:
TURBULENCE, SIMULATION
AND DYNAMICS
No part of this digital document may be reproduced, stored in a retrieval system or transmitted in any form or
by any means. The publisher has taken reasonable care in the preparation of this digital document, but makes no
expressed or implied warranty of any kind and assumes no responsibility for any errors or omissions. No
liability is assumed for incidental or consequential damages in connection with or arising out of information
contained herein. This digital document is sold with the clear understanding that the publisher is not engaged in
rendering legal, medical or any other professional services.
FLUID MECHANICS AND PIPE FLOW:
TURBULENCE, SIMULATION
AND DYNAMICS
DONALD MATOS
AND
CRISTIAN VALERIO
EDITORS
Nova Science Publishers, Inc.
New York
Copyright © 2009 by Nova Science Publishers, Inc.
All rights reserved. No part of this book may be reproduced, stored in a retrieval system or
transmitted in any form or by any means: electronic, electrostatic, magnetic, tape, mechanical
photocopying, recording or otherwise without the written permission of the Publisher.
For permission to use material from this book please contact us:
Telephone 6312317269; Fax 6312318175
Web Site: http://www.novapublishers.com
NOTICE TO THE READER
The Publisher has taken reasonable care in the preparation of this book, but makes no expressed or
implied warranty of any kind and assumes no responsibility for any errors or omissions. No
liability is assumed for incidental or consequential damages in connection with or arising out of
information contained in this book. The Publisher shall not be liable for any special,
consequential, or exemplary damages resulting, in whole or in part, from the readers’ use of, or
reliance upon, this material. Any parts of this book based on government reports are so indicated
and copyright is claimed for those parts to the extent applicable to compilations of such works.
Independent verification should be sought for any data, advice or recommendations contained in
this book. In addition, no responsibility is assumed by the publisher for any injury and/or damage
to persons or property arising from any methods, products, instructions, ideas or otherwise
contained in this publication.
This publication is designed to provide accurate and authoritative information with regard to the
subject matter covered herein. It is sold with the clear understanding that the Publisher is not
engaged in rendering legal or any other professional services. If legal or any other expert
assistance is required, the services of a competent person should be sought. FROM A
DECLARATION OF PARTICIPANTS JOINTLY ADOPTED BY A COMMITTEE OF THE
AMERICAN BAR ASSOCIATION AND A COMMITTEE OF PUBLISHERS.
LIBRARY OF CONGRESS CATALOGINGINPUBLICATION DATA
Fluid mechanics and pipe flow : turbulence, simulation, and dynamics / editors, Donald Matos and Cristian
Valerio.
p. cm.
Includes bibliographical references and index.
ISBN 9781616689902 (EBook)
1. Fluid mechanics. 2. PipeFluid dynamics. I. Matos, Donald. II. Valerio, Cristian.
TA357.F5787 2009
620.1'06dc22
2009017666
Published by Nova Science Publishers, Inc. New York
CONTENTS
Preface vii
Chapter 1 Solute Transport, Dispersion, and Separation in Nanofluidic
Channels
1
Xiangchun Xuan
Chapter 2 H
2
O in the Mantle: From Fluid to HighPressure Hydrous
Silicates
27
N.R. Khisina, R. Wirth and S. Matsyuk
Chapter 3 On the Numerical Simulation of Turbulence Modulation in Two
Phase Flows
41
K. Mohanarangam and J.Y. Tu
Chapter 4 A Review of Population Balance Modelling for Multiphase Flows:
Approaches, Applications and Future Aspects
117
Sherman C.P. Cheung, G.H. Yeoh and J.Y. Tu
Chapter 5 Numerical Analysis of Heat Transfer and Fluid Flow for Three
Dimensional Horizontal Annuli with Open Ends
171
ChunLang Yeh
Chapter 6 Convective Heat Transfer in the Thermal Entrance Region
of Parallel Flow DoublePipe Heat Exchangers
for NonNewtonian Fluids
205
Ryoichi Chiba, Masaaki Izumi and Yoshihiro Sugano
Chapter 7 Numerical Simulation of Turbulent Pipe Flow 231
M. OuldRouis and A.A. Feiz
Chapter 8 Pipe Flow Analysis of Uranium Nuclear Heating with Conjugate
Heat Transfer
269
G.H. Yeoh and M.K.M. Ho
Chapter 9 First and Second Law Thermodynamics Analysis of Pipe Flow 317
Ahmet Z. Sahin
Contents vi
Chapter 10 SinglePhase Incompressible Fluid Flow
in Mini and Microchannels
343
Lixin Cheng
Chapter 11 Experimental Study of Pulsating Turbulent Flow
through a Divergent Tube
365
Masaru Sumida
Chapter 12 Solution of an Airfoil Design Inverse Problem for a Viscous Flow
Using a Contractive Operator
379
Jan Šimák and Jaroslav Pelant
Chapter 13 Some Free Boundary Problems in Potential Flow Regime Using
the Level Set Method
399
M. Garzon, N. BobilloAres and J.A. Sethian
Chapter 14 A New Approach for Polydispersed Turbulent TwoPhase Flows:
The Case of Deposition in PipeFlows
441
S. Chibbaro
Index 455
PREFACE
Fluid mechanics is the study of how fluids move and the forces that develop as a result.
Fluids include liquids and gases and fluid flow can be either laminar or turbulent. This book
presents a level set based methodology that will avoid problems in potential flow models with
moving boundaries. A review of the stateoftheart population balance modelling techniques
that have been adopted to describe the nature of dispersed phase in multiphase problems is
presented as well. Recent works that are aimed at putting forward the main ideas behind a
new theoretical approach to turbulent wallbounded flows are examined, including a stateof
theart review on singlephase incompressible fluid flow.
Recent breakthrough in nanofabrication has stimulated the interest of solute separation in
nanofluidic channels. Since the hydraulic radius of nanochannels is comparable to the
thickness of electric double layers, the enormous electric fields inherent to the latter generate
transverse electromigrations causing chargedependent solute distributions over the channel
crosssection. As a consequence, the nonuniform fluid flow through nanochannels yields
chargedependent solute speeds enabling the separation of solutes by charge alone.
In Chapter 1 we develop a theoretical model of solute transport, dispersion and separation in
electroosmotic and pressuredriven flows through nanofluidic channels. This model provides
a basis for the optimization of solute separation in nanochannels in terms of selectivity and
resolution as traditionally defined.
As presented in Chapter 2, infrared spectroscopic data show that nominally anhydrous
olivine (Mg,Fe)
2
SiO
4
contains traces of H
2
O, up to several hundred wt. ppm of H
2
O (Miller et
al., 1987; Bell et al., 2004; KochMuller et al., 2006; Matsyuk & Langer, 2004) and therefore
olivine is suggested to be a water carrier in the mantle (Thompson, 1992). Protonation of
olivine during its crystallization from a hydrous melt resulted in the appearance of intrinsic
OHdefects (Libowitsky & Beran, 1995). Mantle olivine nodules from kimberlites were
investigated with FTIR and TEM methods (Khisina et al., 2001, 2002, 2008). The results are
the following: (1) Water content in xenoliths is lower than water content in xenocrysts. From
these data we concluded that kimberlite magma had been saturated by H
2
O, whereas adjacent
mantle rocks had been crystallized from waterdepleted melts. (2) Extrinsic water in olivine is
represented by highpressure phases, 10ÅPhase Mg
3
Si
4
O
10
(OH)
2
.nH
2
O and hydrous olivine
n(Mg,Fe)
2
SiO
4
.(H
2
MgSiO
4
), both of which belong to the group of Dense Hydrous
Magnesium Silicates (DHMS), which were synthesized in laboratory highpressure
experiments (Prewitt & Downs, 1999). The DHMS were regarded as possible mineral carriers
for H
2
O in the mantle; however, they were not found in natural material until quite recently.
Donald Matos and Cristian Valerio viii
Our observations demonstrate the first finding of the 10ÅPhase and hydrous olivine as a
mantle substance. (3) 10ÅPhase, which occurred as either nanoinclusions or narrow veins in
olivine, is a ubiquitous nanomineral of kimberlite and closely related to olivine. (4) There are
two different mechanisms of the 10ÅPhase formation: (a) purification of olivine from OH
bearing defects resulting in transformation of olivine to the 10 ÅPhase with the liberation of
water fluid; and (b) replacement of olivine for the 10ÅPhase due to hydrous metasomatism
in the mantle in the presence of H
2
O fluid.
With the increase of computational power, computational modelling of twophase flow
problems using computational fluid dynamics (CFD) techniques is gradually becoming
attractive in the engineering field. The major aim of Chapter 3 is to investigate the Turbulence
Modulation (TM) of dilute two phase flows. Various density regimes of the twophase flows
have been investigated in this paper, namely the dilute GasParticle (GP) flow, Liquid
Particle (LP) flow and also the LiquidAir (LA) flows. While the density is quite high for the
dispersed phase flow for the gasparticle flow, the density ratio is almost the same for the
liquid particle flow, while for the liquidair flow the density is quite high for the carrier phase
flow. The study of all these density regimes gives a clear picture of how the carrier phase
behaves in the presence of the dispersed phases, which ultimately leads to better design and
safety of many twophase flow equipments and processes. In order to carry out this approach,
an EulerianEulerian TwoFluid model, with additional source terms to account for the
presence of the dispersed phase in the turbulence equations has been employed for particulate
flows, whereas Population Balance (PB) have been employed to study the bubbly flows. For
the dilute gasparticle flows, particleturbulence interaction over a backwardfacing step
geometry was numerically investigated. Two different particle classes with same Stokes
number and varied particle Reynolds number are considered in this study. A detailed study
into the turbulent behaviour of dilute particulate flow under the influence of two carrier
phases namely gas and liquid was also been carried out behind a sudden expansion geometry.
The major endeavour of the study is to ascertain the response of the particles within the
carrier (gas or liquid) phase. The main aim prompting the current study is the density
difference between the carrier and the dispersed phase. While the ratio is quite high in terms
of the dispersed phase for the gasparticle flows, the ratio is far more less in terms of the
liquidparticle flows. Numerical simulations were carried out for both these classes of flows
and their results were validated against their respective sets of experimental data. For the
LiquidAir flows the phenomenon of drag reduction by the injection of microbubbles into
turbulent boundary layer has been investigated using an EulerianEulerian twofluid model.
Two variants namely the Inhomogeneous and MUSIG (MUltipleSIzeGroup) based on
Population balance models are investigated. The simulated results were benchmarked against
the experimental findings and also against other numerical studies explaining the various
aspects of drag reduction. For the two Reynolds number cases considered, the buoyancy with
the plate on the bottom configuration is investigated, as from the experiments it is seen that
buoyancy seem to play a role in the drag reduction. The under predictions of the MUSIG
model at low flow rates was investigated and reported, their predictions seem to fair better
with the decrease of the breakup tendency among the microbubbles.
Population balance modelling is of significant importance in many scientific and
industrial instances such as: fluidizations, precipitation, particles formation in aerosols,
bubbly and droplet flows and so on. In population balance modelling, the solution of the
population balance equation (PBE) records the number of entities in dispersed phase that
Preface ix
always governs the overall behaviour of the practical system under consideration. For the
majority of cases, the solution evolves dynamically according to the “birth” and “death”
processes of which it is tightly coupled with the system operation condition. The
implementation of PBE in conjunction with the Computational Fluid Dynamics (CFD) is
thereby becoming ever a crucial consideration in multiphase flow simulations. Nevertheless,
the inherent integrodifferential form of the PBE poses tremendous difficulties on its solution
procedures where analytical solutions are rare and impossible to be achieved. In Chapter 4,
we present a review of the stateoftheart population balance modelling techniques that have
been adopted to describe the phenomenological nature of dispersed phase in multiphase
problems. The main focus of the review can be broadly classified into three categories: (i)
Numerical approaches or solution algorithms of the PBE; (ii) Applications of the PBE in
practical gasliquid multiphase problems and (iii) Possible aspects of the future development
in population balance modelling. For the first category, details of solution algorithms based
on both method of moment (MOM) and discrete class method (CM) that have been proposed
in the literature are provided. Advantages and drawbacks of both approaches are also
discussed from the theoretical and practical viewpoints. For the second category, applications
of existing population balance models in practical multiphase problems that have been
proposed in the literature are summarized. Selected existing mathematical closures for
modelling the “birth” and “death” rate of bubbles in gasliquid flows are introduced.
Particular attention is devoted to assess the capability of some selected models in predicting
bubbly flow conditions through detail validation studies against experimental data. These
studies demonstrate that good agreement can be achieved by the present model by comparing
the predicted results against measured data with regards to the radial distribution of void
fraction, Sauter mean bubble diameter, interfacial area concentration and liquid axial velocity.
Finally, weaknesses and limitations of the existing models are revealed are suggestions for
further development are discussed. Emerging topics for future population balance studies are
provided as to complete the aspect of population balance modelling.
Study of the heat transfer and fluid flow inside concentric or eccentric annuli can be
applied in many engineering fields, e.g. solar energy collection, fire protection, underground
conduit, heat dissipation for electrical equipment, etc. In the past few decades, these studies
were concentrated in twodimensional research and were mostly devoted to the investigation
of the effects of convective heat transfer. However, in practical situation, this problem should
be threedimensional, except for the vertical concentric annuli which could be modeled as
twodimensional (axisymmetric). In addition, the effects of heat conduction and radiation
should not be neglected unless the outer cylinder is adiabatic and the temperature of the flow
field is sufficiently low. As the author knows, none of the open literature is devoted to the
investigation of the conjugated heat transfer of convection, conduction and radiation for this
problem. The author has worked in industrial piping design area and is experienced in this
field. The author has also employed threedimensional bodyfitted coordinate system
associated with zonal grid method to analyze the natural convective heat transfer and fluid
flow inside threedimensional horizontal concentric or eccentric annuli with open ends.
Owing to its broad application in practical engineering problems, Chapter 5 is devoted to a
detailed discussion of the simulation method for the heat transfer and fluid flow inside three
dimensional horizontal concentric or eccentric annuli with open ends. Two illustrative
problems are exhibited to demonstrate its practical applications.
Donald Matos and Cristian Valerio x
In Chapter 6, the conjugated Graetz problem in parallel flow doublepipe heat exchangers
is analytically solved by an integral transform method—Vodicka’s method—and an analytical
solution to the fluid temperatures varying along the radial and axial directions is obtained in a
completely explicit form. Since the present study focuses on the range of a sufficiently large
Péclet number, heat conduction along the axial direction is considered to be negligible. An
important feature of the analytical method presented is that it permits arbitrary velocity
distributions of the fluids as long as they are hydrodynamically fully developed. Numerical
calculations are performed for the case in which a Newtonian fluid flows in the annulus of the
double pipe, whereas a nonNewtonian fluid obeying a simple power law flows through the
inner pipe. The numerical results demonstrate the effects of the thermal conductivity ratio of
the fluids, Péclet number ratio and powerlaw index on the temperature distributions in the
fluids and the amount of exchanged heat between the two fluids.
Many experimental and numerical studies have been devoted to turbulent pipe flows due
to the number of applications in which theses flows govern heat or mass transfer processes:
heat exchangers, agricultural spraying machines, gasoline engines, and gas turbines for
examples. The simplest case of nonrotating pipe has been extensively studied experimentally
and numerically. Most of pipe flow numerical simulations have studied stability and
transition. Some Direct Numerical Simulations (DNS) have been performed, with a 3D
spectral code, or using mixed finite difference and spectral methods. There is few DNS of the
turbulent rotating pipe flow in the literature. Investigations devoted to Large Eddy
Simulations (LES) of turbulence pipe flow are very limited. With DNS and LES, one can
derive more turbulence statistics and determine a wellresolved flow field which is a
prerequisite for correct predictions of heat transfer. However, the turbulent pipe flows have
not been so deeply studied through DNS and LES as the planechannel flows, due to the
peculiar numerical difficulties associated with the cylindrical coordinate system used for the
numerical simulation of the pipe flows.
Chapter 7 presents Direct Numerical Simulations and Large Eddy Simulations of fully
developed turbulent pipe flow in nonrotating and rotating cases. The governing equations are
discretized on a staggered mesh in cylindrical coordinates. The numerical integration is
performed by a finite difference scheme, secondorder accurate in space and time. The time
advancement employs a fractional step method. The aim of this study is to investigate the
effects of the Reynolds number and of the rotation number on the turbulent flow
characteristics. The mean velocity profiles and many turbulence statistics are compared to
numerical and experimental data available in the literature, and reasonably good agreement is
obtained. In particular, the results show that the axial velocity profile gradually approaches a
laminar shape when increasing the rotation rate, due to the stability effect caused by the
centrifugal force. Consequently, the friction factor decreases. The rotation of the wall has
large effects on the root mean square (rms), these effects being more pronounced for the
streamwise rms velocity. The influence of rotation is to reduce the Reynolds stress component
〈V
r
'V
z
'〉 and to increase the two other stresses 〈V
r
'V
θ
'〉 and 〈V
θ
'V
z
'〉. The effect of the Reynolds
number on the rms of the axial velocity (〈V
z
'
2
〉
1/2
) and the distributions of 〈V
r
'V
z
'〉 is evident,
and it increases with an increase in the Reynolds number. On the other hand, the 〈V
r
'V
θ
'〉
profiles appear to be nearly independent of the Reynolds number. The present DNS and LES
predictions will be helpful for developing more accurate turbulence models for heat transfer
and fluid flow in pipe flows.
Preface xi
The field of computational fluid dynamics (CFD) has evolved from an academic curiosity
to a tool of practical importance. Applications of CFD have become increasingly important in
nuclear engineering and science, where exacting standards of safety and reliability are
paramount. The newlycommissioned Open Pool Australian Lightwater (OPAL) research
reactor at the Australian Nuclear Science and Technology Organisation (ANSTO) has been
designed to irradiate uranium targets to produce molybdenum medical isotopes for diagnosis
and radiotherapy. During the irradiation process, a vast amount of power is generated which
requires efficient heat removal. The preferred method is by lightwater forced convection
cooling—essentially a study of complex pipe flows with coupled conjugate heat transfer.
Feasibility investigation on the use of computational fluid dynamics methodologies into
various pipe flow configurations for a variety of molybdenum targets and pipe geometries are
detailed in Chapter 8. Such an undertaking has been met with a number of significant
modeling challenges: firstly, the complexity of the geometry that needed to be modeled.
Herein, challenges in grid generation are addressed by the creation of purposebuilt body
fitted and/or unstructured meshes to map the intricacies within the geometry in order to
ensure numerical accuracy as well as computational efficiency in the solution of the predicted
result. Secondly, various parts of the irradiation rig that are required to be specified as
composite solid materials are defined to attain the correct heat transfer characteristics.
Thirdly, the use of an appropriate turbulence model is deemed to be necessary for the correct
description of the fluid and heat flow through the irradiation targets, since the heat removal is
forced convection and the flow regime is fully turbulent, which further adds to the complexity
of the solution. As complicated as the computational fluid dynamics modeling is, numerical
modeling has significantly reduced the cost and lead time in the molybdenumtarget design
process, and such an approach would not have been possible without the continual
improvement of computational power and hardware. This chapter also addresses the
importance of experimental modeling to evaluate the design and numerical results of the
velocity and flow paths generated by the numerical models. Predicted results have been found
to agree well with experimental observations of pipe flows through transparent models and
experimental measurements via the Laser Doppler Velocimetry instrument.
In Chapter 9, the entropy generation for during fluid flow in a pipe is investigated. The
temperature dependence of the viscosity is taken into consideration in the analysis. Laminar
and turbulent flow cases are treated separately. Two types of thermal boundary conditions are
considered; uniform heat flux and constant wall temperature. In addition, various cross
sectional pipe geometries were compared from the point of view of entropy generation and
pumping power requirement in order to determine the possible optimum pipe geometry which
minimizes the exergy losses.
Chapter 10 aims to present a stateoftheart review on singlephase incompressible fluid
flow in mini and microchannels. First, classification of mini and microchannels is
discussed. Then, conventional theories on laminar, laminar to turbulent transition and
turbulent fluid flow in macrochannels (conventional channels) are summarized. Next, a brief
review of the available studies on singlephase incompressible fluid flow in mini and micro
channels is presented. Some experimental results on single phase laminar, laminar to
turbulent transition and turbulent flows are presented. The deviations from the conventional
friction factor correlations for singlephase incompressible fluid flow in mini and micro
channels are discussed. The effect factors on mini and microchannel singlephase fluid flow
are analyzed. Especially, the surface roughness effect is focused on. According to this review,
Donald Matos and Cristian Valerio xii
the future research needs have been identified. So far, no systematic agreed knowledge of
singlephase fluid flow in mini and microchannels has yet been achieved. Therefore, efforts
should be made to contribute to systematic theories for microscale fluid flow through very
careful experiments.
In Chapter 11, an experimental investigation was conducted of pulsating turbulent flow in
a conically divergent tube with a total divergence angle of 12°. The experiments were carried
out under the conditions of Womersley numbers of α =10∼40, mean Reynolds number of Re
ta
=20000 and oscillatory Reynolds number of Re
os
=10000 (the flow rate ratio of η = 0.5).
Timedependent wall static pressure and axial velocity were measured at several longitudinal
stations and the distributions were illustrated for representative phases within one cycle. The
rise between the pressures at the inlet and the exit of the divergent tube does not become too
large when the flow rate increases, it being moderately high in the decelerative phase. The
profiles of the phaseaveraged velocity and the turbulence intensity in the cross section are
very different from those for steady flow. Also, they show complex changes along the tube
axis in both the accelerative and decelerative phases.
Chapter 12 deals with a numerical method for a solution of an airfoil design inverse
problem. The presented method is intended for a design of an airfoil based on a prescribed
pressure distribution along a mean camber line, especially for modifying existing airfoils. The
main idea of this method is a coupling of a direct and approximate inverse operator. The goal
is to find a pseudodistribution corresponding to the desired airfoil with respect to the
approximate inversion. This is done in an iterative way. The direct operator represents a
solution of a flow around an airfoil, described by a system of the NavierStokes equations in
the case of a laminar flow and by the k−ω model in the case of a turbulent flow. There is a
relative freedom of choosing the model describing the flow. The system of PDEs is solved by
an implicit finite volume method. The approximate inverse operator is based on a thin airfoil
theory for a potential flow, equipped with some corrections according to the model used. The
airfoil is constructed using a mean camber line and a thickness function. The so far developed
method has several restrictions. It is applicable to a subsonic pressure distribution satisfying a
certain condition for the position of a stagnation point. Numerical results are presented.
Recent advances in the field of fluid mechanics with moving fronts are linked to the use
of Level SetMethods, a versatile mathematical technique to follow free boundaries which
undergo topological changes. A challenging class of problems in this context are those related
to the solution of a partial differential equation posed on a moving domain, in which the
boundary condition for the PDE solver has to be obtained from a partial differential equation
defined on the front. This is the case of potential flow models with moving boundaries.
Moreover, the fluid front may carry some material substance which diffuses in the front and is
advected by the front velocity, as for example the use of surfactants to lower surface tension.
We present a Level Set based methodology to embed this partial differential equations
defined on the front in a complete Eulerian framework, fully avoiding the tracking of fluid
particles and its known limitations. To show the advantages of this approach in the field of
Fluid Mechanics we present in Chapter 13 one particular application: the numerical
approximation of a potential flow model to simulate the evolution and breaking of a solitary
wave propagating over a slopping bottom and compare the level set based algorithm with
previous front tracking models.
Preface xiii
Chapter 14 is basically a review of recent works that is aimed at putting forward the main
ideas behind a new theoretical approach to turbulent wallbounded flows, notably pipeflows,
in which complex physics is involved, such as combustion or particle transport. Pipe flows
are ubiquitous in industrial applications and have been studied intensively in the last century,
both from a theoretical and experimental point of view. The result of such a strong effort is a
good comprehension of the physics underlying the dynamics of these flows and the
proposition of reliable models for simple turbulent pipeflows at large Reynolds number
Nevertheless, the advancing of engineering frontiers casts a growing demand for models
suitable for the study of more complex flows. For instance, the motion and the interaction
with walls of aerosol particles, the presence of roughness on walls and the possibility of drag
reduction through the introduction of few complex molecules in the flow constitute some
interesting examples of pipeflows with some new complex physics involved. A good
modeling approach to these flows is yet to come and, in the commentary, we support the idea
that a new angle of attack is needed with respect to present methods. In this article, we
analyze which are the fundamental features of complex twophase flows and we point out that
there are two key elements to be taken into account by a suitable theoretical model: 1) These
flows exhibit chaotic patterns; 2) The presence of instantaneous coherent structures radically
change the flow properties. From a methodological point of view, two main theoretical
approaches have been considered so far: the solution of equations based on first principles
(for example, the NavierStokes equations for a single phase fluid) or Eulerian models based
on constitutive relations. In analogy with the language of statistical physics, we consider the
former as a microscopic approach and the later as a macroscopic one. We discuss why we
consider both approaches unsatisfying with regard to the description of general complex
turbulent flows, like twophase flows. Hence, we argue that a significant breakthrough can be
obtained by choosing a new approach based upon two main ideas: 1) The approach has to be
mesoscopic (in the middle between the microscopic and the macroscopic) and statistical; 2)
Some geometrical features of turbulence have to be introduced in the statistical model. We
present the main characteristics of a stochastic model which respects the conditions expressed
by the point 1) and a method to fulfill the point 2). These arguments are backed up with some
recent numerical results of deposition onto walls in turbulent pipeflows. Finally, some
perspectives are also given.
In: Fluid Mechanics and Pipe Flow ISBN: 9781607410379
Editors: D. Matos and C. Valerio, pp. 126 © 2009 Nova Science Publishers, Inc.
Chapter 1
SOLUTE TRANSPORT, DISPERSION, AND SEPARATION
IN NANOFLUIDIC CHANNELS
Xiangchun Xuan
*
Department of Mechanical Engineering, Clemson University,
Clemson, SC 296340921, USA
Abstract
Recent breakthrough in nanofabrication has stimulated the interest of solute separation in
nanofluidic channels. Since the hydraulic radius of nanochannels is comparable to the
thickness of electric double layers, the enormous electric fields inherent to the latter generate
transverse electromigrations causing chargedependent solute distributions over the channel
crosssection. As a consequence, the nonuniform fluid flow through nanochannels yields
chargedependent solute speeds enabling the separation of solutes by charge alone. In this
chapter we develop a theoretical model of solute transport, dispersion and separation in
electroosmotic and pressuredriven flows through nanofluidic channels. This model provides a
basis for the optimization of solute separation in nanochannels in terms of selectivity and
resolution as traditionally defined.
1. Introduction
Solute transport and separation in microcolumns (e.g., micro capillaries and chipbased
microchannels) have been a focus of research and development in electrophoresis and
chromatography communities for many years. Recent breakthrough in nanofabrication has
initiated the study of these topics among others in nanofluidic channels [13]. Since the
hydraulic radius of nanochannels is comparable to the thickness of electric double layers
(EDL), the enormous electric fields inherent to the latter generate transverse electromigrations
causing chargedependent solute distributions over the channel crosssection [47]. As a
consequence, the nonuniform fluid flow in nanochannels yields chargedependent solute
speeds enabling the separation of solutes by charge alone [8,9]. Such chargebased solute
*
Email address: xcxuan@clemson.edu. Tel: (864) 6565630. Fax: (864) 6567299
Xiangchun Xuan 2
separation was first proposed and implemented by Pennathur and Santiago [10] and Garcia et
al. [11] in electroosmotic flow through nanoscale channels, termed nanochannel
electrophoresis. As a matter of fact, this separation may also happen in pressuredriven flow
along nanoscale channels, termed here as nanochannel chromatography for comparison,
which was first demonstrated theoretically by Griffiths and Nilson [12] and Xuan and Li [13],
and later experimentally verified by Liu’s group [14].
So far, a number of theoretical studies have been conducted on the transport [46,9
13,15,19], dispersion [7,9,12,1519] and separation [4,913,15,19] of solutes in free solutions
through nanofluidic channels. This chapter combines and unites the works from ourselves in
this area [6,13,15,1719], and is aimed to develop a general analytical model of solute
transport, dispersion and separation in nanochannels. It is important to note that this model
applies only to pointlike solutes. For those with a finite size, one must consider the
hydrodynamic and electrostatic interactions among solutes, electric field, and flow fluid, and
as well the Steric interactions between solutes and channel walls etc [20].
2. Nomenclature
a channel halfheight
B
i
defined function, = exp(−z
i
Ψ)
c
b
bulk concentration of the background electrolyte
c
i
concentration of solute species i
C
i
bulk concentration of solute species i
c
i,0
concentration of solute species i at the channel centerline
C
i,0
initial concentration of solute species i at the channel centerline
D
i
solute diffusion coefficient
i
D′ effective solute diffusion coefficient
E axial electric field
E
st
streaming potential field
F Faraday’s constant
h
i
reduced theoretical plate height
j electric current density
K
i
hydrodynamic dispersion
L channel length
P hydrodynamic pressure drop per unit channel length
Pe Peclet number
r
ji
solute selectivity
R Universal gas constant
R
ji
resolution
t time coordinate
T absolute temperature
u
i
axial solute speed
i
u mean solute speed
v
i
solute mobility
W
i
half width of the initially injected solute zone
Solute Transport, Dispersion, and Separation in Nanofluidic Channels 3
x streamwise or longitudinal coordinate
X
i
the central location of the injected solute zone
y transverse coordinate
z
i
valence of ions
Z electrokinetic “figure of merit”
Greek Symbols
β nondimensional product of fluid properties, = λ
b
μ/εRT
ε permittivity
γ apparent viscosity ratio
κ reciprocal of Debye length
χ
i
dispersion coefficient
λ
b
molar conductivity of the background electrolyte
μ fluid viscosity
ψ electrical double layer potential
Ψ nondimensional EDL potential
Ψ
0
EDL potential at the channel center
ρ
e
net charge density
σ
b
bulk electric conductivity of background electrolyte
σ
i
standard deviation of solute peak distribution
σ
t
standard deviation of solute peak distribution in the time domain
ζ zeta potential
ζ
*
nondimensional zeta potential
Subscripts
e electroosmosis related
p pressuredriven related
i solute species i
3. Fluid Flow in Nanochannels
Given the fact that the width (in micrometers) of stateoftheart nanofluidic channels is
usually much larger than their depth (in nanometers) [13], we consider the solute transport in
fluid flow through a long straight nanoslit, see Figure 1 for the schematic. The flow may be
electric fielddriven, i.e., electroosmotic, or pressuredriven. For simplicity, the electrolyte
solution is assumed symmetric with unitcharge, e.g., KCl. As the time scale for fluid flow (in
the order of nanoseconds) is far less than that of solute transport (typically of tens of
seconds), we assume a steadystate, fullydeveloped incompressible fluid motion, which in a
slit channel is governed by
Xiangchun Xuan 4
x
y
u + v
i
z
i
FE a u
Solute zone
Figure 1. Schematic of solute transport in a slit nanochannel (only the top half is illustrated due to the
symmetry).
2
2
0
e
d u
P E
dy
μ ρ + + = (31)
where μ is the fluid viscosity, u the axial fluid velocity, y the transverse coordinate originating
from the channel axis, P the pressure drop per unit channel length, and E the axial electric
field either externally applied in electroosmotic flow or internally induced in pressuredriven
flow (i.e., the socalled streaming potential field) [2124]. The net charge density, ρ
e
, is
solved from the Poisson equation [25]
2
2 e
d
dy
ψ
ρ ε = − (32)
where ε is the fluid permittivity and ψ is the EDL potential. Invoking the noslip condition for
Eq. (31) and the zeta potential condition for Eq. (32) on the channel wall (i.e., y = a), one
can easily obtain
p e
u u u = + (33)
2 2
2
1
2
p
a y
u P
a μ
⎛ ⎞
= −
⎜ ⎟
⎝ ⎠
(34)
1
e
u E
εζ
μ ζ
∗
⎛ ⎞ Ψ
= − −
⎜ ⎟
⎝ ⎠
(35)
where u
p
is the pressuredriven fluid velocity, u
e
the electroosmotic fluid velocity, a the half
height of the channel, and Ψ = Fψ/RT and ζ
*
= Fζ/RT the dimensionless forms of the EDL
and wall zeta potentials with F the Faraday’s constant, R the universal gas constant and T the
absolute fluid temperature. It is noted that the contribution of charged solutes to the net
charge density ρ
e
has been neglected. This is reasonable as long as the solute concentration is
much lower than the ionic concentration of the background electrolyte, which is fulfilled in
typical solute separations. Under such a condition, it is also safe to assume a uniform zeta
potential on the channel wall.
Solute Transport, Dispersion, and Separation in Nanofluidic Channels 5
The nondimensional EDL potential in Eq. (35), Ψ, may be solved from the Poisson
Boltzmann equation [25]
( )
2
2
2
sinh
d
dy
κ
Ψ
= Ψ (36)
where
2
2
b
F c RT κ ε = is the inverse of the socalled Debye screening length with c
b
the
bulk concentration of the background electrolyte. We recognize that the assumed Boltzmann
distribution of electrolyte ions in PoissonBoltzmann equation might be questionable in
nanoscale channels, especially in those with strong EDL overlapping [26,27]. However, this
equation has been successfully used to explain the experimentally measured electric
conductance and streaming current in variable nanofluidic channels [2832], and is thus still
employed here.
For the case of a small magnitude of ζ (e.g., ζ < 25 mV or ζ
*
 < 1) which is actually
desirable for sensitive solute separations in nanochannels as demonstrated by Griffiths and
Nilson [12], one may use the DebyeHuckel approximation to simplify Eq. (36) as [21,25]
2
2
2
d
dy
κ
Ψ
= Ψ (37)
It is then straightforward to obtain
( )
( )
*
cosh
cosh
y
a
κ
ζ
κ
Ψ = (38)
where κa may be viewed as the normalized channel halfheight. It is important to note that for
a given fluid and channel combination, the wall zeta potential will in general vary with κa
[28,31,32]. One option to address this is to use a surfacecharge based potential parameter for
scaling instead of zeta potential [16]. In this work and other studies [614], the zeta potential
is used directly, because it may be readily determined through experiment and provides a
direct measure of the electroosmotic mobility.
The areaaveraged fluid velocity u may be written in terms of the Poiseuille and
electroosmotic components
e p
u u u + = (39)
P
a
u
p
μ 3
2
= (310)
Xiangchun Xuan 6
( ) tanh
1
e
a
u E
a
κ
εζ
μ κ
⎡ ⎤
= − −
⎢ ⎥
⎣ ⎦
(311)
where ( ) ( )
0
a
d y a =
∫
signifies an areaaveraged quantity.
3.1. Electroosmotic Flow
For electroosmotic flow, no pressure gradient is present, and so the fluid motion in Eq.
(33) is described by
0
p
u = and
( )
( )
cosh
1
cosh
e
y
u E
a
κ
εζ
μ κ
⎡ ⎤
= − −
⎢ ⎥
⎣ ⎦
(312)
0
0.2
0.4
0.6
0.8
1
0 0.2 0.4 0.6 0.8 1
E
l
e
c
t
r
o
o
s
m
o
t
i
c
v
e
l
o
c
i
t
y
p
r
o
f
i
l
e
y/a
κa = 1
2
5
10
20
Figure 2. Radial profile of the normalized electroosmotic fluid velocity, u
e
/U
HS
= 1 − cosh(κy)/cosh(κa),
at different κa values. All symbols are referred to the nomenclature.
Figure 2 shows the profile of electroosmotic velocity normalized by the socalled Helmholtz
Smoluchowski velocity U
HS
= −εζE/μ [25], i.e., u
e
/U
HS
= 1 − cosh(κy)/cosh(κa), in a slit
channel with different κa values. When κa > 10, the curves are almost pluglike except near
the channel wall and the bulk velocity is equal to U
HS
. These are the typical features of
electroosmotic flow when there is little or zero EDL overlapping. The profiles at κa < 5
become essentially parabolic resembling the traditional pressuredriven flow. Moreover, the
maximum velocity along the channel centerline is significantly lower than U
HS
and decreases
with κa, indicating a vanishingly small electroosmotic mobility when κa approaches 0.
Solute Transport, Dispersion, and Separation in Nanofluidic Channels 7
3.2. PressureDriven Flow
For pressuredriven flow, the downstream accumulation of counterions results in the
development of a streaming potential field [2124]. This induced electric field, E
st
, can be
determined from the condition of zero electric current though the channel. If an equal
mobility for the positive and negative ions of the electrolyte is assumed, the electric current
density, j, in pressuredriven flow is given as [2124]
( ) cosh
e b st
j u E ρ σ = + Ψ (313)
where σ
b
= c
b
λ
b
is the bulk conductivity of the electrolyte with λ
b
being the molar
conductivity. Referring to Eqs. (32) and (33), one may rewrite the last equation as
( ) ( )
2
2
cosh
p e b b st
d RT
j u u c E
dy F
ε λ
Ψ
= − + + Ψ (314)
Note that the surface conductance of the outer diffusion layer in the EDL has been considered
through the cosine hyperbolic function in Eq. (314) (which reduces to 1 at Ψ = 0). The
contribution of the inner Stern layer conductance [33] to the electric current is, however,
ignored. Readers may be referred to Davidson and Xuan [34] for a discussion of this issue in
electrokinetic streaming effects.
Integrating j in Eq. (314) over the channel crosssection and using the zero electric
current condition in a steadystate pressuredriven flow yield
( )
*
1
*2
2 3
st
b
g
E P
c F g g
ζ
β ζ
=
+
(315)
( )
a
a
g
κ
κ tanh
1
1
− = ,
( )
( ) a a
a
g
κ κ
κ
2 2
cosh
1 tanh
− = and ( )
∫
⎟
⎠
⎞
⎜
⎝
⎛
Ψ =
a
a
y
d g
0
3
cosh (316)
Therefore, the fluid motion in pressuredriven flow is characterized as
2 2
2
1
2
p
a y
u P
a μ
⎛ ⎞
= −
⎜ ⎟
⎝ ⎠
and
( )
( )
cosh
1
cosh
e st
y
u E
a
κ
εζ
μ κ
⎡ ⎤
= − −
⎢ ⎥
⎣ ⎦
(317)
Areaaveraging the two velocity components in the last equation and combining them
with Eq. (315) lead to
Z
u
u
p
e
− = (318)
Xiangchun Xuan 8
where Z is previously termed electrokinetic “figure of merit” as it gauges the efficiency of
electrokinetic energy conversion [35,36], and defined as
( ) ( )
2
1
2
2
2 3
3g
Z
a g g κ β ζ
∗
=
+
(319)
RT
b
ε
μ λ
β = (320)
where κ
2
= 2F
2
c
b
/εRT has been invoked during the derivation, and β is a nondimensional
product of fluid properties whose reciprocal was termed Levine number by Griffiths and
Nilson [37]. Apparently, Z depends on three nondimensional parameters, β, κa and ζ
*
,
among which β spans in the range of 2 ≤ β ≤ 10 and ζ
*
spans in the rage of −8 ≤ ζ
*
≤ 0 [33]
for typical aqueous solutions. Moreover, Z is unconditionally positive and less than unity due
to the entropy generation in nonequilibrium electrokinetic flow [38]. The curves of Z at ζ
*
=
−1 (or ζ ≈ −25 mV) and β = 2 and 10, respectively, are displayed in Figure 3 as a function of
κa. One can see that Z achieves the maximum at around κa = 2, indicating that nanochannels
with a strong EDL overlapping are the necessary conditions for efficient electrokinetic energy
conversion. For more information about Z and its function in electrokinetic energy
conversion, the reader is referred to Xuan and Li [36].
0.00
0.03
0.06
0.09
0.12
0.15
0.1 1 10 100
Z
κa
β = 2
β = 10
Figure 3. The electrokinetic “figure of merit” as a function of κa at ζ
*
= −1 and β = 2 and 10,
respectively.
Solute Transport, Dispersion, and Separation in Nanofluidic Channels 9
Based on Eq. (318), the effects of streaming potential on the average fluid speed, i.e.,
u in Eq. (39), in an otherwise pure pressuredriven flow, are characterized by
1
p
u
Z
u
= − (321)
This equation also provides a measure of the socalled electroviscous effects in
micro/nanochannels [39]. If the concept of apparent viscosity is employed to characterize
such retardation effects, the apparent viscosity ratio γ is readily derived as
( ) 1 1 Z γ = − (322)
For more information on this topic, the reader is referred to Li [39] and Xuan [40].
4. Solute Transport in Nanochannels
Solute transport in nanochannels is governed by the NernstPlanck equation in the
absence of chemical reactions, which under the assumption of fullydeveloped fluid flow is
written as
2 2
2 2
i i i i
i i i i i i
c c c c
u D D v z F c
t x x y y y
ψ ⎛ ⎞ ∂ ∂ ∂ ∂ ∂ ∂
+ = + +
⎜ ⎟
∂ ∂ ∂ ∂ ∂ ∂
⎝ ⎠
(41)
i e p i i
u u u v z FE = + + (42)
where c
i
is the concentration of solute species i, t the time coordinate, u
i
the local solute speed
(a combination of electroosmosis, pressuredriven motion, and electrophoresis), x the axial
coordinate originating from the channel inlet (see Figure 1), D
i
the molecular diffusion
coefficient, v
i
the solute mobility, and z
i
the solute charge number. Note that the product v
i
z
i
F
represents the solute electrophoretic mobility. Integrating Eq. (4.1) over the channel cross
section eliminates the last two terms on the right hand side due to the impermeable wall
conditions
2
2
i i i i
i
c u c c
D
t x x
∂ ∂ ∂
+ =
∂ ∂ ∂
(43)
where again ... indicates the areaaverage over the channel crosssection as defined earlier.
Since the time scale for transverse solute diffusion in nanochannels (characterized by
a
2
/D
i
, which is about 100 μs when a = 100 nm and D
i
= 1×10
−10
m
2
/s) is much shorter than
Xiangchun Xuan 10
that for longitudinal solute transport (typically of tens of seconds), it is reasonable to assume
that solute species are at a quasisteady equilibrium in the y direction, i.e.,
2
2
0
i
i i i i
c
D v z F c
y y y
ψ ⎛ ⎞ ∂ ∂ ∂
= +
⎜ ⎟
∂ ∂ ∂
⎝ ⎠
(44)
Integrating Eq. (4.4) twice and using the NernstEinstein relation [33], v
i
= D
i
/RT, one obtains
( ) ( ) ( )
,0 0
, , , exp
i i i
c x y t c x t z = − Ψ − Ψ ⎡ ⎤
⎣ ⎦
(45)
where c
i,0
is the solute concentration at the channel centerline where the local EDL potential
is defined as Ψ
0
(nonzero in the presence of EDL overlapping). Substituting Eq. (4.5) into
Eq. (4.3) and considering the hydrodynamic dispersion due to the velocity nonuniformity
over the channel crosssection [41,42], one may obtain
2
,0 ,0 ,0
2
i i i
i i
c c c
u D
t x x
∂ ∂ ∂
′ + =
∂ ∂ ∂
(46)
i ip ie i i
u u u v z FE = + + (47)
p i
ip
i
u B
u
B
= and
e i
ie
i
u B
u
B
= (48)
where
i
u is the mean solute speed (i.e., zone velocity) with
ip
u and
ie
u being its components
due to pressuredriven and electroosmotic flows,
i
D′ the effective diffusion coefficient which
is a combination of molecule diffusion and hydrodynamic dispersion and will be addressed in
the next section, and B
i
= exp(−z
i
Ψ) the likeBoltzmann distribution of solutes in the cross
stream direction. It is the dependence of
i
u on the charge number z
i
that enables the charged
based solute separation in nanochannels.
For an initially uniform concentration C
i,0
of solute species i along the channel axis, a
closedform solution to Eq. (46) is given by [5]
,0
,0
erf erf
2 2 2
i
i i i i i
i
i i
C
W x X u t x X u t
c
Dt Dt
⎡ ⎤ ⎛ ⎞ ⎛ ⎞
− + + − −
= + ⎢ ⎥ ⎜ ⎟ ⎜ ⎟
⎜ ⎟ ⎜ ⎟
′ ′
⎢ ⎥
⎝ ⎠ ⎝ ⎠ ⎣ ⎦
(49)
where erf denotes the error function, and W
i
is the half width of the initial solute zone with its
center being located at X
i
. As such, the electrokinetic transport of solute species in
nanochannels is described by
Solute Transport, Dispersion, and Separation in Nanofluidic Channels 11
( ) ( ) , , erf erf exp
2 2 2
i i i i i i
i i
i i
C W x X u t x X u t
c x y t z
Dt Dt
⎡ ⎤ ⎛ ⎞ ⎛ ⎞
− + + − −
= + − Ψ ⎢ ⎥ ⎜ ⎟ ⎜ ⎟
⎜ ⎟ ⎜ ⎟
′ ′
⎢ ⎥
⎝ ⎠ ⎝ ⎠ ⎣ ⎦
(410)
where ( )
,0 0
exp
i i i
C C z = Ψ is the bulk solute concentration at zero potential outside the slit
nanochannel, refer to Eq. (45).
z
v
=+2
z
v
=+1
z
v
=0
z
v
=−1
z
v
=−2
C
max
0.8 C
max
0.6 C
max
0.4 C
max
0.2 C
max
Figure 4. Transport of solutes with z
i
= [+2, −2] through a 100 nm deep channel in nanochannel
chromatography. Other parameters are referred to the text. Reprinted with permission from [13].
Figure 4 illustrates the transport of an initially W
i
= 1 μm wide plug of solutes with z
i
=
[+2, −2] (from top to bottom) through a 100 nm deep (i.e., a = 50 nm) channel 5 s after a
pressure gradient P = 1×10
8
Pam
1
was imposed. Note that only the top half of the channel is
shown due to symmetry. The ionic concentration of the background electrolyte is c
b
= 1 mM,
corresponding to κa ≈ 5. The other two nondimensional parameters are assumed to be β = 4
and ζ
*
= −2 (or ζ = −50 mV), both of which are typical to aqueous solutions as indicated
above. As to the validity of the DebyeHuckel approximation at ζ
*
= −2, we have recently
demonstrated using numerical simulation the fairly good accuracy of Eq. (37) in predicting
the solute migration velocity [5]. As shown, positive solutes are concentrated to near the
negatively charged wall due to the solutewall electrostatic interactions [9,11], or in essence
the transverse electromigration in response to the induced EDL field [5,10,12]. Moreover, the
higher the charge number is, the closer the solutes are to the walls. As the fluid velocity near
noslip walls is slower than its average, positive solutes move slower than neutral solutes that
are still uniformly distributed over the channel crosssection. Conversely, negative solutes are
repelled by the negatively changed walls and concentrated to the region close to the channel
center. Hence, they move faster than neutral solutes as seen in Figure 4.
As the fluid velocity profile is available in Eq. (312) for electroosmotic flow and in Eq.
(317) for pressuredriven flow, the mean speed of solutes,
i
u , is readily obtained from Eq.
(47). Figure 5 compares the mean speed of solutes with z
i
= [−2, +2] in electroosmotic flow
with an electric field of 4 kV/m. One can see that
i
u of all five solutes decreases when κa
gets smaller. This reduction may be explained by the overall lower electroosmotic velocity at
a smaller κa, as demonstrated in Figure 2. When κa > 100, negatively charged solutes move
slower than positive ones due to their opposite electrophoresis to fluid electroosmosis
Xiangchun Xuan 12
(identical to the curve with z
i
= 0). When κa gets smaller than 100, however, negatively
solutes start moving faster than positively solutes as the latter ones are concentrated to the
EDL region within which the fluid has a slower speed than the bulk as explained above. The
relative magnitude of
i
u between the solutes of like charges is, however, a complex function
of both the charge number z
i
, which determines the velocity component due to fluid flow, i.e.,
ip ie
u u + in Eq. (47), and the solute mobility v
i
, which determines the velocity component
due to solute electrophoresis, i.e., the most right term in Eq. (47). When κa further decreases
to less than 1, the EDL potential becomes nearly flat due to the strong EDL overlapping (see
Figure 2), and so the order of
i
u for the three solutes at large κa (i.e., microchannel
electrophoresis) is recovered.
Figure 5. Comparison of the mean speeds of solutes with z
i
= [−2, +2] as a function of κa in
nanochannel electrophoresis. The solute diffusivity is assumed to be constant, D
i
= 5×10
−11
m
2
/s. Other
parameters are referred to the text.
Combining Eqs. (47) and (48) provides a measure of the streaming potential effects on
the solute mean speed in pressuredriven flow,
1
e i i i i st i
ip p i
u B v z F B E u
u u B
+
= + (411)
It is obvious that the last equation is reduced to Eq. (321) for neutral solutes, i.e., z
i
= 0 and
thus B
i
= 1. Figure 6 shows the ratio
i ip
u u as a function of κa. As expected, streaming
Solute Transport, Dispersion, and Separation in Nanofluidic Channels 13
potential effects reduce the solute speed due to the induced electroosmotic backflow. This
reduction varies with the solute charge z
i
and attains the extreme at about κa = 3, where the
figure of merit Z (refer to Figure 3) approaches its maximum indicating the largest streaming
potential effects. In both the high and low limits of κa, streaming potential effects become
negligible, i.e., Z → 0, see Figure 3. Accordingly,
i ip
u u reduces to 1 for all solutes at large
κa while varying with z
i
at small κa because of the finite solute mobility [15].
0.7
0.8
0.9
1
1.1
0.1 1 10 100
u
i
/
u
i
p
κa
z
i
= +2
+1
0
−2
−1
Figure 6. Effects of streaming potential on the solute mean speed in nanochannel chromatography.
5. Solute Dispersion in Nanochannels
As u
p
and u
e
vary over the channel crosssection (refer to Eqs. (34) and (35), and Figure
2), they both contribute to the spreading of solutes along the flow direction, which is termed
hydrodynamic dispersion or Taylor dispersion [41,42]. The general formula for calculating
this dispersion is given by [43,44]
( )
2
1
2
0
y
i i i i
i
i i
B B u u dy
a
K
D B
−
⎡ ⎤
′ −
⎢ ⎥
⎣ ⎦
=
∫
(51)
Referring to Eqs. (42) and (47), one may then rewrite the last equation as
( )
2
2
2
i ip p ipe p e ie e
i
a
K F u F u u F u
D
= + + (52)
Xiangchun Xuan 14
( )
2
1
1
0
y
ip i i p ip i
F B B u u dy B
−
− ∗ ∗
⎡ ⎤
′ = −
⎢ ⎥
⎣ ⎦
∫
(53a)
( )
2
1
1
0
y
ie i i e ie i
F B B u u dy B
−
− ∗ ∗
⎡ ⎤
′ = −
⎢ ⎥
⎣ ⎦
∫
(53b)
( ) ( )
1
1
0 0
2
y y
ipe i p ip i e ie i i
F B u u Bdy u u Bdy B
−
− ∗ ∗ ∗ ∗
⎡ ⎤ ⎡ ⎤
′ ′ = − −
⎢ ⎥ ⎢ ⎥
⎣ ⎦ ⎣ ⎦
∫ ∫
(53c)
where
m m m
u u u
∗
= and
im im m
u u u
∗
= (m = p and e). Note that the three terms, F
ip
, F
ipe
and F
ie
in Eq. (52) represent the contributions to dispersion due to the pressuredriven flow,
the coupling between pressuredriven and electroosmotic flows, and the electroosmotic flow,
respectively.
Hydrodynamic dispersion is often expressed in terms of a nondimensional dispersion
coefficient χ
i
[45],
i i i i
D Pe K
2
χ = (54)
where the Peclet number Pe
i
in this case may be defined with respect to the mean solute
speed, i.e.,
i i i
D a u Pe = [12,19], or to the areaaveraged fluid velocity, i.e.,
( )
i e p i
D a u u Pe + = [1518,45]. Using the solute speedbased Peclet number, χ
i
becomes dependent on the solute diffusivity D
i
which complicates the analysis. This is
because the solute mobility v
i
in
i
u [see Eq. (47)] is coupled to D
i
via the NernstEinstein
relation, v
i
= D
i
/RT. Such dependence doesn’t occur if χ
i
is defined using the fluid velocity
based Peclet number. Here, we employ the latter definition in keeping with the dispersion
studies of neutral solutes in the literature [45]. As such, the dispersion coefficient χ
i
may be
easily obtained from Eq. (52) as
( )
2
2
2
ip p ipe p e ie e
i
p e
F u F u u F u
u u
χ
+ +
=
+
(55)
It is important to note that χ
i
is independent of the solute speed or the driving force of the
flow while K
i
(in the unit of D
i
) not. Instead, χ
i
is primarily determined by the flow type
(pressure or electric fielddriven), channel structure (including shape and size) and solute
charge number.
Solute Transport, Dispersion, and Separation in Nanofluidic Channels 15
5.1. Electroosmotic Flow
In electroosmotic flow, the hydrodynamic dispersion in Eq. (52) is reduced to
2
2
e
i ie
i
a u
K F
D
= (56)
Accordingly, the dispersion coefficient in Eq. (5.5) is simplified as
i ie
F χ = (57)
0.0001
0.001
0.01
0.1
0.1 1 10 100
χ
i
f
o
r
n
a
n
o
c
h
a
n
n
e
l
e
l
e
c
t
r
o
p
h
o
r
e
s
i
s
κa
z
i
= 0
+1
+2
−1
−2
Figure 7. Illustration of dispersion coefficient χ
i
of solutes with z
i
= [−2, +2] in nanochannel
electrophoresis as a function of κa.
Figure 7 shows χ
i
of solutes with z
i
= [−2, +2] in nanochannel electrophoresis as a
function of κa. We see that in the entire range of κa, χ
i
of positive solutes is larger than that
of neutral ones while χ
i
of negative solutes is smaller than the latter. This is because positive
solutes are concentrated to near the channel walls where the velocity gradients are large while
negative ones are concentrated to the channel centerline where the velocity gradients are
small (refer to Figure 4). Moreover, the higher the charge number z
i
, the larger is χ
i
for
positive solutes and the smaller for negative ones. In the low limit of κa (i.e., the narrowest
channel), the EDL potential is essentially uniform over the channel crosssection (refer to
Figure 2), and so is the solute distribution regardless of the charge number. As a consequence,
χ
i
of all charged solutes approach that of neutral solutes, i.e., 2/105. Note that this value is
equal to the dispersion coefficient of neutral solutes in a pure pressuredriven flow indicating
the resemblance between pressuredriven and electroosmotic flows in very small
Xiangchun Xuan 16
nanochannels. This aspect will be revisited shortly. In the high limit of κa (i.e., the widest
channel), the EDL thickness is so thin compared to the channel height that the solute
distribution becomes once again uniform across the channel (the EDL potential is, here,
uniformly zero while equal to the wall zeta potential in the low limit of κa). Therefore, the
hydrodynamic dispersion in electroosmotic flow, or the socalled electrokinetic dispersion
[46], decreases with the square of κa and ultimately converges to zero [47,48].
5.2. PressureDriven Flow
In pressuredriven flow with consideration of streaming potential, the hydrodynamic
dispersion is obtained from Eq. (52) as
( )
2
2
2
2 3
1
p
i ip i i
i
a u
K F Z Z
D
δ δ = − + (58)
2 i ipe ip
F F δ = and
3 i ie ip
F F δ = (59)
during which Eqs. (317) and (318) have been invoked and Z is the electrokinetic “figure of
merit” as defined in Eq. (319). It is apparent that the streaming potential induced
electroosmotic backflow produces two additional dispersions in pressuredriven flow: one is
the electrokinetic dispersion due to electroosmotic flow itself, the term with δ
i3
in Eq. (58),
which tends to increase the total dispersion, and the other is due to the coupling of pressure
driven and electroosmotic flows, the term with δ
i2
in Eq. (58), which tends to decrease the
total dispersion. The latter phenomenon has been employed previously to reduce the
hydrodynamic dispersion in capillary electrophoresis where a pressuredriven backflow is
intentionally introduced to partially compensate the nonuniformity in electroosmotic velocity
profile [49,50]. If streaming potential effects are ignored, i.e., for a pure pressuredriven flow
with Z = 0, Eq. (58) reduces to
2
2
p
ip ip
i
a u
K F
D
= (510)
such that
2
2 3
1
i
i i
ip
K
Z Z
K
δ δ = − + (511)
Similarly, the dispersion coefficient in Eq. (55) for real pressuredriven flow is obtained
as
( )
2
2 3
2
1
1
i i
i ip
Z Z
F
Z
δ δ
χ
− +
=
−
(512)
Solute Transport, Dispersion, and Separation in Nanofluidic Channels 17
where F
ip
= χ
ip
is the dispersion coefficient for a pure pressuredriven flow by analogy to Eq.
(57) in a pure electroosmotic flow. We thus have
( )
2
2 2 3
2
1
1
i i i i
ip ip
Z Z K
K
Z
χ δ δ
γ
χ
− +
= =
−
(513)
Therefore, χ
i
/χ
ip
differs from K
i
/K
ip
by only the square of the apparent viscosity ratio γ, see
the definition in Eq. (322). As γ is independent of the solute charge number z
i
, it is expected
that the variation of χ
i
/χ
ip
with respect to z
i
will be identical to that of K
i
/K
ip
.
0.001
0.01
0.1
1
0.1 1 10 100
χ
i
f
o
r
n
a
n
o
c
h
a
n
n
e
l
c
h
r
o
m
a
t
o
g
r
a
p
h
y
κa
z
i
= 0
+1
+2
−1
−2
Figure 8. Dispersion coefficient χ
i
of solutes with z
i
= [−2, +2] in nanochannel chromatography as a
function of κa.
Figure 8 shows χ
i
of solutes with z
i
= [−2, +2] as a function of κa in nanochannel
chromatography. Due to the same reason as stated above for nanochannel electrophoresis, χ
i
of positive solutes is larger than that of neutral ones while χ
i
of negative solutes is the
smallest. In both the high and low limits of κa, the flowinduced streaming potential is
negligible, see Eq. (318) and Figure 3. Hence, the electroosmotic back flow and the induced
solute electrophoresis vanish, yielding χ
i
= 2/105 regardless of the solute charge [51]. It is
important to note that χ
i
of neutral solutes in pressuredriven flow is not uniformly 2/105 as
accepted in the literature. Due to the effects of flowinduced streaming potential, χ
i
is
increased by the electroosmotic backflow [i.e., δ
i3
Z
2
term in Eq. (512)] even though the
coupled dispersion term [i.e., −δ
i2
Z term in Eq. (512)] drops for neutral solutes [17].
Xiangchun Xuan 18
0.75
0.8
0.85
0.9
0.95
1
1.05
1.1
1.15
0.1 1 10 100
K
i
/
K
i
p
o
r
χ
i
/
χ
i
p
κa
z
i
z
i
K
i
/K
ip
χ
i
/χ
ip
.
Figure 9. Effects of streaming potential on the ratio of solute dispersion, K
i
/K
ip
, and the ratio of
dispersion coefficient, χ
i
/χ
ip
, in nanochannel chromatography as a function of κa. Adapted with
permission from [18].
Figure 9 displays the effects of streaming potential on the ratio of solute dispersion,
K
i
/K
ip
, and the ratio of dispersion coefficient, χ
i
/χ
ip
, in nanochannel chromatography as a
function of κa. In all cases, K
i
/K
ip
is less than 1 indicating that streaming potential effects
result in a decrease in hydrodynamic dispersion. This reduction, as a consequence of the
induced electroosmotic backflow, gets larger (i.e., K
i
/K
ip
deviates further away from 1) when
the solute charge z
i
increases. The optimum κa at which K
i
/K
ip
achieves its extreme increases
slightly with z
i
. In contrast to the decrease in solute dispersion, the dispersion coefficient is
increased by the effects of streaming potential, i.e., χ
i
/χ
ip
> 1. These dissimilar trends stem
from the dependence of γ on κa, see Eqs. (319), (322) and (513). As streaming potential
effects increase (or in other words, the electrokinetic “figure of merit” Z increases), the
electroosmotic backflow increases causing a decrease in K
i
/K
ip
(and K
i
/K
ip
< 1) while an
increase in γ (and γ > 1). The net result is the observed variation of χ
i
/χ
ip
with respect to κa.
The increase in χ
i
/χ
ip
is more sensitive to z
i
than the decrease in K
i
/K
ip
. However, the trend
that χ
i
/χ
ip
varies with respect to z
i
is consistent with K
i
/K
ip
as pointed out earlier. In addition,
χ
i
/χ
ip
attains a maximum at a larger value of κa than that at which K
i
/K
ip
is minimized.
5.3. Neutral Solutes
For neutral solutes (z
i
= 0), closedform formulae are available for the functions F
m
(m =
ip, ipe, ie) defined in Eq. (53) which in turn determine δ
i2
and δ
i3
through Eq. (59).
Specifically, we find [17]
Solute Transport, Dispersion, and Separation in Nanofluidic Channels 19
2 105
ip
F = (514)
( )
( ) ( )
2 4
2 2 6 1
1 15
ipe
F
a a
ω ω
ω ω
κ κ
⎡ ⎤
− ⎛ ⎞
= − + ⎢ ⎥
⎜ ⎟
−
⎝ ⎠
⎢ ⎥
⎣ ⎦
(515)
( ) ( ) ( ) ( ) ( )
2
2 2 2 2
2 2
1 2 3 1
3
1 2 2 cosh
ie
F
a a a a
ω
ω κ ω κ ω κ κ
⎡ ⎤
= + − − ⎢ ⎥
−
⎢ ⎥
⎣ ⎦
(516)
( ) tanh a
a
κ
ω
κ
= (517)
Figure 10 compares the magnitude of δ
i2
and δ
i3
[see their definitions in Eq. (59)] for
neutral solutes as a function of κa. As shown, δ
i2
is always larger than δ
i3
. In the low limit of
κa, δ
i2
approaches 2 while δ
i3
approaches 1, and the square bracketed terms in Eq. (52) thus
reduces to (<u
p
> + <u
e
>)
2
reflecting the similarity of pressuredriven and electroosmotic flow
profiles in very narrow nanochannels. In the high limit of κa, both δ
i2
and δ
i3
approach zero
because the streaming potential is negligible and the electroosmotic velocity profile becomes
essentially pluglike. Note that Eq. (514) gives the wellknown hydrodynamic dispersion
coefficient of neutral solutes in a pure pressuredriven flow between two parallel plates [51].
