions to the value of the arbitrary
exponent.
Equation (99) differs from all prior expressions for the turbulent shear
stress, except for Eq. (93), by virtue of its presumed generality for all values
of y and all values of a300 (and perhaps even for a145), and its
incorporation of all of the known theoretical structure, namely, Eq. (32) for
y0, Eq. (35) for ya, and 1 B/y for the regime of overlap. It
supersedes all existing correlating equations for the eddy viscosity and the
mixing length. Although a correlating equation with the same generality
may be constituted for the eddy viscosity by the combination of Eqs. (99)
and Eq. (86), and for the mixing length by combination of Eqs. (99) and
(87), such expressions would not appear to serve any useful purpose.
3. New Correlating Equations for the Velocity Distribution
and the Friction Factor
Although numerical values for the velocity distribution and the friction
factor may be determined simply by evaluating the integrals of Eqs. (89) and
(91), respectively, using values of (u'v') from Eq. (99), generalized correlat-
ing equations for these two quantities may be constructed for convenience.
Churchill and Chan [71, 77] developed such correlating equations using
Eq. (93) for (u'v'). Their expression is not reproduced here, since these same
forms have recently been updated by Churchill [58] using Eq. (99) and the
experimental data of Zagarola [73]. The resulting nal expressions for u
and u
"
are
u
_
(y)`
1 yexp1.75(y/10)"}
`
_
1
0.436
ln
1 14.48y
1 0.301(e/a)a
6.824
y
a
5.314
y
a
`
`
(100)
and
u
"
3.30
227
a

50
a
`
1
0.436
ln
a
1 0.301(e/a)a
. (101)
Equation (100) has the form of Eq. (71) with u
"
adapted from the limiting
form of Eq. (33) with : 710" and [0, and u
adapted from Eq. (97).

The modied form of u
"
and the added value of unity in the argument of
the logarithm of u
are simply mathematical contrivances to avoid singul-

arities in ranges of y for which these terms do not contribute signicantly.
The coefcient of 14.48 corresponds to 6.13 in Eq. (79), that is, (1/
0.436) ln14.48} 6.13. The combining exponent of 3 was chosen by
Churchill and Chan [71] on the basis of various early sets of experimental
data (see their Fig. 1). The term 10.301(e/a)a was included in the
argument of the logarithmic term to extend the applicability of Eq. (100) to
commercial (naturally rough) piping, at least for ye. The coefcient of
0.301 as compared to 0.304 in Eq. (74) represents the slightly rened
expressions for the components for smooth and rough pipes. The leading
coefcient of 3.30 for Eq. (101) is adopted from Eq. (79) of Zagarola, but the
terms in (a) and (a)` are a necessary consequence of uy near
the wall, although such corrections have generally been overlooked. The
coefcients 227 and 50 are based on the numerical integration of Eq. (99)
since the experimental data of Zagarola do not encompass the regime of a
for which these terms are the most signicant. The term in (a)"" in Eq.
(80) is a purely empirical approximation for this behavior, and as was noted
in Section II, A, 10, is based on integration of Eq. (83) of Spalding. Again,
the term 10.301(e/a)a was incorporated in the argument of the logarith-
mic term to extend the applicability of Eq. (101) to encompass commercial
(naturally rough) piping.
Numerical integration of Eq. (90) using (u'v') from Eq. (99) predicts
values of the constant in Eq. (97) that vary from 5.39 to 5.76 with a, and
integration of Eq. (92) predicts values of the constant in Eq. (101) that vary
only slightly about 2.71. In this instance, the experimentally based values of
6.13 and 3.30 were given preference. The discrepancy between the predicted
and the experimentally determined leading constants is presumed to be due
to a slight underprediction of Eq. (99) in the regime of interpolation.
Churchill [58] compared the prediction of u
'
by Eq. (100) with the
experimental values of Zagarola and found an average absolute deviation of
only 0.17% and a maximum deviation of 0.39%. The deviations for small
and intermediate values of y were even less on the whole. The prediction
of u
"
by Eq. (101) was found to have an average absolute deviation of only
0.22%and a maximum deviation of 0.5%. In both instances these deviations
are less than those corresponding to the correlating equations of Zagarola
himself.
4. New Formulations and Correlating Equations for Other Geometries
Equation (85) with t/t
"
substituted for 1 (y/a) is applicable for all
one-dimensional fully developed turbulent ows. However, the variation of
t/t
"
with distance from a wall is known a priori only for forced ow in a
round tube and between identical parallel plates (both smooth or both
equally rough) and in planar Couette ow (induced by the movement of one
plate parallel and uniformly with respect to an identical one).
On the basis of the analogy of MacLeod, Eqs. (99) and (100) are
presumed to be directly applicable for forced ow between identical parallel
plates if b is simply substituted for a, while the friction factor may be
represented by
2
f
`
u
"
4.615
155
b

1
0.436
ln
b
1 0.301(e/b)b
. (102)
The value of 4.615 for the constant term as well as that of 155 for the
coefcient are based on computations by Danov et al. [85] using Eq. (99),
since experimental values of u
"
for parallel plates of accuracy and modern-
ity comparable to those of Zagarola for round tubes do not appear to exist.
Churchill [83], Chapter 3, constructed a theoretically based correlating
equation for u in planar Couette ow, for which t t
"
at all locations
within the uid. A corresponding correlating equation for (u'v') might be
postulated, but experimental data to test such an expression critically over
a wide range of conditions do not appear to be available.
Churchill [86] also constructed a generalized empirical expression from
which t/t
"
but not u or (u'v') may be estimated for forced ow through
a circular concentric annulus. The construction of such expressions for
combined forced and induced ow between parallel plates and for rotational
annular ow is not yet feasible because of the lack of appropriate data for
u'v' and/or u.
5. Recapitulation
The primary purposes in deriving Eqs. (99)(102) was to provide the basis
for the development of the corresponding expressions for forced convection
in a round tube and between parallel plates. In this regard, the path of their
derivation is as relevant as their nal form. However, over and above this
objective, these four expressions are presumed to be the most accurate and
comprehensive ones in the literature for turbulent ow, at least for a and
b300. Indeed, the rst of these, for the turbulent shear stress, has no
counterpart in the current literature, while Eq. (101) for the friction factor
in a round tube may be considered to be an improvement upon as well as
a replacement for all current graphical and algebraic correlations.
Although the structure of Eqs. (99)(102) represents the current state of
the art, and the constants, coefcients, and exponents therein are based on
the best available experimental data and computed values, these expressions
and values should all be considered to be subject to improvement on the
basis of future contributions, both theoretical and experimental.
III. The Quantitative Representation of Fully Developed
Turbulent Convection
The history and present state of predictive and correlative expressions for
the turbulent forced convection of energy in a round tube differ greatly from
those described previously for ow. One reason is the greater difculty in
characterizing the process of thermal convection experimentally. For
example, (1) the thermal conductivity and the viscosity both vary signi-
cantly with the primary dependent variable, the temperature of the uid,
forcing the use of small overall temperature differences, whereas the viscosity
does not vary with the velocity for ordinary uids; (2) the mixed-mean
temperature must be determined by integrating the product of the tempera-
ture and the velocity over the cross-section at a series of axial distances,
whereas the mixed-mean velocity is invariant with axial length and may be
determined externally and directly with a owmeter; (3) the heat ux
density, which is difcult to measure accurately, and/or the temperature
varies along the wall, whereas the velocity is zero at the wall and the shear
stress at the wall, which is ordinarily invariant with distance, may readily be
determined from a single measurement of the axial pressure drop; (4) the
heat ux density within the uid, which is almost impossible to measure
accurately, varies complexly with the distance from the wall and depends on
both the Reynolds number and the thermal boundary condition, whereas
the shear stress within the uid is known a priori to vary linearly with
radius; (5) the extent of the deviation from fully developed convection is
more difcult to determine than that for fully developed ow because the
latter condition is dened simply as a negligible change in the velocity
distribution and/or the axial pressure gradient with axial distance, whereas
the former is ordinarily dened as a negligible change in
T
"
T
T
"
T
"
or
T
"
T
T
"
T
'
and/or h
j
"
T
"
T
"
while T r}, T
'
, T
"
, and j
"
or T
"
are still varying individually; and nally (6)
the experiments must be repeated for a series of different uids encompass-
ing a wide range of values of the Prandtl number. Corresponding complex-
ities arise in modeling, as revealed in the sections that immediately follow.
These complexities and uncertainties appear to have inspired rather than
discouraged the development of purely empirical and semiempirical express-
ions for heat transfer since their number and variety far exceed those for
ow. Another difference is the focus of the work in ow and convection. In
many applications of ow, the velocity distribution is of equal or greater
interest than the friction factor, whereas in most applications of forced
convection interest in the heat transfer coefcient greatly exceeds that in the
temperature distribution.
A somewhat different order and scheme of presentation is followed for
turbulent convection than that for turbulent ow. The essentially exact
structure is rst examined in order to provide a framework and standard
for evaluation of the early work. Thereafter new developments are con-
sidered within this same framework as well as in terms of historical
precedents.
A. EssrN1i:iix Ex:c1 Foxii:1ioNs
1. New Differential Formulations
The general differential equation for the conservation of energy in a
moving uid with constant density, viscosity, and thermal conductivity may
be expressed in cylindrical coordinates as follows:
jc

cT
ct
u
'
cT
cr

u
r
cT
c0
cT
cz
k
_
1
r
c
cr
r
cT
cr
1
r`
c`T
c0
`

c`T
cz`
2j
cu
'
cr
`
_
1
r
cu
c0
u
'
cu
cz
`
cu
cz

1
r
cù
c0
`
`
cu
cr

cu
'
cz
`
_
1
r
cu
'
c0
r
c
cr
v
r
`
. (103)
The only new variables as compared to Eqs. (1)(3) are the temperature
T, the thermal conductivity k, and the specic heat capacity c
. Viscous
dissipation, as represented by the terms with the viscosity j as a coefcient,
is signicant only for very high velocities and for very viscous uids. Such
conditions and uids will not be considered herein. Hence these terms with
j as a coefcient will be dropped. The time- averaged form of the remaining
terms of Eq. (103) for steady, fully developed ow and fully developed
thermal convection may then be expressed as
jc
u
cT
cz

1
r
c
cr
kr
cT
cr
jc
rT 'v'
, (104)
where for consistancy with Eq. (13), the substitutions u u
and v u
'
have been made. Equation (104) may be integrated formally to obtain
jc
r ,
'
"
u
cT
cz
rdr k
cT
cy
j c
T 'v' . (105)
The terms k(dT /dy) and jc
T 'v' represent the heat ux densities in the

y-direction (negative-r-direction) due to thermal conduction and the turbu-
lent uctuations, respectively. The integral term on the left-hand side of Eq.
(105) represents the axial heat ux in the central core of uid with a radius
r. It may also be interpreted, by virtue of Eq. (105) itself, as the total heat
ux density j at r in the negative-r-direction. Equation (105) may also be
expressed in the dimensionless form
j
j
"
[1 (T 'v')]
dT
dy
, (106)
where
T k(T
"
T )(t
"
j)`/jj
"
(T 'v') jc
T 'v'/j
and
j
j
"
1
R ,
^
"
(cT /cx)
(cT
"
/cx)
u
u
"
dR`. (107)
This denition of T was chosen in order to achieve the same form for Eq.
(106) as for Eq. (85) and to result in T 0 at y0 in analogy to u0
at y0. The term (T 'v') is also analogous to (u'v') in the sense that
it is the local fraction of the heat ux density due to the turbulent
uctuations. Equation (107) was constructed by noting that, according to
Eq. (105), the total heat ux density at the wall may be expressed as
j
"
jc
a ,
"
u
cT
cz
rdr
jc
a
2 ,
"
u
cT
cz
dR`
jc
au
"
2
cT
"
cz
. (108)
Equation (108) may also be considered, as indicated, to dene the velocity-
weighted (mixed) mean of the longitudinal temperature gradient.
The only explicit difference between Eqs. (85) and (106) is that j/j
"
is given
by Eq. (107) in the latter whereas t/t
"
is simply equal to 1 (y/a) R
in the former. This difference is, however, a source of great complexity in the
expressions for convection. Another implicit difference is that although the
velocity is ordinarily postulated to be zero at the wall, a temperature varying
along the wall or a uniform or varying heat ux density along the wall may
be specied. Although T remains zero at the wall, T
"
itself may vary. The
thermal boundary condition thus becomes a parameter. An implicit differ-
ence of even greater signicance is the dependence of (T 'v') on a
parameter Pr c
j/k, called the Prandtl number, as well as on y and a.

It follows that T depends on Pr, and in general so does j/j
"
. This
parametric dependence is not avoidable in general simply by some other
choice of dimensionless variable, although it may vanish in certain narrow
regimes.
The measured values of T 'v' or T and j that are required to evaluate
(T 'v') are too limited in scope and accuracy to support the construction
of a generalized correlating equation in terms of y, a, and Pr comparable
to Eq. (98) for (u'v'). This deciency may be alleviated somewhat by
reexpressing Eq. (106) as
j
j
"
[1 (u'v')]
Pr
'
Pr

dT
dy
(109)
where
Pr
'
Pr

1 (T 'v')
1 (u'v')
(110)
Since (u'v') is implied to be known in advance, the net effect of this
substitution is to replace (T 'v') by Pr
'
/Pr as an unknown. The quantity
Pr
'
/Pr, as dened by Eq. (110), may be recognized in physical terms as the
ratio of the local fractions of the transport of energy and momentum by
molecular motion. This quantity suffers from the same uncertainties as
(T 'v') as well from the lesser ones associated with (u'v'), but has, as
will be demonstrated, a more constrained behavior for small and moderate
values of Pr. Therein lies its principal merit as a characteristic quantity. The
following alternative to both Eqs. (106) and (109) also has some advantages:
j
j
"
_
1
Pr
Pr
'

(u'v')
1 (u'v')
dT
dy
. (111)
Here
Pr
'
Pr

(u'v')
(T 'v')
1 (T 'v')
1 (u'v')
. (112)
The quantity Pr
'
/Pr, as dened by Eq. (112), may be recognized in physical
terms as the ratio of the transport of momentum by molecular and eddy
motions, divided by the equivalent ratio for the transport of energy.
Although Eq. (112) appears to be more complex than Eq. (110), and Eq.
(111) more complex than either Eq. (106) or (109), Pr
'
proves to be
essentially constant for large Pr, which results in a signicant simplication.
Elimination of (T 'v') between Eqs. (110) and (112) or of j/j
"
(dy/dT )
between Eqs. (109) and (111) results in
1
Pr
'
(u'v')
Pr
'
1 (u'v')
Pr
. (113)
This relationship between Pr
'
and Pr
'
will subsequently prove very useful.
Since j/j
"
differs only moderately from t/t
"
R, it is convenient to
introduce the variable , dened by
1
j
j
"

t
"
t
j/j
"
R

j/j
"
1 (y/a)
. (114)
Substituting for j/j
"
in Eq. (109) from Eq. (114) results in
(1 )
1
y
a
[1 (u'v') ]
Pr
'
Pr

dT
dy
. (115)
Comparison of Eqs. (115) and (85) indicates more explicitly the complica-
tions associated with convection than does comparison of Eqs. (106) and
(85). Substitution for j/j
"
from Eq. (111) in Eq. (114) results in
(1 )
1
y
a
_
1
Pr
Pr
'

(u'v')
1 (u'v')
dT
dy
. (116)
Equations (115) and (116) are the starting points for the subsequent exact
formulations for fully developed thermal convection.
Reichardt [87] in 1951 was apparently the rst to propose as a
correlative quantity. Rohsenow and Choi [88] in 1961 subsequently sugges-
ted the use of M1 as an alternative quantity for correlation. Although
the effects represented by and M are generally signicant, as will be shown
on the basis of experimental and computed values, they have been over-
looked or ignored in many analyses of convection.
2. New Integral Formulations
Equation (115) may be reexpressed in terms of R and then integrated
formally to obtain the following exact expression for the temperature
distribution:
T
a
2 ,
(1 ) [1 (u'v')]
Pr
'
Pr
dR`. (117)
Integration of this expression for T , weighted by u/u
"
, over the cross-
section of the pipe then gives the following expression for the mixed-mean
temperature
T
"

a
2 ,
"
_,
^
(1 )[1 (u'v')]
Pr
'
Pr
dR`

u
u
"
dR`, (118)
from which it follows that
Nu
j
"
D
k(T
"
T
"
)

2a
T
"
4
,
"
_,
^
(1 )[1 (u'v')]
Pr
'
Pr
dR`
u
u
"
dR`
. (119)
The quantity Nu, called the Nusselt number, may be interpreted as the
dimensionless rate of convective heat transfer. The corresponding express-
ions for T and Nu in terms of Pr
'
rather than Pr
'
are
T
a
2 ,
^
(1 )dR`
1
Pr
Pr
'

(u'v')
1 (u'v')
(120)
and
4
Nu
2T
"
a

,
"
_
,
^
(1 )dR`
1
Pr
Pr
'

(u'v')
1 (u'v')

u
u
"
dR`. (121)
Equations (117)(121) are the nal, general integral formulations herein for
the temperature distribution, mixed-mean temperature, and Nusselt num-
ber. However, some reductions are possible for particular boundary condi-
tions and particular values of the Prandtl number as described in the
immediately following sections.
It may be inferred that the analytical or numerical evaluation of T by
means of Eqs. (117) or (120) requires (u'v') as a function of y and a,
and and Pr
'
or Pr
'
as a function of Pr and the thermal boundary condition
as well as of y and a. The evaluation of T
"
by means of Eqs. (118) or
(120), and Nu by means of Eqs. (119) or (121), may further be inferred to
require a relationship for u as a function of y and a and u
"
as a function
of a. However, these requirements may be relaxed somewhat. First, u and
u
"
are given exactly as integral functions of (u'v') by Eqs. (89) and (91),
respectively, and approximately but probably with sufcient accuracy for all
practical purposes by Eqs. (100) and (101), respectively. Yahkot et al. [89]
assert, although they do not prove, that Pr
'
is a universal function of j
'
/j
and Pr for all geometries and boundary conditions. By virtue of Eq. (86),
this generality, if valid, must extend to Pr
'
as a function of (u'v') and Pr.
Although this assertion of Yahkot et al. has been implied in a number of
analyses, Abbrecht and Churchill [22] appear to have provided the only
experimental conrmation. They found Pr
'
to be invariant with axial
distance in developing thermal convection following a step in wall tempera-
ture in fully developed turbulent owa severe test of independence from
the thermal boundary condition. Their results for a round tube were also
found to agree closely with those of Page et al. [90] for heat transfer from
a plate at one uniform temperature to a parallel one at a different uniform
temperature for ows at the same values of a and ba severe test of
independence from geometry as well as from the thermal boundary condi-
tion. Thus, the evaluation of T , T
"
, and Nu only requires (u'v') as a
function of y and a, Pr
'
or Pr
'
as a function of (u'v') and Pr, and as
a function of y, a, Pr and the thermal boundary condition. The depend-
ence of on Pr vanishes under some circumstances. Finally, as will be
shown, the relationship for , although quite complex, is known exactly,
whereas those for Pr
'
and Pr
'
are highly uncertain both theoretically and
experimentally.
a. A Uniform Heat Flux Density from the Wall As noted in the rst
paragraph of Section III, fully developed convection is ordinarily dened as
the attainment of essentially unchanging values of
T
"
T
T
"
T
"
or
T
"
T
T
"
T
'
and/or of h j
"
/(T
"
T
"
)
with axial distance. Although the exact point of this attainment is ill dened,
the concept is a useful one in both an analytical and an applied sense. The
majority of the theoretical semitheoretical solutions and correlations in the
literature for the Nusselt number in turbulent ow are for this regime, which
prevails or is closely approached over most of the length of ordinary
industrial heat exchangers.
A uniform heat ux density from the wall to the uid may be attained
approximately by passing an electrical current axially through the metal
wall of a heat exchanger, which thereby functions as an electrical resistance.
Small deviations from uniform heating of the uid may then occur because
of end effects, for example, thermal conduction along the tube wall or
nonuniform heat losses to the surroundings. A uniform heat ux density
from the wall to the uid may also be closely approached in the inner pipe
of a concentric circular double-pipe heat exchanger operated in equal
countercurrent ow. In this case small deviations may be expected due to
variations in the local overall heat transfer coefcent as a consequence of
entrance effects in ow and the variation of the physical properties of the
two uids with temperature. If the heat transfer coefcent h j
"
/(T
"
T
"
)
approaches an asymptotic value with axial distance for a uniform heat ux
density, T
"
T
"
must as well. Then, if (T
"
T )/(T
"
T
"
) approaches an
asymptotic value,
c
cz
T
"
T
T
"
T
"
1
(T
"
T
"
)
cT
"
cz

cT
cz
0 (122)
and
c
cz
(T
"
T
"
)
cT
"
cz

cT
"
cz
0. (123)
Hence, for fully developed convection with uniform heating,
cT
cz
5
cT
"
cz
5
cT
"
cz
. (124)
Then from Eqs. (107) and (114)
1
1
R` ,
^
"

u
u
"
dR`. (125)
By virtue of Eq. (125), Eq. (118) may be integrated by parts to obtain
8
Nu
4a
T
"
"
(1 )`[1 (u'v')]
Pr
'
Pr
dR". (126)
The analog of Eq. (126) in terms of Pr
'
is readily shown to be
8
Nu
4T
"
a

,
"
(1 )`dR"
1
Pr
Pr
'

(u'v')
1 (u'v')
. (127)
The evaluation of Nu from Eqs. (126) or (127) appears to involve only a
single integration. However, the quantity must be evaluated by integration
for each value of a, as indicated by Eq. (125). This latter relationship may
be expressed directly in terms of (u'v') by substituting for u and u
"
from
Eqs. (89) and (91), respectively, integrating by parts, and simplifying to
obtain
1R`
R` ,
^`
"
[1(u'v')]dR"
,
^`

1 R`
R`
[1 (u'v')]dR"
,
"
[1 (u'v')]dR"
.
(128)
The behavior of for two special cases is worthy of note. From Eq. (125)
it is apparent that is zero for all values of y only for the hypothetical case
of plug ow. On the other hand, it is apparent from Eqs. (125), (89), and
(91) that for R0, for which u approaches a nearly constant value,
1
u
'
u
"
"
[1 (u'v')]dR`
,
"
[1 (u'v')]R`dR`
. (129)
Equation (129), which may also be derived directly but by a considerably
longer path from Eq. (128), denes the maximum value of for each value
of a and thereby characterizes the magnitude of the deviation of j/j
"
from
t/t
"
1/R.
Equations (117), (125), and (126), together with Eq. (128), constitute the
nal exact and completely general formulations herein for fully developed
turbulent convection in a uniformly heated round tube. Their numerical
evaluation requires only an expression such as Eq. (99) for (u'v') and one
for Pr
'
or Pr
'
, presumably as a function only of (u'v') and Pr. [Actually,
Eqs. (117), (126), (127), and (128) are also applicable for fully developed
laminar convection with uniform heating as well. For this case, (u'v')0,
Eq. (113) gives Pr/Pr
'
1, Eq. (128) gives 1 R`, and both Eqs. (126)
and (127) give Nu 48/11.]
In the limit of Pr 0, Eq. (113) reduces to
Pr
Pr
'
1 (u'v'). (130)
Substitution of this expression in Eq. (117) gives
T Pr 0}
a
2 ,
^
(1 )dR`
a
2
(1 R`)(1
mR`
), (131)
where
"^
is seen to be the integrated-mean value from R` to 1. Since is
nite and positive for all R1 for both laminar and turbulent ow, the
term
"^
is nite and positive as well. Similarly, substitution of
Pr
Pr
'
from
Eq. (130) in Eq. (126) gives
8
NuPr 0}
4T
"
Pr 0}
a

,
"
(1 )`dR" (1 )`
"^`
, (132)
where (1 )`
"^`
is seen to be the integrated mean of (1 )` over R" from
0 to 1.0. Equation (131) provides an upper bound for T /a and Eq. (132)
a lower bound for Nu for all Pr as a function of a. In all of these
expressions, 1 represents the effect of the deviation of j/j
"
from t/t
"
(which has often been neglected) while (1 )` includes the effect of the
velocity distribution as well.
Insofar as Pr
'
approaches a nite value as y0, the corresponding
asymptotic solution may be derived for Pr . For this case, the entire
temperature development takes place within the viscous boundary layer
where y/a may be neglected, 0, and (u'v') may be approximated by
the rst term on the right-hand side of Eq. (32). Equation (116) thereby
reduces to
dT
dy

1 :(y)`
1
Pr
Pr
'
1
:(y)`
. (133)
The function on the right-hand side of Eq. (133) may be integrated
analytically if Pr
'
is postulated to be invariant with respect to y. With the
boundary condition u0 at y0, the result is
T

Pr
Pr
'
3:`
Pr
Pr
'
1
"` _
1
2
ln
(1 z)`
1 z z`
3` tan
2z 1
3`
3`
6
y
Pr
Pr
'
1
, (134)
where
z :`
Pr
Pr
'
1
`
y.
Near the wall for very large values of Pr, the last term on the right-hand
side of Eq. (134) may be dropped. Finally, letting z gives the following
expression for the fully developed temperature, which differs negligibly from
the mixed-mean temperature, and thereby:
T
5T
"

2a
Nu

2
Pr
Pr
'
3``:`
Pr
Pr
'
1
"`
. (135)
For : 710", it follows that
Nu0.07343
1
Pr
'
Pr
"`
Pr
Pr
'
`
Re
f
2
`
0.07343
Pr
Pr
'
`
Re
f
2
`
(136)
The more general form of Eq. (136) was apparently rst derived by
Churchill [91], but the equivalent of the limiting form, usually with Pr
'
postulated to be unity, was derived much earlier by Petukhov [92] and
others. The utility of the term
1
Pr
'
Pr
"`
is in providing a rst-order correction for the effect of a nite value of Pr
and conversely of dening the lower limit of applicability of this limiting
form with respect to nite values of Pr. It may be inferred from the absence
of a and that Eqs. (134)(136) are applicable for fully developed turbulent
convection in any fully developed shear ow and for any thermal boundary
condition, not just for a uniformly heated round tube. As shown subsequent-
ly, the one speculative element in the derivation of Eqs. (134) and (135),
namely the attainment of a nite asymptotic value for Pr
'
as y0 and Pr
increases, is supported by some sets of experimental data and direct
numerical simulations but is contradicted by others.
The postulate that Pr
'
Pr requires, by virtue of Eq. (113), that Pr
'
Pr
as well. Insofar as Pr
'
Pr for all y, Eq. (117) reduces to
T Pr
'
Pr
'
Pr}
a
2 ,
^
(1 )[1 (u'v')]dR`, (137)
which, by virtue of Eq. (89), may be expressed as
T Pr
'
Pr
'
Pr} u(1
""^
) (138)
where
""^
is the integrated mean of , weighted with respect to 1(u'v')
over R` from R` to 1.0. The deviation of the T y, a} from uy, a}
for Pr
'
Pr is seen to be wholly a consequence of the factor 1 and thus
wholly due to the deviation of j/j
"
from t/t
"
. The similarity of the
distribution of T to that of u, as represented by Eq. (138), is one reason
for the arbitrary denition of T herein. The postulate of Pr
'
Pr
'
Pr
for all y allows the reduction of both Eqs. (126) and (127) to
8
NuPr
'
Pr
'
Pr}
4 T
"
Pr
'
Pr
'
Pr}
a

,
"
(1 )`[1(u'v')]dR".
(139)
Comparison of Eqs. (139) and (91) reveals the following similarity for T
"
and u
"
:
T
"
Pr
'
Pr
'
Pr} u
"
(1 )`
""^`
(140)
Here (1 )`
""^`
is the integrated mean of (1 )`, weighted with respect
to 1 (u'v'), over R" from 0 to 1.0. It follows that
NuPr
'
Pr
'
Pr}
2a
u
"
(1 )`
""^`
Re( f /2)
(1 )`
""^`
. (141)
Equation (141) is a surprising and remarkable result. It has the same explicit
functional dependence on ow as the famous analogy of Reynolds [18],
namely
Nu Re
f
2
Pr, (142)
but occurs at Pr
'
Pr
'
Pr instead of Pr 1 and differs by the factor
(1 )`
""^`
. The speculative element upon which the derivation of Eqs.
(137)(141) is based, namely the invariance of Pr
'
with y for the particular
value of Pr
'
Pr
'
Pr, has some experimental and semitheoretical support
for Pr 5 0.87, in particular over the turbulent core. The observed behavior
of Pr
'
and Pr
'
in the viscous sublayer and the buffer layer is not necessarily
contradictory, just uncertain.
Despite the indicated uncertainties with regard to Eqs. (136) and (141),
these two expressions, together with Eq. (132), prove to be invaluable in
evaluating approximate and speculative formulations and solutions and in
constructing generalized correlating equations.
b. A Uniform Wall Temperature Next to uniform heating, the most
frequently postulated thermal boundary condition in analytical formula-
tions for convective heat transfer in a round tube is a uniform temperature
on the wall, higher (or lower) than that of the entering uid. This boundary
condition is closely approximated in real exchangers cooled or heated on
the outer surface of the tubes(s) by a boiling liquid or condensing vapor,
respectively. Deviations from a uniform wall temperature may occur as a
result of a nite value of the outer heat transfer coefcient and of end effects.
For a uniform wall temperature, fully developed convection may be charac-
terized by
c
cz
T
"
T
T
"
T
"
cT
cz
T
"
T
"
(T
"
T )
cT
"
cz
(T
"
T
"
)`
0, (143)
which implies that
cT/cz
cT
"
/cz
T
"
T
T
"
T
"
T
T
"
. (144)
Substituting this expression in Eq. (107) and then that result in Eq. (114)
leads to
1
1
R` ,
^
"
T
T
"

u
u
"
dR`
1
R`T
"
u
"
,
^
"
T udR`. (145)
Seban and Shimazaki [93] were apparently the rst to identify the equival-
ent of Eqs. (143) and (144) as characterizing convection with a uniform wall
temperature.
It is apparent from Eq. (145) that for a uniform wall temperature, as
contrasted with a uniform heat ux density from the wall, is nite even for
the hypothetical case of plug ow. Furthermore, the maximum value of
1 may be inferred from Eq. (145) to be equal to (T
'
/T
"
)(u
'
/u
"
) and
therefore greater than for uniform heating by the factor T
'
/T
"
. It follows
that the error due to neglecting is greater for a uniform wall temperature.
Equations (117)(121) are directly applicable for a uniform wall tempera-
ture, but because of the dependence of on T , as expressed by Eq. (145),
an iterative process of solution is required. For example, for a specied value
of Pr, a correlating equation for Pr
'
/Pr and an arbitrary postulated
expression
"
y} for y}, T y} may be calculated from Eq. (117), T
"
from Eq. (118), and then
y} from Eq. (145). These calculations are

repeated, starting with
1
y}, and continued until convergence is achieved.
Now consider the three special cases of Pr 0, Pr , and
Pr
'
Pr
'
Pr for a uniform wall temperature. For Pr , Eqs. (133)
(136), which are independent of , remain directly applicable. For Pr 0,
must be determined iteratively from Eq. (145) using T from Eq. (131) and
T
"
from the following reduced form of Eq. (118):
T
"
Pr 0}
a
2 ,
"
_,
(1 )dR`

u
u
"
dR`. (146)
Similarly, for Pr
'
Pr
'
Pr, must be determined iteratively from Eq.
(144) using T from Eq. (136) and T
"
from the following reduced form of
Eq. (118):
T
"
Pr
'
Pr
'
Pr}
a
2 ,
"
_,
^
(1 )[1 (u'v')]dR`
u
u
"
dR`.
(147)
c. Generalized Expressions An alternative form of expression for T
"
and
Nu is useful for interpretation if not for numerical evaluations. Setting R0
in the lower limit of the integral of Eq. (117) results in the following
expression for the temperature at the centerline:
T
'

a
2 ,
"
[1 ][1 (u'v')]
Pr
'
Pr
dR`. (148)
From this it follows that
Nu
2a
T
"
2a
T
'

T
'
T
"
4(T
'
/T
"
)
,
"
[1 ][1 (u'v')]
Pr
'
Pr
dR`
. (149)
For Pr 0, Eq. (149) reduces to
NuPr 0}
2a
T
"
Pr 0}
4(T
'
/T
"
)
,
"
[1 ]dR`
4(T
'
/T
"
)
(1 )
"^
, (150)
where here (1 )
""^
is the integrated mean of 1 over R` from 0 to 1,
whereas for Pr
'
Pr
'
Pr Eq. (150) reduces to
NuPr
'
Pr
'
Pr}
4(T
'
/T
"
)
,
"
(1 )[1 (u'v')]dR`
T
'
T
"

u
"
u
'

Re
f
2
(1 )
""^
,
(151)
where (1 )
""^
is the integrated mean of 1, weighted by [1(u'v')],
over R` from 0 to 1.0. The factors T
'
/T
"
and u
"
/u
'
may be expected to
compensate for each other to some extent, although T
'
/T
"
is always larger.
Equations (149)(151) do not have any merit relative to Eqs. (119), (121),
(126), (127), (132), (139), and (141) as far as numerical calculations are
concerned because of the presence of T
'
/T
"
. However, these formulations
will be shown subsequently to be invaluable in terms of constructing a
theoretically based correlating equation.
As mentioned previously, the factor (1 ) represents in all cases the
effect of the deviation of the heat ux density ratio from the shear stress
ratio, while the factor (1 )` represents the effect of the velocity distribu-
tion as well. Equations (148)(151) are applicable for both uniform heating
and uniform wall temperature. This approach does not result in an alterna-
tive expression for NuPr } since the postulate that T
'
T
"
is
inherent for that limiting case. It may be inferred that the effects of these two
thermal boundary conditions are exerted wholly through (T
'
/T
"
)/
(1 )
""^
for Pr 0 and
T
'
/T
"
u
'
/u
"

(1 )
""^
for Pr
'
Pr
'
Pr.
3. Parallel Plates and Other Geometries
Insofar as the analogy of MacLeod is applicable for (u'v') and u, all
of the previous expressions in Section III for T and dT /dy are directly
applicable for fully developed convection from parallel plates heated equally
on both surfaces (either uniformly or isothermally) if a is simply replaced
by b and R by Z1 (y/b) wherever they appear. The expressions for
and for T
"
are, however, different because they invoke integrations over
a planar rather than a circular area. As an example, for equal uniform
heating on both plates, the expression analogous to Eq. (126) is
12
Nu
"'
3T
"
b

,
"
(1 )`[1 (u'v')]
Pr
'
Pr
dZ`. (152)
where here, as contrasted with Eq. (128),

1 Z
Z ,
"
[1 (u'v')]dZ`
,

1 Z
Z
[1 (u'v')]dZ`
,
"
[1 (u'v')]dZ`
.
(153)
Again, as for a round tube, increases monotonically from 0 at Z0 to
u
'
/u
"
at Z1.
For Pr 0, Eq. (152) reduces to
Nu
"'
Pr 0}
12
,
"
(1 )`dZ`
12
(1 )`
"
, (154)
where (1 )`
"
is the integrated-mean value over Z`. On the other hand,
for Pr Pr
'
Pr, Eq. (152) reduces, by analogy with Eq. (141), to
Nu
"'
Pr Pr
'
Pr
'
}
12
,
"
(1 )`[1 (u'v')]dZ`
Re
"'

f
2
(1 )`
""
,
(155)
where (1 )`
"
is the integrated-mean value, weighted by [1 u'v'],
over Z`. Equation (136) for Pr is directly applicable in terms of Nu
"'
and Re
"'
.
For parallel plates at different uniform temperatures, j is uniform across
the channel and 0. It follows that
T
,
'
"
[1 (u'v')]
Pr
'
Pr
dZ (156)
and that
Nu
'
j
"
b
k(T
"
T
'
)

b
T
'
1
,
"
[1 (u'v')]
Pr
'
Pr
dZ
. (157)
For Pr 0, Eq. (157), by virtue of Eq. (130), reduces to simply
Nu
'
Pr 0} 1. (158)
The half-spacing b was chosen as the characteristic dimension in order to
achieve this particular, obvious result. For Pr Pr
'
Pr
'
, Eq. (157) re-
duces to
Nu
'
1
,
"
[1 (u'v')]dZ
1
1 (u'v')]
"
, (159)
where [1 (u'v')] is the integrated-mean value over the channel. For
Pr , Eq. (136) is directly applicable in terms of Nu
'
and Re
'
.
Expressions for equal uniform wall-temperatures may readily be for-
mulated by analogy to those for an isothermal round tube in Section III, A,
2, b. but are not included here in the interests of brevity. Expressions for
parallel-plate channels analogous to Eqs. (149)(151) are also omitted, even
though they are referred to subsequently, since their form is readily inferred.
Equivalent formulations for fully developed convection are possible for all
one-dimensional ows, but their implementation is dependent upon individ-
ual expressions for (u'v'), as discussed in II, B, 4, and in turn for .
4. Alternative Models and Formulations
None of the differential models that have been proposed in the past for
the heat transfer in turbulent ow appear to provide any improvement over
the dimensionless turbulent heat ux density, (T 'v'), or its exact equival-
ents in terms of the dimensionless turbulent shear stress, (u'v'), and Pr
'
or Pr
'
. However, the eddy conductivity, k
'
, and its implementation are
described here in some detail because of the widespread use of this quantity
or its exact equivalent, :
'
k
'
/jc
, the thermal eddy diffusivity, for correla-

tion and prediction in the past and present literature. The eddy conductivity
itself may be dened by
k
'
dT
dy
jc
(T 'v') (160)
and incorporated in the elementary differential energy balance to obtain, in
dimensionless form,
j
j
"
1
k
'
k
dT
dy
. (161)
Elimination of dT /dy between Eqs. (161) and (106) reveals that
k
'
k

(T 'v')
1 (T 'v')
. (162)
Equations (160)(162) are directly analogous to Eqs. (37), (38), and (86),
respectively, for momentum transfer. Since, from physical considerations,
(T 'v') must be greater than zero and less than unity at all locations within
the uid in a round tube, k
'
may be inferred to be positive, bounded, and
interchangeable with (T 'v') in this geometry.
Equation (161) has often been expanded as
j
j
"
_
1
c
j
k
k
'
c
j
'
j
'
j
dT
dy

_
1
Pr
Pr
'

j
'
j
dT
dy
(163)
or as
j
j
"
_
c
j
k
k k
'
c
(j j
'
)
jj
'
j
dT
dy

Pr
Pr
'

j
'
j
dT
dy
. (164)
where here Pr
'
c
j
'
/k
'
, Pr
'
c
(j j
'
)/(k k
'
), j
'
j j
'
, and
k
'
k k
'
. These denitions of Pr
'
and Pr
'
are consistent in every respect
with those of Eqs. (112) and (110), respectively. Such transformations were
of course initially made with the expectation that Pr
'
/Pr or Pr
'
/Pr would
be more constrained in its behavior than k
'
/k.
Equation (163) may be integrated formally to obtain
T
,
'
"

j
j
"
dy
_
1
Pr
Pr
'

j
'
j
, (165)
and then T from Eq. (165), weighted by u/u
"
, may be integrated formally
over the cross-section of the round tube to obtain
2a
Nu
T
"

,
"
_
,
'
"

j
j
"
dy
_
1
Pr
Pr
'

j
'
j

u
u
"
dR`. (166)
For uniform heating it has been the custom, when utilizing the eddy
conductivity ratio k
'
/k or its equivalent such as (Pr/Pr
'
)(j
'
/j), to substitute
from Eq. (125) for j/j
"
, thereby transforming Eq. (165) to
T
a
^
_,
^
"

u
u
"
dR`
dR
R
_
1
Pr
Pr
'

j
'
j
(167)
and Eq. (166) to
2
Nu
T
"
a

,
"
^
_,
^
"

u
u
"
dR`
dR
R
_
1
Pr
Pr
'

j
'
j

u
u
"
dR`.
(168)
Lyon [94] in 1951 recognized that changing the order of integration allows
reduction of Eq. (168) to
Nu
2
,
"
_,
^
"

u
u
"
dR`
` dR
R
_
1
Pr
Pr
'

j
'
j
, (169)
which requires far less computation for numerical evaluations than does the
triple integral of Eq. (168). The analogous reduction of Eq. (119) to (126)
was much simpler and more obvious because of the use of (u'v') rather
than both u and j
'
/j as variables. Lyon further recognized that setting
Pr 0 in Eq. (169) gives an expression for the lower limiting value of Nu
that varies with a (or Re) only by virtue of the variation in the velocity
distribution. He further inferred (incorrectly, as will be shown) that this
limiting value is approached asymptotically as RePr approaches zero.
For a uniform wall temperature, Eqs. (165) and (166) become, by virtue
of Eq. (145),
T
a

,
^
_,
^
"
T
T
"

u
u
"
dR`
dR
R
_
1
Pr
Pr
'

j
'
j
(170)
and
Nu
2
,
"
^
_,
^
"
T
T
"

u
u
"
dR`
dR
R
_
1
Pr
Pr
'

j
'
j
u
u
"
dR`
.
(171)
Equations (165)(171) may be expressed in terms of Pr
'
rather than Pr
'
simply by replacing
1
Pr
Pr
'
j
'
j
by
Pr
Pr
'
j
'
j
.
The preceding expressions in terms of j
'
, u, and Pr
'
or Pr
'
are exact but
more cumbersome than the corresponding ones in terms of (u'v'), , and
Pr
'
or Pr
'
. An expression for j
'
/j could be derived from a correlating
equation for u by virtue of Eq. (38), but in most applications, for some
unexplained reason, separate, incongruent correlating equations have been
used for u and j
'
. Despite appearances to the contrary, the use of (u'v')
rather than j
'
/j does not decrease the number of integrations to determine
values of T and Nu; the integration or integrations of u/u
"
are simply
performed separately in the process of evaluating .
The c and u'v' T 'v' models do not appear to have a useful role for
convection in round tubes or parallel-plate channels, but the latter one has
promise for circular annuli despite the considerable empiricism involved in
the implementation of the supplementary equations of transport.
B. EssrN1i:iix Ex:c1 Nixric:i Soii1ioNs
The integral formulations of Section III, A are exact, except possibly the
reduced ones for the special case of Pr Pr
'
Pr
'
, which incorporate the
postulate of invariance of Pr
'
and Pr
'
with y. In addition, the closed-form
solution for Pr is subject to the asymptotic attainment of a nite value
for Pr
'
as y0. Some uncertainty arises in the numerical evaluation of the
integral expressions for Nu for all nite values of Pr by virtue of the
empirical expression, such as Eq. (99), that is used for (u'v'), but the net
effect is presumed to be completely negligible. On the other hand, the
uncertainty introduced by the expressions utilized for Pr
'
or its equivalent
[Pr
'
, k
'
, k
'
or (T 'v')] is potentially very signicant. The uncertainty
associated with expressions for Pr
'
or its equivalent extends to all prior
numerical results for turbulent convection, other than those from direct
numerical simulations, as well to those subsequently presented herein. The
estimation of values of Pr
'
or its equivalent is therefore given rst attention
in this section.
1. Expressions for the Turbulent or Total Prandtl Number
As noted heretofore, Pr
'
is presumed on the basis of theoretical conjec-
tures, as well as experimental evaluations, to be the same unique function of
(u'v') (or j
'
/j) and Pr for all geometries and all thermal boundary
conditions. This presumption, which has been overlooked or denied impli-
citly by many prior investigators, greatly simplies the task of correlation
for Pr
'
as well as the integrations for T and Nu.
The high degree of uncertainty of the various expressions for Pr
'
arises on
the one hand from the very severe requirements for precision in the
measurements of either T 'v' or dT/dy, and on the other hand from the lack
of a universally accepted theoretical model. A vast but generally disappoint-
ing body of literature exists on this subject (see, for example, Reynolds [95]
and Kays [95a]). Only a few directly relevant contributions will be noted
here.
Jischa and Rieke [96] and others have successfully correlated the exten-
sive data for the turbulent core for uids with Pr 0.7 by means of a simple
algebraic expression, such as
Pr
'
0.85
0.015
Pr
. (172)
Over the purported range of validity of Eq. (172), the turbulent Prandtl
number is predicted to vary only from 0.85 to 0.87 and to be independent
of y and a [or (u'v')]. Such constrained behavior, insofar as this
prediction is valid, appears to justify the use of Pr
'
rather than (T 'v') or
k
'
/k or even Pr
'
as a variable for correlation. The nominal restriction of Eq.
(172) to the turbulent core may be attributed in part to the widespread
scatter of the experimental data for Pr
'
in the viscous sublayer and the buffer
layer rather than wholly to its inapplicability in those regimes. Kays [95a]
proposed the extension of Eq. (172) for small values of Pr by replacing the
constant 0.85 by Aj
'
/j with a value for A of 0.7 based on direct numerical
simulations or 2.0 based on experimental data. Because of the excessive
values of Pr
'
predicted by this modication of Eq. (172) very near the wall,
he proposed to set Pr
'
1 in that region.
A more complex empirical expression is that of Notter and Sleicher [97],
which may be rewritten in terms of (u'v') rather than j
'
/j as follows:
Pr
'
1 90Pr``
u'v')
1 (u'v')
"
1
10
35
(u'v')
1(u'v')
0.025 Pr
u'v')
1(u'v')
90Pr``
u'v')
1(u'v')
"
.
(173)
Equation (173) predicts an asymptotic value of Pr
'
`
"
0.778 as y0
for large values of Pr, which supports the critical postulate in the derivation
of Eq. (136). However, it predicts higher values than Eq. (172) for the
turbulent core and values of Pr Pr
'
that depend slightly on (u'v') even
in the turbulent core, which is not in accord with the critical postulate in
the derivation of Eq. (141). Yahkot et al. [89] used renormalization group
theory to derive
1.1793
1
Pr
'
1.1793
1
Pr

"'`
2.1793
1
Pr
'
2.1793
1
Pr

"``
1 (u'v') (174)
Equation (174) is implied by the authors to be applicable for all geometries,
all thermal boundary conditions, and all values of Pr and (u'v'). Further-
more, they assert that this expression is free of any experimentally adjusted
parameters. However, they undermine its credibility somewhat by suggest-
ing, in a footnote added in proof, the change of a theoretical index in their
derivation from 7 to 4, which appears to have signicant numerical
consequences. The dependence of Pr
'
on the rate of ow and on location
within the uid stream may be inferred from Eq. (174) itself to be charac-
terized wholly by (u'v'). The dependence of Pr
'
on Pr and (u'v') only,
if valid, must extend to Pr
'
by virtue of Eq. (113). The limited experimental
support for these various presumptions has already been discussed in the
paragraph following Eq. (121). In any event, Eq. (174) is attractive in terms
of simplicity and purported generality, and its predictions appear to be
qualitatively if not quantitatively correct. For example, it predicts
Pr Pr
'
Pr
'
over the entire cross-section of ow for Pr 0.848, but on
the other hand an obviously low value of Pr
'
0.39 at the wall for
asymptotically large values of Pr. Elperin et al. [97a] showed that only one
of the constants and exponents of Eq. (174) is independent, and determined
an improved value thereof. However, the latter value does not eliminate
the indicated shortcomings.
The nal recent contribution to be examined here is that of Papavassiliou
and Hanratty [98], who used both Lagrangian and Eulerian direct numeri-
cal simulations to predict Pr
'
for heat transfer between parallel plates for a
series of values of Pr from 0.05 to 2400, but only for b150, which is just
above the minimum value for fully developed turbulence. The title of the
previously cited paper by Einstein [4] on Brownian motion does not suggest
any relevance to turbulent ow and convection, but its statistical develop-
ment serves as the origin of the Lagrangian DNS methodology of Papavas-
siliou and Hanratty. Their predictions appear to be in fair qualitative and
quantitative agreement with Eqs. (172)(174) for most conditions, but for
Pr 100 they indicate an increase without limit in Pr
'
as y0. This latter
prediction, as represented (in thermal terms) by
Pr
'
Pr