Moreover, Eq. (516) is identical to that derived by Griffiths and Nilson [47,48] which gives
the electrokinetic dispersion coefficient of neutral solutes in a pure electroosmotic flow
between two parallel plates.
0
0.5
1
1.5
2
0.1 1 10 100 1000
δ
κa
δ
i2
δ
i3
Figure 10. Plot of δ
i2
and δ
i3
for neutral solutes as a function of κa. Adapted with permission from [17].
Xiangchun Xuan 20
6. Solute Separation in Nanochannels
Solute separation is typically characterized by retention, selectivity, plate height (or plate
number), and resolution [43], of which retention and plate height are related to only one type
of solutes. In contrast, selectivity and resolution are both dependent on two types of solute
species, and thus provide a direct measure of the separation performance of solutes. As plate
height is involved in the definition of resolution, see Eq. (68), it will still be considered
below along with the selectivity and resolution for a comprehensive understanding of solute
separation in nanofluidic channels.
In order to emphasize the advantage of electrophoresis and chromatography in
nanochannels over those taking place in microcolumns, we focus on the solutes with a
similar electrophoretic mobility, or specifically, v
i
z
i
F = constant. This is equivalent to
assuming a constant chargetosize ratio or a constant product, Dz = D
i
z
i
, of solute charge
and diffusivity because solute size is inversely proportional to its diffusivity via the Nernst
Einstein relation [33]. Such solutes are unable to be separated in free solutions through
pressuredriven or electroosmotic microchannel flows. A typical value of the solute charge
diffusivity product, Dz = 1×10
−10
m
2
/s, was selected in the following demonstrations while
the solute charge number z
i
may be varied from −4 to +4. The ratio of channel length to half
height was fixed at L/a = 10
4
for convenience even though we recognize that fixing the
channel length might be a wiser option when the channel height is varied.
6.1. Selectivity
Selectivity, r
ji
, is defined as the ratio of the mean speeds of solutes i and j
j i ji
u u r = (61)
and should be larger than 1 as traditionally defined [43]. A larger r
ji
indicates a better
separation. Figure 11 compares the selectivity, r
ji
, of (a) positive and (b) negative solutes in
nanochannel chromatography (solid lines) and nanochannel electrophoresis (dashed lines). It
is important to note that the indices of r
ji
, which indicate the charge values of the two solutes
to be separated, are switched between positive and negative solutes in order that r
ji
> 1 as
traditional defined [43]. Specifically, we use r
21
, r
32
, and r
43
for positive solutes (or more
generally, solutes with z
i
ζ
*
< 0) as those with higher charges are concentrated in a region of
smaller fluid speed (i.e., closer to the channel wall) and thus move slower. Note that the
solute electrophoretic mobility has been assumed to remain unvaried. In contrast, negative
solutes (or solutes with z
i
ζ
*
> 0) with higher charges appear predominantly in the region of
larger fluid speed (closer to the channel center) and thus move faster. Therefore, we need to
use r
12
, r
23
, and r
34
for negative solutes. This index switch also applies to the resolution, R
ji
,
which will be illustrated in Figure 12.
Solute Transport, Dispersion, and Separation in Nanofluidic Channels 21
1
2
3
4
0.1 1 10 100
S
e
l
e
c
t
i
v
i
t
y
,
r
j
i
κa
(a)
r
21
r
32
r
43
r
21
r
32
r
43
1.0
1.1
1.2
0.1 1 10 100
S
e
l
e
c
t
i
v
i
t
y
,
r
j
i
κa
(b)
r
12
r
23
r
34
r
12
r
23
r
34
Figure 11: Selectivity, r
ji
, of (a) positive and (b) negative solutes in nanochannel chromatography (solid
lines) and nanochannel electrophoresis (dashed lines). Reprinted with permission from [19].
One can see in Figure 11a that the selectivity, r
ji
, of positive solutes in nanochannel
chromatography is always greater than that of the same pair of solutes in nanochannel
electrophoresis. This discrepancy gets larger when the solute charge number z
i
increases.
Meanwhile, the optimal κa value at which r
ji
is maximized increases for both chromatography
and electrophoresis though it is always smaller in the former case. The discrepancy between
Xiangchun Xuan 22
these two optimal κa also increases with the rise of z
i
. For negative solutes, Figure 11b shows
a significantly lower r
ji
than positive solutes in nanochannel chromatography. Moreover, r
ji
decreases when the solute charge number increases. The optimal κa at which r
ji
is maximized
is also smaller than that for positive solutes, and decreases (but only slightly) with z
i
. All
these results apply equally to r
ji
of negative solutes in nanochannel electrophoresis except at
around κa = 0.6 where r
ji
varies rapidly with κa. Within this region of κa, the electrophoretic
velocity of negative solutes is close to the fluid electroosmotic velocity [more accurately,
ie
u
in Eq. (47)] while in the opposite direction. Therefore, the real solute speed is essentially so
small that even a trivial difference in the solute speed (essentially the difference in
ie
u as
the solute electrophoretic velocity is constant due to the fixed chargetosize ratio) could yield
a large r
ji
.
It is, however, important to note that the speed of negative solutes could be reversed in
nanochannel electrophoresis when κa is less than a threshold value (e.g., κa = 0.6 in Figure
11b). In other words, solutes migrate to the anode side instead of the cathode side along with
the electrolyte solution. In such circumstances, it is very likely that only one of the two solute
species migrates toward the detector no matter the detector is placed in the cathode or the
anode side of the channel. Another consequence is that the maximum r
ji
in nanochannel
electrophoresis might be achieved with a fairly long analysis time, which makes the
separation practically meaningless. We therefore expect that solutes with a constant
electrophoretic mobility can be better separated in nanochannel chromatography than in
nanochannel electrophoresis. Moreover, solutes with z
i
ζ
*
< 0 can be separated more easily
than can those with z
i
ζ
*
> 0.
6.2. Plate Height
Plate height, H
i
, is the spatial variance of the solute peak distribution,
2
i
σ , divided by the
migration distance, L, within a time period of t
i
. It is often expressed in the following
dimensionless form of a reduced plate height, h
i
[42,45]
i
i i i i i
i
u a
D
aL
t D
aL a
H
h
′
=
′
= = =
2 2
2
σ
(62)
( )
2
1
i i i i i i
D D K D Pe χ ′ = + = + (63)
where
i
D′ is the effective diffusion coefficient due to a combination of hydrodynamic
dispersion K
i
[see Eq. (54)] and molecular diffusion D
i
. Note that other sources of dispersion
such as injection and detection (refer to [46,52,53] for detail) have been neglected for
simplicity.
Following Griffiths and Nilson’s analysis [12], we may combine Eqs. (62) and (63) to
rewrite the reduced plate height as
Solute Transport, Dispersion, and Separation in Nanofluidic Channels 23
( )
i i i i
Pe Pe h χ + = 1 2 (64)
Therefore, h
i
attains its minimum
i i
h χ 4
min ,
= at
i opt i
Pe χ 1
,
= (65)
In other words, there exists an optimal value for the mean solute speed,
i i opt i
a D u χ =
,
,
and thus an optimal electric field in nanochannel electrophoresis or an optimal pressure
gradient in nanochannel chromatography, at which the separation efficiency is maximized. As
h
i
is a function of solely the dispersion coefficient χ
i
that has been demonstrated in Figures 7
and 8 for nanochannel electrophoresis and chromatography, respectively, its variations with
respect to z
i
and κa are not repeated here for brevity.
6.3. Resolution
Resolution, R
ji
, can be defined in two different ways: the one introduced by Giddings
[54], i.e., Eq. (66), and the one adopted by Huber [55] and Kenndler et al. [5658], i.e., Eq.
(67),
( )
j t i t
i j
ji
t t
R
, ,
2 σ σ +
−
= (66)
i t
i j
ji
t t
R
,
σ
−
= (67)
where t is the migration time as defined in Eq. (62) and σ
t
is the standard deviation of solute
peak distribution in the time domain. Consistent with the solute selectivity r
ji
, a larger value
of R
ji
indicates a better separation. Substituting
i i
u L t = ,
j j
u L t = and
i i i t
u σ σ =
,
into
the last equation leads to
( ) ( )
1 1
ji ji ji
i i
L a L
R r r
h σ
= − = − (68)
Referring back to Eq. (65), it is straightforward to obtain
( )
,max
,min
1
ji ji
i
L a
R r
h
= − at
i opt i
Pe χ 1
,
= (69)
Xiangchun Xuan 24
because the selectivity r
ji
is independent of the solute Peclet number. Therefore, when the
plate height of one type of solute is minimized, the separation resolution of this solute from
another type of solute may reach the maximum value.
Figure 12 compares the maximum resolution, R
ji,max
, of positive and negative solutes (as
labeled) in nanochannel chromatography (solid lines) and nanochannel electrophoresis
(dashed lines). The indices of R
ji,max
are assigned following those of the selectivity, r
ji
, in
Figure 11, to ensure R
ji,max
> 0. One can see that R
ji,max
of positive solutes in chromatography
is larger than that of negative ones throughout the range of κa. In electrophoresis, the former
also yield a better resolution if κa > 1. When κa < 1, R
ji,max
of negative solutes increases and
reaches the extremes at κa = 0.6 due to the sudden rise in the selectivity (refer to Figure 11b)
as explained above. Within the same range of κa, R
ji,max
of positive solutes continues
decreasing when κa decreases and thus becomes smaller than that of negative solutes.
Interestingly, chromatography and electrophoresis offer a comparable resolution for both
types of solutes in nanoscale channels if κa > 1.
1
10
100
0.1 1 10 100
M
a
x
i
m
u
m
r
e
s
o
l
u
t
i
o
n
,
R
j
i
,
m
a
x
κa
R
32
R
43
R
21
R
21
R
32
R
43
R
34
R
12
R
23
Negative solutes
Positive solutes
Figure 12. Maximum resolution, R
ji,max
, of positive and negative solutes in nanochannel
chromatography (solid lines) and nanochannel electrophoresis (dashed lines). Reprinted with
permission from [19].
It is also noted in Figure 12 that the optimum channel size for both separation methods
appears to be 1 < κa < 10. In other words, the best channel halfheight for solute separation in
nanochannels will be 10 nm < a < 100 nm if 1 mM electrolyte solutions are used. In this
context, the optimum Peclet number to achieve the maximum resolution in a channel of κa =
5 (or a ≈ 50 nm) will be Pe
i,opt
= O(4) because h
i,min
= O(1). Although this Peclet number
(corresponding to the mean solute speed of the order of 8 mm/s) seems a little too high in
current nanofluidics, it indicates that large fluid flows are preferred in both nanochannel
Solute Transport, Dispersion, and Separation in Nanofluidic Channels 25
chromatography and nanochannel electrophoresis for high throughputs and separation
efficiencies.
7. Conclusion
We have developed an analytical model to study the transport, dispersion and separation
of solutes (both charged and noncharged) in electroosmotic and pressuredriven flows
through nanoscale slit channels. This model explains why solutes can be separated by charge
in nanochannels, and provides compact formulas for calculating the migration speed and
hydrodynamic dispersion of solutes. It also presents a simple approach to optimizing the
separation performance in nanochannels, which has been applied particularly to solutes with a
similar electrophoretic mobility. In addition, we would like to point out that the model or the
approach developed in this work can be readily extended to onedimensional round nanotubes
[12,1517,19] and to even twodimensional rectangular nanochannels [7,59,60].
References
[1] Prakash, S.; Piruska, A.; Gatimu, E. N.; Bohn, P. W. et al. IEEE Sensor. J. 2008, 8, 441
450.
[2] Abgrall, P.; and Nguyen, N. T. Anal. Chem. 2008, 80, 23262341.
[3] Schoh, R. B.; and Han, J. Y.; Renaud, P. Rev. Modern Phys. 2008, 80, 839883.
[4] Pennathur, S.; and Santiago, J. G. Anal. Chem. 2005, 77, 67726781.
[5] Petsev, D. N. J. Chem. Phys. 2005, 123, 244907.
[6] Xuan, X.; and Li, D. Electrophoresis 2006, 27, 50205031.
[7] Dutta, D. J. Colloid. Interf. Sci. 2007, 315, 740746.
[8] Pennathur, S.; Baldessari, F.; Santiago, J. G.; Kattah, M. G. et al. Anal. Chem. 2007, 79,
83168322.
[9] Das, S.; and Chakraborty, S. Electrophoresis, 2008, 29, 11151124.
[10] Pennathur, S.; and Santiago, J. G. Anal. Chem. 2005, 77, 67826789.
[11] Garcia, A. L.; Ista, L. K.; Petsev, D. N. et al. Lab Chip 2005, 5, 12711276.
[12] Griffiths, S. K.; and Nilson, R. N. Anal. Chem. 2006, 78, 81348141.
[13] Xuan, X.; and Li, D. Electrophoresis 2007, 28, 627634.
[14] Wang, X.; Kang , J.; Wang, S.; Lu, J.; and Liu, S. J Chromatography A 2008, 1200,
108113.
[15] Xuan, X. J. Chromatography A 2008, 1187, 289292.
[16] De Leebeeck, A.; and Sinton, D. Electrophoresis 2006, 27, 49995008.
[17] Xuan, X.; and Sinton, D. Microfluid. Nanofluid. 2007, 3, 723728.
[18] Xuan, X. Anal. Chem. 2007, 79, 79287932.
[19] Xuan, X. Electrophoresis 2008, 29, 37373743.
[20] Israelachvili, J. Intermolecular & Surface Forces, Academic Press, 2
nd
edition, San
Diego, CA, 1991.
[21] Burgreen, D., and Nakache, F. R., J. Phys. Chem. 1964, 68, 10841091.
[22] Rice, C. L. and Whitehead, R., J. Phys. Chem. 1965, 69, 40174024.
[23] Hildreth, D., J. Phys. Chem. 1970, 74, 20062015.
Xiangchun Xuan 26
[24] Li, D. Colloid. Surf. A 2001, 191, 3557.
[25] Hunter, R. J. Zeta potential in colloid science, principles and applications, Academic
Press, New York, 1981.
[26] Qu, W.; and Li, D. J. Colloid and Interface Sci. 2000, 224, 397407.
[27] Taylor, J.; and Ren, C. L. Microfluid. Nanofluid. 2005, 1, 356363.
[28] Stein, D.; Kruithof, M.; and Dekker, C. Phys. Rev. Lett. 2004, 93, 035901.
[29] Karnik, R.; Fan, R.; Yue, M. et al. Nano Lett. 2005, 5, 943948.
[30] Fan, R.; Yue, M.; Karnik, R. et al. Phys. Rev. Lett. 2005, 95, 086607.
[31] Van der Heyden, F. H.J.; Stein, D.; and Dekker, C. Phys. Rev. Lett. 2005, 95, 116104.
[32] Van Der Hayden, F. H. J.; Bonthius, D. J.; Stein, D.; Meyer, C.; and Dekker, C. Nano
Lett. 2007, 7, 10221025.
[33] Probstein, R. F. Physicochemical hydrodynamics, John Willey & Sons, New York,
1995.
[34] Davidson, C., and Xuan, X., Electrophoresis 2008, 29, 11251130.
[35] Morrison, F. A.; and Osterle, J. F. J. Chem. Phys. 1965, 43, 21112115.
[36] X. Xuan, and D. Li, J. Power Source, 156 (2006) 677684.
[37] Griffiths, S. K.; and Nilson, R. H. Electrophoresis 2005, 26, 351361.
[38] Xuan, X.; and Li, D J Micromech. Microeng. 2004, 14, 290298.
[39] Li, D., Electrokinetics in Microfluidics, Elsevier Academic Press, Burlington, MA 2004.
[40] Xuan, X. Microfluid. Nanofluid. 2008, 4, 457462.
[41] Taylor, G. I. Proc. Roy. Soc. London A 1953, 219, 186203.
[42] Aris, R. Proc. Roy. Soc. London A 1956, 235, 6777.
[43] Giddings, J. C. Unified separation science, John Wiley & Sons, Inc., New York, 1991.
[44] Martin, M.; Giddings, J. C. J. Phys. Chem. 1981, 85, 727733.
[45] Dutta, D.; Ramachandran, A.; and Leighton, D. T. Microfluid. Nanofluid. 2006, 2,
275290.
[46] Ghosal, S., Annu. Rev. Fluid Mech. 2006, 38, 309338.
[47] Griffiths, S. K., Nilson, R. H. Anal. Chem. 1999, 71, 55225529.
[48] Griffiths, S. K., and Nilson, R. N., Anal. Chem. 2000, 72, 47674777.
[49] Datta, R. Biotechnol. Prog. 1990, 6, 485493.
[50] Datta, R.; and Kotamarthi, V. R AICHE J. 1990, 36. 916926.
[51] Wooding, R. A., J. Fluid. Mech. 1960, 7, 501515.
[52] Gas, B.; Stedry, M.; and Kenndler, E. Electrophoresis 1997, 18, 21232133.
[53] Gas, B.; Kenndler, E. Electrophoresis 2000, 21, 38883897.
[54] Giddings, J. C. Sep. Sci. 1969, 4, 181189.
[55] Huber, J. F. K. Fresenius' Z. Anal. Chem. 1975, 277, 341347.
[56] Kenndler, E. J Cap. Elec. 1996, 3, 191198.
[57] Schwer, C.; and Kenndler, E. Chromatographia 1992, 33, 331335.
[58] Kenndler, E.; and Fridel, W. J Chromatography 1992, 608, 161170.
[59] Dutta, D. Electrophoresis 2007, 28, 45524560.
[60] Dutta, D. Anal. Chem. 2008, 80, 47234730.
In: Fluid Mechanics and Pipe Flow ISBN: 9781607410379
Editors: D. Matos and C. Valerio, pp. 2739 © 2009 Nova Science Publishers, Inc.
Chapter 2
H
2
O IN THE MANTLE: FROM FLUID
TO HIGHPRESSURE HYDROUS SILICATES
N.R. Khisina
1,*
, R. Wirth
2
and S. Matsyuk
3
1
Institute of Geochemistry and Analytical Chemistry of Russian Academy of Sciences,
Kosygin st. 19, 119991 Moscow, Russia
2
GeoForschungZentrum Potsdam, Germany
3
Institute of Geochemistry, Mineralogy and Ore Formation, National Academy of
Sciences of Ukraine, Paladin Ave., 34, 03680 Kiev142, Ukraine
Abstract
Infrared spectroscopic data show that nominally anhydrous olivine (Mg,Fe)
2
SiO
4
contains
traces of H
2
O, up to several hundred wt. ppm of H
2
O (Miller et al., 1987; Bell et al., 2004;
KochMuller et al., 2006; Matsyuk & Langer, 2004) and therefore olivine is suggested to be a
water carrier in the mantle (Thompson, 1992). Protonation of olivine during its crystallization
from a hydrous melt resulted in the appearance of intrinsic OHdefects (Libowitsky & Beran,
1995). Mantle olivine nodules from kimberlites were investigated with FTIR and TEM
methods (Khisina et al., 2001, 2002, 2008). The results are the following: (1) Water content in
xenoliths is lower than water content in xenocrysts. From these data we concluded that
kimberlite magma had been saturated by H
2
O, whereas adjacent mantle rocks had been
crystallized from waterdepleted melts. (2) Extrinsic water in olivine is represented by high
pressure phases, 10ÅPhase Mg
3
Si
4
O
10
(OH)
2
.nH
2
O and hydrous olivine
n(Mg,Fe)
2
SiO
4
.(H
2
MgSiO
4
), both of which belong to the group of Dense Hydrous Magnesium
Silicates (DHMS), which were synthesized in laboratory highpressure experiments (Prewitt
& Downs, 1999). The DHMS were regarded as possible mineral carriers for H
2
O in the
mantle; however, they were not found in natural material until quite recently. Our
observations demonstrate the first finding of the 10ÅPhase and hydrous olivine as a mantle
substance. (3) 10ÅPhase, which occurred as either nanoinclusions or narrow veins in olivine,
is a ubiquitous nanomineral of kimberlite and closely related to olivine. (4) There are two
different mechanisms of the 10ÅPhase formation: (a) purification of olivine from OH
bearing defects resulting in transformation of olivine to the 10 ÅPhase with the liberation of
water fluid; and (b) replacement of olivine for the 10ÅPhase due to hydrous metasomatism in
the mantle in the presence of H
2
O fluid.
*
Email address: khisina@geokhi.ru
N.R. Khisina, R. Wirth and S. Matsyuk 28
Introduction
The presence of water in the mantle, either as H
2
O molecules or OH

groups, has been the
subject of longterm interest in geochemistry and geophysics because of the dramatic H
2
O
influence on the melting and the physical properties of mantle rocks. Nowdays the concept of
“wet” and heterogeneous mantle is universally accepted among petrologists (Kamenetsky et
al., 2004; Sobolev & Chaussidon, 1996; Katayama et al., 2005). However, which minerals
could serve as deep water storage in the mantle is yet to be a widelydiscussed topic. The
problem is that the mantle material available for a direct investigation is very limited and
restricted by mantle nodules trapped by kimberlitic or basaltic magma from the depth. Among
the presumed candidates considered for water storage in the mantle were the socalled DHMS
phases (dense hydrous magnesium silicates) synthesized in laboratory experiments at PT
conditions of the mantle (Prewitt & Downs, 1999); however, DHMS phases have not yet been
found as macroscopic minerals in mantle material. The main mineral of the mantle is olivine
(Mg,Fe)
2
SiO
4
, which is stable at mantle pressures up to ~ 15 GPa; with increasing pressure
the olivine ( αphase) transforms to higher density structure of wadsleite (βphase). The PT
conditions of αβ transition in olivine correspond to the depth of ~ 400 km; according to
geophysical data, the mantle has a discontinuity at this depth, specified as a boundary
between the upper mantle and transition zone. Infrared spectroscopic data show that olivine
contains traces of H
2
O up to several hundreds wt. ppm of H
2
O (Bell et al., 2003; KochMüller
et al., 2006; Kurosawa et al., 1997; Matsyuk & Langer, 2004; Miller et al., 1987); therefore,
nominally anhydrous olivine is considered as water storage in the mantle (Thompson, 1992).
The highest water content, such as about 400 wt. ppm of H
2
O, was registered for olivine
samples from mantle peridotite nodules in kimberlites. Water in olivine occurs as either OH
or H
2
O. There are two modes of “water” occurrence in olivine: intrinsic and extrinsic. An
intrinsic mode represents the OH

incorporated into the olivine structure and is considered a
waterderived defect complex either associated with a metal vacancy {v
Me
, 2OH

} or by a
vacancy at the Si site {v
Si
,4OH

} (Beran & Putnis, 1983; Beran & Libovitzky, 2006; Lemaire
et al., 2004). Extrinsic water is possessed by inclusions, either solid or fluid, and occurs as
OH

or H
2
O. Recent TEM investigations of olivine nodules from Udachnaya kimberlite
(Yakuyia) revealed the nanoinclusions of hydrous magnesium silicates represented by high
pressure phases, 10ÅPhase and hydrous olivine, partially replaced by lowpressure
serpentine + talc assemblage (Wirth & Khisina, 1998; Khisina et al., 2001; Khisina & Wirth,
2002, 2008a, 2008b). 10ÅPhase and hydrous olivine belong to the group of DHMS phases.
This first finding of DHMS phases in mantle material (Wirth & Khisina, 1998; Khisina et al.,
2001; Khisina & Wirth, 2002, 2008a, 2008b) provides direct evidence of the DHMS
occurrence in the mantle and specifies them as nanominerals of nonmagmatic origin, closely
related to olivine.
The amount of water incorporated into the olivine structure depends on the PT
conditions as well as on the chemical environment and olivine composition (Fe/Mg ratio in
olivine), and increases with increasing water activity, oxygen fugacity, pressure and
temperature (Kohlstedt et al., 1996; Bai & Kohlstedt, 1993; Zhao et al., 2004). Diffusion of
hydrogen in olivine is very fast (Kohlstedt & Mackwell, 1998); therefore, due to interaction
of olivine and surrounding melt the initial water content in olivine can be changed under
changes of either P, or T, or fO
2
or fH
2
O during a postcrystallization stage. Olivine remained
H
2
O in the Mantle: From Fluid to HighPressure Hydrous Silicates 29
in a host melt can loose water due to decompression; olivine trapped by a foreign melt can
become either deprotonated or secondarily protonated, depending on whether the foreign melt
is lower or higher by water concentration (by water activity) in comparison to the host melt,
correspondingly. Experiments by Peslier and Luhr (2006) and Mosenfelder et al. (2006a,
2006b) demonstrate rapid processes of deprotonation and secondary protonation of olivine.
FTIR and TEM data provide the information about H
2
O content and a mode of water
occurrence in olivine. Here we suggest a way to reconstruct the Phistory of the kimberlite
process and elucidate the water behavior at different stages of the kimberlite process. The
collected data on the H
2
O occurrence in mantle olivine nodules represented by xenoliths,
xenocrysts and phenocrysts from Yakutian kimberlites (Udachnaya, Obnazennaya, Mir,
Kievlyanka, Slyudyanka, Vtorogodnitza and Bazovayua pipes) are used here as a guide for
tracing H
2
O behavior from fluid to highpressure DHMS phases in the mantle. We show here
that nominally anhydrous olivine is a carrier of water in the mantle and can be used as
indicator of Pf(H
2
O) regime in the mantle.
Samples and Collected Data
Sample Description
Typical kimberlites сontain xenocrysts and xenoliths of mantle and crustal origin
embedded into a fine to coarsegrained groundmass of crystallized kimberlite melt (Sobolev
V.S. et al, 1972; Sobolev N.V., 1974; Pokhilenko et al., 1993; Ukhanov et al., 1988; Matsyuk
et al., 1995).
We collected the mantle olivine samples represented by xenoliths, xenocrysts and
phenocrysts from kimberlite pipes of Yakutian kimberlite province (Udachnaya,
Obnazennaya, Mir, Slyudyanka, Vtorogodniza and Kievlyanka). Olivine samples are
classified as xenoliths, xenocrysts and phenocrysts on the base of petrographic examination.
Phenocrysts are olivine single grains disintegrated from groundmass of kimberlite rock.
Xenoliths are fragments of adjacent mantle and crustal rocks trapped by kimberlite magma
and transported from the depths during kimberlite eruption. Xenocrysts are olivine single
grains disintegrated from adjacent rocks, either mantle or crustal, and lifted from the depths
during kimberlite eruption. Xenoliths are of several cm in size. Xenocrysts are less than 1 cm
in size and comparable by size with phenocrysts.
The samples were studied with optical microscopy, FTIR and TEM.
H
2
O Content in the Olivine Samples
The highest H
2
O content in the mantle olivine samples from Yakutian kimberlites was
measured as 400–420 wt. ppm of H
2
O (KochMuller et al., 2006; Matsyuk & Langer, 2004).
H
2
O contents in xenoliths from Udachnaya and Obnazennaya pipes vary between 14 and 246
wt. ppm (Table 1).
Present FTIR study on the H
2
O content in xenoliths from Udachnaya and Obnazennaya,
together with previous Infrared spectroscopic data on xenocrystic and phenocrystic olivine
samples (Matsyuk and Langer, 2004; KochMuller et al., 2006) show the wide variation of
N.R. Khisina, R. Wirth and S. Matsyuk 30
the H
2
O content in the olivine samples, such as between 0–3 and 400 wt. ppm.of H
2
O. The
results are shown in histograms (Figure 1).
Summarized FTIR data on the H
2
O content in the all studied mantle olivine samples
including xenoliths, xenocrysts and phenocrysts from Yakutian kimberlites (Matsyuk and
Langer, 2004; KochMüller et al., 2006; present study) are represented at histogram (Figure
1a). The same data is plotted individually for phenocrysts (Figure 1b) and xenoliths together
with xenocrysts (Figure 1c). For comparison, the H
2
O contents in mantlederived olivine
megacrysts from the Monastery kimberlite, South Africa (Bell et al., 2004) are plotted on the
histogram in Figure 1d. The most frequent values of H
2
O in olivine samples from Monastery
kimberlite are 150–200 wt. ppm of H
2
O.
Table 1. Water contents in olivine from mantle xenoliths
Xenolith sample Locality Xenolith description
H
2
O content,
wt. ppm
Ob152 Obnazennaya pipe Garnet lerzolite 246
Ob105 ″ Garnet lerzolite 14–65
Ob174 ″ Garnet lerzolite 143(15)
Ob312 ″ Garnet lerzolite 45
U47/76 Udachnaya pipe 9.5 cm in size
megacrystalline . diamondbearing
xenolith. Harzburgitedunite
18
9206 ″ 5.2 cm in size coarsegrained
xenolith. Garnetfree harzburgite
29
0 100 200 300 400
H
2
O, wt. ppm
0
4
8
12
0 100 200 300 400
H
2
O, wt. ppm
0
4
8
12
(A) (B)
Figure 1. Continued on next page.
H
2
O in the Mantle: From Fluid to HighPressure Hydrous Silicates 31
0 100 200 300 400
H
2
O, wt. ppm
0
4
8
12
0 50 100 150 200 250 300
H
2
O, wt. ppmÐ2Ùá öåþ ççüÐ2Ùá öåþ ççü
0
2
4
6
8
10
(C) (D)
Figure 1. Histograms of the H
2
O contents in mantle olivine samples from kimberlites. a – c: The FTIR
data for Yakutiyan olivine samples from Udachnaya, Obnazennaya, Mir, Vtorogodnitza, Kievlyanka
and Sluydyanka kimberlites (present data; Matsyuk & Langer, 2004; KochMüller et al., 2006);
a – xenoliths, xenocrysts and phenocrysts all together; b – phenocrysts; c – xenoliths and xenocrysts;
d – Xenoliths from Monastery kimberlite, South Africa (Bell et al., 2003).
Extrinsic H
2
O in Olivine Samples
TEM examination of the olivine samples (Khisina et al., 2001; Khisina & Wirth, 2002;
Khisina et al., 2008) revealed the OH segregation resulted in nanoheterogeneity of several
types: (i) nanoinclusions; (ii) lamellar precipitates; (iii) veins developed along healed
microcracks. All kinds of heterogeneity are associated with deformation slip bands in the
samples understudy (Khisina et al., 2008).
(i) Nanoinclusions were observed in both xenoliths and xenocrysts. They are several tens of
nanometers in size and have a shape of pseudohexagonal negative crystals. The
nanoinclusions are often arranged in arrays along [100], [011], [101] and [101]
crystallographic directions of the olivine host. The phase constituents of
nanoinclusions were identified from TEM data (Wirth & Khisina, 1998; Khisina et
al., 2001; Khisina & Wirth, 2002; Khisina et al., 2008) as 10ÅPhase
Mg
3
Si
4
O
10
(OH)
2
.
nH
2
O, where n = 0.65, 1.0, and 2.0, and hydrous olivine
(MgH
2
SiO
4
)
.
n(Mg
2
SiO
4
), both of them represent highpressure DHMS phases (Bauer
& Sclar, 1981; Khisina & Wirth, 2002; Churakov et al., 2003). 10ÅPhase is a
typical constituent of nanoinclusions in xenocrysts, while hydrous olivine is more
common for nanoinclusions in xenoliths. Both 10ÅPhase and hydrous olivine are
replaced often by a lowpressure assemblage of serpentine Mg
3
Si
2
O
5
(OH)
4
+ talc
Mg
3
Si
4
O
10
(OH)
2
. Highpressure phases in nanoinclusions as well as their
replacement products are strictly aligned relative to the crystallographic directions of
the olivine matrics, with a
ol
║ a
hy
║ c
10Å
║ c
tc
║ c
serp
(hy is abbreviation of hydrous
N.R. Khisina, R. Wirth and S. Matsyuk 32
Figure 2. Array of nanoinclusions in olivine xenocryst from Udachnaya kimberlite. Nanoinclusions are
composed of the 10ÅPhase in the middle part of the inclusions, and filled by H
2
O fluid in the regions
bordered the adjacent olivine (white areas of nanoinclusions).
Figure 3. Healed microcrack filled by 10ÅPhase in olivine xenolite sample from Udachnaya
kimberlite.
olivine), which is indicative of the topotaxic character of these intergrowths. A
characteristic feature of nanoinclusions in xenocrystic olivine samples is the presence
of voids unfilled with solid material (Figure 2). Solid phase fills the equatorial area
of the inclusions parallel to the (100) plane of the olivine host, and voids in
nanoinclusions are observed at the polar areas bordering the olivine matrix (Figure
H
2
O in the Mantle: From Fluid to HighPressure Hydrous Silicates 33
2). The observations led to the conclusion that these voids are not artifacts and were
not produced during sample preparation. According to TEM observations, the
nanoinclusions are not connected with the surface of the crystals through channels,
which could have served as pathways for the transport of H
2
O from the external
medium during the formation of inclusions.
(ii) (100) Lamellar precipitates of hydrous olivine, several nanometers in thickness were
observed in xenolithic olivine (Khisina et al., 2001; Khisina & Wirth, 2002;
Churakov et al., 2003).
Veins filled by 10ÅPhase + talc, were observed in xenolithic olivine (Figure 3). These
veins were developed along (100) healed microcracks.
Discussion
Olivine as Water Storage in the Mantle
Histogram of the all H
2
O content measurements for olivine samples, as summarized for
xenoliths, xenocrysts and phenocrysts together (Figure 1a) reveals two maximums, at 0–50
wt. ppm of H
2
O and 200–250 wt. ppm of H
2
O. The same maximums are pronounced at the
histogram represented the H
2
O contents in xenoliths and xenocrysts together (Figure 1c),
whereas only one maximum at 200–250 wt. ppm of H
2
O is observed at the histogram
represented the H
2
O data for phenocrystic olivine (Figure 1b). Histograms at Figure 1b,c
show that the maximum between 0–50 wt. ppm of H
2
O corresponds to the most frequent H
2
O
contents in xenoliths and xenocrysts, while the maximum between 200–250 wt. ppm of H
2
O
corresponds to the most frequent H
2
O values in phenocrysts. We suggest that the most
frequent H
2
O contents correspond to the initial H
2
O contents incorporated by the olivine
samples during crystallization. Hence, the initial H
2
O content in olivine derived from adjacent
and kimberlite rocks has been different. We conclude that compared to xenolithic and
xenocrystic samples, the olivine phenocrysts have incorporated more amount of water during
crystallization. According to the experimental data on the Pf(H
2
O)–dependence of the OH
solubility in olivine (Kohlstedt et al, 1996), this could mean that the trapped fragments of
adjacent and kimberlite rocks have been crystallized either at different depths or from
different melts. On the basis of the data by Kohlstedt (1996) the pressure conditions of olivine
crystallization from watersaturated melt can be estimated as < 1 GPa for xenocrysts and
xenoliths and as 4–4.5 GPa for phenocrysts. The pressure of 4–4.5 GPa estimated as the
pressure of crystallization of phenocrysts corresponds to the depth of 120–135 km that is
consistent with the upper mantle depths and is in good agreement with estimations of the
depths of kimberlite formation as between 120 and 230 km (Pochilenko et al., 1993). The
pressure < 1 GPa estimated as the pressure of crystallization of mantle olivine samples
represented by xenoliths and xenocrysts, is less than the lowest pressures in the mantle and
corresponds to the depth of <30 km that is consistent with crustal depths; this is not
compatible with petrographic identification of these samples as derived from mantle rocks.
Therefore the difference between mantle xenoliths (and xenocrysts) on the one hand, and
phenocrysts on the other hand, in respect of initial water contents can not be explained by
different depths of their crystallization. Consequently, this led us to the conclusion that the
N.R. Khisina, R. Wirth and S. Matsyuk 34
mantle adjacent rocks and the kimberlites have been crystallized from different melts, water
saturated and water depleted, respectively.
PostCrystallization H
2
O Behavior in Olivine
Postcrystallization H
2
O behavior in the olivine samples is pronounced mostly in
xenoliths and xenocrysts and manifests itself in the following processes.
(i) Secondary protonation of olivine samples derived from adjacent mantle rocks. Secondary
protonation of the olivine derived from adjacent mantle rocks could take place at the
contact of adjacent rocks and the parent kimberlite magma chamber. Interaction of
adjacent rocks with watersaturated parent kimberlite melt resulted in H
2
O
enrichment up to 150–200 wt. ppm in olivine (Figure 1).
(ii) Internal redistribution of H
2
O within the olivine host resulted in nucleation of
nanoinclusions consisted of highpressure DHMS phases represented by10ÅPhase
and hydrous olivine, with the H
2
O fluid separation from the solid inside of the
nanoinclusions.
(iii) Subsequent replacement of DHMS by lowpressure assemblage of serpentine + talc in
nanoinclusions.
(iv) Olivine hydration due to interaction of adjacent rocks with the H
2
O fluid collected in
kimberlite magma chamber; this resulted in the olivine replacement by 10ÅPhase
along microcracks;
(v) Subsequent dehydration of the 10ÅPhase developed along micrcracks in olivine.
H
2
O Fluid in Kimberlite Melt
During a crystallization of early generation of olivine from the parent watersaturated
kimberlite melt in the magma chamber the separation of H
2
O fluid from the melt occurred
that led to the internal pressure increase in the kimberlite magma chamber. The bulk H
2
O
contents in the olivine phenocrystic samples vary between150 and 300 wt. ppm about the
most frequent content as 200–250 wt. ppm of H
2
O. This variations may indicate that the
pressure in kimberlite magma chamber has been oscillated with time by magnitude between
at least of 3 GPa and 5.2 GPa because of the alternating processes of pressure increasing due
to the H
2
O fluid separation and accumulation in the chamber followed then by decompression
during chamber expansion and H
2
O fluid liberation. The pressure oscillation with time took
place in kimberlite chamber prior to explosion, and resulted in the deformation of olivine and
subsequent appearance of DHMS nanoinclusions associated with deformation slip bands in
olivine. The alteration of adjacent rocks took place at the contact with the kimberlite magma
chamber before the explosion and eruption of kimberlite, and resulted in (i) secondary
protonation of olivine due to interaction with watersaturated kimberlite melt and (ii)
metasomatic hydration of olivine resulted in the olivine replacement by the 10ÅPhase due to
interaction of the olivine with H
2
O fluid. The reaction of hydration of olivine may be written
as
4(Mg,Fe)
2
SiO
4
+ 3H
2
O(fluid) = Mg
3
Si
4
O
14
H
6
+ 5(Fe,Mg)O (1)
H
2
O in the Mantle: From Fluid to HighPressure Hydrous Silicates 35
The kimberlite explosion happened because the internal pressure increased as the H
2
O
fluid separated from the melt inside of the kimberlite magma chamber. This event resulted
when the pressure corresponded to the highest H
2
O content in olivine phenocrysts measured
as 400–420 wt. ppm (Figure 1) attained at least 6.5 GPa, as it may be estimated from the
pressure dependence on the H
2
O content in olivine (Kohlstedt et al., 1996). The subsequent
eruption and rapid lifting of the mantle material from the depth of 120135 km was
accompanied by destruction of adjacent rocks through the flow path. The pressure increasing
inside the chamber prior to explosion followed then by rapid decompression during eruption,
when the 10ÅPhase substituted for talc Mg
3
Si
4
O
10
(OH)
2
(lowpressure phase) with H
2
O fluid
liberation:
Mg
3
Si
4
O
14
H
6
= Mg
3
Si
4
O
10
(OH)
2
+ H
2
O (fluid)↑ (2)
According to Chinnery et al. (1999), dehydration of 10ÅPhase (reaction 2) proceeds at
pressures lower than 3.5–5.2 GPa at T <700°C.
OHBearing Nanoinclusions and Intracrystalline H
2
O Fluid
The nucleation of OHbearing nanoinclusions could be initiated by deformation of
olivine that is evident from the similar arrangement of arrays of nanoinclusions and the
optically visible slip bands (Khisina et al., 2008), both along common crystallographic
directions in the olivine host. The slip bands in olivine could be produced by the internal
pressure progressively increasing in the kimberlite magma chamber due to separation of the
H
2
O fluid from the magma melt. Slip bands are zones of high density of dislocations in a
crystal and, consequently, contain relatively high concentration of OHbearing point defects.
Nanoinclusions composed of DHMS phases could be nucleated within the former slip bands
under subsequent decompression and cooling. Further, the DHMS phases were substituted by
a lowpressure serpentine + talc assemblage.
The question is what was a driving force for the exsolution of highpressure hydrous
silicate phases from the host olivine that occurs within slip bands. If it is assumed that the
process of selfpurification of olivine from water took a place under decompression, then how
to explain that this process resulted in creation of highpressure phases? Are they in
equilibrium with the olivine host?
In order to explain the mechanism and reactions of formation and further transformation
of the primary highpressure phases, 10ÅPhase and hydrous olivine, to the lowpressure
serpentine + talc assemblage, we need to understand the nature of voids in the nanoinclusions
(Figure 2), the source of H
2
O, and the mechanism of formation of nanoinclusions themselves.
TEM observations show the topotaxic relationships of the olivine/10ÅPhase intergrowth,
the exact Mg/Si stoichiometric ratio and the lack of channels connecting the inclusions with
the grain surface. Therefore we propose that no fluid infiltration occurred, and the
nanoinclusions observed in olivine samples might have formed due to internal isochemical
process within a crystal. Consequently, a crystal should have been water saturated prior to the
nucleation of inclusions. We suggest that nanoinclusions in olivine have been formed at the
postcrystallization stage under the pressure oscillation in the kimberlite chamber and the
formation of the nanoinclusions proceed through the diffusion and segregation of intrinsic
OHbearing point defects, i.e., without any infiltration from outside. Diffusion experiments
N.R. Khisina, R. Wirth and S. Matsyuk 36
have shown that hydrogen can rapidly diffuse through olivine at temperatures >800°C
(Kohlstedt & Mackwell, 1998). The diffusion process could result in an aggregation of the
OH and associated metal vacancies within the OHsaturated slip bands of the olivine host.
We consider the formation of OHbearing nanoinclusions in olivine as a process of self
purification of olivine from the OHbearing defects by the coupled diffusion mechanism Mg
2+
═ {v
Mg
,2H
+
}.With the assumption that the cation composition of the inclusion corresponds to
the stoichiometry of the 10Å Phase (Mg/Si = 3:4), the reaction of the nanoinclusion formation
by unmixing of OHbearing olivine solid solution is the following one:
OHolivine (s.s.) = [stoichiometric olivine]
(matrix)
+ [10ÅPhase + H
2
O(fluid)]
(inclusion)
(3a)
or
5y[Mg
2x
v
x
SiO
4
H
2x
] = (5y4yx)[Mg
2
SiO
4
]
(matrix)
+ yx[Mg
3
Si
4
O
14
H
6
+ 2H
2
O]
(inclusion)
(3b)
Taking into account the variations of the 10ÅPhase chemical composition with respect
to H
2
O content (Bauer & Sclar, 1981; Chinnery et al., 1999), the bracketed part of the right
side of the reaction (3b) should be more correctly presented as Mg
3
Si
4
O
10
(OH)
2
.
nH
2
O + (4
n)H
2
O, where n is 0.65; 1.0; 2.0. Reaction (3) occurs in a closed volume limited by the size of
the inclusion. The plausibility of the proposed model for the nanoinclusion formation can be
checked by the criterion of volume conservation, i.e., in reaction (3) the volume of reactants
must be equal to the volume of products. The volume of the inclusion can be expressed as a
function of the molar volume of olivine as 4yx[V
u.c.(Ol)
/Z]
.
N, and the volume occupied by the
10ÅPhase formed in reaction (3) is yx[V
u.c.(10ÅPhase)
/Z]
.
N, where N is Avogadro’s number,
V
u.c.
is a unit cell volume and Z is a number of chemical formula units per unit cell. The
difference of these volumes is ΔV = 52yxN under normal PTconditions. With the
assumption that the volume ΔV is filled with the H
2
O fluid released owing to reaction (3), we
calculated the density of the fluid by the equation ρ = M/Vmol, where Vmol is the molar
volume per one molecule of H
2
O in the inclusion, which equals ΔV/2yx in accordance with
Figure 4. Fluid inclusion in olivine xenolith (Bullfontain kimberlite, South Africa). 10ÅPhase is
observed as a shell of the inclusion at the contact between the inclusion and olivine matrix.
H
2
O in the Mantle: From Fluid to HighPressure Hydrous Silicates 37
reaction (3), and M is the formula weight of H
2
O (M=18). For n of 0.65, 1.0 and 2.0 in the
chemical formula of the 10ÅPhase, the number of formula units of H
2
O fluid is 3.35, 3.0,
and 2.0, and the calculated fluid density ρ is approximately 1.9, 1.7, and 1.1 g/cm
3
,
respectively. The value ρ=1.1 g/cm
3
corresponding to the highest H
2
O content in the 10Å
Phase (n=2) is most consistent with the density of water under normal PTconditions (ρH
2
O
= 1.0 g/cm
3
), which allows consideration of the proposed model of inclusion formation as
rather realistic at n=2.
It should be mentioned here that the 10ÅPhase together with H
2
O fluid was also
observed in the shellrim of fluid inclusion occurring in the mantle olivine sample from
Bullfontain kimberlite (South Africa) (Figure 4). In this case, the 10ÅPhase was most likely
created due to recrystallization of the trapped fluid inclusion.
Conclusion
The collected data show that xenoliths, xenocrysts and phenocrysts from kimberlite
represent specimens of the socalled “wet” olivine; xenoliths and xenocrysts have a number
of common features, such as frequently occurring deformation slip bands and the OHbearing
nanoinclusions that are filled by high and lowpressure hydrous magnesium silicates together
with the H
2
O fluid. The observations show that the 10ÅPhase is closely related to olivine in
both kimberlite and adjacent rocks and, therefore, the 10ÅPhase could be considered
ubiquitous nanomineral of kimberlites marking at about 4–5 GPa a certain stage of the
kimberlite process.
On the other hand, xenoliths and xenocrysts differ from phenocrysts by the water bulk
content, as well as by the mechanism of the 10ÅPhase formation. On the basis of collected
FTIR data, the xenoliths and xenocrysts on one hand, and phenocrysts on the other hand, are
considered crystallized from different melts, such as waterdepleted magma and water
saturated kimberlite melt, correspondingly.
The pressure regime inside the kimberlite magma chamber was controlled by H
2
O fluid,
with pressure oscillation with time between of at least 3 GPa and 5.2 GPa due to alternating
processes of pressure increasing during the H
2
O fluid separation and accumulation in the
chamber followed then by decompression due to chamber expansion and H
2
O fluid liberation.
The released H
2
O fluid participates in metasomatic hydration of the olivinebearing adjacent
mantle rocks, resulting in the olivine replacement for the 10ÅPhase.
Reviewed by
Prof. O. Lukanin
Institute of Geochemictry and Analytical Chemistry,
Russian Academy of Sciences, Kosygin st. 19,
119991 Moscow Russia
N.R. Khisina, R. Wirth and S. Matsyuk 38
References
[1] Bai, Q., and Kohlstedt, D.L. (1993): Effects of chemical environment on the solubility
and incorporation mechanism for hydrogen in olivine. Phys. Chem. Minerals, 19, 460
471.
[2] Bauer, J.F., and Sclar, C.B. (1981): The 10 Å Phase in the system MgOSiO
2
H
2
O. Am.
Mineral., 66, 576585.
[3] Bell, D.R., Rossman, G.R., Maldener, J., Endisch, D., and Rauch, F. (2003): Hydroxide
in olivine: a quantitative determination of the absolute amount and calibration of the IR
Spectrum. J. Geophys. Res., 108, B2:21052113.
[4] Beran, A., and Putnis, A. (1983): A model of the OH positions in olivine derived from
infraredspectroscopic investigations. Phys. Chem. Minerals, 9, 5760.
[5] Beran A., and Libowitzky E. (2006). Water in natural mantle minerals II: olivine, garnet
and accessory minerals. In: Water in nominally anhydrous minerals. (H.Klepper and
J.R.Smyth ed.) Rev. in Mineralogy and Geochemistry, Geochemical Soc. and
Mineralogical Soc. of America, v.62, p.169191.
[6] Chinnery N.J., Pawley, A.R., and Clark S.M. (1999): In situ observation of 10 Å Phase
from talc + H
2
O at mantle pressures and temperatures. Science, 286, 940942.
[7] Churakov, S.V., Khisina, N.R., Urusov, V.S., and Wirth, R. (2003): Firstprinciples
study of (MgH
2
SiO
4
).n(Mg
2
SiO
4
) hydrous olivine structures. 1. Crystal structure
modelling of hydrous olivine Hy2a (MgH
2
SiO
4
).3(Mg
2
SiO
4
). Phys. Chem. Minerals,
30, 111.
[8] Kamenetsky M.B., Sobolev A.V., and Kamenetsky V.S. (2004) Kimberlite melts rich in
alkali chlorides and carbonates: a potent metasomatic agent in the mantle. Geology, 32,
10, 845 – 848.
[9] Katayama I., Karato Si., and Brandon M. Evidence of high water content in the deep
upper mantle inferred from deformation microstructures. (2005): Geology, 33, 613616.
[10] Khisina, N.R., and Wirth, R. (2002): Hydrous olivine (Mg
1y
Fe
2+
y
)
2x
v
x
SiO
4
H
2x
– a new
DHMS phase of variable composition observed as sizedsized precipitations in mantle
olivine: Phys. Chem. Minerals, 29, 9811.
[11] Khisina, N.R., Wirth, R., Andrut, M., and Ukhanov, A.V. (2001): Extrinsic and intrinsic
mode of hydrogen occurrence in natural olivines: FTIR and TEM investigation. Phys.
Chem. Minerals, 28, 291301.
[12] Khisina N.R., and Wirth R. (2008) Nanoinclusions of highpressure hydrous silicate,
Mg
3
Si
4
O
10
(OH)
2
.
nH
2
O (10ÅPhase), in mantle olivine: mechanisms of formation and
transformation. Geochemistry International, 46, 4, 319 – 327.
[13] Khisina N.R., Wirth R., Matsyuk S., and KochMüller M. (2008) Microstructures and
OHbearing nanoinclusions in “wet” olivine xenocrysts from Udachnaya kimberlite.
Eur. J. Mineral. (in press),
[14] KochMüller, M., Matsyuk, S.S., Rhede, D., Wirth, R., and Khisina, N. (2006):
Hydroxyl in mantle olivine xenocrysts from the Udachnaya kimberlite pipe. Phys.
Chem. Minerals, 33, 276287.
[15] Kohlstedt, D.L., Keppler, H., and Rubie, D.C. (1996): Solubility of water in the α, β and
γ phases of (Mg,Fe)
2
SiO
4
. Contrib. Mineral. Petrol., 123, 345357.
H
2
O in the Mantle: From Fluid to HighPressure Hydrous Silicates 39
[16] Kurosawa, M., Yurimoto, H., and Sueno, S. (1997): Patterns in the hydrogen and trace
element compositions of mantle olivines. Phys. Chem. Minerals, 24, 385395.
[17] Lemaire C., Kohn S.C., and Brooker R.A. (2004): The effect of silica activity on the
incorporation mechanisms of water in synthetic forsterite: a polarized infrared
spectroscopic study. Contrib. Mineral. Petrol. 147, 4857.
[18] Matsyuk, S.S. (1995): Olivine macrocrysts from kimberlites of Yakutiya.
Mineralogicheski journal, 17, 3, 4457. (in Russian).
[19] Matsyuk, S.S., and Langer, K. (2004): Hydroxyl in olivines from mantle xenoliths in
kimberlites of the Siberian platform. Contrib. Mineral. Petrol., 147, 413437.
[20] Miller, G.H., Rossman, G.R., and Harlow, G.E. (1987): The natural occurrence of
hydroxide in olivine. Phys. Chem. Minerals, 14, 461472.
[21] Peslier, A.N., and Luhr, J.F. (2006): Hydrogen loss from olivines in mantle xenoliths
from Simcoe (USA) and Mexico: mafic alkalic magma ascent rates and water budget of
the subcontinental lithosphere. Earth and Planetary Science Letters, 242, 302319.
[22] Pokhilenko, N.P., Sobolev, N.V., Boyd, F.R., Pearson, D.G., and Shimizu, N. (1993):
Megacrystalline pyrope peridotites in the lithosphere of the Siberian Platform:
mineralogy, geochemical peculiarities and the problem of their origin. Geol. Geophys.,
34/1, 5062. (in Russian).
[23] Prewitt, C.T., and Downs, R.T. (1999): Highpressure crystal chemistry. in J. Hemley
(Ed), Rev. Mineral., vol. 37, pp. 301312.
[24] Sobolev, V.S., Dobretsov, N.L., and Sobolev N.V. (1972): Classification of deep seated
xenoliths and the type of the upper mantle. Geol. Geophys., 12, 3742 (in Russian).
[25] Sobolev, N.V. (1974): Deepseated inclusions in kimberlites and the problem of the
composition of the upper mantle. Nauka, Novosibirsk. 264 pp. (in Russian).
[26] Sobolev A.V., and Chaussidon M. (1996) H
2
O concentrations in primary melts from
suprasubduction zones and midocean ridges: Implications for H
2
O storage and
recycling in the mantle. Earth and Planetary Sci. Lett., 137, 45 – 55.
[27] Thompson, A.B. (1992): Water in the Earth upper mantle. Nature, 358, 295302.
[28] Ukhanov, A.V., Ryabchikov, I.D., and Kharkiv, A.D. (1988): Lithospheric mantle of the
Yakutiya kimberlite province . Nauka, Moscow, 286 pp.
[29] Wirth, R., and Khisina, N.R. (1998): OHbearing crystalline inclusions in olivine from
kimberlitic peridotite (UdachnayaEast, Yakutiya). Suppl. EOS Trans., 79,45, T32B17.
[30] Zhao, Y.H., Ginsberg, S.B., and Kohlstedt, D.L. (2004): Solubility of hydrogen in
olivine: dependence on temperature and iron content. Contrib. Mineral. Petrol., 147,
155161.
In: Fluid Mechanics and Pipe Flow ISBN: 9781607410379
Editors: D. Matos and C. Valerio, pp. 41115 © 2009 Nova Science Publishers, Inc.
Chapter 3
ON THE NUMERICAL SIMULATION OF TURBULENCE
MODULATION IN TWOPHASE FLOWS
K. Mohanarangam and J.Y. Tu
*
School of Aerospace, Mechanical and Manufacturing Engineering
RMIT University, Vic. 3083, Australia
Abstract
With the increase of computational power, computational modelling of twophase flow
problems using computational fluid dynamics (CFD) techniques is gradually becoming
attractive in the engineering field. The major aim of this book chapter is to investigate the
Turbulence Modulation (TM) of dilute two phase flows. Various density regimes of the two
phase flows have been investigated in this paper, namely the dilute GasParticle (GP) flow,
LiquidParticle (LP) flow and also the LiquidAir (LA) flows. While the density is quite high
for the dispersed phase flow for the gasparticle flow, the density ratio is almost the same for
the liquid particle flow, while for the liquidair flow the density is quite high for the carrier
phase flow. The study of all these density regimes gives a clear picture of how the carrier
phase behaves in the presence of the dispersed phases, which ultimately leads to better design
and safety of many twophase flow equipments and processes. In order to carry out this
approach, an EulerianEulerian TwoFluid model, with additional source terms to account for
the presence of the dispersed phase in the turbulence equations has been employed for
particulate flows, whereas Population Balance (PB) have been employed to study the bubbly
flows. For the dilute gasparticle flows, particleturbulence interaction over a backwardfacing
step geometry was numerically investigated. Two different particle classes with same Stokes
number and varied particle Reynolds number are considered in this study. A detailed study
into the turbulent behaviour of dilute particulate flow under the influence of two carrier phases
namely gas and liquid was also been carried out behind a sudden expansion geometry. The
major endeavour of the study is to ascertain the response of the particles within the carrier
(gas or liquid) phase. The main aim prompting the current study is the density difference
between the carrier and the dispersed phase. While the ratio is quite high in terms of the
dispersed phase for the gasparticle flows, the ratio is far more less in terms of the liquid
particle flows. Numerical simulations were carried out for both these classes of flows and
*
Email address: Jiyuan.Tu@rmit.edu.au. Tel: +61399256191. Fax: +61399256108. Mailing address: Prof
Jiyuan Tu, School of Aerospace, Mechanical and Manufacturing Engineering, RMIT University, PO Box 71,
Bundoora Vic 3083, AUSTRALIA
K. Mohanarangam and J.Y. Tu 42
their results were validated against their respective sets of experimental data. For the Liquid
Air flows the phenomenon of drag reduction by the injection of microbubbles into turbulent
boundary layer has been investigated using an EulerianEulerian twofluid model. Two
variants namely the Inhomogeneous and MUSIG (MUltipleSIzeGroup) based on Population
balance models are investigated. The simulated results were benchmarked against the
experimental findings and also against other numerical studies explaining the various aspects
of drag reduction. For the two Reynolds number cases considered, the buoyancy with the plate
on the bottom configuration is investigated, as from the experiments it is seen that buoyancy
seem to play a role in the drag reduction. The under predictions of the MUSIG model at low
flow rates was investigated and reported, their predictions seem to fair better with the decrease
of the breakup tendency among the microbubbles.
Introduction
The basics of twophase flows have received interdisciplinary attention from chemical,
mechanical and industrial engineers for decades as they play a major role in a variety of
industrial and design processes. Among these dilute particulate flows are encountered in a
variety of industrial (Kolaitis & Founti, 2002) and natural processes irrespective of their
carrier phase being gas or liquid. Particles constantly interact with the gas in industries
through sand blasting equipments (Qianpu et al., 2005), pneumatic transport equipments (Gil
et al., 2002), and also play a major role in the safe operation of the power plants (Tu, 1999),
gas turbine engines (Awatef et al., 2005) and helicopters. In chemical industries they appear
as reactants and catalysts, thereby controlling the order and the fate of the chemical reaction.
There are even found in nature as dust dispersed in the room and as pollen released from the
plants carried away by the wind. Lately there has been continued interest in these classes of
flows in Biomedical applications, to study the dust deposition patterns in realistic human
nasal airway (Inthavong et al., 2006) and also to aid better delivery of the medications into the
human nose. In industrial and engineering applications a better understanding of the physics
of these flows will not only lead to better operating efficiency by improved equipment
designs but also increase the longevity accompanied with lower maintenance costs and better
operational improvements.
Flows with particles amid liquid are an important class of twophase flows classified
under as slurry flows, whose flow systems are representative of many mineral processing
operations and also provide useful operation correlations for such processes. They form an
important class of flows encompassing pneumatic conveying system, turbines and
machineries operating in particulateladen environments. These flows provide a useful tool in
the simulation of sprays in industrial and natural processes, as they have comparable phase
density ratios. Comparable densities are of particular interest, since all the effects of
interphase momentum transfer are important (Parthasarathy and Faeth, 1987). They also serve
as a good test of methods to predict particle motion in turbulent environments (Parthasarathy
and Faeth, 1987) as they exhibit high relative turbulence intensities for particle motion, which
influence particle drag properties (Clift et al., 1978).
Over the past few decades there have also been inter disciplinary research to study the
mechanism of drag reduction with a mission to adopt them as tangible practice for a wide
range of applications using microbubbles. While surfactant and polymer injections along the
turbulent boundary layer, contribute a major portion towards this endeavour. Nonetheless,
they are limited in terms of realistic applications due to their inherent ability to cause
On the Numerical Simulation of Turbulence Modulation in TwoPhase Flows 43
environmental pollution and there by causing the destruction of the oceans flora and fauna.
Drag reduction by the injection of microbubbles along the boundary layer seems to be a
realistic approach because fixing the dispersed phase as air for drag reduction devices may
sound quite pragmatic; due to its wide spread availability unlike polymers or surfactants.
Although there may be some overhead losses in placing them along the boundary layer, they
are still undoubtedly the cheapest source, with no requirement of large storage space unlike
the use of other low density inert (environment friendly) gases to form microbubbles.
With the increase in the computational power and efficiency, computational fluid
dynamics (CFD) have taken a centre point in offering effective solutions to not only single
phase flows, but also for the simulation of wide range of twophase flows viz., gasparticle,
liquidparticle and liquidgas flows. They not only offer a cheap solution, but also help
scientists and engineers probe into places prohibitive by experimental methods or hostile
environments unsuitable for human life forms. Numerical simulation of these twophase
flows lay broadly into two major categories namely the EulerianEulerain twofluid model
approach and the EulerianLagrangian particle tracking approach.
In the Lagrangian particle tracking approach, each and every particle is tracked, thereby
providing a detailed behaviour of their trajectories, velocities, bounce back angles and other
parameters. Although they encompass a great deal of information within the domain, they
prove to be rather cumbersome for multidimensional problems for the same reason being not
able to track rather large number of particles given the computational power such as to obtain
a good statistical information of the dispersed phase. Besides this, there is the problem of
representing turbulent interactions between two phases (twoway coupling), which
necessitates the need to fully understand the interactions of particles with individual vortices
(Fessler and Eaton, 1997).
EulerianEulerian approach constituted by Anderson and Jackson (1967), Ishii (1975),
regard the carrier and the dispersed phases as two interacting fluids with momentum and
energy exchange between them. One major advantage of using the Eulerian approach is that
the wellproven numerical procedures for singlephase flows can be directly extended to the
secondary phase with the effects of turbulent interactions between the two fluids, lately there
are considerations of extending the Eulerian twofluid model by adopting the large eddy
simulation (LES) approach (Pandya and Mashayek, 2002). Shirolkar et al (1996) in their
paper stated that Eulerian models have problems to account for the particle history effects as
they do not retrace the motion of individual particles together; they also suffer from
continuum assumption problems with respect to particles, as the particles equilibrate with
neither local fluid nor each other when flowing through the flow field, in addition, crossing
trajectories become more pronounced as particle inertia increases and Eulerian methods may
become less accurate with increasing Stokes number, so a priori and rudimentary Lagrangian
calculations should always be performed to check its validity.
The above argument may be true for dispersed phase flows that contain solid particles
that does not undergo any shape changes, the same was used to simulate the bubbly flows not
time long, however with the advances in CFD in relation to bubble dynamics, it is envisioned
that the constant bubble size may only be valid for problems where the dispersed phase does
not undergo deformation. Directly adopting the model for flow problems, wherein dispersed
phase undergoes constant change in the shape, may introduce substantial error into the final
predictions. The MUSIG model was thereby developed to circumvent the above shortcoming
(Lo, 1996). Theoretically speaking, the model works in a manner in which it resolves the
K. Mohanarangam and J.Y. Tu 44
bubble size mechanistically while working within a range specified groups for the dispersed
phase. However, each group possess an individual equation to be solved, which increases the
turn around time for each simulation, there by making the parametric study of the problem
computationally expensive.
Numerical modelling of these classes of flows poses a problem, considering the diversity
of twophase flows, wherein the addition of a dispersed phase to a flow greatly increases the
parameter space of the problem. In addition to the complexity involved in tracking the
dispersed phase, their mere presence can drastically change the characteristics of the flow
itself (Fessler and Eaton, 1995). The turbulent dispersion of these dispersed phase
particles/bubbles within the carrier phase can be studied by two orders, one by presuming that
the presence of the dispersed phase does not have any effect on the turbulent carrier flow field
which is considered to be ‘one way coupling’ and the second by considering a feedback of the
dispersed phase into the carrier flow field, in addition to the dispersed phase getting affected
by the carrier flow field, this is considered to be ‘twoway coupling’, which is most of the
times the precise phenomena taking place in the real world.
Turbulence Modulation (TM) which redefines the carrier phase both at the velocity and
at the turbulence level in the presence of dispersed phase is crucial in the design of
engineering applications. However, this study is paralysed owing to the complexities of the
flows and limitations of the instruments. A nearly homogenous flow like liquidparticle flow
can circumvent this problem, wherein all turbulence properties are attributed due to the
relative motion of the particles; thereby any change felt due to the dispersed phase on the
carrier phase is a direct result of only the TM phenomenon (Parthasarathy and Faeth, 1990).
This phenomenon have been exploited by many experimental researchers (Parthasarathy and
Faeth, 1987; Parthasarathy and Faeth, 1990; Alajbegovic et al, 1994; Rashidi et al, 1990; Sato
and Hishida, 1996; Ishima et al, 2007; Borowsky and Wei, 2007; Righetti and Romano, 2007)
to not only investigate, study and understand the basic features of TM but also to aid in the
better formulation of numerical models. A number of previous studies have examined particle
response for gasparticle flows in a sudden expansion flow experimentally (Ruck and
Makiola, 1988; Hishida and Maeda, 1999; Fessler and Eaton, 1997). Whereas for the liquid
particle flows, although there have been studies in channel flow geometries, but their
publication is limited for a sudden expansion geometry except for Founti & Klipfel (1998).
Reynoldsaveraged NavierStokes (RANS) equations are one of the well known
numerical approaches to predict the Turbulence Modulation (TM) of the carrier phase in
many industrial flow applications. They stem out as a consequence of time average to yield
the constitutive mean flow equations, however the turbulent stresses which arise as a direct
consequence should be modeled with some degree of approximation. Specification of the
turbulent eddy viscosity serves as the turbulent closure and the kε turbulence formulation
specifies this by solving two additional transport equations for turbulent kinetic energy and
the eddy dissipation. This being the case for single phase flows, for dilute nonreacting
particulate flows, these pose a problem in the sense the momentum of the carrier phase
undergoes a phenomenal change due to the presence of particles, this is also reflected in terms
of TM of the carrier phase.
The major aim of this article is to investigate the various density regimes of the two
phase flows and to study the effects of the dispersed phase onto the carrier phase at the
velocity and at the turbulence level (Turbulence Modulation), for the dilute gasparticle flow,
liquidparticle flow and also the airliquid flow. While the density is quite high for the
On the Numerical Simulation of Turbulence Modulation in TwoPhase Flows 45
dispersed phase flow for the gasparticle flow, the density ratio is almost the same for the
liquid particle flow, while for the airliquid flow the density is quite high for the carrier phase
flow. The study of all these density regimes gives a clear picture of how the carrier phase
behaves in the presence of the dispersed phases, which ultimately leads to better design and
safety of many twophase flow equipments.
Conservation Equations
3.1. GasParticle and LiquidParticle Flows
The modified Eulerian twofluid model developed by Tu and Fletcher (1995) and Tu
(1997) is employed for the simulation of gasparticle and liquidparticle flows, which
basically considers the carrier and the dispersed phases as two interpenetrating continua.
Hereby, a two way coupling is achieved between the dispersed and the carrier phases.
The underlying assumptions employed in the current study are:
1) The particulate phase is dilute and consists of mono disperse spherical particles.
2) For such a dilute flow, the gas volume fraction is approximated by unity.
3) The viscous stress and the pressure of the particulate phase are negligible.
4) The flow field is isothermal.
3.1.1. Governing Equations for Carrier Phase Modeling
The governing equations in Cartesian form for steady, mean turbulent gas flow are
obtained by Favre averaging the instantaneous continuity and momentum equations
0 = )
u
(
x
i
g
g
i
ρ
∂
∂
(3.1)
Di
i
g
j
g g
j
i
g
j
g
gl
j i
g
i
g
j
g
g
j
F u u (
x
)
u
x
(
x
+
x
p
 = )
u u
(
x
−
∂
∂
−
∂
∂
∂
∂
∂
∂
∂
∂
) ' ' ρ ν ρ ρ
(3.2)
Eq. (3.1) and (3.2) respectively are the continuity and momentum equation of the carrier
gas phase, where
g g g g
p and u u ' , , ρ
are the bulk density, mean velocity, fluctuating velocity
and mean pressure of the gas phase, respectively. ν
gl
is the laminar viscosity of the gas phase. F
Di
is the Favreaveraged aerodynamic drag force due to the slip velocity between the two phases
and is given by
t
)
u

u
( f
= F
p
i
p
i
g
p Di
ρ (3.3)
where the correction factor f is selected according to Schuh et al (1989)
K. Mohanarangam and J.Y. Tu 46
⎪
⎪
⎩
⎪
⎪
⎨
⎧
<
≤ < +
≤ < +
=
p p
p p p
p p
f
Re 2500 Re 0167 . 0
2500 Re 200 Re 0135 . 0 Re 914 . 0
200 Re 0 Re 15 . 0 1
282 . 0
687 . 0
(3.4)
with the particle response or relaxation time given by ) 18 /( d t
2
p p gl g s
ν ρ ρ = , wherein d
p
is
the diameter of the particle.
3.1.2. Governing Equations for Particulate Phase Modeling
After Favre averaging, the steady form of the governing equations for the particulate
phase is
0 = )
u
(
x
i
p p
i
ρ
∂
∂
(3.5)
WMi Di Gi
i
p
j
p p
j
i
p
j
p
p
j
F + F + F + u u (
x
 = )
u u
(
x
) ' ' ρ ρ
∂
∂
∂
∂
(3.6)
where
p p p
u and u ' , ρ
are the bulk density, mean and fluctuating velocity of the particulate
phase, respectively. In equation (3.6), there are three additional terms representing the gravity
force, aerodynamic drag force, and the wallmomentum transfer force due to particlewall
collisions, respectively. The gravity force is
g F
p
ρ =
Gi
, where g is the gravitational acceleration.
3.1.3. Turbulence Modeling for Carrier Phase
For the carrier gas phase, which uses an eddyviscosity model, the Reynolds stresses are
given by
ij g g
i
j
g
j
i
g
gt g
j
g
i
g g
k
3
2
)
x
u
x
u
( ' u ' u δ ρ
∂
∂
∂
∂
ν ρ ρ + + − = (3.7)
where ν
gt
is the turbulent or ‘eddy’ viscosity of the gas phase, which is computed by
ν ε
μ gt
C = ( / ) k
g g
2
. The kinetic energy of the turbulence, k
g
and its dissipation rate, ε
g
is
governed by separate transport equations. The RNG theory, models the k
g
and ε
g
transport
equations (3.8) & (3.9) respectively by taking into account the particulate turbulence
modulation, in which α is the inverse Prandtl number.
k g g kg
g
gt g
j
g
S P
k
k u + − + = ε ρ
∂
∂
ν αρ
∂
∂
ρ
∂
∂
)
x
(
x
) (
x
j
g
j
g
j
(3.8)
On the Numerical Simulation of Turbulence Modulation in TwoPhase Flows 47
ε ε ε
ρ ε ρ
ε
∂
∂ε
ν αρ
∂
∂
ε ρ
∂
∂
S + R ) C P C (
k
)
x
(
x
) (
x
g 2 kg 1
g j
g
j
g
j
g g
g g
gt g
j
g
u − − + =
(3.9)
The rate of strain term R in the ε
g
equation is expressed as
( )
R
C
k
g
=
−
+
μ
η η η
βη
ε
3
0
3
2
1
1
g ,
η
ε
∂
∂
∂
∂
= = +
⎛
⎝
⎜
⎜
⎞
⎠
⎟
⎟
k u
x
u
x
g g
i
j
g
j
i g
ij ij
S S ( ) ,
/
2
1
2
2 1 2
(3.10)
where β = 0.015, η
0
= 4.38. The major endeavour of including this term is to take into
account the effects of rapid strain rate along with the streamline curvature, which in many
cases the standard kε turbulence model fails to predict. The constants in the turbulent
transport equations are given by α = 1.3929, C
μ
= 0.0845, C
ε1
= 1.42 and C
ε2
= 1.68 as per the
RNG theory (Yakhot & Orszag, 1986).
For the confined twophase flow, the effects of the particulate phase on the turbulence of
the gas phase are taken into account through the additional terms S
k
and S
ε
in the k
g
and ε
g
equations which arise from the correlation term given by
) (
2
) ' ' (
gp g p
p
Di
i
g k
k k
t
f
F u S − − = − = ρ (3.11)
in the k
g
equation and
) (
t
f 2
x x
' F ' u
2 S
gp g p
p j j
Di
i
g
gl
ε ε ρ ν
ε
− − =
∂ ∂
∂ ∂
− =
(3.12)
in the ε
g
equation, where k
gp
and ε
gp
will be presented in the next following section discussing
the particulate turbulence modeling.
3.1.4. Turbulence Modeling for the Dispersed Phase
The transport equation governing the particulate turbulent fluctuating energy can be
written as follows:
gp kp
j
p
p
pt
p
j
p
j
p p
j
I P
x
k
x
k u
x
− + = ) ( ) (
∂
∂
σ
ν
ρ
∂
∂
ρ
∂
∂
(3.13)
The turbulence production P
kp
of the particulate phase is given by
k
i
p
k
k
p
pt p ij p
k
i
p
i
j
p
j
i
p
pt p kp
x
u
x
u
k
x
u
x
u
x
u
P
∂
∂
∂
∂
ν δ ρ
∂
∂
∂
∂
∂
∂
ν ρ ) (
3
2
) ( + − + =
(3.14)
and the turbulence interaction between two phases I
gp
is given by
K. Mohanarangam and J.Y. Tu 48
) (
2
gp p p
p
gp
k k
t
f
I − = ρ (3.15)
Here
i
p
u
i
g
u
gp
k ' '
2
1
=
is the turbulence kinetic energy interaction between two phases.
The transport equation for the gasparticle covariance i
p
u
i
g
u ' '
can be again derived (Tu,
1997), to obtain the transport equation governing the gasparticle correlation which is given
by
gp gp p gp
j
gp
p
pt
g
gt
p
j
gp
j
p
j
g p
j
II P
x
k
x
k u u
x
− − + + = + ε ρ
∂
∂
σ
ν
σ
ν
ρ
∂
∂
ρ
∂
∂
) ( ] ) ( [ (3.16)
where the turbulence production by the mean velocity gradients of two phases is
) ) (
3
1
3
2
) ( }( {
j
i
p
j
i
g
k
k
p
pt
k
k
p
gt ij p gp ij p
i
j
p
pt
j
i
g
gt p gp
x
u
x
u
x
u
x
u
k
x
u
x
u
P
∂
∂
∂
∂
∂
∂
ν
∂
∂
ν δ ρ δ ρ
∂
∂
ν
∂
∂
ν ρ + + − − + =
(3.17)
The interaction term between the two phases takes the form
] 2 2 2 ) 1 [(
2
p g gp p
p
gp
k m k k m
t
f
II − − + = ρ (3.18)
here m is the mass ratio of the particle to the gas, m=ρ
p
/ρ
g
. The dissipation term due to the
gas viscous effect is modeled by
) exp(
g
g
p g gp
k
t B
ε
ε ε
ε
− =
(3.19)
where B
ε
=0.4.
The turbulent eddy viscosity of the particulate phase, ν
pt
, is defined in a similar way as
the gas phase as:
ν
pt p pt pt p
k t l k = =
2
3
2
3
(3.20)
The turbulent characteristic length of the particulate phase is modeled by
) , min(
s pt pt
D l l ′ =
where
pt
l′ is given by
)] (
 ' 
 ' 
exp[ ) cos 1 (
2
2
p g
g
r
gp
gt
pt
k k sign
u
u
B
l
l − − + = ′ θ
(3.21)
where θ is the angle between the velocity of the particle and the velocity of the gas to account
for the crossing trajectories effect (Huang et al, 1993). B
gp
is an experimentally determined
On the Numerical Simulation of Turbulence Modulation in TwoPhase Flows 49
constant, which takes a value of 0.01. D
s
is the characteristic length of the system and
provides a limit to the characteristic length of the particulate phase.
The relative fluctuating velocity is given by
p g r
u u u ' ' ' − =
(3.22)
and
) 2 (
3
2
' ' ' 2 '  ' 
2 2
p gp g
p
p g
g
r
k k k u u u u u + − = + − = (3.23)
The solution process of the above stated numerical formulation can be depicted in the
form of a flow chart as shown in figure 3.1.
No
Initial flow
setup
Solve gasphase momentum
equations
Solve pressure correction for
the gas phase
Solve particlephase
momentum equations
Solve particle concentration
equation
Solve turbulence equations for
the gas phase (k
g
, e
g
)
Solve turbulence equations for
the particle phase (kp
)
Solve turbulence equations for
gasparticle coupling (kgp, egp
)
Update primitive variables
(concentration, viscosity)
Twoway coupling terms
Stop
A
A
B
B
Overall
convergence
Yes
Figure 3.1. Solution procedure for Eulerian twofluid model.
3.2. LiquidAir Flows (Microbubble)
3.2.1. Inhomogeneous TwoFluid Model
3.2.1.1. Mass Conservation
Numerical simulations presented in this paper are based on the twofluid model using
EulerianEulerian approach. The liquid phase is treated as continuum while the gas phase
(bubbles) is considered as dispersed phase (ANSYS, 2006). Under isothermal flow condition,
K. Mohanarangam and J.Y. Tu 50
with no interfacial mass transfer, the continuity equation of the twophases with reference to
Ishii (1975) and Drew and Lahey (1979) can be written as:
( )
( ) 0 = ⋅ ∇ +
∂
∂
i i i
i i
u α ρ
t
α ρ
(3.24)
where α, ρ and u
is the void fraction, density and velocity of each phase. The subscripts i = l
or g denotes the liquid or gas phase.
3.2.1.2. Momentum Conservation
The momentum equation for the twophase can be expressed as follow:
( )
( ) ( ) ( ) [ ]
lg
T
i i
e
i i i i i i i i i
i i i
F u u μ α g ρ α P α u u α ρ
t
u α ρ
+ ∇ + ∇ ⋅ ∇ + + ∇ − = ⋅ ∇ +
∂
∂
(3.25)
On the right hand side of Eq. (3.25), F
lg
represents the total interfacial force calculated
with averaged variables, g
is the gravity acceleration vector and P is the pressure. The term
F
lg
represents the interphase momentum transfer between gas and liquid due to the drag force
resulted from shear and drag which is modelled according to Ishii and Zuber (1979) as:
( )
l g l g l if D
u u u u ρ a C F F
− − = − =
8
1
gl lg
where
D
C is the drag coefficient which can be evaluated by correlation of several distinct
Reynolds number regions for individual bubbles proposed by Ishii and Zuber (1979).
3.2.1.3. Interfacial Area Density
In Eq. (3.25), interfacial momentum transfer due to the drag force is directly dependent
on the contact surface area between the two phases and is characterized by the interfacial area
per unit volume between gas and liquid phase, named as the interfacial area density a
if
. Based
on the particle model, assuming that liquid phase is continuous and the gas phase is dispersed,
the interfacial area per unit volume is then calculated based on the Sauter mean bubble
diameter d
g
given by
g
g
if
d
a
*
6α
= where
⎪
⎩
⎪
⎨
⎧
>
−
−
<
=
) ( ) ,
1
1
max(
) ( ) , max(
max min max
max
max min
*
α α α α
α
α
α α α α
α
g
g
g g
g
if
if
On the Numerical Simulation of Turbulence Modulation in TwoPhase Flows 51
The nondimensional interphase transfer coefficients can be correlated in terms of the
particle Reynolds number and is given by
l
g l g l
d U U
μ
ρ −
=
lg
Re
where µ
l
is the viscosity of the liquid phase.
3.2.2. MUSIG Model
To account for nonuniform bubble size distribution, the MUSIG model employs multiple
discrete bubble size groups to represent the population balance of bubbles. Assuming each
bubble class travel at the same mean algebraic velocity, individual number density of bubble
class i based on Kumar and Ramkrishna (1996a) can be expressed as:
( )
i
j
j i g
i
R n u
t
n
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
= ⋅ ∇ +
∂
∂
∑
(3.26)
where ( )
i
j
j
R
∑
represents the net change in the number density distribution due to
coalescence and breakup processes. The discrete bubble class between bubble volumes
i
v
and
1 + i
v is represented by the centre point of a fixed nonuniform volume distributed grid
interval. The interaction term ( ) ( )
B C B C
i
j
j
D D P P R − − + =
∑
contains the source rate of
C
P ,
B
P ,
C
D and
B
D , which are, respectively, the production rates due to coalescence and
breakup and the death rate due to coalescence and breakup of bubbles.
3.2.2.1. MUSIG Breakup Rate
The production and death rate of bubbles due to the turbulent induced breakage is
formulated as:
( )
j i j
N
i j
B
n v v Ω P :
1
∑
+ =
=
i i B
n Ω D = with
∑
=
=
N
k
ki i
Ω Ω
1
(3.27)
Here, the breakup rate of bubbles of volume
j
v into volume
i
v is modelled according
to the model developed by Luo and Svendsen (1996). The model is developed based on the
assumption of bubble binary breakup under isotropic turbulence situation. The major
difference is the daughter size distribution which has been taken account using a stochastic
breakage volume fraction f
BV
. By incorporating the increase coefficient of surface area, c
f
=
K. Mohanarangam and J.Y. Tu 52
[
3 2 /
BV
f +(1f
BV
)
2/3
1], into the breakage efficient, the breakup rate of bubbles can be obtained
as:
( )
( )
( )
dξ
d βρ
c
ξ
ξ
d
ε
C F
n α
v v Ω
l
f
ξ
j
B
j g
i j
min
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
− ×
+
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
=
−
∫ 3 / 11 3 / 5 3 / 2
1
3 / 11
2
3 / 1
2
12
exp
1
1
:
ξ ε
σ
where
j
d / λ ξ = is the size ratio between an eddy and a particle in the inertial subrange
and consequently
j min min
d / λ ξ = and C and β are determined, respectively, from
fundamental consideration of drops or bubbles breakage in turbulent dispersion systems to be
0.923 and 2.0.
3.2.2.2. MUSIG Coalescence Rate
The number density of individual bubble groups governed by coalescence can be
expressed as:
∑∑
= =
=
i
k
i
l
j i ij jki C
n n P
1 1
2
1
χ η
) /( ) (
1 1 − −
− − +
i i i k j
ν ν ν ν ν if
i k j i
ν ν ν ν < + <
−1
=
jki
η ) /( ) (
1 1 i i k j i
ν ν ν ν ν − + −
+ +
if
1 +
< + <
i k j i
ν ν ν ν
0 otherwise
∑
=
=
N
j
j i ij C
n n D
1
χ
From the physical point of view, bubble coalescence occurs via collision of two bubbles
which may be caused by wake entrainment, random turbulence and buoyancy. However, only
turbulence random collision is considered in the present study as all bubbles are assumed to
be spherical (wake entrainment becomes negligible). Furthermore, as all bubbles travel at the
same velocity in the MUSIG model, buoyancy effect is also eliminated. The coalescence rate
considering turbulent collision taken from Prince and Blanch (1990) can be expressed as:
[ ] ( )
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
− + + =
ij
ij
tj ti j i C ij
t
u u d d F
τ
π
χ exp
4
5 . 0
2 2 2
where
ij
τ is the contact time for two bubbles given by
3 / 1 3 / 2
/ ) 2 / ( ε
ij
d and
ij
t is the time
required for two bubbles to coalesce having diameter d
i
and d
j
estimated to be
) / ln( ] 16 / ) 2 / [(
0
5 . 0 3
f l ij
h h d σ ρ . The equivalent diameter d
ij
is calculated as suggested by
On the Numerical Simulation of Turbulence Modulation in TwoPhase Flows 53
Chesters and Hoffman (1982):
1
) / 2 / 2 ( (
−
+ =
j i ij
d d d . According to Prince and Blanch
(1990), for airwater systems, experiments have determined the initial film thickness h
o
and
critical film thickness h
f
at which rupture occurs as
4
10 1
−
× and
8
10 1
−
× m respectively. The
turbulent velocity u
t
in the inertial subrange of isotropic turbulence (Rotta, 1972) is given
by:
3 / 1 3 / 1
2 d u
t
ε = .
Numerical Procedure
All the transport equations are discretized using a finite volume formulation in a
generalized coordinate space, with metric information expressed in terms of area vectors. The
equations are solved on a nonstaggered grid system, wherein all primitive variables are stored
at the centroids of the mass control volumes. Thirdorder QUICK scheme is used to
approximate the convective terms, while secondorder accurate central difference scheme is
adopted for the diffusion terms. The velocity correction is realized to satisfy continuity
through SIMPLE algorithm, which couples velocity and pressure. At the inlet boundary the
particulate phase velocity is taken to be the same as the gas velocity. The concentration of the
particulate phase is set to be uniform at the inlet. At the outlet the zero streamwise gradients
are used for all variables. The wall boundary conditions are based on the model of Tu and
Fletcher (1995).
All the governing equations for both the carrier and dispersed phases are solved
sequentially at each iteration, the solution process is started by solving the momentum
equations for the gas phase followed by the pressurecorrection through the continuity
equation, turbulence equations for the gas phase, are solved in succession. While the solution
process for the particle phase starts by the solution of momentum equations followed by the
concentration then gasparticle turbulence interaction to reflect the twoway coupling, the
process ends by the solution of turbulence equation for the particulate phase. At each global
iteration, each equation is iterated, typically 3 to 5 times, using a strongly implicit procedure
(SIP).The above solution process is marched towards a steady state and is repeated until a
converged solution is obtained.
Numerical Predictions
Gas –Particle Flow
4.1. Code Verification
In this section the code is validated for mean streamwise velocities and fluctuations for
both the carrier and dispersed phases against the benchmark experimental data of Fessler and
Eaton (1995). This task is undertaken to verify the fact that these two classes of particles,
which share the same Stokes number but varied particle Reynolds number can be handled by
the code. Figure 4.1 show the backward facing step geometry used in this study, which is
similar to the one used in the experiments of Fessler and Eaton
(1995), which has got a step
height (h) of 26.7mm. As the span wise zdirection perpendicular to the paper is much larger
K. Mohanarangam and J.Y. Tu 54
than the ydirection used in the experiments, the flow is considered to be essentially two
dimensional. The backward facing step has an expansion ratio of 5:3. The Reynolds number
over the step works out to be 18,400 calculated based on the centerline velocity and step
height (h). The independency of grid over the converged solution was checked by refining the
mesh system through doubling the number of grid points along the streamwise and the lateral
directions. Simulations results revealed that the difference of the reattachment length between
the two mesh schemes is less than 3%.
x
y H
h
H=40mm h=26.7mm
x=35h
Figure 4.1. Backward facing step geometry.
4.1.1. Mean Streamwise Velocities
The mean streamwise velocities for the gas phase have been shown in Figure 4.2 for
various stations along the backwardfacing step geometry, it can be generally seen that there
is fairly good agreement with experimental findings of Fessler and Eaton (1995). This is then
followed by the streamwise velocities for the two classes of particles considered in this study,
whose properties are tabulated in Table 1. The broad varying characteristics of different
particle sizes and material properties can be unified by a single dimensionless parameter; it is
also used to quantify the particles responsivity to fluid motions and this nondimensional
parameter is the Stokes number (St) and is given by the ratio of particle relaxation time to
time that of the appropriate fluid time scale, St = t
p
/t
s
. In choosing the appropriate fluid time
scale, the reattachment length has not been considered as it varies due to addition of particles
and is not constant in this study, rather a constant length scale of five step heights, which is in
accordance to the reattachment length is used. The resulting time scale is given by t
s
=5h/U
o
,
in accordance with the experiments of Fessler and Eaton (1995).
Table 4.1 Properties of the dispersed phase particles
Nominal Diameter(μm) 150 70
Material Glass Copper
Density(kg/m
3
) 2500 8800
Stokes Number (St) 7.4 7.1
Particle Reynolds number (Re
p
) 9.0 4.0
On the Numerical Simulation of Turbulence Modulation in TwoPhase Flows 55
0.00
0.50
1.00
1.50
2.00
2.50
Distance along the step (x/h)
H
e
i
g
h
t
(
y
/
h
)
x/h=1 x/h= x/h=9
x/h=
x/h= x/h=
0
x/h=1
u/U
b
1.5
Figure 4.2. Mean streamwise gas velocities.
0.00
0.50
1.00
1.50
2.00
2.50
Distance along the step (x/h)
H
e
i
g
h
t
(
y
/
h
)
x/h=1 x/h= x/h=5
x/h= x/h=9
0
u/U
b
1.5
Figure 4.3.a. Streamwise mean velocity for 70μm copper particles.
○ Experimental Numerical
○ Experimental Numerical
K. Mohanarangam and J.Y. Tu 56
The significance of the Stokes number is that a particle with a small Stokes number
(St<<1) are found to be in near velocity equilibrium with the surrounding carrier fluid, there
by making them extremely responsive to fluid velocity fluctuations, in fact Laser Doppler
velocimetry (LDV) takes advantage of this to deduce local fluid velocity from the measured
velocity of these very small Stokes number particles, however on the other hand for a larger
Stokes number (St>>1) particles are found be no longer in equilibrium with the surrounding
fluid phase as they are unresponsive to fluid velocity fluctuations and they will pass
unaffected through eddies and other flow structures.
Figures 4.3a & 4.3b shows the mean streamwise velocity profiles for the two particle
classes considered in this study, it can be seen that there is generally a fairly good agreement
with the experimental results. From the particle mean velocity graphs of the carrier and
dispersed phases it can be inferred, that the particle streamwise velocity at the first station
x/h=2 is lower than the corresponding gas velocities, this is in lines with the fully developed
channel flow reaching the step as described in the experiments of Kulick et al (1994), wherein
the particles at the channel centerline have lower streamwise velocities than that of the fluid
as a result of crossstream mixing. However the gas velocity lags behind the particle
velocities aft of the sudden expansion as the particles inertia makes them slower to respond to
the adverse pressure gradient than the fluid.
0.00
0.50
1.00
1.50
2.00
2.50
Distance along the step (x/h)
H
e
i
g
h
t
(
y
/
h
)
x/h= x/h= x/h=7 x/h=9 x/h=1
0 u/U
b
1.5
Figure 4.3.b. Streamwise mean velocity for 150μm glass particles.
4.1.2. Mean Streamwise Fluctuations
The code has been further validated for mean streamwise fluctuations and Figure 4.4
shows the mean streamwise fluctuations for the gas phase against the experimental data. It is
seen that there is a general under prediction of the simulated data with the experimental
○ Experimental Numerical
On the Numerical Simulation of Turbulence Modulation in TwoPhase Flows 57
results and this is more pronounced towards the lower wall for a height of up to y/h≤2,
however the pattern of the simulated results have been found to be in tune with its
experimental counterpart.
0.00
0.50
1.00
1.50
2.00
2.50
Distance along the step (x/h)
H
e
i
g
h
t
(
y
/
h
)
x/h= x/h= x/h x/h x/h=1
0 u'/U
b
0.2
Figure 4.4. Fluctuating streamwise gas velocities particles.
0.00
0.50
1.00
1.50
2.00
2.50
Distance along the step (x/h)
H
e
i
g
h
t
(
y
/
h
)
x/h=2 x/h=5
x/h=7
x/h=9
x/h=12
0 u'/U
b
0.3
Figure 4.5.a. Fluctuating streamwise particle velocities for 70μm copper particles.
○ Experimental Numerical
K. Mohanarangam and J.Y. Tu 58
0.00
0.50
1.00
1.50
2.00
2.50
Distance along the step (x/h)
H
e
i
g
h
t
(
y
/
h
)
0 u' /U
b
0.3
x/h= x/h=5 x/h=7 x/h=9
x/h=1
Figure 4.5.b. Fluctuating streamwise particle velocities for 150μm glass particles.
Figures 4.5a & 4.5b show the streamwise fluctuating particle velocities for the two
different classes of particles considered, there has been a minor underprediction until stations
y/h ≤ 1, over all a fairly good agreement can be observed. It can be also seen that for y/h >
1.5, the particle fluctuating velocities are considerably larger than those of the fluid. This
again is in accordance with channel flow inlet conditions, where the particles have higher
fluctuating velocities than those of the fluid owing to crossstream mixing. All the
experimental results used for comparison of particle fluctuating velocities correspond to
maximum mass loadings of particles as reported in the experiments of Fessler & Eaton
(1999).
4.2. Results and Discussion
4.2.1. Turbulence Modulation (TM)
The plot depicted in the following sections to represent the Turbulent Modulation (TM)
of the carrier gas phase is given by the ratio of the laden flow r.m.s streamwise velocity to the
unladen r.m.s streamwise velocity. These plots signify that any turbulence modulation felt in
the carrier phase is reflected as an exit of the ratio from unity.
4.2.1.1. Analysis of Experimental Data
Plots 4.6a and 4.6b depict the experimental data as obtained from the experiments of
Fessler and Eaton (1995). Figures 4.6a&b represent the mean streamwise particle velocities
○ Experimental Numerical
On the Numerical Simulation of Turbulence Modulation in TwoPhase Flows 59
and fluctuations respectively for the two sets of dispersed phase particles i.e., copper and
glass particles. It can be well seen that the mean velocities and to a similar extent the
fluctuations for these two classes seem to behave analogous to each other. However from
figures 4.7ac, which shows the experimental TM for the carrier phase in the presence of the
dispersed phase at the same mass loading of 40% seem to behave in contrary to the above
findings. In all these stations considered in this study the glass particle seems to attenuate the
carrier phase more than copper particles. It is also be seen at the station x/h=2, the attenuation
0.00
0.50
1.00
1.50
2.00
2.50
Distance along the step (x/h)
H
e
i
g
h
t
(
y
/
h
)
x/h=2 x/h=5 x/h=7
x/h=9
0 u/U
b
1.5
Figure 4.6.a. Experimental mean streamwise particle velocity.
of the turbulence for glass particle is totally opposite in relation to copper particle for location
y/h>1.25. At station x/h=7 for certain regions along the height of the step the turbulence
attenuation for the two particle classes seems to be in phase, whereas before and after this
small region of unison the glass particle seem to attenuate more than the copper particles. At
the station x/h=14, it can be clearly seen, that there is a uniform degree of difference in
attenuation all along the step, this is more attributed to the uniform distribution of the
particles seen along the step. The maximum turbulence attenuation can be seen for the 150μm
particles, which up to 35% as reported by Fessler and Eaton (1999).
From the above analysis, it is quite clear that particles with the same Stokes number
modulate the carrier phase turbulence in a totally different fashion. This makes us conclude
that although Stokes number can be generalized to account for mean values of velocity and
fluctuations, it cannot be generalized when it comes to TM, in which case something more
than Stokes number is required to define one’s particle response to surrounding carrier phase
turbulence either to attenuate or enhance it.
70μm copper 150μm glass
K. Mohanarangam and J.Y. Tu 60
0.00
0.50
1.00
1.50
2.00
2.50
Distance along the step (x/h)
H
e
i
g
h
t
(
y
/
h
)
x/h= x/h= x/h=7
x/h=
0 u'/U
b
0.3
Figure 4.6.b. Experimental fluctuating streamwise particle velocity.
0.60
0.70
0.80
0.90
1.00
1.10
0.00 0.50 1.00 1.50 2.00 2.50
Hei ght (y/h)
T
u
r
b
u
l
e
n
c
e
M
o
d
u
l
a
t
i
o
n
(
T
M
)
x/h=2
Figure 4.7.a. Experimental Turbulence Modulation at x/h=2.
70μm copper 150μm glass
On the Numerical Simulation of Turbulence Modulation in TwoPhase Flows 61
0.60
0.70
0.80
0.90
1.00
1.10
0.00 0.50 1.00 1.50 2.00 2.50
Height (y/h)
T
u
r
b
u
l
e
n
c
e
M
o
d
u
l
a
t
i
o
n
(
T
M
)
x/h=
Figure 4.7.b. Experimental Turbulence Modulation at x/h=7.
0.60
0.70
0.80
0.90
1.00
1.10
0.00 0.50 1.00 1.50 2.00 2.50
Height (y/h)
T
u
r
b
u
l
e
n
c
e
M
o
d
u
l
a
t
i
o
n
(
T
M
)
x/h=14
Figure 4.7.c. Experimental Turbulence Modulation at x/h=14.
70μm copper 150μm glass
K. Mohanarangam and J.Y. Tu 62
4.2.2. TM & (Particle Number Density) PND Results
In this section we have tried to derive an understanding for the Turbulence Modulation
(TM) of the carrier gas phase in the presence of particles using our turbulent formulation,
along with its corresponding PND results for the two classes we have considered in this
study. The simulated results of the above two parameters are plotted along side the
experimental findings of Fessler and Eaton (1995). The experimental values for the
modulation are plotted with error bars, as significant scatter are apparent in these plots,
thereby making any variations on the order of ±5% insignificant (Fessler & Eaton, 1995).
0.00
2.00
4.00
6.00
8.00
10.00
12.00
0.00 0.50 1.00 1.50 2.00 2.50
Height (y/h)
P
a
r
t
i
c
l
e
N
u
m
b
e
r
d
e
n
s
i
t
y
(
P
N
D
)
0.00
0.20
0.40
0.60
0.80
1.00
1.20
T
u
r
b
u
l
e
n
c
e
m
o
d
i
f
i
c
a
t
i
o
n
(
T
M
)
Figure 4.8.a. Turbulence Modulation & Particle Number Density for 70μm copper particles at x/h=2.
The plots 4.8ac shows the combined numerical and experimental results of TM
(secondary axis) and PND (primary axis) for three sections along the step viz. x/H=2, 7 &14
for copper particles. It can be noted at station x/h=2 just aft of the step, for y/h<1 there exist
very few particles, but however the experimental TM shows a wavy behavior in comparison
to the simulated results which seem to behave in unison with the PND, however aft of this
section, the simulated values compare well with the experimental data. In the middle of the
step at station x/h=7, there is generally a good comparison of the experimental and simulated
values for both the TM and PND. At the exit (x/h=14) however, there seems to be general
under prediction of TM for y/h>1.5, while the PND seem to vary from under predicting to
over predicting the experimental data. Figures 4.9ac shows plots of the glass particles for the
same three sections along the step. At section x/h=2, the simulated values seem to over
predict the experimental data for y/h>1, however this distinct behavior of maximum
attenuation as reported by Fessler and Eaton (1995) occurs here, this under prediction is not
quite in terms with the PND results which seem to show a uniform distribution. A fairly good
●ExperimentalTM NumericalTM ▲ExperimentalPND NumericalPND
On the Numerical Simulation of Turbulence Modulation in TwoPhase Flows 63
agreement is with both turbulence modulation and PND can be seen for stations x/h=7 and
14.
0.00
2.00
4.00
6.00
8.00
10.00
12.00
0.00 0.50 1.00 1.50 2.00 2.50
Height (y/h)
P
a
r
t
i
c
l
e
N
u
m
b
e
r
d
e
n
s
i
t
y
(
P
N
D
)
0.00
0.20
0.40
0.60
0.80
1.00
1.20
T
u
r
b
u
l
e
n
c
e
m
o
d
i
f
i
c
a
t
i
o
n
(
T
M
)
`
Figure 4.8.b. Turbulence Modulation & Particle Number Density for 70μm copper particles at x/h=7.
It is generally seen that for the two classes of particles considered, there exist
considerably more particles in the region y/h>1 for locations x/h=2 and 7, and the particles
exhibit a uniform distribution by the time it reaches the location x/h=14 due to its uniform
spreading action, but despite a uniform distribution the turbulence attenuation is still small for
y/h<1 at the x/h=14, and this is attributed to the noneffectiveness of the particles to large
scale vortices which are reported to exist downstream of the single phase backward facing
step (Le et al., 1997). From the PND results for the two classes of particles, both simulated
and experimental plots show that there exits very few particles in the region y/h<1 before the
reattachment point and a significant spreading has taken place by, at locations x/h=9 and 14,
considering this behavior along side with the turbulence attenuation one can state that lack of
turbulence modulation is not simple due to the absence of particles but clearly a difference on
response of the turbulence in a specific region due to the presence of particles
(Fessler &
Eaton, 1995) and this has been explicitly seen in our simulated values using our turbulence
formulation mentioned in the previous section.
●ExperimentalTM NumericalTM ▲ExperimentalPND NumericalPND
K. Mohanarangam and J.Y. Tu 64
0.00
2.00
4.00
6.00
8.00
10.00
12.00
0.00 0.50 1.00 1.50 2.00 2.50
Height (y/h)
P
a
r
t
i
c
l
e
N
u
m
b
e
r
d
e
n
s
i
t
y
(
P
N
D
)
0.00
0.20
0.40
0.60
0.80
1.00
1.20
T
u
r
b
u
l
e
n
c
e
m
o
d
i
f
i
c
a
t
i
o
n
(
T
M
)
Figure 4.8.c. Turbulence Modulation & Particle Number Density for 70μm copper particles at x/h=14.
4.2.3. Effect of Particle Reynolds Number on TM
In this section, we have tried to explain the phenomenon of the carrier phase TM by
particles in relation to particle Reynolds number. Looking back at Table 4.1, it is seen that for
the glass particles the Re
p
is almost 125% more than for the copper particles and also
considering the size of the particles which is around 114% more for the glass particles in
comparison to the copper particles. The presented results on the degree of carrier phase
turbulence attenuation also seem to throw some interesting pattern especially in regards to
copper and glass particles, which almost share the same Stokes number. Based on this
similarity one would expect that their response to the turbulence be the same, although this
fact seems to be quite in lines with the particle mean and fluctuating velocities but their
behavior in regards to turbulence modulation paint a totally different picture, which makes us
conclude that the turbulence attenuation has a direct link of the particles onto the turbulence
rather a derived effect on the mean flow modifications (Fessler and Eaton, 1999). And this
attenuation cannot follow suit with increasing loading and particle size as there will be an
increase in the turbulence as reported by Hetsroni (1989), it is also interesting to note from
the same author that apart from the Stokes number, particle Reynolds number plays an
important role in the systematic behavior of the particles, the same has been reported from the
experimental findings of Fessler and Eaton
(1995), this is in conformity to our simulated
results wherein the two classes of copper and glass behave differently although their particle
Stokes number remain the same.
●ExperimentalTM NumericalTM ▲ExperimentalPND NumericalPND
On the Numerical Simulation of Turbulence Modulation in TwoPhase Flows 65
0.00
2.00
4.00
6.00
8.00
10.00
12.00
0.00 0.50 1.00 1.50 2.00 2.50
Height(y/h)
P
a
r
t
i
c
l
e
N
u
m
b
e
r
d
e
n
s
i
t
y
(
P
N
D
)
0.00
0.20
0.40
0.60
0.80
1.00
1.20
T
u
r
b
u
l
e
n
c
e
M
o
d
i
f
i
c
a
t
i
o
n
(
T
M
)
Figure 4.9.a. Turbulence Modulation & Particle Number Density for 150μm glass particles at x/h=2.
It is known from previous studies that there is a physical increase in the carrier phase
turbulence and this is attributed to the wake formation behind these large particles similar to
the ones behind the cylinders encountered in single phase flows. Experimental studies from
Eaton et al (1999) state that for Re
p
around 10, the modulation is caused due to the strong
asymmetric wake distortion extending many particle diameters downstream, whereas for
Re
p
=1, the flow distortion is highly localized to a nearly symmetric region extending only a
few particle diameters. Taking this into consideration one would expect a significant
enhancement of the carrier phase turbulence for the glass particles but in contrast significant
attenuation have taken place, the maximum attenuation is noted at location x/h=2 and the
attenuation of glass particles is roughly about 44% more than its counterpart copper particles.
This behavior of the glass particle to attenuate rather than enhance the turbulence of the
carrier phase can be explained by the fact that, due to strong asymmetric wake, the distortion
caused by this class of particles is dissipated more easily within the carrier gas phase, this is
in sharp contrast to the parallel experiments run by Sato et al (2000) using water as the carrier
phase fluid, although similar Stokes number were achieved, much higher Re
p
was realized,
this in total showed significant vortices behind the particles and also an enhancement in the
carrier phase turbulence, which is primarily caused due to the different density ratios between
the carrier and dispersed phases, which makes one to justify that Stokes number alone will not
be able to describe the parameter space of the dispersed twophase flow problems, Re
p
should
also be taken into consideration.
●ExperimentalTM NumericalTM ▲ExperimentalPND NumericalPND
K. Mohanarangam and J.Y. Tu 66
0.00
2.00
4.00
6.00
8.00
10.00
12.00
0.00 0.50 1.00 1.50 2.00 2.50
Height(y/h)
P
a
r
t
i
c
l
e
N
u
m
b
e
r
d
e
n
s
i
t
y
(
P
N
D
)
0.00
0.20
0.40
0.60
0.80
1.00
1.20
T
u
r
b
u
l
e
n
c
e
M
o
d
i
f
i
c
a
t
i
o
n
(
T
M
)
Figure 4.9.b. Turbulence Modulation & Particle Number Density for 150μm glass particles at x/h=7.
0.00
2.00
4.00
6.00
8.00
10.00
12.00
0.00 0.50 1.00 1.50 2.00 2.50
Height(y/h)
P
a
r
t
i
c
l
e
N
u
m
b
e
r
d
e
n
s
i
t
y
(
P
N
D
)
0.00
0.20
0.40
0.60
0.80
1.00
1.20
T
u
r
b
u
l
e
n
c
e
M
o
d
i
f
i
c
a
t
i
o
n
(
T
M
)
Figure 4.9.c. Turbulence Modulation & Particle Number Density for 150μm glass particles at x/h=14.
Liquid–Particle Flow
This section describes the numerical investigation of the turbulent gasparticle flow over
a backward facing step, using the EulerianEulerian model as described in the previous
section. The numerical procedure is almost the same as of section 3.1. However, two
●ExperimentalTM NumericalTM ▲ExperimentalPND NumericalPND
On the Numerical Simulation of Turbulence Modulation in TwoPhase Flows 67
backwardfacing step geometries are utilized in this study one to validate the GP flow and the
other for the LP flow. Figure 5.1a shows the backward facing step geometry, which is similar
to the one used in the experiments of Fessler and Eaton (1995), comprising of a step height
(h) of 26.7mm. As the span wise zdirection perpendicular to the paper is much larger than
the ydirection used in the experiments, the flow is considered to be essentially two
dimensional. The backward facing step has an expansion ratio of 5:3. The Reynolds number
over the step works out to be 18,400 calculated based on the centerline velocity and step
height (h). The experimental set up of Founti and Klipfel (1998) consisted of a pipe flow with
a sudden expansion ratio of 1:2, with a step height of 25.5mm, as depicted in figure 5.1b,
working at a Reynolds number of 28,000. The summary of the flow conditions along with the
properties of the dispersed phase particles used in this study are summarized in Table 5.1.
x
y H
h
H=40mm h=26.7mm
x=35h
Figure 5.1.a. Backward facing step geometry (Fessler & Eaton; 1995).
y
x
H
h
H=25.5mm h=25.5mm
x=35h
Figure 5.1.b. Backward facing step geometry
(Founti & Klipfel; 1998).
Table 5.1. Flow properties of carrier and dispersed phases for LP & GP flows
Parameters
GasParticle (GP) flow
Fessler & Eaton (1995)
LiquidParticle (LP) flow
Founti & Klipfel (1998)
Reynolds number (Re
h
) 18,700 28,000
Geometry BFS Pipe (BFS)
Continuous Phase Air Diesel Oil
Mass loading 20% 15%
Particle Density 2500 2500
Particle Diameter 150 micron 450 micron
Phasedensity ratio 2137:1 3.0:1
K. Mohanarangam and J.Y. Tu 68
5.1. Analysis of Experimental Data
In this section, the experimental data for mean velocities and fluctuations are analysed, in
order to understand the particle behaviour in relation to its carrier phase namely the gas and
the liquid. For this purpose, three sections were selected aft of the sudden expansion one near
the step, another almost at the middle and the other farther away from the step nearing the exit
of the geometry.
An important, dimensionless scaling parameter in defining on how the fluidparticle
behave with the flow field is the Stokes number (St), which is given by the ratio of the
particle relaxation time to a time characteristic of the fluid motion, i.e., St = t
p
/t
s
. This
determines the kinetic equilibrium of the particles with the surrounding liquid. In choosing,
the fluid time scale t
s
, amidst the complexity of having two different geometries with
different expansion ratios and also with the reattachment length varying with the addition
of the particles, the fluid time scales were determined by t
s
=5h/U
o
in lieu with the
experimental conditions of Fessler and Eaton (1997). A small stokes number (St << 1)
signifies that the particles are in near velocity equilibrium with the carrier fluid. For larger
stokes number (St >> 1) particles are no longer in equilibrium with the surrounding fluid
phase, which will be exemplified in the later sections. Based on the above definition, the
Stokes number for the GP and the LP flow examined in our study worked out to be 14.2
and 0.59 respectively.
Figures 5.2ac shows the mean streamwise velocities of the liquid and particle phase
flows as presented from the experiments of Founti & Klipfel (1998). It can be seen at section
x/h=0.7, the particles seem to exhibit a higher negative velocity for a section of y/h<1, after
this height the particles seem to surpass the liquid velocities for the section y/h>1, while for a
small region at the proximity of the step they seem to exhibit a homogeneous behaviour. At
section x/h=7.8, which is almost the middle section, the liquid phase have a higher negative
velocity than that of the particulate phase, this behaviour seems to follow for a height of up to
y/h=1, after which the both the phases seem to behave in unison. The final section is x/h=
15.7, herein again the liquid seems to exhibit a higher velocity than that of the particles, with
unified flow at the top.
Figures 5.3ac shows the mean velocities of the GP flow as obtained from the
experiments of Fessler and Eaton (1999)
.
At section x/h=2, just aft of the step the particle
velocities seem to lag behind the gas velocities, whereas at section x/h=7 in the middle of
the geometry, the particle velocities seem to ‘catch up’ with that of the gas phase and their
velocities are more or less the same, however at the exit of the backwardfacing step
geometry (x/h=14), a clear marked difference in velocities is observed, wherein the
particles seem to overtake the gas due to its inertial. From the two sets of experimental
results outlined above, one could observe that at near the inlet sections, where the re
circulation is quite predominant for both the cases, particles seem to lag behind the gas for
the GP flows, where as the particles seem to exhibit a higher inertial with respect to the LP
flows, with particles leading throughout the height of the step. Overall, with respect to the
magnitude of the mean velocities, the particle seem to exhibit more or less a change in the
pattern from ‘lead’ to ‘lag’, as we proceed along the step for LP flows, whereas the
particles seem to exhibit the opposite pattern from ‘lag’ to ‘lead’ for the GP flows along the
step. The presence of a clear shear layer just aft of the step is quite prominent for the
On the Numerical Simulation of Turbulence Modulation in TwoPhase Flows 69
particles at sections all along the step for LP flow, which in single phase is the major source
of turbulence generation due to shear.
In order to understand this behaviour better we now turn to the mean streamwise
fluctuations for these two kinds of flows. Figures 5.4ac shows the fluctuating velocities of
LP flows and it can be seen that at the entry the particles seem to ‘lag’ behind the liquid for a
height of y/h>1, while the showing an increase with respect to liquid in the lower part. At the
middle section considered, for a height of about y/h>1 they exhibit a homogenous flow
behaviour, whereas at the lower part the particles again seem to exceed its liquid counterpart.
Whereas near the exit, both the continuous and dispersed phases have almost the same pattern
prompting to the fact that the particle ‘catch up’ with the liquid phase, mimicking a
homogenous flow pattern.
Figures 5.5ac show the fluctuation results for the GP flow, at the entry section x/h=2, it
can be seen that the particulate phase has a higher fluctuation in comparison to the gas phase,
while at the middle section x/h=7, the gas phase seem to catch up with the particulate phase,
while at the exit at x/h=14, it is observed that both the dispersed and the continuous phase
seem to fluctuate in unison. From the above experimental results of the fluctuation, it can be
ascertained that the particle ‘lag’ behind with respect to the continuous phase in terms of LP
flows, whereas they ‘lead’ in terms of the gasparticle flows.
0.00
0.25
0.50
0.75
1.00
1.25
1.50
1.75
2.00
0.20 0.00 0.20 0.40 0.60 0.80 1.00
x/h=0.7
y
/
h
u/U
o
Figure 5.2.a. Experimental mean streamwise velocities at x/h=0.7 for liquidparticle flow.
Liquid Particle
K. Mohanarangam and J.Y. Tu 70
0.00
0.25
0.50
0.75
1.00
1.25
1.50
1.75
2.00
0.20 0.00 0.20 0.40 0.60 0.80 1.00
x/h=7.8
y
/
h
u/U
o
Figure 5.2.b. Experimental mean streamwise velocities at x/h=7.8 for liquidparticle flow.
0.00
0.25
0.50
0.75
1.00
1.25
1.50
1.75
2.00
0.20 0.00 0.20 0.40 0.60 0.80 1.00
y
/
h
u/U
o
x/h=15.7
Figure 5.2.c. Experimental mean streamwise velocities at x/h=15.7 for liquidparticle flow.
Liquid Particle
On the Numerical Simulation of Turbulence Modulation in TwoPhase Flows 71
0.00
0.50
1.00
1.50
2.00
2.50
0.25 0.00 0.25 0.50 0.75 1.00
x/h=2
y
/
h
u/U
o
Figure 5.3.a. Experimental mean streamwise velocities at x/h=2 for gasparticle flow.
0.00
0.50
1.00
1.50
2.00
2.50
0.00 0.25 0.50 0.75 1.00
x/h=7
y
/
h
u/U
o
Figure 5.3.b. Experimental mean streamwise velocities at x/h=7 for gasparticle flow.
Gas Particle
K. Mohanarangam and J.Y. Tu 72
0.00
0.50
1.00
1.50
2.00
2.50
0.00 0.20 0.40 0.60 0.80 1.00
x/h=14
y
/
h
u/U
o
Figure 5.3.c. Experimental mean streamwise velocities at x/h=14 for gasparticle flow.
0.00
0.25
0.50
0.75
1.00
1.25
1.50
1.75
2.00
0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16
u' /U
o
y
/
h
x/h=0.7
Figure 5.4.a. Experimental mean fluctuating velocities at x/h=0.7 for liquidparticle flow.
Liquid Particle
Gas Particle
On the Numerical Simulation of Turbulence Modulation in TwoPhase Flows 73
0.00
0.25
0.50
0.75
1.00
1.25
1.50
1.75
2.00
0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16
u' /U
o
y
/
h
x/h=7.8
Figure 5.4.b. Experimental mean fluctuating velocities at x/h=7.8 for liquidparticle flow.
0.00
0.25
0.50
0.75
1.00
1.25
1.50
1.75
2.00
0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16
u' /U
o
y
/
h
x/h=15.7
Figure 5.4.c. Experimental mean fluctuating velocities at x/h=15.7 for liquidparticle flow.
Liquid Particle
K. Mohanarangam and J.Y. Tu 74
0.00
0.50
1.00
1.50
2.00
2.50
0.00 0.05 0.10 0.15 0.20
x/h=2
y
/
h
u' /U
o
Figure 5.5.a. Experimental mean fluctuating velocities at x/h=2 for gasparticle flow.
0.00
0.50
1.00
1.50
2.00
2.50
0.00 0.05 0.10 0.15 0.20
x/h=7
y
/
h
u' /U
o
Figure 5.5.b. Experimental mean fluctuating velocities at x/h=7 for gasparticle flow.
Gas Particle
On the Numerical Simulation of Turbulence Modulation in TwoPhase Flows 75
0.00
0.50
1.00
1.50
2.00
2.50
0.00 0.05 0.10 0.15 0.20
x/h=14
y
/
h
u' /U
o
Figure 5.5.c. Experimental mean fluctuating velocities at x/h=14 for gasparticle flow.
5.2. Numerical Code Validation
In this section the code is validated for mean streamwise velocities and fluctuations for
both the carrier and dispersed phases against the benchmark experimental data of Fessler and
Eaton (1995) for GP flow and the experimental data of Founti and Klipfel (1998) for the LP
flow. This task is undertaken to verify the fact that particulate flows with two varied carrier
phases can be handled by the code.
The ability of the numerical code to replicate the experimental results of GP (Fessler
and Eaton, 1999)
and LP (Founti and Klipfel, 1998) flows are tested. Figure 5.6a shows the
numerical findings of single phase (Diesel oil) mean velocities against the experimental
data, although the overall behaviour is replicated numerically there have been some under
prediction for a height of y/h>1 for midsection of the geometry, while a minor over
prediction is felt along the entire height at section x/h=15.7. Figure 5.6b shows the
fluctuating liquid velocities along the step compared against the experimental findings,
there have been some minor under prediction for a height of y/h<1 at some sections, the
majority show a good comparison with the experimental data. Figure 5.6c and figure 5.6d
depicts the experimental and numerical comparison of particle mean and fluctuating
velocities and it can be seen that overall numerical results have a good agreement with the
experimental data.
Gas Particle
K. Mohanarangam and J.Y. Tu 76
0.00
0.50
1.00
1.50
2.00
Dist ance along t he st ep (x/h)
H
e
i
g
h
t
(
y
/
h
)
x/h=0 x/h=5 x/h=7 x/h=11 x/h=15
0 1.5 u/U
Figure 5.6.a. Axial liquid velocities along the step for LP flows.
0.00
0.50
1.00
1.50
2.00
Distance along the step (x/h)
H
e
i
g
h
t
(
y
/
h
)
u'/U
o
0 0.
x/h=0.7 x/h=5.9
x/h=7. x/h=11. x/h=15.
Figure 5.6.b. Fluctuating axial liquid velocities along the step for LP flows.
○ Experimental Numerical
On the Numerical Simulation of Turbulence Modulation in TwoPhase Flows 77
0.00
0.50
1.00
1.50
2.00
Distance along the step (x/h)
H
e
i
g
h
t
(
y
/
h
)
u/U
o
0 1.5
x/h=0.7 x/h=5.9 x/h=7.8 x/h=11.8 x/h=15.7
Figure 5.6.c. Axial particle velocities along the step for LP flows.
0.00
0.50
1.00
1.50
2.00
Distance along the step (x/h)
H
e
i
g
h
t
(
y
/
h
)
u'/U 0 0.3
x/h=0.
x/h=5.
x/h=7. x/h=11. x/h=15.
Figure 5.6.d. Fluctuating axial particle velocities along the step for LP flows.
○ Experimental Numerical
K. Mohanarangam and J.Y. Tu 78
0.00
0.50
1.00
1.50
2.00
2.50
Distance along the step (x/h)
H
e
i
g
h
t
(
y
/
h
)
0
u/U
o 1.5
x/h=2 x/h=5 x/h=7 x/h=9 x/h=14
Figure 5.7.a. Streamwise gas velocities along the step for GP flows.
0.00
0.50
1.00
1.50
2.00
2.50
Dist ance along t he st ep (x/h)
H
e
i
g
h
t
(
y
/
h
)
x/h= x/h= x/h= x/h x/h=1
0
u' /U
o
0.2
Figure 5.7.b. Fluctuating streamwise gas velocities along the step for GP flows.
○ Experimental Numerical
On the Numerical Simulation of Turbulence Modulation in TwoPhase Flows 79
0.00
0.50
1.00
1.50
2.00
2.50
Distance along the step (x/h)
H
e
i
g
h
t
(
y
/
h
)
x/h= x/h= x/h= x/h=9 x/h=1
4
0 u/U
o
1.5
Figure 5.7.c. Streamwise mean velocity for 150μm glass particles.
0.00
0.50
1.00
1.50
2.00
2.50
Distance along the step (x/h)
H
e
i
g
h
t
(
y
/
h
)
0 u' /U
o
0.3
x/h=2 x/h=5 x/h=7 x/h=9
x/h=1
Figure 5.7.d. Fluctuating streamwise particle velocities for 150μm glass particles.
○ Experimental Numerical
K. Mohanarangam and J.Y. Tu 80
With the LP flow showing satisfactory agreement, the code is further validated to
substantiate the numerical findings of GP flows against the experimental data of Fessler and
Eaton (1999). Figure 5.7a shows the numerical comparison against the experimental findings
and it can be seen that a fairly good agreement have been obtained between the experimental
and its numerical counterpart, minor under prediction have been observed within a height of
y/h<1 for the first two sections considered along the step. Figure 5.7b shows the mean
streamwise fluctuating velocities along the step and it can be there is a general under
prediction along the length of the step with some minor over prediction for a height of y/h<1
at section x/h=2. But however, the trend as seen from the numerical simulation is in lines with
the experimental data. Figure 5.7c shows the comparison of the mean streamwise particle
velocities for the 150µm glass particle against its experimental counterpart, it can be seen that
there is a good agreement between the experimental and the numerical findings all along the
step. The simulated streamwise fluctuating velocities are compared against the experimental
findings in figure 5.7d and it can be seen that a fairly good agreement is felt along various
sections of the backwardfacing step geometry.
From the comparisons of the GP flows, it is worth while to note, that mean velocities of
the particles are lower at x/h=2 at the entry of the step than the gas phase, this is similar to the
fully developed channel flow before the step, as reported in the experiments of Kulick et al.
(1994), wherein the particles at the channel centreline exhibit lower streamwise velocities
than that of the fluid as a result of crossstream mixing. However, the gas velocity lags behind
the particles aft of the step as particle inertia is slower to respond to the adverse pressure
gradient than that of the fluid. At the fluctuation level, the particles exhibit higher values than
the gas which again is attributed to the crossstream mixing (Kulick et al., 1994).
5.3. Results and Discussion
5.4.1. Particle Response Mean Velocity Level
With the numerical code apt to replicate the experimental findings of the GP and LP flows,
we now embark on a mission to study the response of the particles to the surrounding carrier
phases. In order to proceed with this endeavour, four different Stokes numbers, by invariably
changing the particle response time have been chosen for both the GP and the LP flows. For this
numerical experiment the step geometry of Fessler and Eaton (1995) has been adopted, while
the inlet velocity for the carrier and the particulate phase corresponds to that of the experimental
conditions of Founti and Klipfel (1998). The four different Stokes number chosen corresponds
to 0.05, 0.5, 2.0 and 6.0. Particles with Stokes number 0.05 acts as tracers to the carrier phase
and are widely used by experimentalists for PIV/LDA study. The Stokes number of 0.5
corresponds to unveil realistically the response of particles for St≤1. Stokes number of 2.0 was
chosen such as to fall within the range of overshoot phenomena (Chein and Chung, 1987;
Hishida et al., 1992; Ishima et al., 1993), wherein the particles is said to disperse more readily
than the fluid, while higher Stokes number of 6.0 is to study the independent responsitivity of
the particles in relation to the two different carrier phases namely the air and diesel oil. In order
to represent the results more qualitatively 12 interrogation points made up of a matrix of three
sections (x/h=2, 7 & 14) along the length of the step and four sections along the height of the
step (y/h=0.5, 1.0, 1.5, 2.0) have been considered.
On the Numerical Simulation of Turbulence Modulation in TwoPhase Flows 81
u
p
/U
o
y
/
h
0 0.2 0.4 0.6 0.8
0.00
0.50
1.00
1.50
2.00
2.50
St=0.05
St=0.50
St=2.00
St=6.00
Figure 5.8.a. Mean streamwise particle velocities for varying Stokes number for LP Flows.
TKP/U
o
2
y
/
h
0 0.005 0.01 0.015 0.02 0.025 0.03
0.00
0.50
1.00
1.50
2.00
2.50
St=0.05
St=0.50
St=2.00
St=6.00
Figure 5.8.b. Fluctuating streamwise particle velocities for varying Stokes number for GP flows.
Figure 5.8a shows the mean streamwise particle velocities along a section for the LP flows for
varying Stokes number, the monitoring stations along the height of the step has been depicted
with the help lines running with constant y/h values. Figure 5.8b also shows the monitoring
K. Mohanarangam and J.Y. Tu 82
stations of particle fluctuation for the GP flows along varying Stokes number. From both the
figures shown here, the mean velocities as well as the fluctuations increase with a
corresponding increase in Stokes number for LP and GP flow respectively.
0.20
0.40
0.60
0.80
1.00
1.20
0.00 1.00 2.00 3.00 4.00 5.00 6.00
x/h=2; y/h=0.5
St
u
p
/
U
o
Figure 5.9.a. Mean streamwise particle velocities for varying Stokes number along the height of the
step for x/h=2 & y/h=0.5.
0.20
0.40
0.60
0.80
1.00
1.20
0.00 1.00 2.00 3.00 4.00 5.00 6.00
x/h=2; y/h=1.0
St
u
p
/
U
o
Figure 5.9.b. Mean streamwise particle velocities for varying Stokes number along the height of the
step for x/h=2 & y/h=1.0.