1.71
(y)"``
, (175)
is in accord with the measurements by Shaw and Hanratty [99] of the rate
of electrochemical mass transfer. This result would appear to refute the
validity of Eq. (136), but since convective turbulent heat transfer is usually
limited to uids with Pr 100, Eq. (136) may remain applicable as an
asymptotic element of a correlating equation as long as the latter is not
utilized for higher values.
In view of the contradictions among these representative results, the
principal challenge in turbulent convection appears to be the resolution of
the uncertainties in the qualitative and quantitative dependence of Pr
'
on
(u'v') (or y and a) and Pr. In the following section the effects of this
uncertainty in Pr
'
are avoided insofar as possible by choosing particular
conditions for which its impact is minimal.
Substitution of Pr
'
from Eq. (172) or (173) and of (u'v') from Eq. (99)
in Eq. (112) would yield a direct analog of (u'v') for convection, that is,
an algebraic expression for (T 'v') as a function of y, a, and Pr.
Substitution of Pr
'
from Eq. (110) and again of (u'v') from Eq. (99) in
Eq. (174) or its alternatives would yield much more complex algebraic
expressions for (T 'v'), again as a function of y, a, and Pr. Such
expressions for (T 'v') have not been presented herein because they would
obviously be less convenient to apply than those for Pr
'
and Pr
'
. It is
obvious that such expressions for (T 'v') would incorporate the consider-
able uncertainties of the generating expressions for Pr
'
and Pr
'
as well as
the lesser ones attributable to Eq. (99) and its components.
2. Numerical Results for Nu for a Uniformly Heated Tube
a. Solutions for Particular Conditions Consideration is rst given to those
conditions for which the dependence of Nu on the uncertainty of the values
of Pr
'
is absent or minimal. Heng et al. [100] were the rst to use the new
formulations herein for numerical evaluations of Nu but the later evalu-
ations of Yu et al. [100a] are presented herein since they are presumed to
be slightly more accurate.
For Pr 0, the Nusselt number, as given by Eq. (132), is independent of
both Pr and Pr
'
. Values of Nu, and T
'
/T
"
for a500, 1000, 2000, 5000,
and 10,000, as computed using Eq. (99) for (u'v') and in turn Eq. (128)
for , are listed in Table I along with values for laminar and plug ow. The
TABLE II
Coxii1rn Cn::c1ris1ics or Fiiix Drvrioirn TiniirN1 Fiov 1noicn : RoiNn Tinr
(from Yu et al. [100a])
a u
"
Re10` u
'
/u
"
f 10` Re( f /2)
500 17.0 17.0 1.2696 6.920 58.82
1000 18.815 37.63 1.2375 5.650 106.3
2000 20.518 82.07 1.2148 4.751 194.96
5000 22.69 226.9 1.1926 3.885 440.72
10,000 24.295 485.9 1.1793 3.388 823.21
20,000 25.90 1036 1.1680 2.981 1544.4
50,000 28.01 2801 1.1552 2.549 3570.15
TABLE I
Coxii1rn Tnrx:i Cn::c1ris1ics or Fiiix
Drvrioirn TiniirN1 CoNvrc1ioN iN :
UNiroxix Hr:1rn RoiNn Tinr ro Pr 0
Nu
a T
'
/T
"
(YOC) (KL) (NS)
500 1.862 6.480 6.490 6.82
1000 1.884 6.675 6.695 6.935
2000 1.889 6.808 6.845 7.03
5000 1.911 6.932 6.895 7.175
10,000 1.918 7.004 6.995 7.30
20,000 1.924 7.063
50,000 1.930 7.130
YOC, Yu et al. [100a]; KI, Kays and Leung [101],
interpolated with respect to Re
`
; NS, Notter and
Sleicher [97], interpolated with respect to Re
`
.
values of Re were obtained from 2au
"
and hence may be used to recover,
if desired, the computed values of u
"
(2/ f )`. Values of Nu were obtained
from 2a/T
"
and hence can be used to recover the computed values of T
"
and in turn those of T
'
. The values of Nu attributed to Kays and Leung
[101] and Notter and Sleicher [97] were obtained by theoretically based
interpolation of their actual computed values with respect to Re. In all cases,
the new values lie between the older ones. The models and procedures used
to obtain these earlier computed values are discussed subsequently. The
computed values of u
"
, u
'
/u
"
, Re 2au
"
, f 2/(u
"
)` and Re( f /2)
2a/u
"
corresponding to the computed values of Nu, etc. in Table I are
listed in Table II.
TABLE III
Coxii1rn Tnrx:i Cn::c1ris1ics or Fiiix
Drvrioirn TiniirN1 CoNvrc1ioN iN :
UNiroxix Hr:1rn RoiNn Tinr ro
Pr Pr
'
0.8673 B:srn oN Eq. (172)
Nu
a T
'
/T
"
(YOC) (KL) (NS)
500 1.242 53.63 55.12 52.7
1000 1.222 99.45 102.0 97.1
2000 1.206 185.3 189.5 177.7
5000 1.189 424.1 433.0 401.7
10,000 1.177 796.7 812.4 749.8
20,000 1.167 1502
50,000 1.155 3487
See footnotes in Table I and values of Re
`
and
u
'
/u
"
in Table II. Values of KL and NS were
interpolated for both Re
`
and Pr.
Equation (172) implies that Pr Pr
'
Pr
'
for Pr 0.8673. Values of Nu
computed for this condition using Eq. (139), and again Eq. (99) for (u'v')
and Eq. (128) for , are listed in Table III. The corresponding values of
T
'
/T
"
are also provided. Individual values of T
"
and T
'
may be
determined from the tabulated values of Nu and T
'
/T
"
. The indicated
values of Re( f /2) 2a/u
"
may be used to determine values of (1 )`
""^`
.
The values of Nu attributed to Kays and Leung [101] and Notter and
Sleicher [97] were in this case obtained by interpolating their computed
values with respect to both Pr and Re. Again the new values are intermedi-
ate to the older ones.
b. Solutions for General Values of Pr For Pr 0.867, Eq. (127) may be
rearranged and approximated by
Nu
8
,
"
(1 )`
_
1
Pr
Pr
'

u'v')
1 (u'v')
dR"
, (176)
and hence by
Nu
8
(1 )`
"^
_
1
Pr
Pr
'

u'v')
1 (u'v')
""^`
, (177)
where the weighting factor for the integrated-mean value of the term in
TABLE IV
Prnic1rn Nissri1 Nixnrs ro Fiiix Drvrioirn TiniirN1 CoNvrc1ioN iN :
UNiroxix Hr:1rn RoiNn Tinr vi1n Pr
'
n:srn oN Eq. (172) (from Yu et al. [100a])
Small Pr
a 10" 10` 0.01 0.1 0.7
Nu
500 6.481 6.489 7.073 16.67 48.17
1000 6.675 6.694 7.927 26.70 88.53
2000 6.809 6.848 9.327 44.39 163.7
5000 6.933 7.035 12.95 90.62 371.5
10,000 7.006 7.210 18.25 159.1 694.0
20,000 7.068 7.476 27.63 283.3 1302
50,000 7.141 8.154 51.84 618.6 3006
Large Pr
a 1 10 100 1000 10,000
Nu/(0.07343(Pr/Pr
'
)`Re( f /2)`)
500 0.7462 0.9227 0.9794 0.9935 0.9950 1.0000
1000 0.6957 0.9066 0.9763 0.9934 0.9953 1.0000
2000 0.6510 0.8903 0.9726 0.9928 0.9953 1.0000
5000 0.5993 0.8688 0.9673 0.9918 0.9952 1.0000
10,000 0.5650 0.8529 0.9631 0.9909 0.9948 1.0000
20,000 0.5341 0.8374 0.9588 0.9899 0.9945 1.0000
50,000 0.4979 0.8176 0.9532 0.9886 0.9936 1.0000
square brackets is (1 )`. Since Eqs. (172)(174) all predict increasing
values of Pr
'
as Pr decreases, it may be inferred from Eq. (177) that the effect
of any error in the values of Pr
'
used to compute Nu from Eq. (127) for small
values of Pr will be very limited and will continually decrease as Pr
decreases below a value of 0.867. Insofar as Eq. (172) is valid, Pr
'
only varies
slightly for Pr 0.867, at least in the turbulent core. On the basis of these
considerations for small and large values of Pr, Eq. (172) might, despite its
obvious shortcomings, be expected to result in a reasonable approximation
for Nu for all values of Pr. Values so computed are listed in Table IV. As a
quantitative test, the computations leading to the values in Table IV were
repeated using Eq. (173) rather than Eq. (172) for Pr
'
. The results differ
signicantly only for very large a with Pr 0.001, 0.01, 0.1 and 10" or
greater and therefore are not reproduced herein.
c. Prior Computed Values of Nu Many prior analytical and numerical
solutions for turbulent convection have neglected the variation in the total
heat ux density with radius or postulated the same linear variation as for
the total shear stress. Many have postulated Pr
'
1 or some other xed
value for all conditions, and a number have incorporated the postulate that
u'v' is proportional to y" near the wall. Only the two numerical solutions
that avoid all of these gross idealizations will be examined here, namely the
previously mentioned ones of Kays and Leung [101] and Notter and
Sleicher [97].
The solutions of Kays and Leung are ostensibly for circular annuli but
include a uniformly heated round tube and parallel plates as limiting cases.
They carried out numerical integrations of the partial differential energy
balance using separate, incongruent correlating equations for the velocity
and eddy viscosity as well as an expression of unknown accuracy for the
turbulent Prandtl number. Their expressions for j
'
and u are unquestionably
less accurate than the equivalent values used by Yu et al. [100a]. For Pr 0
the errors due to their expressions for the eddy viscosity and the turbulent
Prandtl number phase out, and the slight discrepancies between their values
of Nu and those of Yu et al. must be due to the inaccuracy of their values
of u as compared to those of Yu et al. for (u'v'). The slightly greater
discrepancies in Table III presumably stem only from the values that Kays
and Leung used for j
'
and u since the dependence on Pr
'
is effectively
eliminated.
Notter and Sleicher [97, 111] developed Graetz-type series solutions for
Nu in developing as well as fully developed convection. The correlating
equations that they utilized for u, j
'
/j, and Pr
'
[Eq. (173)] are almost
certainly more accurate than those used by Kays and Leung, but less
accurate, with respect to u and j
'
/j, than the equivalent values used by Yu
et al. The remarks concerning the discrepancies of the values of Kays and
Leung for Nu in Tables I and III are applicable at least qualitatively to those
of Notter and Sleicher.
All in all, the values of Yu et al. for Nu in Tables I, III and IV are
presumed to be more accurate than any previously computed values,
primarily because of the greater accuracy associated with Eq. (99) for
(u'v'). Neither the absolute nor the relative error in Nu associated with
the values used for Pr
'
in the numerical predictions of Yu et al., Kays and
Leung, and Notter and Sleicher can be evaluated with certainty at this time.
Fortunately, the error associated with Pr
'
is reduced in Nu in all cases by
the integration or summation involved in the evaluation of the latter.
Comparison of the numerically computed values of Nu with experimental
data is deferred until after their representation by correlating equations.
TABLE V
Tnr Tnrx:i Cn::c1ris1ics or Fiiix
Drvrioirn TiniirN1 CoNvrc1ioN iN : RoiNn
Tinr vi1n UNirox W:ii-Trxir:1ir ro
Pr 0 (from Yu et al. [100a])
a Nu T
'"
/T
""
500 4.959 2.124
1000 5.055 2.152
2000 5.122 2.171
5000 5.187 2.188
10,000 5.225 2.198
20,000 5.258 2.205
50,000 5.295 2.214
TABLE VI
Tnr Tnrx:i Cn::c1ris1ics or Fiiix
Drvrioirn TiniirN1 CoNvrc1ioN iN : RoiNn
Tinr vi1n UNirox W:ii-Trxir:1ir ro
Pr Pr
'
0.8673 B:srn oN Eq. (172)
a Nu T
'
/T
"
500 52.07 1.276
1000 97.08 1.251
2000 181.5 1.232
5000 416.9 1.210
10,000 784.8 1.196
20,000 1482 1.184
50,000 3447 1.169
3. Numerical Results for an Isothermal Wall-Temperature
Yu et al. [100a] carried out numerical calculations for fully developed
turbulent convection in a round tube following a discrete step in wall
temperature for the same conditions as those of Tables I, III, and IV. Owing
to the presence of T in Eq. (145) for and of in Eq. (120) for T , an
iterative method of solution is required. They concluded that stepwise
solution of the differential equivalents of these two equations for trial values
of T
"
was more efcient computationally than iterative evaluation of the
integrals by quadrature. The results obtained using Eq. (172) for Pr
'
are
summarized in Tables V, VI, and VII. The computed values of Nu for
TABLE VII
Prnic1rn Nissri1 Nixnrs ro Fiiix Drvrioirn TiniirN1 CoNvrc1ioN iN : RoiNn
Tinr vi1n UNirox W:ii-Trxir:1ir vi1n Pr
n:srn oN Eq. (172)

Small Pr
10" 10` 0.01 0.1 0.7
a Nu
500 4.959 4.967 5.488 14.61 46.50
1000 5.055 5.072 6.155 23.89 86.01
2000 5.123 5.157 7.328 40.53 159.7
5000 5.188 5.275 10.50 84.57 364.1
10,000 5.227 5.402 15.27 150.4 682.0
20,000 5.262 5.611 23.89 270.6 1282
50,000 5.304 6.173 46.54 596.8 2966
Large Pr
a 1 10 100 1000 10,000
a Nu/(0.07343(Pr/Pr
'
)`Re( f /2)`)
150 0.8216 0.9440 0.9812 0.9914 0.9930 1.0000
500 0.7270 0.9200 0.9791 0.9935 0.9950 1.0000
1000 0.6811 0.9047 0.9761 0.9934 0.9953 1.0000
2000 0.6394 0.8887 0.9725 0.9928 0.9953 1.0000
5000 0.5904 0.8675 0.9672 0.9918 0.9952 1.0000
10,000 0.5575 0.8518 0.9630 0.9909 0.9948 1.0000
20,000 0.5278 0.8363 0.9588 0.9899 0.9945 1.0000
50,000 0.4928 0.8166 0.9531 0.9886 0.9936 1.0000
uniform wall temperature are observed to be signicantly less than those for
uniform heating only for Pr 1.
The values of Nu in Table VIII for a5000 only were, on the other
hand, calculated using Eq. (173) in order to provide a direct comparison
with the values computed by Notter et al. [97] who used the same
expression. The differences are therefore presumed, if numerical errors
in calculation are negligible, to be wholly a consequence of using (u'v')
from Eq. (99) rather than less accurate and incongruent expressions for j
'
and u.
TABLE VIII
Coxi:isoN or 1nr Coxii1rn V:iirs or Nu ro Fiiix
Drvrioirn TiniirN1 CoNvrc1ioN vi1n : UNirox W:ii-
Trxir:1ir :1 a5000 vi1n Pr
'
B:srn oN Eq. (173)
Nu
Pr Yu et al. [100a] Notter and Sleicher* [97]
0 5.187 5.29
10` 9.350 11.86
10 84.96 65.7
0.7 360.2 332
1.0 454.0 443
8 1306
C
1220
10` 3572 3436
10` 7915 7747
10" 17,119 16,730
*Interpolated semitheoretically with respect to a.
C
Interpolated semitheoretically with respect to Pr.
4. Numerical Results for Nu for Parallel-Plate Channels
a. Equal Uniform Heating on Both Plates Danov et al. [85] utilized the
integral formulations of Eqs. (152)(155) together with (u'v') from Eq.
(99), from Eq. (153), and Pr
'
from Eq. (172) to compute numerical
solutions for fully developed convection in turbulent ow between two
parallel plates heated uniformly and equally. Their results are summarized
in Table IX. The values of Re
"'
4bu
"
in Table IX correspond to values
of u
"
determined by integrating (u'v') as given by Eq. (99) with b
substituted for a. The corresponding values of Re
"'
4u
"
b, u
'
/u
"
,
f 2/(u
"
)` and Re
"'
( f /2) 4b/u
"
are also listed in Table X. The values
of Re
"'
( f /2) were determined from 4b/u
"
and those of
""
from Re
"'
( f /
2)/Nu
"'
Pr 0.867}. Their computed values of Nu
"'
are compared in Table
XI with the earlier ones of Kays and Leung [101] for the other limiting case
of a concentric circular annulus. Interpolation was avoided by utilizing
values of b corresponding to the values of Re
"'
chosen by Kays and Leung.
The comments on the discrepancies between the new and prior results in
Tables I and III are presumed to be directly applicable here.
b. Different Uniform Temperatures on the Two Plates Danov et al. [85]
also carried out numerical integrations for this boundary condition
using the integral formulations of Eqs. (157) and (159). Their results are
TABLE IX
Prnic1rn Nissri1 Nixnrs ro Fiiix Drvrioirn CoNvrc1ioN iN TiniirN1 Fiov
nr1vrrN UNiroxix :Nn Eqi:iix Hr:1rn P::iiri Pi:1rs vi1n Pr
'
n:srn oN Eq. (172)
(from Danov et al. [85])
Nu
"'
for small values of Pr
Pr
b 0 0.001 0.01 0.10 0.70 0.867
500 10.43 10.45 11.46 28.93 90.40 101.3
1000 10.61 10.64 12.76 46.66 166.3 187.9
5000 10.85 11.03 21.15 162.4 701.6 804.7
10,000 10.93 11.27 30.31 288.0 1314 1515
50,000 11.07 12.78 89.95 1142 5732 6668
Nu
"'
for large values of Pr
Pr
b 1.0 10 100 1000 10,000 25,000
500 107.0 280.6 701.8 1535 3335 4531
1000 198.4 547.8 1392 3062 6667 9061
5000 876.2 2672 6927 15,271 33,334 45,288
10,000 1617 5148 13,815 30,466 66,656 90,568
50,000 7176 25,051 68,715 152,087 333,209 452,471
TABLE X
Coxii1rn Cn::c1ris1ics or Fiiix Drvrioirn TiniirN1 Fiov nr1vrrN P::iiri
Pi:1rs ir D:Nov et al. [85]
a u
"
u
'
/u
"
Re
"'
10` f 10` Re
"'
( f /2)
500 18.558 1.1605 37.116 5.81 107.8
1000 20.312 1.1437 81.249 4.85 196.9
5000 24.132 1.1189 482.64 3.43 828.8
10,000 25.738 1.1113 1029.5 3.02 1554
50,000 29.428 1.0974 5885.5 2.31 6796
TABLE XI
Coxi:isoN or Prnic1rn V:iirs or Nu
"'
ro Fiiix Drvrioirn TiniirN1 CoNvrc1ioN
rox Tvo UNiroxix :Nn Eqi:iix Hr:1rn Pi:1rs vi1n Pr
'
n:srn oN Eq. (172)
Nu
"'
Nu
"'
/Pr`
Pr 0 Pr 0.867 Pr 1000
b Re
"'
10` K & L D, A & C K & L D, A & C K & L D, A & C
415 30 10.41 10.40 84.90 85.90 99.90 127.5
1204 100 10.66 10.66 212.3 222.0 288.6 368.6
3244 300 10.74 10.81 514.2 543.1 766.5 991.9
9737 1000 10.90 10.93 1392 1478 2305 2967
K & L, Kays and Leung [101]; D, A & C, Danov, Arai and Churchill [85]; b, is based on
specied values of Re and Eq. (102).
summarized in Table XII. No appropriate prior results were identied for
comparative purposes.
C. Cori:1ioN ro Nu
Although direct numerical integrations such as those of Eqs. (89) for uand
(91) for u
"
using Eq. (99) for (u'v') are perhaps feasible on demand for each
particular condition of interest, those required for Nu for each value of aand
Pr are somewhat more demanding because of the added dependence on and
Pr
'
. Correlating equations are therefore convenient for computed values as
well as for experimental data for convection. By denition, correlating
equations for computed values necessarily incorporate some empiricism. That
empiricismmay, however, often be minimized by the appropriate use of exact
or nearly exact asymptotic expressions within the structure of the correlating
equation. Such theoretically structured expressions are more reliable func-
tionally and usually more accurate and general than purely empirical ones.
1. Dimensional Analysis
Dimensional and speculative analysis proved to be very helpful in
constructing the nal correlating equations for turbulent ow. It is, however,
much less helpful in turbulent convection, again because of the added
dependence on Pr, Pr
'
, and .
The heat transfer coefcient for fully developed convection in a smooth
round tube and for invariant physical properties might, quite justiably, be
postulated to be a function only of D, u
"
, t
"
, j, j, k, and c
. However, since
TABLE XII
Prnic1rn Nissri1 Nixnrs ro Fiiix Drvrioirn iN TiniirN1 CoNvrc1ioN nr1vrrN
Pi:1rs :1 UNrqi:i UNirox Trxir:1irs vi1n Pr
'
n:srn oN Eq. (172)
(from Danov et al. [85])
Nu
'
for small values of Pr
Pr
b 0 0.001 0.01 0.10 0.70 0.867
500 1.0 1.002 1.113 3.337 14.16 16.43
1000 1.0 1.003 1.230 5.609 26.74 31.20
5000 1.0 1.018 2.126 21.86 118.8 139.8
10,000 1.0 1.036 3.186 40.54 226.9 267.6
50,000 1.0 1.175 10.83 175.3 1028 1218
Nu
'
for large values of Pr
Pr
b 1.0 10 100 1000 10,000 25,000
500 18.11 66.58 169.7 382.1 833.3 1132
1000 34.49 31.07 338.1 761.3 1665 2264
5000 155.3 630.5 1671 3797 8319 11,307
10,000 180.0 1240 3380 7588 16,624 92,617
50,000 1360 5993 16,656 37,915 82,855 113,154
t
"
/ju`
"
is known to be a unique function of D(t
"
j)`/j, one of the ve
variables in these latter two groupings is redundant in the listing for h. For
example, eliminating t
"
, u
"
, and j individually allows the following three
different dimensionless groupings to be derived:
hD
k

Du
"
j
j
,
c
j
k
or Nu Re, Pr} (178)
hD
k

D(t
"
j)`
j
,
c
j
k
or Nu
Re
f
2
`
, Pr
(179)
and
hD
k

Dt
"
ju
"
,
c
j
k
or Nu
Re
f
2
, Pr
. (180)
These three expressions are functionally equivalent to one another by virtue
of the relationship between t/ju`
"
and D(t
"
j)`/j, but on speculative
reduction they lead in some but not all cases to fundamentally different
results. For example, the further speculation of independence of h from D
leads, respectively, to
Nu RePr} (181)
NuRe
f
2
`
Pr} (182)
and
NuRe
f
2
Pr}. (183)
Equation (182) has been shown [see Eq. (136)] to be a valid asymptote for
Pr , whereas Eq. (183) provides a rst-order expression for
Pr Pr
'
Pr
'
(neglecting the dependence on ). On the other hand, Eq.
(181) does not appear to be valid for any condition. The speculation of
independence from c
in Eqs. (178)(180) leads to the limiting solutions for

Pr 0, but elimination of j, k, j, and t
"
individually does not appear to
lead to valid expressions. These limited results are to be contrasted with the
extensive set of asymptotes obtained for ow by speculative analysis.
Nusselt [102] in 1909 was apparently the rst to apply dimensional
analysis to turbulent convection in a round tube. Unfortunately, he postul-
ated a power dependence of h on each of the dependent variables and
thereby obtained the equivalent of
Nu ARePr" (184)
rather than simply Eq. (178). On the basis of the various exact integral
expressions of Section III, A, Nu does not appear to be a xed power of Re
for any condition. Since the friction factor, f, was found in Section II to be
a non-power-function of Re, the proportionality of Nu to Re( f /2) or
Re( f /2)` does not constitute a power-dependence on Re. Similarly, Nu was
found to be a xed power of Pr only in the limit of Pr .
The use of Eq. (184) for correlation of experimental data has actually
impeded the representation, understanding, and prediction of turbulent
convection, as illustrated in the following paragraphs.
2. Purely Empirical Correlating Equations
Nusselt [102] tted the constants of Eq. (184) using his own experimental
data for turbulent convection in gases in a round tube and determined an
exponent n 0.786 for Re and inexplicably the same value for the exponent
of Pr. Based on experimental data for Re 10" and various gas and liquids
Fic. 15. Determination of power dependence of Nu on Pr for Re 10". (From Sherwood
and Petrie [104], Figure 1.)
with 0.7 Pr 100, Dittus and Boelter [103] in 1930 recommended
A0.0265 and m0.3 for cooling, and A0.0243 and m0.4 for
heating. Sherwood and Petrie [104] in 1932 plotted experimental values of
Nu for Re 510" versus Pr in logarithmic coordinates, as shown in Fig. 15,
and determined a power dependence on Pr of 0.4. The straight line in Fig.
16 represents the following expression:
Nu 0.024Re"`Pr"". (185)
A later correlation of this type from Coulson and Richardson [105] is
shown in Fig. 16. The data appear to be well represented on the mean for
large Re by Eq. (184) with A0.023, m0.4, and n 0.8, but the scatter
is suppressed visually by the logarithmic coordinates and furthermore is
undoubtedly due in part to the oversimplied form of correlation as well as
to experimental inaccuracy.
Colburn [106] in 1933 noted the similarity of Eq. (185) to the following
empirical expression for the friction factor:
f
2
0.023Re"`. (186)
Fic. 16. Test of Eq. (185) with experimental data. (From Coulson and Richardson [105], p.
166.)
He thereby inferred that
Nu
RePr`
f
2
. (187)
He chose the exponent of
`
for Pr, not on theoretical grounds but simply as
a compromise for the values of Dittus and Boelter and others ranging from
0.3 to 0.4. Equation (187) predicts the wrong functional dependence for Nu
on Re except as a rst-order approximation for Pr O1} and on Pr except
in the asymptotic limit of Pr .
3. Numerical Predictions for L ow-Prandtl-Number Fluids
Equations (185) and (187) failed utterly to predict the convective behavior
of liquid metals in nuclear reactors in the 1950s, thereby stimulating the new
theoretical analyses described in the following section. It was soon recog-
nized that the distinctive thermal characteristic of liquid metals was their
relatively high thermal conductivity (or low Prandtl number), which results
in a signicant contribution by thermal conduction even in the turbulent
core.
Martinelli [107], Lyon [94], and others derived numerical solutions for
turbulent convection in round tubes, using the eddy conductivity and
accounting for thermal conduction over the entire cross section. These
solutions predicted a lower limiting value for Nu as Pr 0 and an improved
representation for liquidmetal heat transfer. They also had the effect of
establishing the credibility of theoretical predictions as compared to purely
empirical correlations of experimental data. However, Lyon conjectured
that his computed values of Nu would be a function only of RePr, that is, to
be independent of the viscosity, thereby leading to miscorrelations such as
that of Lubarsky and Kaufman [108], as shown in Fig. 17. The dashed and
dotted lines represent Eq. (185) for Pr 0.006 and Pr 0.03, respectively.
The other two curves represent purely empirical correlating equations.
Sleicher and Tribus [109] carried out numerical calculations for Nu for a
wide range of values of Re and Pr, developing as well as fully developed
convection, and a number of thermal boundary conditions, using a Graetz-
type series expansion and more accurate values of u, j
'
, Pr, and Pr
'
than
those of prior investigators. They concluded from their results that the scatter
in Fig. 17 and similar plots was due in part to a parametric dependence on Pr
beyond that of RePr, as well as to incomplete thermal development.
Notter and Sleicher [97, 110, 111] subsequently improved somewhat
upon these solutions. Their numerical results are probably the most reliable
in the literature other than those of Yu et al. [100a], which are based on the
equivalent of more accurate values of u and j
'
.
Churchill [112] correlated all of the computed values of Nu of Notter and
Sleicher for fully developed convection as well as culled experimental data
with the expression
Nu Nu
"
0.079Re
f
2
`
Pr
[1 Pr"`]`'
(188)
with Nu
"
4.8 and 6.3, respectively, for a uniform wall temperature and
uniform heating. He also extended this expression to encompass laminar
and transitional ow as follows:
Nu"Nu"
'

_
exp(2200Re)/366}
Nu`
'
Nu
"
0.079Re( f /2)`Pr`
[1 Pr"`]`'
`
`
.
(189)
Fic. 17. Representation of experimental values of Nu for liquid metals as a function of Pe RePr by Eq. (185): (- - -) Pr 0.006; ( ) Pr 0.03. (From
Lubarsky and Kaufman [108], Figure 42.)
Fic. 18. Representation of culled experimental data and predicted values for Nu and Sh by
Eq. (189). (From Churchill [112], Figure 1.)
Here Nu
'
, the solution for laminar ow, equals 3.657 for a uniform wall
temperature and 4.364 for uniform heating. Equation (189) is seen in Fig.
18 to represent the computed values of Notter and Sleicher as well as the
experimental data very well.
4. Mechanistic Analogies
Equation (187) is commonly known as the Colburn analogy because it was
constructed by postulating the same empirical dependence for Nu/RePr`
and f /2 on the Reynolds number. Many other analogies for turbulent
convection have been devised by postulating similar mechanisms of trans-
port for momentum and energy. Although Eqs. (136) and (141) and their
counterparts relate Nu and f, those relationships are simply a consequence
of the dependence of the rate of heat transfer on the velocity eld rather than
an explicit analogy. Several of these mechanistic analogies are described
briey because of the understanding they convey, and one in detail because
it proves remarkably useful for correlation.
Reynolds [18] in 1874, as mentioned in the Introduction, made a signif-
icant contribution to turbulent convection by postulating equal rates of
transport of momentum and energy from the bulk of the uid to the wall
by means of the uctuating eddies. His result, in modern notation, takes the
form of Eq. (142).
Prandtl [113] in 1910 attempted to improve upon the Reynolds analogy,
by postulating that the transport of momentum and energy by the eddies
extends only to the edge of a boundary layer and that the completion of the
transport to the wall occurs by linear molecular diffusion. His result may be
expressed as
Nu
Re( f /2)Pr
1 o(Pr 1)( f /2)`
, (190)
where oo(t
"
j)`/j. The major contribution of the Prandtl analogy, Eq.
(190), is the prediction that the dependence of Nu on Re shifts from
proportionality to Re( f /2) to proportionality to Re( f /2)` as Pr increases
from unity to very large values. It implies that the Reynolds analogy is
valid only for Pr 1. As contrasted with the Reynolds analogy, which
is free of explicit empiricism, the Prandtl analogy contains an empirical
factor o.
Thomas and Fan [114] attempted to improve upon one other deciency
of the Reynolds analogy by applying the penetration and surface renewal
model of Higbie [115] and Danckwerts [116] to account for transport from
the eddies to the wall when they reach it. Their result is
Nu Re
f
2
Pr`. (191)
Comparison of Eqs. (142), (190), and (191) with Eqs. (132)(136) and
(141) reveals that three analogies all incorporate the postulates of 0 and
Pr 1. They all fail outright for small values of Pr because of the failure to
account for conduction in the turbulent core. For large values of Pr, the
Reynolds analogy fails because of the neglect of the boundary layer, the
Prandtl analogy fails because of the neglect of turbulent transport within the
boundary layer, and the Thomas and Fan analogy fails because periodic
transient conduction is simply a very poor model for the combination of
turbulent and molecular transport in the boundary layer. Perhaps the
greatest lasting value of these analogies is the understanding provided by
analysis of the reasons for their failure.
5. A Useful Differential Analogy
Reichardt [87] in 1951 derived an analogy based on the differential
momentum and energy balances in time-averaged form. He utilized the eddy
viscosity model for turbulent transport, but his derivation will be outlined
here in terms of (u'v'). Taking the ratio of Eqs. (115) and (116),
respectively, with Eq. (85) gives
dT
du
(1 )
Pr
'
Pr

1
1 (u'v')(Pr/Pr
'
)(u'v')
. (192)
Integrating the rightmost form from the centerline to the wall results in
T
'

,
u
'
"
_
(1 )
1 (u'v')
Pr
Pr
'
(u'v')
du. (193)
His ingenious expansion of the equivalent integrand in terms of j
'
may be
rephrased in terms of (u'v') as follows:
T
'

,
u
'
"
1 (u'v')
Pr
Pr
'
(u'v')
Pr
'
Pr
1
Pr
'
Pr
1
Pr
Pr
'
u'v'
1u'v'
du.
(194)
In order to obtain a solution in closed form, Reichardt suggested that for
moderate and large values of Pr the leftmost term of the integrand be
approximated by (Pr
'
/Pr) since 0 for small values of u while (u'v')
51 for uu
'
. He also concluded that the rightmost term was negligible
except very near the wall where du5dy. He further postulated Pr
'
/Pr
to be invariant over the cross-section. Had he utilized the limiting form of
Eq. (93), that is, Eq. (194) for (u'v') in the rightmost term, he would have
obtained
Nu
2a
T
"
2a
T
'

T
'
T
"
1
(1 )
"
Pr
'
Pr
u
"
u
'

T
'
T
"
Re
f
2
1
Pr
'
Pr
3``:`
2
T
'
T
"
Pr
Pr
'
`
Re
f
2
`
, (195)
where (1 )
"
is the integrated-mean value over u. By virtue of Eq. (88),

this latter term may also be interpreted as the integrated mean, weighted by
1 (u'v'), over R`. Equation (195) is, by virtue of the limits of integration
and several of the approximations, applicable for a uniform wall tempera-
ture as well as for uniform heating.
6. T heoretically Based, Generalized Correlating Equations
On the basis of the asymptotic expressions for Pr Pr
'
and Pr for
uniform wall temperature, namely, Eqs. (151) and (136), Eq. (195) may be
interpreted as
Nu
1
Pr
'
Pr
1
Nu
1
Pr
'
Pr
1
Nu
,
(196)
where Nu
signies NuPr Pr
'
} and Nu
signies NuPr }. Accord-

ingly, Eq. (196) may be postulated to be applicable for uniform heating with
Nu
and Nu
from Eqs. (141) and (136), respectively. The analogy of

Prandtl [Eq. (190)] may be noted to have the form of Eq. (196) with,
however, the implicit postulates of Pr
'
1 and 0, and a missing
dependence on Pr` for Pr . Equation (195) may be interpreted on the
basis of the Prandtl analogy as the consequence of the resistances for
Pr
'
Pr and Pr in series, or alternatively as an application of Eq. (71)
with n 1 and limiting solutions of (Pr/Pr
'
) Nu
and Nu
/(1 Pr
'
/Pr).
Equations (195) and (196) are limited to Pr Pr
'
, which according to the
expressions of Yahkot et al. [89] and Jischa and Rieke [96] means to
Pr 0.848 and 0.867, respectively.
By analogy to Eq. (196), rearranged as
Nu Nu
"
Nu
Nu
1
1
Pr
'
Pr Pr
'
Nu
Nu
(197)
Churchill et al. [117] speculated that
Nu Nu
"
Nu
Nu
"
1
1
Pr
Pr
'
Pr
Nu
Nu
"
(198)
might be applicable as a correlating equation for Pr Pr
'
. However, Eq.
(198) was not found to be sufciently accurate and in addition to result in
a discontinuity in the derivative of Nu with respect to Pr/Pr
'
at Pr Pr
'
.
Accordingly, they introduced an arbitrary coefcient : as a multiplier of
(Pr/Pr
'
Pr)(Nu
/Nu
"
) and evaluated it functionally to provide a continu-
ous derivative. The resulting expression
Nu Nu
"
Nu
Nu
"
1
1
Pr
Pr
'
Pr
Nu
Nu
Nu
Nu
"
Nu
Nu
"
(199)
where Nu
Nu
Pr Pr
'
} 0.07343Re( f /2)`, has proven as successful
as Eq. (196). Although Eq. (199) lacks the theoretical basis of Eq. (196) it is
free of any explicit empiricism.
Because of the generality of their structure and components, Eqs. (196)
and (199) might be speculated to be applicable for all thermal boundary
conditions and for all channels. As will be shown, this conjecture is
conrmed for all of the yet available numerical results.
Although Eq. (136) is presumed to be universally applicable for Nu
,
different expressions are required for Nu
and Nu
"
in Eqs. (196) and (199)
for each case, as discussed next.
a. Uniformly Heated Round Tubes In order to utilize Eqs. (196) and (199)
for values of a intermediate to those of Tables IIV, it is necessary to have
supplementary correlating equations for Nu
and Nu
"
. The following purely
empirical expressions, together with u
"
from Eq. (102) (with a modied
leading constant of 3.2) reproduce the values of Nu in Tables I and III
almost exactly:
Nu
"
8
1
7.7
(u
"
)`"
(200)
Nu
Re( f /2)
1
185
(u
"
)``
2a/u
"
1
185
(u
"
)``
(201)
The choice of u
"
rather than u
'
, u
'
/u
"
, a or Re as the independent
variable in Eqs. (200) and (201) is arbitrary since they all bear a one-to-one
correspondence. The leading constant of 3.3 in Eq. (102) was chosen on the
basis of the experimental data of Zagarola [73], while the recommended
value here of 3.2 corresponds more closely to the computed values of u
"
in
Table II and is thereby self-consistent with the computed values of Nu.
b. Isothermally Heated Round Tubes Separate correlating equations might
have been devised for T
'
/T
"
and u
'
/u
"
as well as for (1 )
"^
and
(1 )
""^
in Eqs. (150) and (151). However, in the interests of simplicity,
the following overall expressions were derived:
Nu
"
8
1
1.538
(u
"
)`
(202)
Nu
Re( f /2)
1
148
(u
"
)`"
2a/u
"
1
148
(u
"
)`"
(203)
Equations (202) and (203) reproduce the values of Nu in Tables V and VI,
respectively, almost exactly.
c. Uniform and Equally Heated Parallel Plates The corresponding expres-
sions are
Nu
"
12
1
5.71
(u
"
)`"
(204)
and
Nu
Re( f /2)
1
90
(u
"
)``
4b/u
"
1
90
(u
"
)``
(205)
Here, Nu and Re are based on a characteristic length of 4b and Eq. (102) is
to be used for u
"
.
d. Convection between Isothermal Plates at Different Temperatures In this
case, b is chosen as the characteristic length in order that Nu
"
1. The
corresponding expression is
Nu
Re
f
2
1
11.707
u
"
b/u
"
1
11.707
u
"
(206)
Equation (206) reproduces the values in Table XII for Pr 0.867 very
closely.
e. Test of the Correlating Equations Figure 19 provides a test of Eqs. (196)
and (199) for the computed values of Nu for a uniformly heated round tube
and for parallel plates, both uniformly and equally heated and at different
uniform temperatures in terms of Pr/Pr
'
and Figure 20 for an isothermal
tube in terms of Pr with Pr
'
estimated from Eq. (172). The agreement is very
good. The slight discrepancy for Pr 0.01 and a50,000 is presumed to
result from the simplications made by Reichardt [87] in deriving the
equivalent of Eq. (195).
f. Interpretation of Correlating Equations Equation (199) predicts a rapid
increase in Nu as Pr increases followed by a point of inection and a
decreasing rate of increase as Pr Pr
'
. Equation (196) similarly predicts a
more rapid increase beyond Pr Pr
'
followed by a second point of
inection and a decreased rate of increase approaching a one-third-power
dependence. The changes for Pr Pr
'
are smaller than those for Pr Pr
'
and indeed almost indistinguishable in the scale of Figs. 19 and 20.
Such behavior, which is presumed to be real, is far more complex than
could ever be deduced from experimental or even precise computed values
and is an illustration of the value of theoretically structured equations for
correlation.
Equations (196) and (199) together with Eqs. (136), (172) and (200)(206)
are presumed to predict more accurate values of Nu than any prior
correlating equations. They are subject to signicant improvement primarily
with respect to Eq. (172). A more accurate expression for Pr
'
not only affects
the predictions of Eqs. (196) and (199) but also the numerically computed
values upon which Eqs. (136) and (200)(206) are based.
IV. Summary and Conclusions
A. TiniirN1 Fiov
1. A New Model for the Turbulent Shear Stress
The new and improved representatives proposed in Section I for fully
developed turbulent ow in a channel are a direct consequence of the
observation by Churchill and Chan [77] that the local, dimensionless,
time-averaged shear stress, namely (u'v')ju'v'/t
"
, constitutes a better
variable for this purpose than traditional mechanistic and heuristic quanti-
ties such as the mixing length and the eddy viscosity. Churchill [80]
Fic. 19. Representation of numerically predicted values of Nu by Yu et al. [100a] and
Danov et al. [85] with Eqs. (196) and (199) for a and b5000. [x, equally and uniformly
heated parallel plates (Nu 4bu
"
/v); , uniformly heated round tube (Nu 2au
"
/v); ,
parallel plates at different uniform temperature (Nubu
"
/v)].
subsequently noted that the local fraction of the shear stress due to
turbulence, namely (u'v')ju'v'/t, is an even better choice.
The presentation of new integral formulations and algebraic correlations
for fully turbulent ow based on the time-averaged partial differential
equations of conservation might appear to be atavistic in view of the recent,
presumably exact, solutions of these equations in their unreduced time-
dependent form. However, the use of integral and algebraic structures based
on the time-averaged equations may be expected to persist into the
Fic. 20. Representation of numerically predicted values by Yu et al. [100a] of Nu for a
uniform wall temperature by Eqs. (196), (199), and (172).
foreseeable future for two reasons. First, the exact numerical solutions,
which have been attained only by direct numerical simulation, are currently
very limited in scope by their computational requirements and perhaps their
inherent structure. Second, even if these limitations are eventually eliminated
or at least eased by improved computer hardware and software as well as
better inherent representations, or even if the DNS calculations are replaced
or supplemented by some other methodology, the results will be in the form
of discrete instantaneous or time-averaged values of u', v', u'v', and u for a
particular condition and therefore not directly useful for applications such
as the design of hydrodynamic piping. Correlating equations will accord-
ingly continue to be useful if not essential to summarize and generalize the
vast quantity of information that is generated. Theoretically based algebraic
structures will likewise continue to be useful in constructing forms for these
correlating equations.
2. Integral Formulations in Terms of the Turbulent Shear Stress
An unexpected result from the use of (u'v') as a variable was the
realization that, by virtue of integration by parts, u
"
as well as u may be
expressed as simple, single integrals of this quantity. The possibility of such
a simplication by means of integration by parts was apparently rst
discovered by Kampe de Fe riet [81] in the context of u'v' and a parallel-
plate channel. This suggestion was rst implemented by Pai [82] for both
parallel-plate channels and round tubes, but with very poor representations
for u'v'. The advantage of using u'v' rather than u as a primary variable was
noted by Bird et al. [35], p. 175, but only in connection with the cited work
of Pai, and even then incorrectly. The major contribution of Churchill and
Chan [77] in this context was the recognition that an accurate and
generalized correlating equation for (u'v') was the key to successful
implementation of the integral formulation.
An inherent advantage of correlating equations for (u'v') or (u'v') over
those for u apart from simplicity, is that integration is a smoothing
process and somewhat dampens any minor error in the integrand. Hence,
the predictions of Eqs. (89) and (90), using Eq. (99) for (u'v'), are
inherently more accurate than those of Eqs. (100) and (101).
3. T he Development of Correlating Equations for (u'v'), u, and u
"
The structure of an almost exact correlating equation for (u'v'), namely
Eq. (99), was developed by Churchill and Chan [71] from a number of
asymptotic and speculative expressions for the time-averaged velocity as
well as for the time-averaged turbulent shear stress. The empirical coefcient
of the asymptotic solution for y0 was evaluated using the several sets
of results obtained by DNS while those for intermediate values of y and
a were evaluated from the experimental time-averaged velocity distribu-
tions measured by Nikuradse [46]. These latter constants were subsequently
reevaluated by Churchill [80] using the recent improved measurements of
the time-averaged velocity distribution by Zagarola [73]. The incorporation
of the equivalent of the semilogarithmic expression for the time-averaged
velocity nominally restricts this correlating equation to a300, but it
provides a very good approximation for ya even down to a145,
the lower limit of fully turbulent ow.
The calculation of u and u
"
from the integral formulations using Eq.
(99) for (u'v') is feasible even with a handheld calculator. Hence, separate
correlating equations for u and u
"
are not really required. However, such
expressions were constructed in the name of convenience and tradition.
Equations (100) and (101) are presumed to be the most accurate expressions
in the literature for u and u
"
, respectively, for both smooth and naturally
rough pipe. Since u
"
(2/ f )`, the correlating equation for u
"
serves as
one for the Fanning friction factor as well.
Equations (99), (100), and (101) are subject to renement, at least in terms
of the coefcients, constants and combining exponents, upon the appearance
of better values for (u'v'), u, u
"
, and the roughness e
'
from either
experimentation or numerical simulations. The tabulated values of e
'
in the
current literature are very old and are almost certainly not representative
for modern piping. The reevaluation of these roughnesses for representative
materials and conditions would appear to have a high priority.
4. T he Analogy of MacL eod
The analogy attributed to MacLeod [60] was crucial to the development
of the just-mentioned correlations for (u'v') and u in that it allowed
experimental data and computed values for round tubes and parallel plates
to be used interchangeably. This little-known analogy appears to be
validated within the accuracy of the experimental values for u'v' and u, but
has no theoretical rationale. A critical test may be beyond the accuracy of
present experimental means, but should be possible, at least for a and
b300, by DNS. Such a resolution would appear to have great merit.
5. Obsolete and L imited Models
Science and engineering progress by discarding obsolete models as well as
by new discoveries. One of the rst discoveries resulting from the use of
(u'v') as the primary dependent variable was that the mixing length is
unbounded at one location in the uid in all channels and in addition is
negative over a nite adjacent region in all channels other than round tubes
and parallel plates. Although the eddy viscosity is well behaved in round
tubes and parallel-plate channels, it shares the failure of the mixing length
in all other channels.
How did these anomalies completely escape attention for 75 years? The
explanation has three elements. First, the initial numerical evaluation of the
mixing length by Nikuradse [45, 46] not only was based on experimental
data of insufcient precision but also was conditioned by a preconceived
notion concerning the behavior. Second, the subsequent acceptance of the
mixing length by later investigators is simply inexcusable, since the
aforementioned failures of this concept are readily apparent from a critical
examination of most of their own sets of measurements of the velocity
distribution as well as from the predictions thereof. Third, the anomalies are
much more apparent in terms of (u'v') than in terms of earlier formula-
tions.
When it was rst introduced by Launder and Spalding [42], the c
model appeared to have great promise for the predictions of turbulent ows,
but it has ultimately proven to have no real utility. For round tubes and
parallel-plate channels the c model, in all of its manifestations, not only
invokes a great deal of empiricism and approximation, but is unneeded. In
all other channels it shares the failure of the eddy-viscosity model, to which
it is directly linked. The c u'v' model avoids this linkage and thereby
has a possible role, despite its high degree of empiricism, for geometries,
such as circular annuli, in which the variation of the total shear stress is not
known a priori. The large eddy simulation (LES) methodology avoids the
need for time-averaging, at least in the turbulent core, and has a wider range
of applicability than the DNS methodology, but at the price, at least at the
present time, of a considerable degree of empiricism and approximation for
the region near a surface.
Barenblatt [57] and co-workers have recently attempted to resuscitate the
power-law correlation of Nikuradse [46] and Nunner [56] for the time-
averaged velocity, and Zagarola [73] has demonstrated that it is more
accurate than the semilogarithmic model for a narrow range of values of y.
This improvement is accomplished at the price of considerable empiricism,
functionally as well as numerically, and very poor behavior outside that
narrow range. Hence it does not appear to have any utility as an element of
overall correlating equations for (u'v') and u.
B. TiniirN1 CoNvrc1ioN
1. Initial Perspectives
As a result of the great success described earlier in developing simple
formulations and improved correlating equations for fully developed turbu-
lent ow, the development of analogous expressions for fully developed
turbulent convection was undertaken with consideration condence and
great expectations. Unfortunately, it soon became apparent that turbulent
convection is much more complex than turbulent ow even in the simplest
of contexts, and that the data base, both experimental and computational,
and the known asymptotic structure are much more limited.
Turbulent convection would be expected to be responsive to the same new
numerical methodologies used for ow, such as DNS, but so far the greater
inherent complexity of the behavior has limited the scope and accuracy of
such results.
2. New Differential Models
Time-averaging of the partial differential energy balance, followed by one
integration and expression in terms of dimensionless variables, results in Eq.
(106), in which (T 'v')jc
T 'v'/j, the fraction of the heat ux density due