On the Numerical Simulation of Turbulence Modulation in TwoPhase Flows 83
0.20
0.40
0.60
0.80
1.00
1.20
0.00 1.00 2.00 3.00 4.00 5.00 6.00
x/h=2; y/h=1.5
St
u
p
/
U
o
Figure 5.9.c. Mean streamwise particle velocities for varying Stokes number along the height of the
step for x/h=2 & & y/h=1.5.
0.20
0.40
0.60
0.80
1.00
1.20
0.00 1.00 2.00 3.00 4.00 5.00 6.00
x/h=2; y/h=2.0
St
u
p
/
U
o
Figure 5.9.d. Mean streamwise particle velocities for varying Stokes number along the height of the
step for x/h=2 & y/h=2.0.
In this part of the results and discussion section the response of the particles to the mean
flow with two different carrier mediums are investigated in relation to varying Stokes
number. Three sections along the step and four along the height of the step have been
investigated. Figures 5.9ad shows the plot of the normalized mean particle velocities for
x/h=2 section along varying Stokes number across the step height. The solid lines with circles
in the figure depict the particle behaviour in GP flow, while the broken lines with squares
depict again the particle behaviour but in LP flow. It can be seen from the figures that the
mean particle velocities increase with a corresponding increase in its Stokes number for both
K. Mohanarangam and J.Y. Tu 84
the carrier phases with the maximum particulate velocity occurring for the LP flows as one
progresses along the height. There is along a steady increase in the particulate velocities along
the height of the step for up to y/h=1.5 after which there is a meagre drop for y/h=2.0 where
in the wall boundary conditions try to retard the flow. The particle velocities for the GP tend
to overtake its counterpart LP flow at section y/h=2.0 but however not for the maximum
Stokes number considered in our study.
0.30
0.40
0.50
0.60
0.70
0.80
0.00 1.00 2.00 3.00 4.00 5.00 6.00
x/h=7; y/h=0.5
St
u
p
/
U
o
Figure 5.10.a. Mean streamwise particle velocities for varying Stokes number along the height of the
step for x/h=7 & y/h=0.5.
0.30
0.40
0.50
0.60
0.70
0.80
0.00 1.00 2.00 3.00 4.00 5.00 6.00
x/h=7; y/h=1.0
St
u
p
/
U
o
Figure 5.10.b. Mean streamwise particle velocities for varying Stokes number along the height of the
step for x/h=7 & y/h=1.0.
u
p
/
U
o
On the Numerical Simulation of Turbulence Modulation in TwoPhase Flows 85
0.30
0.40
0.50
0.60
0.70
0.80
0.00 1.00 2.00 3.00 4.00 5.00 6.00
x/h=7; y/h=1.5
St
u
p
/
U
o
Figure 5.10.c. Mean streamwise particle velocities for varying Stokes number along the height of the
step for x/h=7 & y/h=1.5.
0.30
0.40
0.50
0.60
0.70
0.80
0.00 1.00 2.00 3.00 4.00 5.00 6.00
x/h=7; y/h=2.0
St
u
p
/
U
o
Figure 5.10.d. Mean streamwise particle velocities for varying Stokes number along the height of the
step for x/h=7 & y/h=2.0.
Figures 5.10ad shows the mean particulate velocities at section x/h=7 along the height of
the step geometry, it can also be seen here that the velocities keep increasing for a height of
about y/h=1.5 after which there is a small drop due to the wall interference. However in this
section of the step geometry, it can seen that the particulate velocities for the GP flows most
of the time exceed than that of the particles in the LP except for section y/h=0.5. This is
attributed to the fact that the GP flow particles are free to move in a less restricted
K. Mohanarangam and J.Y. Tu 86
environment like air while particles in the LP flow move in a highly viscous environment,
which fundamentally restricts it motion.
0.40
0.50
0.60
0.70
0.00 1.00 2.00 3.00 4.00 5.00 6.00
x/h=14; y/h=0.5
St
u
p
/
U
o
Figure 5.11.a. Mean streamwise particle velocities for varying Stokes number along the height of the
step for x/h=14 & y/h=0.5.
0.40
0.50
0.60
0.70
0.00 1.00 2.00 3.00 4.00 5.00 6.00
x/h=14; y/h=1.0
St
u
p
/
U
o
Figure 5.11.b. Mean streamwise particle velocities for varying Stokes number along the height of the
step for x/h=14 & y/h=1.0.
On the Numerical Simulation of Turbulence Modulation in TwoPhase Flows 87
0.40
0.50
0.60
0.70
0.00 1.00 2.00 3.00 4.00 5.00 6.00
x/h=14; y/h=1.5
St
u
p
/
U
o
Figure 5.11.c. Mean streamwise particle velocities for varying Stokes number along the height of the
step for x/h=14 & y/h=1.5.
0.40
0.50
0.60
0.70
0.00 1.00 2.00 3.00 4.00 5.00 6.00
x/h=14; y/h=2.0
St
u
p
/
U
o
Figure 5.11.d. Mean streamwise particle velocities for varying Stokes number along the height of the
step for x/h=14 & y/h=2.0.
Figures 5.11ad shows the plot of particle velocities for the section x/h=14 along the step,
here also it can be seen that similar trend corresponding to two previous sections along the
step is felt, with the increase in the particle velocities along the height and also with the
particles in GP flow exceeding than that of the LP flow. For all the three x/h sections
considered along the length of the step it can be seen that there is decrease in the particle
velocities, which is attributed to the fact that particle losing their momentum as they travel
along the step of the geometry.
K. Mohanarangam and J.Y. Tu 88
5.4.2. Particle ResponseTurbulence Level
In this part of the results and discussion section, particles turbulent behaviour is
investigated for two variants of carrier phase flows namely the GP and the LP flows. Here
again three sections along the length of the step and four along its height have been
considered. Figures 5.12ad shows the turbulent kinetic energy of the particles normalized by
its free stream velocity at section x/h=2 for four different heights and it can be seen that for
all sections considered there has been an increase in the particle kinetic energy with a
subsequent increase in the Stokes number for the GP flows but however for the LP flows
there has been a decrease with subsequent increase in Stokes number. It can also be seen
along the height of the step a decrease in the particulate kinetic energy is pronounced for both
GP as well as LP flows.
Figures 5.13ad shows the particulate turbulent kinetic energy for the section x/h=7 of the
step. While there have been a steady increase in the kinetic energy with a rise in the Stokes
number for the GP flows, they seem to work in the opposite fashion for the LP flows which
show a decrease with increase in Stokes number. It can also be seen that the magnitude of the
kinetic energy is a fold less than at the section x/h=2 and that also a decrease is felt with the
increase in the height of the step for the corresponding Stokes number. Figures 5.14ad also
shows a similar pattern for the section x/h=14 at the farther end near the exit of the geometry.
The magnitude of the particulate turbulent kinetic energy is lesser than the previous section of
x/h=7 and x/h=2. However, the turbulent kinetic energy of the particles seems to decrease
with the subsequent increase in the Stokes number for LP flows.
0.000
0.010
0.020
0.030
0.040
0.050
0.00 1.00 2.00 3.00 4.00 5.00 6.00
x/h=2; y/h=0.5
T
K
P
/
U
o
2
Figure 5.12.a. Fluctuating streamwise particle velocities for varying Stokes number along the height of
the step for x/h=2 & y/h=0.5.
St
On the Numerical Simulation of Turbulence Modulation in TwoPhase Flows 89
0.000
0.010
0.020
0.030
0.040
0.050
0.00 1.00 2.00 3.00 4.00 5.00 6.00
x/h=2; y/h=1.0
T
K
P
/
U
o
2
Figure 5.12.b. Fluctuating streamwise particle velocities for varying Stokes number along the height of
the step for x/h=2 & y/h=1.0.
0.000
0.010
0.020
0.030
0.040
0.050
0.00 1.00 2.00 3.00 4.00 5.00 6.00
x/h=2; y/h=1.5
T
K
P
/
U
o
2
Figure 5.12.c. Fluctuating streamwise particle velocities for varying Stokes number along the height of
the step for x/h=2 & y/h=1.5.
St
St
K. Mohanarangam and J.Y. Tu 90
0.000
0.010
0.020
0.030
0.040
0.050
0.00 1.00 2.00 3.00 4.00 5.00 6.00
x/h=2; y/h=2.0
T
K
P
/
U
o
2
Figure 5.12.d. Fluctuating streamwise particle velocities for varying Stokes number along the height of
the step for x/h=2 & y/h=2.0.
0.000
0.005
0.010
0.015
0.020
0.025
0.030
0.00 1.00 2.00 3.00 4.00 5.00 6.00
x/h=7; y/h=0.5
T
K
P
/
U
o
2
Figure 5.13.a. Fluctuating streamwise particle velocities for varying Stokes number along the height of
the step for x/h=7 & y/h=0.5.
St
St
On the Numerical Simulation of Turbulence Modulation in TwoPhase Flows 91
0.000
0.005
0.010
0.015
0.020
0.025
0.030
0.00 1.00 2.00 3.00 4.00 5.00 6.00
x/h=7; y/h=1.0
T
K
P
/
U
o
2
Figure 5.13.b. Fluctuating streamwise particle velocities for varying Stokes number along the height of
the step for x/h=7 & y/h=1.0.
0.000
0.005
0.010
0.015
0.020
0.025
0.030
0.00 1.00 2.00 3.00 4.00 5.00 6.00
x/h=7; y/h=1.5
T
K
P
/
U
o
2
Figure 5.13.c. Fluctuating streamwise particle velocities for varying Stokes number along the height of
the step for x/h=7 & y/h=1.5.
St
St
K. Mohanarangam and J.Y. Tu 92
0.000
0.005
0.010
0.015
0.020
0.025
0.030
0.00 1.00 2.00 3.00 4.00 5.00 6.00
x/h=7; y/h=2.0
St
T
K
P
/
U
o
2
Figure 5.13.d. Fluctuating streamwise particle velocities for varying Stokes number along the height of
the step for x/h=7 & y/h=2.0.
0.000
0.005
0.010
0.015
0.020
0.00 1.00 2.00 3.00 4.00 5.00 6.00
x/h=14; y/h=0.5
T
K
P
/
U
o
2
Figure 5.14.a. Fluctuating streamwise particle velocities for varying Stokes number along the height of
the step for x/h=14 & y/h=0.5.
St
St
On the Numerical Simulation of Turbulence Modulation in TwoPhase Flows 93
0.000
0.005
0.010
0.015
0.020
0.00 1.00 2.00 3.00 4.00 5.00 6.00
x/h=14; y/h=1.0
T
K
P
/
U
o
2
Figure 5.14.b. Fluctuating streamwise particle velocities for varying Stokes number along the height of
the step for x/h=14 & y/h=1.0.
0.000
0.005
0.010
0.015
0.020
0.00 1.00 2.00 3.00 4.00 5.00 6.00
x/h=14; y/h=1.5
T
K
P
/
U
o
2
Figure 5.14.c. Fluctuating streamwise particle velocities for varying Stokes number along the height of
the step for x/h=14 & y/h=1.5.
St
St
K. Mohanarangam and J.Y. Tu 94
0.000
0.005
0.010
0.015
0.020
0.00 1.00 2.00 3.00 4.00 5.00 6.00
x/h=14; y/h=2.0
T
K
P
/
U
o
2
Figure 5.14.d. Fluctuating streamwise particle velocities for varying Stokes number along the height of
the step for x/h=14 & y/h=2.0.
5.4.3. Summary of Particulate Responsitivity
The above two sections which outlined the particle response at the mean velocity level
and at the turbulence level shows that the mean velocity of the particles increase with a
subsequent increase in the Stokes number for both the carrier phases namely the gas and the
liquid. The mean particulate velocity not only increase with the Stokes number but is also
higher than its corresponding carrier phase velocities for the three Stokes number viz St=0.5,
2.0 & 6.0 considered in our study. This is quite in lines with the recent experimental data of
Ishima et al (2007) and the phenomenon is explained with the help of the particle terminal
velocity which gives a rough approximation as a percentage of how much the particle velocity
exceeds the carrier phase velocity. The other reason for the particle velocity to lead the carrier
phase is the attribute of the particulate phase to respond slowly to the adverse pressure
gradient dominant in shear flow geometries like backward facing step, in lieu to the carrier
phase.
The particle turbulent kinetic energy plots for both the GP and the LP flows depict the
response of the particles at the turbulence level, across these plots it can be summarized that
while there is a steady increase in the particulate turbulence for the GP flows with successive
increase in Stokes number, with some sections showing even a 100% increase between the
minimum and the maximum Stokes number considered. However, for the LP flows, the
magnitude of the increase in the particulate turbulence across the increasing Stokes number is
not as characteristic as its counterpart. Across the same sections for LP flows the majority of
the trend shows a decrease after which they more of less remain a constant. Previous studies
of liquidparticle flows in vertical channel (Ishima et al., 2007; Borowsky & Wei, 2007)
St
On the Numerical Simulation of Turbulence Modulation in TwoPhase Flows 95
shows that with the increase in the Stokes number there is usually a corresponding increase in
the particulate turbulence, however the flow considered in this study is a shear flow geometry,
which basically depicts a totally different flow feature unlike the simple channel geometry.
In these lines even the Turbulence Modulation (TM) of the shear flow geometry for the
well established gasparticle flows using the same set of experimental data employed in this
study
does not seem to well correspond with the models been employed and formulated for
the vertical channel flows (Mohanrangam & Tu, 2007). Thereby, given the complexity of the
problem much deeper understanding and experimental data may be required to ascertain the
same. The readers are also advised that the two sets of experimental data (Fessler & Eaton,
1997; Founti & Klipfel, 1998) used in his study were not alike in all respects within the flow
field with parity only in carrier phases. However, for the current set of results obtained the
particle response to the turbulence for the LP flow in comparison to the GP flow may be
explained in terms of the carrier phase employed to study the particle response. Firstly, the
density and the viscosity used to study LP flow are approximately 709 and 257 times higher
than that of the GP flow, which basically prohibits the fluctuating motion of the particle.
Another rationale being, in regions of strong mean velocity gradient, the streamwise particle
fluctuating velocities are determined more by the mean gradient than by the actual response
of the particles to turbulent fluctuations. In the absence of the same the particle velocity
fluctuations tend to be lower than the fluid velocity fluctuations
as noted from the
experiments of Fessler and Eaton (1995), it was also stated by the same authors that in their
experiments wallnormal fluctuating particle velocities were lower that fluid, which was
attributed to the large inertia of the particles making them unresponsive to many of the fluid
motions. From these conclusions the eventual decrease in the particle fluctuation is more or
less attributed to the decrease in the velocity gradient with a corresponding increase in Stokes
number. The crossstream mixing, which attributes towards higher particle fluctuating
velocities in GP flow may be prohibitive in LP flow considering the elevated density and
viscosity of LP flows.
AirLiquid Flows
In the modelling of microbubble laden Air_liquid flows, two sets of governing equations
for momentum were solved. The generic CFD code ANSYS CFX 11 (ANSYS, 2006) was
employed as a platform for twofluid flow computation. The builtin Inhomogeneous and
MUSIG models had been adopted for our numerical simulations. Figure 6.1a shows the
schematic diagram of the numerical model used in our computations. Numerical simulations
were performed with a velocity inlet and a pressure outlet, on the left and right side of the 2D
computational domain respectively. The top wall is modelled as a friction free boundary
condition, wherein the height of the computational domain reflects only half the height of the
original test section. The bottom part of the domain has been divided into three distinct
sections, section 1 & 3 were modelled as noslip walls for liquid and freeslip for micro
bubbles, emulating the experimental boundary conditions. The section 2 is specified as the
inlet boundary condition for our gas inlet imitating the experimental conditions of gas
injection thought the porous plate.
K. Mohanarangam and J.Y. Tu 96
Tractionfree opening
Velocity
Inlet
x
y 0.057m
Pressure
outlet
0.280m 0.178m 0.254m
Section 1
Section 2 Section 3
Figure 6.1.a. Schematic diagram of the numerical model.
A uniform liquid velocity was specified at the inlet of the test section, different gas flow
rates were specified along the section 2 of the computational domain, the free stream
velocities and the gas injection rates used in the simulations are summarized in Table 1. An
area permeability of 0.3 which lies in line with the sintered metal used in the experiments and
also employed in the numerical work of Kunz et al. (2003) is used all along section 2 for gas
injection purposes. At the outlet, a relative averaged static pressure of zero was specified. For
all flow conditions, reliable convergence were achieved within 2500 iterations when the RMS
(root mean square) pressure residual dropped below 1.0×10
7
. A fixed physical time scale of
0.002s is adopted for all steady state simulations.
Table 6.1. Input boundary conditions for the computational model
Case Air flow rate Q
a
(m
3
/s)
Water free stream
velocity (m/s)
Re
L
based on the total
plate length
Q0V9.6 (C
fo
) 0 9.6 7.66 x 10
6
Q1V9.6 0.001 9.6 7.66 x 10
6
Q2V9.6 0.0015 9.6 7.66 x 10
6
Q3V9.6 0.002 9.6 7.66 x 10
6
Q4V9.6 0.0025 9.6 7.66 x 10
6
Q5V9.6 0.003 9.6 7.66 x 10
6
Q014.2(C
fo
) 0 14.2 1.13 x 10
7
Q1V14.2 0.001 14.2 1.13 x 10
7
Q2V14.2 0.0015 14.2 1.13 x 10
7
Q3V14.2 0.002 14.2 1.13 x 10
7
Q4V14.2 0.0025 14.2 1.13 x 10
7
Q5V14.2 0.003 14.2 1.13 x 10
7
Section 1
Section 2
Section 3
Figure 6.1.b. Computational grid used for computations.
On the Numerical Simulation of Turbulence Modulation in TwoPhase Flows 97
In handling turbulent microbubble flow, unlike single phase fluid flow problem, no
standard turbulence model has been customized for twophase (liquidair) flow. Nevertheless,
numerical investigation revealed that standard kε model predicted an unrealistically high gas
void fraction peak close to wall (Frank et al., 2004, Cheung et al., 2006). The kω based Shear
Stress Transport (SST) model by Menter (1994) provided more realistic prediction of void
fraction close to wall.
0.00
5.00
10.00
15.00
20.00
25.00
30.00
1 10 100 1000 10000 100000
U
+
=y
+
; U
+
= (1/k)Ln(Ey
+
)
Numerical
y
+
U
+
Figure 6.2. Comparison of simulated boundary layer velocity profile with the standard law of curves for
U
∞
=14.2m/s.
The SST model is a hybrid version of the kε and kω models with a specific blending
function. Instead of using empirical wall function to bridge the wall and the faraway
turbulent flow, the kω model solves the two turbulence scalars right up to the wall boundary.
This approach eliminates errors arising from empirical wall function and thus provides better
prediction at the near wall region. The SST model is thereby employed in the present study.
Moreover, to account for the effect of bubbles on liquid turbulence, the Sato’s bubbleinduced
turbulent viscosity model (Sato et al., 1981) has been adopted as well.
Figure 6.1b shows the mesh distribution within the computational model, wherein a non
uniform orthogonal mesh with 151x101 grid points was spanned over the whole computation
domain. Wallnormal clustering was used in order to resolve the boundary layer and the
height of the two closest cells next to the walls was designed to be y
+
≤1, the height of which
can be approximated from the logarithmic law
K. Mohanarangam and J.Y. Tu 98
κ υ κ
κ υ κ
τ
τ
τ
1
ln
1
1
ln
1
− +
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
=
− + ⎟
⎠
⎞
⎜
⎝
⎛
=
∞
∞
∞
B
U
u L U
B
L u
u
U
Using the relation
2 / 1
2
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
=
∞
f
C u
U
τ
(3.28)
and substituting it to the above equation, we get
κ κ
1
2
Re ln
1 2
2 / 1
2 / 1
− +
⎥
⎥
⎦
⎤
⎢
⎢
⎣
⎡
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
=
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
∞
B
C
C
f
f
(3.29)
The skin friction coefficient (C
f
) can be solved from the above equation (3.29) by
substituting values of B=5.0 and κ=0.41. u
τ
can be solved from the equation (3.28) and the
distance of the first grid point from the wall can be solved from the equation
τ
υ
u
y
y
first
+
=
6.1. Results and Discussion
In order to better understand the drag reduction phenomenon and to better comprehend
the various mechanisms behind it, a prognostic approach has been carried out in throughout
this section starting from the single phase and then to the airliquid microbubble flows,
wherein at each juncture our numerical outcomes have been verified and validated against
well established experimental and numerical findings.
Before investigating the physical phenomenon of the drag reduction, it is very crucial to
ascertain adequate grid spacing has been generated to resolve the inner, buffer and the outer
boundary layer. Figure 6.2 shows the single phase boundary layer velocity profile of the
outlet compared against the standard law of the wall curves for the maximum free stream
velocity of 14.2 m/s. The excellent agreement with the standard curve clearly confirmed that
sufficient mesh resolution has been spanned throughout the boundary layer to capture the
associated velocity gradient. The 9.6 m/s case had the similar y
+
values and hence been not
shown here for brevity.
On the Numerical Simulation of Turbulence Modulation in TwoPhase Flows 99
6.1.1. Experimental Validation (Inhomogeneous Model)
With the mesh spacing sufficiently fine enough to resolve the turbulent boundary layer
attention is then focussed on the microbubbles drag reduction mechanism. Figure 6.3 shows
the comparison of the predicted skin friction ratios of the twofluid inhomogeneous model
against the experimental data of Madavan et al. (1984) along various gas injection rates (Q1
Q5) for both the freestream velocities of 9.6 and 14.2m/. Herein, C
f
& C
fo
are the skinfriction
coefficients with and without the gas injection respectively. The skinfriction coefficient
throughout our numerical study have been obtained by averaging out the entire flat plate of
‘section 3’. It can be observed from the figure that satisfactory agreement was obtained using
the inhomogeneous twofluid model for both the free stream velocities considered in our
study.
0.50
0.60
0.70
0.80
0.90
1.00
0.0000 0.0005 0.0010 0.0015 0.0020 0.0025 0.0030 0.0035
14.2m/s (Expt)
14.2m/s (Num)
9.6m/s (Expt)
9.6m/s (Num)
C
f
/
C
f
o
Q
a
Figure 6.3. Comparison of skin friction coefficient with the experimental findings for U
∞
=14.2m/s &
9.6m/s.
Readers are advised that these results have been compared treating that the experimental
errors encountered during various process of measurements are nil (error percentages of the
experimental data). However, hypothetical they may seem, the numerical predictions were in
remarkable agreement with the measurements. It can be seen that there is always an increase
in the drag reduction (DR) or a decrease in the C
f
/C
fo
ratio with a corresponding increase in
the gas flow rates for both the free stream velocities considered in our simulation, which
again is congruent to the previous microbubble studies. In all the simulations presented,
buoyancy was included as a necessary force, as one could find a marked change from the
experimental results with the plate on ‘Top’ and ‘Bottom’ configuration.
K. Mohanarangam and J.Y. Tu 100
0.000
0.005
0.010
0.015
0.020
0.025
6.00 8.00 10.00 12.00 14.00 16.00
Streamwise liquid velocity (m/s)
Y
(
m
)
Q4V14.2
Q1V14.2
Q2V14.2
Q3V14.2
Q5V14.2
Figure 6.4.a. Velocity profiles for varying gas injection rates for free stream velocity U
∞
= 14.2m/s.
0.50
0.55
0.60
0.65
0.70
0.75
0.80
0.85
0.90
0.95
1.00
1 10 100 1000
Q4V14.2
Q1V14.2
Q2V14.2
Q3V14.2
Q5V14.2
u
l
/
u
u
l
y
+
Figure 6.4.b. Change in the mean flow velocity for the carrier phase along the boundary layer for U
∞
=
14.2m/s.
On the Numerical Simulation of Turbulence Modulation in TwoPhase Flows 101
6.1.2. Investigation of Mechanisms of Drag Reduction
With the skin friction coefficients showing fairly good comparison for the twofluid
inhomogeneous model, it can be further investigated to study the various mechanisms of drag
reduction. To begin with the mean streamwise velocities of the carrier phase are scrutinised.
Figure 6.4a shows the mean streamwise liquid velocity profiles along varying gas injection
rates for the high Reynolds number case of 14.2m/s. As depicted, a clearly marked change in
the velocity profile can be seen with a subsequent increase in the gas flow injection rates,
which is certainly in relation to the large amount of microbubbles present along the boundary
layer as shown from the void fraction profiles in 6.7b. The streamwise velocities reveal that
with the increase in the gas flow rates, there is a subsequent and a gradual increase in the
streamwise velocities in the outer layer. In contrast, a close investigation of the velocities
within the buffer and inner layer shows an opposite phenomenon. In figure 6.4b, the velocity
profiles of the liquid phase for the microbubbles laden flows have been normalized with the
corresponding single phase liquid velocities along the length of the boundary layer, where by
any velocity change felt in the carrier liquid phase is reflected as an exit of the ratio from
unity, from the figure it can be revealed, that there is a marked decrease in the mean
streamwise velocities with a subsequent increase of the gas injection rates. It can also be seen,
that the flow undergoes a maximum reduction in the mean velocity of about 40% for the
highest gas flow rate, while it is quite nominal and about 13% for the lowest gas injection
rate, this trend keeps increasing until a y
+
value of 100, aftermath of which there is a spike for
the largest of the three gas flow rates and then a downward trend follows.
0.00
0.01
0.02
0.03
0.04
0.05
0.06
1 10 100 1000
N
o
r
m
a
l
l
i
q
u
i
d
v
e
l
o
c
i
t
y
(
m
/
s
)
y+
Q4V14.2
Q1V14.2
Q2V14.2
Q3V14.2
Q5V14.2
Q0V14.2
Figure 6.4.c. Liquid normal velocity for the carrier phase along the boundary layer for U
∞
= 14.2m/s.
K. Mohanarangam and J.Y. Tu 102
These findings reported above are in lines with the DNS findings of Ferrante and
Elghobashi (2004), wherein the presence of microbubbles in the turbulent boundary layer
results in a local positive divergence of the fluid velocity, 0 U > • ∇ , creating a positive
mean velocity normal to (and away from) the wall which in turn, reduces the mean
streamwise velocity and displaces the quasistreamwise longitudinal vortical structures away
from the wall. The shifting of the vortical structures away from the wall indicates that the
‘sweep’ and ‘ejection’ events (Robinson, 1991), which are located respectively at the
downward and upward sides of these longitudinal vortical structures, are moved farther away
from the wall, thereby reducing the intensity of wall streaks along the wall and consequently
decreasing the skinfriction. It was also reported that there is shift with respect to the location
of peak Reynolds stress production away from the wall, thus reducing the production rate of
turbulence kinetic energy and enstrophy.
Figure 6.4.d. Air void fraction contour plot for Q5V14.2.
Figure 6.4c shows the plot of water normal velocity through varying gas injection rates, it
can be seen that there is generally an increasing trend in the velocities and then a decrease
which is followed by a maximum peak, and this is in direct relation to the loss incurred by the
flow along the streamwise direction, across varying gas injection rates. There is also a sudden
spike in the normal velocities within a y
+
range of 60120. However, the onset of the increase
and the occurrence of the maximum differ in accordance to the gas injection rates. For the
Outlet
Water flow U
∞
x
y
On the Numerical Simulation of Turbulence Modulation in TwoPhase Flows 103
three higher injection rates (Q3Q5) the location of the start of sudden increase and the
occurrence of maximum seems to occur more or less in unison, but their magnitude of
maximum normal velocities differ wide apart. While for the lower gas injection rates (Q1 &
Q2) the location and the magnitude are more distinct and separated wide apart. It can also be
seen that the unladen wall normal velocity is quite smaller in lieu with the laden normal
velocities. This can be further confirmed from the contour plot of the air void fraction along
the boundary layer as shown in figure 6.4d, for the highest air flow rate considered in our
study, where there is a small layer of water, which is immediately followed on the top by a
thick layer of air and then followed again by water, herein due to the inherent presence of the
microbubbles in the middle section, caused an upward shift in the water normal velocities.
0.00
5.00
10.00
15.00
20.00
25.00
30.00
35.00
40.00
1 10 100 1000 10000 100000
y
+
U
+
U
+
=y
+
; U
+
= (1/k)ln(Ey
+
)
Q4V14.2
Q1V14.2
Q2V14.2
Q3V14.2
Q5V14.2
Figure 6.5. Change in the boundary layer for varying gas flow rates for U
∞
= 14.2m/s.
Figure 6.5 shows the plot of nondimensional streamwise velocity profiles along the
boundary layer for varying gas flow rates at the middle of ‘section 3’. The presence of the
micro bubbles can be felt for a y
+
≥10, where in there is a gradual thickening of the viscous
zone with an upward shift of the logarithmic region, while the inner layer seems more or less
unaltered. With these findings, it can be ascertained that the important aspect in achieving
drag reductions is the accumulation of the micro bubbles within a critical zone in the buffer
layer. This is in lieu with the experimental findings of Villafuerte & Hassan (2006), whereby
high drag reductions were reported due the accumulation of the micro bubbles within a range
of 15 ≥ y
+
≤ 30.
6.1.3. Turbulence Modulation (TM)
The plot depicted in the figure 6.6 demonstrates the turbulent modulation (TM) of the
liquid phase in the presence of the microbubbles and is given by the ratio of the micro
K. Mohanarangam and J.Y. Tu 104
bubble laden flow r.m.s streamwise velocity to the unladen r.m.s streamwise velocity. These
plots signify that any TM felt in the carrier liquid phase is reflected as an exit of the ratio
from unity. It can be seen from the plot, across various gas injection rates a marked
attenuation is felt up to a distance along the boundary layer and then a subsequent increase,
which is attributed towards the turbulence enhancement of the liquid phase. It is also
worthwhile to note that the flow has a tendency to attenuate more for higher gas flow rates.
On the other hand there is a turbulence augmentation effect pronounced more in the outer
layer of the boundary. The marked attenuation felt for a small distance from the wall is
attributed to the presence of a thin lining of liquid all along the wall (as explained above).
However, in order to explain the augmentation of the turbulence felt within the boundary, the
“bubblerepelling” and the “bubblerising” events observed from the experiments of Y.Murai
et al. (2006) is used. From their findings, it is outlined that the vertical rise velocity of the
bubble or the “bubblerising” event towards the wall is only about 5% of the streamwise
velocity, after which the bubble reaches equilibrium with its surroundings and starts its
journey back through the “bubblerepelling” event away from the wall, but however this
downward journey accounts for 25% of the streamwise velocity.
0.00
0.20
0.40
0.60
0.80
1.00
1.20
1.40
1.60
1.80
2.00
1.00 10.00 100.00 1000.00 y
+
u
'
/
u
'
l
Q4V14.2
Q1V14.2
Q2V14.2
Q3V14.2
Q5V14.2
Figure 6.6. Turbulence Modulation (TM) along the boundary layer for U
∞
= 14.2m/s.
Although the aforementioned experimental observations refer to individual bubble
motion which can only be tracked numerically using Lagrangian approach, the phenomenon
of turbulence augmentation in the carrier phase is taken care in our simulation through the
SATO (Sato et al; 1981) model, which accounts for the additional viscosity generated through
the bubble slip velocity, wherein the vortices are formed behind the bubbles by their motion,
thereby causing an increase in the turbulence levels in the outer layer of the boundary. It can
On the Numerical Simulation of Turbulence Modulation in TwoPhase Flows 105
also be seen that this turbulence enhancement is more pronounced in the outer layer of the
boundary, while most part of the inner and the buffer layer experiences attenuation.
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.00 500.00 1000.00 1500.00 2000.00
V
o
l
u
m
e
f
r
a
c
t
i
o
n
o
f
a
i
r
Q4V9.6
Q1V9.6
Q2V9.6
Q3V9.6
Q5V9.6
y
+
Figure 6.7.a. Volume fraction of air along the outlet for U
∞
= 9.6m/s.
Figure 6.7a shows the void fraction profiles for the dispersed phase for the freestream
velocity of 9.6m/s, along the outlet plane of the geometry, it can be seen that there is a sharp
increase in the void fraction for a y
+
value of about 200 for the maximum flow rate, with the
maximum occurring there, where as from figure 6.7b, it can be seen that the void fraction
profiles for the dispersed phase occurring in a slightly different pattern with the maximum
occurring at a distance around y
+
=150. This occurrence of maximum void fraction can be
best used to explain the degree of drag reduction between the two Reynolds numbers
considered in our study. For the 9.6m/s case where the maximum void fraction occurs at a
distance higher than that of the 14.2m/s, a higher degree of drag reduction is seen and this is
attributed to the fact that the bubbles can rise and distribute themselves within the boundary
layer there by expanding the boundary layer (buffer and outer layer) and thus causing a
greater drag reduction, but however for the high Reynolds number 14.2m/s case the bubbles
have lower residence time to redistribute themselves within the boundary layer there by
causing a smaller drag reduction and consequently a higher C
f
/C
fo
value (Kodama et al. 2000).
From figures 6.7a and 6.7b, it can be seen that with the increase in the gas flow rates the void
fraction of the injected air increases for all the cases and later flatten out due to the dispersion
and also the washing down effect of the continuous phase. From the close examination of the
graphs it can be seen that there exists a thin film of liquid covering all over the wall
irrespective of the injected gas flow rates, these finding are in accordance to the experimental
findings of Murai et al. (2007), who later concluded that this liquid film thickness gradually
becomes thinner from the front to the rear part of the bubble such that the skin friction varies
along the coordinate. From the work of Tinse et al. (2003) it can be deduced that thinner this
K. Mohanarangam and J.Y. Tu 106
liquid film becomes the lower the shear stress the film has. This has also been verified from
our numerical findings that the skin friction ratio decreases with the increase in the gas flow
due to the inherent thinning of the liquid film adjacent to the wall.
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.00 500.00 1000.00 1500.00 2000.00 2500.00
Q4V14.2
Q1V14.2
Q2V14.2
Q3V14.2
Q5V14.2
y
+
V
o
l
u
m
e
f
r
a
c
t
i
o
n
Figure 6.7.b. Volume fraction of air along the outlet for U
∞
= 14.2m/s.
6.1.3. Effect of Bubble Coalescence and Breakup in Drag Reduction
In the light of the aforementioned results, one should notice that a prescribed bubble
diameter has been specified throughout all numerical simulations for Inhomogeneous two
fluid model. Although encouraging results have been presented in previous sections, the value
of bubble diameter (i.e. 500 µm) unfortunately at best served as a fair engineering estimation
which is calibrated against experimental data based on trailanderror without solid physical
interpretations. As discussed before, the constant diameter assumption may introduce
numerical errors if the bubble coalescence and breakup become dominant in the problem,
especially when the air injection rate is considerably high. In attempting to overcome this
problem, the MUSIG model is introduced into the simulation allowing bubble diameter to be
evaluated mechanistically using the coalescence and breakage kernels.
Figure 6.8a shows the numerical comparison against its experimental counterpart for a
free stream velocity of 9.6m/s, the MUSIG model in addition to Inhomogeneous model is
used for comparison. Herein the MUSIG model had been specified 10 groups of bubbles,
diameters ranging from 100µm1000µm. As depicted in the figure, at high flow rates (i.e. Q4
and Q5), MUSIG model gives the best agreement while the inhomogeneous model tends to
slightly overpredict the skin friction coefficients. However, using the default coefficients for
On the Numerical Simulation of Turbulence Modulation in TwoPhase Flows 107
breakup and coalescence models serious underprediction has been observed for the MUSIG
model for low gas flow rates (i.e. Q1Q3). Figure 6.8b shows the similar comparison of
experimental data with higher freestream velocity of 14.2 m/s. Analogy to the previous
result, predictions of the MUSIG model appear marginally superior to the inhomogeneous at
high gas injection rates (i.e. Q4 & Q5), while considerably under predictions have shown for
lower gas injection rates.
0.50
0.60
0.70
0.80
0.90
1.00
0.0000 0.0005 0.0010 0.0015 0.0020 0.0025 0.0030 0.0035 0.0040
Expt
TwoFluidInhomogenous
MUSIG
C
f
/
C
f
o
Flow rate of Air (Q
a
)
9.6m/s
Figure 6.8.a. Comparison of computed skinfriction coefficient Inhomogeneous & MUSIG models
U
∞
= 9.6m/s.
U=14.2m/s
0.5
0.6
0.7
0.8
0.9
1.0
0.0000 0.0005 0.0010 0.0015 0.0020 0.0025 0.0030 0.0035
Expt
TwoFluidInhomogenous
MUSIG
Flow rate of Air (Q
a
)
C
f
/
C
f
o
Figure 6.8.b. Comparison of computed plate drag coefficient Inhomogeneous & MUSIG models U
∞
=
14.2m/s.
K. Mohanarangam and J.Y. Tu 108
One possible reason attributed to the underprediction of the MUSIG model for low gas
injection rates could be the overestimation of bubble breakup rate which sequentially
introduced more small bubbles into calculations. These additional small bubbles were thereby
dispersed with the boundary layer caused greater drag reduction on the surface. It should be
emphasized that default model parameters of the MUSIG have been adopted directly in the
above numerical investigation. These parameters were calibrated with bubbly flow condition
where isotopic turbulence was assumed. At high air injection rate, such assumption may be
quite close the physical behaviour as higher turbulence modulation has been introduced by
the presence of bubbles. However, it may become invalid for low flow rates. In essence, it is
well known that these model parameters may vary from cases to cases which should be re
calibrated for particular flow condition.
0.50
0.60
0.70
0.80
0.90
1.00
0.0000 0.0005 0.0010 0.0015 0.0020 0.0025 0.0030 0.0035 0.0040
Expt
MUSIG Breakup coefficient=1.0
MUSIG Breakup coefficient=0.05
C
f
/
C
f
o
Flow rate of Air (Q
a
)
9.6m/s
Figure 6.9.a. Comparison of skin friction coefficients with different breakup coefficients for U
∞
=
9.6m/s.
Based on the above argument, also serves as a confirmation of the above observations,
another set of simulations have been carried out with the break up coefficient deliberately
decreased 20 times to minimize the resultant bubble breakup rate. Figure 6.9a shows the
corresponding predictions of the skinfriction coefficient plots for the free stream velocity of
9.6m/s. Compared with the default MUSIG model, the predicted results were generally in
satisfactory agreement with measurements across varying flow rates, while a considerably
improvements have been obtained for the lower air injection rates. Similar observations can
also be found for the freestream velocity of 14.2m/s showing in Figure 6.9b.
The effect of the reduced breakup rate can be exemplified by a closer visualization of the
predicted bubble size distributions of the two numerical results. Figure 6.10a shows the
predicted bubble size distribution at the outlet obtained from the default and the recalibrated
MUSIG models. With the default breakup coefficient, in both freestream velocities, the
relatively high volume fraction of small bubble clearly demonstrated that the default model
On the Numerical Simulation of Turbulence Modulation in TwoPhase Flows 109
tends to create additional small bubble via the dominating breakup mechanism. In contrast,
by limiting the breakup rate, bubble coalescence overcomes the breakup mode forming
relatively higher volume fraction for the larger bubbles.
0.5
0.6
0.7
0.8
0.9
1.0
0.0000 0.0005 0.0010 0.0015 0.0020 0.0025 0.0030 0.0035 0.0040
Expt
MUSIG breakup coefficient=1.0
MUSIG breakup coefficient=0.05
Flow rate of Air (Q
a
)
C
f
/
C
f
o
14.2 m/s
Figure 6.9.b. Comparison of skin friction coefficients with different breakup coefficients for U
∞
=
14.2m/s.
0.0%
1.0%
2.0%
3.0%
4.0%
5.0%
6.0%
7.0%
8.0%
100 200 300 400 500 600 700 800 900 1000
MUSIG breakup coefficient=0.05
MUSIG breakup coefficient=1.0
d (µm)
V
o
l
u
m
e
f
r
a
c
t
i
o
n
Figure 6.10.a. Bubble diameter distribution function for Q1V9.6.
K. Mohanarangam and J.Y. Tu 110
In the present study, the deduction of breakup coefficient arose only as an engineering
estimation. In fact, as the air injection rates are considerably low, interactions between
bubbles are relatively insignificant compared with that for the high injection rates. It is
thereby unsurprising to recalibrate the model constants for obtaining a “better” comparison.
Although revealing the shortcomings of the current kernels is certainly one of the findings,
this drawback of the model should be circumvented by refining the model assumption and the
mechanism which unfortunately is left far beyond the focus of this paper.
Directing back to the theme of current work, one could easily state that good predictions
of the skinfriction coefficients can be obtained by specifying a proper bubble size for each
simulation. Nevertheless, one should also be reminded that bubble sizes may change
significantly which is impossible to be represented by a fixed average value. This problem is
further exacerbated; if rigorous bubble interactions are involved at high air injection rate. In
practical microbubble problems, it is more easily to acquire the range of bubble size rather
than exact bubble diameter. The MUSIG model which tailored to resolve the bubble size
distribution mechanistically within a given range of bubble size appeared as the best
candidate to resolve the physics embedded in microbubble drag reduction problems.
0.0%
1.0%
2.0%
3.0%
4.0%
5.0%
6.0%
7.0%
8.0%
9.0%
10.0%
100 200 300 400 500 600 700 800 900 1000
MUSIG breakup coefficient=0.05
MUSIG breakup coefficient=1.0
d (µm)
V
o
l
u
m
e
f
r
a
c
t
i
o
n
Figure 6.10.b. Bubble diameter distribution function for Q2V14.2.
Conclusion
In this article outlined above, a lot of work was undertaken numerically to study the
behavior of twophase turbulent flows of varying density regimes viz., GasParticle, Liquid
Particle and LiquidAir flows. In addition to the carrier and the dispersed phases mean and
turbulent behavior, Turbulence Modulation (TM) has also been investigated. It is given by the
On the Numerical Simulation of Turbulence Modulation in TwoPhase Flows 111
change in the carrier phase amidst the dispersed phase or the effect of the dispersed phase on
to the carrier phase at the turbulence level.
For the GasParticle flow, the particleturbulence twophase flow interaction has been
successfully investigated with the Eulerian twofluid model. The numerical code has been
validated against the experimental results of Fessler and Eaton (1999) for mean velocity and
the fluctuating velocities for both carrier and the dispersed phases. Two classes of particles
sharing the same Stokes number, but different particle Reynolds number has also been
investigated in this study. The majority of the results agree well with the experimental data;
however there have been some minor discrepancies felt at the proximity of the experimental
results.
The Turbulence Modulation (TM) of the carrier phase for these two classes of particles
have been studied along the three sections that is near the inlet (x/h=2), in the midsection
(x/h=7) and just aft of the exit (x/h=14). It can be concluded that even though the 70μm
copper and 150μm glass particles share the same Stokes number, their behavior seems to be
quite different, which suggests that Stokes number alone does not characterize the particle
behavior, thereby making particle Reynolds number an important parameter in classifying the
way the particles behave.
Particles response to turbulent GP (GasParticle) and LP (LiquidParticle) flow, behind a
turbulent backwardfacing step geometry have also been successfully analysed and simulated
numerically using an Eulerian twofluid model. A significant amount of work was undertaken
to provide an indepth understanding of the particle response, amidst turbulent flow
conditions for two different carrier phases namely the gas and the liquid (diesel oil). From
the two sets of experimental data, at the mean velocity level, the particles seem to ‘lead’ and
later ‘catch up’ with the carrier phase for the LP flow, whereas they ‘lag’ behind and later
‘lead’ for the GP flow. While at the turbulence level, the particles seem to ‘lag’ and then
‘catch up’ for the LP flow while they ‘lag’ and phenomenally ‘lead’ for the GP flow. The
detailed study and also the numerical diagnosis undertaken for turbulent particulate flows
with two different carrier phases, to study the particle response both at the mean velocity and
at the turbulence level, behind a shear flow sudden expansion geometry is quite unique and
one of its kind, as there is no current published work dealing with the analysis and numerical
validation of the same.
Numerically the code was validated against the benchmark experimental data of Fessler
and Eaton (1997) for GP and the experimental data of Founti & Klipfel (1998) for the LP
flows. The numerical results revealed good agreement with the experimental data. From there
the code was further used to investigate Stokes number effect on the two different carrier
phases both at the mean velocity and at the turbulence level, for this exercise the experimental
geometry of the GP flow and the inlet conditions of the LP flow were used. In order to
present the results in a more methodical manner, 12 points consisting of a matrix of three
sections along the length of the step and four along the height of the step were used to study
the Stokes number effect on the two types of flows.
At the mean velocity level the particles seem to move faster that the carrier phases for
both the GP and the LP flow. However, at the particle fluctuation level, although the GP flow
show an escalation with the increase in the Stokes number, the same feature seem to be absent
in the LP flow, wherein the particle fluctuation seem to decrease and almost flatten out with
the increase in its Stokes number. The main reason for this behaviour is the difference in the
physical characteristics of the carrier phase namely the liquid, which is far denser than the
K. Mohanarangam and J.Y. Tu 112
gas, this eventually changes the crossstream and the mean gradient behaviour, which is
shown to cause elevated particle fluctuations in the GP flow.
For the LiquidAir flows, the turbulent microbubble laden flow has been investigated
with the help of two numerical models namely the twofluid Inhomogeneous and MUSIG
models, for two different freestream velocities. Inhomogeneous model, which uses a fixed
bubble diameter, shows a very good comparison of the skinfriction coefficients with the
experiment. This model is further probed to study the various physical phenomenon’s causing
the drag reduction along the boundary layer, firstly it was observed that there is drop in the
mean streamwise water velocities with a subsequent increase in the normal along varying gas
injection rates. Secondly, the presence of the microbubbles caused turbulence attenuation for
some distance along the boundary layer and later an augmentation was felt due to the
shedding of the vortices behind the bubbles. Thirdly, the peak of the void fractions seem to
differ in relation to the degree of drag reduction along the two freestream velocities
considered in our study.
However, with respect to the drag reduction caused due to the presence of microbubbles
in the turbulent boundary layer MUSIG model seem to show good predictions for higher gas
flow rates while under predicting for lower gas flow rates. This poor prediction of the model
at low flow rates was investigated to be the dominating breakup phenomenon that was taking
place within the flow. This was done in order to represent the actual flow condition, where by
groups of bubbles of varying bubble sizes are found within the boundary layer. Thereby
allowing the MUSIG model to resolve the bubble size distribution mechanistically within a
given range of bubble size and feeding it back to the Inhomogeneous model appeared as the
best candidate to resolve the hidden physics in microbubble induced drag reduction
problems.
References
Alajbegovic, A, Assad, A, Bonetto, F, & Lahey, Jr RT 1994, ‘Phase distribution and
turbulence structure for solid/fluid upflow in a pipe’, International Journal of Multiphase
flow, Vol. 20, pp. 453479.
Anderson, TB, & Jackson, R 1967, ‘A fluid mechanical description of fluidized beds:
equations of motion’, Industrial and Engineering Chemistry Fundamentals, Vol. 6, pp.
527–539.
ANSYS 2006, CFX11 User Manual, ANSYS CFX.
Awatef, AH, Widen, T, Richard, BR, Kaushik, D & Puneet, A 2005, 'Turbine Blade Surface
Deterioration by Erosion', Journal of Turbomachinery, Vol. 127, pp. 445452.
Borowsky, J, & Wei, T 2007, ‘Kinematic and Dynamic Parameters of a LiquidSolid Pipe
Flow using DPIV/Accelerometry’, Journal of Fluids Engineering. Vol. 129, pp. 1415
1421.
Chan, CK, Zhang, HQ, & Lau, KS 2001, ‘Numerical simulation of gasparticle flows behind
a backwardfacing step using an improved stochastic separated flow model’,
Computational Mechanics, Vol. 27, pp. 412–417.
Chein R, & Chung, JN 1987, ‘Effects of Vortex pairing on particle dispersion in turbulent
shear flows’, International Journal of Multiphase Flow, Vol. 13, pp. 785802.
On the Numerical Simulation of Turbulence Modulation in TwoPhase Flows 113
Chen, CP, & Wood, PE 1985, ‘A Turbulence closure Model for Dilute GasParticle Flows’,
Canadian Journal of Chemical Engineering, Vol. 63, pp. 349360.
Chesters, AK, & Hoffman, G 1982, ‘Bubble coalescence in pure liquids’, Applied Scientific
Research, Vol. 38, pp. 353361.
Cheung, SCP, Yeoh, GH, & Tu, JY 2006. ‘On the modelling of population balance in
isothermal vertical bubbly flows – average bubble number density approach’, Chemical
Engineering Process, in Press
Clift, R, Grace, JR, & Weber ME, ‘Bubbles, Drops and Particles’, Academic Press, New
York, 1978.
Drew, DA, & Lahey Jr, RT 1979. ‘Application of general constitutive principles to the
derivation of multidimensional twophase flow equation’, International Journal
Multiphase Flow, Vol. 5, pp. 243264.
Eaton, JK, Paris AD, & Burton TM 1999, ‘Local distortion of turbulence by dispersed
particles’, AIAA Paper, 1999; 993643.
Ferrante, A, & Elghobashi, S 2004. ‘On the Physical Mechanisms of Drag Reduction in a
Spatially Developing Turbulent Boundary Layer Laden with Microbubbles’. Journal of
Fluid Mechanics, Vol. 503, pp. 345–355.
Fessler JR, Eaton JK 1995, ‘Particleturbulence interaction in a backwardfacing step flow’,
Mechanical Engineering Department Report MD70, Stanford University, Stanford,
California.
Fessler JR, & Eaton JK 1997, ‘ParticleResponse in a Planar Sudden Expansion Flow’,
Experimental Thermal and Fluid Science, Vol. 15, pp. 413423.
Fessler JR, & Eaton JK 1999. Turbulence modification by particles in a backward facing step
flow. Journal of Fluid Mechanics, Vol. 394, pp. 97117.
Founti, M, & Klipfel, A 1998. Experimental and computational investigations of nearly dense
twophase sudden expansion flows. Experimental Thermal and Fluid science, Vol. 17,
pp. 2736.
Frank, T, Shi, J, & Burns, AD 2004, ‘Validation of Eulerian multiphase flow models for
nuclear safety application’, Proceeding of the Third International Symposium on Two
Phase Modelling and Experimentation, Pisa, Italy.
Gil, A, Cortes, C, Romeo, LM, & Velilla, J 2002, ‘Gasparticle flow inside cyclone diplegs
with pneumatic extraction’, Powder Technology, Vol. 128, pp. 7891.
Hetsroni G 1989, ‘Particlesturbulence interaction’, International Journal of Multiphase
Flow, Vol. 15, pp. 735746.
Hishida K, Ando A, & Maeda, M 1992. ‘Experiments on particle dispersion in a turbulent
mixing layer’, International Journal of Multiphase Flow, Vol. 18, pp. 181194.
Hishida K, & Maeda, M 1999. ‘Turbulence Characteristics of ParticleLaden Flow Behind a
Reward Facing Step’. ASME FED, Vol. 121, pp. 207212.
Huang, XY, Stock, DE, & Wang, LP 1993, ‘Using the MonteCarlo process to simulate two
dimensional heavy particle dispersion’, ASMEFED, GasSolid flows, Vol. 166, pp. 153
160.
Inthavong, K, Tian, ZF, Li, HF, Tu, JY, Yang, W, Xue, CL & Li, CG 2006, ‘A Numerical
Study of Spray Particle Deposition in a Human Nasal Cavity’, Aerosol Science and
Technology, Vol. 40, pp. 10341045.
Ishii, M 1975, ThermalFluid Dynamic Theory of Twophase Flow, Eyrolles, Paris.
K. Mohanarangam and J.Y. Tu 114
Ishii, M, & Zuber, N 1979. ‘Drag coefficient and relative velocity in bubbly, droplet or
particulate flows’. AIChE Journal, Vol. 5, pp. 843855.
Ishima, T, Hisanobu, K, Naoki, S, & Tomio, O 2007, ‘Turbulence characteristics in two
phase pipe flow lading with various particles’, S3_Thu_B_50, ICMF 2007, Leipzig,
Germany, July 913.
Ishima, T, Hishida K, & Maeda, M 1993, ‘Effect of particle residence time on particle
dispersion in a plane mixing layer’, Journal of Fluids Engineering, Vol. 115, pp. 751
759.
Kodama, Y., Kakugawa, A., Takahashi, T., Kawashima, H., 2000. Experimental study on
microbubbles and their applicability to ships for skin friction reduction. International
Journal of Heat and Fluid Flow 21, 582588.
Kolaitis, DI, & Founti, MA 2002, ‘Modeling of the gasparticle flow in industrial
classification chambers for design optimization’, Powder Technology, Vol. 125, pp. 298
305.
Kumar, S, & Ramkrishna, D 1996. ‘On the solution of population balance equations by
discretisation – I. A fixed pivot technique’, Chemical Engineering Science, Vol. 51, pp.
13111332.
Kulick, JD, Fessler, JR, & Eaton, JK 1994‚ ‘Particle response and turbulence modification in
fully developed channel flow’, Journal of Fluid Mechanics, Vol. 277, pp. 109134.
Le, H, Moin, P, & Kim, J 1997, ‘Direct numerical simulation of turbulent flow over a
backwardfacing step’, Journal of Fluid Mechanics, Vol. 330, pp. 349374.
Lo, S 1996, ‘Application of the MUSIG model to bubbly flows’, AEAT1096, AEA
Technology.
Luo, H, & Svendsen, H 1996, ‘Theoretical model for drop and bubble breakup in turbulent
dispersions’, AIChE Journal, Vol. 42, pp. 12251233.
Madavan, NK, Deutsch, S, & Merkle, CL 1984, ‘Reduction of turbulent skin friction by
microbubbles’, Physics of Fluids, Vol. 27, pp. 356363.
Menter, FR 1994, ‘Twoequation eddy viscosity turbulence models for engineering
applications’, AIAA Journal, Vol. 32, pp. 15981605.
Murai, Y, Fukuda, H, Oishi, Y, Kodama, Y, & Yamamoto, F 2007, ‘Skin Friction Reduction
by Large Air Bubbles in a Horizontal Channel Flow’, International Journal of
Multiphase Flow, Vol. 33, pp. 147163.
Pandya RVR, & Mashayek F 2002, ‘TwoFluid Largeeddy Simulation approach for Particle
laden Turbulent Flows’, International Journal of Heat and Mass Transfer, Vol. 45, pp.
4753–4759.
Parthasarathy RN, & Faeth GM 1987, ‘Structure of ParticleLaden Turbulent Water Jets in
Still Water’, International Journal of Multiphase Flow, Vol. 13, pp. 699716.
Parthasarathy RN, & Faeth GM 1990, ‘Turbulence Modulation in homogenous dilute
particleladen flows’, Journal of Fluid Mechanics, Vol. 220, pp. 485514.
Prince, MJ, & Blanch, HW 1990, ‘Bubble coalescence and breakup in air sparged bubble
columns’, AIChE Journal, Vol. 36, pp. 14851499.
Qianpu, W, Morten, CM, Sunil, RDS & Guifang, T 2005. 'Experimental and computational
studies of gas–particle flow in a ‘sandblast’ type of erosion tester', Powder Technology,
Vol. 160, pp. 93102.
Rashidi, M, Hetsroni, G, & Banerjee, S 1990, ‘ParticleTurbulence Interaction in a Boundary
Layer’, International Journal of Multiphase Flow, Vol. 16, pp. 935949.
On the Numerical Simulation of Turbulence Modulation in TwoPhase Flows 115
Righetti, M, & Romano, GP 2007, ‘Particle fluid interactions in a plane nearwall turbulent
flow’, Journal of Fluid Mechanics, Vol. 505, pp. 93121.
Robinson, S.K., 1991. Coherent motions in the turbulent boundary layer. Annual Review of
Fluid Mechanics 23, 601639.
Ruck, B, & Makiola, B 1988, ‘Particle Dispersion in a SingleSided BackwardFacing Step
Flow’, International Journal of Multiphase Flow, Vol. 14, pp. 787800.
Sato, Y, Sadatomi, M, Sekoguchi, K 1981, ‘Momentum and heat transfer in twophase bubbly
flow – I’, International Journal of Multiphase Flow, Vol. 7, pp. 167178.
Sato, Y, Hishida, K 1996, ‘Transport process of Turbulence energy in particleladen turbulent
flow’, International Journal of Heat & Fluid Flow, Vol. 17, pp. 202210.
Sato, Y, Fukuichi, V, & Hishida, K 2000, ‘Effect of InterParticle Spacing on Turbulence
Modulation by Lagrangian PIV’, International Journal of Heat and Fluid Flow, Vol. 21,
pp. 554561.
Schuh, MJ, Schuler, CA, & Humphrey JAC 1989, ‘Numerical calculation of particleladen
gas flows past tubes’, AIChE Journal, Vol. 35, pp. 466480.
Shirolkar, JS, Coimbra, CFM, McQuay, Q 1996, ‘Fundamental Aspects of modeling turbulent
Particle Dispersion in Dilute Flows’, Progress in Energy Combustion and Science, Vol.
22, pp. 363–399.
Tisne, P, Aloui, F, & Doubliez, L 2003. ‘Analysis of wall shear stress in wet foam flow using
the electrochemical method’, International Journal of Multiphase Flow, Vol. 29, pp.
841–854.
Tu, JY, & Fletcher, CAJ 1995, ‘Numerical computation of turbulent gassolid particle flow in
a 90
o
bend’, AIChE Journal, Vol. 41, pp. 21872197.
Tu, JY 1997. ‘Computational of turbulent twophase flow on overlapped grids’, Numerical
Heat Transfer Part B.; Vol. 32, pp. 175–195.
Villafuerte, JO, Hassan, YA 2006, ‘Investigation of microbubble boundary layer using
particle tracking velocimetry’, Journal of Fluids Engineering, Vol. 129, pp. 6679.
Yakhot, V, & Orszag, SA 1986, ‘Renormalization group analysis of turbulence I. basic
theory’, Journal of Scientific Computing, Vol. 1, pp. 351.
In: Fluid Mechanics and Pipe Flow ISBN: 9781607410379
Editors: D. Matos and C. Valerio, pp. 117169 © 2009 Nova Science Publishers, Inc.
Chapter 4
A REVIEW OF POPULATION BALANCE MODELLING
FOR MULTIPHASE FLOWS: APPROACHES,
APPLICATIONS AND FUTURE ASPECTS
Sherman C.P. Cheung
1
, G.H. Yeoh
2
and J.Y. Tu
1*
1
School of Aerospace, Mechanical and Manufacturing Engineering, RMIT University,
Victoria 3083, Australia
2
Australian Nuclear Science and Technology Organisation (ANSTO), PMB 1, Menai,
NSW 2234, Australia
Abstract
Population balance modelling is of significant importance in many scientific and industrial
instances such as: fluidizations, precipitation, particles formation in aerosols, bubbly and
droplet flows and so on. In population balance modelling, the solution of the population
balance equation (PBE) records the number of entities in dispersed phase that always governs
the overall behaviour of the practical system under consideration. For the majority of cases,
the solution evolves dynamically according to the “birth” and “death” processes of which it is
tightly coupled with the system operation condition. The implementation of PBE in
conjunction with the Computational Fluid Dynamics (CFD) is thereby becoming ever a
crucial consideration in multiphase flow simulations. Nevertheless, the inherent
integrodifferential form of the PBE poses tremendous difficulties on its solution procedures
where analytical solutions are rare and impossible to be achieved. In this article, we present a
review of the stateoftheart population balance modelling techniques that have been adopted
to describe the phenomenological nature of dispersed phase in multiphase problems. The main
focus of the review can be broadly classified into three categories: (i) Numerical approaches
or solution algorithms of the PBE; (ii) Applications of the PBE in practical gasliquid
multiphase problems and (iii) Possible aspects of the future development in population
balance modelling. For the first category, details of solution algorithms based on both method
of moment (MOM) and discrete class method (CM) that have been proposed in the literature
are provided. Advantages and drawbacks of both approaches are also discussed from the
theoretical and practical viewpoints. For the second category, applications of existing
*
Email address: jiyuan.tu@rmit.edu.au. Phone no.:+6139925 6191. Fax no.:+6139925 6108. Corresponding
Author: Prof. Jiyuan Tu, SAMME, RMIT University, Bundoora, Melbourne, Victoria 3083, Australia
Sherman C.P. Cheung, G.H. Yeoh and J.Y. Tu 118
population balance models in practical multiphase problems that have been proposed in the
literature are summarized. Selected existing mathematical closures for modelling the “birth”
and “death” rate of bubbles in gasliquid flows are introduced. Particular attention is devoted
to assess the capability of some selected models in predicting bubbly flow conditions through
detail validation studies against experimental data. These studies demonstrate that good
agreement can be achieved by the present model by comparing the predicted results against
measured data with regards to the radial distribution of void fraction, Sauter mean bubble
diameter, interfacial area concentration and liquid axial velocity. Finally, weaknesses and
limitations of the existing models are revealed are suggestions for further development are
discussed. Emerging topics for future population balance studies are provided as to complete
the aspect of population balance modelling.
Keywords: Population balance; Computational Fluid Dynamics; bubbly flow.
1. Introduction
Particles embedded within flow structures are featured in a wide diversity of industrial
systems, such as gassolid dispersion in combustors, catalytic reactions in fluidized beds,
liquidliquid dispersion in stirring tanks, microbial processes in bioreactors, and gasliquid
heat and mass transfer in bubble column reactors. In most cases, these particles (regardless
whether they are inherently presence within the system or deliberately introduce into the
system) are often the dominant factor affecting the behaviour of the systems. Such mounting
industrial interests have certainly stimulated numerous scientific and engineering studies
attempting to synthesize the behaviour of the population of particles and its dynamical
evolution subject to the system environments in which has resulted in a widely adopted
concept known as Population Balance.
The population balance of any system is a record for the number of particles, which may
be solid particles, liquid nuclei, bubbles or, variables (in mathematical terms) whose presence
or occurrence governs the overall behaviour of the system under consideration. In most of the
systems under concern, the record of these particles is dynamically depended on the “birth”
and “death” processes that terminate existing particles and create new particles within a finite
or defined space. Mathematically, dependent variables of these particles may exist in two
different coordinates: namely “internal” and “external” coordinates (Ramkrishna and
Mahoney, 2002). The external coordinates refer to the spatial location of each particle which
is governed by its motion due to convection and diffusion flow behaviour. On the other hand,
internal coordinates concerns the internal properties of particles such as: size, surface area,
composition and so on. Figure 1 shows an example of the internal and external coordinates
involve in the population balance for gassolid particles flows.
From a modelling perspective such as demonstrated in Figure 1, enormous challenges
remain in fully resolving the associated nucleation, growth and agglomeration processes of
particles within the internal coordinates and flow motions of external coordinates which are
subjected to interfacial momentum transfer and turbulence modulation between gas and solid
phases. Owing to the significant advancement of computer hardware and increasing
computing power over the past decades, Direct Numerical Simulations (DNS), which attempt
to resolve the whole spectrum of possible turbulent length scales in the flow, provide the
propensity of describing the complex flow structures within the external coordinates (Biswas
et al., 2005; Lu et al., 2006). Nevertheless, practical multiphase flows that are encountered in
A Review of Population Balance Modelling for Multiphase Flows 119
natural and technological systems generally contain millions of particles that are
simultaneously varying along the internal coordinates. Hence, the feasibility of DNS in
resolving such flows is still far beyond the capacity of existing computer resources. The
population balance approach, which records the number of particles as an averaged function,
has shown to be a more promising way in handling the flow complexity because of its
comparatively lower computational requirements. It is envisaged that the next stages of
multiphase flow modelling in research and in practice would most probably concentrate on
the development of more definitive efficient algorithms for solving the population balance
equation (PBE).
Figure 1. An exmple of the internal and external coordinates of population balance for gassolid particle
flows.
The development of population balance model has a long standing history. Back to the
end of 18
th
century, the Boltzmann equation, devised by Ludwig Boltzmann, could be
regarded as the first population balance equation which can be expressed in terms of
statistical distribution of molecules or particles in a state space. Nonetheless, the derivation of
a generic population balance concept was actually initiated from the middle of 19
th
century. In
1960s, Hulburt and Katz (1964) and Pandolph and Larson (1964), based on the statistical
mechanics and continuum mechanical framework respectively, presented the population
balance concept to solve particle size variation due to nucleation, growth and agglomeration
processes. A series of research development were thereafter presented by Fredrickson et al.
(1967), Ramkrishna and Borwanker (1973) and Ramkrishna (1979, 1985) where the treatment
of population balance equations were successfully generalized with various internal
coordinates. A number of textbooks mainly concerning the population balance of
Sherman C.P. Cheung, G.H. Yeoh and J.Y. Tu 120
aerocolloidal systems have also been published (Hidy and Brock, 1970; Pandis and Seinfeld,
1998; Friedlander, 2000). The flexibility and capability of population balance in solving
practical engineering problems has not been fully exposed, until recently, where Ramkrishna
(2000) wrote a textbook focusing on the generic issues of population balance for various
applications.
Although the concept of population balance has been formulated over many decades,
implementation of population balance modelling was only realized until very recent times.
Such dramatic breakthrough was made possible by the rapid development of computational
fluid dynamics (CFD) and insitu experimental measuring techniques. The availability of
CFD softwares certainly facilitated a solid foundation in obtaining useful PBE solutions. With
the field information provided by the CFD framework, external variables of the PBE can be
easily acquired by decoupling the equation from external coordinates which can then enabled
solution algorithms to be developed within internal coordinates. The capacity to measure
particle sizes or other population balance variables from experiment is also of significant
importance. These experimental data not only allow the knowledge of particle sizes and their
evolution within systems to be realized but also provide a scientific basis for model
calibrations and validations.
In view of current developments of the stateoftheart, this paper aims to further exploit
the methodology of population balance modelling for multiphase flows from several aspects.
First, it attempts to elucidate the implementation of population balance in conjunction with
CFD techniques in handling practical industrial problems. Second, it seeks to reviews the
solution algorithms of the PBE that have been proposed in literatures and discuss the
advantages and drawbacks of each from the theoretical and practical viewpoints. Third, it
demonstrates the model’s capability in predicting gasliquid bubbly flows with or without
heat and mass transfer through rigorous validation studies against various experimental data.
Finally, it seeks to outline the fundamental weakness and limitations of current model
development. Particular focus will be centred on the possible directions for further
development.
2. Population Balance Approaches
2.1. Population Balance Equation
The foundation development of the PBE stems from the consideration of the Boltzman
equation. Such equation is generally expressed in an integrodifferential form describing the
particle size distribution (PSD) as follow:
( )
( )
∫
′ ′ ′ − ′ ′ − = ⋅ ∇ +
∂
∂
ξ
ξ ξ ξ ξ ξ ξ ξ ξ ξ
ξ
0
) , ( ) , ( ) , (
2
1
) , , ( ) , , (
, ,
d t f t f a t r f t r u
t
t r f
∫
∞
′ ′ ′ ′ − −
0
) , ( ) , ( ) , ( ξ ξ ξ ξ ξ ξ d t f a t f
∫
∞
′ ′ ′ ′ +
ξ
ξ ξ ξ ξ ξ γ ds t f p b ) , ( ) / ( ) ( ) ( ) , ( ) ( t f b ξ ξ − (1)
A Review of Population Balance Modelling for Multiphase Flows 121
where ) , , ( t r f ξ is the particle size distribution dependent on the internal space vectorξ ,
whose components could be characteristics dimensions, surface area, volume and so on. r and
t are the external variables representing the spatial position vector and physical time in
external coordinate respectively. ) , , ( t r u ξ is velocity vector in external space. On the RHS,
the first and second terms denote birth and death rate of particle of space vectorξ due to
merging processes, such as: coalescence or agglomeration processes; the third and fourth
terms account for the birth and death rate caused by the breakage processes respectively.
) , ( ξ ξ ′ a is the coalescence or agglomeration rate between particles of sizeξ andξ′ .
Conversely, ) (ξ b is the breakage rate of particles of sizeξ . ) (ξ γ ′ is the number of
fragments/daughter particles generated from the breakage of a particle of sizeξ′ and
) / ( ξ ξ ′ p represents the probability density function for a particle of sizeξ to be generated by
breakage of a particle of sizeξ′ .
Owing to the complex phenomenological nature of particle dynamics, analytical solutions
only exist in very few cases
of which coalescence and breakage kernels are substantially
simplified (Scott, 1968; McCoy and Madras, 2003). Driven by practical interest, numerical
approaches have been developed to solve the PBEs. The most common methods are Monte
Carlo methods, Method of Moments and Class Methods. Theoretical speaking, Monte Carlo
methods, which solve the PBE based on statistical ensemble approach (Domilovskii et al.,
1979; Liffman, 1992; Debry et al., 2003; Maisels et al., 2004), are attractive in contrast to
other methods. The main advantage of the method is the flexibility and accuracy to track
particle changes in multidimensional systems. Nonetheless, as the accuracy of the Monte
Carlo method is directly proportion to number of simulations particles, extensive
computational time is normally required. Furthermore, incorporating the method into
conventional CFD program is also not straightforward which greatly degraded its
applicability for industrial problems. Because of their relevance in CFD applications,
numerical approaches developed for Method of Moments and Class Methods are then
discussed.
2.2. Method of Moments (MOM) Approach
The method of moments (MOM), first introduced by Hulburt and Katz (1964), has been
considered as one of the many promising approaches in viably attaining practical solutions to
the PBE. The basic idea behind MOM centres in the transformation of the problem into
lowerorder of moments of the size distribution. The moments of the particle size distribution
are defined as:
ξ ξ ξ d t f t m
k k
∫
∞
=
0
) (
) , ( ) ( (2)
From the above equation, the first few moments give important statistical descriptions on
the population which can be related directly to some physical quantities. In the case space
vectorξ represents the volume of particle, the zero order moment (k = 0) represents the total
Sherman C.P. Cheung, G.H. Yeoh and J.Y. Tu 122
number density of population and the fraction moment, k = 1/3 and k = 2/3 gives information
on the mean diameter and mean surface area respectively.
The primary advantage of MOM is its numerical economy which condenses the problem
substantially by tracking the evolution of limited number of moments (Frenklach, 2002). This
becomes a critical value in modeling complex industrial systems when particle dynamics is
coupled with already timeconsuming calculations of turbulence multiphase flows. Another
significance of the MOM is that it does not suffer from truncation errors in the PSD
approximation. Mathematically, the transformation from the PSD space to the space of
moments is rigorous. Unfortunately, throughout the transformation process, fraction
moments, representing mean diameter or surface area, are normally involved posing serious
closure problem (Frenklach and Harris, 1987). In order to overcome the closure problem, in
the early development of MOM, Frencklach and his coworkers (Frenklach and Wang, 1991;
Markatou et al., 1993; Frenklach and Wang, 1994) proposed an interpolative scheme to
determine the fraction moment from integer moments – namely Method of moments with
interpolative closure (MOMIC).
2.2.1. Quadrature Method of Moments
Another different approach for computing the moment is to approximate the integrals in
Eq. (1) using numerical quadrature scheme – the quadrature method of moment (QMOM) as
suggested by McGraw (1997). In the QMOM, instead of space transformation, Gaussian
quadrature closure is adopted to approximate the PSD by a finite set of Dirac’s delta functions
as follow:
) ( ) , (
1
i
M
i
i
x N t f − ≈
∑
=
ξ δ ξ (3)
where N
i
represents the number density or weight of the i
th
class consists of all particles per
unit volume with a pivot size or abscissa,
i
x . A graphical representation of the QMOM in
approximating the PSD is depicted in Figure 2.
Although the numerical quadrature approach suffers from truncation errors, it
successfully eliminates the problem of fraction moment which special closure is usually
required. The closure of the method is then brought down to solving 2M unknowns,
i
x and N
i
.
A number of approaches in the specific evaluation of the quadrature abscissas and weights
have been proposed. McGraw (1997) first introduced the productdifference (PD) algorithm
formulated by Gordon (1968) for solving monovariate problem. Nonetheless, as point out by
Dorao et al. (2006, 2008), the PD algorithm is a numerical illconditioned method for
computing the Gauss quadrature rule (Lambin and Gaspard, 1982). Comprehensive derivation
of the PD algorithm can be found in Bove (2005). In general, the computation of the
quadrature rule is unstable and sensitive to small errors, especially if large number of
moments is used. Later, McGraw and Wright (2003) derived the Jacobian Matrix
Transformation (JMT) for multicomponent population which avoids the instability induced
by the PD algorithm. Very recently, Grosch et al. (2007) proposed a generalized framework
for various QMOM approaches and evaluated different QMOM formulations in terms of
A Review of Population Balance Modelling for Multiphase Flows 123
numerical quadrature and dynamics simulation. Several studies have also been carried out
validating the method against different gassolid particle problems (Barrett and Webb, 1998;
Marchisio et al., 2003a,b,c). Encouraging results obtained thus far clearly demonstrated its
usefulness in solving monovariate problems and its potential fusing within Computational
Fluid Dynamics (CFD) simulations. One of the main limitations of the QMOM is that
moments are adopted to represent the PSD, each moment is “convected” in the same phase
velocity which is apparently nonphysical, especially for gasliquid flow where bubble could
be deformed and travel in different trajectory.
Figure 2. Graphical presentations of the Quadrature Method of Moments (QMOM).
2.2.2. Direct Quadrature Method of Moments (DQMOM)
With the aim to solve multidimensional problems, Marchisio and Fox (2005) extended
the method by developing the direct quadrature method of moment (DQMOM) where the
quadrature abscissas and weights are formulated as transport equations. The main idea of the
method is to keep track the primitive variables appearing in the quadrature approximation,
instead of moments of the PSD. As a result, the evaluation of the abscissas and weights are
solved obtained using matrix operations. Substitute Eq. (3) into Eq. (1) and after some
mathematical manipulations, transport equations for weights and abscissas are given by:
( )
( )
i i
i
a t r N t r u
t
t r N
= ⋅ ∇ +
∂
∂
) , ( ) , (
,
(4)
( )
( )
i i
i
b t r t r u
t
t r
= ⋅ ∇ +
∂
∂
) , ( ) , (
,
ζ
ζ
(5)
Sherman C.P. Cheung, G.H. Yeoh and J.Y. Tu 124
where
i i i
x N = ζ is the weighted abscissas and the terms
i
a and
i
b are related to the “birth” and
“death” rate of population which forms 2M linear equations of which unknowns can be
evaluated via matrix inversion:
d A = α (6)
where the 2M×2M coefficient matrix A= ] A A [
2 1
is given by:
⎥
⎥
⎥
⎥
⎥
⎥
⎦
⎤
⎢
⎢
⎢
⎢
⎢
⎢
⎣
⎡
−
−
−
− =
− − 1 2
2
1 2
1
2
1 1
) 1 ( 2 ) 1 ( 2
0
1
0
1
A
M
M
M
M
x M
x
x M
x
(7)
⎥
⎥
⎥
⎥
⎥
⎥
⎦
⎤
⎢
⎢
⎢
⎢
⎢
⎢
⎣
⎡
− −
=
− − 2 2 2 2
1
1 2
) 1 2 (
2
) 2 (
2
1
0
1
0
A
M
M
M
M
x M
x
x x M
x
(8)
The 2M vector of unknownsα is defined by:
[ ]
⎥
⎦
⎤
⎢
⎣
⎡
= =
b
a
b b a a
T
M M
1 1
α (9)
and the source on the RHS is:
[ ]
T
M
S S d
1 2 0 −
= (10)
The source term for the k
th
moment
k
S is defined by:
ξ ξ ξ d t r S t r S
k
k
) , , ( ) , (
0
∫
∞
= (11)
One attractive feature of the DQMOM is that it permits weights and abscissas to be
varied within the state space according to PSD evolution. Furthermore, different travelling
velocity can be also incorporated into transport equations allowing the flexibility to solve
polydispersed flows where weights and abscissas travel indifferent flow fields (Ervin and
Tryggvason, 1997; Bothe et al., 2006). In summary, the MOM represents a rather sound
A Review of Population Balance Modelling for Multiphase Flows 125
mathematical approach and an elegant tool of solving the PBE with limited computational
burden. Such approach with no doubt is an emerging technique for solving PBE, due to the
considerably short development history, thorough validation studies comparing model
predictions against experimental data are however outstanding. It can be concluded that the
DQMOM or other moment methods still require further assessments and validations for
various multiphase flow problems.
2.3. Class Method (CM) Approach
Instead of inferring the PSD to derivative variables (i.e. moments), the class method
(CM) which directly simulate its main characteristic using primitive variable (i.e. particle
number density) has received greater attention due to its rather straightforward
implementation within CFD software packages. In the method of discrete classes, the
continuous size range of particles is discretized into a series number of discrete size classes.
For each class, a scalar (number density of particles) equation is solved to accommodate the
population changes caused by intra/intergroup particle coalescence and breakage. The
particle size distribution is thereby approximated as follow:
) ( ) , (
1
i
M
i
i
x N t f − ≈
∑
=
ξ δ ξ (12)
The expression is exactly the same with QMOM in Eq. (3), however, the groups (or
abscissas) of class methods are fixed and aligned continuously in the state space. A graphical
representation of the CM in approximating the PSD is depicted in Figure 3.
Figure 3. Graphical presentations of Class Methods (CM).
Sherman C.P. Cheung, G.H. Yeoh and J.Y. Tu 126
2.3.1. Average Quantities Approach
The simplest approach in class methods is adopted an averaged quantity to represent the
overall changes of the particle population. Kocamustafaogullari and Ishii (1995) first derived
an interfacial area concentration (IAC) transport equation for tracking the interfacial area
between gas and liquid phase in bubbly flow problems. They concluded that the governing
factor of the interfacial transfer mechanisms is strongly dominated by the interfacial area
concentration. Modelling of the interfacial area concentration is therefore essential. In
addition, as the population of particle is represented by a single average scalar, such average
quantity approach requires very limited computational time in solving the PBE, which
provides an attractive feature for practical engineering problems.
Following their study, Ishii and his coworkers preformed extended the capability of their
(IAC) transport model to simulate different bubbly flow regime in different flow conditions
(Wu et al, 1998; Hibiki and Ishii, 2002; Fu et al., 2002a,b; Sun et al., 2004a,b). A series of
experimental studies covering a wide range of flow conditions have been carried out in order
to provide a solid foundation for their model development and calibration. Recently, similar
modelling approach has been also adopted by Yao and Morel (2004) with the attempt of
better improving the bubble coalescence and breakage kernels. On the other hand, equivalent
to the formulation of the interfacial area transport equation, an Average Bubble Number
Density (ABND) equation has been proposed very recently in our previous studies (Yeoh and
Tu, 2006; Cheung et al., 2007).
2.3.2. MUltiple SIze Group (MUSIG) model
Besides the proposed the average quantity approach, a more sophisticated model, namely
homogeneous MUltipleSIzeGroup (MUSIG) model which first introduced by Lo (1996) is
becoming widely adopted. Research studies based on the by Pochorecki et al. (2001), Olmos
et al. (2001), Frank et al. (2004), Yeoh and Tu (2005) and Cheung et al. (2007) typified the
application of MUSIG model in bubbly flow simulations. In the MUSIG model, the
continuous particle size distribution (PSD) function is approximated by M number size
fractions; the mass conversation of each size fractions are balanced by the interfraction mass
transfer due to the mechanisms of particle coalescence/agglomeration and breakage
processes. The overall PSD evolution can then be explicitly resolved via source terms within
the transport equations.
Snayal et al. (2005) examined and compared the CM and QMOM in a twodimensional
bubbly column simulation; both methods were found to yield very similar results. The CM
solution has been found to be independent of the resolution of the internal coordinate if
sufficient number of classes were adopted. Computationally speaking, as the number of
transport equations depends on the number of group adopted, the MUSIG model requires
more computational time and resources than the MOM to achieve stable and accurate
numerical predictions. For typical bubbly flow simulations, the model requires around 10
groups to yield accurate results.
A Review of Population Balance Modelling for Multiphase Flows 127
Figure 4. Schemetic diagram of the homogeneous and imhomogeneous MUSIG models.
Nonetheless, unlike the QMOM, CM provides the feasibility of accounting different
bubble shapes and travelling gas velocities. The inhomogeneous MUSIG model developed by
Krepper et al. (2005), which consisted of subdividing the dispersed phase into N number of
velocity fields, demonstrated the practicability of such an extension. This flexibility
represents a robust feature for multiphase flows modelling, especially for bubbly flow
simulations where bubbles may deform into different shapes. Figure 4 shows the concept of
the inhomogeneous MUSIG in comparison to homogeneous MUSIG. Useful information on
the implementation and application of the inhomogeneous MUSIG model can be found in Shi
et al. (2004) and Krepper et al. (2007). In spite of the sacrifices being made to computational
efficiency, the extra computational effort will rapidly diminish due to foreseeable
advancement of computer technology; the class method should therefore suffice as the
preferred approach in tackling more complex multiphase flows.
The formulation of the MUSIG model originates from the discretised PSE is given by:
( ) ( )
i ph
i
i i
i
R R n u
t
n
) ( − = ⋅ ∇ +
∂
∂
∑
(13)
To ensure overall mass conservation for all polydispersed vapour phases, the above
bubble number density equation for the inhomogeneous MUSIG model can be reexpressed in
terms of the volume fraction and size fraction of the bubble size class i, ] , 1 [
j
M i ∈ , of
velocity group j, [ ] N j , 1 ∈ according to:
( )
i ph i i j j i j g
i j j
R m S u f α ρ
t
f α ρ
) (
,
− = ⋅ ∇ +
∂
∂
(14)
with additional relations and constraints:
∑∑ ∑
= = =
= =
N
j
M
i
ij
N
j
j g
j
1 1 1
α α α ;
∑
=
=
j
M
i
i j
1
α α
Sherman C.P. Cheung, G.H. Yeoh and J.Y. Tu 128
1 = +
l g
α α ;
∑
=
=
j
M
i
i
f
1
1 (15)
where m
i
is the mass fraction of the particular size group I and
ph
R is the mass transfer rate
due to phase changes which will be discussed in the coming sections.
On the right hand side of Eq. (14) , the term ( ) ( )
B C B C
i
i i j
D D P P R m S − − + = =
∑ ,
represents the net mass transfer rate of the bubble class i resulting from the source of
C
P ,
B
P ,
C
D and
B
D , which are the production rates due to coalescence and breakage and the
death rate due to coalescence and breakage of particles respectively. They can be formulated
as:
∑∑
= =
=
i
k
i
l
j i ij jki C
n n P
1 1
2
1
χ η
) /( ) (
1 1 − −
− − +
i i i k j
ν ν ν ν ν if
i k j i
ν ν ν ν < + <
−1
=
jki
η ) /( ) (
1 1 i i k j i
ν ν ν ν ν − + −
+ +
if
1 +
< + <
i k j i
ν ν ν ν
0 otherwise
∑
=
=
N
j
j i ij C
n n D
1
χ
( )
j i j
N
i j
B
n v v Ω P :
1
∑
+ =
=
i i B
n Ω D = with
∑
=
=
N
k
ki i
Ω Ω
1
(16)
with
∑
=
=
j
M
i
i j j
S S
1
,
and 0
1
=
∑
=
N
j
j
S (17)
3. Population Balance Modelling for Bubbly Flows
In the previous section, approaches of population balance modelling have been reviewed.
One should now realize the broad application of population balance modelling in practical
engineering problems. In essence, population balance has been applied in numerous
multiphase flow systems. For example: predicting the soot formation rate from incomplete
A Review of Population Balance Modelling for Multiphase Flows 129
combustion processes (Frenklach and Wang, 1994; Zucca et al., 2007); solving the turbulent
reacting flow in fluidized bed (Lakatos et al., 2008; Khan et al., 2007; Fan et al., 2004);
evaluating the bubble side distribution and interfacial area in bubble column (Sha et al., 2006;
Borel et al., 2006; Jia et al., 2007). In the following sections, to exemplify the capability of
population balance modelling in conjunction with computational fluid dynamics framework,
numerical simulations based on the MUSIG model of the CM were performed to predict the
evolution of bubble size in both isothermal and heat and mass transfer conditions. Predictions
were validated against experimental data measured by Hibiki et al. (2001), Yun et al. (1997)
and Lee et al. (2002). Before presenting the methodology of the numerical study,
phenomenological discussion and background of bubbly flows is firstly provided below.
Figure 5. Schematic of the physical phenomenon embedded in (a) isothermal bubbly flows and (b)
subcooled boiling flows.
3.1. Isothermal Bubbly Flows
For bubbly flows, the dynamic evolution of the PSD is governed predominantly by the
bubble mechanistic behaviours such as bubble coalescence and bubble breakage. Figure 5a
illustrates the physical characteristic of a typical isothermal bubbly flow. Owing to the
positive lift force created by the lateral velocity gradient, small bubbles (under 5.5 mm for
airwater flows) being injected at the bottom have a tendency to migrate towards the channel
walls thereby increasing the bubble number density near the wall region. The net effect of
Sherman C.P. Cheung, G.H. Yeoh and J.Y. Tu 130
bubble coalescence encourages the formation of larger bubbles downstream. Larger bubbles
(above 5.5mm), driven by the negative lift force, will move towards to the centre core of the
channel of which they will further coalesce with other bubbles to yield distorted/cap bubbles.
These mechanisms strongly govern the distribution of bubble size and void fraction of the gas
phase within the bulk liquid; appropriate submodels (or kernels) in describing the bubble
coalescence and bubble breakage are essential to the proper modelling of bubbly flows. The
Prince and Blanch (1990) coalescence and breakage kernels are widely applied for the bubbly
flow simulations. Various models focusing towards modifying the coalescence frequency and
daughter bubble distribution due to breakage have also been proposed such as those from
Chester et al. (1982), Luo and Svendsen
(1996), Lehr et al. (2001), Hagesaether et al. (2002),
Wang et al. (2003, 2005, 2006) and Andersson and Andersson
(2006). Chen et al. (2005)
compared various kernels using a twodimensional twofluid model and showed that although
the mechanistic considerations were different in each model, the predicted results were found
Figure 6. Schematic drawings illustrating (a) the mechanism of bubble departing, sliding and lifting off
from a vertical heated surface and (b) the area of influence due to bubble growth and sliding referring to
Eq. (26).
A Review of Population Balance Modelling for Multiphase Flows 131
to be rather similar across the broad range of models. A number of research studies have also
been performed applying the aforementioned kernels to simulate size distribution evolution in
bubbly turbulent pipe flows (Yeoh and Tu, 2004; Sha et al., 2006; Borel et al., 2006; Jia et al.,
2007). All these works were nonetheless limited in solving isothermal bubbly turbulent co
flow problems with low superficial gas velocity where there was no significant formation of
cap/slug bubbles. It should be noted that the adoption of the above kernels remain debatable
for modelling the bubble dynamics beyond bubbly flow regime, especially for high
superficial gas velocity and flows involving complex interphase exchange of heat and mass
transfer.
Subcooled boiling flow belongs to another special category of bubbly turbulent pipe flow
which embraces the complex dynamic interactions of bubble coalescence and bubble
breakage in the bulk flow as well as the presence of heat and mass transfer occurring in the
vicinity of the heated wall due to nucleation and condensation. Heterogeneous bubble
nucleation occurs naturally within small pits and cavities on the heated surface designated as
nucleation sites as stipulated in Figure 5b, which is in contrast to having the bubbles being
injected externally at the bottom of the flow as illustrated in Figure 5a for isothermal bubbly
turbulent pipe flow. These nucleation sites, activated by external heat, act as a continuous
source generating relatively small bubbles along the vertical wall of the channel. The
presence of a heated wall represents the fundamental difference between isothermal bubbly
and subcooled boiling flows, where the former has a constant bubble injection rate to govern
the overall void fraction of gas phase, but the latter has a variable bubble nucleation rate
which is subjected mainly to the heat transfer and phase changing phenomena. Furthermore,
experimental observations have confirmed that the vapour bubbles, driving by external forces,
have a tendency to travel a short distance away from the nucleation sites, gradually increasing
in size, before lifting off into the bulk subcooled liquid (Klausner et al., 1993). Figure 6
shows a schematic illustration of the bubble motion and its area of influence on the heater
surface. Such bubble motion not only alters the mode of heat transfer on the surface, but also
governs the departure and liftoff diameter of bubbles, which in turn also influences the
bubble distribution in the bulk liquid. In isothermal flow, coalescence prevails in the channel
core where the turbulent dissipation rate is relatively low. Conversely, bubbles tend to
decrease in size for subcooled boiling flow as a result of increasing condensation away from
the heated walls since the temperature in the bulk liquid remains below the saturation
temperature limit. This subcooling effect is a wellknown phenomenon confirmed by various
experiments (Gopinath et al., 2002). Subject to this effect, such flows will certainly yield a
broader range of bubble sizes and possibly even amplifying greater dynamical changes of the
bubble size distribution when compared to isothermal bubbly turbulent pipe flow. Given the
challenging task of modelling the sophisticated phenomena, empirical equations (Anglart and
Nylund, 1996) were inevitably employed to determine the bubble diameter in the gas phase.
In order to make progress, a modified version of the MUSIG model incorporating nucleation
at the heated wall and condensation in the subcooled liquid were developed to better resolve
the problem (Yeoh and Tu, 2005; Yeoh and Tu, 2006). Although some empirical equations
were still retained to determine the bubble nucleation and detachment, the potential of
adopting population balance approach in subcooled boiling flow demonstrated considerable
success in aptly predicting the bubble Sauter diameter distribution of the gas bubbles for
vertical boiling flows.
Sherman C.P. Cheung, G.H. Yeoh and J.Y. Tu 132
4. Mathematical Models
Referring back to the formulation of PBE, one should notice that the lefthand side of the
equation denotes the time and spatial variations of the PSD which depends on the external
variables. By incorporating the PBE within CFD solver, external variables can be obtained. In
this section, governing equations of the two fluid model and its associated model for handling
interfacial momentum and mass transfer are introduced.
4.1. TwoFluid Model
The threedimensional twofluid model solves the ensembleaveraged of mass,
momentum and energy transport equations governing each phase. Denoting the liquid as the
continuum phase (α
l
) and the vapour (i.e. bubbles) as disperse phase (α
g
), these equations can
be written as:
Continuity Equation of Liquid Phase
( )
lg
Γ = ⋅ ∇ +
∂
∂
l l l
l l
u α ρ
t
α ρ
(18)
Continuity Equation of Vapour Phase
( )
lg
Γ − = ⋅ ∇ +
∂
∂
i i g i g g
i g g
f S u f α ρ
t
f α ρ
(19)
Momentum Equation of Liquid Phase
( ) g P u u α ρ
t
u α ρ
l l l l l l l
l l l
ρ α α + ∇ − = ⋅ ∇ +
∂
∂
(20)
lg gl lg
) ( )] ) ( ( [ F u u u u
l g
T
l l
e
l l
+ Γ − Γ + ∇ + ∇ ∇ +
μ α
Momentum Equation of Vapour Phase
( ) g P u u α ρ
t
u α ρ
g g g g g g g
g g g
ρ α α + ∇ − = ⋅ ∇ +
∂
∂
(21)
gl g l
T
g g
e
g g
F u u u u + Γ − Γ + ∇ + ∇ ∇ + ) ( )] ) ( ( [
lg gl
μ α
Energy Equation of Liquid Phase
( ) ) ( )] ( [
lg gl g l l
e
l l l l l l
l l l
H H T H u α ρ
t
H α ρ
Γ − Γ + ∇ ∇ = ⋅ ∇ +
∂
∂
λ α
(22)
A Review of Population Balance Modelling for Multiphase Flows 133
Energy Equation of Vapour Phase
( ) ) ( )] ( [
lg gl g l g
e
g g g g g g
g g g
H H T H u α ρ
t
H α ρ
Γ − Γ + ∇ ∇ = ⋅ ∇ +
∂
∂
λ α
(23)
On the righthand side of equation (4), S
i
represents the additional source terms due to
coalescence and breakage. For isothermal bubbly turbulent pipe flows, it should be noted that
the mass transfer rate
lg
Γ and
gl
Γ are essentially zero. The total interfacial force
lg
F appearing
in equation (5) is formulated according to appropriate consideration of different subforces
affecting the interface between each phase. For the liquid phase, the total interfacial force is
given by:
dispersion
lg
n lubricatio
lg
lift
lg
drag
lg lg
F F F F F + + + = (24)
The subforces appearing on the right hand side of equation (9) are: drag force, lift force,
wall lubrication force and turbulent dispersion force. More detail descriptions of these sub
forces can be found in Anglart and Nylund (1996). Note that for the gas phase, F
gl
=  F
lg
.
The interfacial mass transfer rate due to condensation in the bulk subcooled liquid in
equation (4) can be expressed by:
fg
l sat if
h
T T a h ) (
lg
−
= Γ (25)
Here, h indicates the interphase heat transfer coefficient which is correlated in terms of
the Nusselt number (Tu and Yeoh, 2002). The wall generation rate for the vapour is modelled
in a mechanistic manner by considering the total mass of bubbles detaching from the heated
surface as:
) (
l sat pl fg
e
gl
T T C h
Q
− +
= Γ (26)
Here, Q
e
refers as the heat transfer rate due to evaporation. For subcooled boiling flows,
the wall nucleation rate is accounted in equation (4) as a specified boundary condition
apportioned to the discrete bubble class based on the size of the bubble liftoff diameter,
which is evaluated from the improved wall heat partition model. The term
lg
Γ
i
f represents the
mass transfer due to condensation. The gas void fraction along with the scalar size fraction f
i
are related to the number density of the discrete bubble ith class n
i
(similarly to the jth class
n
j
) as
i i i g
n f ν α = .
Sherman C.P. Cheung, G.H. Yeoh and J.Y. Tu 134
4.1.1. Turbulence Modelling for TwoFluid Model
In handling bubble induced turbulent flow, unlike single phase fluid flow problem, no
standard turbulence model is tailored for multiphase flow. For simplicity, the standard kε
model has been employed with encouraging results in early studies (Schwarz and Turner,
1988; Davidson, 1990). Nonetheless, based on our previous study (Cheung et al, 2006), the
Menter’s (1994) kω based Shear Stress Transport (SST) model were found superior to the
standard kε model. Similar observations have been also reported by Frank et al. (2004).
Based on their bubbly flow validation study, they discovered that standard kε model
predicted an unrealistically high gas void fraction peak close to wall. Interestingly, they also
found that the two turbulence models behaved very similar by reducing the inlet gas void
fraction to a negligible value. This could be attributed to a more realistic prediction of
turbulent dissipation close to wall provided by the kω formulation. It revealed that further
development should be focused on multiphase flow turbulence modelling in order to better
understand or improve the existing models.
The SST model is a hybrid version of the kε and kω models with a specific blending
function. Instead of using empirical wall function to bridge the wall and the faraway
turbulent flow, it solves the two turbulence scalars (i.e. k and ω) explicitly down to the wall
boundary. The ensembleaveraged transport equations of the SST model are given as:
( )
l l l l k l l
k
l t
l l l l l l
l l l
k ρ P k α k u α ρ
t
k α ρ
ω β α
σ
μ
μ ′ − +
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
∇ + ⋅ ∇ = ⋅ ∇ +
∂
∂
,
3
,
) (
( )
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
∇ + ⋅ ∇ = ⋅ ∇ +
∂
∂
l
l t
l l l l l l
l l l
α u α ρ
t
α ρ
ω
σ
μ
μ ω
ω
ω
) (
3
,
2
3 , 3
2
1 1
1
) 1 ( 2
l l l k
l
l
l
j
l
j
l
l
l
ρ P
k x x
k
F ω β
ω
γ α
ω
ω σ
α ρ
ω
− +
∂
∂
∂
∂
− − (27)
where
3 k
σ ,
3 ω
σ ,
3
γ and
3
β are the model constants which are evaluated based on the blending
function F
1
. The shear induced turbulent viscosity
l ts,
μ is given by:
) , max(
2 1
1
,
SF a
k a
l
l
l ts
ω
ρ
μ = ,
ij ij
S S S 2 = (28)
The success of SST model hinges on the use of blending functions of F
1
and F
2
which
govern the crossover point between the kω and kε models. The blending functions are given
by:
) tanh(
4
1 1
Φ = F ,
⎥
⎥
⎦
⎤
⎢
⎢
⎣
⎡
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
= Φ
+ 2
2
2
1
4
,
500
,
09 . 0
max min
n
l l
n l l
l
n l
l
d D
k
d d
k
ω ω
σ
ρ
ω ρ
μ
ω
A Review of Population Balance Modelling for Multiphase Flows 135
) tanh(
2
2 2
Φ = F ,
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
= Φ
2 2
500
,
09 . 0
max
n l l
l
n l
l
d d
k
ω ρ
μ
ω
(29)
Here, default values of model constants were adopted. More detail descriptions of these
model constants can be found in Menter (1994). In addition, to account the effect of bubbles
on liquid turbulence, the Sato’s bubbleinduced turbulent viscosity model was also employed
(Sato et al., 1981). The turbulent viscosity of liquid phase is therefore given by:
l td l ts l t , , ,
μ μ μ + = (30)
and the particle induced turbulence can be expressed as:
l g S g l p l td
U U D C
− = α ρ μ
μ ,
(31)
For the gas phase, dispersed phase zero equation model was adopted and the turbulent
viscosity of gas phase can be obtained as:
g
l t
l
g
g t
σ
μ
ρ
ρ
μ
,
,
= (32)
where
g
σ is the turbulent Prandtl number of the gas phase.
4.2. Coalescence and Breakage Kernels
4.2.1. Bubble Coalescence and Breakage Kernels
Bubble breakage rate of volume
j
v into volume
i
v is modelled according to Luo and
Svendsen (1996), which is based on the assumption of bubble binary breakage under isotropic
turbulence situation. The daughter size distribution is accounted using a stochastic breakage
volume fraction
BV
f . Denoting the increase coefficient of surface area as c
f
= [
3 / 2
BV
f +(1
BV
f )
2/3
1], the breakage rate can be obtained as:
( )
( )
( )
ξ
ξ ε
σ
ξ
d
d βρ
c
ξ
ξ
d
ε
C F
n α
v v Ω
l
f
j
B
j g
i j
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
− ×
+
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
=
−
∫ 3 / 11 3 / 5 3 / 2
1
3 / 11
2
3 / 1
2
12
exp
1
1
:
min
(33)
where
j
d / λ ξ = is the size ratio between an eddy and a particle in the inertial subrange and
consequently
j
d /
min min
λ ξ = and C and β are determined from fundamental consideration of
Sherman C.P. Cheung, G.H. Yeoh and J.Y. Tu 136
drops or bubbles breakage in turbulent dispersion systems to be 0.923 and 2.0. F
B
is the
breakage calibration factor.
Bubble coalescence occurs via collision of two bubbles which may be caused by wake
entrainment, turbulence random collision and buoyancy. Only turbulence random collision is
considered in the present study as all bubbles are assumed to be of spherical shape (wake
entrainment becomes negligible). The coalescence rate considering the turbulent collision
taken from Prince and Blanch (1990) can be expressed as:
[ ] ( )
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
− + + =
ij
ij
tj ti j i C ij
t
u u d d F
τ
π
χ exp
4
5 . 0
2 2 2
(34)
where
ij
τ is the contact time for two bubbles given by
3 / 1 3 / 2
/ ) 2 / ( ε
ij
d and
ij
t is the time
required for two bubbles to coalesce having diameter d
i
and d
j
estimated to be
) / ln( ] 16 / ) 2 / [(
0
5 . 0 3
f l ij
h h d σ ρ . The equivalent diameter d
ij
is calculated as suggested by
Chesters and Hoffman (1982):
1
) / 2 / 2 ( (
−
+ =
j i ij
d d d . According to Prince and Blanch
(1990), experiments have determined the initial film thickness h
o
=
4
10 1
−
× m and critical
film thickness h
f
=
8
10 1
−
× m at which rupture for airwater systems. The turbulent velocity
u
t
in the inertial subrange of isotropic turbulence (Rotta, 1972) which is given
by:
3 / 1 3 / 1
2 d u
t
ε = . F
C
is the coalescence calibration factor.
4.2.2. Bubble Source and Sink Due to Phase Change in Subcooled Boiling Flows
The term
i ph
R ) ( in Eq. (14) constitutes the essential formulation of the source/sink rate
for the phase change processes associated with subcooled boiling flow. Considering the
condensation of bubbles, the bubble condensation rate in a control volume for each bubble
class can be determined from:
dt
dR
A
V
n
B
B
i
COND
− = φ (35)
The bubble condensation velocity
(Gopinath et al., 2002) is obtained from:
fg g
l sat
h
T T h
dt
dR
ρ
) ( −
= (36)
Substituting Eq. (21) into Eq. (20), and given that the bubble surface are A
B
and volume
V
B
which are based on Sauter mean bubble diameter, Eq. (20) can be rearranged to yield:
A Review of Population Balance Modelling for Multiphase Flows 137
i
fg
l sat if
i g
COND i ph
n
h
T T ha
R
⎥
⎥
⎦
⎤
⎢
⎢
⎣
⎡ −
− = =
) (
1
) (
α ρ
φ (37)
At the heated surface, bubbles at the nucleation sites are formed through the evaporation
processes. The bubble nucleation rate from these sites can be expressed as:
C
H
WN
A
Nafξ
φ = (38)
where Na , f,
H
ξ and A
C
is the active nucleation site density, the bubble generation frequency
from the nucleation sites, the heated perimeter, and the crosssectional area of the boiling
channel, respectively. Since the bubble nucleation process is taken to occur only at the heated
surface, this heated wall nucleation rate is treated as a specified boundary condition to Eq.
(14) apportioned to the discrete bubble class i, based on the bubble liftoff diameter
determined from the mechanistic wall heat partition model.
4.3. Mechanistic Wall Heat Partition Model
To determine the bubble generation frequency and liftoff diameter for the boundary
condition of the MUSIG model, an improved wall heat partition model based on a
mechanistic approach is proposed (Yeoh et al., 2008) and briefly discussed in this section.
With the presence of convective force or buoyancy force acting upon a vertically orientated
boiling flow as depicted in Figure 6a, vapour bubble departs from its nucleation site, slides
along the heating surface and continues to grow downstream until it lifts off from the surface
(Klausner et al., 1993). The motion of the travelling bubble affects the heat transfer at the
heated wall according to two mechanisms: (i) the latent heat transfer due to microlayer
evaporation and (ii) transient conduction as the disrupted thermal boundary layer reforms
during the waiting period (i.e. incipience of the next bubble at the same nucleation site).
4.3.1. Transient Conduction Due to Bubble Motion
Transient conduction occurs in regions at the point of inception and in regions being
swept by sliding bubbles. For a stationary bubble, the heat flux is given by:
f t
D
K N R T T
t
C k
Q
w
d
a f l s
w
pl l l
tc
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
− =
4
) ( 2
2
π
π
ρ
) 1 (
4
) ( 2
2
f t
D
N R T T
t
C k
w
d
a f l s
w
pl l l
−
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
− +
π
π
ρ
(39)
where D
d
is the bubble departure diameter, T
s
is the temperature of the heater surface and T
l
is
the temperature of the liquid. Eq. (24) indicates that some fraction of the nucleation sites will
Sherman C.P. Cheung, G.H. Yeoh and J.Y. Tu 138
undergo transient conduction while the remaining will be in the growth period. For a sliding
bubble, the heat flux due to transient conduction that takes place during the sliding phase and
the area occupied by the sliding bubble at any instant of time is given by:
f KDt l N R T T
t
C k
Q
w s a f l s
w
pl l l
tcsl
) ( 2 − =
π
ρ
) 1 (
4
) ( 2
2
f t
D
ft N R T T
t
C k
w sl a f l s
w
pl l l
−
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
− +
π
π
ρ
(40)
where the average bubble diameter D is given by ( ) 2
l d
D D D + = and D
l
is the bubble
liftoff diameter. In this study, a value of 1.8 is assumed for the area of influence, K
(Judd
and Hwang, 1976). The reduction factor R
f
appearing in equations (24) and (25) depicts the
ratio of the actual number of bubbles lifting off per unit area of the heater surface to the
number of active nucleation sites per unit area, viz., ( ) s l R
s f
/ 1 = where l
s
is the sliding
distance and s is the spacing between nucleation sites. In the present study, the spacing
between nucleation sites is approximated as
a
N s 1 = (Basu et al., 2005). The active
nucleation site density, N
a
, is expressed by ( )
805 . 1
] 185 [
sat w a
T T N − = (Končar et al., 2004).
The factor R
f
is obtained alongside with the sliding distance evaluated from the force balance
model described below.
4.3.2. Force Convection for Single Phase Component
Forced convection always prevails at all times in areas of the heater surface that are not
influenced by the stationary and sliding bubbles (see also in Figure 6b). The fraction of the
heater area for stationary and sliding bubbles is given by:
) 1 (
4 4
1 1
2 2
f t
D
N f t
D
K N R A
w
d
a w
d
a f q
−
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
+
⎢
⎢
⎣
⎡
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
− = −
π π
⎥
⎥
⎦
⎤
−
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
+ + ) 1 (
4
2
f t
D
ft N f KDt l N
w
d
sl a w s a
π
(41)
The heat flux due to forced convection can be obtained according to the definition of
local Stanton number St for turbulent convection is:
) )( 1 (
l w q l pl l c
T T A u C St Q − − = ρ (42)
where u
l
is the adjacent liquid velocity.
A Review of Population Balance Modelling for Multiphase Flows 139
4.3.3. Latent Heat Due to Vapour Evaporation Processes
The heat flux attributed to vapour generation is given by the energy carried away by the
bubbles lifting off from the heated surface. It also represents the energy of vaporization
whereby the bubble size of D
l
is produced, which is expressed as:
fg g
l
a f e
h
D
f N R Q ρ
π
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
=
6
3
(43)
The total wall heat flux Q
w
is the combination of the following heat flux components: Q
w
= Q
e
+ Q
tc
+ Q
tcsl
+ Q
c
.
Mechanistic Approach for Bubble Frequency Evaluation
The bubble nucleation rate in Eq. (23) requires the knowledge of the bubble frequency
(f). Within the wall partition model, the bubble frequency is determined by a mechanistic
approach based on the description of an ebullition cycle in nucleate boiling, which is
formulated as:
w g
t t
f
+
=
1
(44)
The waiting period (t
w
) and the growth period of vapour bubbles (t
g
) is derived from the
transient conduction and force balance model respectively.
When transient conduction occurs, the boundary layer gets disrupted and cold liquid
comes in contact with the heated wall. Assuming that the heat capacity of the heater wall
ρ
s
C
ps
δ
s
is very small, the conduction process can be modelled by considering onedimensional
transient heat conduction into a semiinfinite medium with the liquid at a temperature T
l
and
the heater surface at a temperature T
s
. The wall heat flux can be approximated by:
l
l s l
w
T T k
Q
δ
) ( −
= (45)
where δ
l
(= t πη ) is the thickness of the thermal boundary layer. If the temperature profile
inside this layer is taken to be linear (Hsu and Graham, 1976), it can be expressed as:
l
l s
w b
x T T
T T
δ
) ( −
− = (46)
where x is the normal distance from the wall. Based on the criterion of the incipience of
boiling from a bubble site inside the thermal boundary layer, the bubble internal temperature
for a nucleus site (cavity) with radius r
c
is
Sherman C.P. Cheung, G.H. Yeoh and J.Y. Tu 140
g fg c
sat
sat b
h r C
T
T T
ρ
σ
2
2
− = at
c
r C x
1
= (47)
where θ θ sin / ) cos 1 (
1
+ = C and θ sin / 1
2
= C . The angle θ represents the bubble
contact angle. By substituting Eq. (32) into Eq. (31), the waiting time t
w
can be obtained as
2
2
1
/ 2 ) (
) ( 1
⎥
⎥
⎦
⎤
⎢
⎢
⎣
⎡
− −
−
= =
c fg g sat sat w
c l s
w
r h C T T T
r C T T
t t
ρ σ πη
(48)
The cavity radius r
c
can be determined by applying Hsu’s criteria and tangency condition
of equations (32) and (33), viz.,
πη πη σ
ρ
2
2
2
2
2
2 1
) ( ) (
2
l s
w
l l s
sat
c fg g
T T
Q
k T T
T
r h C C
t
−
⎥
⎦
⎤
⎢
⎣
⎡
=
−
⎥
⎥
⎦
⎤
⎢
⎢
⎣
⎡
= (49)
From the above equation,
2 / 1
2
⎥
⎥
⎦
⎤
⎢
⎢
⎣
⎡
=
w fg g
l sat
c
Q h
k T
F r
ρ
σ
(50)
where,
2 / 1
2
2 / 1
2 1
cos 1
sin 1
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
+
=
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
=
θ
θ
C C
F (51)
According to Basu et al. (2005), the factor F indicates the degree of flooding of the
available cavity size and the wettability of the surface. If the contact angle θ → 0, all the
cavities will be flooded. Alternatively, as θ → 90
o
, F → 1, all the cavities will not be flooded
(i.e. they contain traces of gas or vapour).
4.3.4. Force Balance Model and the Bubble Growth Time
The bubble growth time is correlated to its liftoff diameter which depends on various
forces acting on the bubble in the directions parallel and normal to a vertical heating surface.
Figure 7 illustrates the forces acting on the bubble in the xdirection and ydirection which
can be formulated according to the studies performed by Klausner et al. (1993) and Zeng et
al. (1993):.
ΣF
x
=F
sx
+ F
dux
+ F
sL
+ F
h
+ F
cp
(52)
A Review of Population Balance Modelling for Multiphase Flows 141
and
ΣFy =F
sy
+ F
duy
+ F
qs
+ F
b
(53)
where F
s
is the surface tension force, F
du
is the unsteady drag due to asymmetrical growth of
the bubble and the dynamic effect of the unsteady liquid such as the history force and the
added mass force, F
sL
is the shear lift force, F
h
is the force due to the hydrodynamic pressure,
F
cp
is the contact pressure force accounting for the bubble being in contact with a solid rather
than being surrounded by liquid, F
qs
is the quasi steadydrag in the flow direction, and F
b
is
the buoyancy force. In addition, g indicates the gravitational acceleration; α, β and θ
i
are the
advancing, receding and inclination angles respectively; d
w
is the surface/bubble contact
diameter; and d is the vapour bubble diameter at the wall.
Figure 7. Schematic drawings illustrating the forces balance of a growing vapour bubble attached to the
heated surface.
The forces acting in the xdirection can be estimated from:
] cos [cos α β
β α
π
σ −
−
− =
w sx
d F ; θ cos
du dux
F F − =
2 2
2
1
r U C F
l L sL
π ρ Δ = ;
4 4
9
2
2 w
l h
d
U F
π
ρ Δ =
Sherman C.P. Cheung, G.H. Yeoh and J.Y. Tu 142
r
w
cp
r
d
F
σ π 2
4
2
= (54)
The forces acting in the xdirection can be estimated from:
] sin [sin
) (
) (
2 2
β α
β α π
β α π
σ +
− −
−
− =
w sy
d F ; θ sin
du duy
F F − =
r U C F
l D qs
π μ Δ = 6 ; ( )g r F
g l b
ρ ρ π − =
3
3
4
(55)
From the various forces described along the xdirection and ydirection, r is the bubble
radius, ΔU is the relative velocity between the bubble centre of mass and liquid, C
D
and C
L
are the respective drag and shear lift coefficients and r
r
is the curvature radius of the bubble at
the reference point on the surface x = 0, which is r
r
∼ 5r (Klausner et al., 1993).
The growth force F
du
is modelled by considering a hemispherical bubble expanding in an
inviscid liquid, which is given by Zeng et al. (1993) as:
⎟
⎠
⎞
⎜
⎝
⎛
+ = r r r C r F
s l du
2 2
2
3
π ρ (56)
where ( ˙ ) indicates differentiation with respect to time. The constant C
s
is taken to be 20/3
(Zeng et al., 1993). In estimating the growth force, additional information on the bubble
growth rate is required. As in Zeng et al. (1993), a diffusion controlled bubble growth
solution by Zuber (1961) is adopted:
t Ja
b
t r η
π
2
) ( = ;
fg g
sat pl l
h
T C
Ja
ρ
ρ Δ
= ;
pl l
l
C
k
ρ
η = (57)
where Ja is the Jakob number, η is the liquid thermal diffusivity and b is an empirical
constant that is intended to account for the asphericity of the bubble. For the range of heat
fluxes investigated in this investigation, b is taken to be 0.21 based on a similar subcooled
boiling study performed by Steiner et al. (2006), which has been experimentally verified
through their inhouse measurements with water as the working fluid.
While a vapour bubble remains attached to the heated wall, the sum of the parallel and
normal forces must satisfy the following conditions: ΣF
x
= 0 and ΣF
y
= 0. For a sliding
bubble case, the former establishes the bubble departure diameter (D
d
) while the latter yields
the bubble liftoff diameter (D
l
). The growth period t
g
appearing in Eq. (29) can be readily
evaluated based on the availability of the bubble size at departure from its nucleation site
through Eq. (42). Details of the present wall partition model can also be found in literature
(Yeoh et al., 2008) and the references therein.
A Review of Population Balance Modelling for Multiphase Flows 143
4.4. Experimental Details
Brief discussions of the experimental setup for the isothermal and subcooled bubbly
turbulent pipe flows are provided below.
Isothermal bubbly turbulent pipe flow experiments have been performed at the Thermal
Hydraulics and Reactor Safety Laboratory in Purdue University (2001). The test section
comprised of an acrylic round pipe with an inner diameter D = 50.8 mm and a length of 3061
mm. Temperature of the apparatus was kept at a constant temperature (i.e. 20
o
C) within a
deviation of ±0.2
o
C controlled by a heat exchanger installed in a water reservoir. Local flow
measurements using the double sensor and hotfilm anemometer probes were performed at
three axial (height) locations of z/D = 6.0, 30.3 and 53.5 and 15 radial locations of r/R = 0 to
0.95. Experiments at a range of superficial liquid velocities j
f
and superficial gas velocities j
g
were performed covering most bubbly flow regions including finely dispersed bubbly flow
and bubblytoslug transition flow regions.
A series of subcooled boiling experiments have been performed by Yun et al. (1997) and
Lee et al. (2002). The experimental setup consisted of a vertical concentric annulus with an
inner rod of 19 mm outer diameter uniformly heated by a 54 kW DC power supply. This
heated section comprised of a 1.67 m long Inconel 625 tube with a 1.5 mm wall thickness
filled with magnesium oxide powder insulation. The outer wall comprised of two stainless
steel tubes with 37.5 mm inner diameter. Demineralised water was used as the working fluid.
Local gas phase parameters such as radial distribution of the void fraction, bubble frequency
and bubble velocity were measured by a twoconductivity probe method located 1.61 m
downstream of the beginning of the heated section. The bubble Sauter diameters (assuming
spherical bubbles) were determined through the IAC, calculated using the measured bubble
velocity spectrum and bubble frequency.
4.5. Numerical Details
For isothermal gasliquid bubbly turbulent pipe flow, the generic CFD code ANSYS
CFX 11 (2006) was utilised to handle the two sets of equations governing conservation of
mass and momentum. Numerical simulations were performed on a 60
o
radial sector of the
pipe with symmetry boundary conditions imposed at the end vertical sides. At the test section
inlet, uniformly distributed superficial liquid and gas velocities, void fraction and bubble size
were specified. Details of the boundary conditions for different flow conditions are
summarized in Table 1.
At the pipe outlet, a relative averaged static pressure of zero was specified. A three
dimensional mesh containing hexagonal elements was generated resulting in a total of
108,000 elements over the entire pipe domain. For all flow conditions, bubble size in the
range of 010 mm was discretised into 10 bubble classes as tabulated in Table 2. The
prescribed 10 size groups were further divided into two velocity fields for the inhomogeneous
MUSIG model to predict the transition of bubblyslug flow. Reliable convergence were
achieved within 2500 iterations when the RMS (root mean square) pressure residual dropped
below 1.0×10
7
. A fixed physical time scale of 0.002s was adopted for all steady state
simulations.
Sherman C.P. Cheung, G.H. Yeoh and J.Y. Tu 144
Table 1. Bubbly flow conditions and its inlet boundary conditions employed in the
present study
Superficial liquid velocity,
f
j (m/s)
Superficial gas velocity,
g
j (m/s)
Hibiki et al. (2001) experiment Bubbly flow Regime Transition Regime
0.491
[
0 0. D / z
g
α
=
(%)]
[
0 0. D / z
S
D
=
(mm)]

0.0556
[10.0]
[2.5]

0.986
[
0 0. D / z
g
α
=
(%)]
[
0 0. D / z
S
D
=
(mm)]
0.0473
[5.0]
[2.5]
0.1130
[10.0]
[2.5]
0.242
[20.0]
[2.5]
Table 2. Diameter of each discrete bubble class for MUSIG model
Class No. Central Class Diameter d
i
(mm)
Isothermal Subcooled Boiling
1 0.5 0.45
2 1.5 0.94
3 2.5 1.47
4 3.5 2.02
5 4.5 2.58
6 5.5 3.14
7 6.5 3.71
8 7.5 4.27
9 8.5 4.83
10 9.5 5.40
11  5.96
12  6.53
13  7.10
14  7.66
15  8.23
For subcooled boiling flow, numerical solutions were obtained from the two sets of
transport equations governing not only mass and momentum but also energy using the generic
CFD code CFX 4.4. A total number 15 bubble classes were specified for the dispersed phases
in the homogeneous MUSIG model (see Table 2). Similar to the isothermal flow simulations,
only one quarter of the annulus geometry was modelled due to the uniform prescription of the
heat flux at the inner wall. A bodyfitted conformal system was employed to generate the
threedimensional mesh within the annular channel resulting in a total of 13 (radial) × 30
A Review of Population Balance Modelling for Multiphase Flows 145
(axial) × 3 (circumference) control volumes. A standard kε model was applied for both
phases while additional turbulent viscosity induced by bubbles was included using Sato’s
model
95
. Since wall function was used in the present study to bridge the wall and the fully
turbulent region away from heater surface, the normal distance between the wall and the first
node in the bulk liquid should be such that the corresponding x
+
was greater than 30. Grid
independence was examined. In the mean parameters considered, further grid refinement did
not reveal significant changes to the twophase flow parameters. Convergence was achieved
within 1500 iterations when the mass residual dropped below 1.0×10
7
. Experimental
conditions used for comparison with the simulated results are tabulated in Table 3.
Table 3. Subcooled boiling flow conditions measured by Yun et al. (1997) adopted in the
present numerical study
Case P
inlet
[Mpa] T
inlet
[°C]
T
sub
(inlet)
[°C]
Q
w
[kW/m
2
] G [kg/m
2
s]
L1 0.143 96.9 13.4 152.9 474.0
L2 0.137 94.9 13.8 197.2 714.4
L3 0.143 92.1 17.9 251.5 1059.2
5. Results and Discussion
5.1. Isothermal Bubbly Turbulent Pipe Flow Results
Preliminary numerical simulations performed in this study over a range of superficial gas
and liquid velocities within the bubbly flow regime have consistently resulted in the prediction
of larger than expected bubble sizes. However, these results were found to clearly contradict the
measurements performed by Hibiki et al. (2001) as well as some experimental observations
(Bukur et al., 1996; George et al., 2000). One plausible explanation for this discrepancy could
possibly be the error embedded in the turbulent dissipation rate prediction (Bertola et al., 2003)
as a consequence of the turbulence model being applied contributing in turn to excessively high
coalescence rates in the MUSIG model. As reported in Chen et al. (2005), similar observations
also confirmed the high coalescence rates that were experienced in their bubble column study.
They argued that the local coalescence rate should be reduced by an order of magnitude lower
than the local breakage rate by a factor of about 10. Olmos et al. (2001) further demonstrated
the need to prescribe suitable calibration factor in order to aptly predict the bubble size
distributions in their bubble column flow configuration. A value of 0.075 was assumed for the
coalescence calibration factor. According to similar arguments stipulated above, the coalescence
and breakage calibration factors (i.e. F
C
and F
B
), were set as 0.05 and 1.0 throughout all
intended simulations of isothermal flow conditions (Cheung et al., 2007b).
5.1.1. Local Distributions of Void Fraction snd Interfacial Gas Velocity
Depending on the range of superficial velocities of gas and liquid experienced in the
bubbly flow regime, the phase patterns of the vapour void fraction can be broadly categorized
Sherman C.P. Cheung, G.H. Yeoh and J.Y. Tu 146
into four types of distributions: “wall peak”, “intermediate peak”, “core peak” and
“transition” (Serizawa an Kataoka, 1988). Figure 8 compares the gas void fraction profiles
obtained from the homogeneous MUSIG model with the experimental data measured at the
location of z/D = 53.5 (i.e. close to the exit of channel) for three different bubbly turbulent
pipe flow conditions. The high void fractions close to wall proximity typically characterised
the “wall peak” behaviour. Its phenomenological establishment can be best described by the
balance between the positive lift force that acted to impel the bubbles away from the central
core of the flow channel and the opposite effect being imposed by the lubrication force
preventing the bubbles from being obliterated at the channel walls. In spite of similar trends
predicted, the predicted void fraction peaks, on closer examination, appeared to be leaning
more towards the channel wall in contrast to the actual bubble distributions observed during
experiments. Against the assessment on other wall lubrication models by Frank et al. (2004)
and Tomiyama (1998), a much lower than expected wall force determined via Antal et al.
(1991) model purported to be the most probable cause of the discrepancy. As aforementioned,
the twophase turbulence modelling that could affect the magnitude of the turbulence induced
dispersion force could also contribute to the additional modelling uncertainty.
Figure 8. Predicted radial void fraction distribution and experimental data of Hibiki et al.(2001) at
measuring station of z/D =53.5
A Review of Population Balance Modelling for Multiphase Flows 147
Figure 9. Predicted interfacial gas velocity distribution and experimental data of Hibiki et al.
(2001) at
measuring station of z/D =53.5.
Figure 9 illustrates the local interfacial gas velocity distributions at the measuring station
of z/D = 53.5, close to the outlet of the pipe. Unlike in single phase flows, the presence of
bubbles has the tendency to enhance the liquid flow turbulence intensity (Serizawa et al.,
1990; Hibiki et al., 2001). Additional turbulence being experienced at the core flattened the
liquid velocity profile as expected. Through the interfacial momentum transfer, such effects
are brought down to the gas phase yielding similar interfacial gas velocity profiles. In general,
the interfacial gas velocity profiles for all flow conditions were found to be in good
agreement with measurements, especially at the channel core. In essence, the prediction of the
local bubble sizes evaluated by the coalescence and breakage kernels of the MUSIG model
has a coupling effect on the phases velocities. In better determining the Sauter mean bubble
diameter, a more accurate description of the interfacial forces between the two phases should
surface in more enhanced liquid/gas velocity predictions.
Sherman C.P. Cheung, G.H. Yeoh and J.Y. Tu 148
5.1.2. Local Distributions of Sauter Mean Bubble Diameter, Interfacial Area
Concentration and the Evolution of Bubble Size Distribution
Figure 10 illustrates the predicted and measured Sauter mean bubble diameter radial
profiles at the location of z/D = 53.5. As measured by Hibiki et al. (2001), the Sauter mean
bubble diameter profiles remained roughly unchanged throughout the whole channel. Overall,
bubble size changes were found mainly due to the bubble expansion caused by the static
pressure variation along the axial direction. As demonstrated by the good agreement between
the predicted and the measured bubble diameters at the channel core, the ensemble bubble
expansion effect was adequately captured by the twofluid approach via the ideal gas
assumption. Near the wall region, slightly larger bubbles were formed due to the tendency of
small bubbles migrating towards the wall thereby creating higher concentration of bubbles,
increasing the likelihood of possible bubble coalescence. Remarkable agreement with the
measurement at the wall region clearly illustrated the enlargement of bubble size due to
bubble coalescence and bubble breakage successfully represented by the kernels in the
MUSIG model. As the Sauter mean bubble diameter is closely related to the interfacial
momentum forces (i.e. drag and lift forces), appropriate evaluation of the bubble diameter is
thus crucial for the prediction of interfacial gas velocities.
Figure 10. Predicted Sauter mean bubble diameter distribution and experimental data of Hibiki et al.
(2001)
at measuring station of z/D =53.5.
A Review of Population Balance Modelling for Multiphase Flows 149
Figure 11. Predicted size fraction of each bubble classes and its evolution along radial direction at the
measuring station of z/D =53.5.