to turbulence, is a new variable analogous to (u'v'). However, the heat
ux density ratio, j/j
"
, is a dependent variable, given in general by Eq. (107)
as contrasted with t/t
"
1 y/a for ow. Furthermore, very few data
have been obtained for T 'v' or correlated in terms of (T 'v'). Accordingly,
Eq. (106) was reexpressed as Eq. (109) with the expectation that the
behavior of Pr
'
/Pr 1 (T 'v')/1 (u'v') would be more constrained
than that of (T 'v'). The terms Pr
'
/Pr and j/j
"
represent the complications
associated with turbulent convection as compared with turbulent ow.
3. T he Heat Flux Density Ratio
For a uniform heat ux from the wall, the heat ux density ratio is a
function only of the time-averaged velocity distribution, and Eq. (109) may
be reduced to Eq. (115) with given by Eq. (125), which may also be
expressed in terms of (u'v') [see Eq. (128)]. Owing to the accuracy and
generality of Eq. (99), the uncertainty associated with Eq. (207) and thereby
with the prediction of Nu herein is essentially conned to Pr
'
/Pr. Many past
semitheoretical expressions for Nu have, however, also been in error to an
unknown degree because of the implicit postulate of 0.
4. T he Turbulent Prandtl Number
One of the initial objectives of the investigation of turbulent convection
that culminated in this article was to eliminate Pr
'
or its equivalent, Pr
'
[see
Eq. (113)]. However, an important discovery resulting from the use of
(u'v') and (T 'v') as primary variables is that Pr
'
and Pr
'
bear a
one-to-one correspondence to (u'v') and (T 'v') and are therefore
independent of their heuristic diffusional origin.
It has generally been postulated that Pr
'
and Pr
'
are functions only of
(u'v') (or j
'
/j) and Pr and thus independent of the thermal boundary
condition. For example, this postulate is inherent in the solutions of Notter
and Sleicher [97, 110, 111] for developing thermal convection with both
uniform heating and a uniform wall temperature. It is implied by Eq. (172)
of Jischa and Rieke [96], Eq. (173) of Sleicher and Notter [97], and Eq.
(174) of Yahkot et al. [89]. The last imply that their expression is also
independent of geometry. The only direct experimental conrmation of
either postulate appears to be that of Abbrecht and Churchill [22], who
found the eddy conductivity to be independent of length in developing
thermal convection in an isothermal round tube as well as identical to that
of Page et al. [90] for fully developed heat transfer across a parallel-plate
channel at equal values of a and b, respectively. The publication of this
result was greeted with two contradictory responses: one that it was
obvious, and the other that it was obviously wrong.
Despite the great simplication provided by these generalizations, neither
a completely satisfactory correlating equation nor a universally accepted
theoretical expression for Pr
'
or Pr
'
appears to exist. This is the principal
unresolved problem of turbulent convection, at least in round tubes and
parallel-plate channels, and is worthy of renewed effort, experimentally,
theoretically, and computationally. Thermal calculations by DNS, LES, and
cu'v' T 'v' generate values of (T 'v') or the equivalent and therefore
do not require a separate expression for Pr
'
. However, with the exception
of the predictions of Papavassiliou and Hanratty [98], which are limited to
b150, these methodologies have not yet produced reliable values of
(T 'v') or Pr
'
for a broad range of Pr and (u'v') or y and a.
5. Integral Formulations for Nu
Because of the great simplication in the expression for the heat ux
density ratio that is possible for uniform heating, most theoretical solutions,
including the present ones, have been restricted to this condition. By virtue
of Eq. (125), Nu may be represented by the single integral of Eq. (126) in
terms of Pr
'
and of Eq. (127) in terms of Pr
'
. Such a simplication has
apparently not been achieved before because of the greater complexity of
the formulations in terms of j
'
/j and u as compared to these in terms of
(u'v') only.
Because of the uncertainty in the various expressions for Pr
'
and Pr
'
,
particular attention has been given herein to three cases for which that
uncertainty is eliminated or greatly reduced, namely Pr 0, PrPr
'
Pr
'
,
and Pr while y0. The second of these conditions is implied by Eq.
(172) to occur for Pr 0.8673, by Eq. (174) for Pr 0.848, and by Eq. (173)
for values of Pr varying from 0.8 to 0.9, depending upon the value of
(u'v').
Equation (172) implies a limiting value of Pr
'
0.85 for Pr and Eq.
(173) a limiting value of Pr
'
0.78 for y0 for large Pr, but the
calculations of Papavassiliou and Hanratty [98] using DNS suggest that
such a nite limiting value is attained only for Pr 100. The postulate that
Pr
'
0.85 as y0 and Pr allows the derivation of an analytical
solution in closed form, as represented by Eq. (136), which, however, may
be valid only for large values of Pr but less than 100.
6. Numerical Solutions for Nu
Numerical solutions for Nu have been carried out by Heng et al. [100]
for a uniformly heated round tube and by Yu et al. [100a] for both a
uniformly heated and an isothermal tube, in both instances for a complete
range of values of Pr and a wide range of values of a using Eq. (172) for
Pr
'
. The results of Yu et al., including the three limiting cases described in
the preceding section are summarized in Tables IVIII. Similar results for
parallel-plate channels, as obtained by Danov et al. [85], are summarized in
Tables IXXII. Despite the uncertainty associated with the use of Eq. (172)
for Pr
'
, these values of Nu are presumed to be more accurate than any prior
ones because of the essentially exact representation in every other respect.
They are of course subject to improvement and should be updated when
more accurate values or expressions for Pr
'
or Pr
'
become available.
7. Final Correlating Equations
Because of a lack of data of proven reliability and broad scope, a new
correlating equation was not devised for (T 'v') or Pr
'
. For the same
reason, new correlating equations were not constructed for T . Instead
attention was focused directly on Nu.
The integral formulations for Pr 0 and Pr Pr
'
Pr
'
and the ana-
lytical solution for Pr imply that all prior correlating equations for Nu,
including the Colburn analogy, are in signicant error functionally as well
as numerically even over their own purported range of validity.
A new simple but very general correlating equation for Nu for Pr 0.867
was devised on this basis of the analogy of Reichardt. This expression, Eq.
(196), represents all of the computed values of Yu et al. and Danov et al. for
Pr 0.867 quite accurately and is presumed to be applicable for other
conditions as well. A supplementary empirical correlating equation was
devised for Pr 0.867. This expression, Eq. (199), also represents all of the
computed values very well. The overall success of Eqs. (196) and (199) is
displayed in Fig. 19. in terms of Pr/Pr
'
, and in Fig. 20. The close represen-
tation of the computed values of Nu in Fig. 20 does not constitute a critical
test of the absolute values of Nu because Eq. (172) was used for Pr
'
in both
cases. However, the accuracy of the predictions of Nu by Eqs. (199) and
(198) is presumed to be independent of the expression used for Pr
'
.
References
1. Lamb, H. (1945). Hydrodynamics, 1st American ed. Dover Publications, New York, p. 663.
2. Richter, J. P., ed. (1970). T he Notebooks of L eonardo da V inci, Vol. 1. Dover Publications,
New York.
3. Chandrasekhar, S. (1949). On Heisenbergs elementary theory of turbulence. Proc. Roy.
Soc. (London) A 200, 2033.
4. Einstein, A. (1905). [Engl. transl. On the motion required by the molecular kinetic theory
of heat of particles suspended in uids at rest.] Ann. Phys. 17, 549560.
5. Heisenberg, W. (1924). U ber Stabilita t und Turbulenz von Flu ssigkeitsstro men. Ann.
Phys. 74, 577627.
6. Kapitsa, P. L. (1947). Theoretical and empirical formulas for heat transfer in two-
dimensional turbulent ow. Doklady AN SSSR 55, 595602.
7. Landau, L. D. (1944). Turbulence. Doklady AN SSSR 44, 339342.
8. Lorentz, H. A. (1907). U ber die Entstehung turbulenter Flu ssigkeitsbewegungen und u ber
den Einuss dieser Bewegungen bei der Stro mung durch Rohren. Abh. Theor. Phys.,
L eipzig 1, 4371.
9. Newton, I. (1701). Scala graduum Caloris. Philos. Trans. Roy. Soc. (London) 22, 824829.
10. Rayleigh, Lord (J. W. Strutt) (1880). On the stability or instability of certain uid motions.
Proc. London Math. Soc. 11, 5170.
11. Sommerfeld A. (1909). A Contribution to the Hydrodynamical Explanation of Turbulent
Fluid Motions. Atti del IV. Cong. intern. dei matematici, Roma, pp. 116124.
12. Uhlenbeck, G. (1980). Some notes on the relation between uid mechanics and statistical
physics. Ann. Rev. Fluid Mech. 12, 19.
13. Mises, R. von (1941). Some remarks on the laws of turbulent motion in channels and
circular tubes. In Th. von Krmn Anniversary Volume, Calif. Inst. Techn. Press, Pasadena,
CA, pp. 317327.
14. Weizsa cker, C. F. von (1948). Das Spectrum der Turbulenz bei grossen Reynoldsschen
Zahlen, Z. Phys. 124, 614627.
15. Zeldovich, Ya. B. (1937). Limiting laws for turbulent ows in free convection. Zh. Eksp.
Teor. Fiz. 7, 14631465.
16. Schlichting, H. (1979). Boundary Layer Theory, 7th ed., translated by J. Kestin. McGraw-
Hill Book Co., New York, p. xxii.
17. Reynolds, O. (1895). On the dynamical theory of incompressible viscous uids and the
determination of the criterion. Philos. Trans. Roy. Soc. (London) 186, 123161.
18. Reynolds, O. (1874). On the extent and action of the heating surface of steam boilers.
Proc. Lit. Soc. Manchester 14, 712.
19. Navier, C.-L. M. N. (1822). Me moire sur les lois du mouvement des uides. Mm. Acad.
Roy. Sci. 6, 389416.
20. Stokes, G. G. (1845). On the theories of internal friction of uids in motion, and of the
equilibrium and motion of elastic solids. Trans. Cambridge Philos. Soc. 8, 287319.
21. Barenblatt, G. I., and Goldenfeld, N. (1995). Does fully developed turbulence exist?
Reynolds number independence versus asymptotic covariance. Phys. Fluids 7,
30783082.
22. Abbrecht, P. H., and Churchill, S.W. (1960). The thermal entrance region in fully
developed turbulent ow. AIChE J. 6, 268273.
23. Fourier, J. B. (1822). Thorie analytique de la chaleur, Gauthie r-Villais, Paris.
24. Rayleigh, Lord (J. W. Strutt) (1915). The principle of similitude. Nature 95, 6668.
25. Rayleigh, Lord (J. W. Strutt) (1892). On the question of the stability of the ow of uids.
Philos. Mag. 34, 5970.
26. Churchill, S. W. (1981). The use of speculation and analysis in the construction of
correlations. Chem. Eng. Commun. 9, 1938.
27. Churchill, S. W. (1997). A new approach to teaching dimensional analysis. Chem. Eng.
Educ. 30, 158165.
28. Reynolds, O. (1883). An experimental investigation of the circumstances which determine
whether the motion of water shall be direct or sinuous, and the law of resistance in parallel
channels. Philos. Trans. Roy. Soc. (London) 174, 935982.
29. Prandtl, L. (1926). ber die ausgebildete Turbulenz, Verhdl. 2. Kong. Techn. Mechanik,
Zurich, p. 62.
30. Millikan, C. B. (1938). A critical discussion of turbulent ows in channels and circular
tubes. Proc. Fifth Intern. Congr. Appl. Mech., Cambridge, MA, pp. 386392.
31. Ka rma n, Th. von (1930). Mechanische A hnlichten und Turbulenz. Proc. Third Intern.
Congr. Appl. Mech., Stockholm, Part 1, pp. 8593.
32. Murphree, E. V. (1932). Relation between heat transfer and uid friction. Ind. Eng. Chem.
24, 726736.
33. Boussinesq, J. (1877). Essai sur la the orie des eaux courantes. Mm. prsents divers savants
Acad. Sci. Inst. Fr. 23, 1680.
34. Prandtl, L. (1925). Bericht u ber Untersuchungen zur ausgebildeten Turbulenz. Zeit.
angew. Math. Mech. 6, 136139.
35. Bird, R. B., Stewart, W. E. and Lightfoot, E. N. (1960) Transport Phenomena, John Wiley
& Sons, New York.
36. Rotta, J. C. (1950). Das in Wandnahe gultige Geschwindigkeitsgesetz turbulenter
Stro mungen. Ing. Arch. 18, 277280.
37. Prandtl, L. (1933). Neuere Ergebnisse der Turbulenzforschung. Zeit. Vereines Deutscher
Ingenieure 77, 105114.
38. van Driest, E. R. (1956). On turbulent ow near a wall. J. Aeronaut. Sci. 13, 10071011.
39. Kolmogorov, A. N. (1941). The local structure of turbulence in an incompressible viscous
uid at very large Reynolds numbers. Doklady AN SSSR 30, 301304.
40. Prandtl, L. (1945). U ber ein neues Formelsystem fu r die ausgebildete Turbulenz. Nachr.
Ges. Wiss. Gttingen Math.-Phys. Klasse, pp. 619.
41. Batchelor, G. K. (1953). The Theory of Homogeneous Turbulence. Cambridge Univ.
Press.
42. Launder, B. E. and Spalding, D. B. (1972). Mathematical Models of Turbulence, Academic
Press, London.
43. Hanjalic , K., and Launder, B. E. (1972). A Reynolds stress model of turbulence and its
application to thin shear ows. J. Fluid Mech. 52, 609638.
44. Satake, S., and Kawamura, H. (1995). Large eddy simulation of turbulent ow in
concentric annuli with a thin inner rod. In Turbulent Shear Flows 9, (Ed. F. Durst, N.
Kasagi, B. E. Launder, F. W. Schmidt, K. Suzuki, and J. N. Whitelaw). Springer, Berlin,
pp. 259281.
45. Nikuradse, J. (1930). Widerstandgesetz und Geschwindigkeitsverteilung von turbulenten
Wasserstro mungen in glatten und rauhen R ohren. Proc. Third Intern. Congr. Appl. Mech.,
Stockholm, Part 1, pp. 239248.
46. Nikuradse, J. (1932). Gesetzma ssigkeiten der turbulenten Stro mung in glatten Rohren,
Ver. Deutsch. Ing. Forschungsheft, 356.
47. Nikuradse, J. (1933). Stro mungs gesetze in rauhen Rohren. Ver. Deutsch. Ing. Forschung-
sheft, 361.
48. Miller, B. (1949). The laminar lm hypothesis. Trans. ASME 71, 357367.
49. Robertson, J. M., Martin, J. D., and Burkhart, T. H. (1968). Turbulent ow in rough pipes.
Ind. Eng. Chem. Fundam. 7, 253265.
50. Lynn, S. (1959). Center-line value of the eddy viscosity. AIChE J. 5, 566567.
51. Lindgren, E. R., and Chao, J. (1969). Average velocity distribution of turbulent pipe ow
with emphasis on the viscous sublayer. Phys. Fluids 12, 13641371.
52. Hinze, J. O. (1963). Turbulent pipe ow. In Mcanique de la turbulence. Editions CNRS,
Paris, pp. 130165.
53. Blasius, H. (1913). Das A hnlichkeitsgesetz bei Reibungsvorga ngen in Flu ssigkeiten. Ver.
Deutsch. Ing. Forsch-arb. Ing.-wes., No. 131.
54. Freeman, J. R. (1941). Experiments upon the Flow of Water in Pipes and Fittings, ASME,
New York.
55. Prandtl, L. (1929). U ber den Reibungswiderstand Stro mender Luft. Ergebn. Aerodyn.
Versanst., Gttingen 3, 15. (First mentioned in I. Lieferung (1921), p. 136.)
56. Nunner, W. (1956). Wa rmeu bertragung und Druckabfall in rauhen Rohren. Ver. Deutsch.
Ing. Forschungsheft, 455.
57. Barenblatt, G. I. (1993). Scaling laws for fully developed turbulent shear ows. Part 1.
Basic hypotheses and analysis. J. Fluid Mech. 248, 513520.
58. Churchill, S. W. (2000). An appraisal of new experimental data and predictive equations
for fully developed turbulent ow in round tubes, in review.
59. Rothfus, R. R., and Monrad, C. C. (1955). Correlation of turbulent velocities for tubes and
parallel plates. Ind. Eng. Chem. 47, 11441149.
60. MacLeod, A. L. (1951). Liquid turbulence in a gasliquid absorption system. Ph.D.
Thesis, Carnegie Institute of Technology, Pittsburgh, PA.
61. Whan, G. A., and Rothfus, R. R. (1959). Characteristics of transition ow between parallel
plates. AIChE J. 5, 204208.
62. Senecal, V. E., and Rothfus, R. R. (1955). Transition ow of uids in smooth tubes. Chem.
Eng. Progr. 49, 533538.
63. Colebrook, C. F. (19381939). Turbulent ow in pipes with particular reference
to the transition region between smooth and rough pipe laws. J. Inst. Civ. Eng. 11,
133156.
64. Churchill, S. W. (1973). Empirical expressions for the shear stress in turbulent ow in
commercial pipe. AIChE J. 19, 375376.
65. Churchill, S. W. and Usagi, R. (1972). A general expression for the correlation of rates of
transfer and other phenomena. AIChE J. 18, 11211128.
66. Orszag, S. D. and Kells, L. C. (1980). Transition to turbulence in plane Poiseuille and
plane couette ow. J. Fluid Mech. 96, 159205.
67. Kim, J., Moin, P., and Moser, R. (1987). Turbulence statistics in fully developed channel
ow at low Reynolds numbers. J. Fluid Mech. 177, 133166.
68. Lyons, S. L., Hanratty, T. J., and McLaughlin, J. B. (1991). Large-scale computer
simulation of fully developed turbulent channel ow with heat transfer. Int. J. Num.
Methods Fluids 13, 9991028.
69. Rutledge, J., and Sleicher, C. A. (1993). Direct simulation of turbulent ow and heat
transfer in a channel. Part I. Smooth walls. Int. J. Num. Methods Fluids 16, 10511078.
70. Eckelmann, H. (1974). The structure of the viscous sublayer and the adjacent wall region
in turbulent channel ow. J. Fluid Mech. 65, 439459.
71. Churchill, S. W., and Chan, C. (1995). Theoretically based correlating equations for the
local characteristics of fully turbulent ow in round tubes and between parallel plates. Ind.
Eng. Chem. Res. 34, 13321341.
72. Groenhof, H. (1970). Eddy diffusion in the central region of turbulent ow in pipes and
between parallel plates. Chem. Eng. Sci. 25, 10051014.
73. Zagarola, M. V. (1966). Mean-ow scaling of turbulent pipe ow. Ph.D. Thesis, Princeton
University, Princeton, NJ.
74. Spalding, D. B. (1961). A single formula for the Law of the Wall. J. Appl. Mech. 28E,
455458.
75. Churchill, S. W. and Choi, B. (1973). A simplied expression for the velocity distribution
in turbulent ow in smooth pipes. AIChE J. 19, 196197.
76. Reichardt, H. (1951). Vollsta ndige Darstellung der turbulenten Geschwindigkeitsver-
teilung in glatten Leitungen. Zeit. angew. Math. Mech. 31, 201219.
77. Churchill, S. W., and Chan, C. (1995). Turbulent ow in channels in terms of local
turbulent shear and normal stresses. AIChE. J. 41, 25132525.
78. Churchill, S. W., and Chan, C. (1994). Improved correlating equations for the friction
factor for fully developed turbulent ow in round tubes and between identical parallel
plates, both smooth and rough. Ind. Eng. Chem Res. 33, 20162019.
79. Churchill, S. W. (1997). New simplied models and formulations for turbulent ow and
convection. AIChE J. 42, 11251140.
80. Churchill S.W. (1996). A critique of predictive and correlative models for turbulent ow
and convection. Ind. Eng. Chem. Res. 35, 31223140.
81. Kampe de Fe riet, J. (1948). Sur le coulement dun uide visqueux incompressible entre
deux plaques paralle lel indenies. La Houille Blanche 3, 509517.
82. Pai, S. I. (1953) On turbulent ow between parallel plates. J. Appl. Mech. 20, 109114;
(1953). On turbulent ow in circular pipe. J. Franklin Inst. 236, 337352.
83. Churchill, S. W. (1994). Turbulent Flows. The Practical Use of Theory. Notes, The
University of Pennsylvania, Philadelphia, Chapter 5.
84. Wei, T., and Willmarth, W. W. (1980). Reynolds-number effects on the structure of a
turbulent channel ow. J. Fluid Mech. 204, 5795.
85. Danov, S. N., Arai, N., and Churchill, S. W. (2000). Exact formulations and nearly exact
solutions for convection in turbulent ow between parallel plates. Int. J. Heat Mass
Transfer 43, 27672777.
86. Churchill, S. W. (1997). New wine in new bottles; unexpected ndings in heat transfer.
Part II. A critical examination of turbulent ow and heat transfer in circular annuli.
Therm. Sci. Eng. 5, 112.
87. Reichardt, H. (1951). Die Grundlagen des turbulenten Wa rmeu berganges. Arch. ges.
Wrmetechn. 2, 129142.
88. Rohsenow, W. M., and Choi, H. Y. (1961). Heat, Mass, and Momentum Transfer,
Prentice-Hall, Englewood Cliffs, NJ, p. 183.
89. Yahkot, V., Orszag, S. A., and Yahkot, A. (1987). Heat transfer in turbulent uids 1.
Pipe ow. Int. J. Heat Mass Transfer 30, 1522.
90. Page, F., Jr., Schlinger, W. G., Breaux, D. K., and Sage, B. H. (1952). Point values of eddy
conductivity and eddy viscosity in uniform ow between parallel plates. Ind. Eng. Chem.
44, 424430.
91. Churchill, S. W. (1997). New wine in new bottles; unexpected ndings in heat transfer.
Part III. The prediction of turbulent convection with minimal explicit empiricism. Therm.
Sci. Eng. 5, 1330.
92. Petukhov, B. S. (1970). Heat transfer and friction in turbulent pipe ow with variable
physical properties. Adv. Heat Transfer 6, 503562.
93. Seban, R. A., and Shimazaki, T. T. (1951). Heat transfer to a uid owing turbulently in
a smooth pipe with walls at constant temperature. Trans. ASME 73, 803808.
94. Lyon, R. N. (1951). Liquid metal heat-transfer coefcients. Chem. Eng. Progr. 47, 7579.
95. Reynolds, A. J. (1975). The prediction of turbulent Prandtl and Schmidt numbers. Int. J.
Heat Mass Transfer 18, 10551069.
95a. Kays, W. M. (1994). Turbulent Prandtl number Where are we? J. Heat Transfer, Trans.
ASME 116, 284295.
96. Jischa, M., and Rieke, H. B. (1979). About the prediction of turbulent Prandtl and
Schmidt numbers from modied transport equations. Int. J. Heat Mass Transfer 22,
15471555.
97. Notter, R. H., and Sleicher, C. A. (1972). A solution to the turbulent Graetz problemIII.
Fully developed and entry region heat transfer rates. Chem. Eng. Sci. 27, 20732093.
97a. Elperin, T., Kleeorin, N., and Rogachevskii, I. (1996). Isotropic and anisotropic spectra
of passive sealer uctuations in turbulent lm ow. Phys. Rev. E 53, 34313441.
98. Papavassiliou, D. V., and Hanratty, T. J. (1997). Transport of a passive scalar in a
turbulent channel ow. Int. J. Heat Mass Tranfer 40, 13031311.
99. Shaw, D. A., and Hanratty, T. J. (1978). Turbulent mass transfer to a wall for large
Schmidt numbers. AIChE J. 23, 2837.
100. Heng, L., Chan, C. and Churchill, S.W. (1998). Essentially exact characteristics of
turbulent convection in a round tube. Chem. Eng. J. 71, 163173.
100a. Yu, B., Ozoe, H., and Churchill, S. W. (2000). The characteristics of fully developed
turbulent convection in a round tube, in review.
101. Kays, W. M., and Leung, R. Y. (1963). Heat transfer in annular passages Hydrodynami-
cally developed turbulent ow with arbitrarily prescribed heat ux. Int. J. Heat Mass
Transfer 6, 537557.
102. Nusselt, W. (1909). Der Wa rmeu bergang in Rohrleitungen. Mitt. Forsch.-Arb. Ing.-wes.,
89, 17501787.
103. Dittus, R. W., and Boelter, L. M. K., (1930). Heat Transfer in Automobile Radiators of the
Tubular Type, University of California Publications in Engineering 2, 443461.
104. Sherwood, T. K., and Petrie, J. M. (1932). Heat transmission to liquids owing in pipes.
Ind. Eng. Chem. 24, 736748.
105. Coulson, J. M., and Richardson, J. F. (1954). Chemical Engineering, Vol. 1. McGraw-Hill
Book Co., New York, p. 166.
106. Colburn, A. P. (1933). A method for correlating forced convection heat transfer data and
a comparison with uid friction. Trans. AIChE 29, 174210.
107. Martinelli, R. C. (1947). Heat transfer to molten metals. Trans. ASME 69, 947959.
108. Lubarsky, B., and Kaufman, S. J. (1956). Review of experimental investigations of
liquid-metal heat transfer. Nat. Advisory Comm. Aeronaut. Report 1270, Washington,
D.C.
109. Sleicher, C. A., and Tribus, M. (1957). Heat transfer in a pipe with turbulent ow and
arbitrary wall-temperature distribution. Trans. ASME 79, 789797.
110. Notter, R. H., and Sleicher, C. A. (1971). The eddy diffusivity in the turbulent boundary
layer near a wall. Chem. Eng. Sci. 26, 161172.
111. Notter, R. H., and Sleicher, C. A. (1971). A Solution to the turbulent Graetz problem by
matched asymptotic expansions II. The case of uniform heat ux. Chem. Eng. Sci. 26,
559565.
112. Churchill, S.W. (1977). Comprehensive correlating equations for heat, mass, and momen-
tum transfer in fully developed ow in smooth tubes. Ind. Eng. Chem. Fundam. 16,
109116.
113. Prandtl, L. (1910). Eine Beziehung zwischen Wa rmeaustausch and Stro mungswiderstand
der Flu ssigkeiten. Phys. Z. 11, 10721078.
114. Thomas, L. C., and Fan, L. T. (1971). Heat and momentum analogy for incompressible
boundary layer ow. Int. J. Heat Mass Transfer 14, 715717.
115. Higbie, R. (1935). The rate of pure gas into a still liquid during short periods of exposure.
Trans. Amer. Inst. Chem. Engrs. 31, 365389.
116. Danckwerts, P. V. (1957). Signicance of liquid lm coefcients in gas absorption. Ind.
Eng. Chem. 43, 14601467.
117. Churchill, S. W., Shinoda, M., and Arai, N. (2000). A new concept of correlation for
turbulent convection. Thermal Sci. Eng., in press.
a
Progress in the Numerical Analysis of Compact
Heat Exchanger Surfaces
R. K. SHAH
Delphi Harrison Thermal Systems
Lockport, New York 14094
M. R. HEIKAL
University of Brighton
Brighton, United Kingdom
B. THONON AND P. TOCHON
CEA-Grenoble
DTP/GRETh
38054 Grenoble, France
I. Introduction
Compact heat exchangers (CHEs) are characterized by a large heat
transfer surface area per unit volume of the exchanger, resulting in reduced
space, weight, support structure and footprint, energy requirements, and
cost, as well as improved process design, plant layout, and processing
conditions, together with low uid inventory compared to conventional
designs such as shell-and-tube heat exchangers.
Somewhat arbitrarily, a gas-to-uid exchanger is referred to as a compact
heat exchanger if it incorporates a heat transfer surface with area density
above about 700 m`/m` (213 ft`/ft`) or the hydraulic diameter D
"
6 mm
(1/4 in.) for operating in a gas stream and above about 400 m`/m` (122
ft`/ft`) for operating in a liquid or phase-change stream. In contrast, a
typical process industry shell-and-tube exchanger has a surface area density
of less than 100 m`/m` on one uid side with plain tubes, and two to three
363
0065-2717/01 $35.00
Fic. 1. Plate-n geometries: (a) offset strip n, and (b) louver n.
Fic. 2. Tube-n geometries: (a) wavy n on round tubes, (b) louver n on round tubes, and
(c) louver n on elliptical tubes.
times that with high-n-density low-nned tubing. A typical plate heat
exchanger has about two times the heat transfer coefcient h or the overall
heat transfer coefcient U compared to that for a shell-and-tube exchanger
for water/water applications. For phase-change applications, even higher
heat transfer coefcients are achieved compared to a shell-and-tube ex-
changer. A laminar ow heat exchanger (also referred to as a meso heat
exchanger) has a surface area density on one uid side greater than about
3000 m`/m` (914 ft`/ft`) or 100 jmD
"
1 mm. A heat exchanger is
referred to as a micro heat exchanger if the surface area density on
one uid side is greater than about 15,000 m`/m` (4570 ft`/ft`) or 1 jm
D
"
100 jm. A compact heat exchanger is not necessarily of small bulk and
mass. However, if it did not incorporate a surface of high area density, it
would be much more bulky and massive.
Plate-n, tube-n, and rotary regenerators are examples of compact heat
exchangers for gas ow on one or both sides; whereas gasketed, welded,
brazed plate, and printed circuit heat exchangers are examples of compact
heat exchangers for liquid ows. Typical n geometries used in plate-n
and tube-n exchangers are shown in Figs. 1 and 2, and plate geometries
used in plate heat exchangers (PHEs) are shown in Fig. 3. The most
commonly used n geometries for plate-n exchangers are offset strip ns
364 . x. sn:n r1 :i.
Fic. 3. Plate heat exchanger plate geometries: (a) washboard, (b) zigzag, and (c) chevron.
and louver ns (referred to as multilouver ns in the automobile industry).
A considerable amount of experimental results are available in the
literature for ow and heat transfer phenomena in complex ow passages of
compact heat exchanger surfaces. Starting with the description of some of
the complex ows in compact heat exchanger surfaces, it is explained that
ows in compact heat exchanger surfaces are dominated by swirl and
vortices in uninterrupted ow passages, and by boundary layer ows and
wake regions (separation, recirculation, and reattachment) for interrupted
ow passages. Although unsteady laminar ows are relatively easy to
analyze, swirl and low Reynolds number turbulent ows are difcult to solve
numerically because of the lack of appropriate turbulence models. This is
the reason for the very slow progress in the numerical analysis of compact
heat exchanger surfaces.
A comprehensive experimental study of the performance of CHE surfaces
is very expensive because of the high cost of the tools needed to produce a
wide range of geometric variations. Numerical modeling, on the other hand,
has the potential of offering a exible and cost-effective means for such a
parametric investigation, with the added advantage of reproducing ideal
geometries and boundary conditions, and exploring the performance behav-
ior in specic and critical areas of ow geometry.
Thus, the objective of this work is to provide a comprehensive state-of-
the-art review on numerical studies of single-phase velocity and temperature
elds, and heat transfer and ow friction characteristics of compact heat
exchanger surfaces, as well as to provide specic comparisons to evaluate
the accuracy of numerical work where experimental data are available. The
surfaces include offset strip and louver ns used in plate-n exchangers,
wavy ns/channels used in tube-n exchangers and plate heat exchangers,
365 Nixric:i :N:ixsis or cnr sir:crs
and chevron (stamped) plates used in plate heat exchangers. First, a
description of some of the complex ows in such surfaces is presented. Next,
some highlights are presented for the numerical analysis of compact heat
exchanger surfaces. Since separation, recirculation, and reattachment as well
as large eddies and small-scale turbulence generation are common features
in CHE surfaces, a comprehensive but concise overview of turbulence
models/methods is presented next to illustrate the current capabilities and
limitations of these models. The rest of the paper covers numerical work
reported in the literature on the following CHE surfaces: offset strip ns,
louver ns, wavy ns/channels, and chevron trough plates. For each surface,
the numerical analysis is described in sufcient detail, and comparisons are
presented with experimental measurements where available. Thus, based on
the insight gained from numerical and experimental results, the performance
(uid ow and heat transfer) behavior of these CHE surfaces is discussed
and summarized. Also, briey mentioned is the proposed mode of research
that combines numerical analysis, sophisticated experimentation on the
small sample n geometries, and performance testing of actual heat ex-
changer cores.
II. Physics of Flow and Heat Transfer of CHE Surfaces
In this section, the current understanding of the physics of ow and heat
transfer in compact heat exchanger surfaces is described in order to set the
stage for the task of numerical analysis. The description is divided into
interrupted and uninterrupted complex ow passages, followed by charac-
terization into laminar unsteady and low Reynolds number turbulent ows.
A. IN1rii1rn Fiov P:ss:crs
The two most common interrupted n geometries are the offset strip n
and louver n geometries as shown in Fig. 1. Here the n surface is broken
into a number of small sections. For each section, a new leading edge is
encountered, and thus a new boundary layer development begins, and is
then abruptly disrupted at the end of the n offset length l. The objective
for such ow passages is not to allow the boundary layers to thicken, thus
resulting in the high heat transfer coefcients associated with thin boundary
layers. However, the interruptions create the wake region, and self-sustained
ow unsteadiness (see Figs. 4 and 5). As a result, the models based on the
boundary layer development are not adequate and do not accurately predict
366 . x. sn:n r1 :i.
Fic. 4. Flow phenomena in an offset strip n geometry.
Fic. 5. Flow phenomena in louver n geometry: (a) conventional louvers (section AA of Fig.
1b but not the same number of louvers); (b) and (c) CFD results of typical ow path in a louver
n array at Re
l
10 and Re
l
1600, respectively [113]; (d, see color insert) Flow visualization
in a louver n geometry (courtesy of Hitachi Mechanical Engineering Lab).
the heat transfer coefcients (Nusselt number Nu or Colburn factor j ) and
friction factors.
Separation, recirculation, and reattachment are important ow features in
most interrupted heat exchanger geometries [1]. Consider, for example, the
ow at the leading edge of a n of nite thickness. The ow typically
encounters such a leading edge at the heat exchanger inlet or at the start of
new ns, offset strips, or louvers. For most Reynolds numbers, a geometric
ow separation will occur at the leading edge because the ow cannot turn
the sharp corner of the n as shown in Fig. 4. Downstream from the leading
edge, the ow reattaches to the n. The uid between the separating
streamline (see Fig. 4) and the n surface is recirculating. This region is
called a separation bubble or recirculation zone. Within the recirculation
zone, a relatively slow-moving uid ows in a large eddy pattern. The
boundary between the separation bubble and the separated ow (along the
separation streamline) consists of a free-shear layer. Since free shear layers
are highly unstable, velocity uctuations develop in the free shear layer
downstream from the separation point. These perturbations are advected
downstream to the reattachment region, and there they result in an
increased heat transfer. The n surface in contact with the recirculation zone
is subject to lower heat transfer because of the lower uid velocities and the
thermal isolation associated with the recirculation eddy. The separation
bubble increases the form drag, and thus usually represents an increase in
pumping power with no corresponding gain in heat transfer. If the ow does
not reattach to the surface from which it separates, a wake results.
A free shear layer is also manifested in the wake region at the trailing edge
of a n element. Depending on the Reynolds number and geometry, the wake
from the upstream ns can have a profound impact on the downstream n
elements. The highly unstable wake can promote strong mixing that destroys
the boundary layers from the upstream ns, causing downstream heat
transfer enhancement. However, at low Reynolds numbers, or for very close
streamwise spacing of n elements, the shear layers might not be destroyed
or the next n element might be embedded in the wake of an upstream n
element. In such cases, the low velocity and near-n temperature of the wake
will have a detrimental effect on the downstream heat transfer. Wake
management in complex heat exchanger passages poses a difcult challenge,
especially at moderate and high Reynolds numbers where numerical simula-
tion is difcult to perform. Nevertheless, wake management appears to be the
key to further progress in improved heat exchanger surface design.
1. Offset Strip Fins
The ow phenomenon for the offset strip n geometry is described by
Jacobi and Shah [1] as follows. The ow unsteadiness begins at relatively
368 . x. sn:n r1 :i.
low Reynolds numbers (Re -100) as waviness in the wake of the n
elements. As the Reynolds number increases, oscillating ow develops in the
wake region. At higher Reynolds numbers, individual strips shed vortices at
regular intervals. These vortices are transverse to the main ow, and as they
are carried through the n array, they refresh the boundary layer to produce
a time-averaged thinner boundary layer. For deep arrays, vortex shedding
begins at the downstream ns and moves upstream as the ow rate is
increased (see Fig. 6 of [1]). At low Reynolds numbers (less than 400), ow
through the offset strip n geometry is laminar and nearly steady, and the
boundary layer effects dominate the heat transfer and friction. For inter-
mediate Reynolds numbers (roughly 400 Re 1000), the ow remains
laminar, but unsteadiness and vortex shedding become important. For
example, at Re 850, boundary layer restarting causes roughly a 40%
increase in heat transfer over the plain channel with vortex shedding causing
an additional 40% increase. Unfortunately, there is a commensurate in-
crease in the pressure drop due to boundary layer restarting and vortex
shedding. For Reynolds numbers greater than 1000, the ow becomes
turbulent in the array, and chaotic advection may be important in the low
Reynolds number turbulent regime. A factor of 2 or 3 increase in heat
transfer and pressure drop over plain ns can be obtained as a result of the
turbulent mixing. The important variables affecting the wake region identi-
ed are the strip length l, the n spacing s, and the n thickness o. The n
spacing s and the strip length l are responsible for the boundary layer
interactions and wake dissipation; the n thickness o introduces form drag
and also affects the heat transfer performance. Higher aspect ratios (s/b
or b/s), shorter strip lengths l, and thinner ns (o) are found to provide
higher heat transfer coefcients (Nu or j ) and friction factors f.
2. L ouver Fins
Flow through louver n geometries is similar to the ow through offset
strip n geometries, with boundary layer interruption and vortex shedding
playing potentially important roles. However, another important aspect of
louver n performance is the degree to which the uid follows the louvers.
At low Reynolds numbers (Re 200), boundary layer growth between
neighboring louvers becomes pronounced, and a signicant blockage effect
can result. Thus, at very low Reynolds numbers, the uid tends to ow
mostly between the ns forming the channel without following the louvers.
This ow is referred to as the duct ow (see Fig. 5a). At intermediate
Reynolds numbers, when the boundary layers are thinner, the ow tends to
more closely follow the louvers. This ow is referred to as the louver ow
(see Fig. 5a). At high Reynolds numbers (5000), the louvers act as a
rough surface, and the duct ow oscillates after the rst bank of louvers
in a n geometry. Effectively, at all Reynolds numbers, both the duct ow
and louver ow components exist, but the relative amount depends on the
Reynolds number. Sketches of possible ow patterns in louver ns are
shown in Figs. 5b and 5c, and a ow visualization picture of ow through
louvers is shown in Fig. 5d (see color insert).
To effectively exploit high heat transfer associated with short ow lengths
(louvers act as short at plates), it is important that the uid follow the
louver (louver ow) rather than passing between two ns (duct ow) to
obtain high Nu (and the resultant high f factors). The degree of the ow
deection by the louvers is determined by the relative hydraulic resistance
to the ow for the louver ow vs duct ow. This is dependent upon the n
geometry and the ow Reynolds number. The degree to which the uid
follows the louvers is sometimes called the ow efciency, which can be
dened as the mean angle of the ow divided by the louver angle. The
behavior of the ow efciency and its relation to heat transfer has been
examined by Cowell et al. [2]. For louver angles from 15 to 35 and n
pitch-to-louver length ratios (p
'
/l) from 1 to 2.5, the ow efciency drops
dramatically for Re
l
100. See typical results presented later in Fig. 14.
Flow efciency is nearly at its maximum by Re
l
200 and is almost
independent of the Reynolds number for higher ow rates. The ow does
not align with the louver array at the louver inlet and it takes a few louvers
to turn the ow. The heat transfer and pumping power performance is
strongly dependent on this ow-directing properties of the louver array.
Surfaces that cause the ow to follow the louvers, i.e., those with high ow
efciency, generally perform better than those in which the ow does not
follow the louvers. However, the exact heat transfer performance of these
surfaces is less well understood. Although the qualitative effect of the degree
of ow alignment on heat transfer is accepted, more accurate quantication
of these effects is needed. The degree of attening of the Stanton number
curve at low Reynolds numbers [2] should be examined further with a view
to determining, more accurately, the critical Reynolds number at which this
attening starts and the effect of the different geometrical parameters on this
phenomenon.
The experimental work of Chang and Wang [3] demonstrates clearly that
general correlating equations for predicting the heat transfer and pressure
drop performance of these surfaces are far from being achieved. This is
mainly due to the fact that the performance of these surfaces is a function
of a large number of geometric parameters and that a number of variants
of the n geometry are in use. Additionally, the manufacturing tolerances in
the production of the ns and variations in the test conditions also play a
part in producing different performances for supposedly similar ns at the
same ow conditions.
370 . x. sn:n r1 :i.
Fic. 6. (a) Plate-n exchanger, (b) tube-n exchanger with at ns. At section AA: (c) wavy
corrugated passage, and (d) wavy furrowed passage.
A novel approach for the optimization of the heat transfer performance
of ns with variable louver angles was presented by Cox et al. [4] as an
alternative to numerical modeling. Their method utilizes the Reynolds
analogy to obtain heat transfer performance characteristics from measure-
ment of the forces acting on the louvers in a 20 : 1 large-scale model of a
typical matrix. The model allows the angle of individual louver rows to be
driven automatically to specic angles. Force data logging and angle control
were performed automatically under computer control implementing opti-
mization strategies for the maximization of heat transfer performance as a
function of louver angles. Results for xed angle arrays based on known
geometries showed good agreement with previously obtained experimental
data based on thermal experiments on full-size matrix sections. Although
the model used had too few ns to be representative of an innite array, the
authors demonstrated the viability of such method for the optimization of
variable louver ns.
B. UNiN1rii1rn Coxiirx Fiov P:ss:crs
In this case, the heat transfer surface (the n or the prime surface) is not cut,
but convoluted such that the ow passage geometry does not allow the
boundary layer growth. For a plate-n geometry (Fig. 6a), two ow passages
are possible: wavy corrugated and wavy furrowed cross section (of Fig. 6a) as
shown in Fig. 6c and 6d. For plate heat exchangers, the cross section of plates
having intermating troughs (washboard design) is shown in Fig. 7a and those
of plates having chevron troughs are shown in Figs. 7b and 7c. The physics of
ow of these surfaces is discussed next. The Reynolds number for the plate
heat exchanger is commonly dened with one of two characteristic lengths:
the hydraulic diameter (D
"
four times the channel volume divided by the
total heat transfer surface area) or the equivalent diameter (D
twice the
plate spacing). The ratio D
/D
"
characterizes the surface extension ratio
(A
developed
/A
projected
), and it ranges from 1.1 up to 1.4 for industrial plates.
Fic. 7. Cross-section of two neighboring plates: (a) intermating troughs, (b) and (c)
chevron troughs.
1. Wavy Corrugated and Furrowed Channels
Corrugated and furrowed channels, as shown in Fig. 6, differ from plain
channels of constant cross-section. Wavy geometries provide little advan-
tage at low Reynolds numbers, and maximum advantage at transitional
Reynolds numbers. However, at higher Reynolds numbers, periodic shed-
ding of transverse vortices increases the Nusselt number with a considerable
increase in the friction factor. The following are important ow mechanisms
associated with wavy ns [1]: At low Re
(200), steady recirculation

zones form in the troughs of the wavy passages and heat transfer is not
enhanced. For higher Reynolds numbers, the free shear layer becomes
unstable; vortices roll up and are advected downstream, thus enhancing the
heat transfer. Transition to turbulence occurs at Re -1200, depending on
the geometry. It appears that chaotic advection may contribute to the heat
transfer in the transitional Reynolds number range. For Reynolds numbers
of 4000 and over, the ow is fully turbulent with very high pressure drops.
Thus, wavy channels provide higher heat transfer rates than plain channels,
but with higher pressure drops. Ali and Ramadhyani [5] and Gschwind et
al. [6] found that a streamwise Go rtler-like vortex system forms in the
transitional Reynolds number range. Although the impact on heat transfer
and pressure drop is not completely clear, such a vortex system is known to
increase heat transfer.
Wavy passages clearly offer heat transfer enhancement over plain channel
passages; however, they do not offer a performance advantage (heat transfer
relative to the pressure drop) over interrupted passages.
2. Intermating and Chevron Trough Plates
The high heat transfer coefcients obtained in plate heat exchangers are
a direct result of the corrugated plate patterns. A cross-section of one
corrugation along the ow length of an intermating trough design is shown
372 . x. sn:n r1 :i.
in Fig. 7a. The uid ows through wavy passages in which, depending upon
the Reynolds number Re, ow will separate in hills and valleys where
TaylorGo rtler vortices are generated. The ow separation and vortices are
responsible for the high performance of these surfaces. The increases in j, f,
and j/ f are generally higher than those for ow over a plate having a dimple
surface.
In chevron plate (see Fig. 3c with [ dened there) design, the ow
geometry is 3D and quite complex. The typical cross-sectional geometries
for chevron plates at [90 are shown in Figs. 7b and 7c. In other
geometries, the furrows in the bottom plate have continuous path at angle
[ and the mating top plate has furrows at an angle 180 [; thus, the uid
moves in different directions in the ow passages of mating plates. Because
of the criss-crossing (three-dimensional, 3D) nature of corrugations of the
mating plates, the secondary ows induced are swirl ows, which are
generally superior in terms of an increase in heat transfer over friction.
Hence, the relative performance of chevron plates is superior to all other
corrugation patterns and thus it is now most commonly used heat transfer
surface in plate heat exchangers. A much better understanding of the ow
patterns in chevron plates and subsequent enhancement is now available
[79]. Flow visualization by Focke and Knibbe [10] and Hugonnot [9] in
a larger-scale channel clearly shows recirculation areas downstream of the
corrugation edges. These areas are large at low Reynolds numbers, but the
transition to turbulent ow (which occurs at Re
200) reduces the size of

these areas. The recirculation area induces degradation of the kinetic energy
of mean ow and reduces heat transfer. Experimental information on local
heat transfer coefcient distribution has been obtained by Gaiser and
Kottke [11]. They indicate that the pitch-to-hydraulic diameter ratio and
the angle of corrugation have some inuence. They observed that the heat
transfer coefcient distribution is more homogeneous at high corrugation
angles, but there are still some weak areas.
Local measurements by laser pulse and thermographic analysis in a
two-dimensional (2D) channel [12] show poor heat transfer coefcients
downstream of the corrugation. It can also be seen that the local Nusselt
number tends to be more homogeneous while increasing the Reynolds
number. Be reiziat et al. [13] have measured the wall shear rate and observed
some similar recirculating areas in a 3D channel (: 60). These areas of
low heat transfer coefcient could be limited by a proper design of the
corrugation shape. More recently, Stasiek et al. [14] have performed
measurements of the local heat transfer coefcients by applying a thermoch-
romic method. Their results (corrugations angles 30) show that the heat
transfer coefcients are linked to the ow pattern and to the mixing intensity
in the channel with variations of _50% where measured.
C. UNs1r:nx L:xiN: vrsis Lov RrxNoins Nixnr TiniirN1 Fiov
In the numerical analysis as well as in the heat exchanger design, it is
essential to characterize whether the ow is unsteady laminar or low
Reynolds number turbulent ow. In case of numerical analysis, the unsteady
laminar ow is much easier and accurate to analyze compared to the low
Reynolds number turbulent ow. For the latter case, the large eddy
simulation model and the direct numerical simulation (refer to Section IV
for turbulence models) could be used for analyzing the ow. For heat
exchanger design, the pressure drop associated with the unsteady laminar
ow is lower than that for the low Re turbulent ow. As we understand
today, the following is the characterization of these ows.
Unsteady laminar ow may be seen in the wake of a bluff body, or a
streamlined body inclined to a ow, when the Reynolds number is sufcient-
ly low. The unsteady motion is orderly and contains structures (vortices)
that are generated continuously at a characteristic frequency and are of a
size closely related to the width of the body projected normal to the ow.
In contrast, low Reynolds number turbulent ow contains haphazard or
chaotic motions in addition to the underlying steady (or unsteady) ow.
These chaotic motions encompass a range of length and time scales, the
range being related to the Reynolds number, and so, unlike unsteady
laminar ow behind a bluff body, cannot be characterized by a single size
and frequency.
In DNS studies, we have access to time and space values; the statistical
analysis of uctuations can provide information on the ow structure. At
the moment, these statistical calculations can only give information for a
specic case, and the criteria found for the change in the regime cannot be
extended to other geometries. But such analysis could be generalized to
experiments and DNS calculations to nd objective criteria based on time
and space uctuations.
As far as we know, only a few authors have tried to apply these methods
to numerical modeling, but for experimental work this has already been
done for two-phase ow structures. It is essential that sophisticated experi-
mentation and careful numerical analysis be conducted to characterize what
type of ows occur at low Re in interrupted n geometries, wavy ns/
channels, and chevron plates for its far-reaching impact on compact heat
exchanger design.
Classical theories of turbulence require very high Reynolds number as a
prerequisite for the occurrence of the phenomenon. However, recent obser-
vations from compact heat exchanger studies (Jacobi and Shah [1] or Focke
and Knibbe [10]) seem to suggest that the Reynolds number need not be
very high to have turbulence. This observation, which we will refer to as low
374 . x. sn:n r1 :i.
Reynolds number turbulence, certainly deserves more studies in the context
of the classical notions of turbulence. This is particularly necessary since
some of the standard turbulence modeling procedures are based on the
assumption of an innitely high Reynolds number.
III. Numerical Analysis
Numerical analysis of compact heat exchanger surfaces started about 20
years ago with signicant progress in this time frame. However, the
problems analyzed numerically are simpler models of the real complex ows
within the CHE surfaces. As a result, many of the phenomena observed in
CHE surfaces by ow visualization and partial/full-scale testing have not yet
been duplicated by the numerical analysis. In spite of this, CFD (computa-
tional uid dynamics) codes allow some modeling of 2D or 3D ows for
better understanding of the basic mechanisms of heat transfer and pressure
drop in CHE surfaces. These methods can be used as guidelines for a
parametric design study and the study of new geometries.
The most common numerical approaches used for the analysis of CHE
surfaces are the nite difference and nite volume methods, and only in
some cases, nite element. The algorithm used for the solution of the partial
differential equations is the pressure-based method because of the low Mach
numbers in CHEs. Also, in most analyses, structured grid is used for the
analysis. Unstructured, adaptive, and composite grids have been rarely used
in analyzing compact heat exchanger surfaces. Refer to Heikal et al. [15] for
the governing equations, solution algorithm, 2D and 3D models for numeri-
cal meshes, boundary conditions, and the determination of performance
parameters (such as Nu, St, Re, h, f ) for multilouver n geometries.
As mentioned earlier, the ow and heat transfer performance of CHE
surfaces is mainly dictated by the boundary layer behavior over the
interruptions or in complex ow passages, and ow separation, recircula-
tion, reattachment, and vortices in the wake region. Careful consideration
must therefore be given to the grid used. Adequate grid renement is needed
to capture the boundary layer growth and separation, and this is not always
possible with moderate computing resources. One must also determine
whether steady-state solutions are adequate or a time-dependent model is
needed to capture the correct ow behavior. The decision to perform 2D or
3D modeling must also be made and assessed against grid size requirement
and speed of computation. However, the accurate prediction of local and
overall Nu (or j ) and f factors for CHE surfaces will only be possible by
analyzing time-dependent 3D ows.
A. Mrsn GrNr:1ioN
The numerical solution of the governing CFD equations in arbitrarily
shaped regions requires the generation of numerical grids. A grid is a
discrete representation of the continuous eld phenomena being modeled. It
is the structure on which the numerical solution is built (Thompson et al.
[16]) and should therefore accurately represent the geometrical congur-
ation of the domain and the physics of the problem. The mesh density and
structure have a signicant inuence on the accuracy and stability of the
solution. The optimum mesh should be ne enough to reduce the discretiz-
ation error and resolve ow and heat transfer details, especially in the areas
of sharp gradients. It is important to keep the grid as orthogonal as possible,
and to avoid cell aspect ratios signicantly larger or smaller than unity.
The nite-difference CFD algorithms for complex geometries require grid
generation techniques that transform a curvilinear nonuniform grid into a
uniform rectangular one in the computational space (structured grids). The
boundary conditions can be accurately represented, in this case, as some
coordinate line (or surface in 3D) coincides with a boundary of the physical
region (body-tted coordinates). Although the body-tted structured grids
are widely used for both nite-difference and nite-volume algorithms, the
meshing of the complex geometries found in most compact heat exchanger
surfaces can consume considerable time and effort. Finite-volume methods
enable the use of unstructured grids, which allow more meshing exibility.
Also, Cartesian grids with boundary cells aligned to the surface are among
current trends in grid generation (Anderson [17], Melton [18]). The grid
generation strategy is determined according to the size and location of ow
features such as shear layers, separated regions, boundary layers, and mixing
zones. For wall-bounded ows, the grid size at the wall can affect the
accuracy of the computed shear stress and heat transfer coefcient. One
must address the specic requirements of the wall functions used (see
Section IV). For example, using a classical kc model, the grid point closer
to the wall must be inside the buffer zone of the boundary layer. Because of
the strong interaction of the mean ow and turbulence, the numerical results
for turbulent ows tend to be more susceptible to grid dependency than
those for laminar ows.
B. BoiNn:x CoNni1ioNs
The boundary conditions are very important for computational uid
dynamic techniques as they govern the solutions. Usually, inlet conditions
are uniform bulk velocity (based on the specied ow rate) and temperature
or xed velocity/temperature distribution, although time-dependent condi-
376 . x. sn:n r1 :i.
tions are becoming more common. No-slip velocity conditions are used at
wall as a ow condition, whereas uniform temperature or heat ux at wall
is specied as a thermal one. For the outlet, a zero spatial derivative in a
direction normal to the boundary is specied (Shaw [19]). As the pressure
is obtained by the solution of the Navier-Stokes equation, a uniform
arbitrary pressure is usually xed at the outlet of the computational domain.
However, this condition is sometimes unsuitable when the reattachment
point of a separated ow is near the outlet or when an eddy structure exists
through it. For these cases, special conditions are used: uniform streamwise
pressure gradient (Mercier and Tochon [20]) or Sommerfeld radiative
conditions (Orlanski [21]) in the outer part of the domain. For the lateral
part, two conditions could be used: periodicity or symmetry (free slip
condition). The former is based on a direct pressure coupling between the
two lateral sides and is well suited for deviated ows (louver ns, for
example). The latter is usually used for spatially developed ows inside a
symmetrical geometry (offset strip ns, for example). To simulate fully
developed ows, a cyclic condition could be used. In this case, the velocity
and temperature proles at the outlet of the domain are placed at the inlet
at each time step.
With the use of (cu/cx) 0 and v 0 for the outow condition, a longer
wake region downstream of the surface is required to get reasonable results,
especially for unsteady ow. For example, to simulate ow past a cylinder
at Re 300, the ow becomes unsteady with vortices in the downstream
region. If we set the length of the downstream wake region as smaller than
20D (D is the cylinder diameter), the simulation may diverge or the results
for ow performance (such as Strouhal number, ow friction, or pressure
drop) may differ from what they should be by using the preceding outow
condition. This is because the actual ow cannot meet the condition of
(cu/cx) 0 and v 0 at the boundary. Thus, if we use (cu/cx0 and
v 0) for the downstream boundary, the downstream wake region would
be longer (for example, 30D) for more accurate results. If we use the
boundary layer approximations for the outow condition, the downstream
wake region in the foregoing example can be reduced to lower than 20D for
accurate results, thus reducing both the number of grid cells and the
computation time. The 3D boundary layer momentum and energy equa-
tions for the outow condition are
c
ct
(ju)
c
cx
(ju`)
c
cy
(jvu)
c
cz
(jwu)
cP
cx

c
cy
j
cu
cy
c
cz
j
cu
cz
c
ct
(jv)
c
cx
(juv)
c
cy
(jv`)
c
cz
(jwu)
c
cy
j
cv
cy
c
cz
j
cv
cz
c
ct
(jw)
c
cx
(juw)
c
cy
(jvw)
c
cz
(jw`)
c
cy
j
cw
cy
c
cz
j
cw
cz
c
ct
(jc
T )
c
cx
(jc
uT )
c
cy
(jc
vT )
c
cz
(jc
wT )
c
cy
z
cT
cy
c
cz
z
cT
cz
. (1)
These equations are solved in the last cell near the boundary, and hence the
following terms are not present in the preceding equations (they are present
when solving the boundary layer equations in the interior domain):
cù
cx`
0,
c`v
cx`
0,
c`w
cx`
0,
cP
cy
0,
cP
cz
0,
c`T
cx`
0. (2)
The pressure at the outow boundary is assumed uniform and used to
compute the pressure correction P' at all interior grid points. Then the
corrected P at the last node before the outow boundary is used to solve
Eq. (3), the nite difference form of the foregoing boundary layer equations,
for the last cell near the boundary to get a better value of u at the outow
boundary node. This iteration between correction P' and rened value of u
continues until the convergence, yielding the correct values of u and pressure
eld. Kieda et al. [22] implemented the boundary layer approximation for
the velocity components, and Xi [23] extended the concept by the compu-
tation of the pressure eld.
The foregoing boundary conditions and the computational domain em-
ployed by Xi and Shah [90] are shown as an example in Fig. 8 for a 3D
analysis of the offset strip n geometry.
C. Soii1ioN Aicoi1nx :Nn Nixric:i Scnrxr
The delity of the results from computational uid dynamics techniques
for turbulent ows is largely determined by the solution algorithm and the
numerical scheme. This is true for Reynolds averaged numerical simulation
(RANS), large eddy simulation (LES), and direct numerical simulation
(DNS). LES methods need to solve accurately motions over a wide range
of scales (although not wide as for DNS) and require spatial and temporal
discretization schemes that are at least second-order accurate. Generally, the
time discretization schemes used are the AdamsBashford, RungeKutta,
or Leap Frog schemes. Mostly explicit schemes are used except viscous
terms, which are treated implicitly when they require smaller time steps. For
ows of practical interest, mostly nite-volume methods are used, since
378 . x. sn:n r1 :i.
Fic. 8. Boundary conditions and the computational domain for an offset strip n geometry
analyzed by Xi and Shah [90].
nite-element applications need a higher cost per node in terms of computer
memory and CPU time are requirements. Moreover, the sub-grid-scale-
stress (SGS) effect is essentially dissipative. So, mostly second-order central
differencing schemes are used for the discretization of convection terms,
since the numerical error is dispersive. Third-order (such as the Quick
scheme described by Leonard [24]) or fth-order upwind differencing
schemes are used too, but Moin [25] nds them too dissipative in ow-past-
cylinder calculations.
The DNS calculations are usually based on the nite difference or spectral
element scheme for the calculation of ows in CHEs, since their geometry
is very complex. In particular, the pseudospectral method, which is com-
monly used for the DNS of homogeneous ows, cannot be used for CHE
calculations because of the presence of solid walls. And, as for LES models,
explicit schemes or part implicit schemes are used for the DNS methods.
IV. Turbulence Models
In the design of compact heat exchangers, complex geometries are often
used to promote high heat transfer rates. These geometries involve non-
straight channels or ducts as described in Section II earlier. The complex
ow phenomena in such systems have profound inuence on the heat
transfer and ow friction. Indeed, even though the mean ow is represented
by a Reynolds number based on the mean velocity and hydraulic diameter,
Lane and Loehrke [26] and Ota et al. [27] dened a Reynolds number
based on half the height of the surface roughness prole to describe the
separated ow. According to these authors, when this specic Reynolds
number is greater than 300, for discrete rib roughness, the ow separates at
the leading edge of the ns in a laminar manner but reattaches in a turbulent
way. This phenomenon creates coherent eddy structures, which increase the
local heat transfer. Thus, because of the complex geometries of compact heat
exchangers, even for ows at low Reynolds numbers based on the hydraulic
diameter (at so-called classical laminar or transitional ows), some turbu-
lent phenomena can appear. Although the global turbulence that is
observed in a pipe ow for Re 2300 is also found in many heat ex-
changers, the quasi-coherent structures of the type of von Ka rma n streets
are more important in the CHE surfaces. The selection of an appropriate
models to calculate low-Re turbulent ow, transition ow, and turbulent
ows is one of the key factors in obtaining reliable prediction of ow friction
and heat transfer in CHEs. For engineering applications, obtaining empiri-
cal data for heat transfer and uid ow is quite cumbersome and costly; as
a result, the scope of evaluating various geometries and operating par-
ameters is limited. Therefore, the development of reliable computational
techniques is required to evaluate heat transfer rates and pressure drops for
the CHE surfaces.
In this section, the most commonly used turbulence models/methods for
computational uid dynamics analyses are described. Although not all
turbulence models are commonly used in CHE analysis, it is imperative that
we provide a comprehensive but concise review to challenge the CFD and
CHE researchers to advance the CFD technology for CHE applications.
The methods for calculating turbulence can be divided into the following
three broad categories.
380 . x. sn:n r1 :i.
(
Reynolds averaged NavierStokes (RANS) models of turbulence such as
the kc model or Reynolds stress closure model (RSM), which consists
of the second moment turbulence modeling
(
L arge eddy simulation (L ES) techniques
(
Direct numerical simulation (DNS) techniques
For more information on turbulence models, refer to Rodi [28], Halbaeck et
al. [29], and Wilcox [30]. We introduce the equations governing the total
(mean plus uctuating) ow and the heat transfer, then describe the
turbulence models/methods.
A. RrxNoins Avr:crn N:virS1oxrs (RANS) Eqi:1ioNs
In this method [31], the instantaneous solution variables in the governing
equations (NavierStokes equations, continuity, and energy) are decom-
posed into their mean and uctuating components. For an incompressible
uid, the instantaneous velocity obeys the following equation (in Cartesian
tensor form):
cu
'
ct

c
cx
'
(u
'
u
'
)
1
j
cp
cx
'
c
cx
'
_
v
cu
'
cx
'
cu
'
cx
'
. (3)
The velocity components and scalar quantities such as pressure are decom-
posed as
u
'
U
'
u'
'
(4)
p Pp', (5)
where U
'
and P are the mean components and u'
'
and p' the uctuating
components. By substituting u'
'
and p' of Eqs. (4) and (5) into Eq. (3) and
time (or ensemble) averaging, the mean velocity equations can be written as
DU
'
Dt

1
j
cP
cx
'
c
cx
'
_
v
cU
'
cx
'
cU
'
cx
'
2
3
o
''
cU
'
cx
'
c
cx
'
(u'
'
u'
'
). (6)
The mean continuity equation for an incompressible uid can be written as
c
cx
'
(jU
'
) 0. (7)
Equations (6) and (7) are called the Reynolds-averaged NavierStokes
equations. They have the same form as the laminar NavierStokes equa-
tions with the velocities and other variables representing time-averaged (or
ensemble-averaged) values. However, an additional term appears in Eq. (6),
which represents the effect of turbulence and is called the Reynolds stress
tensor: (u'
'
u'
'
). This term needs to be modeled in order to close the system
of equations. Several approaches already exist for this purpose: (1) eddy
viscosity models (EVMs), (2) algebraic stress models (ASMs), and (3)
Reynolds stress transport models (RSMs). These are now briey described.
All these approaches require a special treatment of turbulent ows near the
wall, and some of the models for wall effects are summarized in Subsection
4 of this section.
1. Eddy V iscosity Models (EV M)
This is the most common way to model the Reynolds stresses. It is based
on the Boussinesq hypothesis, which assumes that the Reynolds stresses are
related to the mean velocity gradients by the empirical formula
u'
'
u'
'
2
3
ko
''
v
'

cU
'
cx
'
cU
'
cx
'
(8)
where
o
''
0
1
for i j
for i j
, (9)
and k is the turbulent kinetic energy. Equation (8) is valid for incompressible
uid only; Eq. (7) is already incorporated in Eq. (8). In this approach, a
turbulent viscosity v
'
is introduced and needs to be determined. The
advantage of this method is the relatively low computational cost associated
with the calculation of the turbulent viscosity using one of the following
three methods. The rst method does not need an additional equation,
whereas the other two need one and two additional equations, respectively.
a. Zero-Equation Models In these models, no additional differential equa-
tions are needed to obtain the turbulent viscosity v
'
, which is dened as a
function of the mean ow. The BaldwinL omax model [32] is one such
model based on the Prandtl mixing-length model. In this model, two
different expressions are given for the turbulent viscosity.
(
For the inner zone (0 y y
'
),
v
'
l`c (10)
where the mixing length l and the vorticity c are given by
l Ky
1 exp
y
A
(11)
c
,
cU
'
cx
'
cU
'
cx
'
cU
'
cx
'
cU
'
cx
'
cU
'
cx
'
cU
'
cx
'
`
(12)
with A26, and the von Ka rma n constant K0.42.
382 . x. sn:n r1 :i.
(
For the outer region (y
'
y o),
v
'
F(U`
'
U`
'
U`
'
) (13)
where y is the distance normal to the wall, y
'
the inner zone (viscous
sublayer) thickness, and o the boundary layer thickness. More complex
models for the outer region are also published.
The main drawback of these models is that the mixing length is only
computed inside the boundary layer of the ow. It is therefore difcult to
generalize the model for the complex geometries found in CHEs. So, this
model is rarely used for CHE applications.
b. One-Equation Models In this method, v
'
of Eq. (10) is given by an
additional transport equation. Usually, it is the transport equation for the
turbulent kinetic energy k. In this case, v
'
k`l, and an ad hoc specica-
tion for l is still needed. Spalart and Allmaras [33] have developed an
example of such models.
This model was designed specically for aerospace applications involving
wall-bounded ows and has been shown to give good results for boundary
layers subjected to adverse pressure gradients. In its original form, the
SpalartAllmaras model is a low Reynolds number model, requiring the
viscous-affected region of the boundary layer to be properly solved. When
the mesh resolution is not sufciently ne, some commercial CFD software
uses specic wall functions. However, one-equation models are often
criticized for their inability to accommodate rapid changes in the length
scale, such as might be necessary when the ow changes abruptly from a
wall-bounded to a free shear ow, a phenomenon that is encountered
frequently in CHEs. For this reason, this model is rarely used in CHE
applications.
c. Two-Equations Models In these models, two separate transport equa-
tions determine independently the turbulent velocity and length scales.
These models are usually implemented as kc, kc, or kl. In the kc
models, the two transport equations are for the turbulent kinetic energy k
and its dissipation rate c. The kc models are based on the Boussinesq
hypothesis and assume that the turbulence is isotropic. As a result, they are
expected to perform poorly in curved geometries and ows with directional
inuence. For example, they cannot calculate the ows shown later in Figs.
15 and 16 because of the rotational body forces. The standard kc model
is used for practical engineering ow calculations as proposed by Launder
and Spalding [34]. This is an economic and numerically robust model,
which gives reasonably accurate results for a wide range of turbulent ows
that do not involve too much rotational ow. Nevertheless, it is commonly
used in complex ows of industrial applications and heat transfer simula-
tions that have rotational ows. It is a semiempirical model whose strengths
and weaknesses have become known [35]. Several improvements have been
made to obtain better performance with this model.
The Standard kc Model. The standard kc model proposed by
Launder and Spalding [34] determines the turbulent kinetic energy k and
its dissipation rate c from the following transport equations:
Dk
Dt

c
cx
'
_
v
v
'
o
'
ck
cx
'
G
'
G
'
c (14)
Dc
Dt
c
cx
'
_
v
v
'
o
'
cc
cx
'
C
'
c
k
(G
'
C
`'
G
'
) C
`'
c`
k
. (15)
In these equations, G
'
represents the generation of turbulent kinetic energy
by the mean velocity gradients, G
'
is the generation of turbulent kinetic
energy by buoyancy, C
'
, C
`'
, C
`'
are constants, and o
'
, o
'
are the turbulent
Prandtl numbers for k and c, respectively. The turbulent viscosity is given by
v
'
C
`
k`
c
(16)
where C
`
is a constant often equal to 0.09 in practical applications.
The standard kc model is commonly used for analyzing ow inside most
CHE geometries; for example, corrugated wavy channels (Hugonnot [9]
and Ergin et al. [130]), louver ns (Achaichia et al. [109]), offset strip ns
(Michallon [87]), or chevron trough plates (Fodemsky [151]), Ciofalo et al.
[152], or Hessami [153]).
The RNG kc Model. The RNG-based kc turbulence model is derived
from the instantaneous NavierStokes equations using a rigorous statistical
technique called renormalization group theory. The model equations are
Dk
Dt

c
cx
'
_
v
'
o
'
ck
cx
'
G
'
G
'
c (17)
Dc
Dt
c
cx
'
_
v
'
o
'
cc
cx
'
C
'
c
k
(G
'
C
`'
G
'
) C
`'
c`
k
R. (18)
Here the term R is given by
R
C
`
p`(1 p/p
"
)
(1 [
"
p`)
c`
k
, (19)
where p is given by
p
k
c _
cU
'
cx
'
cU
'
cx
'
cU
'
cx
'
`
. (20)
384 . x. sn:n r1 :i.
The quantities [
"
and p
"
are constants having values as 0.012 and 4.38,
respectively. For the RNG kc model, the eddy viscosity expression is
v
'
v
_
1
C
`
v
` k
c`
`
. (21)
The RNG theory and its application to turbulence are described by Yakhot
and Orszag [37]. The scale elimination procedure in the RNG theory results
in a differential equation for turbulent viscosity, which is integrated to
obtain an accurate description of how the effective turbulent transport varies
with the effective Reynolds number and near-wall ows. In the high
Reynolds number limit, the expression of turbulent viscosity is the same as
in the standard kc model.
The RNG model is similar in form to the standard kc model, but
includes the following improvements:
(
The RNG model has an additional model term R in the c equation that
signicantly improves the accuracy for rapidly strained ows
(
The effect of swirl on turbulence is included in the RNG model,
enhancing accuracy for swirl ows
(
The RNG theory provides an analytical formula for turbulent Prandtl
numbers, whereas the standard kc model uses constant values
(
Whereas the standard kc model is a high Reynolds number model,
the RNG theory provides an analytically derived differential formula
for effective viscosity that takes into account low Reynolds number
effects
Thus, the RNG model is more accurate and reliable for a wider class of
ows than the standard kc model. It is well suited for corrugated n
surfaces where the hydraulic diameters are small and the Reynolds numbers
are low. Because this kind of modeling is relatively recent, few computations
on CHE geometries have been performed. Sunde n [157] used this model for
chevron trough plates geometry and obtained more accurate results than
those with the standard kc model.
However, the accuracy for predicting the turbulent ows using the RNG
model is reported as poorer than that for two other kc models for vortex
shedding behind the bluff objects, such as square rods and circular tubes in
a heat exchanger. In these cases, the separated ows are not well predicted
as in chevron or corrugated plate geometry. Such vortex shedding has low
frequency modulations. Saha et al. [38] compared three turbulence models
to capture the essence of time-averaged ow quantities in a vortex shedding
dominated ow eld through the turbulence models in two dimensions.
They used the Launder and Spalding [34] standard kc model, the
KatoLaunder kc model [39], and the RNG kc model of Yakhot et al.
[40]. In terms of the parameters such as the Strouhal number and lift and
drag coefcients, the predictions due to the KatoLaunder and the standard
kc models were close to each other, and reasonably close to experiments
of Lyn et al. [41]. However, the predictions due to RNG kc models were
not close to the experimental values. A detailed comparison of velocity
proles revealed the KatoLaunder model to have the closest agreement
with the experiments. A comparison between the computations and the
experiment were also made for the time averaged kinetic energy variation
along the centerline of the domain of interest. The KatoLaunder model
predicted the peak value of turbulent kinetic energy in good agreement with
the experiments. The peak value of the turbulent kinetic energy due to the
RNG kc model showed a signicant departure from the experimental
value. Thus, the comparison of these turbulence models indicates that the
accuracy of the models may depend upon the geometry investigated; a more
thorough investigation is needed for establishing the utility of specic
models for specic geometries.
The Realizable kc Model. The realizable kc model (Shih et al. [42])
is a recent development that satises certain mathematical constraints on
the normal stress consistent with the physics of turbulent ows contrary to
the standard and RNG kc models. In practice, the principle differences
between the realizable model and the other kc model are that the former
contains a different formulation for the turbulent viscosity and that another
transport equation has been used for the dissipation rate. This equation has
been derived from an exact equation for the transport of the mean-square
vorticity uctuation. The transport equations for the turbulent kinetic
energy k and for the dissipation rate c are
Dk
Dt

c
cx
'
_
v
v
'
o
'
ck
cx
'
G
'
G
'
c (22)
Dc
Dt
c
cx
'
_
v
v
'
o
'
cc
cx
'
C
'
c
k
C
`'
G
'
C
`'
c`
k vc
C
Sc (23)
where S is a scalar value of the strain tensor. The turbulent viscosity v
'
is
calculated from Eq. (16), but C
`
is no longer a constant. It is given by
C
`
1
A
"
A
'
U*k
c
(24)
386 . x. sn:n r1 :i.
where A
"
4.04, and A
'
and U* [U* dened in Eq. (29)] are functions of
both the mean strain and rotation rates, the angular velocity of the system
rotation, and the turbulence eld (k and c) [42].
This model is more accurate for predicting in the spreading rate of planar
or round jets than the standard kc model. It is likely to provide superior
performance for ows involving rotation, boundary layers under strong
adverse pressure gradients, separation, and recirculation. At present, this
model still requires validation for industrial applications and remains a eld
of research.
The k
Model for L ow Reynolds Number. Hwang and Lin [43]

proposed an improved low Reynolds number k
turbulence model to
describe thermal eld. By adopting the Boussinesq approximation, the
turbulent heat ux is approximated as:
u'
'
T ' :
'
cT
cx
'
(25)
where :
'
is the thermal diffusivity and T
is the mean temperature.

In the standard kc model, :
'
is adopted to be proportional to the ratio
of the turbulent viscosity and the Prandtl number. Most calculations of
CHE have used constant turbulent Prandtl number Pr
'
; it might be more
appropriate to use a variable one for CHE. See Kays and Crawford [44] for
variable Pr
'
. In the k
model, the thermal diffusivity is expressed as a

function of the velocity scale and of the thermal and mechanical time scales
to take into account the variations of the turbulent Prandtl number and the
fact that the thermal diffusivity is not necessarily related to the eddy
diffusivity. Moreover, to give a correct asymptotic behavior in the vicinity
of the wall, the dissipation rate is decomposed into two parts in this model
(Jones and Launder [36]):
c c` c` (26)
Here c` is the dependent variable in the dissipation rate equation (23) and
c` 2v(ck/cx
'
)`. With this decomposition, c` reaches zero at the wall and
c` equals c for y15. This model has been validated on experimental and
DNS data for duct ows.
For CHE applications, the preceding linear eddy viscosity model by
Launder and Sharma [45] simplies the wall boundary condition of the
dissipation equation. However, one of its main limitations is that it gives far
too large near-wall length scales in impinging or recirculation ows. As a
remedy to this, Yap [46] introduced an extra source term into the dissi-
pation equation. With Yaps correction, near-wall turbulent length scale can
be reduced in a separated ow, particularly near the ow reattachment
point around where the maximum heat transfer occurs. Some nonlinear
eddy viscosity (kc) models have been developed to essentially capture the
nonisotropic behavior of the ows as encountered in shear ow, recircula-
tion ow, or swirling ow [47].
For CHE applications, low Reynolds number models such as the k
model have been used for internal ows inside chevron trough plates
(Ciofalo et al. [152], Hessami [153], or Sunde n [157]) and corrugated wavy
channels (Yang et al. [126] and Ergin et al. [132]). However, it still requires
validation for shear ows, which are encountered in offset strip n and
louver n geometries.
Other Common Eddy V iscosity Models. Many other models, based on
the eddy viscosity concept, exist such as the kc model and the shear stress
transport (SST) model. For more details, refer to Wilcox [30] or Menter
[48].
2. Algebraic Stress Models (ASM) or Nonlinear Eddy V iscosity Models
(NL EVMs)
The ASM or nonlinear eddy viscosity model (NLEVM) [4951] is an
intermediary model between the EVM and the RSM. In this model,
Reynolds stresses are represented as a tensor polynomial expansion in terms
of the mean strain rate and rotation rate tensors. The expansion coefcients
are determined from the simplied differential Reynolds stress transport
equation. The ASM is less sensitive to rotation than the EVM and need less
computational cost than the RSM, but this kind of model is still not
commonly used in industrial applications because they require too many
empirical parameters, which have to be adjusted for each application.
3. Reynolds Stress Models (RSM)
In this method, the Reynolds stress is determined by solving the differen-
tial transport equations for each components of Reynolds stresses (Launder
et al. [52]; Gibson and Launder [53] ; Launder [54]). Taking moments from
the exact momentum equations may derive the exact form of the Reynolds
stress transport equation. The momentum equations are multiplied by a
uctuating property and then are time-averaged. But the Reynolds stress
transport equation contains several unknown terms that need to be modeled
388 . x. sn:n r1 :i.
in order to close the equations:
c
ct
(u'
'
u'
'
)
(1)
c
cx
'
(U
'
u'
'
u'
'
)
(2)
c
cx
'
_
u'
'
u'
'
u'
'
P
j
"
(o
''
u'
'
o
''
u'
'
)
(3)
c
cx
'

v
c
cx
'
(u'
'
u'
'
)
(4)
u'
'
u'
'
cU
'
cx
'
u'
'
u'
'
cU
'
cx
'
(5)
[(g
'
u'
'
0
' g
'
u'
'
0
')
(6)
P
j
"

cu'
'
cx
'
cu'
'
cx
'
(7)
2v
cu'
'
cx
'
cu'
'
cx
'
(8)
2D
'
(u'
'
u'
"
c
''"
u'
'
u'
"
c
''"
)
(9)
. (27)
Here, [ the volumetric expansion coefcient, 0
' is the uctuating uid

temperature, and D the rotation vector. In the preceding equation, the
following terms do not require any models: (1) local time derivative; (2)
convection; (4) molecular diffusion; (5) stress production; and (9) production
by system rotation. However, in order to close the equation set, the
following terms need to be modeled: (3) turbulent diffusion due to triple
correlations and pressure uctuations; (6) buoyant production; (7) pressure
strain; and (8) dissipation.
Since the RSM takes into account the effects of streamline curvature,
swirl, rotation, and rapid changes in the strain rate in a more rigorous
manner than other RANS models, it supposedly gives more accurate results
for complex ows. But the RSM predictions are still limited by the closure
assumptions used for various terms of Eq. (27), especially the pressure-strain
and the dissipation-rate terms, designated as (7) and (8).
The RSM does not always yield results superior to the simpler models in
all classes of ows. Compared with the kc models, the RSM requires
additional memory and CPU time because of a large number of transport
equations computed. Furthermore the RSM could need more iterations
than kc models because of the strong coupling between the Reynolds
stresses and the mean ow.
The use of the RSM is interesting when the ow features studied are the
results of strong anisotropy in the Reynolds stresses (cyclone ows, rotating
ows, etc.). At present, these models are still not used for industrial
applications and remain a eld of research. Indeed, they are based on too
many adjustable parameters (unknown quantities) that could be determined
for simple geometries but that are not available for complex ones. Also, it
lacks the generality of the model assumptions, which researchers have tried
to overcome by including higher-order terms in the model equations.
4. Models for Wall Effects
Turbulent ows are signicantly affected by the presence of walls as the
mean velocity eld is affected through the no-slip condition at the wall.
Numerous experiments have shown that the near-wall region can be largely
subdivided into three layers:
(
The viscous sublayer, where the ow has laminar properties and the
viscosity has a dominant role in the momentum and heat transfer
(
The buffer region, where the effects of molecular viscosity and turbu-
lence are equally important
(
The fully turbulent layer, where the effects of turbulence are dominant
There are two standard methods to take into account wall effects in
numerical simulations: wall-function modeling and the use of low Reynolds
number turbulence models.
a. Wall-Function Models Wall functions are a collection of semiempirical
formulas and functions that link the solution variables at the near-wall cells
and the corresponding parameters on the wall. They are composed of laws
of the wall for mean velocity and temperature, and formulas for near-wall
turbulent quantities. For industrial ows, Launder and Spalding [35] wall
functions can be used. Therefore, the law of the wall for the mean velocity is
U*
1
K
ln (Ey*), (28)
where
U*
UC"
`
k`
t
"
/j
"
, y*
yC"
`
k`
v
, K0.42, E9.81, (29)
and y is the normal distance from the wall. The value of C
`
is nearly a
constant, often equal to 0.09 in practical applications. Nevertheless, for the
RNG model, Eq. (24) gives the correct value. This logarithmic law for the
mean velocity is known to be valid for y* 3060. For lower y* values, one
can apply the laminar stressstrain relationship that can be written as
U* y*.
390 . x. sn:n r1 :i.
Similar to this law of the wall for the mean velocity, the law of the wall
for temperature can comprise two equations: a linear law for the thermal
conduction sublayer where conduction is important, and a logarithmic law
for the turbulent region where effects of turbulence dominate conduction.
In high Reynolds number ows, the wall function approach substantially
saves computational resources, as the viscosity affected near-wall region
does not need to be solved. The wall-function approach is economical,
robust, and reasonably accurate. It is a practical option for the near-wall
treatments for industrial ow simulations. Some variations of the wall
functions based on the same concept are used in all CFD codes.
b. Turbulence Models for Low Reynolds Number Flows When low
Reynolds number effects are important in the ow domain, the hypothesis
underlying the wall functions cease to be valid. Therefore, models such as
the two-layer model (Iacovides and Launder [55], Rodi [56]) can be used.
Similar to Rodis model [56], Chen and Patels model [57] resolves the near
wall region by the transport equation for k only while the energy dissipation
and the eddy viscosity are prescribed in an algebraic manner. In these
models, the kc model is combined with one equation model near the wall
so that the dissipation rate and the turbulence viscosity near the wall are
calculated with the prescribed length scales l
'
and l
'
as
v
'
C
`
kl
`
, c
k``
l
'
, (30)
where the length scales l
'
and l
`
contain the damping effects in the near wall
region and are calculated from
l
'
K
C`"
`
y[1 exp(0.236y*)] (31)
l
`
K
C`"
`
y[1 exp(0.016y*)], (32)
where y* yk/v is the dimensionless distance [a denition different from
that in Eq. (29)] and y is the normal distance from the wall. C
`
0.09 and
K0.42, respectively. The Chen and Patel model [57] is observed to be
robust from the point of view of numerical stability and also capable of
predicting separated ows and ows over rough surfaces.
Another commonly used low Reynolds number turbulence model is by
Lam and Bremhorst [58].
5. Assessment of RANS Models
(
Among the RANS models, the kc model is the standard model used
for practical engineering ow calculations as it gives reasonably
accurate results for a wide range of turbulent ows without demanding
excessive CPU time and memory.
(
Other eddy viscosity models do not provide accurate enough results
for turbulent ows, even though the SpalartAllmaras [33] model or
the kc model [30, 48] yields good results for boundary layers
subjected to adverse pressure gradients and is well adapted for aero-
space applications.
(
Moreover, comparing with the kc models, the RSM requires signi-
cant additional memory and CPU time. The RSM does not always
give results superior to those of the kc models because of the large
number of adjustable parameters (unknown quantities). The use of
RSM is interesting when the ow features studied are the results of
strong anisotropy in the Reynolds stresses (cyclone ows, rotating
ows, etc.). ASMs are less sensitive to rotation than EVMs and need
less computational cost than the RSM, but this kind of model is still
not commonly used for industrial applications. Again the reason is that
there are too many parameters that need to be adjusted for each
application.
B. L:cr Ennx Sixii:1ioN (LES)
The main idea of this model is to compute the large-scale turbulence and
to model the smaller scales. Here lies a profound similarity between RANS
and LES models. The only difference between them lies in the denition of
a small scale. In RANS models, the effect of all eddies is simulated by the
turbulence model, whereas in LES models, the large scales are simulated and
the scales smaller than the grid size or the lter width are modeled [59]. All
large-scale structures in both RANS and LES models are determined by
solving the governing equations. Numerical models are applied to subscale
structures. As a result, a set of ltered equations with subscale correlations
is obtained. The subscale structures have relatively low energy and their
structure is expected to be rather universal. In the case when the grid size
is less than the Kolmogorov scale p
'
(lRe`"
'
, where the Reynolds number
is based on the mixing length l ), the uid transport properties are controlled
exclusively by molecular processes, the need for modeling of subscale
processes disappears and LES models turn into DNS models, when suf-
ciently small time steps are used. The major interest in the LES models is
the cases when RANS models perform poorly but DNS models are
392 . x. sn:n r1 :i.
prohibitively computer intensive. For such ows, which are encountered in
CHEs, the kc models plus wall functions cannot provide accurate predic-
tions of heat transfer to meet the CHE designers requirements. In addition,
the more advanced nonlinear second moment models and near-wall sub-
layer models usually result in unsatisfactory numerical stability in many
engineering applications. In some CHE applications where turbulence
mixing can be an important factor inuencing the performance, LES can be
used, as it provides not only mean ow mixing characteristics but also
subscale mixing, which can be very valuable. However, LES suffers from its
high computational cost. The choice therefore between RANS (EVM and
ASM) and LES is really dependent on the balance between accuracy and
computational cost (both memory and speed requirements). At present,
RANS is the only engineering tool for design of industrial CHEs. However,
from the point of view that higher and higher accuracy will be required for
CHE modeling and more and more powerful computers will be available to
users in the near future, it might be wise for researchers to investigate the
suitability of these methods for the prediction of the performance of heat
exchanger surfaces. As an example, Ciofalo et al. [152] used the LES
method for a chevron trough plate geometry and obtained the best predic-
tion of friction and heat transfer coefcients compared with other classical
models. However, this kind of approach is still not widespread. LES
methods have also been used with success for the modeling of turbulent
ows in complex geometries (Rodi et al. [60], Moin [61]).
LES requires some averaging of the variables (e.g., velocity) to obtain the
solved quantities. For this, various approaches are used and summarized
next.
a. Schumanns Approach [62] In this approach, instantaneous quantities
(velocities, temperature, etc.) are averaged over a control volume dened by
the numerical grid leading to a piecewise constant distribution of the
velocity. This method is implicitly used in nite volume methods, but the
averaged velocity results from the discretization and changes with it.
b. Filter Approach (Leonard [63]) This approach applies a low-pass lter
(top-hat lter or Gaussian lter) to the solved quantities. It leads to an
averaged velocity that is now a continuous function of the position and is
independent of the physical scale separation and discretization. However, in
practice, the lter does not appear explicitly in many LES codes. In fact, it
is implicit since scales smaller than the grids are automatically disregarded.
So, it is recommended to use a lter width larger than the mesh size in order
to remove the link between the subgrid scale length and the grid size because
the solutions become grid-independent.
The ltered averaged equations for constant uid density are
cu
'
cx
'
0 (33)
cu
'
ct

c
cx
'
(u
'
u
'
)
c
cx
'