The population balance of bubbles within the bubbly turbulent pipe flow can be further
exemplified by tracking the evolution of bubble size distribution at the measuring station.
Figure 11 shows the development of the size fraction of each bubble classes along the radial
direction at the location of z/D = 53.5. Since the turbulence intensity is relatively low at the
channel core (i.e. r/R = 0.05), bubble sizes remained unaffected owning to the insignificant
bubble coalescence and bubble breakage rates. With increasing number of bubbles driven by
the lift force and the rising turbulence intensity within the boundary layer towards the channel
walls (i.e. r/R=0.8 and 0.95), bubble coalescence and bubble breakage became increasingly
noticeable forming larger bubbles and redistributed the BSD to higher bubble classes. Such
behaviour was amplified especially for cases with relatively higher void fraction (e.g. Figure
8a,c).
Based on the assumption where the bubbles are spherical in shape, the local
Interfacial Area Concentration (IAC) profiles may be determined based on the relation
between the local void fraction and Sauter mean bubble diameter according to
s g if
D a / 6α = . The measured and predicted local interfacial area concentration profiles
for the respective flow condition are shown in Figure 12. Similar to the void fraction
profiles, the IAC predictions at the measuring station yielded good agreement against the
measured profiles. The discrepancy between predicted results and measured data could be
attributed to the error introduced by the void fraction prediction owing to the
uncertainties embedded within the turbulence and wall lubrication models. Another
possible cause could be the invalid assumption of spherical bubbles to aptly resolve the
Sherman C.P. Cheung, G.H. Yeoh and J.Y. Tu 150
gasliquid flow where large distorted bubbles prominently featured at high superficial
velocities (i.e. <j
f
>=0.986m/s).
Figure 12. Predicted Interfacial Area Concentration (IAC) distribution and experimental data of Hibiki
et al.
(2001) at measuring station of z/D =53.5.
5.2. Subcooled Boiling Flow Results
Increasing complexity of numerical simulations is now further elaborated by considering
the bubble dynamics in conjunction with the heat and mass transfer processes typified by a
subcooled boiling flow. For the local cases of L1, L2 and L3, the measured and predicted
radial profiles of the Sauter mean bubble diameter, vapour void fraction and interfacial area
concentration located at the measuring plane 1.61 m downstream of the beginning of the
heated section are discussed below. In all the figures, the dimensionless parameter (rR
i
)/(R
o

R
i
) = 1 indicates the inner surface of the unheated flow channel wall while (rR
i
)/(R
o
R
i
) = 0
indicates the surface of the heating rod in the channel.
A Review of Population Balance Modelling for Multiphase Flows 151
Figure 13. Prediction Sauter mean bubble diameter distribution and experimental data at the measuring
station.
5.2.1. Local Distributions of Void Fraction, Sauter Bubble Diameter and Interfacial
Area Concentration
Figure 13 illustrates the predicted Sauter mean bubble diameter profiles at the measuring
plane of the heated annular channel. Experimental data and observations (Bonjour and
Lallemand et al., 2001; Lee et al., 2002) suggested that vapour bubbles, relatively small when
detached from the heated surface, have the tendency of significantly colliding with other
detached bubbles at the downstream and subsequently forming bigger bubbles via
coalescence. For all three cases, the bubble size changes were found to be adequately
predicted by the modified MUSIG model. Observed consistent trends between the predicted
and measured Sauter mean bubble diameter reflected the measure of the modified MUSIG
model in aptly capturing the bubble coalescence especially in the vicinity of the heated wall.
The development of bubbles in this region stemmed from the evaporation process occurring
at the heated wall and forces acting on the vapour bubbles determining the bubble size at
departure or liftoff.
Sherman C.P. Cheung, G.H. Yeoh and J.Y. Tu 152
Table 4. Predicted heat partitions, bubble departure and liftoff diameter of subcooled
boiling flow conditions
Case L1 Case L2 Case L3
Q
c
(W/m
2
) 0% 0% 0%
Q
tc
(W/m
2
) 2.51% 4.56% 6.42%
Q
tcsl
(W/m
2
) 55.07 61.25% 65.58%
Q
e
(W/m
2
) 42.42% 34.19% 28.00%
D
d
(mm) 0.56 0.58 0.57
M
e
a
s
u
r
i
n
g
l
o
c
a
t
i
o
n
D
l
(mm) 1.45 1.31 1.20
Table 4 illustrates the various contributing heat flux components and the associated
bubble departure and liftoff diameters evaluated by the improved heat partition model. On
the basis of the force balance model, the bubble departure diameter were predicted with a size
of approximately 0.56~0.58 mm while the liftoff diameters were found to range from
1.2~1.45 mm. The ratio between the bubble liftoff diameter and bubble departure diameter
was thus ascertained to be between 2 and 3, which incidentally closely corresponded to
experimental observations of Basu et al. (2005). Surface quenching due to sliding bubbles and
evaporation were found to be the dominant modes of heat transfer governing the heat partition
model. The former highlighted the prevalence of bubble sliding motions on the surface
significantly altering the rate of heat transfer and subsequently the resultant vapour generation
rate. Away from the heated wall, bubbles entering the bulk subcooled liquid were condensed
due to the subcooling effect. Predicted trends of the Sauter mean diameter profiles clearly
showed the gradual collapse of the bubbles from the channel centre to the outer unheated
wall. The overprediction of the bubble diameters for cases L1 and L2 could be attributed by
the underestimation of the subcooling condensation, which was determined by an empirical
correlation based on the Nusselt number description.
Although the measured bubble sizes near the heated wall were found to agree rather
well with the measured data confirming to certain extend the appropriate estimation of the
bubble liftoff diameters, a closer examination of the local void fraction profiles at the
measuring station in Figure 14 indicated a less than satisfactory prediction of the void
fraction near the heated surface where they were either over or underpredicted as
exemplified in cases L1 and L3. The void fraction distribution in case L2 compared
nonetheless reasonably well with measurement. This discrepancy could be attributed to the
uncertainties within the heat partition model in specifically evaluating the vapour
generation rate. In the quest of reducing the application of empirical correlations, the
consideration of the active nucleation site density in the present study still depended on the
use of an appropriate relationship, which could be sensitive to the flow conditions. The
significance of active wall nucleation site density linking to the prediction of the IAC has
also been reported in Hibiki and Ishii
43
. Nevertheless, as the population of cavities may
vary significantly between materials and cannot be measured directly, an adequate
expression of the active nucleation site density covering a wide range of flow conditions
remains outstanding and more concerted research is required. The IAC profiles as shown in
A Review of Population Balance Modelling for Multiphase Flows 153
Figure 15 also exhibited similar trends with the void fraction distributions plotted in Figure
14.
Figure 14. Predicted radial void fraction distribution and experimental data at the measuring station.
5.2.2. Evolution of Bubble Size Distribution and Bubble Generation Rate Due to
Coalescence, Breakage and Condensation
Figure 16 shows the bubble size distribution expressed in terms of interfacial area
concentration of individual bubbles classes along radial direction for the case L3 at the
measuring station describes the bubble dynamics caused by coalescence, breakage and
condensation in subcooled boiling flows. Significant vapour bubbles represented from bubble
class 3 in the vicinity of the heated wall essentially indicated the size of the bubble liftoff
diameter which coalesced with downstream/neighbouring bubbles forming larger void
fraction peaks as indicated by bubble classes of 7 and 9. Owing to the high shear stress within
the boundary layer, some bubbles are affected by turbulent impact due to breakage resulting
in the formation of smaller bubbles as evidenced by the significant distributions indicated
Sherman C.P. Cheung, G.H. Yeoh and J.Y. Tu 154
within bubble classes 1 and 2. Away from the heated wall, the condensation process
dominated in reducing the void fraction of each bubble classes and eventually collapsing
majority of the bubbles beyond the position (rR
i
)/(R
o
R
i
)=0.6. The net generation rate due to
coalescence and breakage and condensation rate of selected bubble classes are depicted in
Figure 17. Close to the wall region, the highest generation rate corresponding to the peak
value observed in Figure 16 is represented by bubble class 7; substantial generation rate was
also found for bubble class 3 at the same region. While the coalescence of bubbles was seen
to be governed mainly by bubble classes 3 and 7, bubble classes 3 and 12 also contributed to
the condensation process due to their considerably high number density and interfacial area.
These two figures aptly demonstrated the mechanisms of coalescence, breakage and
condensation in the modified MUSIG model affecting the thermomechanical and
hydrodynamics processes within the subcooled boiling flow.
Figure 15. Predicted interfacial gas velocity distribution and experimental data at the measuring station.
A Review of Population Balance Modelling for Multiphase Flows 155
Figure 16. Predicted IAC of each bubbles class along radial direction for the case L3 at the measuring
station.
Figure 17. Predicted net bubble generation rate due to coalescence and breakage and condensation rate
of selected bubble classes of modified MUSIG model.
Sherman C.P. Cheung, G.H. Yeoh and J.Y. Tu 156
5.3. Limitations and Shortcomings of Existing Models
Encouraging predictions by the homogeneous MUSIG and modified MUSIG models
clearly demonstrated their viable applications in resolving isothermal bubbly turbulent pipe
flow and subcooled boiling conditions. Nevertheless, the flow cases that have been
investigated from above generally possessed only weak bubblebubble interactions and
narrow bubble size distributions. Specifically, the limitation involved solving a single
velocity field for all bubble classes. To circumvent the problem, the inhomogeneous MUSIG
model (Krepper et al., 2005), which divides the gaseous dispersed phase into N number of
velocity groups, was further assessed to evaluate its feasibility in handling bubblytoslug
transition flow regime.
Transitional flow condition with liquid superficial velocity <j
f
>=0.986m/s and gas
superficial velocity <j
g
>=0.242m/s as measured by Hibiki et al. (2001) was investigated for
the application of both the homogeneous and inhomogeneous MUSIG models. Similar to the
bubbly turbulent pipe flow simulations presented in previous section, the same boundary
conditions were specified for both models where details can be referred in Table 1. Bubble
size in the range of 010 mm discretized into 10 subsize groups as tabulated in Table 2 was
similarly considered. For the inhomogeneous model, two velocity fields were solved
representing the travelling speed of small and big bubbles. The first five bubble classes (range
of 0 – 5 mm) were assigned to the first velocity field while the remaining bubble classes
(range of 5 – 10 mm) were assigned to the second velocity field. Sensitivity studies on the
increasing resolution greater than two velocity fields were also performed. With regards to the
mean parameters investigated, negligible differences were nonetheless found.
The measured and predicted local radical void fraction, Sauter mean bubble diameter,
IAC and gas velocity distribution at the measuring station of z/D=53.5 are illustrated in
Figure 18. Comparing the predicted Sauter mean diameters, the inhomogeneous MUSIG
model was found to yield comparatively better prediction when compared against the
measured data. This could be attributed to the merit of splitting the bubble velocity with two
independent fields which facilitated the model to recapture the separation of small and big
bubbles caused by different lift force actuation. Nevertheless, notable discrepancies were
found when comparing against other variables (i.e. void fraction, gas velocity and IAC)
against the experimental measurements. As depicted in Figure 18b, void fractions of both
models were obviously overpredicted at the channel core but underpredicted at the wall
region, which resulted in unsatisfactory IAC predictions (see Figure 18c). Interestingly
enough, the consideration of multiple velocity fields in the inhomogeneous MUSIG model
did not contribute to the desired expected improvements when comparing the gas velocity
predictions against those of the homogeneous MUSIG model (see Figure 18d). This could be
possibly due to the interfacial force models which have been developed principally for
isolated bubbles rather than on a swarm or cluster of bubbles. Direct applications of these
models for high void fraction conditions, where bubbles are closely packed, become
questionable and introduce uncertainties in the model calculations. Such findings have also
been reported lately in the experimental work by Simonnet et al. (2007). Based on their
measurements, they concluded that the aspiration of bubbles in the wake of the leading ones
became dominant if the void fraction exceeded the critical value 15%. This caused a sharp
increase of the relative velocity of bubbles and significantly altered the associated drag
coefficient. Interfacial forces models based on empirical correlations of isolated spherical
A Review of Population Balance Modelling for Multiphase Flows 157
bubbles have also been found to be not appropriate (Jakobsen, 2001; Behzadi et al., 2004) for
high void fraction conditions, As high void fraction (i.e. 20%) condition has been simulated
here, the ambiguity of the interfacial drag forces could plausibly be the main source of error.
Furthermore, the wall lubrication force could be underpredicted by the Antal’s model (1991)
which could represent another source of error in calculating lateral interfacial forces (Lucas et
al., 2007).
Figure 18. Predicted local radical void fraction, Sauter mean bubble diameter, IAC and gas velocity
distribution of transition bubblytoslug flow condition by homogenous and inhomogeneous MUSIG
model.
On the other hand, coalescence due to wake entrainment and breakage of large bubbles
caused by surface instability may prevail beyond the critical void fraction limit. The existing
kernels which only featured coalescence due to random collision and breakage due to
turbulent impact for spherical bubbles have to be extended to account for additional bubble
mechanistic behaviours for cap/slug bubbles. Several papers in the literature have attempted
to deal with this problem by the development of a twogroup interfacial area transport
equation (Fu et al., 2002a,b; Sun et al., 2004a,b). Encouraging results attained thus far not
Sherman C.P. Cheung, G.H. Yeoh and J.Y. Tu 158
only suggested the feasibility of the proposed approach but more importantly the prevalence
of large bubble mechanisms that are substantially different from spherical bubble interactions
in turbulent gasliquid flows.
6. Conclusions and Further Developments
In previous section, a complete threedimensional twofluid model coupled with the class
method of population balance approach was presented to handle the complex hydrodynamics
and thermomechanical processes of various bubbly turbulent pipe flow conditions.
Numerical study of isothermal bubbly turbulent pipe flows in a vertical pipe was first studied
in order to confine the complexity of solely modelling the bubble coalescence and bubble
breakage mechanisms. The homogenous MUSIG model which assumed all bubbles travelling
with the same velocity was applied. Comparison of the predicted results was made against the
measurements of Hibiki et al. (2001). Overall, the homogeneous MUSIG model yielded good
agreement for the local radial distributions of void fraction, interfacial area concentration,
Sauter mean bubble diameter and gas and liquid velocities against measurements. Numerical
results also clearly showed that the range of bubbles sizes existed in the gasliquid flows
required substantial resolution. Numerical results obtained through this study clearly
demonstrated the competence of the MUSIG model and the robustness of the bubble
coalescence and bubble breakage kernels in accommodating the interactions of finely
dispersed bubbles within isothermal bubbly turbulent pipe flows.
The potential of the population balance approach was further exploited in modelling
subcooled boiling flows. Such flows by nature are inherently complex since they
simultaneously embrace all complex flow hydrodynamics, bubble coalescence and bubble
breakage accompanied by heat and mass transfer processes in the bulk flow, and various
surface heat flux characterisations. An improved heat partition model which mechanistically
determined the bubble departure and liftoff diameter was also presented as a closure for the
problem in order to determine the appropriate evaporation rate of the nucleation process.
Numerical results validated against the experiment data of Yun et al. (1997) and Lee et al.
(2002) for lowpressure subcooled boiling annular channel flows showed good agreement for
the local Sauter mean bubble diameter, void fraction, IAC profiles. Detailed vapour size
distribution and its corresponding generation rates were probed via the size fractions of the
modified MUSIG model in order to better envisage the condensation effect in conjunction
with the occurrence of bubble coalescence and bubble breakage within the gasliquid flows.
In addition to the homogeneous MUSIG model, inhomogeneous MUSIG model was also
applied to investigate the feasibility of application in handling transition bubblytoslug flow.
It was observed that the inhomogeneous MUSIG gave better prediction of the Sauter mean
bubble diameter distributions when compared to the homogeneous model. The more
complicated inhomogeneous MUSIG model provided the premise of better capturing the
different effects of the lift force acting on the small and large bubbles. Less encouraging
results were however ascertained between the predicted and measured void fraction,
interfacial gas velocity and IAC profiles.
From the above numerical study, as an example of population balance modelling, some
important numerical issues have been demonstrated. Firstly, it exposed the limitation of the
existing interfacial forces models. Most of the existing interfacial forces models were
A Review of Population Balance Modelling for Multiphase Flows 159
developed and calibrated from isolated single particle which may not be strictly applicable if
particles are closely packed. This is due to the fact that particle motion may be influenced by
neighbouring particles resulting different momentum transfer. According to the recent
experimental study by Simonnet et al. (2007), significant decrement of gas bubble drag
coefficient was found if the gas void fraction exceeded 15%. In fact, very recently, some
research studies have been performed attempting to improve the existing interfacial drag and
lift force models (Hibiki and Ishii, 2007; Marbrouk et al., 2007; Liu et al, 2008)
Secondly, it unveiled the constraint of the current coalescence and breakage kernels. For
bubbly simulations presented in this review, coalescence and breakage kernels were derived
based on the spherical bubble assumption. This assumption limits the kernels to be only
applicable to bubbly flow regime where bubbles are in spherical shape. This also explains
why less encouraging results were obtained when bubblytoslug flow was considered. In
bubblytoslug flows, subject to the balance of the surface tension and surrounding fluid
motion, large bubbles may be deformed into cap or taylor bubbles. Thus, coalescence due to
wake entrainment may become significant which unfortunately was not modelled in current
adopted kernels. Unfortunately, fundamental knowledge of bubble coalescence and breakage
mechanisms still remain elusive forming a bottleneck for the development of more robust
population balance kernels. In essence, not only for gasliquid systems, similar problems and
challenges are also prevalent in other PBE applications; for example: soot formation
prediction in combustion system (Frenklach and Wang, 1994). It is certainly that substantial
research works would be centred in kernels development in near future.
Finally, although encouraging results had been obtained from the MUSIG model based
on CM, one should be also reminded that QMOM or other moment methods which represent
a rather sound mathematical approach and an elegant tool of solving the PBE with limited
computational burden could be alternatively considered for modelling practical multiphase
flow systems for future investigative studies.
Notation
a
if
= interfacial area concentration
A
c
= cross sectional area of the boiling channel
A
q
= fraction of heater area occupied by bubbles
C
1
, C
2
= constants defined in equation (32)
C
D
= drag coefficient
C
L
= shear lift coefficient
C
p
= specific heat
C
s
= constant defined in equation (41)
C
v
= Acceleration coefficient
d = vapor bubble diameter at heated surface
d
w
= surface/bubble contact diameter
D = average bubble diameter
D
c
,D
B
= bubble death rate due to coalescence and breakage
D
b
= departing bubble diameter
D
d
= bubble departure diameter
D
l
= bubble liftoff diameter
Sherman C.P. Cheung, G.H. Yeoh and J.Y. Tu 160
D
s
= Sauter mean bubble diameter
f = bubble generation frequency
f
i
=
scalar size fraction of the ith discrete bubble classes
g i i
f α α / =
F = degree of surface cavity flooding
F
c
,F
B
= calibration factors for coalescence and breakage
F
b
= buoyancy force
F
cp
= contact pressure force
F
du
= unsteady drag force due to asymmetrical growth of the bubble
F
h
= force due to the hydrodynamic pressure
F
qs
= Quasisteadydrag force
F
s
= surface tension force
F
sL
= shear lift force
F
lg
= action of interfacial forces from vapor on liquid
F
gl
= action of interfacial forces from liquid on vapor
drag
F
lg
=
drag force
lift
F
lg
=
lift force
rication
F
lub
lg
=
wall lubrication force
dispersion
F
lg
=
turbulent dispersion force
g = gravitational constant
g
= gravitational vector
G = mass flux
G
s
= dimensionless shear rate
h = interfacial heat transfer coefficient
h
0
, h
f
= initial and critical film thickness
h
fg
= latent heat
H = enthalpy
Ja = Jakob number
k = thermal or turbulent kinetic energy
K = projected area of bubble
h
fg
= latent heat of vaporization
l
s
= sliding distance
n
i
= number density of the discrete bubble i
th
class
N
a
= active nucleation site density
P = pressure
P
k
= turbulent kinetic energy production term
P
c
,P
B
= bubble production rate due to coalescence and breakage
Q
w
= wall heat flux
Q
c
= heat transfer due to forced convection
Q
e
= heat transfer due to evaporation
Q
tc
= heat transfer (transient conduction) due to stationary bubble
Q
tcsl
= heat transfer (transient conduction) due to sliding bubble
r = bubble radius at heated wall or flow spacing within annular
A Review of Population Balance Modelling for Multiphase Flows 161
channel
r
c
= cavity radius at heated surface
r
r
= curvature radius of the bubble at heated surface
R = source/sink term due to coalescence and breakage
Re = bubble Reynolds number
R
f
= ratio of the actual number of bubbles lifting off to the number of
active nucleation sites
R
ph
= source/sink term due to phase change
R
i
= radius of inner heated wall
R
o
= radius of outer unheated wall
s = spacing between nucleation sites
S
i
= additional source terms due to coalescence and breakage
S
ij
= tensor of shear stress
St = Stanton number
t = time
t
g
= bubble growth period
t
sl
= bubble sliding period
t
w
= bubble waiting period
T = temperature
ΔT = difference in temperature
u = velocity
u
= velocity vector
u
τ
= friction velocity
v
i
= specific volume of discrete bubble i
th
class
x = cartesian coordinate along x
x
+
= nondimensional normal distance from heated wall
y = cartesian coordinate along y
Greek Letters
α = advancing angle
α
g
= vapor void fraction
α
l
= liquid void fraction
β = receding angle
χ
= coalescence rate
δ
l
= Thermal boundary layer thickness
ε = turbulent dissipation rate
WN
φ
= bubble nucleation rate
COND
φ
= bubble condensation rate
η = thermal diffusivity or coalescence volume matrix
λ = size of an eddy
e
λ
=
effective viscosity
μ = viscosity
θ = bubble contact angle
Sherman C.P. Cheung, G.H. Yeoh and J.Y. Tu 162
θ
i
= inclination angle
ρ = density
σ = surface tension/Prandtl number
ij
τ
= bubble contact time
ω = turbulent frequency
Ω = breakage rate
ξ = size ratio between an eddy and a particle in the inertial subrange
H
ξ
= heated perimeter of boiling channel
Γ
lg
= interfacial mass transfer from vapor to liquid
Γ
gl
= interfacial mass transfer from liquid to vapor
Subsripts
axial = axial distribution
g = vapor
inlet = channel entrance
l = liquid
local = local distribution
s = surface heater
t = turbulent
sat = saturation
sub = subcooled
sup = superheated
w = heated surface wall
Acknowledgment
The financial support provided by the Australian Research Council (ARC project ID
DP0877734) is gratefully acknowledged.
References
[1] Andersson R, Andersson B (2006). Modeling the breakup of fluid particles in turbulent
flows. AIChE Journal 52, 20312038.
[2] Anglart H, Nylund O (1996). CFD application to prediction of void distribution in two
phase bubbly flows in rod bundles. Nuclear Science and Engineering 163, 8198.
[3] ANSYS (2006). CFX11 User Manual. ANSYSCFX.
[4] Antal SP, Lahey Jr RT, Flaherty JE 1991). Analysis of phase distribution in fully
developed laminar bubbly twophase flow. International Journal of Multiphase Flow
17, 635652.
[5] Barrett JC, Webb NA (1998). A comparison of some approximate methods for solving
the aerosol general dynamics equation. Journal of Aerosol Science 29, 3139.
A Review of Population Balance Modelling for Multiphase Flows 163
[6] Basu N, Warrier GR, Dhir VK (2005). Wall heat flux partitioning during subcooled
flow boiling: Part I – model development. ASME Journal of Heat Transfer 127, 131
140.
[7] Behzadi A, Issa R, Rusche H (2004). Modelling of dispersed bubble and droplet flow at
high phase fractions. Chemical Engineering Science 59, 759770.
[8] Bertola F, Grundseth J, Hagesaether L, Dorao C, Luo H, Hjarbo KW, Svendsen HF,
Vanni MB, Jakobsen HA (2003). Numerical analysis and experimental validation of
bubble size distributions in two phase bubble column reactors, in: Proceedings of the
3
rd
EuropeanJapanese Two Phase Flow Group Meeting, Certosa di Pontignano.
[9] Biswas S, Esmaeeli A, Tryggvason G (2005). Comparison of results from DNS of
bubbly flows with a twofluid model for two dimensional laminar flows. International
Journal of Multiphase Flow 31, 10361048.
[10] Bonjour J, Lallemand M (2001). Twophase flow structure near a heated vertical wall
during nucleate pool boiling. International Journal of Multiphase Flow 28, 119.
[11] Bordel S, Mato R, Villaverade S (2006). Modeling of the evolution with length of
bubble size distributions in bubble columns. Chemical Engineering Science 61, 2663
2673.
[12] Bothe D, Schmidtke M, Warnecke HJ (2006). VOFSimulation of the life force for
single bubbles in simple shear flow. Chemical Engineering Technology 29, 10481053.
[13] Bove S 2005. Computational fluid dynamics of gasliquid flows including bubble
population balance, Ph.D. Thesis, Aalborg University Esbjerg, Denmark.
[14] Bove S, Solberg T, Hjertager BH (2005). A novel algorithm for solving population
balance equations: the parallel parent and daughter classes. Derivation, analysis and
testing. Chemical Engineering Science 60, 14491464.
[15] Bukur DB, Daly JG, Patel SA (1996). Application of γray attenuation for
measurement of gas holdups and flow regime transitions in bubble columns. Industrial
and Engineering Chemical Research 35, 7080.
[16] Chen P, Sanyal J, Duduković MP (2005). Numerical simulation of bubble columns
flows: effect of different breakup and coalescence closures. Chemical Engineering
Science 60, 10851101.
[17] Chesters AK, Hoffman G (1982). Bubble coalescence in pure liquids, Applied
Scientific Research 38, 353361.
[18] Cheung SCP, Yeoh GH, Tu JY (2007). On the numerical study of isothermal vertical
bubbly flow using two population balance approaches. Chemical Engineering Science
62, 46594674.
[19] Davidson, MR (1990). Numerical calculations of twophase flow in a liquid bath with
bottom gas injection: the central plume. Applied Mathematical Modelling 14, 6776.
[20] Debry E, Sportisse B, Jourdain B (2003). A stochastic approach for the numerical
simulation of the general equation for aerosols. Journal of Computational Physics 184,
649669.
[21] Domilovskii ER, Lushnikov AA, Piskunov VN (1979). A monte carlo simulation of
coagulation processes. Izvestigya Akademii Nauk SSSR, Fizika Atmosfery I Okeana 15,
194201.
[22] Dorao CA, Jakobsen HA (2006). Numerical calculation of the moments of the
population balance equation. Journal of computational and applied mathematics 196,
619633.
Sherman C.P. Cheung, G.H. Yeoh and J.Y. Tu 164
[23] Dorao CA, Lucas D, Jakobsen HA (2008). Prediction of the evolution of the dispersed
phase in bubbly flow problems. Applied mathematical modelling 32, 18131833.
[24] Ervin EA, Tryggvason G (1997). The rise of bubbles in a vertical shear flow, Journal
of Fluids Engineering 119, 443449.
[25] Fan R, Marchisio DL, Fox RO (2004). Application of the direct quadrature method to
polydispersed gassolid fluidized beds, Power Technology 139, 720.
[26] Frank T, Shi J, Burns AD (2004). Validation of Eulerian multiphase flow models for
nuclear safety application, in: Proceeding of the Third International Symposium on
TwoPhase Modelling and Experimentation, Pisa, Italy.
[27] Fredrickson AG, Ramkrishna D, Tsuchiya HM (1967). Statistics and dynamics of
procaryotic cell population. Mathematical Bioscience 1, 327374.
[28] Frenklach M, Harris SJ (1987). Aerosol dynamics modelling using the method of
moments. Journal of Colloid and Interface Science 118, 252261.
[29] Frenklach M, Wang H (1991) Detailed modeling of soot particle nucleation and
growth, in: Proceedings of the 23
rd
Symposium on Combustion, Combustion Institute,
University of Orleans, France.
[30] Frenklach M, Wang H (1994) Detailed mechanism and modeling of soot formation, in:
Bockhorn H (Ed.), Soot formation in combustion: Mechanisms and models, Berlin:
Springer.
[31] Frenklach M (2002). Method of Moments with Interpolative closure. Chemical
Engineering Science 57, 22292239.
[32] Friedlander SK (2000). Smoke, dust and haze: Fundamental of aerosol dynamics (2
nd
Edition). Oxford: Oxford University Press.
[33] Fu XY, Ishii M (2002a). Twogroup interfacial area transport in vertical airwater flow
I. Mechanistic model. Nuclear Engineering and Design 219, 143168.
[34] Fu XY, Ishii M (2002b). Twogroup interfacial area transport in vertical airwater flow
II. Model evaluation. Nuclear Engineering and Design 219, 169190.
[35] George DL, Shollenberger KA, Torczynski, JR (2000). Sparger effect on gas volume
fraction distributions in vertical bubblecolumn flows as measured by gamma
densitometry tomography, in: ASME 2000 Fluid Engineering Division Summer
Meeting, Boston, USA.
[36] Gopinath R, Basu N, Dhir VK (2002). Interfacial heat transfer during subcooled flow
boiling. International Journal of Heat and Mass Transfer 45, 39473959.
[37] Gordon RG (1968). Error bounds in equilibrium statistical mechanics. Journal of
Mathematical Physics 9, 655672.
[38] Grosch R, Briesen H, Marquardt W, Wulkow M (2007). Generalization and numerical
investigation of QMOM. AIChE Journal 53, 207227.
[39] Hagesaether L, Jakobsen H, Svendsen HF (2002). A model for turbulent binary
breakup of dispersed fluid particles. Chemical Engineering Science 57, 32513267.
[40] Hazuku T, Takamasa T, Hibiki T, Ishii M (2007). Interfacial area concentration in
annular twophase flow. International Journal of Heat and Mass Transfer 50, 2986
2995.
[41] Hibiki T, Ishii M, Xiao Z (2001). Axial interfacial area transport of vertical bubble
flows. International Journal of Heat and Mass Transfer 44, 18691888.
A Review of Population Balance Modelling for Multiphase Flows 165
[42] Hibiki T, Ishii M (2002). Development of onegroup interfacial area transport equation
in bubbly flow systems. International Journal of Heat and Mass Transfer 45, 2351
2372
[43] Hibiki T, Ishii M (2003). Active nucleation site density in boiling systems.
International Journal of Heat and Mass Transfer 46, 25872601.
[44] Hibiki T, Ishii M (2008). Lift force in bubbly flow systems. Chemical Engineering
Science 62, 64576474.
[45] Hibiki T, Situ R, Mi Y, Ishii M (2003). Experimental study on interfacial area transport
in vertical upward bubbly twophase flow in an annulus. International Journal of Heat
and Mass Transfer 46, 427441.
[46] Hibiki T, Goda H, Kim SJ, Ishii M, Uhle J (2005). Axial development of interfacial
structure of vertical downward bubbly flow. International Journal of Heat and Mass
Transfer 48, 749764.
[47] Hidy GM, Brock JR (1970). The dynamics of aerocolloidal systems, Oxford:
Pergamon.
[48] Hsu YY, Graham RW (1976). Transport process in boiling and twophase systems,
Hemisphere, Washington.
[49] Hulburt HM, Katz S (1964). Some problems in particle technology: A statistical
mechanical formulation. Chemical Engineering Science 19, 55574.
[50] Jakobsen H (2001). Phase distribution phenomena in twophase bubble column
reactors. Chemical Engineering Science 56, 10491056.
[51] Jia X, Wen J, Zhou H, Fend W, Yuan Q (2007). Local hydrodynamics modelling of a
gasliquidsolid three phase bubble column. AIChE Journal 53, 22212231.
[52] Judd RL, Hwang KS (1976). A comprehensive model for nucleate pool boiling heat
transfer including microlayer evaporation. ASME Journal of Heat Transfer 98, 623629.
[53] Kashinsky ON, Randin VV (1999). Downward bubbly gasliquid flow in a vertical
pipe. International Journal of Multiphase Flow 25, 109138.
[54] Khan AA, De Jong W, Spliethoff H (2007). A fluidized bed combustion model with
discretized population balance. 2. Experimental studies and model. Energy and Fuels
21, 37093717.
[55] Klausner JF, Mei R, Bernhard DM, Zeng LZ (1993). Vapor bubble departure in forced
convection boiling. International Journal of Heat Mass Transfer 36, 651662.
[56] Kocamustafaogullari G, Ishii M (1995). Foundation of the interfacial area transport
equation and its closure relations. International Journal of Heat and Mass Transfer 38,
481493
[57] Končar B, Kljenak I, Mavko B (2004). Modeling of local twophase parameters in
upward subcooled flow boiling at low pressure. International Journal of Heat and Mass
Transfer 47, 14991513.
[58] Krepper E, Lucas D, Prasser H (2005). On the modelling of bubbly flow in vertical
pipes. Nuclear Engineering and Design 235, 597611.
[59] Krepper E, Frank T, Lucas D, Prasser H, Zwart PJ (2007). Inhomogeneous MUSIG
model – a Population Balance approach for polydispersed bubbly flows, in: Proceeding
of the 6
th
International Conference on Multiphase Flow, Leipzig, Germany.
[60] Lakatos BG, Sule Z, Mihalyko C (2008). Population balance model of heat transfer in
gassolid particle systems. International Journal of Heat and Mass Transfer 51, 1633
1645.
Sherman C.P. Cheung, G.H. Yeoh and J.Y. Tu 166
[61] Lambin P, Gaspard JP (1982). Continuedfraction technique for tightbinding systems.
A generalizedmoments method, Physical Review B 26, 43564368.
[62] Lee TH, Park GC, Lee DJ (2002). Local flow characteristics of subcooled boiling flow
of water in a vertical annulus. International Journal of Multiphase Flow 28, 13511368.
[63] Lehr F, Mewes D (2001). A transport equation for the interfacial area density applied
to bubble column. Chemical Engineering Science 56, 1159116.
[64] Liffman K (1992). A direct simulation monte carlo method for cluster coagulation.
Journal of Computational Physics 100, 116127.
[65] Liu TJ, Bankoff SG (1993). Structure of airwater bubbly flow in a vertical pipe – I.
Liquid mean velocity and turbulence measurements. International Journal of Heat and
Mass Transfer 36, 10491060.
[66] Liu Y, Hibiki T, Sun X, Ishii M, Kelly JM (2008). Drag coefficient in onedimensional
twogroup twofluid model. International Journal of Heat and Fluid Flow (In press).
[67] Lo S (1996). Application of the MUSIG model to bubbly flows. AEAT1096, AEA
Technology.
[68] Lu J, Biswas S, Tryggvason G (2006). A DNS study of laminar bubbly flows in a
vertical channel. International Journal of Multiphase Flow 32, 643660.
[69] Lucas D, Krepper E, Prasser HM (2007). Use of models for lift, wall and trubulent
dispersion force acting on bubbles for polydisperse flows. Chemical Engineering
Science 62, 41464157.
[70] Luo H, Svendsen H (1996). Theoretical model for drop and bubble breakup in
turbulent dispersions. AIChE Journal 42, 12251233.
[71] Mabrouk R, Chaouki J, Guy C (2007). Effective drag coefficient investigation in the
acceleration zone of upward gassolid flow. Chemical Engineering Science 62, 318
327.
[72] Maisels A, Kruis FE, Fissan H (2004). Direct simulation monte carlo for simultaneous
nucleation, coagulation, and surface growth in dispersed systems. Chemical
Engineering Science 59, 22312239.
[73] Marchisio DL, Fox RO (2005). Solution of population balance equations using the
direct quadrature method of moments. Journal of Aerosol Science 36, 4373.
[74] Marchisio DL, Vigil DR, Fox RO (2003a). Quadrature method of moments for
aggregationbreakage processes. Journal of Colloid Interface Science 258, 322324.
[75] Marchisio DL, Pikturna JT, Fox RO, Vigil RD (2003b). Quadrature method for
moments for populationbalance equations. AIChE Journal. 49, 12661276.
[76] Marchisio DL, Vigil DR, Fox RO (2003c). Implementation of the quadrature method of
moments in CFD codes for aggregationbreakage problems. Chemical Engineering
Science 58, 33373351.
[77] Matos A, de Rosa, ES, Franca FA (2004). The phase distribution of upward cocurrent
bubbly flows in a vertical square channel. Journal of Brazil Society of Mechanical
Science and Engineering 26, 308316.
[78] Markatou P, Wang H, Frenklach M (1993). A computational study of soot limits in
laminar premixed flames of ethane, ethylene, and acetylene. Combustion and Flame 93,
467482.
[79] McCoy BJ, Madras G (2003). Analytic solution for a population balance equation with
aggregation and fragmentation. Chemical Engineering Science 58, 30493051.
A Review of Population Balance Modelling for Multiphase Flows 167
[80] McGraw E, Wright DL (2003). Chemically resolved aerosol dynamics for internal
mixtures by the quadrature method of moments. Journal of Aerosol Science 34, 189
209.
[81] McGraw R (1997). Description of aerosol dynamics by the quadrature method of
moments. Aerosol Science and Technology 27, 255265.
[82] Menter FR (1994). Twoequation eddy viscosity turbulence models for engineering
applications. AIAA Journal 32, 15981605.
[83] Olmos E, Gentric C, Vial Ch, Wild G, Midoux N (2001). Numerical simulation of
multiphase flow in bubble column. Influence of bubble coalescence and breakup.
Chemical Engineering Science 56, 63596365.
[84] Pandis SN, Seinfeld JH (1998). Atmospheric chemistry and physics: From air pollution
to climate change. New York: Wiley.
[85] Pandolph AD, Larson MA (1964). A population balance for countable entities.
Canadian Journal of Chemical Engineering 42, 280281.
[86] Pochorecki R, Moniuk W, Bielski P, Zdrojkwoski A (2001). Modelling of the
coalescence/redisperison processes in bubble columns. Chemical Engineering Science
56, 61576164.
[87] Prince MJ, Blanch HW (1990). Bubble coalescence and breakup in air sparged bubble
columns. AIChE Journal 36, 14851499.
[88] Ramkrishna D, Borwanker JD (1973). A puristic analysis of population balance.
Chemical Engineering Science 28, 14231435.
[89] Ramkrishna D (1979). Statistical models of cell populations, In: Ghose TK, Fiechter A,
Blakebrough N (Eds.), Advances in Biochemical Engineering, vol. 11, Berlin: Springer.
[90] Ramkrishna D (1985). The status of population balances. Reviews in Chemical
Engineering 3, 4995.
[91] Ramkrishna D (2000). Population balances. Theory and applications to particulate
systems in engineering. New York: Academic press.
[92] Ramkrishna D, Mahoney AW (2002). Population balances modeling. Promise for the
future. Chemical Engineering Science 57, 595606.
[93] Rotta JC (1972). Turbulente Stromungen, Teubner B.G., Stuttgart.
[94] Sanyal J, Marchisio DL, Fox RO, Dhanasekharan K (2005). On the comparison
between population balance models for CFD simulation of bubble column. Industrial
and Engineering Chemical Research 44, 50635072.
[95] Sato Y, Sadatomi M, Sekoguchi K (1981). Momentum and heat transfer in twophase
bubbly flow – I. International Journal of Multiphase Flow 7, 167178.
[96] Schwarz, MP, Turner, WJ (1988). Applicability of the standard kε model to gasstirred
baths, Applied Mathematical Modelling 12, 273279.
[97] Scott WT (1968). Analytic studies of cloud droplet coalescence. Journal of the
Atmospheric Science 25, 5465.
[98] Serizawa A, Kataoka I (1988). Phase distribution in twophase flow, in: Afgan NH,
Transient phenomena in multiphase flow, Washington, DC.
[99] Serizawa A, Kataoka I (1990). Turbulence suppression in bubbly twophase flow.
Nuclear Engineering and Design 122, 116.
[100] Sha Z, Laari A, Turunen I (2006). Multiphasemultigroup model for the inclusion of
population balances in to CFD simulation of gasliquid bubble flows. Chemical
Engineering and Technology 29, 550559.
Sherman C.P. Cheung, G.H. Yeoh and J.Y. Tu 168
[101] Shi JM, Zwart PJ, Frank T, Rohde U, Prasser HM (2004). Development of a multiple
velocity multiple size group model for polydispersed multiphase flows, in: Annual
Report of Institute of Safety Research, Forschungszentrum Rossendorf, Germany.
[102] Simonnet M, Gentric C, Olmos E, Midoux N (2007). Experimental determination of
the drag coefficient in a swarm of bubbles. Chemical Engineering Science 62, 858866.
[103] Steiner H, Kobor A, Gebhard L (2005). A wall heat transfer model for subcooled
boiling flow. International Journal of Heat and Mass Transfer 48, 41614173.
[104] Sun X, Kim S, Ishii M, Beus SG (2004a). Modeling of bubble coalescence and
disintegration in confined upward two phase flow. Nuclear Engineering and Design
230, 326.
[105] Sun X, Kim S, Ishii M, Beus SG (2004b). Model evaluation of twogroup interfacial
area transport equation for confined upward flow. Nuclear Engineering and Design
230, 2747.
[106] Takamasa T, Goto T, Hibiki T, Ishii M (2003). Experimental study of interfacial area
transport of bubbly flow in a smalldiameter tube. International Journal of Multiphase
Flow 29, 395409.
[107] Tomiyama A (1998). Struggle with computational bubble dynamics, in: Proceeding of
the Third International Conference on Multiphase Flow. Lyon, France.
[108] Tu JY, Yeoh GH (2002). On numerical modelling of lowpressure subcooled boiling
flows. International Journal of Heat and Mass Transfer 45, 11971209.
[109] Wang T, Wang J, Jin Y (2003). A novel theoretical breakup kernel function for
bubbles/droplets in a turbulent flow. Chemical Engineering Science 58, 46294637.
[110] Wang T, Wang J, Jin Y (2005). Theoretical prediction of flow regime transition in
bubble columns by the population balance model. Chemical Engineering Science 60,
61996209.
[111] Wang T, Wang J, Jin Y (2006). A CFDPBM coupled Model for gasliquid flows.
AIChE Journal 52, 125140.
[112] Wu Q, Kim S, Ishii M, Beus SG (1998). Onegroup interfacial area transport in vertical
bubbly flow. International Journal of Heat and Mass Transfer 41, 11031112.
[113] Yao W, Morel C (2004). Volumetric interfacial area prediction in upwards bubbly two
phase flow. International Journal of Heat and Mass Transfer 47, 307328.
[114] Yeoh GH, Cheung SCP, Tu JY, Ho MKM (2008). Fundamental consideration of wall
heat partition of vertical subcooled boiling flows. International Journal of Heat and
Mass Transfer 51, 38403853
[115] Yeoh GH, Tu JY (2004). Population balance modelling for bubbly flows with heat and
mass transfer. Chemical Engineering Science 59, 31253139.
[116] Yeoh GH, Tu JY (2005). Thermalhydrodynamic modelling of bubbly flows with heat
and mass transfer. AIChE Journal 51, 827.
[117] Yeoh GH, Tu JY (2006). Numerical modelling of bubbly flows with and without heat
and mass transfer. Applied Mathematical Modelling 30, 10671095
[118] Yun BJ, Park GC, Song CH, Chung MK (1997). Measurements of local twophase
flow parameters in a boiling flow channel, in: Proceedings of the OECD/CSNI
Specialist Meeting on Advanced Instrumentation and Measurement Techniques.
[119] Zeitoun O, Shoukri M (1996). Bubble behavior and mean diameter in subcooled flow
boiling. ASME Journal of Heat Transfer 118, 110116.
A Review of Population Balance Modelling for Multiphase Flows 169
[120] Zeng LZ, Klausner JF, Bernhard DM, Mei R (1993). A unified model for the prediction
of bubble detachment diameters in boiling systems – II. Flow boiling. International
Journal of Heat and Mass Transfer 36, 22712279.
[121] Zuber N (1961). The dynamics of vapor bubbles in nonuniform temperature fields.
International Journal of Heat and Mass Transfer 2,8398.
[122] Zucca A, Marchisio DL, Vanni M, Barresi AA (2007). Validation of bivariate
DQMOM for nanoparticle processes simulation, AIChE Journal 53, 918931.
In: Fluid Mechanics and Pipe Flow ISBN: 9781607410379
Editors: D. Matos and C. Valerio, pp. 171204 © 2009 Nova Science Publishers, Inc.
Chapter 5
NUMERICAL ANALYSIS OF HEAT TRANSFER
AND FLUID FLOW FOR THREEDIMENSIONAL
HORIZONTAL ANNULI WITH OPEN ENDS
ChunLang Yeh
*
Department of Aeronautical Engineering, National Formosa University,
Huwei, Yunlin 632, Taiwan, R.O.C.
Abstract
Study of the heat transfer and fluid flow inside concentric or eccentric annuli can be applied in
many engineering fields, e.g. solar energy collection, fire protection, underground conduit,
heat dissipation for electrical equipment, etc. In the past few decades, these studies were
concentrated in twodimensional research and were mostly devoted to the investigation of the
effects of convective heat transfer. However, in practical situation, this problem should be
threedimensional, except for the vertical concentric annuli which could be modeled as two
dimensional (axisymmetric). In addition, the effects of heat conduction and radiation should
not be neglected unless the outer cylinder is adiabatic and the temperature of the flow field is
sufficiently low. As the author knows, none of the open literature is devoted to the
investigation of the conjugated heat transfer of convection, conduction and radiation for this
problem. The author has worked in industrial piping design area and is experienced in this
field. The author has also employed threedimensional bodyfitted coordinate system
associated with zonal grid method to analyze the natural convective heat transfer and fluid
flow inside threedimensional horizontal concentric or eccentric annuli with open ends. Owing
to its broad application in practical engineering problems, this chapter is devoted to a detailed
discussion of the simulation method for the heat transfer and fluid flow inside three
dimensional horizontal concentric or eccentric annuli with open ends. Two illustrative
problems are exhibited to demonstrate its practical applications.
Keywords：threedimensional horizontal concentric or eccentric annuli with open ends,
conjugated heat transfer of convection, conduction and radiation, bodyfitted coordinate
system, zonal grid method.
*
Email address: clyeh@nfu.edu.tw. Tel. No.：88656315527, Fax No.：88656312415.
ChunLang Yeh 172
Nomenclature
p
C specific heat capacity
h
D hydraulic diameter
Ec Eckert number , =
* * * *
2
*
/
ref h w p
k D q C
V
ref
Fr Froude number , =
* * *
/
h
D g V
g gravitational acceleration
jk
g contravariant metric tensor
h convective heat transfer coefficient
J Jacobian
k thermal conductivity
Pr Prandtl number , =
* *
/α υ
p hydraulic pressure , =
2
* *
* * * *
V
y g p
ref
ref
ρ
ρ +
q heat flux
j
q curvilinear coordinate
Φ
R source term
Ra Rayleigh number , =
* * *
3
* * *
/ α υ β T D g
h
Δ
o
Ra modified Rayleigh number , =
* * * *
4
* * *
/ k q D g
w h
α υ β
T temperature
(u,v,w) physical velocity , = (
* * *
, , w v u ) /
*
V
j
V contravariant velocity
*
V characteristic velocity, =
* *
/
h
D α
(x,y,z) Cartesian coordinates , = (
* * *
, , z y x ) /
*
h
D
α thermal diffusivity； also radiation absorptivity
β thermal expansion coefficient
Φ
Γ diffusion coefficient
ε radiation emissivity
θ nondimensionalized temperature , =
* * *
* *
/ k D q
T T
h w
ref
−
μ viscosity
υ kinematic viscosity
ρ density , =
* *
/
ref
ρ ρ
Numerical Analysis of Heat Transfer and Fluid Flow… 173
σ StefanBoltzmann constant
Φ energy dissipation term； also dependent variable
φ azimuthal angle
Subscripts
i inner cylinder
nb neighboring grid points
o outer cylinder
P main grid point
ref reference state（ ） at atmospheric pressure and room temperature
w wall
Superscripts
▬ averaged quantity
* dimensional quantity
Introduction
Heat transfer and fluid flow inside concentric or eccentric annuli can be found in
numerous engineering applications, e.g. solar energy collection, fire protection, underground
conduit, heat dissipation for electrical equipment, etc. In the past few decades, these studies
were concentrated in twodimensional research and were mostly devoted to the investigation
of the effects of convective heat transfer. However, in practical situation, this problem should
be threedimensional, except for the vertical concentric annuli which could be modeled as
twodimensional (axisymmetric); in addition, the effects of heat conduction and radiation
should not be neglected unless the outer cylinder is adiabatic and the temperature of the flow
field is very low. As the author knows, none of the open literature is devoted to the
investigation of the conjugated heat transfer of convection, conduction and radiation for this
problem. Numerous theoretical and experimental studies on natural convection in horizontal
concentric or eccentric annuli have been conducted. In most of these studies, a two
dimensional model was used in which the annuli were assumed to be infinitely long and
coupled with thermal boundary conditions on the cylinder surfaces specified as either with
two constant wall temperatures or one with constant wall temperature while the other with
constant wall heat flux (including adiabatic surface) [156]. Some representative studies are
listed below.
1. Gju et al. [2,3] conducted experiments for the natural convection in horizontal
concentric or eccentric annuli. The inner and outer cylinders were both kept at
constant temperature. The instability and transition of the heat transfer and fluid flow
were investigated. It is found that for a concentric annulus of inner/outer diameter
ratio 2.36, chaos occur near Rayleigh number between 0.9×10
5
and 3.37×10
5
.
ChunLang Yeh 174
2. Kuehn and Goldstein [79] also performed experimental investigations for the natural
convection in horizontal concentric or eccentric annuli. The inner and outer cylinders
were both kept at constant temperature. The instability and transition of heat transfer
and fluid flow were investigated. It is found that for a concentric annulus of
inner/outer diameter ratio 2.6, transition from laminar to turbulent regimes occurs
near Rayleigh number of 4×10
6
.
3. Vafai et al. [1120] studied numerically the two and threedimensional natural
convection in horizontal concentric annuli. The heat transfer and fluid flow under
different inner/outer diameter ratios and Rayleigh numbers were investigated. The
scope of discussion covers laminar and turbulent regimes as well as steady and
transient flows. In respect of the fluid flow field, the interaction of primary and
second flows was examined. With respect to the thermal field, the distribution of
Nusselt number was analyzed. In addition, the influence of geometry on the heat
transfer and fluid flow was also discussed.
4. Yoo et al. [2127] performed twodimensional simulation of natural or mixed
convection in horizontal concentric annuli. The outer cylinder was kept at constant
temperature while the inner cylinder was kept at constant temperature or heat flux.
The heat transfer and fluid flow were analyzed for different Rayleigh numbers,
Reynolds numbers, Prandtl numbers and inner/outer diameter ratios. The scope of
discussion covers the transition of laminar to turbulent regimes as well as the
phenomenon of chaos.
5. Shu et al. [2833] studied numerically the twodimensional natural convection in
horizontal concentric or eccentric annuli. The outer cylinder is a square duct kept at a
lower temperature while the inner cylinder is a circular one kept at a higher
temperature. The heat transfer and fluid flow were investigated for different Rayleigh
numbers, Reynolds numbers, Prandtl numbers, inner/outer diameter ratios and
eccentricity of inner/outer cylinders. In respect of the thermal field, the distribution
of Nusselt number was analyzed.
6. Mujumdar et al. [3436] investigated numerically the twodimensional natural
convection and phase change in horizontal concentric annuli. The outer cylinder is a
square or circular duct while the inner cylinder is a circular or square one. The heat
transfer and fluid flow were examined for different Rayleigh numbers and heating
rates on the inner or outer cylinders.
7. ElShaarawi et al.[3739] studied numerically the laminar mixed convection in
horizontal concentric annuli with nonuniform circumferential heating. Secondary
flow and Nusselt number were analyzed in their study. They also investigated the
transient conjugated natural convection and conduction in openended vertical
concentric annuli, which could be modeled as twodimensional flow because of its
axisymmetry. The authors also analyzed the free convection in vertical eccentric
annuli with a uniformly heated boundary, which has to be simulated by three
dimensional model due to the eccentricity.
8. Mota et al. [4042] simulated the twodimensional natural convection in horizontal
concentric or eccentric annuli of elliptic or circular cross sections. The heat transfer
and fluid flow were analyzed for different inner/outer diameter ratios and eccentricity
of inner/outer pipes. They found that the eccentric elliptic annulus has the lowest heat
loss.
Numerical Analysis of Heat Transfer and Fluid Flow… 175
9. Char et al. [4345] investigated numerically the twodimensional natural or mixed
convection in horizontal concentric or eccentric annuli. The outer cylinder was kept
at a lower temperature while the inner cylinder was rotating and kept at a higher
temperature. Low Reynolds number turbulence model was adopted to account for the
turbulence effects. The heat transfer and fluid flow were analyzed for different
Rayleigh numbers, Reynolds numbers and inner/outer diameter ratios.
10. Ho et al. [4648] performed experimental and numerical investigations for the two
dimensional natural convection in horizontal concentric or eccentric annuli. The
outer cylinder was kept at constant temperature while the inner cylinder was kept at
constant heat flux. The heat transfer and fluid flow were investigated for different
Rayleigh numbers, Prandtl numbers and eccentricity of inner/outer cylinders. They
found that the heat and fluid flow were affected mainly by the Rayleigh number and
eccentricity of inner/outer cylinders, while the Prandtl number had only minor effect
on this flow.
11. Mizushima et al. [49,50] studied the instability and transition of twodimensional
natural convection in horizontal concentric annuli. The outer cylinder was kept at a
lower temperature while the inner cylinder was at a higher temperature. The heat
transfer and fluid flow were analyzed for different Prandtl numbers. The authors also
investigated the linear stability of the flow.
12. Hamad et al. [51,52] conducted experiments for the natural convection in inclined
annuli with closed ends. The inner cylinder was kept at constant heat flux. The heat
transfer and fluid flow were investigated for different Rayleigh numbers, Nusselt
numbers, inclined angles and inner/outer diameter ratios. They found that the heat
and fluid flow were affected mainly by the Rayleigh number and inner/outer
diameter ratios, while the inclined angle had only minor effect on this flow.
13. Raghavarao and Sanyasiraju [53,54] studied numerically the twodimensional natural
convection in horizontal concentric or eccentric annuli. The inner and outer cylinders
were both kept at constant temperature. The heat transfer and fluid flow were
investigated for different inner/outer diameter ratios and eccentricity of inner/outer
cylinders.
14. Choi and Kim [55,56] investigated numerically the threedimensional linear stability
of mixedconvective flow between rotating horizontal concentric cylinders. The inner
cylinder was rotating and kept at constant heat flux. The authors found that the
heating of the inner cylinder delays the formation of Taylor vortices when the
rotating effect is more pronounced than the buoyancy effect. On the other hand,
when rotation and buoyancy are comparable, the flow becomes unstable. This
instability is caused mainly by the buoyancy effect.
From the above literature review, it can be seen that there have been very few three
dimensional investigations of the heat and fluid flow in concentric or eccentric annuli
between two horizontal cylinders with open ends, except with the configuration of cavity type
[1,15,16]. Owing to the broad application in practical engineering problems, this chapter is
devoted to a detailed discussion of the simulation method for the heat transfer and fluid flow
inside threedimensional horizontal concentric or eccentric annuli with open ends, with the
aim to get a more thorough understanding of this problem.
ChunLang Yeh 176
Mathematical Model
For convenience of illustration, take the underground conduit of electrical power cable as
an example. In this section, two situations are discussed. One is a purely natural convective
flow and the other is a conjugated heat and fluid flow which involves conduction, convection
and radiation.
2.1. Natural Convective Flow
Heat is generated from the electrical resistance of the power cable and the heat
dissipation process in the annulus relies on the natural convection heat transfer from both
open ends of the conduit, which penetrate onto the manhole surfaces. The heat dissipation rate
is determined from the ventilation stemmed from natural convection and will affect the
lifetime of the power cable. Owing to the symmetric nature of the flow field with respect to
the two open ends and to a vertical plane crossing the center of the cylinders, the
computational domain is schematically shown in Figure1. Selected transverse and
longitudinal sections are illustrated in Figure 2.
(a)
Figure 1. Continued on next page.
Numerical Analysis of Heat Transfer and Fluid Flow… 177
(b)
Figure 1. Illustration of the computational domain and zonal grid distribution  (a) overall view (b)
zoomin view near the interface of the two zones.
Unlike the vertical annuli in which the fully developed thermal boundary conditions may
be achievable[39], the boundary conditions at the open ends of the present problem are much
more troublesome. Although this plane is named as the outlet (see Figure 1) in consideration
of the heat dissipation route in the conduit, it consists of inflow (fresh flowing fluid) and
outflow (heated flowing fluid) at the same plane. An approach used in the simulations of the
noncavity type, buoyancyinduced flows is the “zonal grid” approach[5760] which extends
the computational domain outside the outlet plane so that the boundary conditions can be
reasonably specified with the ambient flow properties. Typical examples can be found in
Refs.59 and 60. Although this approach requires enormous computations for three
dimensional problems, it provides more reliable results among existing approaches. In this
work, the “zonal grid” approach is adopted to resolve the problem of the outlet boundary
conditions. In the following discussion, two different cases are discussed. The first is a
numerical investigation of the threedimensional natural convection inside a horizontal
concentric annulus with specified wall temperature or heat flux[59] and the second is a three
dimensional natural convection inside horizontal concentric or eccentric annuli with specified
wall heat flux[60].
The boundary conditions for the first problem as illustrated schematically in Figure 1 are
stated as follows. Either adiabatic or isothermal condition is given on the outer cylinder
surface, while on the inner cylinder a constant heat flux is given. For the isothermal condition,
the outer cylinder surface is maintained at T
ref
（ ） i.e. 300 K . No slip condition is given to all
the three components of the velocity on the outer and inner cylinder surfaces. Since no flow
crosses the circumferentially (or longitudinally) symmetric plane, the angular (or axial)
ChunLang Yeh 178
velocity vanishes on that plane. The angular (or axial) derivatives of the remaining velocity
components and temperature also vanish on the circumferentially (or longitudinally)
symmetric plane.
Figure 2. Illustration of the selected  (a) transverse sections for eccentric annulus, (b) transverse
sections for concentric annulus, and (c) longitudinal sections.
The boundary conditions for the second problem are stated as follows. The adiabatic
condition is given on the outer cylinder surface, while on the inner cylinder a constant heat
flux, which in practical situation is resulted from the heat generation of the power cable due
to electrical resistance, is given. No slip condition is given to all the three components of the
velocity on the outer and inner cylinder surfaces. Since no flow crosses the circumferentially
(or longitudinally) symmetric plane, the angular (or axial) velocity vanishes on that plane.
The angular (or axial) derivatives of the remaining velocity components and temperature also
vanish on the circumferentially (or longitudinally) symmetric plane.
Numerical Analysis of Heat Transfer and Fluid Flow… 179
Because the thermal boundary conditions at the inner cylinder surfaces are specified in
terms of heat fluxes instead of temperatures in the above two example problems, a modified
Rayleigh number is defined as follows.
) / (
* * *
* *
3 * * *
ref h w
ref ref
h ref
o
k D q
D g
Ra
α υ
β
= (1)
As pointed out by Kuehn and Goldstein [8], an onset of transition from laminar to
turbulent regimes starts near
6
10 4× = Ra for gases in a concentric annulus of
6 . 2 / =
i o
D D . Moreover, a study by Labonia and Gju [3] indicated that chaotic flows were
observed in the range of
5 5
10 37 . 3 10 9 . 0 × ≤ ≤ × Ra for a concentric annulus of
36 . 2 / =
i o
D D . However, their conclusions were drawn from the experimental observations
and based upon the cases associated with the constant temperature differences between the
outer and inner cylinders. Kumar [6] made a numerical investigation using a twodimensional
model for an infinitely long, horizontal, concentric annulus where the inner cylinder was
specified by a constant heat flux and the outer cylinder was isothermally cooled. He found
that the critical
o
Ra above which the numerical results failed to converge were
5
10 1 . 3 × and
6
10 3× at 5 . 1 / =
i o
D D and 2.6, respectively. Kumar also pointed out that it was hard to
judge whether the flow would become oscillatory or threedimensional beyond the critical
o
Ra for a given ratio of
i o
D D / . The present example problems, which are essentially
threedimensional with the medium of air（ 7 . 0 Pr = ） , encounter the same convergence
difficulty beyond ) 10 (
7
O Ra
o
= , which may imply an onset of transition from steady
laminar to chaotic or even turbulent flows. Here, two cases of Rayleigh numbers,
5
10 =
o
Ra and
6
10 =
o
Ra , are calculated and demonstrated.
The flow pattern of interest here necessitates the solution of threedimensional fully
elliptic type of partial differential equations, which describe the natural convective flow field.