p
j
c
cx
'
(2vS
''
t
''
), (34)
where
S
''
1
2
cu
'
cx
'
cu
'
cx
'
. (35)
The subgrid scale stress t
''
u
'
u
'
u
'
u
'
represents the effect of unresolved
(subgrid scale) motion on the resolved one and is modeled using subgrid
scale (SGS) models.
The task is to determine the subgrid-scale stress and hence to simulate the
effect of the unresolved motion on the resolved motion. This effect is mainly
dissipative, i.e., energy transfer is globally from large to small scales, but
locally and instantaneously transfer could be in the other direction (back-
scatter).
Eddy Viscosity Models. Most SGS models presently use the eddy-
viscosity (Boussinesq) concept:
t
''
1
3
o
''
t
''
2v
'
S
''
. (36)
The task of SGS is now to determine turbulent viscosity v
'
. From a
dimensional analysis,
v
'
lq
''
, (37)
where l is the length scale of unresolved motion (and not the mixing length
here), and q
''
the velocity scale of unresolved motion. Most active unresol-
ved scales are those closest to the cutoff point, so the natural l in LES is
usually the grid size. For determining q
''
, there are various approaches,
similar to the RANS modeling, as follows.
Smagorinsky Model (Smagorinsky [64], Lilly [65]). Similar to the
Prandtl mixing model, the Smagorinsky model is related to the gradient of
the space average velocity:
q
''
AS
''
S
''
. (38)
More recently, the subgrid viscosity is directly related to the strain tensor
[59] as
v
'
C
'
A`S
''
, (39)
394 . x. sn:n r1 :i.
where C
'
is the Smagorinsky constant (C
'
is often taken as 0.1 in practical
applications) and A is the lter width or subgrid scale.
Renormalization Group (Yakhot et al. [66]). The renormalization
group theory can be used to derive a model for the subgrid-scale eddy
viscosity, which results in an effective subgrid viscosity given by
v
''
(A) v
_
1 H
0.12cA"
v`(2)"
C
`
(40)
where
c 2v
''
(A)S`
''
,, H(x)
x x 0
0 x 0
. (41)
Here, H is the Heaviside function with x as a dummy variable, v is the
molecular viscosity, c is the mean dissipation rate, and C is a constant taken
equal to 73.5. For a high Reynolds number ow, the RNG-based subgrid
scale model reduces to the SmagorinskyLilly model. But in the low
Reynolds number ow region, the effective viscosity is equal to the molecu-
lar viscosity, allowing the RNG model to better predict ow in near-wall
regions.
Dynamic Procedure (Germano et al. [67], Ferziger [68]). The basic
idea of this procedure is to use information available from the smallest scales
to determine the coefcient in the SGS model. So, a test lter wider than the
basic LES lter (A 2A) is introduced and the SGS model is applied to
scales between A and A . The dynamic procedure and the Smagorinsky
model are the most widely used subgrid-scale models.
Structure Function Model (Me tais and Lesieur [69]). The eddy viscosity
is computed according to a structure function
v
'
0.063AF
`
(A) (42)
F
`
(r) [u
'
(x r) u
'
(x)]`. (43)
This structure function model has been validated for simple geometries
(backward-facing step, for example; Fallon [70]), but still requires work for
complex ow geometry, especially with thermal effects.
C. Dirc1 Nixric:i Sixii:1ioN
The direct numerical simulation is the simplest way to simulate turbu-
lence because no critical assumptions and no closure equations are needed
(Moin and Mahesh [71]). The classical NavierStokes, continuity, and heat
balance equations are solved directly using a high-order convective scheme.
However, this model requires a very ne mesh because the nest mesh must
be ner than the smallest scale of turbulence (i.e., Kolmogorov scale
p
'
lRe`"
'
, where the Reynolds number is based on the mixing length l ).
But the ratio of the smallest scale and the largest scale is a function of the
Reynolds number Re`"
'
, so the number of grid points in any direction is
Re`"
'
. In practice, a grid scale on the order of ve times the Kolmogorov
scale is usually sufcient (except in the near-wall turbulence). Xi et al. [72],
Mercier and Tochon [20] and Kouidry [73] have used DNS and unsteady
models to predict ow elds for the OSF geometry and corrugated channels,
but only for a 2D approach. So the DNS studies are limited by the
computational means. For instance, let us review the landmarks of DNS
research:
1. The rst DNS study, by Orszag and Patterson [74] on an isotropic
decaying turbulence, used a 32` mesh (Re based on Taylor macroscale
equal to 35)
2. Kim et al. [75] studied a turbulent channel ow with a 192129160
mesh (Re 3300)
3. Spalart [76] studied a turbulent boundary layer at Re 1410 with a
43280320 mesh
So presently, the main applications of DNS are (1) to provide reliable data
for the validation of turbulence models, especially useful where experimental
accuracy is low; (2) to provide data for evaluation of subgrid models for
LES (i.e., dynamic models); and (3) to conduct some fundamental studies of
turbulence. Even modern supercomputers have limited capability for ana-
lyzing complex ows in the CHE surfaces. Classical industrial applications
also require too much computer capability at present; hence, only some
conned geometries have been simulated with the DNS model, and also at
relatively low Reynolds numbers: McNab et al. [137] and Tochon and
Mercier [139] for corrugated wavy channels, Blomerius et al. [156] for
chevron trough plates, and Mercier and Tochon [20] for offset strip-n
geometries.
1. Assessment of Simulation Models
Although the LES and DNS techniques have demonstrated several
advantages over the RANS approach, these simulation techniques require
excessive CPU time and memory for computations of 3D complex ow
problems in the CHEs and other industrial problems, particularly at large
Reynolds numbers. Therefore, these methods are limited to a large-scale
turbulence with relatively low Reynolds numbers such as transition ows in
channels/ducts, ow over a bluff body, or CHE surfaces at low Reynolds
396 . x. sn:n r1 :i.
numbers. Because computer capabilities increase 10 times or more every 5
years, the simulation models represent a promising numerical tool for the
future.
D. CoNciiniNc Rrx:xs oN TiniirNcr MonriiNc
For the past 15 years, RANS models (kc) have been used for modeling
unsteady laminar, transitional, and turbulent ows in compact heat ex-
changers. But the poor description of anisotropic ows requires the use of
more advanced models (LES and DNS). Most of the work on the develop-
ment of new turbulence models has been based on simple geometries, such
as at plates and isolated backward-facing steps, rather than the complex
geometries encountered in CHE systems. For CHE surfaces, anisotropy,
shear ows, and rotational and other body force effects exist that cannot be
satisfactorily analyzed by most of the existing turbulence models. Although
the foregoing review shows promising results from these models, it is clearly
the case that the CHE technology calls for more accurate and dependable
models. The focus of such efforts should center around models that can
handle the following:
(
The multiple and interacting shear layers, separating and reattachment
surfaces
(
The correct near-wall behavior (ow and heat transfer) in complex
geometry situations
The LES procedure has received attention as a potential solution to these
kinds of difculties, and some progress has been made. However, robust
near-wall treatments as well as easy-to-implement averaging procedures for
the inhomogeneous turbulence features of CHE systems are current chal-
lenges. Until these difculties with LES are resolved, coupled with the
inappropriateness of DNS for realistic CHE systems, it would seem that
RANS (with problem-specic parametric optimization) will remain the
CFD tool for design and optimization of compact heat exchanger surfaces.
RANS models have already provided in-depth knowledge of local phenom-
ena on a relative basis leading to improvements in the design of CHE
surfaces.
V. Numerical Results of the CHE Surfaces
In this section, the numerical analysis of some important compact heat
exchanger surfaces is reviewed. Included are offset strip-n surfaces, louver-
n surfaces, wavy-n surfaces and channels, and chevron plates.
A. Orrsr1 S1ii FiNs
Extensive numerical analysis of offset strip n geometry (shown in Figs.
1a and 8) has been conducted since 1977. The details of numerical models,
operating conditions and geometries of offset strip ns analyzed are pro-
vided in Table I.
Numerical solutions for the offset strip-n (OSF) geometry was rst
started with zero n thickness (o 0) and an innite n height by Sparrow
et al. [77]. They varied the nondimensional strip length l/s from 0.2 to 5,
where l is the strip length and s is the transverse spacing of offset strip ns.
Because of the zero n thickness, the impingement region at the leading edge
of a strip and the recirculating region behind the trailing edge were absent.
Hence, the partial differential equations were parabolic and a marching
procedure was used. Patankar and Prakash [78] extended the analysis to
nite n thickness o in terms of dimensionless o/s 0.1, 0.2, and 0.3 for a
fully developed periodic ow for Reynolds number Re in the range
100 Re 2000. They found a small recirculating zone behind the trailing
edge at low Re or low o/s. At high Re or high o/s, this zone extended from
the trailing edge of a strip to the leading edge of the succeeding strip. This
constricted the ow to the minimum ow area, thinned boundary layers,
and resulted in high j and f factors. An increase in f was more pronounced
with increasing o/s, as expected because of the higher form drag. The
prediction for the f factor was in reasonable agreement with experimental
data, but the predicted j factor was about 100% high, and the slopes of j
and f vs Re data were steeper than those for experimental data.
Kelkar and Patankar [79] extended their previous work [78] on offset
strip ns to investigate the effects of the n length and aspect ratio. The
numerical simulations were performed assuming a zero n thickness for a
3D control volume. The grid used was 302030. To characterize the
geometry, they introduced a parameter [* dened by [* lRe/s. The main
conclusions from their work are as follows: The number of cells necessary
to obtain a fully developed ow increases with [*; the 3D aspect of the ow
is negligible for low aspect ratios (0.2); the ow becomes more complex
and three-dimensional for higher aspect ratios. They compared their nu-
merical results with the empirical correlation of Wieting [80]. Their numeri-
cal model underestimated the friction factors by about 15% and
overestimated the Nusselt numbers by 15% compared to those predicted by
the Wieting correlation for the entire range of Reynolds numbers and
geometrical parameters. The agreement appeared to be better for higher
values of the n length parameter ([* 0.01).
Suzuki et al. [81] obtained a numerical solution for combined free and
forced convection in laminar ow through staggered arrays of zero thickness
398 . x. sn:n r1 :i.
TABLE I
Sixx:x or Nixric:i Monris, Oir:1iNc CoNni1ioNs, :Nn Groxr1irs or Orrsr1 S1ii FiNs
Boundary Inlet Reynolds
Authors Model Grid condition condition number, Re Pr Geometry Validation
Patankar andLaminar 2D: 6030 Constant heat Fully 1002000 0.7 o/p
'
0; 0.1; 0.2; 0.3 Pressure drop
Prakash ux developed l/p
'
1 Nusselt number
[78]
Kelkar and Laminar 3D: 302030 Developing 0.7 b/s 0.1 to 1 Wieting [80]
Prakash correlations
[79]
Suzuki et al. Laminar 2D: 33927 Constant wall Uniform 125500 0.7 o 0
[81] temperature velocity and l/p
'
1
temperature
Suzuki et al. Laminar 2D: 33931 Constant heat Uniform 2001500 0.7 o/p
'
0; 0.2; 0.4 Local Nusselt
[82] ux velocity and l/p
'
1 number
temperature
Xi et al. Laminar 2D: 50080 Constant heat Uniform 2501000 0.7 o/l 0; 0.04; 0.08 Local Nusselt
[83] ux velocity and l/p
'
0.8 to 4 number
temperature
Suzuki et al. Unsteady 2D: 380130 Constant wall Uniform 8005000 0.7 o/l 0.0314; 0.126
[86] laminar temperature velocity and
temperature
Xi et al. Unsteady 2D: 380130 Constant wall Uniform 860 and 3430 0.7 o/l 00314; 0.126 Flow visualization,
[72] laminar temperature velocity and velocity etc.
temperature
Michallon Laminar 2D: 4038 Constant wall Fully 3005000 0.7 o/l 0.14 Pressure drop
[87] kc 3D: 303910 temperature developed l/p
'
2.2 heat transfer
Mizuno et al. Laminar 3D: 36000 grid Constant heat Developing 20300 0.7 o/p
'
0.064 Heat transfer
[88] point ux and 7 l/p
'
2.61
Mercier and DNS 2D: 245141 Constant wall Developing 5000 7 o/p
'
0.14 Local and overall
Tochon temperature and fully l/p
'
2.2 (literature)
[20] developed
Xi and Shah Unsteady 3D: Constant wall Uniform 1006000 0.7 o/p
'
0.064; 0.1 Experimental j and
[90] Taminar 4604545 temperature velocity and l/p
'
1.5; 1.95; 2.61 f factors from
temperature literature
3
9
9
offset strip ns. Suzuki et al. [82] extended the analysis to nite thickness
ns and also to free-stream turbulence. To predict the turbulent ow at a
low Reynolds number, a turbulence model derived from the standard kc
model implemented. A relatively good agreement was found between
numerical and experimental values of Nu within the range of Reynolds
number tested (Re 800). In addition, experiments were conducted by
varying the inlet turbulent intensity, and the effect was found to be very
small. Furthermore, the effect of the n thickness was found to be rather
small. The n length and spacing effects were extensively studied from the
heat transfer point of view. Xi et al. [83] provided details on the numerical
treatment of the cells adjoining the n surface and extended the previous
study to a larger number of offset strips (up to nine). The effect of n
thickness was studied in detail and appeared to depend on the n spacing
and length ratio. The detailed explanation on heat transfer enhancement in
a n array was also presented.
An unsteady ow eld in the OSF geometries results in heat transfer
enhancement. Based on ow visualization results, Mochizuki et al. [84] and
Xi et al. [85] reported that the ow in unsteady region has highly periodic
velocity uctuations, and resulting ow instability is strongly dependent on
the Reynolds number and geometric parameters, such as n pitch and n
thickness. In the transition ow region, ow patterns are different for
different rows in OSF arrays. As outlined by Xi et al. [83] and Jacobi and
Shah [1], ow instabilities can appear in the wake of the n even at low
Reynolds numbers. Suzuki et al. [86] and Xi et al. [72] studied unsteady
ow eld in an inline array of three ns. The ow was assumed uniform at
the inlet. The computational grid was 380130 and was selectively nonuni-
form: A ner grid was applied near the n wall surface. To solve the
governing equations, a central nite difference scheme was used for the
diffusion terms. In the ner mesh region, a second-order upwind scheme was
used to solve the convective terms and a third-order scheme was used in the
region far from the wall. The predicted ow eld was compared with
experimental data obtained by Xi et al. [72], and a good agreement was
found for the mean streamwise velocity and the rms values of the uctuating
velocities. A close analysis of the ow pattern and heat transfer near the n
wall surface has led to the conclusion that heat transfer enhancement is
created by self-sustained ow oscillations. Local instabilities near the wall,
created by upstream vortices that impinge on the n, produce dissimilarity
between the momentum transfer and heat transfer, thus reducing the local
skin friction. These phenomena have already been observed for an array of
cylinders. Unsteady ow simulation is very promising as it gives local
instantaneous information. However, to engender condence in the numeri-
cal results, a careful validation should be provided on instantaneous data,
not just on the time averaged values.
400 . x. sn:n r1 :i.
Michallon [87] used a standard CFD code (TRIO-VF, developed by the
French Atomic Commission) to model an offset strip n channel. The fully
developed ow was obtained by reinjection of the outlet ow conditions as
the input in the iterative solution. The numerical simulations were per-
formed for 300 Re 4000. For Re 2000, a laminar model was applied,
whereas for higher Reynolds numbers, the standard kc model was used.
Two- and three-dimensional analyses were performed, and no signicant 3D
effects were noticed. The reason for this may be related to their computa-
tional model, which is only a part of OSF, and boundary conditions in
which they used a reinjection procedure. As mentioned earlier, ow patterns
are different for different rows in OSF arrays, especially when the ow is
unsteady. Therefore, the n height of the OSF channel, n thickness, and
inlet ow condition need to be taken into consideration for a more precise
3D numerical analysis. The comparison with experimental values obtained
by Michallon [87] showed that the friction factors are overestimated by
about 25%, and the Nusselt numbers are overpredicted signicantly, from
30% for low Reynolds numbers (Re 1000) up to a factor of 2 for high
Reynolds numbers (Re 3000). In terms of local interpretation, a recircula-
tion zone was observed at the trailing edge of the n, but the size of this
zone was limited. This result is in agreement with the work of Patankar and
Prakash [78].
Mizuno et al. [88] numerically investigated three-dimensional offset strip
ns in the laminar low Reynolds number regime (Re 300). The thermal
boundary layer developed on the n surfaces and the primary surface
(parting sheets) cannot be considered to be of the same thickness. On the
parting sheet, the thermal boundary layer is thicker and therefore affects
heat transfer. To prove this phenomenon, a three-dimensional analysis
taking into account conduction in the ns was undertaken [88]. A conven-
tional control volume method was applied, assuming a zero n thickness.
For the ow boundary conditions, a uniform velocity was assumed at the
inlet, with a zero velocity gradient at the outlet. The thermal boundary
condition was constant wall temperature on the separation plates. In
parallel with the numerical work, experimental measurements were per-
formed. The numerical results underestimate the pressure drop by 20% at a
Reynolds number of 20 but are in agreement at higher Reynolds numbers
(Re 300). The average Nusselt number is well predicted at Re 100;
below this value, it is overpredicted; and above this value, it is underpredic-
ted. The maximum deviation is around 10%. A parametric study reports on
the effects of the thermal conductivity of the uid and of the n material,
and outlines the fact that for a high value of the uid thermal conductivity
(water), the heat transfer performance is affected by the n material; i.e.,
aluminum ns are better than stainless steel ns. The other results of this
study concern the thermal boundary layer on the parting sheets; its
thickness is comparable to the one of a plain rectangular channel.
Mercier and Tochon [20] have performed a 2D time-dependent analysis
of turbulent ow in an offset-strip n heat exchanger. The turbulent ow
behavior is solved by using both an accurate convective scheme (third order)
and a ne grid. The mesh size must be smaller than the smallest turbulent
scale, and the computational domain must be greater than the largest scale
involved. Due to limited computer capabilities, not all the turbulent scales
could be solved, and the method is referred to as pseudo-direct numerical
simulations. Two cases considered by Mercier and Tochon [20] are a single
n with a uniform ow upstream, and an array of offset strip ns under
developing and fully developed ows. On a single n, the time-dependent
evolution gives fundamental information on the ow structure (see Fig. 4).
The ow, hitting the front edge, separates and creates a recirculation zone.
At the reattachment point, which oscillates, one part of the ow is convected
downward and the other part of the ow upward in the recirculation zone.
At the trailing edge, vortices are convected, and a von Ka rma n street is
formed. These phenomena are in agreement with visual observation per-
formed on similar geometries. When the size of the recirculation zone is
compared with data from the literature, good agreement is found. Concern-
ing the array of ns, two basic inlet ow conditions have been studied:
developing ow and fully developed (reinjection). For developing ows, the
vortices created by the rst row of ns impinge on the downward ns and
suppress all organized structure (highly turbulent ow). The comparison of
the calculated time average friction factors and Nusselt numbers with
measured values show poor agreement for the developing ow. Applying a
reinjection procedure (fully developed ow) gave relatively good agreement
with the literature data. Despite a 2D approach, the results appear qualitat-
ively correct, and local and overall information on ow structure and
thermal and hydraulic performances can be obtained. For a single n, the
numerical velocity eld was compared with the data of Ota and Itasaka
[89], and the agreement is qualitatively correct with a mean deviation of
20%. For the array of ns, the predicted values of the friction and
Colburn factors are compared to several correlations and results from the
open literature. The predicted friction factors are underestimated by 5 to
30%, and the Colburn factors are underestimated by 30 to 50%.
Xi and Shah [90] conducted a 3D numerical analysis of OSF geometries
(Fig. 8) that differed from that of Michallon [87] and Mizuno et al. [88] in
the following aspects: (1) time-dependent, i.e., unsteady laminar ow; (2)
nite n thickness; and (3) computational domain having upstream ow
region (2l), OSF region (9l), and downstream wake region (10l) so that
the effects of form drag, velocity defect, and temperature excess in the wake
are properly taken into account. In order to suppress an overshoot of the
402 . x. sn:n r1 :i.
solutions and reduce the magnitude of numerical viscosity, a mixed usage of
two upwind schemes for convection terms was applied for the momentum
equations (third order) and energy equation (second order). The nite
difference equations of fully implicit forms were solved step by step along
the time axis with the evaluation of pressure by the SIMPLE algorithm and
the alternating direction implicit (ADI) method for each relaxation. In order
to optimize the computational time for minimizing the maximum and total
residual errors of the computational domain, the following scheme was
adopted: (a) The computation was rst carried for 15,000 time steps. In each
time step, one iteration was done. (b) Then, the computation was carried for
another 2000 time steps in which ve iterations were done. (c) Finally, the
computation was carried out for additional 2000 time steps (with ve
iterations per time step) to get the mean values for the results presented here.
The computed results of the ow and thermal elds are little affected by
what pattern of spatial distributions used as the initial condition. Refer to
Xi and Shah [90] for further numerical details.
Three comparisons were made by Xi and Shah [90]: numerical results of
Mizuno et al. [88], experimental results of an idealized OSF (Mochizuki et
al. [84]), and a real OSF (London and Shah [91]). The 3D numerical results
obtained by Xi and Shah [90] are in better agreement with the experimental
data of Mizuno et al. than are the the latter groups own 3D numerical data.
The primary reasons are that Xi and Shah considered a nite (actual) n
thickness of the OSF geometry, considered unsteady laminar ow in the
analysis, and employed 2l upstream and 10l downstream regions as part of
the OSF domain, whereas Mizuno et al. considered zero n thickness,
steady laminar ow, and employed no upstream and downstream regions
outside the OSF region.
A comparison between the experimental results of Mochizuki et al. and
the 3D numerical results of Xi and Shah for an idealized OSF shows an
excellent agreement in the j factors for Re 6000 and in f factors for
Re 4000, as shown in Fig. 9. The f factors computed for Re 6000 have
possible numerical convergence error due to high Re where the aforemen-
tioned iteration scheme may not be adequate; and a turbulence model may
be required as explained in [90]. A comparison of experimental results for
the real OSF surface of London and Shah [91] and the 3D numerical results
of Xi and Shah for the idealized n geometry (without burrs at edges, surface
roughness, etc.) are shown in Fig 10. For f factors, there is an excellent
agreement between experimental and numerical results for Re 1000.
However, the numerical results have lower values than the experimental f
factors for 1000Re5000 (maximum difference of 35% at Re 2500).
The burrs and bent leading and trailing edges of an OSF and surface
roughness due to brazing would result in a larger effective n thickness. This
Fic. 9. A comparison of 3D numerical results of Xi and Shah [90] with experimental results
of idealized OSF of Mochizuki et al. [84].
Fic. 10. A comparison of 3D numerical results of Xi and Shah [90] with experimental results
of a real OSF of London and Shah [91].
404 . x. sn:n r1 :i.
in turn has an important effect on the ow instability and in the transition
region than in the laminar (low Re) region [90]. For j factors, the numerical
results show much larger values than the existing experimental results and
correlations (about 50% at Re 400), except that the slopes of the correla-
tion by Mochizuki et al. [84] for idealized geometries and the numerical
results are closer. At present, the reasons for this large difference in
numerical vs experimental j factors are not known. The reduction in j factors
for the real OSF surface compared numerical values for the idealized
geometry may be in part due to passage-to-passage nonuniformity; this
nonuniformity has a signicant effect on j factors and negligible effect on f
factors at low Re (Shah [93]). In addition, numerical analysis shows another
important difference between the real and idealized OSF geometries as
follows. The transition for f factors (deviation from the straight f vs Re line
for laminar ow on a log log plot) begins in the range Re 1000 1200
for the London and Shah surface and in the range Re 1600 2500 for
numerical results as shown in Fig. 10. The transition for j factors begins at
Re 1500 2000 for the London and Shah surface and between Re values
of 2500 and 3300 for the numerical results. The reason is due to burrs and
bent edges and surface roughness for real surfaces. Another interesting
observation is that the transition in f factors starts at lower Re than that
for j factors. This feature indicates that the unsteady ow rst enhances ow
friction and then enhances heat transfer as a function of Re. It has been
observed in visualization experiments reported by Mochizuki et al. [84] and
Xi et al. [85] that the unsteady ow rst occurs downstream of n arrays
(affecting only AP and f factors) and then progresses upstream into n
arrays as the Reynolds number is increased.
A comparison of 2D and 3D numerical results for the Shah and London
surface indicate that the effect of 3D geometry is smaller in the laminar ow
region (Re
`'
1600) and the 2D computations are quite accurate at low
Re.
1. Concluding Remarks on Offset Strip Fin Performance/Analysis
The studies just summarized for offset strip n geometry indicate that a
considerable amount of numerical and experimental work has been conduc-
ted on this simple and high-performance compact heat exchanger surface.
As outlined by Xi et al. [72] and Mercier and Tochon [20], unsteady
analysis is required to accurately calculate turbulent ow in compact
geometries. A 3D model is also required to simulate the correct phenomena
[90]. Nevertheless, the main physical phenomena can be predicted by these
methods, and the predicted values are in relatively good agreement with
experimental data.
Fic. 11 Typical louver n geometries: (a) corrugated n, (b) at n.
B. Loivr FiNs
Since the early 1980s, numerous attempts have been made to develop 2D
numerical models of louver n surfaces. In all louver n analyses reported
in this paper and in the literature, the geometry analyzed is a at n as
shown in Fig. 11b; no 3D numerical studies are reported for Fig. 11a. The
details of the geometries analyzed, numerical models employed, and the
Reynolds number range investigated are summarized in Table II. Initially,
the models were based on the idealized zero n thickness and laminar ow
with periodic boundary conditions and constant wall temperature. Kajino
and Hiramatsu [94] and Tomoda and Suzuki [95] solved the stream
function and vorticity equations for incompressible steady laminar 2D ow
over at louver ns using nite difference methods. They presented stream-
lines, velocity proles, and Nusselt number distributions on the louver
surface for one louver geometry only at a single Reynolds number value.
The authors stated that the computational results showed trends similar to
the ow visualization results, but no quantitative validation was made. They
concluded that although ow visualization is a useful tool for assessing the
performance of louver ns, numerical calculations are necessary for more
quantitative evaluation. Their main conclusion was that for high-perform-
ance ns with small louver pitch, best results are obtained if the n pitch is
matched to that of the louver in such a way that the uid ows along the
louver. However, no evidence was presented in the paper to support this
statement, and it would appear that the numerical solution was only used
to prove the experimental ndings.
406 . x. sn:n r1 :i.
TABLE II
Sixx:x or Groxr1irs, Nixric:i Monris, :Nn RrxNoins Nixnr R:Ncrs Covrrn ro Fi:1 Loivr FiNs
Reference Geometry Numerical model Re
l
Kajino and Hiramatsu p
'
1.0 mm l 1.0 mm 2D, stream function and vorticity, laminar, grid not described 500
[94] 0
26
o 0.1 mm
Achaichia and Cowell o 0 2D, nite volume, laminar steady, Cartesian grid, periodic 201500
[96] 0
15 to 35 boundary conditions, didnt solve the energy equation

p
'
/l1.0 to 2.5
Hiramatsu et al. [98] 0
0 to 50 2D, nite difference, body tted rectangular grid, stream 1001000

p
'
/l1.0 and 2.0 function and vorticity
Ha et al. [100] 1000 and 1050 ns/m for 0
23 2D, body tted generalized grid, laminar steady ow 1761006

1000 ns/m for 0
31
1000 ns/m for 0
31
Baldwin et al. [103] 0
15 and 30 2D, nite difference, Cartesian grid, staircase louver edge 2006000
l30 mm representation
Suga et al. [106] 0
30, 26 2D, nite difference, steady laminar ow, overlaid 64450

l10mm, o0.8 mm, p
'
10 mm Cartesian grids
p
'
/l1.0, 1.125 and 1.75
Suga and Aoki [107] 0
20, 26 and 30, same as Ref. [106] above 64450

p
'
/l0.5 to 1.125, o0.8 mm
l10 mm, p
'
10 mm
Ikuta et al. [108] o0.115 mm, l1.3 mm, 2D, body tted grid, nite difference, steady laminar ow 417
p
'
1.1, 1.5 and 1.9 mm,
0
15, 20, 25 and 30

Achaichia et al. [109] 0
20 to 40, p
'
/l1.7. 2D, body tted grid, kc turbulence model for Re
l
1200, 102400
energy equation not solved
Itoh et al. [110] p
'
/l1.22, o/l0.625 2D, steady laminar, grid details not given 645 and 595
Atkinson et al. [111] p
'
1.5 to 2.5 mm, 2D, steady laminar nite volume, automatic body tted grid 1003200
and l0.9, 1.1 and 1.4 mm, generation
Drakulic et al. [112] 0
12 to 28.5, o0.05 to 0.1

Drakulic [113] 2D: same as Refs. [111, 112] Same as Refs. [111, 112] 2D analysis:
3D: p
'
8 and 14 mm Conjugate heat transfer in the 3D case 1003200
p
'
2.17 mm, l1.1 mm 3D analysis:
0
22, o0.05 mm 100, 400, and 1600

4
0
7
A more comprehensive study was carried out by Achaichia and Cowell
[96]. Using nite difference techniques, they modeled one louver in the
periodic fully developed region assuming cyclic boundary conditions,
laminar steady ow, and zero n thickness. Since only one louver was
modeled, the authors were able to use a ne Cartesian grid normal to the
louver surface. Results were presented for n-to-louver pitch ratios of 1 to
2.5, louver angles 15 to 35, and louver pitches based Reynolds number Re
l
from 20 to 1500. The results conrmed the phenomenon rst identied by
Davenport [97], namely the existence of two distinct ow regimes with duct
ow occurring at low Reynolds numbers and louver (aligned) ow prevail-
ing at high Reynolds numbers. This result, which is clearly demonstrated by
the sketch of Fig. 5b, has since been accepted and conrmed by a number
of other studies both experimentally and numerically [98100]. From the
numerical velocity distributions, Achaichia and Cowell [101] quantied this
ow alignment property of the louvers by introducing the concept of the
mean ow angle. They plotted this parameter against the Reynolds number
and showed that the ow is aligned with the louver to within a few degrees
at high Reynolds numbers, and the degree of alignment begins to fall off as
the Reynolds number is decreased. They also found that the Reynolds
number at which this trend starts is a function of the n-to-louver pitch and
derived a simple correlation for the mean ow angle in terms of the
Reynolds number, the n-to-louver pitch ratio, and the louver angle. They
also found that their model could predict recirculation behind the louvers
at high louver angles even for the fully developed periodic case.
The effect of this ow-directing behavior is that the friction factor curves
exhibit a steep slope at low Reynolds numbers as the less aligned uid
impinges on the louver surface as it ows down the duct between the ns.
At high Reynolds numbers, the curve shows a clear attening as the ow is
aligned with the louvers. This ow behavior is also responsible for the
attening of the Stanton number curve observed by Achaichia and Cowell
[101]. A ow efciency term was devised by Webb [102] and Webb and
Trauger [99] to describe this ow behavior and predict the onset of this
attening in the Stanton number curve.
Finite n thickness models were analyzed in a number of numerical
studies on ow through complete louver arrays for a few isolated congur-
ations and/or ow rates. In all cases, a symmetrical array having two-bank-
deep louvers was modeled. Baldwin et al. [103] used a Cartesian grid and
solid cells with zero porosity to dene the n. The louver surface was
therefore represented by a series of staircase-type steps with the grid spacing
selected to make the solid cells approximate the geometry. A nite volume
solution of the ow equations was carried out using the commercial
PHOENICS code. Results were presented for only two louver congur-
408 . x. sn:n r1 :i.
ations, each at a single Reynolds number, and compared with the LDA
measurements of Button et al. [104] and the ow visualization results of
Hiramatsu and Ota [105]. The numerical results displayed the same basic
phenomena as the experimental studies, showing ow separation after the
rst louver and almost complete alignment after the third. As the ow
entered the second bank of louvers, it took longer to become aligned relative
to the rst bank.
Suga et al. [106] presented a nite difference numerical model for a
complete louver n (two louver banks) over a limited range of Reynolds
numbers (64Re
l
450). The model assumed 2D steady laminar ow. They
used an elaborate system of overlaid grids to overcome the nite n
thickness problem and divided the solution domain into regions that were
represented by Cartesian grids. Complex communication between the grids
was achieved by bilinear interpolation of the dependent variables at the grid
boundaries. The authors identied the possibility of interpolation errors at
the false boundaries between grids and stated that they could be reduced by
the careful choice of the grid sizes in the overlapping region. Another
possible source of error was the use of triangular cells to join the bent parts
of the rst half louver. The predicted velocity distributions in the regions
between the louvers were compared with LDV measurements at Re
l
64.
However, the computational results did not display the characteristic duct
ow expected at such a low Reynolds number. The calculated mean Nusselt
number for each louver was also compared with that measured by means of
a nickel lm sensor (which acted both as a heater and a resistance
thermometer). A remarkable degree of agreement was shown even though a
uniform n surface temperature was assumed in the processing of the
experimental results.
Using the same numerical model, Suga and Aoki [107] investigated the
effect of grid renement as expressed by a grid Reynolds number Re
, and
concluded that a grid independent solution was obtained at values of
Re
7. They then carried out a numerical study of the effect of n

parameters on heat transfer performance. The range of the Reynolds
number for this study was also 64Re
l
450, louver angles 200
30,
n thickness to louver pitch ratio 0.04o/l 0.08, and high-density ns
with n pitch to louver pitch ratio 0.5 p
'
/l 1.125. They conjectured that,
provided that the ow is aligned with the louvers, the performance of the
n is dominated by the thermal wake behind the louvers. They then
concluded that for each louver angle, there was an optimum value for the
louver-pitch-to-n-pitch ratio that caused the thermal wake behind the
louvers to ow along a line halfway between two louvers further down-
stream. They presented a formula for the calculation of this optimum ratio
as a function of the louver angle. It is interesting that the value of the
optimum ratio was observed to be independent of the Reynolds number in
the range Re
l
450.
Body-tted coordinates have been used to facilitate the grid generation
for a full n with nite thickness. Ikuta et al. [108] developed a 2D model
using nite difference techniques in a curvilinear coordinate system with a
relatively coarse grid. They displayed the velocity vectors diagram for a
single geometry at one value of Reynolds number showing ow separation
from the rst three louvers with complete alignment subsequently. This
process was shown to repeat in the second bank of louvers.
The oblique grid and coordinate transformation system, adopted by
Hiramatsu et al. [98], offered a more versatile method for the nite
difference modeling of louver n geometries. In this system, two mesh
structures were used with an oblique grid in the zone of the inclined louvers
and a rectangular grid over the horizontal inlet and turn round parts of the
n. The laminar 2D steady ow model was considered for the analysis.
Some mesh renement was done by increasing the number of grid points
near the louver edges. A comparison between the predicted streamlines and
ow visualization in a water channel showed good agreement. The stream-
line plots showed increasing ow alignment with increasing n-pitch-to-
louver-pitch ratio and Reynolds number. The effect of varying the louver
angle was not shown. Calculated values of local heat transfer distribution
on both n surfaces were presented for one geometry at a single value of the
Reynolds number. The results showed a decrease in the mean value of the
mean Nusselt number as the n pitch increased and with decreasing
Reynolds number. This was attributed to the reduced degree of ow
alignment under these conditions. Although the calculated friction factors
did not agree well with their experiments, the mean Nusselt number showed
excellent agreement.
Achaichia et al. [109] described a novel body-tted grid topology that
extended over a number of ns, thus simplifying the introduction of nite
n thickness and geometrical variations. Two-dimensional steady-state nite
volume calculations were performed, using the commercial nite volume
code PHOENICS on a two-bank louver n array. A number of different
louver angles were considered in the range 20 to 40, a xed-n-to-louver-
pitch of 17 for Re
l
between 10 and 2400. The ow was assumed to be
laminar at low values of Reynolds numbers. The kc turbulence model was
applied above the critical Reynolds number of 1200. The mean ow angle
was calculated along the n at different velocities and the results clearly
showed a gradual alignment of the ow along the n. The effect of the
Reynolds number on the degree of alignment was also demonstrated and
quantied. The effect of the louver angle on the distribution of the local skin
friction on the upper and lower n surfaces was shown at Re
l
10 and 600.
410 . x. sn:n r1 :i.
At the lower Reynolds number and for all louver angles, high values of local
skin friction were observed at the trailing edge of the louvers as the duct
directed ow impinged on the surface. This effect was less noticeable at the
higher Reynolds number since the degree of ow alignment was much
higher. The skin friction was markedly higher for the 40 louver, but only
for the lower Reynolds numbers. The authors explained this observation in
terms of the increased ow between the louvers (and less duct ow) as a
result of the larger gap between them at this angle and concluded that
increasing the louver angle has the desired effect of increasing the degree of
ow alignment at the same Reynolds number.
Itoh et al. [110] reported a difference of about 6% in the value of the
average Nusselt number when the temperature dependence of the physical
property was included in the solution procedure.
Ha et al. [100] dened the governing equations in generalized coordinates
in order to overcome grid generation difculties. They used the commercial
nite volume code FLUENT to solve these equations assuming a laminar 2D
steady ow model. However, their grid structure produced a coarse distorted
mesh between the louvers, where high velocity gradients are expected, thus
signicantly reducing the numerical accuracy of their model. Nevertheless,
the authors conrmed the ow directing properties of the louver n. They
also observed a developing ow regime at the inlet with ow separation at
both leading and trailing edges of the louvers. Calculated friction factor and
Nusselt number results were reported but not validated experimentally.
Atkinson et al. [111] and Drakulic et al. [112] reported a novel automatic
grid generator that produced 2D grids for any louver n geometry from
parametric input of the n data. The resulting mesh had a block structure
with three different types of blocks corresponding to the inlet region of the
n, the individual louvers, and the turnaround region. These blocks were
divided into a number of quadrilateral cells. Nearly all the cells were
rectangular, giving maximum possible numerical accuracy. With grid lines
parallel and normal to the louver, the boundary layer proles and integral
parameters could easily be extracted.
Two of the most comprehensive numerical studies of louver n character-
istics are by Drakulic [113] and Atkinson et al. [114]. In these studies, 2D
and 3D numerical modeling was performed for a number of experimentally
tested louver ns using the automatic grid generator described earlier. These
louver ns were at n type used in a at tube and at n construction (see
Fig. 9b). The STAR-CD CFD code was used for the analysis for the
Reynolds number Re
l
range of 503200 for 2D and Re
l
100, 400, and
1600 for 3D. In the numerical modeling, the computation domain included
upstream and downstream region each as l and the louvered n had two
banks of louvers as shown in Fig. 5a with each bank having 5, 10, 13, or 19
Fic. 12. A comparison of friction factor and Stanton number experimental results with 2D
and 3D numerical results of multilouver ns, having p
'
2.17 mm, l 1.1 mm, and 0
22:
(a) p
'
8 mm and (b) p
'
14 mm [113].
full louvers. A constant wall-temperature boundary condition was applied
on the primary surface. A conjugate model of thermal conduction through
the n and the convection from its surface was used. This is because the
temperature distribution in the n and the thermal boundary layer on the
tube wall result in complex temperature proles that cannot be predicted
using 2D models and constant n wall temperature. Time-dependent nu-
merical solutions of the louver n clearly showed vortex shedding from the
trailing edges of the rst two louvers in the upstream and downstream
banks, resulting in a rather long unsteady wake behind these louvers.
Hot-wire measurements of local velocity distributions in louver ns by
Antoniou et al. [115] had showed the same rms values of velocity uctu-
ations, but the ow was wrongly interpreted as being turbulent. The
numerical results clearly showed that this is unsteady laminar ow. The
mean Nusselt number curves computed from the time-dependent solutions
showed the experimentally observed attening at high Reynolds numbers.
Numerical predictions of local and mean ow and heat transfer par-
ameters were compared with the experimental measurements of Achaichia
[116] and Antoniou [117]. The results clearly showed that although
reasonable predictions of the overall friction factor could be obtained, the
Stanton number was overpredicted. For two at multilouver n geometries,
experimental results and 2D and 3D numerical computations are shown in
Fig. 12 [113]. Reviewing these gures, it can be seen that while the friction
412 . x. sn:n r1 :i.
factors agree with experimental values within 010%, the computed Stanton
numbers are about 80% higher compared to experimental values at
Re
l
400. The results presented in Fig. 12 are correct and supersede those
presented in Fig. 10 of [114]. In the numerical analysis, both the air and
wall temperatures vary in 3D computations. To obtain the Stanton number
from the heat ux, one needs to base it on a difference between the wall
(primary surface) temperature and the bulk air temperature. The commer-
cial software does not compute the bulk temperature, only the detailed
temperature distribution in the computational domain. Hence, the mean
temperature difference for the heat transfer coefcient determination was
based on the log-mean temperature difference, where the wall temperature
and the air inlet temperatures are known and the outlet bulk temperature
of air was computed from the numerical results. Since the heat transfer
coefcient based on the LMTD would be different from that based on the
wall temperature minus the bulk air temperature, this may be one reason
for a large discrepancy in the computed versus measured values of St.
However, it cannot account for the large discrepancy; the reasons are not
clear, as was the case for the OSF surface [90].
Most of the previous work on multilouver n geometry is ether experi-
mental or numerical. Only a few studies have combined experimental work
on model-scale and full-scale heat exchangers together with the CFD studies
on the model scale n geometry and the ow visualization setup with a
number of ns. Beamer et al. [118] reported a study on full-scale heat
exchanger performance testing and ow visualization experiments, and a 2D
CFD study on one n (see Fig. 5a) with a periodic boundary condition, and
six- and 12-n geometries to duplicate ow visualization setup and results.
The geometric dimensions used in the 2D study of one n were the same as
those of the actual heat exchanger tested. The CFD domain used for one n
(the central n in Fig. 5a) was extended upstream and downstream of the
n by 20% of the n length (L
'
in Fig. 1b) to take into account the ow
adjustment at louver inlet and exit margins (sections). A known ow rate
was applied at the inlet boundary, and pressure boundary conditions were
prescribed at the outlet boundary. Periodic boundary conditions were
applied on two sides of the complete computational domain (i.e., 1.4 L
'
);
symmetric plane boundary conditions were applied on the top and bottom
control volume surfaces (perpendicular to the plane of the paper in Fig. 5a)
because of the nite volume code. Typically 7000 to 27,000 hexahedral cells
were used in a single array for ow calculations. Blended upwind differenc-
ing with a blending factor of 0.5 was used for the convective differencing
scheme. The results obtained from the 2D CFD study include ow develop-
ment and alignment with the leading bank of louvers, reversal at the center
rib, development and alignment with the trailing banks of louvers, and ow
Fic. 13. A comparison of computational and experimental results for ow efciency of
multilouver ns [118].
exit from the n in a direction parallel to the longitudinal axis. Evidence of
ow separation at the leading edge and vorticity in the wake behind the
trailing edge of the louvers was observed in the CFD vorticity plots (not
shown). The CFD study on the ow visualization setup for 6 and 12 ns
duplicated the observed ow phenomena and indicated that even a 12-n
(note that there are 3 ns shown in Fig. 5a) arrangement does not eliminate
the wall effects in ow through louver ns in a ow visualization setup. A
comparison of ow efciency of their study [118] with those of Webb and
Trauger [99] and Cowell et al. [2, 101] is shown in Fig. 13. Note that
Achaichia and Cowell [101] did not employ a developing ow region before
the beginning of each louver n in their CFD study. As a result, they
overestimated ow efciency as shown in Fig. 13. From this gure, it is
found that the CFD results [118] and ow visualization measurements
show good agreement. Both display an easily discernible knee below which
the ow efciency drops off rapidly. The data of Webb and Trauger [99]
show a similar trend, except that the value of Re for the knee and ow
efciency are considerably higher. This rapid dropoff of the ow efciency
at low Re is attributable to the thickening of the boundary layer between
the louvers, which causes more uid to ow toward the duct ow region.
Thus, the boundary layer growth along the louvers would reduce the heat
transfer coefcient (Nu or j factor). And Beamer et al. [118] found a good
414 . x. sn:n r1 :i.
Fic. 14. Inclined louvers (a counterpart of Fig. 5a).
correlation between the dropoff in the measured j factors in a full-scale heat
exchanger and the knee for dropoff in the ow efciency of Fig. 13 for
Re 150. Thus, they showed an excellent agreement among the CFD study,
ow visualization, and full-scale testing. The reasons for the performance
behavior were explained and an idea for further improvement of the n
geometry was suggested.
1. Concluding Remarks on L ouver Fin Performance/Analysis
The preceding survey shows that although the performance of louver ns
seems to be reasonably well understood, a general correlation for the
accurate quantitative prediction of their performance is not available. Also
at present, no modeling is available for 3D corrugated louver n geometry
(Fig. 1b). Numerical modeling can be used to provide valuable information
on the complex behavior of the ow and heat transfer of these surfaces.
However, careful grid generation strategy must be used to accurately predict
the details of both ow and thermal boundary layers on the n surfaces.
Two-dimensional models provide a fast tool for the assessment of the
relative performance of different geometries. Complete 3D conjugate models
are necessary if performance data are to be predicted accurately. Laminar
time-dependent models, although computationally demanding, can provide
further detailed understanding of some of the unsteady nature of the ow
over the louvers. At this stage, it is essential to conrm with precise
experimentation that whether the ow through the louvers is laminar
unsteady or low Reynolds number turbulent, both from the performance
enhancement and numerical analysis points of view. The best potential
future benet of numerical modeling of these surfaces is in providing
fundamental understanding of the performance of two promising variants of
the basic congurations. The rst is inclined louver ns (which acts like an
offset strip n; see Fig. 14, Tanaka et al. [120], and Suzuki et al. [121]),
which offer a signicant improvement in the ratio of heat transfer to
pressure drop. A second area of potential performance improvement stems
from the recognition that the rst few louvers in an array provide inferior
performance. The possibility of varying louver angle within an array
suggests itself as an area worthy of further consideration. Experimental
studies of these innovations are prohibitively expensive, but they lend
themselves well to numerical analysis.
C. W:vx Cn:NNris
1. Corrugated Wavy Channels
Considerable amount of numerical investigation has been conducted on
this channel geometry used in a plate-n exchanger as well as in a plate
exchanger with the chevron angle of 90. The numerical methods, operating
parameters and the geometries studied are given in Table III. A reference
book by Sunde n and Faghri [122] provides more details on the numerical
methods and analysis of wavy channels discussed later and other selected
compact heat exchangers.
Asako and Faghri [123] developed a solution methodology for laminar
ow and heat transfer in a corrugated duct. A nite volume scheme was
developed to predict fully developed ow, heat transfer coefcients, and
friction factors in a corrugated channel. The basic method was an algebraic
nonorthogonal coordinate transformation, which mapped the corrugated
channel into a rectangular domain. The governing equations of continuity,
momentum, and energy were solved assuming constant thermophysical
properties and excluding natural convection effects. The details on the
transformation of the conservation equations are provided by Faghri et al.
[124]. The boundary conditions used were constant wall temperature and
periodic ow at the inlet and outlet of the domain. As the ow was assumed
to be laminar, no turbulence model was used. Grid size effects were studied
and the maximum change in the Reynolds and Nusselt numbers, between a
1834 mesh and 2650 ne mesh, were within 3 and 5%. The numerical
results were compared with data from the open literature, but as the Re
range does not coincide, only qualitative agreement can be found. For the
range of the Reynolds number studied (100 Re
1000), the numerical

results for the friction factor indicate transition from laminar to turbulent
owregime. The slope of the friction factor vs Reynolds number curve increa-
ses from approximately 0.5 to almost 0 at Re
1000. This latter value

is representative of fully developed turbulent ow in a rough duct. For heat
transfer, in most of the cases studied, the Nusselt number increases with the
Reynolds number. These results indicate that for wavy corrugated channels,
416 . x. sn:n r1 :i.
TABLE III
Sixx:x or Nixric:i Mr1nons, Oir:1iNc CoNni1ioNs, :Nn Groxr1irs or Coic:1rn W:vx Cn:NNris
Boundary Reynolds Prandtl Pitch/height
Author Method Model Grid condition Inlet conditions number number ratio Validation
Asako and Faghri Finite volume Laminar Transformed Constant wall Fully developed 1001500 0.7, 4, 14 Pressure drop,
[123] Cartesian temperature and 8 Nusselt number
Xin and Tao Finite difference Laminar Polar-Cartesian Constant wall Fully developed 1001000 0.7 36 No validation
[128] temperature
Hugonnot [9] Finite difference Laminar Cartesian Adiabatic Fully developed 15010,000 7 3.33 Pressure drop,
kc visualization
Yang et al. [126] Finite volume Low-Re Transformed Constant wall Fully developed 1002500 0.7 2.56 Pressure drop,
model Cartesian temperature visualization
Ergin et al. [130] Finite volume kc Transformed Adiabatic Fully developed 5007000 0.7 1.4 Local velocities,
Cartesian velocity
uctuations
Ergin et al. [132] Finite volume Low-Re Transformed Adiabatic Fully developed 5007000 0.7 1.4 Local velocities,
model Cartesian velocity
uctuations
Kouidry [73] Finite volume DNS Curvilinear Adiabatic Fully developed 500010,000 7 3.33 Local velocities,
developing ow velocity
uctuations
McNab et al. Finite volume 3D: 30500 Cartesian Constant wall Fully developed 2504000 0.7 360 Pressure drop,
[137] meshes temperature Nusselt number
Tochon and Finite volume DNS Curvilinear Constant wall Developing ow 500010,000 7 3.33 Pressure drop,
Mercier [139] temperature Nusselt number
4
1
7
the transition from purely laminar to unsteady laminar and to turbulent
ow occurs at low Reynolds numbers and that the channel geometry has a
strong effect on the transition.
Asako et al. [125] assessed the heat transfer and pressure drop character-
istics of a similar corrugated duct with rounded corners. Computations were
carried out in the Reynolds number range 100 Re
1000 for several

geometric congurations. It was determined that the change in heat transfer
rates caused by rounding the corners of the sharp cornered corrugated duct
depended on the specic ow conditions, geometry, and performance
constraints.
To address the use of a laminar model when turbulence is developing,
Yang et al. [126] have extended the work of Asako and Faghri [123] by
using a low Reynolds number turbulence model. The same numerical
method as Asako and Faghri [123] was used, but the source terms in the
conservation equations were modied to take into account turbulent effects
in the diffusion coefcients. The turbulence was modeled according to the
Lam and Bremhorst [58] model, which is a low Reynolds number form of
the kc formulation. The details of the equations are given in Yang et al.
[126]. In parallel with this numerical work, experiments were performed in
order to validate the model. The test section was a sharp cornered
corrugated channel. The comparison was limited to the laminar model of
Asako and Faghri [123], as the low Reynolds number turbulent ow model
does not allow the modeling of geometries with sharp corners. In term of
the ow pattern, the size of the recirculation areas was well predicted. It
must be noted, however, that the recirculation region reaches a maximum
size at a Reynolds number of 500; the size decreases because of the high
diffusion in turbulent ow for higher Reynolds numbers. The same trends
were observed by Hugonnot [9] in a wavy channel. For friction factors for
a sharp edge channel, there is a good agreement with the laminar model for
Reynolds numbers up to 500, beyond which the friction factors are overes-
timated. Below this transition Reynolds number and for various operating
and geometric parameters, the predicted friction factors for the laminar and
turbulent models are the same. This suggests that the low Reynolds number
turbulent ow model is stable even for laminar ows. Beyond the transition
Reynolds number, the friction factors predicted by the turbulent model are
higher than those predicted by the laminar model. For the Nusselt numbers,
the same trends are observed. Yang et al. [126] claim that the low Reynolds
number turbulent ow model can be used to predict friction factors and
Nusselt number, but no comparison with experimental data is given.
Garg and Maji [127] applied a nite difference scheme (SIMPLEC
algorithm) to laminar ow through a wavy corrugated channel. They
presented results for both developing and fully developed ow for Reynolds
418 . x. sn:n r1 :i.
numbers ranging from 100 to 500. The channel dimensions were varied, and
for one chosen conguration, the detailed behavior of velocity, pressure, and
enthalpy for developing laminar ow were presented.
Xin and Tao [128] performed numerical simulation in wavy channel by
applying a laminar model and using a nite difference method. The domain
was gridded by a combination of polar and Cartesian coordinates. In the
bends, a polar coordinate was applied, and a Cartesian coordinate in the
straight duct between two bends. A fully developed ow was obtained by
periodic boundary conditions at the inlet and outlet of the domain and a
constant wall temperature boundary condition. Several geometric and
operating parameters were studied, but no validation with experimental
results was reported. The channel studied was quite similar to the one used
by Be rieziat [129], and comparisons can be made through the streamline
velocity proles. Xin and Tao [128] claimed that the relative strength and
size of the recirculation zone increase with the Reynolds number. This result
is in partial disagreement with the experimental data of Be reiziat [129] and
the ow visualizations of Hugonnot [9], which indicate that the recircula-
tion zones increase up to a Reynolds number of 200, then vortices are
generated downstream of the corrugation. For higher Reynolds numbers
(Re
350), the ow becomes unstable and turbulence is developing.

According to Be reiziat [129], the ow becomes fully developed turbulent for
Reynolds numbers above 2000. Applying a laminar model for ows devel-
oping turbulence leads to inaccurate predictions, and the conclusions of Xin
and Tao are only valid in the low Reynolds number range (Re
350).
Hugonnot [9] performed numerical simulations in a 2D corrugated
channel using a laminar model at Re
150 and a kc turbulence model

at Re
10,000. The choice of the model was dictated by the previous

observations on the ow structure. A nite difference method was applied
and the ow was considered adiabatic. The results were compared with ow
visualizations and pressure drop measurements in a similar geometry. The
size and the position of the recirculation zones were well predicted at both
Re
150 and Re
10,000. The pressure drop data were compared at

Re
10,000, and a good agreement was found for the overall pressure
drop, but differences can be noticed on the pressure prole essentially in the
recirculation zone. This result is not surprising, as it is well known that
conventional kc models are not accurate in separated ows, and even if the
ow pattern is qualitatively correct, the wall shear stress can be inaccurately
predicted. The comparison of the Hugonnot [9] and Xin and Tao [128]
results for low Reynolds numbers is only qualitative and similar.
Ergin et al. [130] applied the method developed by Faghri and co-
workers [123, 125, 126] to the study of the effect of interwall spacing on
turbulent ow in a sharp-edged corrugated channel. The ow was consider-
ed turbulent (500 Re
7000) and the kc model was adopted for

turbulent closure. Experiments were also conducted to validate the numeri-
cal model. At Re
2000, the ow characteristics given by the model were

in agreement with the visualization experiments and with previous experi-
mental studies [126, 131]. The comparison of the local velocity proles and
turbulent kinetic energy revealed that the proposed method gave accurate
prediction of the velocity proles, but the prediction of the turbulent kinetic
energy was relatively poor. As a result, the friction factors were underpredic-
ted by approximately 33%.
Ergin et al. [132] extended the previous work by applying the Lam
Bremhorst low Reynolds number turbulence model. For a Reynolds number
of 2000, the standard kc model and the low-Re turbulence model give
similar predictions of the ow eld. The comparisons of the predicted
friction factors showed a good agreement with the low-Re turbulence model
in the range 500 Re
3000, whereas the standard kc model is more

accurate for higher Reynolds numbers (Re
3000). As a consequence, the

standard and low-Re turbulence model cannot be applied in the whole range
of the Reynolds number, and this implies selecting a priori the model for the
range of the Reynolds number.
To overcome the problem of simulations of unstable turbulent ow,
Kouidry [73] performed direct numerical simulation of turbulent ow in a
corrugated channel. The geometry was identical to the one of Hugonnot [9]
and Be reiziat et al. [133]. In parallel to this numerical work, local velocity
measurements by laser anemometry were performed. The DNS method
requires a very ne mesh in order to reproduce the smallest turbulent
structures according to the Kolmogorov scale. Based on experimental data,
the smallest turbulent structure was about 95 jm, but limitations of the
workstation led to a 269 jm by 212 jm grid (29395 meshes). As a result,
the smallest turbulent structure was not represented, and the method used
was called pseudo-DNS. The TRIO code based on a nite volume method,
developed by the CEA (French Atomic Commission), was used. The
conservation equations were solved on a control volume with a semi-
implicit time scheme and a ne third-order space scheme (Quick-Sharp).
The advantage of DNS methods compared to time average methods (kc
models) is that they are independent of the ow conguration and that local
instantaneous information is available. Developing ow with a at inlet
velocity prole and periodic ows were studied. The numerical results were
compared to the experiments. For the mean velocity proles, the agreement
was relatively good in the fully developed ow, but differences linked to the
inlet boundary conditions were observed for the developing ow. The
turbulence intensity was analyzed through the mean square velocity uctu-
ations. The experimental uctuations were 30 to 50% around the mean
420 . x. sn:n r1 :i.
velocity, while the numerical simulations gave uctuations up to 150%.
Kouidry [73] claimed that these differences came from the number of time
steps required for the second-order average and by the fact that the
turbulent dissipation was underestimated by the 2D modeling. Despite these
differences, the ow pattern and the vortices are well predicted in terms of
the size and growth.
Farhanieh and Sunde n [134] investigated numerically a three-dimen-
sional fully developed laminar ow in a corrugated square duct. A nite
volume method was applied and the Reynolds number range from 30 to
1000. At constant pressure drop, the grid size effect was studied and
701828 grid points were selected. The numerical model was checked on
a straight square duct, and the predicted friction factor and Nusselt number
were in good agreement with analytical values. A parametric study on the
corrugated duct geometry was performed, but no comparison was given
with experiments.
Asako et al. [135] studied laminar forced convective heat transfer and
uid ow characteristics in a wavy duct with a trapezoidal cross-section.
The algebraic coordinate transformation (Faghri et al. [124]) was applied
to map the trapezoidal cross-section onto a rectangular one. The numerical
predictions of the friction factors were compared for a fully developed ow
in a straight trapezoidal duct, and the agreement with the data of Shah and
Bhatti [136] was within 0.46%. The numerical simulations showed that the
enhancement of heat transfer and pressure drop due to waviness depends
strongly on the Reynolds number. To optimize this geometry for industrial
applications, the authors suggest studying the effect of the wave length, wave
height and corner angle.
McNab et al. [137] computed the ow over herringbone corrugated
(sharp-cornered wavy) channels, using a commercial STAR-CD CFD code.
A 3D approach was adopted and a laminar model was used for Reynolds
numbers (based on D
"
) below 1500, and a high-Re kc model for higher
Reynolds numbers. For Reynolds numbers above 600, difculties appeared
in obtaining a fully converged solution. The authors suggested that ow
unsteadiness may occur for such low Reynolds numbers inducing pressure
uctuations. The maximum uctuation (14%) was observed for a Reynolds
number of 1500, and it was decided not to use time-dependent modeling.
The computed j and f values were compared with the measurements of
Abou-Madi [138] and are in relatively good agreement in the turbulent
regime 17 to 27%, but in laminar regime the differences are 33% for the
friction factor and 54%for the Colburn factor. The authors indicate that for
such low-Re unsteady ow, steady-state modeling may not be appropriate
for low Reynolds number and that the recirculation zone is not well
predicted by the kc model in turbulent ow.
Tochon and Mercier [139] have extended the work of Kouidry [73] using
an approach based on a large eddy simulation without a subgrid model of
turbulent ow. The advantage of using LES is to produce directly the main
part of coherent structures for turbulent regimes, for a better understanding
of the ow pattern. A curvilinear coordinate system has been adopted and
the rened mesh is 90 jm (or ve nondimensional wall units) near the wall
in the spanwise directions. The mesh is nearly constant in the streamwise
direction and the size is 200 jm (or 15 nondimensional wall units). As the
Kolmogorov scale is approximately 95 jm for the considered geometry
(Kouidry [73]), the chosen mesh seems to be sufcient to describe the main
part of the dissipative structures: no subgrid is needed. The mechanism of
eddy generation and heat transfer have been analyzed in the developing ow
(Figs. 15 and 16). The results are qualitatively and quantitatively compared
with experimental data obtained on a specic experimental rig and from
open literature, and the differences in the heat transfer coefcient and
friction factor do not exceed 10%.
The application of numerical modeling of a number of different high-
performance heat transfer surfaces has been presented by Atkinson et al.
[140]. The 2D and 3D models for louvered ns showed that accurate
calculations of overall heat transfer could only be achieved by using 3D
models, which incorporate the effects of tube surface area and n resistance.
Accurate predictions of ow and heat transfer over corrugated ns in the
laminar ow region were also presented. However, they showed that the
high-Re kc model with log-law wall functions gave poor predictions of heat
transfer for the turbulent-ow regime. They suggested that low-Re forms of
the model would be more accurate in this regime.
2. Wavy Furrowed Channels
Sobey [141] conducted a numerical study of ow through furrowed
channels in conjunction with a related experimental investigation [142].
Sobey calculated the ow patterns obtained using both steady and pulsatile
inow. The effects of varying dimensional parameters and Reynolds number
were examined. Furthermore, the ow structures that occur in a channel
with arc-shaped walls were compared with the patterns induced by
sinusoidal curved wavy walls. An oscillatory ow in various types of wavy
passages was examined further in subsequent studies performed by Sobey
[143, 144].
Sparrow and Prata [145] examined a family of periodic ducts, using both
numerical and experimental methods. The periodic duct is a tube consisting
of a succession of alternately converging and diverging conical sections.
Numerical simulations were carried out for fully developed laminar ow in
422 . x. sn:n r1 :i.
Fic. 15. Numerical results for time-dependent evolution of vorticity in a 2D wavy channel
[139].
the Reynolds number range Re 1001000, for various duct congur-
ations. The resulting data indicated that the periodic furrowed tube is not
conducive to heat transfer enhancement for steady laminar ow.
Garg and Maji [127] used a nite difference method to solve the
governing equations for steady laminar ow and heat transfer in a furrowed
wavy channel. Calculations were performed using various wall amplitudes
for Re 100500. Both the developing and fully developed ow regions
were analyzed. The local Nusselt number was observed to uctuate
sinusoidally in the fully developed region. Moreover, the Nusselt number
Fic. 16. Numerical results for time-dependent evolution of temperature in a 2D wavy
channel [139].
increased with the Reynolds number, unlike a constant value for laminar
ow through a straight channel.
Blancher et al. [146] analyzed convective heat transfer by a spectral
method for an expanding vortex in a wavy furrowed channel. The ow is
assumed laminar fully developed with the Reynolds number up to 200. An
algebraic transformation of the coordinate was applied to reduce the
physical domain to a rectangular computational domain. The governing
equations were solved introducing the stream functions, with the details
provided on the mathematical models and transformations. The method is
limited to Reynolds numbers above 200, as the ow becomes unsteady for
higher values. The numerical results (ow pattern, wall shear stress, and
vorticity proles), compared to those of Sobey [141], are in good agreement.
Guzman and Amon [147] reported the transition to chaos for fully
developed ow in furrowed wavy channels. Specically, the Ruelle
TakensNewhouse scenario for the onset of chaos was veried using direct
424 . x. sn:n r1 :i.
numerical simulations. The results were illustrated for various Reynolds
numbers using velocity time signals, Fourier power spectra, and phase space
trajectories.
Guzman and Amon [148] extended their previous study by using DNS
to calculate dynamical system parameters. Dynamical system techniques,
such as time-delay reconstructions of pseudophase spaces, Poincare maps,
autocorrelation functions, fractal dimensions, and Eulerian Lyapunov expo-
nents, were employed to characterize laminar, transitional, and chaotic ow
regimes. Also, 3D simulations were performed to determine the effect of the
spanwise direction on the route of transition to chaos. They have shown that
the transition to chaos takes place at Re 500 following a quasi-periodic
frequency locking route.
Wang and Vanka [149] applied a nite volume method to the study of
convective heat transfer in wavy furrowed channels. To solve the governing
equations, an accurate numerical scheme was applied on a 256128
calculation domain. The simulations were performed for both steady and
unsteady regimes. The ow was observed steady until Re 180. Afterward,
self-sustained oscillations appeared and led to destabilization of the laminar
ow. Physical interpretation of the transition was given and the ow
structure was analyzed for Reynolds numbers up to 1000. The friction factor
calculations, compared to the experimental results of Nishimura et al. [150],
slightly underpredicted the measurements but showed a good agreement
with experimental trend. No comparison with heat transfer experiments was
given in the paper.
D. CnrvoN Toicn Pi:1rs
Chevron plates with a corrugation angle of 90 can be considered as a 2D
wavy corrugated geometry. Numerous investigations have been published
on this specic geometry and are discussed in the previous section. In most
of the cases, the corrugation angles are between 30 and 60 in commercial
chevron plates. As observed from the experimental work, the ow remains
mainly in the furrow of the corrugation for the corrugation angle of 30;
whereas for higher corrugation angles (60), the ow is almost 3D and
highly turbulent even for very low Reynolds numbers (Re
200). Flow
simulation in cross corrugated structures has been reported by Fodemsky
[151], Ciofalo et al. [152], and Hessami [153]. The numerical and ow
conditions are summarized in Table IV.
Fodemsky [151] used a standard kc model to predict ow and heat
transfer in a corrugated channel. A body tted Cartesian grid was applied
to model the channel geometry, and particular attention was paid to the
grid in the corners where the upper and lower plates were in contact. The
results were provided for corrugation angles between 15 and 20, and the
comparison with experimental results was only qualitative. The simulation
provided the heat transfer coefcient eld and outlined large heterogeneities.
The highest values were located at the top of the corrugations where the
cross section is minimized. This result is in agreement with the data of
Gaiser and Kottke [11]. As for the 2D modeling, kc models are not
accurate to predict separated ow where recirculations zones exist, and the
results are only qualitative.
The study of Ciofalo et al. [152] is more extensive. Both numerical and
experimental work was performed and several numerical methods were
applied on a large range of operating and geometric conditions (see Table
IV). Similar to Fodemskys work, a body-tted grid was used to model the
complex geometry. The same grid was used for all the models, except for the
standard kc model, where the wall function must be respected. This implies
that the center of the control volume near the wall lies far out from the
viscous sublayer. Ciofalo et al. [152] used a conventional denition, and the
thickness of the viscous sublayer was based on plain tube correlations. This
assumption is incorrect because the thickness of the viscous sublayer is not
constant, a result of the large recirculation area. The results were compared
to pressure drop measurements of Focke and co-workers [7, 10]. Laminar
and kc computations were unsatisfactory, as they respectively underpredic-
ted and overpredicted the friction factors for the complete range of Reynolds
numbers investigated (1000 Re 5000). The best agreement with the
experimental data was obtained with the low-Re turbulence model or LES;
the difference was about _50%for the complete range of Reynolds numbers
and for various corrugation angles. The agreement was better for Reynolds
numbers over 2000 and for corrugation angles between 15 and 30. For the
average Nusselt number, the laminar and standard kc models are not
satisfactory and the best agreement was found with the low-Re turbulence
model and LES. These two models underpredicted the average Nusselt
number by 20%, but the corrugation angle effect was well taken into
account. For the local ow structure and heat transfer, the comparison was
made with experiments performed by Stasiek et al. [14] on a similar
corrugated channel. The Nusselt number vs Re experimental results were
reasonably well predicted by the low-Re turbulence model and LES,
whereas the laminar and kc models were not satisfactory. The best
agreement was obtained with LES. No quantitative comparison of the ow
structure was obtained and the accuracy of the different models was not fully
validated. Nevertheless, it appears that LES allowed reproducing the ow
structure according to Ciofalo et al. [152]. Finally, Ciofalo et al. compared
LES and DNS and observed that for their conguration, the DNS models
did not improve the results signicantly.
Hessami [153] performed a three-dimensional study of heat transfer and
426 . x. sn:n r1 :i.
TABLE IV
Sixx:x or Nixric:i Mr1nons, Oir:1iNc CoNni1ioNs, :Nn Groxr1irs or CnrvoN Toicn Pi:1rs
Boundary Reynolds Prandtl Corrugation
Author Method Model Grid condition Inlet conditions number number Pitch/height angle Validation
Fodemsky Finite difference kc Body tted Constant wall Fully developed 16003000 0.7 2 1520 No
[151] Cartesian 12` temperature
Ciofalo et al. Finite volume Laminar, Body tted Constant wall Fully developed 10`10" 0.7 24 1575 Nu, f
[152] kc Cartesian 24` temperature developing ow Local Nu
LowRe
LES Body tted Constant wall Fully developed
Cartesian 32` temperature developing ow
Hessami Finite volume Laminar, Body tted Constant wall Developed 2006000 7 3.855.86 30, 45, 60 No
[153] kc Cartesian temperature
LowRe and heat ux
Blomerius Finite volume DNS 0.10.10.05 Constant wall Fully developed 1502000 0.7 3.5, 4.5, 5 45 Nu and f
et al. [156] mm temperature
Sunde n Finite volume laminar 252525 Constant wall Fully developed 40010000 4.33 and 3 0, 22.5, and Nu and f
[157] LowRe Cartesian temperature 12.8 45
RNG-kc body tted
4
2
7
pressure drop in a cross-corrugated heat exchanger, using the commercial
CFD software CFX 4.1 (a version of FLOW3D software). Three models
were considered: laminar, standard kc and low-Re kc turbulent models.
A unit cell was represented by a 221622 grid. Preliminary investigations
showed additional blocks or cells were required at the inlet and outlet of the
unit cell to obtain a stable solution. The ow boundary condition was a fully
developed velocity prole, but no precision was given on the establishment
ow regime. Three corrugation angles (30, 45 and 60) and three pitch-to-
height ratios (3.85, 4.55, and 5.56) were considered; the Reynolds numbers
were covered up to 6000. For a corrugation angle of 60 (hard plates),
comparison of the three models showed large differences. For heat transfer,
both the laminar and standard kc models gave a very low dependency of
the Nusselt number with the Reynolds number (Nu Re"); in contrast,
the low-Re turbulence model was very sensitive (NuRe`). These de-
pendencies with the Reynolds number were not physical and may have come
from the grid or boundary conditions used in this study. The trends on
predicted Nu and f were not in agreement with the experimental work of
Focke et al. [154] and Thonon et al. [12], and neither the Reynolds number
effect nor the corrugation angle effect could be predicted by the model used
by Hessami [153]. Other geometric parameters were studied by Hessami,
but the main result of this study was that the boundary conditions and the
numerical model have a large inuence on the results, and that great care
should be taken in modeling such complex geometries.
Sawyers et al. [155] studied numerically, 2D (sinusoidal), and 3D (egg-
carton, i.e., sinusoidal in the transverse direction as well) corrugated (wavy)
channels. To avoid unsteady ow, the study was limited to low Reynolds
numbers (Re
250). Both algebraic and numerical methods were applied

to predict pathlines, pressure drop, and heat transfer. The authors analyzed
several geometric congurations, but no comparison with experimental
results was provided.
Blomerius et al. [156] presented numerical results on 3D ow eld and
heat transfer in a corrugated channel. The NavierStokes were directly
solved by the code, using an implicitexplicit scheme for the convective
uxes. A reinjection procedure was adopted to obtain a fully developed ow.
Depending on the geometry, 80,000 to 150,000 grid points were used. The
sensitivity of the results (heat transfer and pressure drop) with the gridding
was up to 10% at Re 1500. The corrugation angle was xed at 45, and
the pitch-to-height ratio was 3.5, 4.5, and 5. This latter value was selected
for a comparison with the experimental results of Gaiser and Kottke [11].
For Reynolds numbers below 180270 (depending on the pitch-to-height
ratio), the ow was fully laminar; for higher Reynolds numbers, the ow
became time-periodic self-sustained oscillatory, requiring time-dependent
428 . x. sn:n r1 :i.
calculations. For high Reynolds numbers, a time averaging procedure was
adopted to estimate heat transfer and pressure. At Re 2000, the deviation
between the predicted and experimental results was respectively 5 and 8%
for the Nusselt number and friction factor.
Sunde n [157] analyzed the ow structure and thermal and hydraulic
performances of plate and frame heat exchangers using CFD. The geometry
studied was representative of an industrial plate heat exchanger with two
different corrugation angles (22.5 and 45). A single unit cell was considered
(volume between four contact points of chevron plates) to provide informa-
tion on friction factors and Nusselt numbers. A steady laminar ow model
was applied for Reynolds numbers up to 2000 and a low-Reynolds kc
or an RNGkc model with wall functions for turbulent ows. As outlined
in Section IV, these turbulent models are more accurate for low Reynolds
number unsteady ows or turbulent ows. A 3D body tted mesh was used,
and great care was paid to model the contact points. To avoid singularities,
the contact line between the upper and lower plates was replaced by a
surface contact. For one of the geometries modeled, the numerical results
were compared with experiments. The predicted Nusselt number was
underpredicted by at most 25% while the friction factor was underpredicted
by 1740%. The author suggested using a more advanced turbulence model
to get a higher accuracy.
1. L ocal Analysis of Flow and Heat Transfer Phenomena
in Corrugated Channels
Based on the numerical analysis, local information on the ow structure
is now available and allows analyzing and qualitatively characterizing heat
transfer mechanisms in complex geometries. The experimental studies (see
Section II, B) have shown complex ow patterns and transition from steady
laminar to unsteady laminar and turbulent ow at low Reynolds numbers.
The numerical studies have conrmed these observations, but the mechan-
ism of eddy creation can only be more precisely dened using unsteady
numerical models. The 2D studies were performed by Tochon and Mercier
[139], and 3D studies by Ciafolo et al. [152] and Blomerius et al. [156]. The
information gained from these studies is discussed next.
For the 2D geometry, near the entrance and at each corrugation, the ow
separates from the wall and generates a large recirculation zone. Because of
turbulent instabilities, this zone grows larger and larger and then breaks
suddenly in two parts. One part creates a vortex, which is released in the
ow (eddy no. 1 in Fig. 15) in the downward direction, as the other part
remains in the protected region. Interactions between the rotating eddies in
the center part lead to a pairing (eddy nos. 4 and 5 in Fig. 15). The same
process is repeated. When detached from the wall, eddies are convected in
the ow, under the inuence of other eddies. They can keep their coherent
structure for several microseconds. Their lifetime has been found to be a
little longer numerically than in the experiment. This discrepancy can be
explained by the limitation of the two-dimensional model, although turbu-
lence is three-dimensional in nature. This 3D character is especially true in
the region of an adverse pressure gradient, where partial disorganization of
the coherent structures seems to occur in the experimental visualization.
Nevertheless, the main mechanisms of eddy creation and motion are very
similar to those observed experimentally. This alternative generation of
vortices on the lower and upper walls of the domain of calculation liberates
coherent eddies, with either clockwise or anticlockwise rotational motion.
As a result, it is also apparent that this geometry is able to ensure good
turbulent mixing of the uid after only a few corrugations. Concerning the
thermal aspects (Fig. 16), the alternating vortices lead to a uniform tempera-
ture inside the channel by mixing the warm liquid of the recirculating areas
with the main ow. Indeed, local investigations show that the heat transfer
coefcient is low inside the recirculating areas by means of the conduction
process and is high at the reattachment point. So, because of the turbulent
behavior of the ow, the hot pocket oscillates, becomes unstable, and is
convected to the bulk ow. These new eddies interact with each other and
generate large coherent structures that create a fully developed turbulent
ow. So the vortex shedding from the warm area near the wall to the core
is the main mechanism for heat transfer.
For 3D geometries, the ow is more complex to analyze, as the 3D effects
are present even for low corrugation angles. Ciafalo et al. [152] have
investigated a cross-corrugated channel, using an LES model, with the
corrugation angle between 18.5 and 30. For these typical angles, according
to Focke and Knibbe [10], a furrow ow should exist, and the mixing
intensity should increase with the Reynolds number. From the velocity
uctuations, a Strouhal number can be deduced (Sr Ft, where F is the
frequency and t the time necessary to cross a unitary cell at the average
velocity U). For all values of corrugation angles (18.5 to 30) and Reynolds
numbers (780 to 4250), a constant value of Sr 3 was found. A detailed
examination of the instantaneous and uctuating ow elds (at Re 2450)
indicated that the eddies were highly three-dimensional in the central part
of the cell, having their largest component in the vertical direction (normal
to the channel). These eddies were generated by mixing of two furrow ows
(upper and lower plates). The intensity of the uctuations was found to
increase with both Reynolds number and corrugation angle.
Blomerius et al. [156] investigated a corrugated channel with a 45
430 . x. sn:n r1 :i.
corrugation angle. According to previous visualization experiments, the ow
structure is strongly affected by the Reynolds number and becomes highly
three-dimensional. A Fourier analysis was performed on the u-component
of the velocity (main ow direction). For Reynolds numbers of 255 and 425,
a dominant frequency with corresponding Strouhal numbers of 0.5 and 1.3
was found. For a higher Reynolds number based on the hydraulic diameter
(Re 1700), no dominant frequency could be detected and the ow was
aperiodic, which is the characteristic of turbulent ow. For a low Reynolds
number (Re 180), the midplane shear layer was relatively stable, whereas
for a higher Reynolds number (Re 1800) considerable mixing between
the two streams was noticed. The maximum velocities were nearly twice
the average uid velocity. This higher mixing at Re 1800 led to a much
more homogeneous temperature eld over the whole cross-section. In the
region just downstream from the contact point, the uid temperature was
almost the same as the wall temperature, indicating a stagnant zone. This
size of this region was strongly dependent on the Reynolds number. The
foregoing information is fundamental for selecting the most appropriate
geometry for given process conditions or to develop new geometries that
could limit the size of these stagnant and recirculation areas. In the food
industry, pasteurization and sterilization are often achieved in plate heat
exchangers, and it is fundamental to have an equal (homogeneous) residence
time for all uid particles and no dead zones where fouling could initiate.
The numerical studies have shown that for Reynolds numbers higher than
1500 and for corrugation angles higher than 45, the recirculation zones are
limited in size and the uid temperature is much more homogeneous.
Furthermore, these channels have a very high local shear stress, which help
mitigating fouling.
2. Summary of 2D and 3D Modeling of Wavy Channels
and Chevron Trough Plates
For 2D geometries (corrugated wavy channels), several grid schemes have
been applied, but the major difference lies in the numerical method used.
Laminar models are limited to very low Reynolds numbers (Re 350) as
instabilities occurs for higher Reynolds numbers. Furthermore, in fully
developed turbulent regimes, conventional turbulent models can only give
overall information, as the ow is inaccurately predicted in the recirculation
zones. As a result, to cover the entire range of the Reynolds number and to
be independent of ow patterns, advanced methods that enable prediction
of separated ows must be applied. The low Reynolds number turbulent
ow model [126] appears to constitute progress in the simulation of ows
in complex geometries, but its validity is not established. DNS methods [73]
are interesting, but 2D models underestimate the turbulent diffusion and
efforts must be applied to 3D modeling (i.e., the realistic chevron plates).
Concerning 3D modeling, the studies have covered a wide range of
corrugation angles and Reynolds numbers, and the agreement compared to
experimental results (average values of the friction factor and Nusselt
number) is within 10 to 50%. Local heat transfer coefcients and wall shear
stress are still not accurately predicted.
Hessami [153] and Sunde n [157] used various turbulence models, but the
accuracy was relatively poor. Nevertheless, the general trends for heat
transfer and pressure drop were well predicted. Ciofalo et al. [152] used
several numerical models and claimed that the low Reynolds number
turbulent ow model or LES gave satisfactory results. The latter method is
computationally more expensive but provides instantaneous information.
Blomerius et al. [156] used a DNS method and claimed a maximum
deviation of 5 and 8%, respectively for Nusselt numbers and friction factors.
These results are encouraging, and advanced turbulent models implemented
in commercial CFD codes should be tested for complex geometries such as
cross-corrugated heat exchangers.
As computational costs will decrease, LES or DNS methods must be
developed, and careful validation with local measurements (velocity and
turbulence intensity) must be performed. Progress in modeling ow and
heat transfer will enable study of complex geometries. More sophisticated
experiments will be required for validation of modeling and numerical
results. A comparison of CFD simulations between researchers, using
different research and commercial codes and techniques on standardized
geometries, and a comparison with validated laboratory and industrial data
must be an objective for CFD to be a design, analysis, and optimization tool
for heat exchangers.
VI. Conclusions
A comprehensive review is made of numerical analysis of some of the
important surface geometries of compact heat exchangers: offset and louver
n geometries used in plate-n exchangers, wavy corrugated and furrowed
ns/channels in tube-n and plate heat exchangers, and chevron plates in
plate heat exchangers. Specic understanding obtained from the numerical
studies of each n/plate geometry is summarized at the end of each n/plate
geometry section. Also, some of the known physics of ow from experimen-
tation and ow visualization is presented to further assist in the renement
in the numerical analysis. Although numerical analysis has progressed
432 . x. sn:n r1 :i.
considerably in the past 20 years, most of the numerical analyses reported
are 2D and/or simpler than the real complex ows. Most of the ows
encountered in CHE applications are unsteady, often with ow separation
and recirculation zones, and the Reynolds number range is typically
1002000 for most compact heat exchanger surfaces. The separation and
reattachment of the ow and vortices in the wake of the ns affect both heat
transfer and pressure drop, and these mechanisms cannot be predicted with
time average modeling (conventional kc models). Unsteady laminar model
requires a very ne description of the geometry, and computation capacity
limits its application to 2D modeling. There is no clear evidence as yet that
the ows in this Reynolds number range are unsteady laminar or low
Reynolds number turbulent ows, although it is possible that gradually the
unsteady laminar ows associated with the interruptions become low
Reynolds number turbulent ows. Hence, more sophisticated experimenta-
tion is required to determine the ow characteristics. At the same time, RSM
and LES models should be further developed to simulate the local ow and
heat transfer characteristics of compact heat exchanger surfaces. See Section
IV for specic conclusions/recommendations on turbulence models. Since
the art of compact heat exchanger surfaces has reached an asymptotic level,
further enhancement and innovation in those surfaces will come from the
detailed accurate numerical analysis. Numerical analysis is also needed to
study the effects of minor changes in the n prole (such as those due to the
aging of manufacturing tools). Manufacturers of heat exchangers in the
automotive industry are active in this area, but the numerical methods
proposed in commercial software require some careful validation with
experimental data. A similar conclusion was also reported earlier (which
came to the authors attention) by Baggio and Fornasieri [158]. It is a
challenge to numerical analysts to numerically simulate highly complex
ows accurately to provide a basic understanding of ow and heat transfer
phenomena for their exploitation in designing new and improved heat
transfer surfaces.
Acknowledgments
We gratefully acknowledge Mr. J. P. Chevalier of CEA-Grenoble, DTP/GRETh, Grenoble,
France, for his assistance in preparing Section IV on turbulence models. We appreciate very
much a careful and extensive review of our chapter by Dr. F. Ladeinde of Thaerocomp
Technical Corp of Stony Brook, NY, USA. We are also thankful to Prof. W. Rodi of the
University of Karlsruhe, Germany, Dr. S. Sazhin of the University of Brighton, UK, Prof. G.
Biswas of the Indian Institute of Technology Kanpur, India, and Dr. G. Xi of Daikin Industries,
Inc., Osaka, Japan, for their critical review of Section IV on turbulence models.
Nomenclature
A heat transfer surface area on
one uid side, m`
A constant of Eq. (11), m
A
"
minimum free ow area on one
uid side of a heat exchanger,
m`
b spacing between plates for
corrugated ns, mean gap, or
plate spacing in a plate heat
exchanger, m
c
specic heat of uid at constant

pressure, J/kg K
D particular derivative:
D
Dt
c
ct
u V
, 1/s
D
equivalent diameter, D
2b, m
D
"
hydraulic diameter, D
"
4V/A
4A
"
L /A, m
E constant of Eq. (28)
F frequency, 1/s
F
`
structure function dened in
Eq. (42), m`/s`
f Fanning friction factor,
dimensionless
G uid mass velocity based on
minimum free ow area, Gm /
A
"
, kg/m`s
G
'
generation of turbulent kinetic
energy due to buoyancy, m`/s`
G
'
generation of turbulent kinetic
energy due to the mean velocity
gradients, m`/s`
h convective heat transfer
coefcient, W/m`K
j Colburn factor, St Pr``,
dimensionless
K constant of Eq. (28)
k kinetic energy of turbulence per
unit mass, J/kg (or m`/s`)
l mixing length, m
l offset strip length, louver,
length, or louver pitch, m
m uid mass ow rate, kg/s
Nu Nusselt number, hD
"
/k,
dimensionless
P mean pressure component of
uid static pressure, Pa
Pr uid Prandtl number,
dimensionless
p' uctuating pressure
component, Pa
p
n pitch, m
p
'
tube pitch of at tubes in Fig.
11b, m
q
''
velocity scale, m/s
R correction term in Eq. (18),
m`/s`
Re Reynolds number based on
hydraulic diameter, GD
"
/j,
dimensionless
Re
Reynolds number based on the

equivalent diameter, GD
/j,
dimensionless
Re
l
interrupted length, Gl/j,
dimensionless
Re
'
mixing length and mean
velocity, jUl/j, dimensionless
r correlation distance used for
structure function, m
S strain tensor, m`/s
Sr Strouhal number, Sr Ft
"'
,
dimensionless
St Stanton number, St h/Gc
,
dimensionless
s n spacing, s p
o, m
T uid temperature, K
t time, s
t
"'
characteristic time, s
At characteristic time,
characteristic length of
corrugation divided by the bulk
velocity, see Figs. 15 and 16, s
U mean velocity component, m/s
u', v', w' uctuating velocity
components, m/s
u, v, w velocity components, m/s
V void volume on one uid side,
m`
x axial direction, m
Y
contribution of the uctuating

dilatation in the dissipation
rate, m`/s`
434 . x. sn:n r1 :i.
y normal distance from the wall,
m
y wall coordinate y(t
`
/j)`/v,
dimensionless
Grrx Lr11rs
:
'
thermal diffusivity, m`/s
[ cubic dilatability in Section IV,
1/K
[ corrugation angle for chevron
plates, measured from the
axis parallel to the plate length,
[90, (see Fig. 3), deg
[* lRe/s, dimensionless
A lter width, m
o boundary layer thickness in
Section IV, n thickness in all
other sections, m
c dissipation rate of turbulent
kinetic energy, J/kg s (or m`/s`)
0
uid temperature, K
0
louver angle for multilouver