Considering the steady state flow situation, the governing equations in Cartesian coordinates
read：
( ) ( ) ( ) 0 ρw
z
ρv
y
ρu
x
=
∂
∂
+
∂
∂
+
∂
∂
(2)
( ) ( ) ( )
⎥
⎦
⎤
⎢
⎣
⎡
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
∂
∂
+
∂
∂
+
∂
∂
∂
∂
+
∂
∂
+
∂
∂
+
∂
∂
⋅
+
∂
∂
− =
∂
∂
+
∂
∂
+
∂
∂
z
w
y
v
x
u
x z
u
y
u
x
u
x
p
wu
z
vu
y
uu
x
3
1
Pr
2
2
2
2
2
2
ρ ρ ρ
(3)
ChunLang Yeh 180
( ) ( ) ( )
2 2
2
2
2
2
2
1
3
1
Pr
Fr z
w
y
v
x
u
y z
v
y
v
x
v
y
p
wv
z
vv
y
uv
x
ρ
ρ ρ ρ
−
+
⎥
⎦
⎤
⎢
⎣
⎡
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
∂
∂
+
∂
∂
+
∂
∂
∂
∂
+
∂
∂
+
∂
∂
+
∂
∂
⋅
+
∂
∂
− =
∂
∂
+
∂
∂
+
∂
∂
(4)
( ) ( ) ( )
⎥
⎦
⎤
⎢
⎣
⎡
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
∂
∂
+
∂
∂
+
∂
∂
∂
∂
+
∂
∂
+
∂
∂
+
∂
∂
⋅
+
∂
∂
− =
∂
∂
+
∂
∂
+
∂
∂
z
w
y
v
x
u
z z
w
y
w
x
w
z
p
ww
z
vw
y
uw
x
3
1
Pr
2
2
2
2
2
2
ρ ρ ρ
(5)
( ) ( ) ( )
Π − Φ ⋅ ⋅ +
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
∂
∂
+
∂
∂
+
∂
∂
⋅
+
∂
∂
+
∂
∂
+
∂
∂
=
∂
∂
+
∂
∂
+
∂
∂
Ec
z
p
w
y
p
v
x
p
u Ec
z y x
w
z
v
y
u
x
Pr
2
2
2
2
2
2
θ θ θ
θ ρ θ ρ θ ρ
(6)
where
2
2
2
2
2
2
2
3
2
2
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
∂
∂
+
∂
∂
+
∂
∂
−
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
∂
∂
+
∂
∂
+
⎟
⎠
⎞
⎜
⎝
⎛
∂
∂
+
∂
∂
+
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
∂
∂
+
∂
∂
+
⎥
⎥
⎦
⎤
⎢
⎢
⎣
⎡
⎟
⎠
⎞
⎜
⎝
⎛
∂
∂
+
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
∂
∂
+
⎟
⎠
⎞
⎜
⎝
⎛
∂
∂
= Φ
z
w
y
v
x
u
y
w
z
v
x
w
z
u
x
v
y
u
z
w
y
v
x
u
(7)
and
* * * *
/
w
q g V μ = Π (8)
Note that the Boussinesq approximation usually made in the formulation of the natural
convection is not adopted here and the density is determined using the ideal gas law. The
reason of using the ideal gas law rather than Boussinesq approximation for density
determination is that the Boussinesq approximation is not valid in case of high temperature
difference, which may arise when the Rayleigh number exceeds a critical value.
2.2. Conjugated Heat and Fluid Flow
In section 2.1, the effects of heat conduction and radiation are neglected. This assumption
is valid only if the outer cylinder（concrete conduit） is adiabatic and the temperature of the
flow field is low enough. A practical underground power cable placed in a concrete conduit
can be schematically shown as Figure 3. To give a preliminary estimate for the influence of
the conduction and radiation heat transfer on this problem, the twodimensional simplified
Numerical Analysis of Heat Transfer and Fluid Flow… 181
model shown in Figure 4 is analyzed. Consider conjugated conduction and convection heat
transfer first.
solar beam
concrete
conduit
soil
power cable
Figure 3. Illustration of a practical underground power cable placed in a concrete conduit.
concrete conduit
2cm
power cable
air
r
2
r
1
=d
o
/2
d
i
/2
soil
Figure 4. Illustration of a twodimensional simplified model for the underground power cable placed in
a concrete conduit.
ChunLang Yeh 182
Assume that the concrete conduit is in thermal equilibrium with the surrounding soil,
diameter of the power cable is d
i
=0.1m, the inner radius of the concrete conduit is
r
1
(=d
o
/2)=0.1m，and its thickness is r
2
r
1
=2cm, then the heat transfer（which includes both
conduction and convection）from the soil to the air is[61]
x x m soil
x m soil
x x m soil x m soil
T T
T T
T T T T
x P U q
Δ +
Δ +
−
−
− − −
Δ =


ln
)  ( )  (
) (
,
where the conjugated heat transfer coefficient, U, is
1
2 1
ln
1
1
r
r
k
r
h
U
concrete air
+
=
(9)
k
concrete
=0.7W/m．K is the thermal conductivity of the concrete conduit.
From Figure 4 of Ref.7, the Nusselt number Nu=
10 ≅
k
D h
h o
. Then, the convection
heat transfer coefficient for the air inside the annulus is
h o
D k h / 10 = = 7 . 2 1 . 0 / 10 7 . 2 10
2
= × ×
−
K m W ⋅
2
/ (10)
where
1 . 0 1 . 0 2 . 0 = − = − =
i o h
d d D m;
thermal conductivity for air
2
10 7 . 2
−
× =
air
k K m W ⋅ / . The conjugated heat transfer
coefficient, U, can then be found by substituting
o
h from Eq.(10) into h
air
in Eq.(9).
52 . 2
1 . 0
12 . 0
ln
7 . 0
1 . 0
7 . 2
1
1
=
+
= U
K m W ⋅
2
/
The error caused by the neglect of conduction heat transfer then is
% 1 . 7
52 . 2
52 . 2 7 . 2
=
−
=
−
U
U h
o
The above analysis can give us a preliminary estimate of the influence of the conduction
heat transfer on this problem. In fact, it can be observed from Eq.(9) that the conjugated heat
Numerical Analysis of Heat Transfer and Fluid Flow… 183
transfer coefficient, U, is closely connected with the outer/inner radius ratio（r
2
/r
1
）of the
concrete conduit, which is related to the thickness of the concrete conduit. For a thin concrete
conduit, r
2
/r
1
approaches unity and ln(r
2
/r
1
) is very small. Under such situation, the
conduction heat transfer can be neglected. However, in practical engineering problems, the
concrete conduit can not be too thin because of the high pressure it sustained（due to the soil
for a land cable or the sea water for a submarine cable）. Therefore, the conduction heat
transfer should be taken into account in practical engineering applications.
To illustrate the influence of the radiation heat transfer on this problem, the two
dimensional simplified model（Figure 4）is analyzed again. Assume the surface temperature
of the power cable to be T
i
, radiation emissivity 88 . 0 ≅
i
ε （for rubber）, and the inner
surface temperature of the concrete conduit to be T
o
, radiation emissivity 91 . 0 ≅
o
ε （for
concrete）. Also assume that the surface of the power cable and the inner surface of the
concrete conduit are gray surfaces, i.e. radiation emissivity and absorptivity are
equal（ α ε = ）. Then, the net radiation heat transfer on the surface of the power cable can
be estimated as
pi o i i i po i o o o pi cable r
A F T A F T A q
→ →
− = ×
4 4 ' '
σ ε σ ε
where the view factors
po
pi
i o
A
A
F =
→
， 1 =
→o i
F ，
8
10 67 . 5
−
× = σ
4 2
/ K m W ⋅
Then
] ) ( [
4
4 ' '
o
i
i o o cable r
T
T
T q ε ε σ − =
where
o
i
T
T
can be written as
o o
o
o
i
T
T
T
T T
T
T Δ
+ =
Δ +
= 1
.
From the author’s previous studies[59,60], K T 1 ≈ Δ , K T
o
301 ≈ for
5
10 =
o
Ra ,
and K T 12 ≈ Δ , K T
o
316 ≈ for
6
10 =
o
Ra , where the modified Rayleigh
number（
o
Ra ）is defined in Eq.(1).
This yields
o
T
T Δ
<< 1 ⇒
o o o
i
T
T
T
T
T
T Δ
+ ≈
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛ Δ
+ =
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
4 1 1
4 4
.
When
5
10 =
o
Ra , the net radiation heat transfer from the surface of the power cable
then is
ChunLang Yeh 184
5 . 8 )]
301
1
4 1 ( 88 . 0 91 . 0 [ 301 10 67 . 5
4 8 ' '
≈ × + − × × × =
−
cable r
q W/m
2
(11)
and when
6
10 =
o
Ra , the net radiation heat transfer from the surface of the power cable is
6 . 58 )]
316
12
4 1 ( 88 . 0 91 . 0 [ 316 10 67 . 5
4 8 ' '
− ≈ × + − × × × =
−
cable r
q W/m
2
(12)
On the other hand, from the definition of the modified Rayleigh number（Eq.(1)）, the
heat transfer from the surface of the power cable is
3
*
2
*
* *
* *
3 * * *
* *
* * *
Pr
1
) / ( ) / (
h
ref
ref
o
h ref
o
h ref
ref ref
h ref w
D
g
Ra
D k Ra
D g
D k q
⋅
= =
υ
β β
α υ
where 1 . 0
*
=
h
D m,
2 *
10 7 . 2
−
× =
ref
k K m W ⋅ / for air , Pr = 0.711, and
7
2
*
* *
10 2 . 13 × =
ref
ref
g
υ
β
( K m ⋅
3
/ 1 ) .
Therefore, when
5
10 =
o
Ra , the heat transfer from the surface of the power cable is
29 . 0
*
=
w
q
2
/ m W (13)
and when
6
10 =
o
Ra , the heat transfer from the surface of the power cable is
9 . 2
*
=
w
q
2
/ m W (14)
Comparisons of Eq.(11) with (13) and Eq.(12) with (14) reveal that the radiation heat
transfer plays an important role in this problem.
Although the above analysis is only a twodimensional simplified analysis, it can give us
a preliminary view of the influence of the conduction and radiation heat transfer on this
problem. The analysis shows that conduction and radiation heat transfer should not be
neglected unless the outer cylinder is adiabatic and the temperature of the flow field is very
low.
In respect of the conjugated heat transfer of convection and conduction, Ha and Jung[62]
investigated numerically the threedimensional conjugated heat transfer of conduction and
natural convection in a differentially heated cubic enclosure with a heatgenerating cubic
conducting body. Méndez and Treviño[63] made a numerical study for the conjugated
Numerical Analysis of Heat Transfer and Fluid Flow… 185
conduction and natural convection heat transfer along a thin vertical plate with nonuniform
internal heat generation. Liu et al.[64] performed a twodimensional numerical analysis of the
coupled conductionconvection problem for an underground rectangular duct containing three
insulated cables. Du and Bilgen[65] studied numerically the coupling effect of the wall
conduction with natural convection in a rectangular enclosure. Kim and Viskanta[66]
analyzed numerically the effects of wall conductance on the natural convection in differently
oriented square cavities. A common feature of the above studies for the conjugated
conductionconvection heat transfer is to consider the individual heat transfer mode in each
medium and to keep continuities of the temperature and heat flux at the interfaces of any two
different media. Take the underground power cable as an example, the heat transfer modes
include conduction of the power cable (solid), convection of the air inside the concrete
conduit (gas), conduction of the concrete conduit (solid) and conduction of the soil (solid).
Heat transfer modes in each medium have to be solved separately. Continuities of the
temperature and heat flux at the interfaces of any two different media provide the necessary
boundary conditions.
With respect to the influence of the radiation heat transfer [6773], the zone model and
Monte Carlo method are generally recognized as more accurate ones. However, these two
methods require enormous computing time and storage. In addition, they are not of
differential type and therefore are more cumbersome to incorporate with the differential
equations of fluid motion and energy. Perhaps the most commonly used radiation model for
engineering application is the flux model. Recently, the finite volume method （FVM）and
the discrete ordinates method（DOM）have become more and more popular because of their
success in simulating irregular geometries [6972]. In using the FVM and DOM, the
simulation domain is divided into small blocks. This is quite similar to the concept of the
finite volume method in computational fluid dynamics. The DOM needs a quadrature set
which has a crucial influence on the accuracy of the method. On the other hand, the FVM is
less restricted and is better for the conservation of radiation heat transfer.
As pointed out earlier, an onset of transition from laminar to turbulent regimes starts near
7
10 = Ra . Among the existing turbulence simulation methods, Direct Numerical
Simulation（ ） （ ） DNS and Large Eddy Simulation LES can simulate the actual physics of
turbulence more accurately. However, using them for routinely designing and analyzing
engineering problems can not be achieved nowadays. DNS is limited to low Reynolds
number flows because of the large computer resources needed. Although LES reduces the
requirement to some extent; however, much more sophisticated subgrid scale models and
nearwall treatments are required for many flows of interest. A recent research of the LES
was conducted by Worthy and Rubini [74] who studied LES stress and scalar flux subgrid
scale models in the context of buoyant jets. Reynolds stress model (RSM) is essentially more
realistic than the Eddy viscosity model (EVM) because it involves modeled transport
equations for all of the Reynolds stresses. Yilbas et al. [75] employed the RSM to examine jet
impingement onto a hole with a constant wall temperature and analyzed the influence of the
hole wall temperatures and jet velocities. In their study, the Nusselt number ratio (ratio of the
Nusselt number predicted to the Nusselt number obtained for a fully developed turbulent flow
based on the hole entrance Reynolds number) was computed and the mass flow ratio (ratio of
mass flow rate through the hole to mass flow emanating from the nozzle) was determined. On
the other hand, even though the EVM assumes a crude relation between the turbulence
ChunLang Yeh 186
quantity（ ） （ ） Reynolds stress and the mean flow quantity strain rate , turbulence models
based on this concept, such as the standard kε model, have been widely used for industrial
CFD calculations due to numerical stability and low cost. However, it has long been
recognized that the standard kε model has defects in predicting turbulent flows with large
streamline curvature and rotational effects. Consequently, the more sophisticated two
equation models employ the Low Reynolds Number（ ） LRN Model, the NonLinear Eddy
Viscosity Model（ ） （ ） NEVM or the Algebraic Reynolds Stress Model ARSM to improve the
suitability of the standard kε model. Brescianini and Delichatsios [76] examined several
versions of the kε turbulence model as to their suitability for computational fluid dynamics
modeling of free turbulent jets and buoyant plumes. Yang and Ma [77] investigated the
predictive performance of linear and nonlinear eddyviscosity turbulence models for a
confined swirling coaxial jet. Abdon and Sundén [78] carried out an investigation of a single
round unconfined impinging air jet under different flow and geometrical conditions to assess
the performance of linear and nonlinear twoequation turbulence models. Wei et al. [79]
proposed and formulated an algebraic turbulent mass flux model (AFM), which properly
accounts for swirl–turbulence interactions, and incorporated this model with an algebraic
Reynolds stress model proposed previously to simulate the swirling turbulent flow and
mixing in a combustor with helium/air jet and swirling air stream. In the author’s previous
study[80], the performances of three LEVM, one ARSM and one DRSM（ Differential
Reynolds Stress Model） turbulence models are evaluated for the simulation of the plain
orifice atomizer and the pressureswirl atomizer flows. These models include the standard kε
model[81], LaunderSharma’s LRN kε model[82], NaganoHishida’s LRN kε model[83],
GatskiSpeziale’s ARSM model[84], and RandriamampianinaSchiestelWilson’s DRSM
model[85]. Adequate evaluations can help users to select a suitable turbulence model for their
applications.
Numerical Method
The governing equations stated in last section can be cast into the following general form,
which permits a single algorithm to be applied.
( )
Φ Φ
+
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
∂
Φ ∂
Γ
∂
∂
= Φ
∂
∂
R
x x
v
x
j j
j
j
ρ (15)
To facilitate the handling of complex geometry of the present problem, the bodyfitted
coordinate system is used to transform the physical domain into a computational domain,
which is in a rectangular coordinate system with uniform control volumes. Transformation of
Eq.(15) to the bodyfitted coordinates leads to
( )
Φ Φ
+
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
∂
Φ ∂
Γ
∂
∂
= Φ
∂
∂
JR
q
Jg
q
V J
q
k
jk
j
j
j
ρ (16)
Numerical Analysis of Heat Transfer and Fluid Flow… 187
where
j
q ： （ curvilinear coordinates ζ η ξ , , ）
J： ； Jacobian ≡
ζ ξ η η ζ ξ ξ η ζ η ξ ζ ξ ζ η ζ η ξ
z y x z y x z y x z y x z y x z y x − − − + +
j
V ： contravariant velocity（ U,V,W）
[ ] ) ( ) ( ) (
1
η ζ ζ η ζ η η ζ η ζ ζ η
y x y x w z x z x v z y z y u
J
U − + − + − =
[ ] ) ( ) ( ) (
1
ζ ξ ξ ζ ξ ζ ζ ξ ζ ξ ξ ζ
y x y x w z x z x v z y z y u
J
V − + − + − =
[ ] ) ( ) ( ) (
1
ξ η η ξ η ξ ξ η ξ η η ξ
y x y x w z x z x v z y z y u
J
W − + − + − =
jk
g ： metric tensor
[ ]
2 2 2
2
11
) ( ) ( ) (
1
η ζ ζ η ζ η η ζ η ζ ζ η
y x y x z x z x z y z y
J
g − + − + − =
[ ]
2 2 2
2
22
) ( ) ( ) (
1
ζ ξ ξ ζ ξ ζ ζ ξ ζ ξ ξ ζ
y x y x z x z x z y z y
J
g − + − + − =
[ ]
2 2 2
2
33
) ( ) ( ) (
1
ξ η η ξ η ξ ξ η ξ η η ξ
y x y x z x z x z y z y
J
g − + − + − =
[
] ) )( (
) )( ( ) )( (
1
2
21 12
ζ ξ ξ ζ η ζ ζ η
ξ ζ ζ ξ ζ η η ζ ζ ξ ξ ζ η ζ ζ η
y x y x y x y x
z x z x z x z x z y z y z y z y
J
g g
− − +
− − + − − = =
[
] ) )( (
) )( ( ) )( (
1
2
31 13
ξ η η ξ η ζ ζ η
η ξ ξ η ζ η η ζ ξ η η ξ η ζ ζ η
y x y x y x y x
z x z x z x z x z y z y z y z y
J
g g
− − +
− − + − − = =
[
] ) )( (
) )( ( ) )( (
1
2
32 23
ξ η η ξ ζ ξ ξ ζ
η ξ ξ η ξ ζ ζ ξ ξ η η ξ ζ ξ ξ ζ
y x y x y x y x
z x z x z x z x z y z y z y z y
J
g g
− − +
− − + − − = =
ChunLang Yeh 188
The grid layout is constructed by connecting the grid points in each transverse plane,
which are generated by solving the twodimensional elliptic type of partial differential
equations governing the distribution of the grid points [86]. Numerical calculation of Eq.(16)
is performed using the controlvolume based finite difference procedure. The discretized
governing equations are solved on a nonstaggered grid system in association with the
SIMPLEC algorithm [87] and QUICK scheme [88].
In the use of the “zonal grid” approach[5760], the computational domain is extended
outside the outlet plane and is divided into two subdomains of Zones I and II, as
schematically illustrated in Figure 1. Zone I and Zone II are overlapped by three layers of grid
points. The outer cylinder is flush with the adiabatic solid wall while the inner cylinder
extends to the free boundary. The boundary conditions for Zone I have been stated in
“MATHEMATICAL MODEL” and will not be repeated here. The boundary conditions of
Zone II are as follows. On the surface of the inner cylinder, the same boundary conditions as
specified for Zone I are used. On the adiabatic solid wall, the noslip condition and adiabatic
wall are specified. The condition of zero normal gradients is met on the symmetric plane
except for the normal velocity component, which vanishes naturally. On the free boundaries,
the normal gradients of all the dependent variables except for the temperature are set to be
zero. The temperature conditions at the free boundaries are specified as follows： When the
flow at any such boundary is leaving the domain, the normal temperature gradient is taken as
zero. However, when the flow comes into the domain, its temperature is assigned to that of
the ambient.
The treatment of the interface of the two zones follows the overlapping grid method [57].
Starting with guessed values, solutions in Zone I together with the overlapped region are
updated by one sweep of iteration. The updated solutions at the outlet plane of Zone I are then
interpolated by bilinear interpolation as the boundary conditions of Zone II. The solutions in
Zone II are then updated by one sweep of iteration and are used, in turn, to interpolate the
values in the overlapped region, which provide the new boundary conditions for the
resolutions of Zone I together with the overlapped region. This completes a full solution
cycle. In each solution cycle, continuity of the dependent variables and conservation of fluxes
are preserved across the interface. In fact, this is the key to the success of the approach. The
solution cycle is repeated until the convergence criterion is satisfied. The convergence
criterion is described below.
The general form of the discretized governing equations can be written as
Φ
+ Φ = Φ
∑
b A A
nb
nb nb P P
(17)
Define
Φ Φ
− Φ − Φ =
∑
b A A B
nb
nb nb P P i
(18)
Φ Φ
+ Φ + Φ =
∑
b A A H
nb
nb nb P P i
(19)
Numerical Analysis of Heat Transfer and Fluid Flow… 189
( )
∑
Φ
Φ
Φ
=
i
i
i
H
N
B
1
max
λ (20)
where i is an arbitrary grid point in the computational domain and N is the total number of
grid points. When
4
10 0 . 1
− Φ
× ≤ λ for each dependent variable Φ in both the two zones,
the iteration process is convergent.
Illustrative Problems
In this section, two illustrative problems are discussed. The first is a numerical
investigation of the threedimensional natural convection inside a horizontal concentric
annulus with specified wall temperature or heat flux[59] and the second is a three
dimensional natural convection inside horizontal concentric or eccentric annuli with specified
wall heat flux[60]. Illustration of the computational domain and zonal grid distribution are
schematically shown in Figure 1. The inner and outer diameters of Zone I are 0.1m and 0.2m,
respectively, and its length is 3m (equivalent to 30D
h
*
’s). Zone II is constructed by a cube
with side length of 20 D
h
*
’s to assure its boundary conditions being reasonably specified by
the ambient properties. Numerical tests reveal that the maximum difference in mass inflow
rate at the outlet plane between 51×51×101（ ） radial by angular by axial and 61×61×121
grid meshes for Zone I, while the corresponding grid meshes for zone II are 21×41×21（ x by
y by z） and 31×61×31, respectively, is less than 0.5﹪. Therefore, the former set of grid
mesh is adopted in the present work. Figure 2 illustrates the section positions of the
configuration being studied.
4.1. Natural Convection inside a Horizontal Concentric Annulus with
Specified Wall Temperature or Heat Flux
For the first illustrative problem, both the adiabatic and isothermal conditions for the
outer cylinder surface are examined for the conditions of r
o
/ r
i
= 2 and Ra
o
=10
6
to investigate
the natural convection heat dissipation inside the conduit.
Figure 5 shows the velocity vector plots at selected transverse and longitudinal sections
for the examined cases. Note that the results for sections AA, BB and DD are the vertical
projections of the flow field and therefore the ordinate is y, while the result for section CC is
the horizontal projection of the flow field and therefore the ordinate is x. The flow patterns
for the two cases are rather different. As observed from the velocity vector plots on the
longitudinal sections, there appear secondary flows in all the six longitudinal sections for both
cases. The secondary flow of the adiabatic case is stronger than that of the isothermal case.
Further downstream, the secondary flows evolve into counterrotating recirculation zones for
the adiabatic case. Another interesting phenomenon can be observed from the velocity vector
plots on the transverse sections. For both cases, the inflow paths are clearly observed in the
portion below the inner cylinder, whereas the outflow paths are in the portion above. Such
flow patterns result obviously from the buoyancy effect, which can be observed in thermal
ChunLang Yeh 190
plume phenomena. This can also be seen from the velocity vector plots at a distance slightly
outside the open end (i.e. in Zone II) shown in Figure 6, in which the entrainment effect
caused by the upward motion of the buoyant flows can also be seen.
0 10 20 30
z
0.9
1
1.1
1.2
y
Section BB
0 10 20 30
z
0.6
0.8
1
x
Section CC
0 10 20 30
z
0.2
0.1
0
0.1
y
Section DD
0 10 20 30
z
0.5
0
0.5
1
1.5
y
Section AA adiabatic
1000V
*
0 0.5 1
0.5
0
0.5
1
1.5
x
y
500V
*
z=0
0 0.5 1
0.5
0
0.5
1
1.5
x
y
z=6
0 0.5 1
0.5
0
0.5
1
1.5
y
x
z=12
0 0.5 1
0.5
0
0.5
1
1.5
y
x
z=18
0 0.5 1
0.5
0
0.5
1
1.5
y
x
z=24
0 0.5 1
0.5
0
0.5
1
1.5
y
x
z=30
Figure 5. Continued on next page.
Numerical Analysis of Heat Transfer and Fluid Flow… 191
0 10 20 30
z
0.6
0.8
1
x
Section CC
0 10 20 30
z
0.9
1
1.1
1.2
y
Section BB
0 10 20 30
z
0.5
0
0.5
1
1.5
y
Section AA isothermal
1000V
*
0 10 20 30
z
0.2
0.1
0
0.1
y
Section DD
0 0.5 1
0.5
0
0.5
1
1.5
x
y
1000V
*
z=0
0 0.5 1
0.5
0
0.5
1
1.5
x
y
z=6
0 0.5 1
0.5
0
0.5
1
1.5
y
x
z=12
0 0.5 1
0.5
0
0.5
1
1.5
y
x
z=18
0 0.5 1
0.5
0
0.5
1
1.5
y
x
z=24
0 0.5 1
0.5
0
0.5
1
1.5
y
x
z=30
Figure 5. Velocity vector plots at selected transverse and longitudinal sections.
ChunLang Yeh 192
0 1 2
x
2
0
2
4
y
adiabatic
1000V
*
0 1 2
x
2
0
2
4
isothermal
y
Figure 6. Velocity vector plots near the end of the annulus (z=30.6).
0 0.5 1
0.5
0
0.5
1
1.5
3
5
4
3
2
1
5
Level T(K)
6 315
5 313
4 311
3 309
2 307
1 305
y
x
z=12
0 0.5 1
0.5
0
0.5
1
1.5
2
3
4
5
6
7
8
Level T(K)
9 315
8 314
7 313
6 312
5 311
4 310
3 309
2 308
1 307
x
y
z=0
0 0.5 1
0.5
0
0.5
1
1.5
2
3
4
5
1
6
Level T(K)
6 314
5 312
4 310
3 308
2 306
1 304
y
x
z=24
0 0.5 1
0.5
0
0.5
1
1.5
5
4
3
2
6
1
Level T(K)
6 314
5 312
4 310
3 308
2 306
1 304
y
x
z=30
adiabatic
Figure 7. Continued on next page.
Numerical Analysis of Heat Transfer and Fluid Flow… 193
0 0.5 1
0.5
0
0.5
1
1.5
1
2
3
5
4
Level T(K)
6 302.5
5 302
4 301.5
3 301
2 300.5
1 300
y
x
z=12
0 0.5 1
0.5
0
0.5
1
1.5
1
2
3
5
4
Level T(K)
6 302.5
5 302
4 301.5
3 301
2 300.5
1 300
x
y
z=0
0 0.5 1
0.5
0
0.5
1
1.5
1
2
3
5
4
Level T(K)
6 302.5
5 302
4 301.5
3 301
2 300.5
1 300
y
x
z=24
0 0.5 1
0.5
0
0.5
1
1.5
1
2
3
5
4
Level T(K)
6 302.5
5 302
4 301.5
3 301
2 300.5
1 300
y
x
z=30
isothermal
Figure 7. Temperature contours at selected longitudinal sections.
0 1 2 3 4
x
2
0
2
4
6
1
2
5
7
1
2
3
4 3
Level T(K)
8 308
7 307
6 306
5 305
4 304
3 303
2 302
1 301
y
300 K
adiabatic
0 1 2 3 4
x
2
0
2
4
6
1
1
2
3
9
Level T(K)
12 303
11 302.75
10 302.5
9 302.25
8 302
7 301.75
6 301.5
5 301.25
4 301
3 300.75
2 300.5
1 300.25
isothermal
300 K
y
Figure 8. Temperature contours near the end of the annulus (z=30.6).
ChunLang Yeh 194
0 60 120 180
306
308
310
312
314
316
φ(
o
)
T(K)
z=0
z=12
z=24
z=30
adiabatic
0 60 120 180
301
301.5
302
302.5
φ(
o
)
T(K)
z=0
z=12
z=24
z=30
isothermal
Figure 9. Azimuthal distributions of the inner cylinder surface temperatures at four longitudinal
sections.
Figure 7 displays the temperature contours at four selected longitudinal sections along the
annulus in Zone I (refer to Figure 2 for the section positions), while Figure 8 shows the
temperature contours at a distance slightly outside the open end for the examined cases. It is
observed that higher temperature regions around the inner cylinder surface locate in the upper
portion. This can also be seen from the azimuthal temperature distributions along the inner
cylinder surface shown in Figure 9. As pointed out in the above discussion of the flow pattern
on the transverse sections, the (hot) outflow is observed in the portions above the inner cylinder,
whereas the (cold) inflow in the portions below. The temperature contours at a position slightly
outside the annulus, as shown in Figure 8, consistently reflect the above observation.
Numerical Analysis of Heat Transfer and Fluid Flow… 195
Figure 9 shows the azimuthal temperature distributions along the inner cylinder surface at
four selected longitudinal sections. It is clearly seen that the highest temperature occurs right
at the top of the inner cylinder (i.e.
0
0 = φ ). Also from the figure it is observed that the inner
cylinder surface temperatures decrease toward the outlet plane for the adiabatic case, while
remains relatively constant for the isothermal case. In addition, the variation of the surface
temperatures is smaller for the isothermal case. The maximum inner cylinder surface
temperature of the isothermal case is lower than that of the adiabatic case by about 11 K.
4.2. Natural Convection inside Horizontal Concentric or Eccentric Annuli
with Specified Wall Heat Flux
For the second illustrative problem, two cases of modified Rayleigh
number,
5
10 =
o
Ra and
6
10 =
o
Ra , are examined. However, the temperature variation in
the entire flow field for the case of
5
10 =
o
Ra is not obvious. Therefore, only the case of
6
10 =
o
Ra , which may lead to more distinct temperature variations on the inner cylinder
surface, is demonstrated here. Figure 10 shows the velocity vector plots at selected transverse
and longitudinal sections for the concentric and eccentric annuli. The flow patterns for the
two cases are rather different. As observed from the velocity vector plots on the longitudinal
sections, there appear secondary flows in all the six longitudinal sections for both cases.
0 10 20 30
0
0.2
0.4
0.6
0.8
1
Section AA
eccentric case
1000V
*
0 10 20 30
0
0.25
0.5
0.75
Section BB
0 10 20 30
0.5
0
0.5
Section CC
0 10 20 30
0.75
0.5
0.25
0
0.25
Section DD
Figure 10. Continued on next page.
ChunLang Yeh 196
0 0.5 1
1
0.5
0
0.5
1
x
y
z=6
0 0.5 1
1
0.5
0
0.5
1
x
y
1000V
*
z=0
0 0.5 1
1
0.5
0
0.5
1
y
x
z=12
0 0.5 1
1
0.5
0
0.5
1
y
x
z=18
0 0.5 1
1
0.5
0
0.5
1
y
x
z=24
0 0.5 1
1
0.5
0
0.5
1
y
x
z=30
0 10 20 30
0.9
1
1.1
1.2
Section BB
0 10 20 30
0.6
0.8
1
Section CC
0 10 20 30
0.2
0.1
0
0.1
Section DD
0 10 20 30
0.5
0
0.5
1
1.5
Section AA
concentric case
1000V
*
Figure 10. Continued on next page.
Numerical Analysis of Heat Transfer and Fluid Flow… 197
0 0.5 1
0.5
0
0.5
1
1.5
y
x
z=30
0 0.5 1
0.5
0
0.5
1
1.5
y
x
z=24
0 0.5 1
0.5
0
0.5
1
1.5
y
x
z=18
0 0.5 1
0.5
0
0.5
1
1.5
y
x
z=12
0 0.5 1
0.5
0
0.5
1
1.5
x
y
z=6
0 0.5 1
0.5
0
0.5
1
1.5
x
y
1000V
*
z=0
Figure 10. Velocity vector plots at selected transverse and longitudinal sections.
Further downstream, the secondary flows evolve into counterrotating recirculation
zones. Another interesting phenomenon can be observed from the velocity vector plots on the
transverse sections. For the concentric case, the inflow paths are clearly observed in the
portion below the inner cylinder, whereas the outflow paths are in the portion above. For the
eccentric case, since the inner cylinder sits on the bottom of the outer cylinder, the inflow and
outflow paths coexist in each transverse plane. Such flow patterns result obviously from the
buoyancy effect that can be observed in thermal plume phenomena. This can also be seen
from the velocity vector plots at a distance slightly outside the open end (i.e. in Zone II)
shown in Figure 11, in which the entrainment effect caused by the upward motion of the
buoyant flows can also be seen. Another point to be drawn from Figure 10 is that the axial
flow motion is significant and cannot be neglected. Therefore the adoption of a three
dimensional formulation is necessary.
Figure 12 displays the temperature contours at four selected longitudinal sections along
the annulus in Zone I (see Figure 2 for the section positions), while Figure 13 shows the
temperature contours at a distance slightly outside the open end for the examined
configurations. It is observed that higher temperature regions around the inner cylinder
surface locate in the lower portions of the annulus for the eccentric annulus, whereas in the
upper portion for the concentric annulus. This can also be seen from the azimuthal
temperature distributions along the inner cylinder surface shown in Figure 14. The
temperature contours at a position slightly outside the annulus, as shown in Figure 13,
consistently reflect the corresponding flow pattern at a distance slightly outside the open ends
shown in Figure 11.
ChunLang Yeh 198
Figure 11. Velocity vector plots near the end of the annulus (z=30.6).
Figure 12. Continued on next page.
Numerical Analysis of Heat Transfer and Fluid Flow… 199
Figure 12. Temperature contours at selected longitudinal sections.
Figure 13. Temperature contours near the end of the annulus (z=30.6).
ChunLang Yeh 200
Figure 14. Azimuthal distributions of the inner cylinder surface temperatures at four transverse sections.
Figure 14 shows the azimuthal temperature distributions along the inner cylinder surface
at four selected longitudinal sections. It is clearly seen that the highest temperatures occur
right at the bottom of the inner cylinder (i.e.
0
180 = φ ) for the eccentric annulus, whereas
Numerical Analysis of Heat Transfer and Fluid Flow… 201
occur right at the top of the inner cylinder (i.e.
0
0 = φ ) for the concentric annulus. Also from
the figure it is observed that the inner cylinder surface temperatures decrease toward the
outlet plane. In addition, the variation of the surface temperatures is smaller for the case of
concentric annulus. A more evenly distributed inner cylinder surface temperature is desirable
as far as the lifetime of the inner cylinder is concerned. With the concentric annulus, about 30
K decrement of the maximum inner cylinder surface temperature can be achieved as
compared to the eccentric annulus.
Conclusion
In this chapter, the simulation method for the heat and fluid flow inside threedimensional
horizontal concentric or eccentric annuli with open ends is discussed. The simulation
method discussed includes conjugated heat transfer model, turbulence model and zonal grid
approach, which extends the outlet boundary from the open end of the conduit to a far enough
outside position that can be reasonably specified with the ambient flow properties. Two
illustrative problems are discussed. The first is a threedimensional natural convection inside
a horizontal concentric annulus with specified wall temperature or heat flux and the second is
a threedimensional natural convection inside horizontal concentric or eccentric annuli with
specified wall heat flux. For the first illustrative problem, it is found that higher temperatures
around the inner cylinder occur in the region near its top. The maximum inner cylinder
surface temperature occurs right at the top of the inner cylinder. The inner cylinder surface
temperatures decrease toward the outlet plane for the adiabatic case, while remains relatively
constant for the isothermal case. The variation of the inner cylinder surface temperatures is
smaller for the isothermal case, as compared to the adiabatic case. For the second illustrative
problem, it is found that the eccentric annulus has a poorer natural convection heat dissipation
rate, as compared to the concentric annulus. The inner cylinder surface temperatures decrease
toward the outlet plane. It is also found that higher temperatures around the inner cylinder
occur in the region near its bottom（ ） the contacting point of the inner and outer cylinders for
the eccentric annulus, whereas in the region near its top for the concentric annulus. The
variation of the inner cylinder surface temperatures is smaller for the case of concentric
annulus, as compared to the eccentric annulus. The maximum inner cylinder surface
temperatures occur right at the bottom of the inner cylinder for the eccentric annulus, whereas
right at the top of the inner cylinder for the concentric annulus.
References
[1] Vaidya, N.; Shamsundar, N., Heat Transfer in Microgravity Systems, ASME HTD, 1997,
vol.305, 105120.
[2] Gju, S.; Iannetta, G.; Moretti, G., Experiments in Fluids, 1992, vol.12, 385393.
[3] Labonia, G.; Gju, G., Journal of Fluid Mechanics, 1998, vol.375, 179202.
[4] Chang, T. S.; Lin, T. F., International Journal for Numerical Methods in Fluids, 1990,
vol.10, 443459.
[5] Prusa, J.; Yao, L. S., Journal of Heat Transfer, 1983, vol.105, 108116.
[6] Kumar, R., International Journal of Heat and Mass Transfer, 1988, vol.31, 11371148.
ChunLang Yeh 202
[7] Kuehn, T. H.; Goldstein, R. J., International Journal of Heat and Mass Transfer, 1976,
vol.19, 11271134.
[8] Kuehn, T. H.; Goldstein, R. J., Journal of Heat Transfer, 1978, vol.100, 635640.
[9] Kuehn, T. H.; Goldstein, R. J., Journal of Fluid Mechanics, 1976, vol.74, 695719.
[10] Glakpe, E. K.; Watkins, C. B. Jr.; Cannon, J. N., Numerical Heat Transfer, 1986, vol.10,
279295.
[11] Dyko, M. P.; Vafai, K., Physics of Fluids, 2002, vol.14, no.3, 12911294.
[12] Dyko, M. P.; Vafai, K., Journal of Fluid Mechanics, 2001, vol.445, 3136.
[13] Dyko, M. P.; Vafai, K.; Mojtabi, A. K., Journal of Fluid Mechanics, 1999, vol.381,
2761.
[14] Iyer, S. V.; Vafai, K., Numerical Heat Transfer, Part A, 1999, vol.36, no.2, 115128.
[15] Iyer, S. V.; Vafai, K., International Journal of Heat and Mass Transfer, 1998, vol.41,
no.20, 30253035.
[16] Iyer, S. V.; Vafai, K., International Journal of Heat and Mass Transfer, 1997, vol.40,
no.12, 29012911.
[17] Desai, C. P.; Vafai, K., International Journal of Heat and Mass Transfer, 1994, vol.37,
no.16, 24752504.
[18] Vafai, K.; Ettefagh, J., International Journal of Heat and Mass Transfer, 1991, vol.34,
no.10, 25552570.
[19] Vafai K.; Ettefagh, J., International Journal of Heat and Mass Transfer, 1990, vol.33,
no.10, 23112328.
[20] Vafai, K.; Ettefagh, J., International Journal of Heat and Mass Transfer, 1990, vol.33,
23292344.
[21] Yoo, J. S., International Journal of Heat and Mass Transfer, 2003, vol.46, 24992503.
[22] Yoo, J. S.; Han, S. M., Fluid Dynamics Research, 2000, vol.27, 231245.
[23] Yoo, J. S., International Communications of Heat and Mass Transfer, 1999, vol.26,
no.6, 811817.
[24] Yoo, J. S., International Journal of Heat and Mass Transfer, 1999, vol.42, 32793290.
[25] Yoo, J. S., International Journal of Heat and Mass Transfer, 1998, vol.41, 30553073.
[26] Yoo, J. S., International Journal of Heat and Mass Transfer, 1998, vol.41, no.2, 293
302.
[27] Yoo, J. S., International Journal of Heat and Fluid Flow, 1996, vol.17, 587593.
[28] Ding, H.; Shu, C.; Yeo, K. S.; Lu, Z. L., Numerical Heat Transfer, Part A, 2005, vol.47,
no.3, 291313.
[29] Lee, T. S.; Hu, G. S.; Shu, C., Numerical Heat Transfer, Part A, 2002, vol.41, no.8, 803
815.
[30] Shu, C.; Zhu, Y. D., International Journal for Numerical Methods in Fluids, 2002,
vol.38, 429445.
[31] Shu, C.; Yao, Q.; Yeo, K. S.; Zhu, Y. D., Numerical Heat Transfer, 2002, vol.38, no.7
8, 597608.
[32] Shu, C.; Xue, H.; Zhu, Y. D., International Journal of Heat and Mass Transfer, 2001,
vol.44, 33213333.
[33] Shu, C., International Journal for Numerical Methods in Fluids, 1999, vol.30, 977993.
[34] Khillarkar, D. B.; Gong, Z. X.; Mujumdar, A. S., Applied Thermal Engineering, 2000,
vol.20, 893912.
Numerical Analysis of Heat Transfer and Fluid Flow… 203
[35] Ng, K. W.; Devahastin, S.; Mujumdar, A. S.; Arai, N., Journal of Chemical Engineering
of Japan, 1999, vol.32, no.3, 353357.
[36] Ng, K. W.; Gong, Z. X.; Mujumdar, A. S., International Communications of Heat and
Mass Transfer, 1998, vol.25, no.5, 631640.
[37] Habib, M. A.; Negm, A. A. A., Heat and Mass Transfer, 2001, vol.37, no.45, 427435.
[38] ElShaarawi, M. A. I.; Negm, A. A. A., Heat and Mass Transfer, 1999, vol.35, 133141.
[39] ElShaarawi, M. A. I.; Mokheimer, E. M. A., International Journal of Numerical
Methods for Heat and Fluid Flow, 1998, vol.8, no.5, 488503.
[40] Mota, J. P. B.; Esteves, I. A. A. C.; Portugal, C. A. M.; Esperanca, J. M. S. S.,
International Journal of Heat and Mass Transfer, 2000, vol.43, 43674379.
[41] Mota, J. P. B.; Saatdjian, E., International Journal of Numerical Methods for Heat and
Fluid Flow, 1997, vol.7, no.4, 401416.
[42] Mota, J. P. B.; Saatdjian, E., International Journal of Numerical Methods for Heat and
Fluid Flow, 1995, vol.5, no.1, 312.
[43] Char, M. I.; Hsu, Y. H., International Journal of Heat and Mass Transfer, 1998, vol.41,
no.12, 16331643.
[44] Char, M. I.; Hsu, Y. H., International Journal for Numerical Methods in Fluids, 1998,
vol.26, 323343.
[45] Char, M. I.; Lee, G. C., International Journal of Engineering Science, 1998, vol.36,
no.2, 157169.
[46] Ho, C. J.; Lin, Y. H., International Journal of Heat and Mass Transfer, 1991, vol.34,
no.6, 13711382.
[47] Ho, C. J.; Lin, Y. H.; Chen, T. C., International Journal of Heat and Fluid Flow, 1989,
vol.10, no.1, 4047.
[48] Ho, C. J.; Lin, Y. H., Journal of Heat Transfer, 1988, vol.110, no.4, 894900.
[49] Mizushima, J.; Hayashi, S., Physics of Fluids, 2001, vol.13, no.1, 99106.
[50] Mizushima, J.; Hayashi, S.; Adachi, T., International Journal of Heat and Mass
Transfer, 2001, vol.44, 12491257.
[51] Hamad, F. A.; Khan, M. K., Energy Conversion and Management, 1998, vol.39, no.8,
797807.
[52] Hamad, F. A., Solar & Wind Technology, 1989, vol.6, no.5, 573579.
[53] Raghavarao, C. V.; Sanyasiraju, Y. V. S. S., Computational Mechanics, 1996, vol.18,
no.6, 464470.
[54] Raghavarao, C. V.; Sanyasiraju, Y. V. S. S., International Journal of Engineering
Science, 1994, vol.32, no.9, 14371450.
[55] Choi, J. Y.; Kim, M. U., International Journal of Heat and Mass Transfer, 1995, vol.38,
275285.
[56] Choi, J. Y.; Kim, M. U., International Journal of Heat and Mass Transfer, 1993, vol.36,
41734180.
[57] Rai, M. M., AIAA Paper, 1985, pp.851519.
[58] Lee, D.; Yeh, C. L., Numerical Heat Transfer, 1993, vol.24, 273285.
[59] Yeh, C. L., International Journal of Heat and Mass Transfer, 2002, vol.45, no.4, 775
784.
[60] Yeh, C. L., JSME International Journal, Series B, 2002, vol.45, no.2, 301306.
[61] Özisik, M. N., Heat Conduction, Wiley Interscience, New York, 1980.
ChunLang Yeh 204
[62] Ha, M. Y.; Jung, M. J., International Journal of Heat and Mass Transfer, 2000, vol.43,
42294248.
[63] Méndez, F.; Treviño, C., International Journal of Heat and Mass Transfer, 2000,
vol.43, 27392748.
[64] Liu, Y.; Nhan, P. T.; Kemp, R.; Luo, X. L., Numerical Heat Transfer, Part A, 1997,
vol.31, 411431.
[65] Du, Z. G.; Bilgen, E., International Journal of Heat and Mass Transfer, 1992, vol.35,
no.8, 19691975.
[66] Kim, D. M.; Viskanta, R., Journal of Fluid Mechanics, 1984, vol.144, 153176.
[67] Zheng, B.; Lin, C. X.; Ebadian, M. A., International Journal of Heat and Mass
Transfer, 2000, vol.43, 10671078.
[68] Keramida, E. P.; Liakos, H. H.; Founti, M. A.; Boudouvis, A. G.; Markatos, N. C.,
International Journal of Heat and Mass Transfer, 2000, vol.43, 18011809.
[69] Kim, S. H.; Huh, K. Y., International Journal of Heat and Mass Transfer, 2000, vol.43,
12331242.
[70] Baek, S. W.; Byun, D. Y.; Kang, S. J., International Journal of Heat and Mass Transfer,
2000, vol.43, 23372344.
[71] Baek, S. W.; Kim, M. Y.; Kim, J. S., Numerical Heat Transfer, Part B, 1998, vol.34,
419437.
[72] Liu, J.; Shang, H. M.; Chen, Y. S.; Wang, T. S., Numerical Heat Transfer, Part B, 1997,
vol.31, 423439.
[73] Khail, E. E., Modelling of Furnaces and Combustors, Abacus Press, 1982.
[74] Worthy, J.; Rubini, P., Numerical Heat Transfer, Part B, 2005, vol.48, no.3, 235256.
[75] Yilbas, B. S.; Shuja, S. Z.; Budair, M. O., Numerical Heat Transfer, Part A, 2003,
vol.43, no.8, 843865.
[76] Brescianini, C. P.; Delichatsios, M. A., Numerical Heat Transfer, Part A, 2003, vol.43,
no.7, 731751.
[77] Yang, X.; Ma, H., Numerical Heat Transfer, Part B, 2003, vol.43, no.3, 289305.
[78] Abdon, A.; Sundén. B., Numerical Heat Transfer, Part A, 2001, vol.40, no.6, 563578.
[79] Wei, X.; Zhang, J.; Zhou, L., Numerical Heat Transfer, Part B, 2004, vol.45, no.3, 283
300.
[80] Yeh, C. L., Numerical Heat Transfer, Part A, 2007, vol.51, 11871212.
[81] Launder, B. E.; Spalding, D. B., Computer Methods in Applied Mechanics and
Engineering, 1974, vol.3, 269289.
[82] Launder, B. E.; Sharma, B. I., Letters in Heat and Mass Transfer, 1974, vol.1, 131138.
[83] Nagano, Y.; Hishida, M., Journal of Fluids Engineering, 1987, vol.109, 156160.
[84] Gatski, T. B.; Speziale, C. G., Journal of Fluid Mechanics, 1993, vol.254, 5978.
[85] Randriamampianina, A.; Schiestel, R.; Wilson, M., International Journal of Heat and
Fluid Flow, 2004, vol.25, 897914.
[86] Thompson, J. F.; Warsi, Z. U. A.; Mastin, C. W., Numerical Grid Generation, North
Holland, 1985, pp.188271.
[87] Van Doormaal, J. P.; Raithby, G. D., Numerical Heat Transfer, 1984, vol.7, 147163.
[88] Hayase, T.; Humphrey, J. A. C., and R. Grief, Journal of Computational Physics, 1992,
vol.98, 108118.
In: Fluid Mechanics and Pipe Flow ISBN: 9781607410379
Editors: D. Matos and C. Valerio, pp. 205229 © 2009 Nova Science Publishers, Inc.
Chapter 6
CONVECTIVE HEAT TRANSFER IN THE THERMAL
ENTRANCE REGION OF PARALLEL FLOW
DOUBLEPIPE HEAT EXCHANGERS
FOR NONNEWTONIAN FLUIDS
Ryoichi Chiba
1*
, Masaaki Izumi
2
and Yoshihiro Sugano
3
1
Yamagata University, Jonan 4316, Yonezawa 9928510, Japan
2
Ishinomaki Senshu University, Shinmito 1, Minamisakai, Ishinomaki 9868580, Japan
3
Iwate University, Ueda 435, Morioka 0208551, Japan
Abstract
In this chapter, the conjugated Graetz problem in parallel flow doublepipe heat exchangers is
analytically solved by an integral transform method—Vodicka’s method—and an analytical
solution to the fluid temperatures varying along the radial and axial directions is obtained in a
completely explicit form. Since the present study focuses on the range of a sufficiently large
Péclet number, heat conduction along the axial direction is considered to be negligible. An
important feature of the analytical method presented is that it permits arbitrary velocity
distributions of the fluids as long as they are hydrodynamically fully developed. Numerical
calculations are performed for the case in which a Newtonian fluid flows in the annulus of the
double pipe, whereas a nonNewtonian fluid obeying a simple power law flows through the
inner pipe. The numerical results demonstrate the effects of the thermal conductivity ratio of
the fluids, Péclet number ratio and powerlaw index on the temperature distributions in the
fluids and the amount of exchanged heat between the two fluids.
Key words: heat transfer, forced convection, heat exchanger, nonNewtonian fluid, analytical
solution
*
Email address: rchiba@yz.yamagatau.ac.jp. Telephone: +81238263219. Fax: +81238263205
Ryoichi Chiba, Masaaki Izumi and Yoshihiro Sugano 206
Nomenclature
a: radius of outer pipe, m
b: radius of inner pipe, m
B: radius ratio (= b/a)
c: heat capacity, J/[kg·K]
f: dimensionless total diffusivity
g: expansion coefficient
h: heat transfer coefficient, W/[m
2
·K]
J
0
()/J
1
(): the first kind of Bessel functions of order zero/one
k: overall heat transfer coefficient, W/[m
2
·K]
K: constant in the convective term, Eq. (9)
n: the number of partitions
Nu: Nusselt number defined by Eqs. (25) and (26)
Pe : Péclet number ratio (=
I II II I
/[ ( )] u b u a b α α − )
r: radial coordinate, m
R: heat capacity flow rate ratio of fluids defined by Eq. (C4)
T: temperature, K
u: fully developed velocity profile, m/s
u : average velocity, m/s
U: dimensionless velocity (= / u u )
w: mass flow rate, kg/s
x: axial coordinate, m
X: eigenfunction
Y
0
()/Y
1
(): the second kind of Bessel functions of order zero/one
α: thermal diffusivity, m
2
/s
ε: heat exchanger effectiveness defined by Eq. (29)
ε
H
: eddy diffusivity of heat, m
2
/s
γ: eigenvalue
η: dimensionless radial coordinate (= r/a)
λ: thermal conductivity, W/[m·K]
λ : thermal conductivity ratio (= λ
II
/λ
I
)
ν: powerlaw index
θ: dimensionless temperature (=
II I II
0b 0b 0b
( ) /( ) T T T T − − )
ξ: dimensionless axial coordinate (=
I 2 I
/( ) x a u α )
Ψ: the number of heat transfer units defined by Eq. (C3)
Subscripts
0: entrance
b: bulk
i: region number
Convective Heat Transfer in the Thermal Entrance Region... 207
l: eigenvalue number
∞: asymptotic
Superscripts
I: fluid I
II: fluid II
1. Introduction
The usual doublepipe parallel flow heat exchanger consists of two concentric circular
pipes, and fluids with different temperatures enter the annulus and inner pipe at the same side
of the heat exchanger. As the fluids flow through their respective channels, heat is transferred
from the hot fluid to the cold fluid. The traditional methods for predicting heat transfer in
such situations are based on the assignment of heat transfer coefficients for each stream
independently of the actual coupling of the boundary conditions [1]. The methods are
employed under the two essential assumptions that are customarily made [2]: (i) the heat
transfer coefficients are considered to be insensitive to the longitudinal distributions of both
the heat flux and surface temperature, and (ii) they are taken to be uniform irrespective of the
heat exchanger length.
There is substantial evidence [3] that under turbulent flow conditions the above
mentioned methods are acceptable for nonmetallic fluids because the local heat transfer
coefficients are scarcely influenced by thermal boundary conditions in those cases, and the
thermal entrance region usually covers a small part of the heat transfer length. On the other
hand, under laminar flow conditions, which are encountered in low Reynolds number flow
heat exchangers
1
, local heat transfer coefficients become very sensitive to the thermal
boundary condition. In addition, the thermal entrance length can often be of the same order of
magnitude as the heat exchanger length [2]. This leads to spatial variations in heat transfer
coefficients in much of the heat transfer surfaces. Since the two streams are thermally coupled
via the wall separating them, the varying heat transfer coefficients on both surfaces of the
wall, i.e., overall heat transfer coefficient, cannot be defined a priori.
In the case of spatially varying overall heat transfer coefficient, one has to solve the
coupled forced convection heat transfer problem between heating and heated fluids—that is,
the conjugated Graetz problem—for an accurate evaluation of the performance of the heat
exchanger. Since the influence of coupling of the boundary conditions can be important in the
thermal entrance regions, especially with laminar flow, the conjugated Graetz problem for
parallel flow laminar heat exchangers with a relatively short heat transfer length has been
analytically investigated.
The earliest investigations on the heat transfer problems of parallel flow laminar heat
exchangers were conducted by Stein [4, 5]. Subsequently, the same problem was treated by
Mikhailov et al. [6] using a specialized version of the method for conjugated Graetz
problems. Pagliarini et al. [2] theoretically investigated thermal interaction between the
1
Laminar or low Reynolds number turbulent flows are the consequences of either high viscosity fluids, compact
flow passages (i.e., small hydraulic diameter), or low fluid velocities.
Ryoichi Chiba, Masaaki Izumi and Yoshihiro Sugano 208
streams of laminar flow doublepipe heat exchangers, while considering axial heat conduction
in the wall separating the fluids. Neto et al. [7] employed the generalized integral transform
technique, deriving an analytical solution to a mixed lumpeddifferential formulation of
doublepipe heat exchangers. Plaschko [8] used matched asymptotic expansions to obtain an
approximate solution to the heat transfer in parallel flow heat exchangers with high Péclet
numbers. Huang et al. [9] developed a control algorithm of heat flux imposed on the external
surface of the outer pipe in order to obtain the desired thermal entrance length and fluid
temperatures in parallel flow doublepipe heat exchangers.
The problems of heat transfer in laminar cocurrent flow of two immiscible fluids, such
as those studied in [1014], are also governed by the same type of differential equations as
those for parallel flow heat exchangers, and are therefore included in the conjugated Graetz
problem. Bentwich et al. [10] and Nogueira et al. [13] analyzed the temperature distribution
and heat transfer in coreannular twophase liquidliquid flow subject to constant wall
temperature boundary condition. The heat transfer problem of coreannular laminar flow in a
pipe with constant wall heat flux boundary condition was theoretically investigated by Leib et
al. [11]. The same type of problem in a pipe with the thirdkind boundary condition was
studied by Su [14]. Davis et al. [12] examined three types of conjugated boundary value
problems related to conjugated multiphase heat and mass transfer problems, and developed
systematic procedures for their solutions. Simultaneous heat and mass transfer in internal gas
flow in a duct whose walls are coated with a sublimable material [15] and the membrane gas
absorption process [16] are also known to come under the category of conjugated Graetz
problem.
The abovecited papers considered fluids to be Newtonian. In view of the fact that heat
exchange between Newtonian and nonNewtonian fluids is of importance in engineering
applications as well as that between Newtonian fluids, it is desirable to develop a fully
analytical method that is not restricted by the rheology characteristics of fluids. In this
chapter, the conjugated Graetz problem in parallel flow doublepipe heat exchangers is
analytically solved by an integral transform method—Vodicka’s method, and an analytical
solution to the fluid temperatures varying along the radial and axial directions is obtained in a
completely explicit form. Since the present study focuses on the range of sufficiently large
Péclet number, heat conduction along the axial direction is considered to be negligible. An
important feature of the analytical method presented is that it permits arbitrary velocity
distributions, i.e., arbitrary rheology characteristics, of the fluids as long as they are
hydrodynamically fully developed. Numerical calculations are performed for the case in
which a Newtonian fluid flows in the annulus of the double pipe whereas a nonNewtonian
fluid obeying a simple power law flows through the inner pipe. The numerical results
demonstrate the effects of the thermal conductivity ratio of the fluids, Péclet number ratio and
powerlaw index on the temperature distributions in the fluids and the amount of exchanged
heat between the two fluids.
Convective Heat Transfer in the Thermal Entrance Region... 209
2. Theoretical Analysis
2.1. Analytical Model and Mathematical Formulation
Figure 1 shows the physical model and coordinate system. Two fluids, fluid I and fluid II,
flow concurrently with fully developed velocity distributions u
I
(r) and u
II
(r) in a doublepipe
heat exchanger. The heat exchanger consists of two concentric thin pipes: outer pipe of radius
a and inner pipe of radius b. While the fluid II flows through the inner pipe, the fluid I flows
in the annulus between the pipe walls. The fluid temperatures at the entrance are
I
0
T (r) and
II
0
T (r). The surface at r = a is insulated and heat is exchanged between both fluids through the
inner pipe wall.
u=u
I
(r)
T=
I
0
T (r)
x
r
II
I
u=u
II
(r)
T=
II
0
T (r)
Thin concentric pipes
λ
I
, α
I
λ
II
, α
II
a
b
O
Figure 1. Physical model and related cylindrical coordinate system for a parallel flow doublepipe heat
exchanger.
The following assumptions are introduced:
(i) physical properties are independent of temperature and are therefore constant,
(ii) the heat resistance of the pipes is negligible,
(iii) the axial heat conduction and turbulent axial diffusion are negligible,
(iv) molecular and turbulent transport of momentum and heat are additive,
(v) viscous dissipation and compression work are negligibly small,
(vi) turbulent diffusion of heat can be described by available thermal eddy diffusivity
expressions.
The item (i) indicates that no natural convection occurs in the fluids. Since the effects of
pipe wall conductivity on the heat exchange effectiveness are minor in parallel flow heat
exchangers [2], the item (ii) is a reasonable assumption. The item (iii) is valid only for high
Péclet number flow.
In this case, the steadystate heat balance is expressed as follows [3]:
H
( , ) 1 ( , )
( ) [ ( )]
m m
m m m
T x r T x r
u r r r
x r r r
α ε
⎧ ⎫ ∂ ∂ ∂
= +
⎨ ⎬
∂ ∂ ∂
⎩ ⎭
, m = I, II, (1)
Ryoichi Chiba, Masaaki Izumi and Yoshihiro Sugano 210
where α and ε
H
denote the thermal diffusivity and eddy diffusivity of heat, respectively, and
the superscript m indicates the fluid number. The boundary conditions and continuity
condition at r = b are then
0
(0, ) ( )
m m
T r T r = , m = I, II, (2)
II I
( , 0) ( , )
0
T x T x a
r r
∂ ∂
= =
∂ ∂
, (3)
I II
I II
( , ) ( , ) T x b T x b
r r
λ λ
∂ ∂
=
∂ ∂
,
I II
( , ) ( , ) T x b T x b = . (4)
To simplify manipulations, we obtain the dimensionless forms of Eqs. (1)(4):
( ) I
1
( )
(1 )
( ) II
m
m
m
m
m
m
U m
f
B
U Pe m
B
θ
η
ξ θ
η η
η η η
θ
η
ξ
⎧ ∂
=
⎪
∂ ⎡ ⎤ ∂ ∂ ⎪
=
⎨
⎢ ⎥
∂ ∂
− ∂
⎣ ⎦
⎪
=
⎪
∂
⎩
, (5)
0
(0, ) ( )
m m
θ η θ η = , m = I, II, (6)
II I
( , 0) ( ,1)
0
θ ξ θ ξ
η η
∂ ∂
= =
∂ ∂
, (7)
I II
( , ) ( , ) B B θ ξ θ ξ
λ
η η
∂ ∂
=
∂ ∂
,
I II
( , ) ( , ) B B θ ξ θ ξ = , (8)
where we introduce the following dimensionless quantities: / r a η = , / B b a = ,
( ) ( ) /
m m m
U u r u η = ; m = I, II,
I II II I
/ [ ( )] Pe u b u a b α α = − ,
II I
/ λ λ λ = ,
II I II
0b 0b 0b
( ) / ( )
m m
T T T T θ = − − ; m = I, II,
I 2 I
/ ( ) x a u ξ α = ,
H
( ) 1 ( ) /
m m m
f r η ε α = + ; m = I, II.
Equation (5) is a partial differential equation with variable coefficients; therefore, it is
very difficult to obtain the exact solution for arbitrary velocity profiles and total diffusivity
distribution. In order to solve Eq. (5), we divide the annular and circular channels into n
I
and
n
II
regions in the radial direction (ηaxis direction), respectively, and approximate f
m
(η) and
U
m
(η) as constants f
i
and U
i
in each region [17]. In this case, the dimensionless energy
equation in the ith region (i = 1, 2,..., n
I
+n
II
) is:
Convective Heat Transfer in the Thermal Entrance Region... 211
2
2
1
i i i
i
K
θ θ θ
ξ η η η
∂ ∂ ∂
= +
∂ ∂ ∂
, (9)
where
II
II I II
(1 ) /( ) 1
/ 1
i i
i
i i
U Pe B B f i n
K
U f n i n n
⎧
− ⋅ ≤ ≤ ⎪
=
⎨
+ ≤ ≤ +
⎪
⎩
.
Note that the subscripts of θ, K, λ and η denote the region number, not the fluid number.
The continuity conditions at the imaginary interfaces (including the real interface at η = B)
and boundary conditions are expressed as:
1
1
( , ) ( , )
i i i i i
i
λ θ ξ η θ ξ η
λ η η
+
+
∂ ∂
=
∂ ∂
,
1
( , ) ( , )
i i i i
θ ξ η θ ξ η
+
= , (10)
0
(0, ) ( )
m
i
θ η θ η = , if 1 ≤ i ≤ n
II
, then m = II,
if n
II
+1 ≤ i ≤ n
I
+n
II
, then m = I, (11)
I II
( )
1
( ,1)
( , 0)
0
n n
θ ξ
θ ξ
η η
+
∂
∂
= =
∂ ∂
, (12)
where η
i
is the outer radius of the ith region, that is,
II
n
B η = and
I II
( )
1
n n
η
+
= . The initial
boundary value problem expressed by Eqs. (9)(12) is identical to that for transient heat
conduction case in a composite medium consisting of n
I
+n
II
layers.