ns (see Fig. 5a), deg
v molecular viscosity, m`/s
v
'
turbulent viscosity, m`/s
j uid density, kg/m`
o turbulent Prandtl number
t subgrid scale strain, m`/s`
t
`
wall shear stress, Pa
D rotation velocity, 1/s
c vorticity, s
Sinscii1s
eff effective
i axial direction index
j lateral direction index
k vertical direction index
c related to the dissipation rate
v related to the kinematic
viscosity
Annrvi:1ioNs
ASM algebraic stress model
CFD computational uid dynamics
DNS direct numerical simulation
EVM eddy viscosity model
LES large eddy simulation
NLEVM nonlinear eddy viscosity model
OSF offset strip n
RANS Reynolds averaged numerical
simulation
RNG renormalization group
RSM Reynolds stress model
SGS subgrid scale
References
1. Jacobi, A. M. and Shah, R. K. (1998). Air-side ow and heat transfer in compact heat
exchangers: a discussion of enhancement mechanisms. Heat Transfer Eng. 19(4), 2941.
2. Cowell, T. A. , Heikal, M. R., and Achaichia, A. (1995). Flow and heat transfer in compact
louver n surfaces. Exp. Thermal Fluid Sci. 10, 192199.
3. Chang, Y. J., and Wang, C. C. (1997). A generalized heat transfer correlation for louver
n geometry. Int. J. Heat Mass Transfer, 40, 533544.
4. Cox, S. G., Downie, J. H., Heikal, M. R., and Cowell, T. A. (1997). A novel method for
investigating the performance characteristics of variable louvre angle plate ns. Proc. 5th
UK National Heat Transfer Conf., IChemE, London.
5. Ali, M. M., and Ramadhyani, S. (1992). Experiments on convective heat transfer in
corrugated channels. Exp. Heat Transfer 5, 175193.
6. Gschwind, P., Regele, A., and Kottke, V. (1995). Sinusoidal wavy channels with Taylor
Go rtler vortices, Exp. Thermal Fluid Sci. 11, 270275.
7. Focke, W. W., and Knibbe, P. G. (1984). Flow visualisation in parallel plate ducts with
corrugated walls, CISR Report CENG M-519, Pretoria, South Africa.
8. Shah, R. K., and Focke, W. W. (1988). Plate heat exchanger and their design theory. In
Heat Transfer Equipment Design, Hemisphere Publishing Corporation, New York, pp.
227254.
9. Hugonnot P. (1989). Etude Locale et Performances Thermohydrauliques a` Faibles
Nombres de Reynolds dun Canal Plan Corrugue : Application aux Echangeurs a` Plaques.
The` se Universite de Nancy I, France, Juin.
10. Focke W. W., and Knibbe, P. G. (1986). Flow visualisation in parallel-ducts with
corrugated walls. J. Fluid Mech. 165, 7377.
11. Gaiser, G., and Kottke, E. V. (1990). Effect of corrugation parameters on local and
integral heat transfer in plate heat exchangers and regenerators. Proc. 9th Int. Heat Mass
Transfer Conf., Heat Transfer 1990, 5, 8590.
12. Thonon B., Vidil R., and Marvillet C. (1995). Recent research and developments in plate
heat exchangers. J. Enhanced Heat Transfer 2, 149155.
13. Be reiziat D., Devienne R., and Lebouche M. (1992). Characterising the ow structure
inside a rectangular channel by use of electrochemical method: case of non-Newtonian
uids. First European Heat Transfer Conference, IChemE Series No. 129, 2, 875889.
14. Stasiek, J., Collins, M. W., and Ciofalo, M. (1996). Investigation of ow and heat transfer
in corrugated passages I. Experimental results. Int. J. Heat Mass Transfer 39, 149164.
15. Heikal, M., Drakulic , R., and Cowell, T. A. (2000). Multi-louvred n surfaces. In Recent
Advances in Analysis of Heat Transfer for Fin Type Surfaces (B. Sunde n and P. J. Heggs,
Eds.), WIT Press, Southampton, UK, 211250.
16. Thompson, J. F., Warsi, Z. U. A., and Mastin, C. W. (1985). Numerical Grid Generation.
North-Holland, New York.
17. Anderson, J. D. (1995). Computational Fluid Dynamics. McGraw-Hill, New York.
18. Melton, J. E. (1996). Automated three-dimensional Cartesian grid generation and Euler
ow solutions for arbitrary geometries. Ph. D. thesis, U. C. Davis, California.
19. Shaw, C. T. (1992). Using Computational Fluid Dynamics. Prentice Hall, New York.
20. Mercier, P., and Tochon, P. (1997). Analysis of turbulent ow and heat transfer in
compact heat exchanger by pseudo direct numerical simulation. In Compact Heat
Exchanger for the Process Industries (R. K. Shah, Ed.). Begell House Inc., New York, pp.
223230.
21. Orlanski, I. (1976). A simple boundary condition for unbounded hyperbolic ows. J.
Compu. Phys. 21, 251269.
22. Kieda, S., Suzuki, K., and Sato, T. (1981). Numerical study on ow behavior and heat
transfer in the vicinity of starting point of transpiration. In Numerical Methods in L aminar
and Turbulent Flow (C. Taylor and B. A. Schreer, Eds.). Prineridge Press, Swansea, UK,
pp. 905916.
23. Xi, G. N. (1992). Flow and heat transfer characteristics of n arrays in the low and middle
Reynolds number ranges. Doctoral thesis, Kyoto University, Kyoto.
24. Leonard, B. P. (1979). A stable and accurate convective modeling procedure base and
quadratic upstream interpolation. Compt. Meth. Appl. Mech. 19, 5998.
25. Moin, P. (1997). Numerical and physical issues in large eddy simulation of turbulent ows.
Proc. Int. Conf. on Fluid Eng., Tokyo 1, 91100.
26. Lane, J. C., and Loehrke, R. J. (1980). Leading edge separation from a blunt at plate at
low Reynolds number. ASME J. Fluid Eng. 102, 494496.
27. Ota, T. Asano, Y., and Okawa, W. J. (1981). Reattachment length and transition of
separated ow over blunt at plates. Bull. JSME 24, 941947.
28. Rodi, W. (1993). Turbulence Models and Their Application in Hydraulics, 3rd ed. Balkema,
Rotterdam, The Netherlands.
436 . x. sn:n r1 :i.
29. Halbaeck, M., Henningson, D. S., Johansson, A. V. and Alfredsson, P. H., Eds. (1996).
Turbulence and Transition Modelling. Kluwer Academic Publishers, The Netherlands.
30. Wilcox, D. C. (1993). Turbulence modelling for CFD. DCW Industries, Inc., La Can` ada,
California.
31. Speziale, C. G. (1991). Analytical methods for the development of Reynolds-stress closures
in turbulence, Ann. Rev. Fluid Mechanics, 23, 107157.
32. Baldwin, B. S., and Lomax, H. (1978). Thin layer approximation and algebraic model for
separated turbulent ows. AIAA 16th Aerospace Sciences Meeting, Paper No 78257,
Huntsville, AL, January 1618.
33. Spalart, P., and Allmaras, S. (1992). A one-equation turbulence model for aerodynamic
ows. Technical Report AIAA-92-0439, American Institute of Aeronautics and Astronaut-
ics.
34. Launder, B. E., and Spalding D. B. (1972). Lectures in Mathematical Models of Turbulence.
Academic Press, London.
35. Launder, B. E., and Spalding, D. B. (1974). The numerical computation of turbulent ows.
Comp. Meth. Appl. Mechanics Eng., 3, 269289.
36. Jones, W. P., and Launder, B. E. (1972). The prediction of laminarisation of a two
equation of turbulence. Int. J. Heat Mass Transfer 15, 301314.
37. Yakhot, V., and Orszag, S. A. (1986). Renormalization group analysis of turbulence: I.
Basic theory. J. Sci. Computing 1, 151.
38. Saha, A. K., Biswas, G., and Muralidhar, K. (1999). Numerical study of the turbulent
unsteady wake behind a partially enclosed square cylinder using RANS. Comput. Meth.
Appl. Mech. Eng. 178, 323341.
39. Kato, M., and Launder, B. E. (1993). The modelling of turbulent ow around stationary
and vibrating square cylinders. Proc. 9th Symposium on Turbulent Shear Flows, Kyoto,
Japan, 1014.
40. Yakhot, V., Orszag, S. A., Thangam, S., Gatski, T. B., and Speziale, C. G. (1992).
Development of turbulence models for shear ows by a double expansion technique. Phys.
Fluids A4, 15101520.
41. Lyn, D. A., Einav, S., Rodi, W., and Park, J. -H. (1995). A laser-Doppler velocimetry study
of ensemble-averaged characteristics of turbulent near wake of a square cylinder. J. Fluid
Mech. 304, 285319.
42. Shih, T. H. Liou, W. W. Shabbir, A., and Zhu, J. (1995). A new kc eddy-viscosity model
for high Reynolds number turbulent ows model development and validation. Com-
puter Fluids 24, 3, 227238.
43. Hwang, C. B., and Lin, C. A. (1999). A low Reynolds number two-equation k
model
to predict thermal elds. Int. J. Heat Mass Transfer 42, 32173230.
44. Kays, W. M., and Crawford, M. E. (1993). Convective Heat and Mass Transfer, 3rd ed.
McGraw-Hill, New York.
45. Launder, B. E., and Sharma, B. I. (1974). Application of the energy dissipation model of
turbulence to ow near a spinning disc, L etters in Heat and Mass Transfer 1, No. 2,
131138.
46. Yap, C. R. (1987) Turbulent heat and momentum transfer in recirculating and impinging
ows. Ph. D Thesis, Faculty of Technology, University of Manchester, UK.
47. Suga, K. (1995). Development and application of a non-linear eddy viscosity model
sensitised to stress and strain invariants. UMIST, Dept. of Mechanical Engineering,
Report TFD/95/11.
48. Menter, F. (1992). Performance of popular turbulence models for attached and separated
adverse pressure gradient ows. AIAA J. 30(8), 20662072.
49. Ilyushin, B. B., and Kurbatskii, A. F. (1996). New models for calculation of third
order moments in planetary boundary layer. Izv. RAN. Phys. Atmosphere Ocean. 34, 772781.
50. Gatski, T. B., and Speziale, C. G. (1993). On explicit algebraic stress models for complex
turbulent ows. J. Fluid Mech. 254, 5978.
51. Craft, T. J., Ince, N. Z., and Launder, B. E. (1996). Recent developments in second-
moment closure for buoyancy-affected ows. Dynam. Atmos. Oceans 23, 99114.
52. Launder, B. E., Reece, G. J., and Rodi, W. (1975). Progress in the development of a
Reynolds-stress turbulence closure. J. Fluid Mech. 68, 537566.
53. Gibson, M. M., and Launder, B. E. (1978). Ground effects on pressure uctuations in the
atmospheric boundary layer. J. Fluid Mech. 86, 491511.
54. Launder, B. E. (1989). Second-moment closure: present . . . and future? Int. J. Heat Fluid
Flow 10, 282300.
55. Iacovides, H., and Launder, B. E. (1987). Parametric and numerical study of fully-
developed ow and heat transfer in rotation duct. ASME J. Turbomachinery 113, 331338.
56. Rodi, W. (1991). Experience with two-layer turbulence models combining the kc model
with a one-equation model near the wall. AIAA paper 91-0216.
57. Chen, H. C., and Patel, V. C. (1988). Near wall turbulence models for complex ows
including separation. AIAA J. 26, 641648.
58. Lam, C. K. G., and Bremhorst, K. (1981). A modied form of the kc model for predicting
wall turbulence. ASME J. Fluid Eng. 113, 456460.
59. Orszag, S. A., Staroselsky, I., and Yakhot, V. (1993). Some basic challenges for Large Eddy
Simulation research. In L arge Eddy Simulation of Complex Engineering and Geophysical
Flows (B. Galperin and S. A. Orszag, Eds.). Cambridge University Press, pp. 5578.
60. Rodi, W., Ferziger, J. H., Breuer, M., and Pourquie , M. (1997). Status of large-eddy
simulation: Results of a workshop. J. Fluids Eng. 119, 248262.
61. Moin, P. (1998). Numerical and physical issues in large-eddy simulation of turbulent
ows. JSME Int. J., Series B 41(2), 454463.
62. Schumann, U. (1975). Subgrid scale model for nite difference simulations of turbulent
ows in plane channels and annuli. J. Comput. Phys. 18, 376404.
63. Leonard, A. (1974). Energy cascade in large eddy simulations of turbulent ow. Adv.
Geophys. 18A, 237.
64. Smagorinsky, J. S. (1963). General circulation experiments with the primitive equations:
I The basic experiments. Mon. Weather Rev. 91, 99164.
65. Lilly, D. K. (1966). On the application of the eddy viscosity concept in the inertial
subrange of turbulence. NCAR Manuscript 123.
66. Yakhot, A. Orszag, S. A. Yakhot, V., and Israeli, M. (1989). Renormalization group
formulation of large-eddy simulation. J. Sci. Computing 4, 139158.
67. Germano, M., Piomelli, U., Moin, P., and Cabot, W. H. (1991). A dynamic subgrid-scale
eddy-viscosity model. Physics of Fluid A 3, 17601765.
68. Ferziger, J. H. (1997). Large eddy simulation. In Simulation and Modelling of Turbulent
Flows (T. B. Gatski, M. Y. Husaini, and J. L. Lumley, Eds.), ICASE/LaRC Series in
Computer Science in Engineering. Oxford University Press, pp. 109154.
69. Me tais, O., and Lesieur, M. (1991). Spectral large eddy simulation of isotropic and stable
stratied turbulence. J. Fluid Mech. 239, 15794.
70. Fallon, B. (1994). Simulation des Grande Echelles dEcoulements Turbulents Stratie s en
Densite . The` se de doctorat de lINPG, Grenoble, France.
71. Moin, P., and Mahesh, K. (1997). Direct numerical simulation: A tool in turbulence
research. CTR Manuscript 166, Stanford University, Stanford, CA.
72. Xi, G., Hagiwara, Y., and Suzuki, K. (1995). Flow instability and augmented heat transfer
of n arrays. J. Enhanced Heat Transfer 2, 2332.
438 . x. sn:n r1 :i.
73. Kouidry F. (1997). Etude des Ecoulements Turbulents Charge s de Particules: Application
a` Lencrassement Particulaire des Echangeurs a` Plaques Corrugue es. Ph. D. Thesis,
University Josepf Fouirier, Grenoble, France.
74. Orszag, P. A., and Patterson, G. S. (1972). Numerical simulation of three-dimensional
homogenous isotropic turbulence. Phys. Rev. L ett. 28(2), 7685.
75. Kim, J. Moin, P., and Moser, R. (1987). Turbulence statistics in fully developed channel
ow at low Reynolds number. J. Fluid Mech. 177, 133199.
76. Spalart, P. (1989). Theoretical and numerical study of a three-dimensional turbulent
boundary layer. J. Fluid. Mech. 205, 319359.
77. Sparrow, E. M., Baliga, B. R., and Patankar, S. V. (1977). Heat transfer and uid ow
analysis of interrupted-wall channels with application to heat exchangers. ASME J. Heat
Transfer 99, 411.
78. Patankar, S. V., and Prakash, C. (1981). An analysis of the effect of plate thickness on
laminar ow and heat transfer in interrupted-plate passages, Int. J. Heat Mass Transfer
24, 18011810.
79. Kelkar, K. M., and Patankar, S. V. (1989). Numerical prediction of heat transfer and uid
ow in rectangular offset-n arrays, Numerical Heat Transfer: Part A 15, 149164.
80. Wieting, A. R. (1975). Empirical correlations for heat transfer and ow friction character-
istics of rectangular offset-n plate-n heat exchangers. ASME J. Heat Transfer 97,
488490.
81. Suzuki, K., Hirai, E., Sato, T., and Kieda, S. (1982). Numerical study of heat transfer
system with staggered array of vertical at plates used at low Reynolds number. Proc. 7th
Int. Heat Transfer Conf., Heat Transfer 1982, 3, 483488. Hemisphere Publishing,
Washington, DC.
82. Suzuki, K., Hirai, E., Miyaki, T., and Sato, T., (1985). Numerical and experimental studies
on a two-dimensional model of an offset strip-n type compact heat exchanger used at
low Reynolds number. Int. J. Heat Mass Transfer 28, 823836.
83. Xi, G., Hagiwara, Y., and Suzuki, K. (1992). Effect of n thickness on ow and heat
transfer characteristics of n arrays an offset-n array in the low Reynolds number
range, Heat Transfer Japanese Res. 22, 119.
84. Mochizuki, S., Yagi, Y., and Yang, W-J. (1988). Flow pattern and turbulence intensity in
stacks of interrupted parallel-plate surfaces. Exp. T hermal Fluid Sci. 1, 5157.
85. Xi, G. N., Futagami, S., Hagiwara, Y., and Suzuki, K. (1991). Flow and heat transfer
characteristics of offset-n array in the middle Reynolds number range. Proc. 3rd ASME-
JSME Thermal Engineering Joint Conf. 3, 151156.
86. Suzuki, K., Xi, G. N., Inaoka, K., and Hagiwara, Y. (1994). Mechanism of heat transfer
enhancement due to self-sustained oscillation for an in-line n array. Int. J. Heat Mass
Transfer 37, Suppl. 1, 8396.
87. Michallon, E. (1993). Etude du comportement de le coulement dans des canaux de section
rectangulaire constitue s de plaques et ailettes brase es. Ph.D. thesis, Univerisite de Nancy
I, France, November.
88. Mizuno, M., Morioka, M., Hori, M., and Kudo, K. (1995). Heat transfer and ow
characteristics of offset strip ns in low-Reynolds number region. HTD-Vol. 3171,
ASME, New York, pp. 425432.
89. Ota, T., and Itasaka, M. (1976). A separated and reattached ow on a blunt at plate,
ASME J. Fluid Eng. 98, 7986,
90. Xi, G. N., and Shah R. K. (1999). Numerical analysis of offset strip n heat transfer and
ow friction characteristics. In Proc. Int. Conf. Computational Heat and Mass Transfer
(A. A. Mohamad and I. Sezai, Eds.). Eastern Mediterranean University Printinghouse,
Gazimagusa, Cyprus, pp. 7587.
91. London, A. L., and Shah, R. K. (1968). Offset rectangular plate-n surfaces heat transfer
and ow friction characteristics. ASME J. Eng. Power 90, 218228.
92. Manglik, R. M., and Bergles, A. E. (1995). Heat transfer and pressure drop correlation for
the rectangular offset strip n compact heat exchanger. Exp. T hermal Fluid Sci. 10,
171180.
93. R. K. Shah (1985). Compact heat exchangers. In Handbook of Heat Transfer Applications,
2nd ed. (W. M. Rohsenow, J. P. Hartnett, and E. N. Ganic , Eds.), Chapter 4, Part III, pp.
4174 to 4311. McGraw-Hill, New York.
94. Kajino, M., and Hiramatsu, M. (1987). Research and development of automotive heat
exchangers. In Heat Transfer in High Technology and Power Engineering (Y. J. Yang and
Y. Mori, Eds.). Hemisphere Publishing Corp., pp. 420432.
95. Tomoda, T., and Suzuki, K. (1988). A numerical study of heat transfer on compact heat
exchanger (effect on n shape). 25th National Heat Transfer Symp. of Japan, 175177.
96. Achaichia, A., and Cowell, T. A. (1988). A nite difference analysis of fully developed
periodic laminar ow in inclined louver arrays. 2nd UK National Heat Transfer Conf.,
Glasgow. IMechE, London, pp. 883897.
97. Davenport, C. J. (1980). Heat transfer and uid ow in louvered triangular ducts. Ph. D.
Thesis, CNAA, Lanchester Polytechnic, Coventry, UK.
98. Hiramatsu, M., Ishimaru, T., and Matsuzaki, K. (1990). Research on ns for air
conditioning heat exchangers (1st report, numerical analysis of heat transfer on louver
ns). JSME Int. J. Ser. II 33(4), 749756.
99. Webb, R. L., and Trauger P. (1991). Flow structure in the louver n heat exchanger
geometry. Exp. T hermal Fluid Sci. 4, 205217.
100. Ha, M. Y., Kim, K. C., Koak, S. H., Kim, K. H., Kim, K. I., Kang, J. K., and Park, T. Y.
(1995). Fluid ow and heat transfer characteristics in multi-louvered n heat exchanger.
SAE Paper No. 950115.
101. Achaichia, A., and Cowell, T. A. (1988). Heat transfer and pressure drop characteristics
of at tube and louver plate n surfaces. Exp. Thermal Fluid Sci. 1, 147157.
102. Webb, R. L. (1990). The ow structure in the louver n heat exchanger geometry. SAE
Paper No. 900722.
103. Baldwin, S. J., White, P. R. S., Al-Daini, A. J., and Davenport, C. J. (1987). Investigation
of the gas side ow eld in multilouver ducts with ow reversal. 5th Int. Conf. on
Numerical Methods in Laminar and Turbulent Flow, Montreal, 5, Pt. 1, 482495.
104. Button, B. L., Tura, R., and Wright C. C. (1984). Investigation of the air ow through
louver rectangular ducts using laser Doppler anemometry. Proc. 2nd Int. Symp. On
Applications of Laser Doppler Anemometry to Fluid Mechanics, Lisbon, pp. E2E5.
105. Hiramatsu, M., and Ota, K., (1982). Heat transfer analysis for heat exchanger ns. Paper
B303, 19th National Heat Transfer Symp., Japan.
106. Suga, K., Aoki, H., and Shinagawa, T. (1990). Numerical analysis on two-dimensional ow
and heat transfer of louver ns using overlaid grids. JSME Int. J. Series II 33(1), 122127.
107. Suga, K., and Aoki, H. (1991). Numerical study on heat transfer and pressure drop in
multilouvered ns, ASME/JSME T hermal Eng. Joint Conf. 4, 361368.
108. Ikuta, S., Sasaki, Y., Tanaka, K., Takagi, M., and Himeno, R. (1990). Numerical analysis
of heat transfer around louver assemblies. SAE Paper No. 900081.
109. Achaichia, A., Heikal, M. R., Sulaiman, Y. and Cowell, T. A., (1994). Numerical
investigation of ow and friction in louver n arrays. 10th Int. Heat Transfer Conf., Heat
Transfer 1994 4, 333338.
110. Itoh, S., Kuroda, M., Kato, Y., and Kobayashi, T. (1995). Numerical analysis and
visualisation experiment on ow of air through nned heat exchanger. Trans. JSME, Part
B, 61(582), 564571.
440 . x. sn:n r1 :i.
111. Atkinson, K. N., Drakulic , R., and Heikal, M. R. (1995). Numerical modeling of ow and
heat transfer over louvered plate n arrays in compact heat exchangers. 4th Star-CD
European User Group Meeting, London, 1314 November.
112. Drakulic , R., Atkinson, K. N., and Heikal, M. R. (1996). Numerical prediction of the
performance characteristics of louver n arrays in compact heat exchangers. Paper
presented in the open session, 2nd European Thermal Sciences and 24th UIT National Heat
Transfer Conference, Rome.
113. Drakulic , R., (1997). Numerical modeling of ow and heat transfer in louver n arrays.
Ph. D. Thesis, University of Brighton, Brighton, UK.
114. Atkinson, K. N., Drakulic , R., Heikal, M. R., and Cowell, T. A. (1998). Two- and three-
dimensional numerical models of ow and heat transfer over louvred n arrays in
compact heat exchangers, Int. J. Heat Mass Transfer 41, 40634080.
115. Antoniou A., Heikal, M. R., and Cowell T. A. (1990). Measurements of local velocity and
turbulence levels in arrays of louver plate ns. 9th Int. Heat Transfer Conf., Heat Transfer
1990 4, 105110.
116. Achaichia, A. (1987). The performance of louver tube-and-plate n heat transfer surfaces.
Ph. D. Thesis, CNAA, Brighton Polytechnic, Brighton, UK.
117. Antoniou, A. (1989). Measurements of local heat transfer, velocity and turbulence intensity
values in louver arrays. Ph. D. Thesis, CNAA, Brighton Polytechnic, Brighton, UK.
118. Beamer, H. E., Ghosh, D., Bellows, K. D., Huang, L. J. and Jacobi, A. M. (1998). Applied
CFD and experiment for automotive compact heat exchanger development. SAE Paper
No. 980426.
119. Achaichia, A., and Cowell, T. A. (1988). Heat transfer and pressure drop characteristics
of at tube louvered plate n surfaces. Exp. T hermal Fluid Sci. 1, 147157.
120. Tanaka, T., Itoh M., Kedoh, M. and Akira, T. (1984). Improvement of compact heat
exchangers with inclined louver ns, Bull. JSME 27(224), 219226.
121. Suzuki, K., Nishihara, A., Hiyashi, T., Schuerger, M. J., and Hayashi M. (1990). Heat
transfer characteristics of a two-dimensional model of a parallel louver n. Heat
Transfer Japanese Res. 19(7), 654669.
122. Sunde n, B., and Faghri M., Eds. (1998). Computer Simulations in Compact Heat Ex-
changers. Computational Mechanics Publications, Southampton, UK.
123. Asako, Y., and Faghri, M. (1987). Finite volume solutions for laminar ow and heat
transfer in a corrugated duct. ASME J. Heat Transfer 109, 627634.
124. Faghri, M., Sparrow, E. M., and Prata, A. T. (1984). Finite difference solutions of
convectiondiffusion problems in irregular domains using a non-orthogonal coordinate
transformation. Numerical Heat Transfer 7, 183209.
125. Asako, Y., Nakamura, H., and Faghri, M. (1988). Heat transfer and pressure drop
characteristics in a corrugated duct with rounded corners. Int. J. Heat Mass Transfer 31,
12371245.
126. Yang, L. C., Asako, Y., Yamaguchi, Y., and Faghri, M. (1995). Numerical prediction of
transitional characteristics of ow and heat transfer in a corrugated duct. Heat Transfer
in Turbulent Flows, HTD-Vol. 318, pp. 145152. ASME, New York.
127. Garg, V. K., and Maji, P. K. (1988). Laminar ow and heat transfer in a periodically
convergingdiverging channel. Int. J. Num. Methods Fluids 8, 579597.
128. Xin, R. C., and Tao, W. Q. (1988). Numerical prediction of laminar ow and heat
transfer in wavy channels of uniform cross-sectional area. Numerical Heat Transfer 14,
465481.
129. Be reiziat, D. (1993). Structure locale de le coulement de uides Newtonien et non-
Newtoniens en canaux ondule s: application le changeur a` plaques. Institut National
Polytechnique de Loraine, Nancy, France.
130. Ergin, S., Ota, M., Yamaguchi, H., and Sakamoto, M. (1996). A numerical study of the
effect of interwall spacing on turbulent ow in a corrugated channel, HTD-Vol. 333, 2,
131. Sparrow, E. M., and Comb J. W. (1983). Effect of interwall spacing and uid inlet
conditions on corrugated wall heat exchanger, Int. J. Heat Mass Transfer 26, 9931005.
132. Ergin, S., Ota, M., Yamaguchi, H., and Sakamoto, M. (1997). Analysis of periodically fully
developed turbulent ow in a corrugated duct using various turbulence models and
comparison with experiments. JSME Centennial Grand Congress, Int. Conf. on Fluid Eng.,
15271532, Tokyo, Japan.
133. Be reiziat, D., Devienne, R., and Lebouche , M. (1995). Local ow structure for non-
Newtonian uids in a periodically corrugated wall channel. J. Enhanced Heat Transfer 2,
7177.
134. Farhanieh, B. and Sunde n, B. (1992). Laminar heat transfer and uid ow in streamwise-
periodic corrugated square ducts for compact heat exchangers. In Compact Heat Ex-
changer for Power and Process Industries, HTD-Vol. 201, pp. 3749. ASME, New York.
135. Asako, Y., Faghri, M., and Sunde n, B., (1996). Laminar ow and heat transfer character-
istics of a wavy duct with a trapezoidal cross section for heat exchanger application. In
2nd European Thermal Conference (G. P. Celata, P. Di Marco, and A. Mariani, Eds.), Vol.
2, pp. 10971104. Edizioni ETS, Italy.
136. Shah, R. K., and Bhatti, M. S., (1987). Laminar convective heat transfer in ducts. In
Handbook of Single-Phase Heat Transfer (S. Kaka, R. K. Shah, and W. Aung, Eds.). John
Wiley, New York, Chapter 3.
137. McNab C. A., Atkinson K. N., Heikal M. R., and Taylor N. (1998). Numerical modelling
of heat transfer and uid ow over herringbone corrugated ns. Heat Transfer 1998, Proc.
11th Int. Heat Transfer Conf. 6, 119124.
138. Abou-Madi (1998). A computer model for mobile air conditioning system. Ph. D. Thesis,
University of Brighton, Brighton, UK.
139. Tochon, P., and Mercier, P. (1999). Thermalhydraulic investigations of turbulent ows
in compact heat exchangers. In Compact Heat Exchangers and Enhancement Technology
for the Process Industries (R. K. Shah, Ed.). Begell House Inc., New York, pp. 97101.
140. Atkinson, K. N., Drakulic , R., Heikal, M. R., and McNab, C. A. (1998). Applications of
numerical ow modelling to high performance heat transfer surfaces. Heat Transfer 1998,
Proc. 11th Int. Heat Transfer Conf. 5, 333338.
141. Sobey, I. J. (1980). On ow through furrowed channels: Part 1. Calculated ow patterns.
J. Fluid Mech. 125, 359373.
142. Stephanoff, K. D., Sobey, I. J., and Bellhouse, B. J. (1980). On ow through furrowed
channels. Part 2. Observed ow patterns. J. Fluid Mech. 125, 359373.
143. Sobey, I. J. (1982). Oscillatory ows at intermediate Strouhal number in asymmetric
channels. J. Fluid Mech. 125, 359373.
144. Sobey, I. J. (1983). The occurrence of separation in oscillatory ow. J. Fluid Mech. 134,
247257.
145. Sparrow, E. M., and Prata, A. T. (1983). Numerical solutions for laminar ow and heat
transfer in a periodically convergingdiverging tube, with experimental conrmation.
Num. Heat Transfer 6, 441461.
146. Blancher, S., Batina, J., Creff, R., and Andre, P., (1990). Analysis of convective heat transfer
by a spectral method for an expanding vortex in a wavy-walled channel. Proc. 9th Int.
Heat Transfer Conference, Heat Transfer 1990, 2, 393398.
147. Guzman, A. M., and Amon, C. H. (1994). Transition to chaos in convergingdiverging
channel ows: RuelleTakensNewhouse scenario. Phys. Fluids 6, 19942002.
442 . x. sn:n r1 :i.
148. Guzman, A. M., and Amon, C. H. (1996). Dynamical ow characterization of transitional
and chaotic regimes in converging-diverging channels. J. Fluid Mech. 321, 2557.
149. Wang, G., and Vanka, S. P. (1995). Convective heat transfer in periodic wavy passages.
150. Nishimura, T., Ohori, Y., and Kawamura, Y. (1984). Flow characteristics in a channel
with symmetric wavy wall for steady ow. J. Chem. Eng. Jap. 17, 466471.
151. Fodemsky, T. R. (1990). Computer simulation study of thermal hydraulic performance of
corrugated ducts. Proc. 9th Int. Heat Transfer Conf., Heat Transfer 1990, 3, 241246.
152. Ciofalo, M., Stasiek, J., and Collins, M. W. (1996). Investigation of ow and heat transfer
in corrugated passages II. Numerical results. Int. J. Heat Mass Transfer 39, 165192.
153. Hessami, M. A. (1997). Numerical study of heat transfer and pressure loss in cross-
corrugated plate heat exchangers. In Proc. 3rd ISHMT-ASME Heat Mass Trans. Conf.
(G. Biswas, S. Srinivasa Murthy, K. Muralidhar, and V. K. Dhir, Eds.). Tata-McGraw-
Hill, New Delhi, pp. 795802.
154. Focke, W. W., Zachariades, J., and Olivier, I. (1985). The effect of the corrugation
inclination angle on the thermohydraulic performance of plate heat exchanger. Int. J.
Heat Mass Transfer 28, 14691479.
155. Sawyers, D. R., Sen, M., and Chang, H-S. (1998). Heat transfer enhancement in three
dimensional corrugated channel ow. Int. J. Heat Mass Transfer 41, 35593573.
156. Blomerius, H., Ho sken, C., and Mitra, N. K. (1999). Numerical investigation of ow eld
and heat transfer in cross-corrugated ducts. ASME J. Heat Transfer 121, 314321.
157. Sunde n, B. (1999). Flow and heat transfer mechanisms in plate-frame heat exchangers. In
Heat Transfer Enhancement of Heat Exchangers (S. Kaka, Ed.), NATO ASI Series E
Applied Sciences 355, pp. 185206. Kluwer Academic Publishers, Netherlands.
158. Baggio, E., and Fornasieri, E. (1994). Air-side heat transfer and ow friction: theoretical
aspects. In Recent Developments in Finned Tube Heat Exchangers, (Ch. Marvillet, General
Editor). DTI, Energy Technology, Denmark, pp. 91159.
a
AUTHOR INDEX
Numerals in parentheses following the page numbers refer to reference numbers cited in the text.
A
Abbrecht, P. H., 262(22), 292(22), 310(22),
354(22)
Abdel-Khalik, S. I. (Chapter Author), 145,
147(12), 150(12; 34; 35), 151(12), 152(12),
153(12), 154(12), 155(35), 156(12; 34),
157(12), 158(12), 159(12), 160(12),
161(12), 162(12), 164(12), 167(34),
168(34), 176(35), 177(34; 35), 185(12; 35),
187(12), 188(34), 191(35), 193(132; 135;
139), 196(132), 197(132), 198(132; 135),
199(132), 200(132), 202(132; 135; 139),
203(135; 139), 204(135), 205(132; 135;
139), 210(174), 211(174), 214(174),
215(174), 216(174), 218(174), 219(174),
224(174), 226(208), 227(208), 230(208),
231(208), 233(208)
Abdelmessih, A. H., 198(133)
Abdollahian, D., 225(199), 235(223; 225)
Abdullah, Z., 194(125), 198(125), 200(125),
202(125)
Abou-Madi, 421(138)
Abraham, M. A., 151(36; 37)
Abriola, L. M., 15(49)
Achaichia, A., 370(2), 407(96; 109), 408(96;
101), 410(109), 412(116), 414(101)
Achdou, Y., 69(171)
Achenbach, E., 91(186), 92(186), 95(186)
Acrivos, A., 11(32)
Adler, P. M., 2(13)
Adzerikho, K. S., 57(142), 60(142)
Ahmad, S. Y., 221(194)
Ahmed, N. U., 124(214; 215; 216; 217),
125(214; 225)
Akers, W. W., 187(100)
Akhiezer, A. I., 38(99)
Akira, T., 415(120)
Alamgir, M. D., 229(210), 236(210)
Al-Daini, A. J., 407(103)
Alfredsson, P. H., 381(29)
Al-Hayes, R. A. M., 200(137)
Ali, M. I., 150(29; 30), 155(29; 30), 171(29; 30),
174(30), 176(30), 177(30), 185(29; 30),
189(29; 30), 190(30)
Ali, M. M., 372(5)
Allmaras, S., 383(33), 392(33)
Al-Nimr, M. A., 56(136)
Amon, C. H., 424(147), 425(148)
Amos, C. N., 225(199), 226(204), 227(204),
228(204), 229(204), 232(204), 236(204),
239(204), 240(204), 241(204)
Anderson, J. D., 376(17)
Anderson, T. B., 1(1)
Andre, P., 424(146)
Anisimov, S. I., 39(100; 101)
Anita, S., 125(224)
Antal, S. P., 208(152)
Antonia, R. A., 28(77)
Antoniou, A., 412(115; 117)
Aoki, H., 407(106; 107), 409(106; 107)
Arai, N., 303(85), 333(85), 334(85), 335(85),
336(85), 346(117), 349(85), 356(85)
Ardron, K. H., 232(212)
Arkhipov, V. V., 210(165), 211(165)
Armstrong, R. C., 91(191), 94(191), 95(191)
Arpaci, V. S., 56(136)
Asako, Y., 388(126), 416(123), 417(123; 126),
418(123; 125; 126), 419(123; 125; 126),
420(126), 421(135), 431(126)
Asano, Y., 379(27)
Ashikhmin, S. R., 81(176), 82(176)
Atkinson, K. N., 396(137), 407(111; 112),
411(111; 112; 114), 413(114), 417(140),
421(137; 140)
Avellaneda, M., 69(171)
Aziz, K., 148(13), 158(41), 161(41), 175(41),
177(41)
445
B
Baggio, E., 433(158)
Balakrishnan, A. V., 124(221), 125(221)
Baldwin, B. S., 382(32)
Baldwin, S. J., 407(103)
Baliga, B. R., 398(77)
Barajas, A. M., 150(24), 153(24), 154(24),
158(24), 159(24), 166(24)
Barakat, R., 33(82)
Barenblatt, G. I., 262(21)
Barnea, D., 148(19), 150(23), 154(23), 156(23),
158(23), 161(23), 163(23)
Batchelor, G. K., 274(41)
Batina, J., 424(146)
Beamer, H. E., 413(118), 414(118)
Beavers, G. S., 81(177), 82(177), 83(177),
84(177)
Behringer, R. P., 26(71)
Bejan, A., 112(203)
Bellhouse, B. J., 422(142)
Bellows, K. D., 413(118), 414(118)
Beran, M. J., 98(196)
Be reiziat, D., 373(13), 417(133), 419(129)
Bergles, A. E., 114(209), 193(131), 194(126),
195(129), 196(129; 131), 197(129),
198(129), 206(131), 207(131), 208(131),
210(131; 164; 173), 211(131; 164; 173),
212(173; 179), 213(131; 173), 215(131;
173), 216(173), 219(173)
Bezprozvannykh, V. A., 26(74)
Bhatti, M. S., 421(136)
Bibeau, E. L., 192(118), 200(118)
Bilicki, Z., 236(229)
Bird, R. B., 77(173), 78(173), 79(173), 269(35),
280(35), 351(35)
Bisset, D. K., 28(77)
Biswas, A. K., 385(38)
Black, H. S., 192(112)
Blain, C. A., 2(8), 5(8), 23(8), 60(8)
Blancher, S., 424(146)
Blasick, A. M., 193(139), 202(139), 203(139),
205(139)
Blasius, H., 279(53)
Blinkov, V. N., 236(228)
Blomerius, H., 396(156), 427(156), 428(156),
429(156), 430(156), 432(156)
Boelter, L. M. K., 338(103)
Bohren, C. F., 57(144), 60(144)
Bolle, L., 236(229)
Bolstad, M. M., 184(89)
Bories, S., 150(33), 155(33), 171(33), 176(33),
185(33), 189(33), 190(33)
Bornea, D., 148(17), 156(17), 166(17), 171(17),
173(17)
Botten, L. C., 52(126), 57(155; 156; 157)
Bouassinesq, J., 269(33)
Boure , J. A., 194(126), 232(213)
Bowers, M. B., 185(95), 186(95), 210(95),
211(95), 214(95), 215(95), 219(95),
224(95)
Bowring, R. W., 218(178)
Boyd, R. D., 209(156; 157), 210(156; 157; 167;
168; 178), 211(156; 167), 212(167; 168),
215(156; 157), 216(157)
Brauner, N., 147(10), 161(10)
Breaux, D. K., 310(90), 354(90)
Bremhorst, K., 391(58), 418(58)
Brereton, G. J., 28(76)
Bretherton, F. B., 146(8), 150(8)
Breuer, M., 393(60)
Burkhart, T. H., 275(49)
Burns, J. A., 115(211; 212)
Butkovski, A. G., 124(220)
Butterworth, D., 112(204), 115(204)
Button, B. L., 407(104)
Buyevich, Y. A., 59(165)
C
Caceres, M. O., 37(92)
Caira, M., 218(184)
Calata, G. P., 213(181), 215(181)
Camarero, R., 22(51)
Carbonell, R. G., 7(31), 8(31), 9(31), 15(31; 40;
46), 18(46), 34(40; 46; 89), 35(90),
107(40), 126(40)
Carey, V. P., 191(107), 192(107), 205(107)
Caruso, G., 218(184)
Catton, I. (Chapter Author), 1, 2(16; 17; 18;
19; 20; 21; 22; 26; 27; 28), 3(19; 21),
10(21), 11(16; 18; 20; 26), 15(18), 16(16;
18; 21), 21(21), 23(24), 25(16; 20), 26(16;
17; 18; 19; 20; 21), 30(16; 21), 31(19;
26), 36(16; 17; 20; 26; 27), 51(114), 52(23),
57(19; 20; 28; 114; 158; 159; 161; 163),
60(21; 114), 61(114), 62(21), 65(21; 28),
66(166), 68(16; 20), 69(16; 20; 21; 23; 25;
446 :i1no iNnrx
26), 70(16; 17; 20; 21), 71(16; 20), 79(21),
80(21), 81(23), 96(21), 97(21), 102(21),
110(21), 111(114), 116(16; 20; 21; 23; 28),
118(16; 19), 119(16; 20), 123(16; 17;
23), 124(19)
Celata, G. P., 210(172; 175), 211(172; 175),
212(180), 213(172; 175), 217(172; 175),
221(193), 223(193)
Cerro, R. L., 151(36; 37)
Chan, C., 111(201), 287(71), 295(71; 77; 78),
298(71), 300(71), 302(71; 77), 326(100),
348(77), 351(71; 77), 355(100)
Chandrasekhar, S., 258(3)
Chang, H.-S., 428(155)
Chang, Y. J., 371(3)
Chao, J., 276(51)
Chen, G., 46(110; 111)
Chen, H. C., 391(57), 399(57)
Chen, Z.-H., 184(92), 185(92)
Chen, Z.-Y., 184(92), 185(92)
Cheng, H., 96(193), 97(193)
Cheng, P., 32(79; 80), 145(4)
Chexal, B., 235(223; 225)
Chhabra, R. P., 78(174), 79(174)
Chiffelle, R. J., 37(93), 39(93)
Choi, B., 292(75), 293(75)
Choi, H. Y., 309(88)
Christ, C. L., 240(233)
Churchill, S. W. (Chapter Author), 111(200;
201), 184(93), 255, 262(22), 264(26;
27), 281(58), 285(64; 65), 287(71),
288(58), 292(22; 75), 293(75), 295(71; 77;
78; 79), 296(80), 298(71; 83), 299(80),
300(71), 302(58; 71; 77), 303(58; 83; 85),
304(86), 310(22), 314(91), 326(100; 100a),
327(100a), 329(100a), 330(100a),
331(100a), 332(100a), 333(85; 100a),
334(85), 335(85), 336(85), 340(100a; 112),
342(112), 346(117), 348(77), 349(85;
100a), 350(100a), 351(71; 77; 80), 354(22),
355(100; 100a), 356(85)
Ciofalo, M., 384(152), 388(152), 393(152),
423(14), 425(152), 426(14; 152), 428(152),
429(152), 430(152), 432(152)
Colburn, A. P., 338(106)
Cole, R., 191(108), 205(108)
Colebrook, C. F., 274(63), 286(63)
Collier, J. G., 148(14), 182(14), 191(14)
Collier, R. P., 202(202), 225(202; 203),
226(203), 230(202), 231(202), 235(203;
202), 236(203)
Collins, M. W., 384(152), 388(152), 393(152),
423(14), 425(152), 426(14; 152), 428(152),
429(152), 430(152), 432(152)
Comb, J. W., 420(131)
Coppin, P. A., 25(64; 65; 66)
Cornish, R. J., 189(101)
Cornwell, J. D., 185(94), 186(94)
Corson, D. R., 57(145), 60(145)
Coulson, J. M., 338(105), 339(105)
Cowell, T. A., 370(2), 371(4), 375(15), 407(96;
109), 408(96; 101), 410(109), 411(114),
412(115), 413(114), 414(101)
Cox, S. G., 371(4)
Cox, S. J., 52(120)
Craft, T. J., 388(51)
Crapiste, G. H., 15(41), 22(41), 30(41), 35(41),
64(41)
Crawford, M. E., 387(44)
Creff, R., 424(146)
Crosser, O. K., 187(100)
Cumo, M., 210(172; 175), 211(172; 175),
212(180), 213(172; 175), 217(172; 175),
221(193), 223(193)
Curtain, R. F., 125(222)
D
Dagan, R., 236(227)
Daleas, R. S., 210(164), 211(164)
Damianides, C. A., 147(11), 150(11), 153(11),
154(11), 157(11), 158(11), 159(11),
161(11), 162(11), 164(11)
Danckwerts, P. V., 343(116)
Danov, S. N., 303(85), 333(85), 334(85),
335(85), 336(85), 349(85), 356(85)
Da Prato, G., 124(219)
Davenport, C. J., 407(103), 408(97)
Dawes, D. H., 57(156)
Deans, H. A., 187(100)
Deev, V. I., 210(165), 211(165)
Devienne, R., 373(13), 417(133)
Dhir, V. K., 145(3)
Dittus, R. W., 338(103)
Dix, G. E., 192(117), 199(117)
Dobson, D. C., 52(120)
Dombrovsky, L. A., 56(140), 60(140)
Dowling, M. F., 193(132; 139), 196(132),
447 :i1no iNnrx
197(132), 198(132), 199(132), 200(132),
202(132; 139), 203(139), 205(132; 139)
Downar-Zapolski, Z., 236(229)
Downie, J. H., 371(4)
Drakulic, R., 367(113), 375(15), 407(111; 112;
113), 411(111; 112; 113; 114), 412(113),
413(114), 417(140), 421(140)
Drew, D. A., 22(55), 23(56), 206(148; 149)
Drolen, B. L., 56(135)
Dukler, A. E., 148(15; 16; 17; 18), 156(15; 17),
161(15), 162(15), 166(17), 169(16),
171(17), 173(17)
Dullien, F. A. L., 2(12), 81(180), 85(180)
Duncan, A. B., 191(111)
Dybbs, A., 26(68), 89(184), 91(184), 92(184),
95(184)
E
Eckelmann, H., 287(70)
Edwards, R. V., 26(68)
Einav, S., 386(41)
Einstein, A., 258(4), 325(4)
Ekberg, N. P., 150(35), 155(35), 176(35),
177(35), 185(35), 191(35)
Elias, E., 225(200), 231(200), 232(200),
233(200), 236(227)
Elperin, T., 325(97a)
El-Sayed, M. S., 81(180), 85(180)
Elsayed-Ali, H. E., 39(106), 46(106)
Ergin, S., 67(167), 77(167), 81(167), 384(130),
388(132), 417(130; 132), 419(130),
420(130)
F
Faghri, M., 388(126), 416(122; 123; 124),
417(123; 126), 418(123; 125; 126),
419(123; 125; 126), 420(126), 421(124;
135), 431(126)
Fairbrother, F., 146(5), 150(5)
Fallon, B., 393(70)
Fan, L. T., 343(114)
Fand, R. M., 26(67)
Farhanieh, B., 421(134)
Farone, W. A., 57(153)
Fattorini, H. O., 125(223)
Feburie, V., 232(216), 237(216)
Fedoseev, V. N., 91(190), 94(190), 95(190)
Ferziger, J. H., 393(60), 395(68)
Figotin, A., 43(124; 125), 45(124), 52(122;
124), 54(123; 125), 55(123), 57(122; 124;
125)
Flaherty, J. E., 208(152)
Fleming, W. H., 124(218)
Flik, M. I., 46(112)
Focke, W. W., 373(7; 8; 10), 374(10), 426(7;
10), 428(154), 430(10)
Fodemsky, T. R., 384(151), 425(151), 428(151)
Fornasieri, E., 433(158)
Fourar, M., 150(33), 155(33), 171(33), 176(33),
185(33), 189(33), 190(33)
Fourier, J. B., 264(23)
Fox, R. F., 33(82)
France, D. M., 150(28), 155(28), 171(28),
175(28), 192(113; 114; 115)
Franco, J., 236(229)
Freeman, J. R., 280(54), 281(54)
Friedel, L., 191(105), 226(105), 227(105),
229(105), 230(105), 236(105), 239(105),
240(105)
Fritz, A., 232(211)
Fujii, M., 226(206), 227(206)
Fujimoto, J. G., 39(105)
Fukagawa, M., 114(210)
Fukano, T., 150(25), 154(25), 157(25), 159(25),
161(25), 162(25), 165(25), 169(25),
185(25), 186(25), 187(25)
Fushinobu, K., 37(91)
Futagami, S., 400(85), 405(85)
G
Gaiser, G., 373(11), 426(11), 428(11)
Galitseysky, B. M., 91(189), 93(189)
Galloway, J. E., 221(190)
Garg, C. K., 418(127), 423(127)
Garrels, R. M., 240(233)
Gatski, T. B., 386(40), 388(50)
Gelhar, L. W., 33(83)
Geng, H., 232(217; 218), 236(217), 239(218),
241(218)
Georgiadis, J. G., 26(71)
Germore, 395(67)
448 :i1no iNnrx
Ghiaasiaan, S. M. (Chapter Author), 145,
147(12), 150(12; 34; 35), 151(12), 152(12),
153(12), 154(12), 155(35), 156(12; 34),
157(12), 158(12), 159(12), 160(12),
161(12), 162(12), 164(12), 167(34),
168(34), 176(35), 177(34; 35), 183(86),
185(12; 35), 187(12), 188(34), 191(35),
193(132; 135; 139), 196(132), 197(132),
198(132; 135), 199(132), 200(132),
202(132; 135; 139), 203(135; 139),
204(135), 205(132; 135; 139), 210(174),
211(174), 214(174), 215(174), 216(174),
218(174), 219(174), 224(174), 226(208),
227(208), 230(208), 231(208), 232(217;
218), 233(208), 236(217), 239(218),
241(218)
Ghosh, D., 413(118), 414(118)
Gibson, M. M., 388(53)
Ginzburg, V. L., 38(98)
Giot, M., 232(211; 216), 237(216)
Gladkov, S. O., 40(107), 43(107), 45(107)
Gmitter, T. J., 52(117)
Godin, Yu. A., 43(124), 45(124), 52(124),
57(124)
Goldenfeld, N., 262(21)
Goodson, K. E., 46(112; 113)
Gortyshov, Yu. F., 81(175; 176), 82(176),
91(175), 94(175)
Gose, G. C., 231(219), 233(219)
Gotoh, N., 226(207), 227(207)
Govan, A. H., 224(196)
Govier, F. W., 148(13)
Graham, R. W., 191(106), 205(106), 209(106),
225(106), 231(106), 232(106)
Granger, S., 232(216), 237(216)
Gratton, L., 2(17; 19; 26; 27), 3(19), 11(26),
26(17; 19), 31(19; 26), 36(17; 26; 27),
57(19), 69(26), 70(17), 118(19), 123(17),
124(19)
Gray, W. G., 2(8), 5(8), 15(47; 48; 49; 50),
23(8), 60(8)
Gregory, G. A., 158(41), 161(41), 175(41),
177(41)
Gridnev, S. A., 57(163)
Grifth, P., 147(9), 150(9), 153(9), 154(9),
160(9), 161(9), 165(9)
Groeneveld, D. C., 209(158), 216(158)
Groenhof, H., 287(72)
Grolmes, M. A., 231(220), 233(220)
Grzesik, J., 56(139), 57(139)
Gschwind, P., 372(6)
Gutjahr, A. L., 33(83)
Guzman, A. M., 424(147), 425(148)
H
Ha, M. Y., 407(100), 408(100), 411(100)
Hadley, G. R., 98(198)
Hagiwara, Y., 396(72), 399(72; 86; 83), 400(72;
83; 85; 86), 405(72; 85)
Halbaeck, M., 381(29)
Hall, D. D., 217(183)
Hanjalic, K., 274(43)
Hanratty, T. J., 287(68), 325(98), 326(99),
355(98)
Haramura, Y., 222(195)
Hardy, P., 238(232)
Harimizu, Y., 226(206), 227(206)
Hassanizadeh, S. M., 15(50)
Hayashi, M., 415(121)
Healzer, J., 225(199)
Heikal, M. R. (Chapter Author), 363, 370(2),
371(4), 375(15), 396(137), 407(109; 111;
112), 410(109), 411(111; 112; 114),
412(115), 413(114), 417(140), 421(137;
140)
Heisenberg, W., 258(5)
Hendricks, T. J., 56(132)
Heng, L., 326(100), 355(100)
Henningson, D. S., 381(29)
Henry, R. E., 235(224)
Hessami, M. A., 384(153), 388(153), 425(153),
426(153), 427(153), 428(153), 432(153)
Hewiit, G. F., 224(196)
Hibiki, T., 150(26; 31), 153(26), 154(26),
155(31), 157(26), 158(26), 160(26),
163(26), 169(26; 31), 171(31), 172(31),
174(31), 180(26), 185(26; 31), 187(26),
189(31), 190(31)
Higbie, R., 343(115)
Hijikata, K., 37(91)
Hilfer, R., 53(128)
Himeno, R., 407(108), 409(108)
Hino, R., 198(134)
Hinze, J. O., 277(52)
Hirai, E., 398(81; 82), 399(81), 400(82),
407(82)
449 :i1no iNnrx
Hiramatsu, M., 406(94), 407(94; 98), 408(98),
409(105), 410(98)
Hirata, M., 193(155), 209(155), 210(155),
211(155), 212(155), 215(155)
Hiyashi, T., 415(121)
Hopkins, N. E., 184(90)
Hori, M., 399(88), 401(88), 402(88), 403(88),
407(88)
Horimizu, Y., 226(207), 227(207)
Hosaka, S., 193(155), 209(155), 210(155),
211(155), 212(155), 215(155)
Ho sken, C., 396(156), 427(156), 428(156),
429(156), 430(156), 432(156)
Howell, J. R., 56(131; 132), 60(131)
Howle, L., 26(71)
Hsu, C. T., 32(79; 80)
Hsu, Y. Y., 191(106), 205(106), 209(106),
225(106), 231(106), 232(106)
Hu, H. Y., 209(154)
Hu, K., 2(22), 69(25)
Huang, L. J., 413(118), 414(118)
Huffman, D. R., 57(144), 60(144)
Hugonnot, P., 373(9), 384(9), 417(9), 418(9),
419(9), 420(9)
Hwang, C. B., 387(43)
I
Iacovidea, H., 391(55)
Ichikawa, A., 124(219)
Idelchik, I. E., 233(221)
Ikuta, S., 407(108), 409(108)
Ileslamlou, S., 221(188)
Ilyushin, B. B., 388(49)
Imas, Ya. A., 39(100)
Inaoka, K., 399(86), 400(86)
Inasaka, F., 193(130), 196(130), 197(130),
198(130), 199(130), 200(130), 201(130),
202(130), 203(130), 210(169; 170; 171),
211(169; 170; 171), 212(169; 170; 180),
213(169; 170), 217(169; 170)
Ince, N. Z., 388(51)
Ippen, E. P., 39(105)
Ishii, M., 22(52; 53), 150(32), 155(32), 158(42),
162(42), 163(42), 164(42), 166(42),
171(32), 172(32), 173(32; 42), 174(32),
205(142)
Ishimaru, T., 407(98), 408(98), 410(98)
Ishiyama, T., 226(206), 227(206)
Israeli, M., 395(66)
Itasaka, M., 402(89)
Ito, K., 115(211)
Itoh, M., 415(120)
Itoh, S., 407(110), 411(110)
Iwabuchi, M., 114(210)
J
Jackson, R., 1(1)
Jacobi, A. M., 368(1), 369(1), 372(1), 374(1),
400(1), 413(118), 414(118)
Janssen, E., 225(199)
Jendrzejczyk, J. A., 150(28), 155(28), 171(28),
175(28), 192(113; 114)
Jensen, M. K., 193(131), 196(131), 206(131),
207(131), 208(131), 210(131; 173),
211(131; 173), 212(173), 213(131; 173),
215(131; 173), 216(173), 219(173)
Jensen, P. J., 231(219), 233(219)
Jeter, S. M., 150(35), 155(35), 176(35), 177(35),
185(35), 191(35), 193(132; 135; 139),
196(132), 197(132), 198(132; 135),
199(132), 200(132), 202(132; 135; 139),
203(135; 139), 204(135), 205(132; 135;
139), 210(174), 211(174), 214(174),
215(174), 216(174), 218(174), 219(174),
224(174)
Jischa, M., 324(96), 345(96), 354(96)
Johansson, A. V., 381(29)
John, H., 191(105), 226(105), 227(105),
229(105), 230(105), 236(105), 239(105),
240(105)
John, S., 52(118; 119)
Jones, O. C., 205(145)
Jones, O. C., Jr., 189(102), 190(102), 206(150),
207(150), 236(228)
Jordan, R. C., 184(89)
Joseph, D. D., 41(108)
K
Kaganov, M. I., 38(97), 39(97)
Kajino, M., 406(94), 407(94)
Kampe de Fe riet, J., 297(81), 350(81)
450 :i1no iNnrx
Kang, J. K., 407(100), 408(100), 411(100)
Kang, S., 115(211; 212)
Kanzaka, M., 114(210)
Kapeliovich, B. L., 39(101)
Kapitsa, P. L., 258(6)
Kar, K. K., 89(184), 91(184), 92(184), 95(184)
Kariyasaki, A., 150(25), 154(25), 157(25),
159(25), 161(25), 162(25), 165(25),
169(25), 185(25), 186(25), 187(25)
Kasagi, N., 193(155), 209(155), 210(155),
211(155), 212(155), 215(155)
Kashcheev, V. M., 23(59)
Kastner, W., 226(205), 227(205), 229(205)
Kato, M., 386(39)
Kato, Y., 407(110), 411(110)
Katto, Y., 209(159), 210(166), 211(166),
216(159), 220(159), 221(191; 192),
222(191; 195), 223(191; 192), 224(159)
Kaufman, S. J., 340(108), 341(108)
Kaviany, M., 2(7), 56(134), 60(7)
Kawaji, M., 150(27; 29; 30), 155(27; 29; 30),
171(27; 29; 30), 174(30), 176(30), 177(30),
185(27; 29; 30), 189(27; 29; 30), 190(30)
Kawamura, H., 274(44)
Kawamura, Y., 425(150)
Kays, W. M., 77(172), 81(172), 91(172),
95(172), 111(172), 324(95a), 327(101),
328(101), 330(101), 333(101), 335(101),
387(44)
Kazantseva, N. E., 57(161), 60(161), 66(161)
Kedoh, M., 415(120)
Kefer, V., 226(205), 227(205), 229(205)
Kelkar, K. M., 398(79), 399(79)
Kells, L. C., 286(66)
Kennedy, J. E., 193(132), 196(132), 197(132),
198(132), 199(132), 200(132), 202(132),
205(132)
Khan, E. U., 23(58)
Kharitonov, V. V., 91(190), 94(190), 95(190)
Kheifets, L. I., 2(11), 25(11), 32(11)
Khodyko, Yu. V., 39(100)
Khoroshun, L. P., 97(194; 195)
Kichigan, A. M., 210(161), 211(161)
Kieda, S., 378(22), 398(81), 399(81)
Kim, B. Y. K., 26(67)
Kim, I. C., 34(87)
Kim, J., 287(67), 396(75)
Kim, K. C., 407(100), 408(100), 411(100)
Kim, K. H., 407(100), 408(100), 411(100)
Kim, K. I., 407(100), 408(100), 411(100)
Kim, P. H., 205(143)
Kim, S. J., 67(168)
Klausner, J. F., 205(146)
Kleeorin, N., 325(97a)
Knibbe, P. G., 373(7; 10), 374(10), 426(7; 10),
430(10)
Koak, S. H., 407(100), 408(100), 411(100)
Kobayashi, T., 407(110), 411(110)
Kocamustafaogullari, G., 205(142)
Kodal, A., 28(76)
Koizumi, H., 184(91), 185(91)
Kokorev, V. I., 91(190), 94(190), 95(190)
Kolar, R. L., 2(8), 5(8), 23(8), 60(8)
Kolmogorov, R. R., 273(38)
Kottke, E. V., 373(11), 426(11), 428(11)
Kottke, V., 372(6)
Kouidry, F., 396(73), 417(73), 420(73),
421(73), 422(73), 432(73)
Kra tzer, W., 226(205), 227(205), 229(205)
Kroeger, P. G., 237(230)
Kuchment, P., 43(125), 52(122), 54(123; 125),
55(123), 57(122; 125)
Kudinov, V. A., 98(197)
Kudo, K., 399(88), 401(88), 402(88), 403(88),
407(88)
Kumar, S., 56(133; 137)
Kunevich, A. P., 81(176), 82(176)
Kurbatskii, A. F., 388(49)
Kuroda, M., 407(110), 411(110)
Kurshin, A. P., 84(179)
Kushch, V. I., 11(33; 34), 12(35; 36; 37; 38),
13(34), 102(33; 34)
Kuwahara, F., 108(199), 109(199), 111(199)
Kwok, C. C. K., 184(92), 185(92)
L
Lackme, C., 237(231)
Lahey, R. T., Jr., 22(55), 23(56), 206(148; 149;
150), 207(150), 208(152)
Lahey, T. R., Jr., 22(54), 192(116)
Lai, J., 42(109)
Lakhtakia, A., 57(147), 60(147)
Lam, A. C. C., 26(67)
Lam, C. K. G., 391(58), 418(58)
Lamb, H., 257(1)
Landau, L. D., 258(7)
Lane, J. C., 379(26)
451 :i1no iNnrx
Launder, B. E., 274(42; 43), 352(42), 383(34),
384(34; 35), 386(34; 39), 387(45), 388(51;
52; 53; 54), 390(35), 391(55)
Lazarek, G. M., 192(112)
Lebouche , M., 373(13), 417(133)
Lee, C. H., 221(189), 223(189)
Lee, N., 148(16), 169(16)
Lee, P. C. Y., 15(47; 48)
Lee, R. C., 205(141)
Lee, S. C., 56(138; 139), 57(139)
Lee, S. J., 22(55), 206(150), 207(150)
Lee, S. Y., 225(201), 232(201; 214; 215),
236(201; 214), 237(214; 215)
Legg, B. J., 25(64; 65; 66)
Lehner, F. K., 33(81)
Leijsne, A., 2(8), 5(8), 23(8), 60(8)
Lelluche, G. S., 225(200), 231(200), 232(200),
233(200)
Leonard, A., 393(63)
Leonard, B. P., 379(24)
Lesieur, M., 393(69)
Leung, J. C., 231(220), 233(220)
Leung, R. Y., 327(101), 328(101), 330(101),
333(101), 335(101)
Levec, J., 15(46), 18(46), 34(46)
Levy, S., 194(123), 198(123), 200(123),
201(123), 202(123), 203(123), 206(123)
Li, J.-H., 200(136), 202(136)
Li, R.-Y., 184(92), 185(92)
Lienhard, J. H., 229(210), 236(210)
Lifshitz, I. M., 38(97), 39(97)
Lightfoot, E. N., 77(173), 78(173), 79(173),
269(35), 280(35), 351(35)
Lilly, D. K., 394(65)
Lin, C. A., 387(43)
Lin, L., 205(147)
Lin, S., 184(92), 185(92)
Lin, T. -F., 185(96; 97), 187(96; 97), 188(97)
Lindell, I. V., 57(146), 60(146)
Lindgren, E. R., 276(51)
Liou, W. W., 386(42)
Liu, J., 39(105)
Loehrke, R. J., 379(26)
Lomax, H., 382(32)
London, A. L., 77(172), 81(172), 91(172),
95(172), 111(172), 403(91), 404(91)
Loomsmore, C. S., 210(163), 211(163)
Lopez de Bertodano, M., 22(54; 55)
Lorentz, H. A., 258(8)
Lorrain, P., 57(145), 60(145)
Lowry, B., 150(27), 155(27), 171(27), 185(27),
189(27)
Lu, B., 34(85; 88)
Lubarsky, B., 340(108), 341(108)
Lulinkski, Y., 150(23), 154(23), 156(23),
158(23), 161(23), 163(23)
Lumley, J. L., 26(73)
Luo, K., 42(109)
Lyn, D. A., 386(41)
Lynn, S., 275(50), 277(50)
Lyon, R. N., 322(94), 340(94)
Lyons, S. L., 287(68)
M
Macdonald, I. F., 81(180), 85(180)
MacLeod, A. L., 282(61), 283(61)
Mahesh, K., 393(71)
Maji, P.K., 418(127), 423(127)
Majumdar, A., 37(91; 94), 41(94), 42(109),
56(133)
Malbagi, F., 57(152), 59(152), 61(152), 63(152)
Mali, P., 238(232)
Mandane, J. M., 158(41), 161(41), 175(41),
177(41)
Marchessault, R. N., 146(6), 150(6)
Marcy, G. P., 184(88)
Mariani, A., 210(172; 175), 211(172; 175),
212(180), 213(172; 175), 217(172; 175),
221(193), 223(193)
Marle, C. M., 1(3)
Marsh, W. J., 195(128), 198(128)
Martin, H., 112(205)
Martin, J. D., 275(49)
Martinelli, R. C., 201(138), 340(107)
Marvillet, C., 373(12), 428(12)
Mason, S. G., 146(6), 150(6)
Mastin, C. W., 376(16)
Masuoka, T., 26(69)
Matsumoto, K., 226(206; 207), 227(206; 207)
Matsuo, T., 114(210)
Matsuzaki, K., 407(98), 408(98), 410(98)
Mayeld, M. E., 225(203), 226(203), 235(203),
236(203)
McBeth, R. V., 210(176; 177)
McClure, J. A., 231(219), 233(219)
McFadden, J. H., 231(219), 233(219)
452 :i1no iNnrx
McLaughlin, J. B., 287(68)
McLeond, D., 194(125), 198(125), 200(125),
202(125)
McNab, C. A., 396(137), 417(140), 421(137;
140)
McPhedran, R. C., 52(126), 57(155; 156; 157)
Mei, R., 205(146)
Melton, J. E., 376(18)
Menter, F., 388(48), 392(48)
Mercier, P., 377(20), 396(20; 139), 399(20),
402(20), 405(20), 422(139), 423(139),
424(139), 429(139)
Me tais, O., 393(69)
Michallon, E., 384(87), 399(87), 401(87),
402(87), 407(87)
Mikol, E. P., 184(87)
Miller, B., 275(48)
Miller, C. A., 34(86)
Millikan, C. B., 266(30)
Mishima, K., 22(53), 150(26; 31), 153(26),
154(26), 155(31), 157(26), 158(26; 42),
160(26), 162(42), 163(26; 42), 164(42),
166(42), 169(26; 31), 171(31), 172(31),
173(42), 174(31), 180(26), 185(26; 31),
187(26), 189(31), 190(31)
Mitra, N. K., 396(156), 427(156), 428(156),
429(156), 430(156), 432(156)
Mizuno, M., 399(88), 401(88), 402(88),
403(88), 407(88)
Moalem-Maron, D., 147(10), 161(10)
Mochizuki, S., 400(84), 403(84), 404(84),
405(84)
Moin, P., 287(67), 379(25), 393(61; 71),
396(75)
Moizhes, B. Ya., 98(197)
Monrad, C. C., 282(59)
Moody, F. J., 192(116), 234(222), 235(222)
Morega, A. M., 112(203)
Morioka, M., 399(88), 401(88), 402(88),
403(88), 407(88)
Moser, R., 287(67), 396(75)
Moshaev, A. P., 91(189), 93(189)
Motai, T., 114(210)
Movchan, A. B., 57(155)
Mow, K., 81(180), 85(180)
Moyne, C., 81(181), 85(181)
Mudawar, I., 185(95), 186(95), 195(128),
198(128), 210(95), 211(95), 214(95),
215(95), 217(183), 219(95), 221(189; 190),
223(189), 224(95)
Muller, J. R., 226(208), 227(208), 230(208),
231(208), 233(208)
Muralidhar, K., 385(38)
Muravev, G. B., 81(175), 91(175), 94(175)
Murphree, W. V., 268(32)
N
Nabarayashi, T., 226(206; 207), 227(206; 207)
Nadyrov, I. N., 81(175; 176), 82(176), 91(175),
94(175)
Naff, R. L., 33(83)
Nakamura, H., 418(125), 419(125)
Nakamura, S., 226(207), 227(207)
Nakayama, A., 108(199), 109(199), 111(199)
Nariai, H., 193(130), 196(130), 197(130),
198(130), 199(130), 200(130), 201(130),
202(130), 203(130), 210(169; 170; 171),
211(169; 170; 171), 212(169; 170; 180),
213(169; 170), 217(169; 170)
Narrow, T. L., 150(34), 156(34), 167(34),
168(34), 177(34), 188(34)
Navier, C.-L. M. N., 261(19)
Naviglio, A., 218(184)
Neimark, A. V., 2(11), 25(11), 32(11)
Newton, I., 258(9)
Nicorovici, N. A., 52(126), 57(155; 156; 157)
Nigmatulin, B. I., 236(228)
Nikuradse, J., 274(45; 46; 47), 275(45; 46; 47),
276(47), 277(45; 45), 278(45), 279(46),
280(45; 46; 47), 281(46), 284(47), 286(47),
287(46; 47), 297(46), 298(46), 351(46),
352(45; 46), 353(46)
Nishihara, A., 415(121)
Nishihara, H., 150(31), 155(31), 169(31),
171(31), 172(31), 174(31), 185(31),
189(31), 190(31)
Nishimura, T., 425(150)
Nogotov, E. F., 57(142), 60(142)
Nomolov, E. V., 23(59)
Norris, D. M., 225(202), 226(202), 230(202),
231(202), 235(202; 223; 225)
Norris, P. M., 145(1)
Notter, R. H., 324(97), 327(97), 328(97),
330(97; 111), 332(97), 333(97), 340(97;
109; 110; 111), 354(97; 110; 111)
Nozad, I., 15(40), 34(40), 107(40), 126(40)
Nunner, W., 281(56), 353(56)
453 :i1no iNnrx
Nusselt, W., 337(102)
Nydahl, J. E., 205(141)
O
Ohori, Y., 425(150)
Okawa, W. J., 379(27)
Olek, S., 236(227)
Olivier, I., 428(154)
Orczag, S. A., 310(89), 325(89), 345(89),
354(89)
Orlanski, I., 377(21)
Ornatskiy, A. P., 210(160; 161; 162), 211(160;
161; 162)
Orszag, P. A., 396(74)
Orszag, S. A., 385(37), 386(40), 392(59),
394(59), 395(66)
Orszag, S. D., 286(66)
Ota, K., 409(105)
Ota, M., 384(130), 417(130), 419(130), 420(130)
Ota, T., 379(27), 402(89)
Oto, M., 388(132), 417(132)
Oya, T., 150(22), 153(22; 39)
O zisik, M. N., 37(93), 39(93)
Ozoe, H., 326(100a), 327(100a), 329(100a),
330(100a), 331(100a), 332(100a),
333(100a), 340(100a), 349(100a),
350(100a), 355(100a)
P
Paffenbarger, J., 113(206), 122(206)
Page, F., Jr., 310(90), 354(90)
Pai, S. I., 297(82), 351(82)
Pantankar, S. V., 398(77; 78; 79), 399(78; 79),
401(78)
Panton, R. L., 150(24), 153(24), 154(24),
158(24), 159(24), 166(24)
Papavassiliou, D. V., 325(98), 355(98)
Park, J.-H., 386(41)
Park, T. Y., 407(100), 408(100), 411(100)
Patel, V. C., 27(75), 391(57), 399(57)
Patterson, G. S., 396(74)
Paulsen, M. P., 231(219), 233(219)
Peasa, R. F., 191(109)
Pei, B. S., 221(187)
Peng, X. F., 191(110), 193(110; 153), 208(110;
153), 209(110; 154), 215(110; 153)
Perelman, T. L., 39(101)
Pereverzev, S. I., 56(121), 57(121)
Peterson, 231(219), 233(219)
Peterson, G. P., 191(111)
Peterson, R. B., 37(95), 41(95), 50(95)
Petrie, J. M., 338(104)
Petukhov, B. S., 314(92)
Phan, R. T., 26(67)
Pisano, A. P., 205(147)
Plumb, O. A., 15(44; 45), 18(44)
Poirer, D., 194(125), 198(125), 200(125),
202(125)
Pomeranchuk, I., 38(99)
Pomraning, G. C., 57(148; 149; 150; 151; 152),
58(150; 151), 59(148; 151; 152; 164),
61(152; 164), 63(152; 164)
Ponomarenko, A. T., 51(114; 115), 57(114;
115), 60(114; 115; 159; 160; 161), 61(114;
115), 66(159; 161), 97(115), 111(114; 115)
Pope, D. B., 225(203), 226(203), 235(203),
236(203)
Popov, A. M., 24(60; 61)
Poulikakos, D., 67(169)
Pourquie` , M., 393(60)
Prakash, C., 398(78), 399(78), 401(78)
Prandtl, L., 265(29), 269(24), 270(24; 29),
272(24; 37), 273(40), 280(55), 343(113)
Prata, A. T., 416(124), 421(124), 422(145)
Preziosi, L., 41(108)
Primak, A. V., 2(14; 15), 4(14; 15), 16(14; 15),
25(14; 15), 26(14; 15), 27(14; 15), 29(14;
15; 78), 30(14; 15), 116(14; 15)
Pritchard, A. J., 125(222)
Prosperetti, A., 23(57)
Q
Qin, T. Q., 145(1)
Qiu, T. Q., 39(102; 103; 104), 40(102; 103; 104)
Quereshi, Z. H., 193(132), 196(132), 197(132),
198(132), 199(132), 200(132), 202(132),
205(132)
Querfeld, C. W., 57(153)
Quintard, M., 6(29), 7(29; 30), 8(29), 9(29),
15(30)
454 :i1no iNnrx
R
Radushkevich, L. V., 1(5)
Rajkumar, M., 91(185), 92(185), 95(185)
Ramadhyani, S., 372(5)
Ramakers, F. J. M., 205(140)
Rangarajan, R., 52(119)
Raupach, M. R., 24(63), 25(64; 65; 66), 28(77)
Rayleigh, Lord, 258(10), 264(24; 25), 281(25)
Rayleigh, R. S., 12(39)
Reece, G. J., 388(52)
Regele, A., 372(6)
Reichardt, H., 294(76), 296(76), 298(76),
309(87), 344(87), 348(87)
Reimann, J., 191(105), 226(105), 227(105),
229(105), 230(105), 236(105), 239(105),
240(105)
Reiss, H., 56(141)
Renken, K., 67(169)
Revankar, S. T., 225(201), 232(201; 215),
236(201), 237(215)
Reynolds, A. J., 324(95)
Reynolds, O., 259(17; 18), 261(17), 264(17;
28), 315(18), 343(18)
Rezkallah, K. S., 153(38), 160(38), 161(38),
164(38), 165(38), 166(38)
Richardson, J. F., 338(105), 339(105)
Richter, H. J., 228(209), 236(209)
Richter, J. P., 257(2), 258(2)
Rieke, H. B., 324(96), 345(96), 354(96)
Roach, G. M., Jr., 193(132), 196(132),
197(132), 198(132), 199(132), 200(132),
202(132), 205(132), 210(174), 211(174),
214(174), 215(174), 216(174), 218(174),
219(174), 224(174)
Robertson, J. M., 275(49)
Roch, G. M., Jr., 193(135), 198(135), 202(135),
203(135), 204(135), 205(135)
Rodi, W., 26(72), 27(75), 381(28), 386(41),
387(28), 388(52), 391(56), 393(60)
Rogachevskii, I., 325(97a)
Rogers, J. T., 200(136), 202(136)
Rogers, T. J., 194(125), 198(125), 200(125),
202(125)
Rohsenow, W. M., 23(58), 195(129), 196(129),
197(129), 198(129), 309(88)
Romanov, G. S., 39(100)
Rothfus, R. R., 282(59; 61; 62), 283(61)
Rotstein, E., 15(41), 22(41), 30(41), 35(41),
64(41)
Rotta, J. C., 269(35), 280(35), 351(35)
Rubenstein, J., 34(85)
Rutledge, J., 287(69)
Ryvkina, N. G., 51(115), 57(115; 160; 161;
162), 60(115; 160; 161; 162), 61(115),
66(161), 97(115), 111(115)
S
Sadatomi, M., 150(29; 30), 155(29; 30), 171(29;
30), 174(30), 176(30), 177(30), 185(29;
30), 189(29; 30), 190(30)
Sadatomi, Y., 190(104)
Sadowski, D. L., 147(12), 150(12; 34), 151(12),
152(12), 153(12), 154(12), 156(12; 34),
157(12), 158(12), 159(12), 160(12),
161(12), 162(12), 164(12), 167(34),
168(34), 177(34), 185(12), 187(12),
188(34), 226(208), 227(208), 230(208),
231(208), 233(208)
Sage, B. H., 310(90), 354(90)
Saha, P., 194(121), 198(121), 199(121),
203(121), 222(121)
Sakamoto, M., 384(130), 388(132), 417(130;
132), 419(130), 420(130)
Salcudean, M., 192(118; 119; 120), 194(125),
198(125), 200(118; 119; 120; 125),
202(125)
Samaddar, S. N., 57(154)
Sangani, A. S., 11(32)
Saruwatari, S., 190(104)
Sasaki, Y., 407(108), 409(108)
Satake, S., 274(44)
Sato, T., 378(22), 398(81; 82), 399(81),
400(82), 407(82)
Sato, Y., 190(104)
Sawyers, D. R., 428(155)
Scheurer, G., 27(75)
Schlichting, H., 68(170), 258(16)
Schlinger, W. G., 310(90), 354(90)
Schrock, V. E., 225(201), 226(204), 227(204),
228(204), 229(204), 232(201; 204; 214;
215), 236(201; 204; 214), 237(214; 215),
239(204), 240(204), 241(204)
Schuerger, M. J., 415(121)
Schumann, U., 393(62)
Schwartz, F. W., 33(84), 34(84)
Schwellnus, C. F., 236(226)
455 :i1no iNnrx
Scott, P. M., 225(203), 226(203), 235(203),
236(203)
Seban, R. A., 316(93)
Sen, M., 428(155)
Senecal, V. E., 282(62)
Seynhaever, J. M., 232(216), 237(216)
Shabanskii, V. P., 38(98)
Shabbir, A., 386(42)
Shah, A. K., 385(38)
Shah, M. M., 220(185)
Shah, R. K. (Chapter Author), 363, 368(1),
369(1), 372(1), 373(8), 374(1), 378(90),
379(90), 399(90), 400(1), 402(90), 403(90;
91), 404(90; 91), 405(90; 93), 413(90),
421(136)
Sharma, B. I., 387(45)
Shaw, C. T., 377(19)
Shaw, D. A., 326(99)
Shaw, R. H., 24(63)
Shcherban, A. N., 2(14; 15), 4(14; 15), 16(14;
15), 25(14; 15), 26(14; 15), 27(14; 15),
29(14; 15), 30(14; 15)
Sherwood, T. K., 338(104)
Shevchenko, V., 57(161; 162), 60(161; 162),
66(161)
Shi, Z., 42(109)
Shih, T. H., 386(42)
Shimazaki, T. T., 316(93)
Shimura, T., 193(130), 196(130), 197(130),
198(130), 199(130), 200(130), 201(130),
202(130), 203(130), 210(169), 211(169),
212(169), 213(169), 217(169)
Shin, T. S., 205(145)
Shinagawa, T., 407(106), 409(106)
Shinoda, M., 346(117)
Shoukri, M., 236(226)
Shultze, H. D., 207(151)
Shvab, V. A., 26(74)
Siegel, R., 56(131), 60(131)
Sihvola, A. H., 57(146), 60(146)
Simoncini, M., 221(193), 223(193)
Singh, B. P., 56(134)
Skinner, B. C., 210(163), 211(163)
Slattery, J. C., 1(2), 2(6), 5(6), 60(6)
Sleicher, C. A., 287(69), 324(97), 327(97),
328(97), 330(97; 111), 332(97), 333(97),
340(97; 109; 110; 111), 354(97; 110; 111)
Smagorinsky, J. S., 394(64)
Smith, L., 33(84), 34(84)
Snoek, C. W., 209(158), 216(158)
Sobey, I. J., 422(141; 142; 143; 144), 424(141)
Sommerfeld, A., 258(11)
Sonin, A. A., 23(58)
Souto, H. P. A., 81(181), 85(181)
So zen, M., 145(2)
Spalart, P., 383(33), 392(33), 396(76)
Spalding, B. D., 91(191), 94(191), 95(191)
Spalding, D. B., 274(42), 292(74), 293(74),
352(42), 383(34), 384(34; 35), 386(34),
390(35)
Sparrow, E. M., 81(177), 82(177), 83(177),
84(177), 398(77), 416(124), 420(131),
421(124), 422(145)
Spedding, P. L., 150(21)
Spence, D. R., 150(21)
Speziale, C. G., 381(31), 386(40), 388(50)
Staroselsky, I., 392(59), 394(59)
Stasiek, J., 384(152), 388(152), 393(152),
423(14), 425(152), 426(14; 152), 428(152),
429(152), 430(152), 432(152)
Staub, F. W., 194(124), 198(124), 200(124),
202(124)
Stephanoff, K. D., 422(142)
Stewart, W. E., 77(173), 78(173), 79(173),
269(35), 280(35), 351(35)
Stokes, G. G., 261(20)
Stralen, S. V., 191(108), 205(108)
Strutt, J. W., 258(10), 264(24; 25), 281(25)
Stubbs, A. E., 146(5), 150(5)
Stuben, F. B., 225(203), 226(203), 235(203),
236(203)
Su, B., 59(164), 61(164), 63(164)
Subbotin, V. I., 23(59), 91(190), 94(190),
95(190), 210(165), 211(165)
Suga, K., 388(47), 407(106; 107), 409(106; 107)
Sugawara, S., 224(197; 198)
Sulaiman, Y., 407(109), 410(109)
Sunde n, B., 385(157), 388(157), 416(122),
421(134; 135), 428(157), 429(157),
432(157)
Suo, M., 147(9), 150(9), 153(9), 154(9), 160(9),
161(9), 165(9)
Suzuki, K., 378(22), 396(72), 398(81; 82),
399(72; 81; 83; 86), 400(72; 82; 83; 85; 86),
405(72; 85), 406(95), 407(82), 415(121)
T
Taborek, J., 91(191), 94(191), 95(191)
Taitel, Y., 148(15; 16; 17; 18; 20), 150(23),
456 :i1no iNnrx
154(23), 156(15; 17; 23), 158(23), 161(15;
23), 162(15; 20), 163(23), 166(17),
169(16), 171(17), 173(17)
Takagi, M., 407(108), 409(108)
Takatsu, Y., 26(69)
Tanaka, K., 407(108), 409(108)
Tanaka, T., 415(120)
Tanaka, Y., 226(206; 207), 227(206; 207)
Tanatarov, L. V., 38(97), 39(97)
Tang, D., 33(84), 34(84)
Tao, W. Q., 417(128), 419(128)
Tapucu, 22(51)
Taylor, G. I., 146(7), 150(7)
Taylor, N., 396(137), 421(137)
Tchmutin, I. A., 57(159; 160; 161; 162; 163),
60(159; 160; 161; 162), 66(159; 161)
Teo, K. L., 116(213), 124(213; 214; 215; 216;
217), 125(214)
Teyssedou, A., 22(51)
Thangam, S., 386(40)
Theofanous, T. G., 59(165)
Thochon, P. (Chapter Author), 363
Thomas, L. C., 343(114)
Thome, J. R., 148(14), 182(14), 191(14)
Thompson, B., 210(176)
Thompson, J. F., 376(16)
Thonon, B. (Chapter Author), 363, 373(12),
428(12)
Thulasidas, M. A., 151(36; 37)
Tien, C. L., 26(70), 39(102; 103; 104), 40(102;
103; 104), 46(110), 56(130; 133; 135),
145(1)
Tochon, P., 377(20), 396(20; 139), 399(20),
402(20), 405(20), 422(139), 423(139),
424(139), 429(139)
Todreas, N. E., 23(58)
Tomoda, T., 406(95)
Tong, L. S., 194(126), 217(182)
Torquato, S., 34(85; 86; 87; 88)
Tran, T. N., 192(113; 114; 115)
Trauger, P., 408(99), 414(99)
Travkin, V. S. (Chapter Author), 1, 2(14; 15;
16; 17; 18; 19; 20; 21; 22; 26; 27; 28),
3(19; 21), 4(14; 15), 10(21), 11(16; 18; 20;
26; 33; 34), 13(34), 15(18), 16(14; 15;
16; 18; 21), 21(21), 23(24), 25(14; 15; 16;
20), 26(14; 15; 16; 17; 18; 19; 20; 21),
27(14; 15), 29(14; 15; 78), 30(14; 15; 16;
21), 31(19; 26), 36(16; 17; 20; 26; 27),
51(114; 115), 52(23), 57(19; 20; 28; 114;
115; 158; 159; 160; 161; 162; 163), 60(21;
114; 115; 159; 160; 161), 61(114; 115),
62(21), 65(21; 28), 66(159; 161; 166),
68(16; 20), 69(16; 20; 21; 23; 25; 26),
70(16; 17; 20; 21), 71(16; 20), 79(21),
80(21), 81(23), 96(21), 97(21; 115),
102(21; 33; 34), 110(21), 111(114; 115),
116(14; 15; 16; 20; 21; 23; 28), 118(16; 19),
119(16; 20), 123(16; 17; 23), 124(19)
Tretyakov, S. A., 57(146), 60(146)
Tribus, M., 340(109)
Triplett, K. A., 147(12), 150(12), 151(12),
152(12), 153(12), 154(12), 156(12),
157(12), 158(12), 159(12), 160(12),
161(12), 162(12), 164(12), 185(12),
187(12)
Tromov, V. P., 57(142), 60(142)
Tsay, R., 111(202)
Tuckermann, D. B., 191(109)
Tura, R., 407(104)
Tzou, D. Y., 37(93; 96), 39(93), 40(96)
U
Udell, K. S., 205(147)
Ueda, T., 198(134)
Uehara, K., 210(170), 211(170), 212(170),
213(170), 217(170)
Umtsev, P. Y., 56(121), 57(121)
Uher, C., 96(192)
Uhlenbeck, G., 258(12), 261(12)
Unal, H. C., 194(122), 198(122), 205(144)
Ungar, K. E., 185(94), 186(94)
Usagi, R., 285(65)
V
Vafai, K., 26(70), 67(168), 145(2)
van de Hulst, H. C., 57(143), 60(143)
Vandervort, C. L., 193(131), 196(131),
206(131), 207(131), 208(131), 210(131;
173), 211(131; 173), 212(173), 213(131;
173), 215(131; 173), 216(173), 219(173)
van Dreist, R. R., 273(38)
Vanka, S. P., 425(149)
Van Stralen, S. J. D., 205(140)
Varadan, V. K., 57(147), 60(147)
Varadan, V. V., 57(147), 60(147)
457 :i1no iNnrx
Vidil, R., 373(12), 428(12)
Vinyarskiy, L. S., 210(162), 211(162)
Viskanta, R., 85(182; 183), 91(182; 183; 187),
93(183; 186; 187), 95(187; 188)
Vittanen, A. J., 57(146), 60(146)
von Ka rma n, T., 266(31), 269(31), 272(31)
von Mises, R., 258(13)
von Weizsa cker, C. F., 258(14)
Voskoboinikov, V. V., 91(190), 94(190),
95(190)
W
Wacholder, E., 236(227)
Wambsganss, M. W., 150(28), 155(28),
171(28), 175(28), 192(113; 114; 115)
Wang, B.-X., 191(110), 193(110; 153), 208(110;
153), 209(110; 154), 215(110; 153)
Wang, C. C., 371(3)
Wang, C. Y., 145(4)
Wang, G., 425(149)
Wang, S. K., 206(150), 207(150)
Ward, J. C., 83(178)
Warsi, Z. U. A., 376(16)
Webb, R. L., 114(207; 208), 187(98), 188(98),
408(99; 102), 414(99)
Wei, T., 299(84)
Weinbaum, S., 111(202)
Weisman, J., 221(186; 187; 188)
Westacott, J. L., 231(219), 233(219)
Westphal, F., 191(105), 226(105), 227(105),
229(105), 230(105), 236(105), 239(105),
240(105)
Westwater, J. W., 147(11), 150(11), 153(11),
154(11), 157(11), 158(11), 159(11),
161(11), 162(11), 164(11)
Whan, G. A., 282(60), 352(60)
Whitaker, S., 1(4), 2(9; 10), 5(10), 6(29), 7(29;
30; 31), 8(29; 31), 9(29; 31), 15(10; 30;
31; 40; 41; 42; 43; 44; 45), 18(42; 44),
22(41), 23(10; 42), 30(41), 34(40; 89),
35(41), 60(10), 64(41), 107(40), 116(10;
42), 126(40)
White, P. R. S., 407(103)
White, S., 56(139), 57(139)
Wieting, A. R., 398(80)
Wilcox, D. C., 381(30), 388(30), 392(30)
Willmarth, W. W., 299(84)
Wilmarth, T., 150(32), 155(32), 171(32),
172(32), 173(32), 174(32)
Winterton, R. H. S., 200(137)
Wio, H. S., 37(92)
Wright, C. C., 407(104)
Wu, Z. S., 116(213), 124(213)
X
Xi, G. N., 378(23; 90), 379(90), 396(72),
399(72; 83; 86; 90), 400(72; 83; 85; 86),
402(90), 403(90), 404(90), 405(72; 85; 90),
413(90)
Xiang, X., 125(225)
Xin, R. C., 417(128), 419(128)
Y
Yablonovitch, E., 52(116; 117)
Yadigaroglu, G., 194(127)
Yagi, Y., 400(84), 403(84), 404(84), 405(84)
Yahkot, A., 310(89), 325(89), 345(89), 354(89)
Yahkot, V., 310(89), 325(89), 345(89), 354(89)
Yakhot, A., 395(66)
Yakhot, M., 386(39)
Yakhot, V., 385(37), 386(40), 392(59), 394(59),
395(66)
Yamada, T., 24(62)
Yamaguchi, H., 384(130), 388(132), 417(130;
132), 419(130), 420(130)
Yamaguchi, Y., 388(126), 417(126), 418(126),
419(126), 420(126), 431(126)
Yan, Y.-Y., 185(96; 97), 187(96; 97), 188(97)
Yang, C.-Y., 187(98), 188(98)
Yang, L. C., 388(126), 417(126), 418(126),
419(126), 420(126), 431(126)
Yang, S. R., 205(143)
Yang, W.-J., 400(84), 403(84), 404(84), 405(84)
Yao, G., 183(86)
Yap, C. R., 387(46)
Yin, S. T., 198(133)
Yoda, M., 150(35), 155(35), 176(35), 177(35),
185(35), 191(35)
Yokohama, K., 184(91), 185(91)
Yokoya, S., 210(166), 211(166)
458 :i1no iNnrx
Younis, L. B., 91(187), 93(186; 187), 95(187;
188)
Yu, B., 326(100a), 327(100a), 329(100a),
330(100a), 331(100a), 332(100a),
333(100a), 340(100a), 349(100a),
350(100a), 355(100a)
Yu, F., 81(175; 176), 82(176), 91(175), 94(175)
Yurev, Yu. S., 23(59)
Z
Zachariades, J., 428(154)
Zagarola, M. V., 288(73), 289(73), 290(73),
291(73), 300(73), 302(73), 347(73),
351(73), 353(73)
Zeldovich, Ya. B., 258(15)
Zeng, L. Z., 205(146)
Zhang, D. Z., 23(57)
Zhao, L., 153(38), 160(38), 161(38), 164(38),
165(38), 166(38)
Zhu, J., 386(42)
Zijl, W., 205(140)
Zivi, S. M., 187(99), 234(99)
Zolotarev, P. P., 1(5)
Zuber, N., 194(121), 198(121), 199(121),
203(121), 222(121)
Zummo, G., 221(193), 223(193)
459 :i1no iNnrx
a
SUBJECT INDEX
A
Asymptotic dimensional analysis, 264269
B
Baldwin-Lomax model, 382383
Bandgaps, 5356
Boiling
nucleate, 195198
subcooled
forced ow
bubble nucleation, 205209
general issues, 191192
instability, 198205
nucleate onset, 195198
signicant void, 198205
void fractions, 192195
Boundaries
CHE surfaces, 376377
gain, 5354
Boundary L ayer T heory, 258
Bubble nucleation, 205209
C
Chandrasekhar, Subrahmanyan, 258
Channels
furrowed, 372
heat transfer, 429430
wavy
corrugated, 372, 416422
furrowed, 422425
via chevron plates, 429430
CHE. see Compact heat exchange
Chevron plates
description, 425429
local analysis, 429430
wavy channels via, 431432
CHF. see Critical heat ux
Closure theories, 3237
Colburn analogy, 342343
Colebrook equation, 284286
Compact heat exchange
characterization, 363
models
control problems, 123
current practice, 113116
development, 111112
optimization, 124127
VAT-based
equations, 117122
surfaces
chevron plates, 425430
experiments, 365366
interrupted ow passages
general, 366367
louver ns, 369371
offset strip ns, 368369
louver ns, 406416
numerical analysis
boundary conditions, 376377
general issues, 375
mesh generation, 376
solution algorithm, 376377
turbulence models
DNS, 392395
eddy viscosity, 381388
LES, 392395
Reynolds number ow, 391392
Reynolds stress, 388390
wall effects, 390392
uninterrupted complex
furrowed channel, 372
general issues, 371
intermating plates, 372373
Reynolds number, 374375
unsteady laminar, 374375
wavy corrugated channel, 372
wavy channels
furrowed, 422425
Composite media, 103108
461
Conductivity
composite media, 103108
hyperbolic heat, 4142
pure phase media, 101103
two-phase media
conventional formulation, 9798
local distribution, 96
piecewise distribution, 96
VAT considerations, 99101
Conservation
differential, 236240
electron, 4647
energy, 259
mass, 29, 258259
momentum, 258259
two-temperature, 4345
Convection
integrals, 311317
turbulent
alternative models, 320323
correlating equations, 356
differential models, 353354
differentials, 305309
geometry formulations, 318320
heat ux density ratio, 354
initial perspectives, 353
integrals
equations, 309310
expressions, 317318
isothermal wall, 331332
Nu correlations
differential analogy, 344345
dimensional analysis, 335337
empirical, 337339
integral, 355
low-Prandtl-number uids, 339342
mechanistic analogies, 342344
numerical, 355356
theoretically-based, 344348
parallel plates
channels, 333, 335
geometries, 318320
Prandtl number
convection, 323326
elimination, 354355
uids, 339342
structure development, 259260
uncertainty, 323324
uniformly heated tube
Nu values, 330
particular conditions, 326328
Pr values, 328329
Corrugated channels
heat transfer, 429430
wavy, 372, 416422
Cracks
microchannel
differential conservation, 236240
experiments, 225230
integral models
isentropic homogeneous-equilibrium,
232233
LEAK, 235236
Moodys, 234235
numerical models, 236240
Critical heat ux
microchannels
empirical correlations, 216220
experiments, 210215
mass, 215216
noncondensables, 215216
pressure, 215216
theoretical models, 221224
trends, 210215
Crystals
photonic, bandgap, 5356
subcrystalline single, 4546
D
Detailed micromodeling
description, 52
uid phase one, 108
porous media conductivity, 108, 110
radiative heat transport
heterogeneous media, 5758
porous media, 5758
thermal conductivity, 97
-VAT, mismatches, 5253
Differential conservation
microchannel, 236240
turbulent convection, 305309
Direct numerical modeling
description, 52
uid phase one, 108, 110
porous media conductivity, 108, 110
radiative heat transport, 5758
462 sinrc1 iNnrx
thermal conductivity, 97
VAT
mismatches, 5253
verication, 1213
Direct numerical simulation
convection, 259, 260
ow, 259
Dissipation, 306
Distribution, 292294
DMM. see Detailed micromodeling
DNM. see Direct numerical modeling
DNS. see Direct numerical simulation
Drift ux model, 172173, 175
Dynamic procedure model, 395
E
Eddy viscosity
description, 269
lter approach model, 394
one-equation models, 383
two-equation models
advantages, 383384
low Reynolds numbers, 387
realizable kc, 386387
RNG kc;, 384386
standard kc, 383384
zero-equation models, 382383
Einstein, Albert, 258
Electrodynamics, nonlocal
VAT-governing equations
photonic crystals bandgap, 5455
superstructures
acoustical phonon, 4950
electromagnetic, 5051
electron conservation, 4647
uid momentum, 4748
gas energy, 49
longitudinal phonon, 49
Electron conservation, 4647
Electron gas energy, 49
Ensemble averaging, 59
F
Filtered media, 27
Fins
louver, 369371, 406416
offset strip, 368369, 398406
Flow
CHE surface
interrupted
general, 366367
louver ns, 369371
complex passages
uninterrupted
general issues, 371
forced
subcooled boiling
linear models, 1
microchannel
annular, 170177
characteristics, 146147
CHF
experiments, 210215
mass, 215216
pressure, 215216
trends, 210215
conditions, 147148
correlations, 161166
in cracks
experimental data, 225230
integral models, 232236
denition, 148149
microgravity, 159161
narrow rectangular, 170177
noncondensable release, 178179
pressure drop
experiment review, 184191
fractional, 180183
463 sinrc1 iNnrx
Flow (Continued)
general issues, 180
regimes, 150153
rod bundle patterns, 166168
in slits
experiments, 225230
subcooled boiling
instability, 198205
surface wettability, 158159
transitions models, 161166
trends, 153, 156
resistance
porous media
experimental assessment, 6769
momentum in 1D membrane, 6975
pressure loss, 7780
simulation procedures, 8084
turbulent
asymptotic, 263269
dimensional, 263269
dimensional models, 273274
eddy viscosity, 269
exact structure, 260263
friction factor
description, 283286
geometry correlations, 303304
MacLeod analogy, 282283, 352
mixing length, 269273
model-free formulations, 294295
near center line values, 287
near wall values, 286287
new formulations, 303304
Nikuradse data, 274276
numerical simulations, 274
power-law models, 278282
recapitulation, 304
rough piping, 283286
shear stress
integral formulations, 350351
limited models, 352353
MacLeod analogy, 352
new model, 348349
obsolete models, 352353
shear stress correlations, 299301
speculative analyses, 263269
study, history, 257259
velocity
distribution, 292294
Zagarola data, 287292
Flux, 311316, 354
Fractions
heated channels, 192195
microchannels, 169170
Friction
convection, 284286
factors, 301303
Furrowed channels, 372
G
Gas energy, electron, 49
Gas-to-uid exchanger, 463464
Geometry
convection, 318320
turbulent ow, 303304
Grain boundaries, 5354
H
Harmonic eld equations, 6465
Heat
conductivity
composite media, 103108
hyperbolic, 4142
porous media, 108111
pure phase media, 101103
two-phase media
effective modeling, 9697
464 sinrc1 iNnrx
critical ux
microchannels
experiments, 210215
mass ux, 215216
pressure, 215216
trends, 210215
ux, 311316, 354
radiative, transport
heterogeneous media
issues, 5758
nonlocal volume, 6064
porous media
harmonic eld equations, 6465
issues, 5758
linear transfer, 58
transport
CHE models, 124125
microscale, 3743
wave transport
CHE models
development, 111112
VAT-based, 117122
crystal, 4546
superstructures
electron conversion, 4647
uid momentum, 4748
gas energy, 49
Heat transfer
CHE surface
interrupted
general, 366367
louver ns, 369371
general issues, 371
coefcient, 335, 337
in corrugated channels, 429430
porous media, coefcients
assumptions, 85
models, 8689
Heisenberg, Werner, 258
Heterogeneous media
2-phase, 11
issues, 5758
Heterogeneous media modeling, 5253
Highly porous media
turbulent transport
model development
additive components, 29
rst level hierarchy, 27
free stream, 28
mass conservation, 29
momentum equations, 3032
scalar diffusion, 29
separate obstacle, 28
theoretical bases, 2627
High-temperature superconductors, 101
HMM. see Heterogeneous media modeling
Homogeneous isotropic media, 11
HTSC. see High-temperature
superconductors
Hydrodynamics, 257
Hyperbolic heat conduction, 4142
I
Integral models
isentropic homogeneous-equilibrium, 232
233
LEAK, 235236
Moodys, 234235
Isentropic homogeneous-equilibrium model,
232233
K
Kapitsa, Pyotr, 258
465 sinrc1 iNnrx
L
Laminar ow
nonlinear uid medium
concentration value, 20
homogeneous phase diffusion, 20
mass transport, 21
momentum diffusion, 2021
Navier-Stokes equations, 1920
steady-state momentum, 21
porous media
VAT
diffusion equation, 18
divergence form, 17
uid phase, 17
impermeable interface, 18
solid phase, 17
Landau, Lev, 258
Large eddy simulations
CHE surfaces
basic features, 392393
DNS, 395397
lter approach, 393395
numerical scheme, 378380
Schumanns approach, 393
convection, 353
Law of the wall, 265, 283
LEAK model, 235236
Linear models, 1
Linear particle transport, 5859
Linear Stokes equations, 1516
Lorentz, Hendrik, 258
Louver ns, 406416
M
MacLeod analogy, 282283, 352
Mass ux, 215216
Mesh generation, 376
Microchannel ow
characteristics, 146147
two-phase media
annular, 170177
CHF
experiments, 210215
mass, 215216
pressure, 215216
trends, 210215
conditions, 147148
in cracks
denition, 148149
pressure drop
fractional, 180183
general issues, 180
in slits
experiments, 225230
subcooled boiling
instability, 198205
trends, 153, 156
Micro-rod bundles, 1661687
Microscale heat transport
heuristic approach, 37
traditional descriptions
coupling factor, 40
elastic lattice vibration, 3839
heat balance, 38
hyperbolic heat conduction, 4142
in metals, 3940
phonon radiative transfer, 41
466 sinrc1 iNnrx
in solids, 39
two-uid model, 40
Mises, Richard von, 258
N
Navier-Stokes equations
description, 261
Reynolds averaged
algebraic stress models, 388
description, 381382
eddy viscosity models, 382388
stress models, 388392
Newton, Isaac, 258
Noncondensables, 215216
Nucleate boiling, 195198
Nucleation, bubble, 205209
Nu values
convection
parallel-plate channels, 333, 335
uniformly heated tube, 330
correlations
empirical equations, 337339
theoretically based
components, 345346
interpretation, 348
isothermal plates, 347348
parallel plates, 347
round tubes, 346347
structure, 345346
test, 348
numerical solutions, 355356
O
Offset strip ns, 398406
P
Partial differential equations
CHE models, 115116
PDE. see Partial differential equations
Phonon, 41, 4950
Photography, strobe ash, 171
Photonic crystals bandgap, 5256
Plates
chevron
description, 425429
local analysis, 429430
wavy channels via, 431432
exchanger, 364
ne heat exchangers, 114115
isothermal, 347348
parallel
convection, 318320
equal uniform heating, 333
uniformly heated, 347
Porous media
ow resistance
momentum in 1D membrane
equations, 6975
model 1, 75
model 2, 75
model 4, 7576
pressure loss, 7780
heat transfer coefcients
assumptions, 85
uid phase one, 108111
models
conventional, 87
correct form, 8687
full energy equation, 8889
nonlinear uctuations, 88
liquid-impregnated, 66
nonlinear transport, 1517
issues, 5758
linear transfer, 5859
transport
closure theories, 3237
linear/nonlinear, 1517
Power-law models, 278282
Prandtl, Ludwig, 259
Prandtl analogy, 343
Prandtl number
convection, 323326
467 sinrc1 iNnrx
Prandtl number(Continued)
uids, 339342
tubes, 328329
turbulent, 354355
values, 328329
Pressure
drop, microchannel ow
fractional, 180183
general issues, 180
microchannel ow, 215216
Pressure loss experiments, 7780
PU-BTPFL-CHF Database, 217219
R
Radiative transport
heterogeneous media
issues, 5758
phonon, 41
porous media
issues, 5758
linear transfer, 58
VAT basis, 3
Rayleigh, Lord, 258
Renormalization group model, 395
Representative elementary volume
averaging types
differentiation conditions, 56
xed space, 45
lemma, 89
porous medium, 4
scale variables, 10
virtual, 78
heat transfer, 4445
transport averaging, 3
Resistance, ow
porous media
pressure loss, 7780
REV. see Representative elementary volume
Reynolds, Sir Osborne, 259
Reynolds analogy, 260, 343
Reynolds numbers
CHE surfacess, 374375, 391392
louver ns, 369370
Navier-Stokes equations
description, 381382
eddy viscosity models, 382388
stress models, 388390
wall effect models, 390392
Reynolds stress, 27
Rough piping, 284286
Round tubes, 346347
S
Scaling, 7780
Shear stress
local, equations, 299301
turbulent ow
new model, 348349
Signicant void, 198205
Slits
microchannel
experiments, 225230
integral models
isentropic homogeneous-equilibrium,
232233
LEAK, 235236
Moodys, 234235
Smagorinsky model, 394395
Sommerfeld, Arnold, 258
Space averaging, 261262
Spalart-Allmaras model, 383
Speculation, 263264
Stress
algebraic models, 388
Reynolds, 262, 388390
shear
equations, 299301
turbulent ow
468 sinrc1 iNnrx
new model, 348349
Strobe ash photography, 171
Structure function model, 395
Subcooled boiling
forced ow
instability, 198205
Subcrystalline single crystals, 4546
Surfaces
CHE
interrupted ow passages
general, 366367
louver ns, 369371
louver ns, 406416
numerical analysis
general issues, 375
mesh generation, 376
turbulence models
algebraic stress, 388
DNS, 392395
LES, 392395
Reynolds number ow, 391392
general issues, 371
wavy channels
furrowed, 422425
wettability, 158159
T
Temperatures
logitudinal phonon, 49
uniform wall, 316317
Transfer, heat
CHE surface
interrupted
general, 366367
louver ns, 368370
general issues, 371
coefcient, 335, 337
corrugated channels, 429430
Transport
averaging
REV
description, 3
differentiation conditions, 56
xed space, 45
lemma, 89
porous medium, 4
scale variables, 10
virtual, 78
heat wave
CHE models
development, 111112
VAT-based, 117122
superstructures
uid momentum, 4748
gas energy, 49
linear particle, 5859
microscale heat
heuristic approach, 37
traditional descriptions
469 sinrc1 iNnrx
Transport (Continued)
coupling factor, 40
elastic lattice vibration, 3839
heat balance, 38
hyperbolic heat conduction, 4142
in metals, 3940
phonon radiative transfer, 41
in solids, 39
two-uid model, 40
porous media
closure theories, 3237
nonlinear, 1517
raditative heat
issues, 5758
linear transfer, 5859
VAT, 3
radiative heat
heterogeneous media
issues, 5758
Tubes
round, 346347
uniformly heated
Nu values, 330
Pr values, 328329
Turbulence
CHE surfaces
DNS, 395397
LES, 392395
models
algebraic stress, 388
zero-equation, 382383
Turbulent convection
alternative models, 320323
differentials, 305309
geometry formulations, 318320
heat ux density ratio, 354
initial perspectives, 353
integrals
general equations, 309310
generalized expressions, 317318
uniform wall
heat ux, 311316
temperature, 311316
models, 353354
Nu correlations
integral, 355
theoretically based
components, 345346
interpretation, 348
isothermal plates, 347348
parallel plates, 347
round tubes, 346347
structure, 345346
test, 348
parallel plates
different uniform temperatures, 333, 335
equal uniform heating, 333
Prandtl number, 323326, 354355
uncertainty, 323324
uniformly heated tube
Nu values, 330
Pr values, 328329
Turbulent ow
asymptotic analysis, 26369
dimensional models, 273274
eddy viscosity, 269
exact structure, 260263
friction factor
description, 283286
geometry correlations, 303304
mixing length, 269273
model-free formulations, 294295
near centerline values, 287
near wall values, 286287
new formulations, 303304
Nikuradse data, 274276
numerical simulations, 274
470 sinrc1 iNnrx
power-law models, 278282
recapitulation, 304
rough piping, 283286
shear stress
new model, 348349
speculative analyses, 263269
study, history, 257259
velocity distribution
description, 292294
Zagarola data, 287292
Turbulent transport
porous media
nonlinear, 1417
theory, 2627
Two-phase media
microchannel ow
annular, 170177
CHF
experiments, 210215
mass, 215216
pressure, 215216
trends, 210215
conditions, 147148
in cracks, 232236
denition, 148149
pressure drop
fractional, 180183
general issues, 180
regimes, 150153, 156
in slits
experiments, 225230
subcooled boiling
instability, 198205
trends, 153, 156
thermal conductivity
Two-temperature conservation, 4345
U
Uhlenbeck, George, 258
V
VAT. see Volume averaging theory
Velocity
distribution, 292294
Vinci, da Leonardo, 258
Viscosity
dissipation, 264269, 306
eddy
description, 393
lter approach model, 394
one-equation models, 383
two-equation models
advantages, 383
low Reynolds numbers, 387
471 sinrc1 iNnrx
Viscosity (Continued)
realizable kc, 386387
RNG kc, 384386
standard kc, 383384
zero-equation models, 382383
Viscous shear stress law, 269
Void fractions, 169170, 192195
Volume averaging theory
development, 12
electrodynamics, nonlocal, 4647
features, 12
heat wave transport
CHE models
design problems, 123
development, 111112
equations, 117122
subcrystalline crystal, 4546
superstructures
uid momentum, 4748
gas energy, 49
heterogeneous media, 99101
highly porous medium turbulent
model development
additive components, 29
rst level hierarchy, 27
free stream, 28
mass conservation, 29
scalar diffusion, 29
separate obstacle, 28
nonlinear uid medium
laminar ow
concentration value, 20
homogeneous phase diffusion, 20
mass transport, 21
momentum diffusion, 2021
Navier-Stokes equations, 1920
steady-state momentum, 21
photonic crystals bandgap
DMM-DMN mismatches, 5253
governing equations, 5455
porous media
data reduction, 6667
internal heat transfer
assumptions, 85
models, 8689
laminar ow
diffusion equation, 18
divergence form, 17
uid phase, 17
impermeable interface, 18
solid phase, 17
linear Stokes equations, 15
pressure loss, 7778
porous medium transport, 3237
porous medium turbulent, 2126
basis, 3
heterogeneous media
issues, 5758
porous media
issues, 5758
linear transfer, 58
theorem verication
1D Cartesian coordinate version, 11
1-dimensional cases, 12
DMA, 12132
integral terms, 11
2-phase heterogeneous medium, 1112
3-phase homogeneous medium, 12
solid-phase equation, 10
steady-state conduction, 11
two-temperature conservation, 4345
W
Wall
effects, model
features, 390
function, 390391
turbulence, 391392
isothermal, 331332
law of, 265, 283
near, convection, 286287
472 sinrc1 iNnrx
uniform density, 311316
uniform temperature, 316317
Wavy channels
furrowed, 422425
via chevron plates, 431432
Weizscker, C.R. von, 258
Wettability, surface, 158159
Z
Zeldovich, Yakob, 258
Zero-equation models, 382383
473 sinrc1 iNnrx
a

Advances in Heat Transfer - Advances in Heat Transfer Volume 34 2000 PDF

Încărcat de

Informații document

Descriere originală:

Titlu original

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

Advances in Heat Transfer - Advances in Heat Transfer Volume 34 2000 PDF

Încărcat de

Drepturi de autor:

Formate disponibile

ADVANCES IN

f (t, x) dc. (5)

) is a periodic function of zero mean

cannot always be equal to zero, because it

value should not be zero. If it is

yields the intrinsic average

(see Fig. 1) results in

in terms of the position vector,

as indicated in Fig. 3 (see Fig. 1), or

) continues to be dependent on the

, which may have changed location within

, which is the subregion occupied by phase

in the heterogeneous two-phase medium (Fig.

is assumed to have different values at

, then the averaged function (x)}

of variable averaged in the REV

. According to Carbonell and Whitaker [31] and Quintard and

within a boundary cAD

which belongs to some REV AD

/2), with c being an

an innitely small distance, as does y

is reached, which means that the averaged value }

of the REV can be arbitrarily moved,

. In other words, for

there is only one x

that represents the value

on the upper eld level (macroscale eld) if both are mapped on

) c, then there is another

is a constant value, the

/AD, and the volume fraction of ller, or s-phase, is

is the unit vector outward to the s-phase.

VT ,, and for the stationary case there results

prescribed at the sample boundaries, which,

, and for the half-space below the BB plane

0, 0.2, 1, 10, and 10,000, are presented in

Vp V [j(VV (VV )*)] j

and z are scalar elds and nonlinear diffusion coefcients for

, that is not equal to the turbulent eddy

)' and the effective

being the area within the volume AD

occupied by air. Raupach

that yielded the averaged equation for

is an element drag coefcient and S

is particle Reynolds number.

and its uctuation represent any scalar eld that might be

and the phonon tem-

(T ) is the electronic specic heat,

in the following formula

is the thermal conductivity of the electron gas. Using the

, equivalent thermal diffusiv-

is the directional-spectral phonon intensity, V

(T (x)) is the equilibrium intensity corresponding

is the electron optical

is the optical acoustical relaxation time, and k

is the n-doping concentration, N

is the p-doping concentration, and p is the hole number density.

V(icE), V(VE) V(icH), (168)

ic(icH) c`H (169)

, as a factor on the right-hand side of each of the

(r) the absorption and

is the complex dielectric function c

(x, c). In many