2.2. Derivation of Analytical Solution
We employ Vodicka’s method [18] to derive an analytical solution of temperatures in the
fluids and their related Nusselt numbers. Since it allows us to easily analyze the transient heat
conduction in a composite medium with a large number of subregions, this method is applied
to the analysis of temperature fields in a wide range of objects [17, 19]. Although it is
possible, in principle, to analyze the problem under consideration using Laplace transform
[20], the inverse transform is mathematically quite complicated for a large number of sub
regions. Therefore, it is the authors opinion that Vodicka’s method is to be preferred due to its
advantage in facility. Note that the analytical treatment proposed by Mikhailov et al. [21] may
be used as an alternative.
Using Vodicka’s method, the solution to Eqs. (9)(12) is assumed to be
1
( , ) ( ) ( )
i l il
l
X θ ξ η φ ξ η θ
∞
∞
=
= +
∑
, (13)
Ryoichi Chiba, Masaaki Izumi and Yoshihiro Sugano 212
where θ
∞
is the temperature of the fluids for ξ → ∞. An overall heat balance shows that with
an infinite heat transfer area, the outlet temperature of each stream in parallel flow is
1
1
B
Pe B B
θ
λ
∞
+
=
+ +
. (14)
The eigenfunction ( )
il
X η is obtained as
0 0
( ) ( ) ( )
il il i l il i l
X A J K B Y K η γ η γ η = + , (15)
where J
0
() and Y
0
() are the first and second kind of Bessel functions of order 0, respectively.
The constants A
il
and B
il
in Eq. (15) are determined from the conditions that are obtained
by substituting Eqs. (13) and (15) into the continuity and boundary conditions, Eqs. (10) and
(12). Consequently, the continuity conditions of the temperature field at the interfaces
between neighboring regions can be fulfilled. The eigenvalues γ
l
(l = 1, 2,...) are obtained
from the condition under which all the A
il
and B
il
values are nonzero and are therefore
positive roots of the following transcendental equation (for details, see the appendix A):
I II I II
1 2
( 1) ( )
const.
0
n n n n + − +
⎡ ⎤
⋅ ⋅ =
⎢ ⎥
⎣ ⎦
E E E b , (16)
where
1 i i i +
= ⋅ E C D ,
I II
I II
I II
1
( )
1
( )
( )
l
n n
n n
l
n n
Y K
J K
γ
γ
+
+
+
⎡ ⎤
⎢ ⎥
=
⎢ ⎥
−
⎣ ⎦
b ,
1 1 0
1 1 0
/ ( ) ( )
/ ( ) ( )
i i i l i l i i l i
i
i i i l i l i i l i
K Y K Y K
K J K J K
λ λ γ γ η γ η
λ λ γ γ η γ η
+
+
⎡ ⎤
− −
= ⎢ ⎥
⎢ ⎥
⎣ ⎦
C ,
0 1 0 1
1
1 1 1 1 1 1
( ) ( )
( ) ( )
i l i i l i
i
i l i l i i l i l i
J K Y K
K J K K Y K
γ η γ η
γ γ η γ γ η
+ +
+
+ + + +
⎡ ⎤
= ⎢ ⎥
− − ⎢ ⎥
⎣ ⎦
D , (17)
and J
1
() and Y
1
() are the first and second kind of Bessel functions of order 1, respectively.
By substituting Eq. (13) into Eq. (11), the following equation is obtained:
0
1
( ) (0) ( ) ( )
m
l il
l
G X η φ η θ η θ
∞
∞
=
= = −
∑
, m = I or II. (18)
Convective Heat Transfer in the Thermal Entrance Region... 213
The eigenfunction X
il
(η) given by Eq. (15) has an orthognal relationship with
discontinuous weight functions; it is expressed as follows (see the appendix B):
I II
1
1
const. ( )
( ) ( )d
0 ( )
i
i
n n
i i il ik
i
l k
K X X
l k
η
η
λ η η η η
−
+
=
= ⎧
=
⎨
≠
⎩
∑
∫
. (19)
G(η) can be expanded into an infinite series of X
il
(η) as follows:
1
( ) ( )
l il
l
G g X η η
∞
=
=
∑
, (20)
where the expansion coefficient g
l
is given by
[ ]
I II
1
I II
1
1
2
1
( ) ( )d
( ) d
i
i
i
i
n n
i i il
i
l
n n
i i il
i
K G X
g
K X
η
η
η
η
λ η η η η
λ η η η
−
−
+
=
+
=
=
∑
∫
∑
∫
. (21)
Taking Eq. (15) into account, we substitute Eq. (13) into Eq. (9). This yields a firstorder
linear ordinary differential equation for φ
l
(ξ) as follows:
2
d
0
d
l
l l
φ
γ φ
ξ
+ = . (22)
By solving Eq. (22) with the condition φ
l
(0) = g
l
, which is obtained from the comparison
between Eqs. (18) and (20), we obtain φ
l
(ξ) as
2
( )
l
l l
g e
γ ξ
φ ξ
−
= . (23)
Finaly, the dimensionless temperature in the ith region η ∈ [η
i−1
, η
i
] inside the double
pipe heat exchanger, θ
i
(ξ, η), is derived as:
2
0 0
1
1
( , ) ( ) ( )
1
l
i l il i l il i l
l
B
g e A J K B Y K
Pe B B
γ ξ
θ ξ η γ η γ η
λ
∞
−
=
+
⎡ ⎤
= + +
⎣ ⎦
+ +
∑
. (24)
Local Nusselt numbers for outer and inner streams are defined in the usual manner as
I I
I
I I
b
2 ( ) 2( 1)
B B
h a b B
Nu
η η
θ
λ θ θ η
= =
− − ∂
= =
− ∂
for the outer stream, (25)
Ryoichi Chiba, Masaaki Izumi and Yoshihiro Sugano 214
II II
II
II II
b
2 2
B B
h b B
Nu
η η
θ
λ θ θ η
= =
∂
= =
− ∂
for the inner stream. (26)
3. Numerical Calculation
In this study, there is no analytical restriction on the form of the velocity profiles of the
flowing fluids. Thus, as a numerical example, a doublepipe heat exchanger in which a
Newtonian fluid flows in the annulus whereas a nonNewtonian fluid flows through the inner
pipe is considered here. The nonNewtonian fluid is assumed to obey a power law, which can
approximate the nonNewtonian viscosity of many types of fluids with good accuracy over a
wide range of shear rates. The fluids have mutually different uniform temperatures at the
entrance, that is,
I
0
θ (η) = 1 and
II
0
θ (η) = 0. Under laminar flow conditions,
I II
H H
0 ε ε = = and
the fully developed velocity profile for each fluid is [1, 22]:
2 2
I I
2 2
(1 ) ln ( 1) ln
( ) 2
( 1) ( 1) ln
B B
u u
B B B
η η
η
− + −
=
− − +
, (27)
1
II II
3 1
( ) 1
1
u u
B
ν
ν
ν η
η
ν
+
⎡ ⎤
+ ⎛ ⎞
⎢ ⎥
= −
⎜ ⎟
⎢ ⎥ +
⎝ ⎠
⎣ ⎦
, (28)
where ν is the powerlaw index. For values of ν less than unity, the behavior is
pseudoplastic, whereas forν greater than unity the behavior is dilatant. For ν = 1, it reduces to
Newton’s law of viscosity.
In the numerical calculations, we radially divide the each channel into equal intervals
with the number of partitions n
I
= n
II
= 20. Our ealier work [17] showed that the number of
partitions 20 per a channel provides sufficiently accurate results. The number of terms in the
infinite series in Eq. (24) is taken as 200. It should be noted that this value is used for the
verification of sufficient convergence of the numerical results.
4. Results and Discussion
To show the accuracy of the present analytical solution, we first compare local Nusselt
numbers, bulk temperatures and wall temperatures to existing results [23, 24]. The values
shown in Table 1 are in good agreement between the present analytical solution and the
existing solutions: the similarity approach solution [23] and the eigenvalue solution given in
[24].
Table 1. A comparison of the entrance region results obtained by the present analytical method and existing methods [23, 24] for a
parallel flow doublepipe heat exchanger with ν = 1, Pe = 2/3, λ = 2 and B = 0.5
x [23, 24] ξ
II
B
θ
II
B
θ * θ
η = B
θ
η = B
* Nu
I
Nu
I
* Nu
II
Nu
II
*
0.001 0.00016667 0.010589 0.0103 0.43057 0.426 24.123 23.96 16.224 16.42
0.002 0.00033333 0.016574 0.0163 0.43345 0.431 19.215 19.15 12.892 12.98
0.01 0.0016667 0.047178 0.0470 0.44917 0.449 11.560 11.56 7.6210 7.64
0.02 0.0033333 0.073899 0.0737 0.46062 0.460 9.4119 9.42 6.1655 6.17
0.04 0.0066667 0.11529 0.115 0.47611 0.476 7.7722 7.78 5.0914 5.09
0.1 0.016667 0.20501 0.205 0.50591 0.506 6.2719 6.28 4.1998 4.20
0.2 0.033333 0.31095 0.311 0.54061 0.541 5.6245 5.63 3.9329 3.93
0.4 0.066667 0.45267 0.453 0.59330 0.594 5.3742 5.38 3.9267 3.92
0.6 0.10000 0.54102 0.541 0.62927 0.629 5.3390 5.35 3.9403 3.94
* Existing solutions, the data cited from [23] for 0.01 x ≤ and [24] for 0.02 x ≥ .
Ryoichi Chiba, Masaaki Izumi and Yoshihiro Sugano 216
0
0.2
0.4
0.6
0.8
1
10
4
10
3
10
2
10
1
10
0
ν =0.3
3
θ
ξ
Fluid I
Fluid II
Inner pipe wall
η =
0,
0.2,
0.4,
0.5,
0.6,
0.8,
1
(a)
0
0.2
0.4
0.6
0.8
1
10
4
10
3
10
2
10
1
10
0
ν =0.3
3
θ
ξ
Inner pipe wall
Fluid II
Fluid I
η =
0,
0.2,
0.4,
0.5,
0.6,
0.8,
1
(b)
Figure 2. Continued on next page.
Convective Heat Transfer in the Thermal Entrance Region... 217
0
0.2
0.4
0.6
0.8
1
10
4
10
3
10
2
10
1
10
0
ν =0.3
3
θ
ξ
Inner pipe wall
Fluid I
η =
0,
0.2,
0.4,
0.5,
0.6,
0.8,
1
(c)
Figure 2. Temperature variation along the streamwise direction at different radial locations with B = 0.5
and λ =10, for (a) Pe = 0.5, (b) Pe = 2 and (c) Pe = 10.
0
0.5
1
1.5
2
2.5
3
0 0.2 0.4 0.6 0.8 1
ν =3
1
0.3
U
I
I
η/B
Figure 3. Velocity profiles for different powerlaw index values.
Figure 2 shows the axial distributions of dimensionless temperature θ at some radial
locations. In this figure, the solid lines and dashed lines represent the cases for ν = 0.3 and ν
= 3, respectively, for which fluid II exhibits the velocity profiles shown in Figure 3. It is
observed that for a constant thermal conductivity ratio λ , a greater Péclet number ratio Pe
produces a larger temperature variation in fluid I. Moreover, although we omit the graphical
representation, for a constant Pe a greater λ causes a larger temperature variation in fluid I.
These results can be predicted easily from Eq. (14). With regard to the effect of ν on the
Ryoichi Chiba, Masaaki Izumi and Yoshihiro Sugano 218
temperature distributions, an increase in ν increases the values of θ in both fluids throughout
the heat exchanger length except the vicinity of the inner pipe axis. Especially, the effect is
pronounced at η = B, or the inner pipe wall, and it decreases with distance from the inner pipe
wall. Not surprisingly, the temperatures of fluid II, a powerlaw fluid, are directly affected by
ν. However, we can also observe a slight influence of ν on fluid I in the temperature
distributions.
0
0.2
0.4
0.6
0.8
1
10
4
10
3
10
2
10
1
10
0
ν =0.3
3
θ

η
=
B
ξ
Pe =0.1, 0.2, 0.5, 1, 2, 5, 10
(a)
0
0.2
0.4
0.6
0.8
1
10
4
10
3
10
2
10
1
10
0
ν =0.3
3
θ

η
=
B
ξ
Pe =0.1, 0.2, 0.5, 1, 2, 5, 10
(b)
Figure 4. Continued on next page.
Convective Heat Transfer in the Thermal Entrance Region... 219
0
0.2
0.4
0.6
0.8
1
10
4
10
3
10
2
10
1
10
0
ν =0.3
3
θ

η
=
B
ξ
Pe =0.1, 0.2, 0.5, 1, 2, 5, 10
(c)
Figure 4. Temperature variation along the streamwise direction at the inner pipe wall with B = 0.5, for
(a) λ = 0.1, (b) λ = 1 and (c) λ = 10.
The temperatures of the inner pipe wall, which are the most susceptible to the effect of ν,
are plotted against ξ in Figure 4. As λ becomes small, θ of the inner pipe wall increases. In
addition, the larger the value of ν is, the higher θ always becomes. It is interesting that the
variation behavior of θ along the ξaxis direction depends on the value of Pe ; while θ of the
inner pipe wall increases monotonically for a small Pe , it may have a maximal value for a
large Pe , decreasing from a certain axial location.
This is due to the following reasons: in the immediate vicinity of the entrance of the heat
exchanger, large hot/cold heat flows from fluid I to fluid II because of great radial
temperature gradients near the inner pipe wall. This leads to an increase in θ of the inner pipe
wall. Additionally, since the temperature variation in fluid II becomes smaller than that in
fluid I for a large Pe (see Figure 2c), fluid I greatly varies its temperature over the full
section of the annulus along the axial direction. In contrast, fluid II varies its temperature only
in the immedeate vicinity of the inner pipe wall. Up to a downstream location, θ of the inner
pipe wall remains elevated due to high heat flux passing through the wall. When the fluids
reach downstream to some extent, the heat flux passing through the wall, i.e., radial
temperature gradients decrease and meanwhile heat conduction in the negative η direction
gradually increases in fluid II. As a result, the temperature of the inner pipe wall begins to
decrease and then it converges at the asymptotic temperature given by Eq. (14).
It is known that in parallel flow heat exchangers, the bulk temperatures of heating and
heated fluids vary monotonically from each entrance temperature to the asymptotic one.
However, local temperatures, especially those near the inner pipe wall, do not necessarily
Ryoichi Chiba, Masaaki Izumi and Yoshihiro Sugano 220
vary in a monotone along the streamwise direction and may have a maximal value at a certain
axial location. This phenomenon tends to occur in the case of large Péclet number ratio.
0
5
10
15
20
25
30
10
4
10
3
10
2
10
1
10
0
ν =3
1
0.3
Const. heat flux B.C.
Const. temperature B.C.
N
u
I
ξ
(a)
0
5
10
15
20
25
30
10
4
10
3
10
2
10
1
10
0
ν =3
1
0.3
Const. heat flux B.C.
Const. temperature B.C.
N
u
I
I
ξ
(b)
Figure 5. Local Nusselt numbers of (a) outer stream and (b) inner stream for B = 0.5, λ = 1 and Pe = 1.
Convective Heat Transfer in the Thermal Entrance Region... 221
The local Nusselt numbers defined by Eqs. (25) and (26) are shown in Figure 5(a) and
(b), respectively. The discrete plots in Figure 5 indicate the local Nusselt numbers of single
flows for a Newtonian fluid with constant temperature/constant heat flux boundary condition.
The data are cited from [25] for flow in the pipe and [26, 27] for flow in the annulus. Figure 5
demonstrates that the local Nusselt number of the outer stream (fluid I) is little affected by ν.
On the other hand, the Nusselt number of the inner stream (fluid II) depends on the value of
ν, increasing with a decrease in ν. This trend can be also found in the case of single flow [28].
In summary, the local Nusselt number of each stream in parallel flow heat exchangers is
determined by own rheology characteristics, irrespective of the rheology characteristics of
counterpart.
Next, we compare the curve for ν = 1 with the discrete plots in both streams. In the inner
stream, the local Nusselt number obtained from the present numerical calculations falls
somewhere in between the two local Nusselt numbers for single flow, being, on the whole,
closer to the one for constant temperature boundary condition. On the other hand, in the outer
stream, the Nusselt number obtained from the present numerical calculations is smaller than
both Nusselt numbers for single flow; it is closer to the one for constant temperature
boundary condition again. However, it is found difficult to accurately approximate local
Nusselt numbers in parallel flow heat exchangers with those for constant temperature/heat
flux boundary condition.
0
0.2
0.4
0.6
0.8
1
10
4
10
3
10
2
10
1
10
0
ν =0.3
3
θ
I
b
,
θ
I
I
b
ξ
θ
I
b
θ
II
b
(a)
Figure 6. Continued on next page.
Ryoichi Chiba, Masaaki Izumi and Yoshihiro Sugano 222
0
0.2
0.4
0.6
0.8
1
10
5
10
4
10
3
10
2
10
1
10
0
ν =0.3
3
θ
I
b
,
θ
I
I
b
ξ
θ
I
b
θ
II
b
(b)
Figure 6. Bulk temperatures of the heating and heated fluids with different powerlaw index values for
B = 0.5, λ = 1 and (a) Pe = 1 and (b) Pe = 10.
Figure 6 shows the axial variations in the bulk temperatures of heating and heated fluids.
Since a smaller ν leads to a larger Nusselt number in the inner stream, as shown in Figure 5
(b), heat exchange between both fluids is found to be enhanced for ν = 0.3. From Figure 6(a),
maximum difference between the bulk temperatures for ν = 0.3 and 3 is calculated to be
approximately 1% for
I
b
θ and 10% for
II
b
θ , and from Figure 6(b) approximately 7% for both
I
b
θ and
II
b
θ .
Figure 7 makes comparisons of the heat exchanger effectiveness
1
ε defiined by
I
b
1 ( )
( )
1
θ ξ
ε ξ
θ
∞
−
=
−
, (29)
between the value by the present analytical solution and that calculated by means of the bulk
temperatures of the fluids and a constant overall heat transfer coefficient (measured by overall
Nusselt number in the figure). θ
∞
in Eq. (29) is already given by Eq. (14). For the derivation
details of the latter heat exchanger effectiveness, q.v. the appendix C. Figure 7 demonstrates
that it is impossible to approximate well the heat exchanger effectiveness throughout all
2
The heat exchanger effectiveness is defined as the ratio of the actual overall rate of heat transfer to the maximum
possible as computed for an exchanger with the same operating conditions but with infinite heat transfer area
[29].
Convective Heat Transfer in the Thermal Entrance Region... 223
values of ξ with any constant overall heat transfer coefficients. Hence, in designing heat
exchangers with laminar flow, i.e., low Reynolds number flow heat exchangers, it is crucial to
consider streamwise variations in the overall heat transfer coefficient.
0
0.2
0.4
0.6
0.8
1
10
4
10
3
10
2
10
1
10
0
10
1
Present analytical solution
Bulk temp. & const. overall Nu
ε
ξ
Overall Nu =1
2
3
4
5
(a)
0
0.2
0.4
0.6
0.8
1
10
4
10
3
10
2
10
1
10
0
10
1
Present analytical solution
Bulk temp. & const. overall Nu
ε
ξ
Overall Nu =2
4
6
8
10
(b)
Figure 7. Axial variation of heat exchanger effectiveness with B = 0.5, Pe = 1 and ν = 1, for (a) λ = 1
and (b) λ = 10.
Ryoichi Chiba, Masaaki Izumi and Yoshihiro Sugano 224
5. Conclusion
In this chapter, the conjugated Graetz problem in parallel flow doublepipe heat
exchangers has been analytically solved by an integral transform method—Vodicka’s
method—and an analytical solution to the fluid temperatures varying along the radial and
axial directions has been obtained in a completely explicit form. An important feature of the
analytical method presented is that it permits arbitrary velocity distributions of the fluids as
long as they are hydrodynamically fully developed. Numerical calculations have been
performed for the case in which a Newtonian fluid flows in the annulus of the double pipe,
whereas a nonNewtonian fluid obeying a simple power law flows through the inner pipe.
The numerical results have demonstrated the effects of the thermal conductivity ratio of the
fluids, Péclet number ratio and powerlaw index, on the temperature distributions in the fluids
and the amount of exchanged heat between the two fluids.
Acknowledgements
The authors would like to thank Dr. S. Jian, Universidade Federal do Rio de Janeiro, for
his useful comments. Thanks are also given to Prof. A. HajiSheikh, the University of Texas
at Arlington, for helpful suggestions.
Reviewed by
This manuscript was peerreviewed by Dr. Su Jian, Universidade Federal do Rio de
Janeiro, Brazil.
Appendix A: The Eigenvalue Equation
The eigenfunctions X
il
(i = 1, 2,..., n
I
+n
II
) must satisfy
2
1
( ) ( ) ( ) 0
il il i l il
X X K X η η γ η
η
′′ ′ + + = , (A1)
with the continuity and boundary conditions
( 1)
( ) ( )
il i i l i
X X η η
+
= , (A2)
1 ( 1)
( ) ( )
i il i i i l i
X X λ η λ η
+ +
′ ′ = , (A3)
1
(0) 0
l
X′ = , (A4)
I II
( )
(1) 0
n n l
X
+
′ = , (A5)
Convective Heat Transfer in the Thermal Entrance Region... 225
where primes denote differentiation with respect to η. Substituting Eq. (15) into Eqs. (A2)
(A5) yields
0 1 0 1
0 0
0 1 1 0 1 1 0 2 1 0 2 1
1 0 1 1 1 0 1 1 2 0 2 1 2 0 2 1
0 2 2 0 2 2 0 3 2 0 3 2
2 0 2 2 2 0 2
( ) ( ) 0 0
( ) ( ) ( ) ( ) 0 0
( ) ( ) ( ) ( ) 0 0
0 0 ( ) ( ) ( ) ( )
( ) (
l l
l l l l
l l l l
l l l l
l
J K Y K
J K Y K J K Y K
J K Y K J K Y K
J K Y K J K Y K
J K Y K
η η
γ η γ η
γ η γ η γ η γ η
λ γ η λ γ η λ γ η λ γ η
γ η γ η γ η γ η
λ γ η λ γ
= =
′ ′
− −
′ ′ ′ ′ − −
− −
′ ′
2 3 0 3 2 3 0 3 2
) ( ) ( ) 0 0
0
l l l
J K Y K η λ γ η λ γ η
⎡ ⎤
⎢ ⎥
⎢ ⎥
⎢ ⎥
⎢ ⎥
⎢ ⎥
⎢ ⎥
⎢ ⎥
′ ′ − − ⎢ ⎥
⎢ ⎥
⎢ ⎥
⎢ ⎥
⎢ ⎥
⎣ ⎦
I II
I II
1
1
2
2
( )
( )
0
0
0
0
0
0
l
l
l
l
n n l
n n l
A
B
A
B
A
B
+
+
⎧ ⎫
⎧ ⎫
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪ ⎪ ⎪
⋅ =
⎨ ⎬ ⎨ ⎬
⎪ ⎪ ⎪ ⎪
⎪ ⎪ ⎪ ⎪
⎪ ⎪ ⎪ ⎪
⎪ ⎪ ⎪ ⎪
⎪ ⎪ ⎪ ⎪
⎩ ⎭
⎩ ⎭
(A6)
Since Eq. (A6) is a homogeneous system of equations, the determinant of the matrix must
be zero to have a nontrivial solution for all A
il
and B
il
. However, the procedure for obtaining
the determinant is complicated. Therefore, we perform the manipulation shown below.
The boundary condition Eq. (A5) is expressed as:
I II
( )
0
n n +
⋅ = G a , (A7)
where
I II I II
1 1
( ) ( )
l l
n n n n
J K Y K γ γ
+ +
⎡ ⎤
=
⎣ ⎦
G ,
I II I II I II
T
( ) ( ) ( ) n n n n l n n l
A B
+ + +
⎡ ⎤
=
⎣ ⎦
a and the
superscript T is the transpose operator. The continuity conditions Eqs. (A2) and (A3) are
expressed as:
1 1 i i i i + +
⋅ = ⋅ F a D a , (A8)
where
0 0
1 1 1 1
( ) ( )
/ ( ) / ( )
i l i i l i
i
i i i l i l i i i i l i l i
J K Y K
K J K K Y K
γ η γ η
λ λ γ γ η λ λ γ γ η
+ +
⎡ ⎤
= ⎢ ⎥
− − ⎢ ⎥
⎣ ⎦
F , (A9)
and D
i+1
is given in Eq. (17). With
1
1 i i i
−
+
= ⋅ H F D , Eq. (A8) can be written as
I II I II
1
1 ( )
i i i
n n n n
+
+ − +
= ⋅ ⋅ a H H H a . (A10)
Ryoichi Chiba, Masaaki Izumi and Yoshihiro Sugano 226
From Eq. (A7), we obtain
I II
I II I II
I II
1
T
( ) ( )
1
( )
1
( )
l
n n
n n n n l
l
n n
J K
A
Y K
γ
γ
+
+ +
+
⎡ ⎤
⎢ ⎥ = −
⎢ ⎥
⎣ ⎦
a . (A11)
Since the constant B
1l
must be zero due to the nature of Bessel functions
2
,
[ ]
T
1 1
0
l
A = a . This leads to the following equation, considering Eqs. (A10) and (A11):
I II
I II I II
I II
1
1
1 2
1 ( )
1
1
( )
0
( )
l
l
n n
n n n n l
l
n n
A
J K
A
Y K
γ
γ
+
+ − +
+
⎡ ⎤
⎢ ⎥
⎡ ⎤
= ⋅ ⋅
⎢ ⎥
⎢ ⎥
−
⎣ ⎦
⎢ ⎥
⎢ ⎥
⎣ ⎦
H H H . (A12)
In order that all the A
il
and B
il
(i = 1, 2,..., n
I
+n
II
) are nonzero, the eigenvalues γ
l
must
satisfy Eq. (A12) with
I II
( )
0
n n l
A
+
≠ . Here one can write as H
i
= C
i
·D
i+1
/det F
i
= E
i
/det F
i
(C
i
and E
i
are given in Eq. (17)), and therefore obtain Eq. (16).
Appendix B: Orthogonality of the Eigenfunctions
The orthogonality of the eigenfunctions X
il
(i = 1, 2,..., n
I
+n
II
) will now be established.
Equation (A1) is written for l = j and then for l = k with j ≠ k. The equation for l = j is
multiplied by ηX
ik
, and the equation for l = k is multiplied by ηX
ij
. The two resulting
equations are subtracted, multiplied by λ
i
, and then integrated between η = η
i−1
and η = η
i
.
The following results:
1
1
2 2
( ) d ( ) ( ) ( ) ( )
i
i
i
i
i i k j ij ik i ik ij i ij ik
K X X X X X X
η
η
η
η
λ γ γ η η λη η η λη η η
−
−
′ ′ ⎡ ⎤ − = −
⎣ ⎦ ∫
, (B1)
where primes denote differentiation with respect to η. Taking the summation of both sides of
Eq. (B1) for i = 1, 2,..., n
I
+n
II
,
{
I II I II
1
2 2
1 1
( ) d ( ) ( ) ( ) ( )
i
i
n n n n
i i k j ij ik i i ik i ij i ij i ik i
i i
K X X X X X X
η
η
λ γ γ η η λ η η η η η
−
+ +
= =
′ ′ ⎡ ⎤ − = −
⎣ ⎦
∑ ∑
∫
}
1 1 1 1 1
( ) ( ) ( ) ( )
i ik i ij i ij i ik i
X X X X η η η η η
− − − − −
′ ′ ⎡ ⎤ − −
⎣ ⎦
. (B2)
2
In other words, X
1l
(η) must have a finite value at η = 0.
Convective Heat Transfer in the Thermal Entrance Region... 227
Equations (A2) and (A3) are written for l = j and then for l = k, respectively. The
equation (A2) for l = j is multiplied by the equation (A3) for l = k at each side, and similarly
the equation (A2) for l = k is multiplied by the equation (A3) for l = j. The two resulting
equations make Eq. (B2) a simpler form:
I II
1
2 2
1
( ) d
i
i
n n
i i k j ij ik
i
K X X
η
η
λ γ γ η η
−
+
=
− =
∑
∫
I II I II I II I II I II
( ) ( ) ( ) ( ) ( )
(1) (1) (1) (1)
n n n n k n n j n n j n n k
X X X X λ
+ + + + +
⎡ ⎤
′ ′ −
⎣ ⎦
. (B3)
Considering Eq. (A5), Eq. (B3) reduces to
I II
1
1
d 0
i
i
n n
i i ij ik
i
K X X
η
η
λ η η
−
+
=
=
∑
∫
. (B4)
For the case of j = k = l, Eq. (A1) leads to a normalizing factor defined as the
denominator of Eq. (21).
Appendix C: Heat Exchanger Effectiveness for a Constant
Overall Heat Transfer Coefficient
Ignoring the radial distributions of velocity and temperature of the fluids, we derive the
heat exchanger effectiveness by means of their crosssectional average values and a constant
overall heat transfer coefficient. The bulk temperatures of the fluids at the axial location x in
the parallel flow heat exchanger shown in Figure 1 can be written in dimensionless forms as
[30]
( 1) ( )
I
b
1
( )
1
R
R e
R
ξ
θ ξ
− + ⋅Ψ
+ ⋅
=
+
, (C1)
I
II b
b
1 ( )
( )
R
θ ξ
θ ξ
−
= , (C2)
where Ψ(ξ) and R are the number of heat transfer units and heat capacity flow rate ratio
which occur frequently in heat exchanger analysis, respectively; they are given by
II II
2 2
( )
(1 )
k b x Nu
w c Pe B
π
ξ ξ
λ
⋅ ⋅ ⋅
Ψ = =
⋅ −
, (C3)
Ryoichi Chiba, Masaaki Izumi and Yoshihiro Sugano 228
II II
I I
1
w c Pe B
R
w c B
λ ⋅ ⋅
= =
+
. (C4)
Nu in Eq. (C3) is the overall Nusselt number based on the outer stream properties being
defined as Nu = ka/λ
I
and k denotes the usual overall heat transfer coefficient of the inner pipe
wall. Consequently, the heat exchanger effectiveness ε expressed by Eq. (29) is obtained as a
function of the heat exchanger length ξ as follows:
( 1) ( )
( ) 1
R
e
ξ
ε ξ
− + ⋅Ψ
= − . (C5)
References
[1] Mikhailov, M. D., and Ozisik, M. N.: Unified Analysis and Solutions of Heat and Mass
Diffusion, Dover Publications, New York (1994)
[2] Pagliarini, G., and Barozzi, G. S.: Thermal Coupling in Laminar Flow DoublePipe Heat
Exchangers. Trans ASME Journal of Heat Transfer, 113, 526534 (1991)
[3] Stein, R. P.: Liquid Metal Heat Transfer. In: Irvine, T. F., Hartnett, J. P. (eds.) Advances
in Heat Transfer, Vol. 3, Academic Press, New York, London (1966)
[4] Stein, R. P.: Heat Transfer Coefficients in Liquid Metal Concurrent Flow Double Pipe
Heat Exchangers. Chemical Engineering Progress Symposium Series, 59, 6475 (1965)
[5] Stein, R. P.: The Graetz Problem in Concurrent Flow Double Pipe Heat Exchangers.
Chemical Engineering Progress Symposium Series, 59, 7687 (1965)
[6] Mikhailov, M. D., and Shishedjiev, B. K.: Coupled at boundary mass or heat transfer in
entrance concurrent flow. International Journal of Heat and Mass Transfer, 19, 553557
(1976)
[7] Neto, F. S., and Cotta, R. M.: Lumpeddifferential analysis of concurrent flow double
pipe heat exchanger. Canadian Journal of Chemical Engineering, 70, 592595 (1992)
[8] Plaschko, P.: High Péclet number heat exchange between cocurrent streams. Archive of
Applied Mechanics (Ingenieur Archiv), 70, 597611 (2000)
[9] Huang, C. H., and Yeh, C. Y.: An optimal control algorithm for entrance concurrent
flow problems. International Journal of Heat and Mass Transfer, 46, 10131027 (2003)
[10] Bentwich, M., and Sideman, S.: Temperature distribution and heat transfer in annular
twophase (liquidliquid) flow. Canadian Journal of Chemical Engineering, 42, 913
(1964)
[11] Leib, T. M., Fink, M., and Hasson, D.: Heat transfer in vertical annular laminar flow of
two immiscible liquids. International Journal of Multiphase Flow, 3, 533549 (1977)
[12] Davis, E. J., Venkatesh, S.: The solution of conjugated multiphase heat and mass
transfer problems. Chemical Engineering Science, 34, 775787 (1979)
[13] Nogueira, E., Cotta, R. M.: Heat transfer solutions in laminar cocurrent flow of
immiscible liquids. Heat and Mass Transfer, 25, 361367 (1990)
[14] Su, J.: Exact Solution of Thermal Entry Problem in Laminar Coreannular Flow of Two
Immiscible Liquids. Chemical Engineering Research and Design, 84, 10511058 (2006)
[15] Sparrow, E. M., and Spalding, E. C.: Coupled laminar heat transfer and sublimation
mass transfer in a duct. Trans ASME Journal of Heat Transfer, 90, 115124 (1968)
Convective Heat Transfer in the Thermal Entrance Region... 229
[16] Wang, W. P., Lin, H. T., and Ho, C. D.: An analytical study of laminar cocurrent flow
gas absorption through a parallelplate gasliquid membrane contactor. Journal of
Membrane Science, 278, 181189 (2006)
[17] Chiba, R., Izumi, M., and Sugano, Y.: An analytical solution to nonaxisymmetric heat
transfer with viscous dissipation for nonNewtonian fluids in laminar forced flow.
Archive of Applied Mechanics (Ingenieur Archiv), 78, 6174 (2008)
[18] Vodicka, V.: Linear heat conduction in laminated bodies (in German). Mathematische
Nachrichten, 14, 4755 (1955)
[19] Chiba, R.: Stochastic heat conduction analysis of a functionally graded annular disc with
spatially random heat transfer coefficients. Applied Mathematical Modelling, 33, 507
523 (2009)
[20] Carslaw, H. S., and Jaeger, J. C.: Conduction of Heat in Solids, 2nd Edition, Clarendon
Press, Oxford (1986)
[21] Mikhailov, M. D., Ozisik, M. N., and Vulchanov, N. L.: Diffusion in composite layers
with automatic solution of the eigenvalue problem. International Journal of Heat and
Mass Transfer, 26, 11311141 (1983)
[22] Bird, R. B., Stewart, W. E., and Lightfoot, E. N.: Transport Phenomena, John Wiley &
Sons Inc, New York (1960)
[23] Gill, W. N., Porta, E. W., and Nunge, R. J.: Heat transfer in thermal entrance region of
cocurrent flow heat exchangers with fully developed laminar flow. International Journal
of Heat and Mass Transfer, 11, 14081412 (1968)
[24] Nunge, R. J., and Gill, W. N.: An analytical study of laminar counterflow doublepipe
heat exchangers. AIChE Journal, 12, 279289 (1966)
[25] Shah, R. K.: Thermal entry length solutions for the circular tube and parallel plates. 3rd
National Heat Mass Transfer Conference, 1, HMT1175 (1975)
[26] Kays, W. M., Lundberg, R. E., and Reynolds, W. C.: Heat transfer with laminar flow in
concentric annuli with constant and variable wall temperature and heat flux. NASA
Technical reports, AHT2 (1961)
[27] WorsoeSchmidt, P. M.: Heat transfer in the thermal entrance region of circular tubes
and annular passages with fully developed laminar flow. International Journal of Heat
and Mass Transfer, 10, 541551 (1967)
[28] Cotta, R. M., and Ozisik, M. N.: Laminar forced convection of powerlaw non
Newtonian fluids inside ducts. Heat and Mass Transfer, 20, 211218 (1986)
[29] Stein, R. P., and Sastri, V. M. K.: A heat transfer analysis of heat exchangers with
laminar tubeside and turbulent shellside flows–A new heat exchanger Graetz problem.
AIChE Symposium Series, 118, 8189 (1972)
[30] Quick, K.: Direct calculation of exchanger exit temperatures in cocurrent flow.
Chemaical Engineering, 93, 92 (1986).
In: Fluid Mechanics and Pipe Flow ISBN: 9781607410379
Editors: D. Matos and C. Valerio, pp. 231267 © 2009 Nova Science Publishers, Inc.
Chapter 7
NUMERICAL SIMULATION OF TURBULENT
PIPE FLOW
M. OuldRouis and A.A. Feiz
Université ParisEst, Laboratoire Modélisation et Simulation Multi Echelle,
MSME, FRE3160 CNRS, 77454 MarnelaVallée, France
Abstract
Many experimental and numerical studies have been devoted to turbulent pipe flows due to
the number of applications in which theses flows govern heat or mass transfer processes: heat
exchangers, agricultural spraying machines, gasoline engines, and gas turbines for examples.
The simplest case of nonrotating pipe has been extensively studied experimentally and
numerically. Most of pipe flow numerical simulations have studied stability and transition.
Some Direct Numerical Simulations (DNS) have been performed, with a 3D spectral code, or
using mixed finite difference and spectral methods. There is few DNS of the turbulent rotating
pipe flow in the literature. Investigations devoted to Large Eddy Simulations (LES) of
turbulence pipe flow are very limited. With DNS and LES, one can derive more turbulence
statistics and determine a wellresolved flow field which is a prerequisite for correct
predictions of heat transfer. However, the turbulent pipe flows have not been so deeply
studied through DNS and LES as the planechannel flows, due to the peculiar numerical
difficulties associated with the cylindrical coordinate system used for the numerical
simulation of the pipe flows.
This chapter presents Direct Numerical Simulations and Large Eddy Simulations of fully
developed turbulent pipe flow in nonrotating and rotating cases. The governing equations are
discretized on a staggered mesh in cylindrical coordinates. The numerical integration is
performed by a finite difference scheme, secondorder accurate in space and time. The time
advancement employs a fractional step method. The aim of this study is to investigate the
effects of the Reynolds number and of the rotation number on the turbulent flow
characteristics. The mean velocity profiles and many turbulence statistics are compared to
numerical and experimental data available in the literature, and reasonably good agreement is
obtained. In particular, the results show that the axial velocity profile gradually approaches a
laminar shape when increasing the rotation rate, due to the stability effect caused by the
centrifugal force. Consequently, the friction factor decreases. The rotation of the wall has
large effects on the root mean square (rms), these effects being more pronounced for the
streamwise rms velocity. The influence of rotation is to reduce the Reynolds stress component
〈V
r
'V
z
'〉 and to increase the two other stresses 〈V
r
'V
θ
'〉 and 〈V
θ
'V
z
'〉. The effect of the Reynolds
M. OuldRouis and A.A. Feiz 232
number on the rms of the axial velocity (〈V
z
'
2
〉
1/2
) and the distributions of 〈V
r
'V
z
'〉 is evident,
and it increases with an increase in the Reynolds number. On the other hand, the 〈V
r
'V
θ
'〉
profiles appear to be nearly independent of the Reynolds number. The present DNS and LES
predictions will be helpful for developing more accurate turbulence models for heat transfer
and fluid flow in pipe flows.
Nomenclature
C
f
=2τ
w
/ρU
b
2
friction coefficient
L
z
length of the computational domain
N rotation number
N
θ
, N
r
, N
z
grids in θ, r and z directions
r dimensionless coordinate in radial direction scaled by the pipe
radius
R pipe radius (m)
Re
b
=U
b
D/ν Reynolds number based on bulk velocity
Re
τ
=u
τ
D/ν Reynolds number based on friction velocity
Re
p
=U
p
D/ν Reynolds number based on U
p
.
S
ij
rate of strain tensor
u
τ
shear stress velocity
U
b
=U
p
/2 bulk velocity
U
p
centreline streamwise velocity of the laminar Poiseuille flow
v’
r
,v’
θ
, v’
z
fluctuating velocity components
y dimensionless distance from the wall, y=1r
y
+
=(1r) u
τ
/ν distance from the wall in viscous wall units
z coordinate in axial direction
Greek letters
δ* defined by δ*(2Rδ*) = 2
0
R
∫
r(1V
z
(r)/U
p
)dr
Δ=(r Δr Δθ Δz)
1/3
characteristic gridspacing
Δr gridspacing in radial direction
Δθ gridspacing in circumferential direction
Δz gridspacing in axial direction
ν kinematic viscosity
θ coordinate in circumferential direction
θ∗ defined by θ*(2Rθ*) = 2
0
R
∫
r V
z
(r)/U
p
(1V
z
(r)/U
p
)dr
Numerical Simulation of Turbulent Pipe Flow 233
1. Introduction
The turbulent circular pipe flow has attracted the interest of many investigators. The
simplest case of nonrotating pipe has been extensively studied experimentally [Laufer
(1954), Lawn (1971)] and numerically. Most of pipe flow numerical simulations have studied
stability and transition [Itoh (1977), Patera and Orszag (1981)]. Some Direct Numerical
Simulations (DNS) have been performed. Using mixed finite difference and spectral methods,
Nikitin (1993) was able to obtain satisfactory agreement with experimental data, inside the
Reynolds number range of 22505900. Unger et al. (1993) obtained excellent agreement with
experiments, using a second order accurate finite difference method. They confirmed that
pipe flow at low Reynolds number deviates from the universal logarithmic law. Zhang et al.
(1994) reported simulation of low to moderate Reynolds number turbulent pipe flow obtained
with a 3D spectral code. Their initial results agree satisfactorily with both experiments and
previous numerical simulations. Eggels et al. (1994a) have carried out DNS and experiments
in order to investigate the differences between fully developed turbulent flow in an
axisymmetric pipe and a plane channel geometry. Most of the statistics on fluctuating
velocities appear to be less affected by the axisymmetric pipe geometry.
When a flow is introduced to an axially rotating pipe, fluids are given a tangential
component of velocity by the moving wall and the flow in the pipe exhibits a complicated
threedimensional nature. The high levels of turbulence and large shearing rates associated
with swirling flows enhance the mixing process and provide a more homogeneous flow of
fluids. The role of the swirl flow is of great importance for the overall performance of the gas
turbine. Recently, the numerical simulation of turbulent rotating pipe flow has received some
interest. Eggels et al. (1994b) used a DNS of the turbulent rotating pipe flow for moderate
values of the rotation number. They confirmed numerically the drag reduction observed in
experiments. Orlandi and Fatica (1997) have also performed DNS of the turbulent rotating
pipe flow. Their investigation was devoted to the study of the range of the rotation number,
N, not considered by Eggels et al. (1994b), that is the investigation of the flow field at high
values of N (N ≤ 2) but not enough high to include relaminarization, and to analyze the
modifications of the nearwall vortical structures, for a more satisfactory explanation of the
drag reduction. They showed that a degree of drag reduction is achieved in the numerical
simulations just as in the experiments, and that the changes in turbulence statistics are due to
the tilting of the nearwall streamwise vortical structures in the direction of rotation. The
more recent study by Orlandi and Ebstein (2000) is an extension of the previous one. N has
been increased up to 10. These authors have evaluated the budgets for the Reynolds stresses
at high rotation rates. These budgets are useful to those interested in developing new one
closure turbulence models for rotating flows.
Using a modified mixing length theory, Kikuyama et al. (1983) conducted calculations of
the flow in an axially rotating pipe in a region far downstream from its inlet section. They
observed that when a turbulent level is introduced into the rotating pipe, a flow laminarization
is set up through an increase in the rotational speed of the pipe while destabilizing effects
occur when the flow is initially laminar. Malin and Younis (1997) used two Reynolds stress
transport closures for modeling flow and heat transfer in fully developed axially rotating pipe
flow. They showed that both models reproduce the observed influences of rotation. These
include reduction in skin friction and wall heat transfer, suppression of radial turbulent
M. OuldRouis and A.A. Feiz 234
transport of heat and momentum, laminarization of the flow. An encouraging level of
agreement between calculations and available findings of literature was generally found.
Speziale et al. (2000) presented both the analysis and modeling of turbulent flow in an
axially rotating pipe flow. A particular attention was paid to tracing the origin of each of the
two central physical features: the rotationally dependent axial mean velocity and the
rotationally dependent mean azimuthal or swirl velocity relative to the rotating pipe, in order
to gain a better physical insight into this turbulent flow. It was shown that secondorder
closure models provide good description of this flow and can describe both these features
fairly well. In all the previous numerical studies, only one Reynolds number has been
considered.
There is very limited literature on investigations devoted to Large Eddy Simulations
(LES) of turbulence pipe flow. The first LES work on fully developed turbulent pipe flow is
given by Unger and Friedrich (1991). LES have been applied to flows in complex geometries
to a very limited extent. The major reasons for this are due to the need for describing the non
trivial geometry accurately whilst limiting the number of computational grid points. LES
predictions on turbulent pipe flow with rotation are extremely rare. There are only three
works which deal with the turbulent rotating pipe flow. In 1993, Eggels and Nieuwstadt
performed Large Eddy Simulations (LES) for the fully developed turbulent flow inside an
axially rotating pipe. The main objective of this study was to investigate the influences of
pipe rotation on mean flow properties and mean velocity profiles as well as on the profiles of
the Reynolds stress components. They showed that the mean flow properties are fairly well
predicted by LES, especially about the reduction of wall friction, deformation of the mean
axial velocity profile and parabolic distribution of the mean circumferential velocity. Yang
and McGuirk (1999) reported LES of turbulent pipe flow for the rotating and non rotating
cases. Their numerical results compare reasonably well with the experimental data. They
confirmed the experimental observations that turbulence decreases with an increase in pipe
rotation due to the stability effect of the centrifugal force.
In the present study, DNS and LES of fully developed turbulent pipe flow are performed
to report the effects of the rotation and Reynolds numbers on the flow characteristics. The
direct simulations have been carried out at two Reynolds numbers, Re=4900 and Re=7400,
for different rotation rates ranging from N = 0 to N= 18. To elucidate the impact of higher
Reynolds and rotation numbers, large eddy simulations with dynamic model have been
conducted for a Reynolds number up to 20600. The present paper is organized as follows: the
mathematical formulation and the numerical methods (DNS and LES) are described in section
2. Section (3.1) presents DNS predictions of the turbulent pipe flow, and compares our
computations to the results reported in the archival literature. Section (3.2) deals with the LES
predictions of the turbulent pipe flow. The validation of DNS and LES approaches are
achieved by comparing the predicted profiles and statistics to the numerical and experimental
data available in the literature. The effects of the Reynolds and rotation numbers on different
thermal statistics (mean velocity profiles, root mean square of fluctuating velocity
components, Reynolds shear stresses, skewness and flatness factors, velocity and vorticity
fields) are investigated and discussed in section 3. The conclusion in section 4 ends this work.
Numerical Simulation of Turbulent Pipe Flow 235
2. Governing Equations and Numerical Method
2.1. Direct Numerical Simulation
The continuity and momentum equations governing 3Dincompressible turbulent flow
are written in a cylindrical coordinate system, Figure 1, in terms of the variables q
r
=r.V
r
,
q
θ
=V
θ
and q
z
=V
z
, in order to avoid the singularity at the axis r=0. The dimensionless
equations are obtained using U
p
, the centerline streamwise velocity of the laminar Poiseuille
flow, and the pipe radius R as velocity and length scales respectively:
(1)
(2)
(3)
(4)
where the stresses, τ
ij
=2S
ij
, are given by:
(5)
M. OuldRouis and A.A. Feiz 236
Figure 1. Sketch of control volumes.
A mean pressure gradient in the q
z
equation maintains a constant bulk velocity, U
b
. The
Reynolds number and the rotation rate are defined as Re
b
=U
b
D/ν and N=ΩR/U
b
, respectively.
2.2. Large Eddy Simulation
The governing equations for Large Eddy Simulation (LES) are given below. The overbar
indicates the filtering of the instantaneous fields which leads to the resolved scale fields (the
smaller scales of turbulent motion being removed by the spatial filtering approach).
0 =
∂
∂
+
∂
∂
+
∂
∂
z
z
q
r
q
r
r
q
θ
θ
,
ˆ ˆ
1
ˆ r
1 1 1
2
1
2
2 ⎥
⎥
⎦
⎤
⎢
⎢
⎣
⎡
∂
∂
+
∂
∂
+
∂
∂
+
∂
∂
− = +
∂
∂
+
∂
∂
+
∂
∂
+
∂
∂
z
z
r r
r
r
Re
P
r
r
q N
z
z
q q q q
r
r
q q r
r
t
q
θ
τ
θ
θθ
τ
θ
τ
θ
θ
θ
θ θ θ θ
, ˆ
ˆ ˆ ˆ r
1

⎥
⎥
⎦
⎤
⎢
⎢
⎣
⎡
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
−
∂
∂
+
∂
∂
+
∂
∂
+
∂
∂
= −
∂
∂
+
∂
∂
+
∂
∂
+
∂
∂
θθ
τ
τ
θ
θ
τ τ
θ θ
θ
θ
θ
z
rz
r
r
r
rr
Re r
P
r Nr q q
z
z
q
r
q
r
r
q q
r
r
q
r
q
r t
r
q
q
.
ˆ ˆ
1
ˆ r
1 1
1 1
⎥
⎥
⎦
⎤
⎢
⎢
⎣
⎡
∂
∂
+
∂
∂
+
∂
∂
+
∂
∂
− =
∂
∂
+
∂
∂
+
∂
∂
+
∂
∂
z
zz z
r r
rz
r Re z
P
z
z
q
z
q
z
q q
r r
z
q
r
q
r t
z
q
τ
θ
θ
τ τ
θ
θ
Numerical Simulation of Turbulent Pipe Flow 237
The total stresses ij ij ij
' ˆ τ τ τ + =
are
ij
S Re
T ij
) 1 ( ˆ ν τ + =
where the strain tensor
expressed by the variables
i
q
is:
The eddy viscosity ν
T
has different expressions according to the subgrid model used.
Smagorinsky Model
In this model, the subgrid scale eddy viscosity is related to the deformation of the
resolved velocity field as:
[ ]
2 / 1
2
2
(
2
(
ij
S
ij
S Δ)
s
C S Δ)
s
C
T
ν = =
In the present study, the Smagorinsky coefficient C
S
is set equal to 0.15. For a discussion
on the value and the interpretation of this constant, we refer to Mason and Callen (1986). This
subgrid model largely used in LES of isotropic turbulence produced good results. When it
applied to inhomogeneous, and in particular to wall bounded flows, the constant was
modified.
Dynamic Eddy Viscosity Model
The dynamic model provides a methodology for determining an appropriate local value
of the Smagorinsky coefficient. The model was proposed by Germano et al. (1991), with
important modifications and extensions provided by Lilly (1992). In this model, the constant
C
d
is not given a priori, but is computed during the simulation from the flow variables. The
turbulent viscosity is expressed using an eddy viscosity assumption as:
[ ]
2 / 1
2
2
ij
S
ij
S ) (Δ
d
C
T
ν =
but C
d
is dynamically determined as follows.
M. OuldRouis and A.A. Feiz 238
Two different filter widths are introduced; the test filter Δ
~
is larger than the
computational filter
Δ
and it is applied to the momentum equations. Germano et al. (1991)
derived an exact relationship between the subgrid scale stress tensors at the two different filter
widths (Germano identity). Substitution of a Smagorinsky form
ij
S
ij
S S 2 =
for the subgrid
scale stress into Germano identity, along with some additional assumptions [Lilly (1992)],
leads to the expression for the constants
ij
M
ij
M
ij
M
ij
L
Δ
d
C
2
2
1
− =
(1)
where the second order tensors L
ij
and M
ij
are given as follows:
,
2
2
~ ~ ~
ij
M Δ
d
C 
j
q
i
q
j
q
i
q
ij
L = − =
~ ~ ~
~
2
2
ij
S S
ij
S S
Δ
Δ
ij
M − =
The constant could be positive or negative. The positive values are linked to energy
flowing from large to small scales and the negative to energy going from small to large scales
(backward energy transfer). The angled brackets in equation (1) denotes averages in the
homogeneous direction.
The governing equations are discretized on a staggered mesh in cylindrical coordinates.
The numerical integration is performed by a finite difference scheme, secondorder accurate
in space and in time. The timeadvancement employs a fractionalstep method. A thirdorder
Runge Kutta explicit scheme and a CrankNicolson implicit scheme are used to evaluate the
nonlinear and viscous terms respectively. Uniform computational grid and periodic boundary
conditions are applied to the circumferential and axial directions. In the radial direction, non
uniform meshes specified by a hyperbolic tangent function are employed. On the pipe wall,
the usual noslip boundary condition is applied. For Re
b
= 4900 and Re
b
= 7400, we
performed simulations on a pipe of length L
z
=20R using a 65x39x65 grid in the θ, r and z
directions. We have investigated the influence of different grids on the accuracy of the
solution. The finest grid (129x49x129) leads to well resolved simulations. However, the grid
65x39x65 seems to predict and capture all features of the flow although small differences
occur between some of the statistics obtained with these two grids. Since the fine grid
requires much larger CPUtime and storage requirements, we performed calculations on the
65x39x65 grid which gives a good compromise between the required CPUtime and
accuracy. Tables 1 and 2 list the details of the grid resolutions used in the simulations for the
other Reynolds numbers (Re
b
= 5300, Re
b
=10300 and Re
b
= 20600). The final statistics are
accumulated by spatial averaging in the homogeneous streamwise and circumferential
directions, and by timeaveraging.
Numerical Simulation of Turbulent Pipe Flow 239
Table 1. Grids resolutions for LES (Dynamic model)
Model Re
b
N Re
τ
(L
θ
, L
r
, L
z
) Grid
0 628 128x95x128
10300
6 607
0 1242
Dynamic
Model (D. M.)
20600
6 1218
(2π, 1, 15)
128x129x256
Table 2. Grids resolutions for DNS
Re
b
Re
τ
N (L
θ
, L
r
, L
z
) Grid
360 0 256x257x256
360 2 256x257x256
334 6 256x257x256
5300
231 18
(2π, 1, 15)
128x257x256
3. Results and Discussion
3.1. Direct Numerical Simulations
Mean Velocity Profiles
The axial mean velocity profiles (mean axial velocity normalized by the bulk mean
velocity U
b
) are compared in Figure 2a with the DNS and experimental data by Eggels et al.
(1994a) for a stationary pipe flow. The calculated profiles are in satisfactory agreement with
these data. For both Reynolds numbers, the streamwise velocity increases near the centre and
decreases near the wall when the pipe is rotating. An examination of the velocity profiles
shows a gradual approach towards a parabolic shape (Poiseuille profile) when increasing N,
and correspondingly the effect of turbulence suppression due to the pipe rotation becomes
more and more noticeable. Rotation has thus a very marked influence on the damping of the
turbulent motion and drag reduction. This is in agreement with the experiments conducted by
Nishibori et al. (1987), Reich and Beer (1989) and Imao and Itoh (1996). At Re=7400, the
computed velocities have lower values than those for Re=4900 in the central region of the
pipe, whatever the rotation rate is. A similar observation has been reported in the experiments
by Reich and Beer (1989). It appears thus that the Reynolds number dependence of the mean
velocity profile decreases when the rotation rate increases. To better investigate the effect of
rotation on the turbulent pipe flow, we conducted four DNS with a rotation rate changing
gradually from zero to a limit value N = 18. For this value (N = 18), a tendency to the
relaminarization of the flow is observed. Table 2 summarizes the grid resolutions. As shown
in Figure 2b, the mean velocity profile is significantly affected by the Coriolis force. It is
worth pointing that the extend of the nearwall region is reduced when N increases, in
accordance with the experimental observation of Kikuyama et al. (1983).
M. OuldRouis and A.A. Feiz 240
Figure 2a. Axial mean velocity normalized by the bulk velocity U
b
as function of the wall distance for 0
≤ N ≤ 2.
Figure 2b. Axial mean velocity normalized by the bulk velocity U
b
as function of the wall distance for
N=0 (─), N=2 (), N=6 (···) and N=18 (·─·).
Numerical Simulation of Turbulent Pipe Flow 241
Mean Axial Velocity Normalized on the Friction Velocity
In Table 3, the mean flow parameters are compared to those of literature, for Re
b
≈ 5300.
The DNS calculations are first compared to the findings of Eggels et al. (1994a) for which the
Re
τ
is close to our Re
τ
. There is a good agreement between the present predictions and the
results of Eggels et al. (1994a). The slight difference between the data sets can be attributed to
the difference in the grid resolutions between the two simulations. The axial mean velocity
normalized by the friction velocity u
τ
versus the distance from the wall is shown in Figure 3a
in wall units y
+
. This computed profile is consistent with the DNS profile of Eggels et al.
(1994a). The validation of the present predictions has also been achieved by comparing our
calculated kinetic energy, Figure 3b, and vorticity fluctuations, Figure 3c, to those of Eggels
et al. (1994a).
Table 3. Meanflow parameters for Direct Numerical Simulation (DNS) compared to
literature, at Re
b
= 5300 and N=0
DNS
Eggels et al.
(1994b)
Loulou (1996)
Westerweel et
al. (1996)
[PIV]
Westerweel et
al. (1996)
[LDA]
N
θ
256 128 160  
N
r
257 96 72  
N
z
256 256 192  
RΔθ
+
max
8.89 8.84 7.50  
Δr
+
min
0.11 0.94 0.39  
Δr
+
max
4.03 1.88 5.70  
Δz
+
14.10 7.00 9.90  
Re
p
6954 6950 7248 7100 7200
Re
b
5299 5300 5600 5450 5450
Re 362 360 380 366 371
U
c
/u 19.19 19.31 19.12 19.38 19.39
U
b
/u 14.63 14.73 14.77 14.88 14.68
U
c
/U
b
1.31 1.31 1.29 1.30 1.32
C
f
(x10
3
) 9.35 9.22 9.16 9.03 9.28
*/R 0.127 0.127 0.121 0.124 0.130
θ*/R 0.069 0.068 0.066 0.068 0.071
H= */θ* 1.85 1.86 1.84 1.83 1.83
M. OuldRouis and A.A. Feiz 242
Eggels et al, 1994a.
Figure 3a. Axial mean velocity normalized by the friction velocity u
τ
as function of the distance for
N=0 and Re
b
=5300.: lignes (present DNS), symbols.
Eggels et al, 1994a.
Figure 3b. Kinetic energy profile for N=0: lignes (present DNS), symbols.
Numerical Simulation of Turbulent Pipe Flow 243
Eggels et al, 1994a.
Figure 3c. Vorticity fluctuations versus the wall distance for N=0 and Re
b
=5300.: lignes (present DNS),
symbols.
Figure 3d. Axial mean velocity normalized by u
τ
: a comparison between the present DNS and the data
of literature.
M. OuldRouis and A.A. Feiz 244
Figure 3e. Axial mean velocity normalized by u
τ
as function of the distance (in wall units) from the wall
with Re
b
and N as parameters.
To ascertain the reliability and accuracy of the present numerical simulation, the present
predictions are compared to the available results of literature at Re=7442 in the case of a
stationary pipe: the experimental data of Eckelmann (1974) and Kim et al. (1987) for a
channel flow. They are also compared to the LDAmean flow distribution by Durst et al.
(1995) for a turbulent stationary pipe flow. Figure 3d shows that all the velocity profiles meet
fairly well with our numerical predictions. The satisfactory agreements confirm that the mean
flow field is well predicted by the present numerical simulations. In Figure 3e, the solid line
represents the universal velocity distributions in the viscous sublayer, in the buffer layer and
in the inertial sublayer. The viscous sublayer is well resolved in the numerical simulations,
yielding the linear velocity distribution V
z
+
=y
+
for y
+
<5. The buffer region is also well
predicted in accordance with the loglaw V
z
+
=3.05+5*ln y
+
. For the two Reynolds numbers
considered, the agreement with the loglaw at larger distances from the wall (y
+
>30) is less
for the present DNS results, and also for the experimental data by Eggels et al. (1994a). When
the pipe rotates, the differences between the computed mean velocity and the loglaws are due
to the relaminarization of the flow when the rotation rate increases. Similar observations have
been reported by Orlandi and Fatica (1997) and by Zhang et al. (1994). The reason is that the
loglaws are not observed in the pipe flow for Re
b
≤ 9600 (Re
b
=U
b
D/ν), in contrast to plane
channel flows. Theoretically, the loglaws are only justified at large Reynolds numbers
(Tennekes and Lumley, 1972). The variation of the velocity friction with the rotation number
N is sketched in Table 4, for Re
b
= 5300. Here, each velocity friction
u
τ
is normalized by that
of the nonrotating case,
0
u
τ
. Note that the friction velocity diminishes with an augmentation
in the rotation number, inducing a marqued reduction in the friction coefficient at the wall.
Numerical Simulation of Turbulent Pipe Flow 245
The friction factor decreases with an increase in the rotation number and this tendency
becomes more remarkable for larger values of the Reynolds number. For the highest rotation
number, N = 18, the reduction in the friction coefficient f is about 60%. For stationary pipe
flows, the simulation predicts a friction factor close to the value evaluated by using the
Blasius relation for Re=4900 (f = 0.3164 Re
1/4
= 0.0378).
Table 4. Variation of the velocity friction with the rotation number at Re
b
= 5300.
Re
b
N u
/ u
τ
0
0 1.00
2 1.00
6 0.93
5300
18 0.64
Root Mean Square (rms)
The radial distribution of the root mean square (rms) of the fluctuating velocities in a
nonrotating pipe are plotted in Figure 4a, for Re
b
= 5300, along with DNS calculations of
Eggels et al. (1994a). There is a satisfactory agreement between the present predictions and
the measurements and DNS by Eggels et al. (1994a). The rms values of all velocity
components are also depicted in Figures 4bd and compared to the numerical computation of
Kim et al. (1987), to the measured rms of Kreplin and Eckelmann (1979) obtained with a X
hot film probe and to the LDA measurements by Durst et al. (1995). The comparison between
the present DNS results and the available experimental and numerical data of literature seems
Figure 4a. Rootmeansquare profiles of azimuthal, radial and axial velocity components for N=0.
Lines: Present DNS; Symbols (Eggels et al., 1994b): (Δ) V
r , rms
, (O) V
θ , rms
, ( ) V
z , rms.
M. OuldRouis and A.A. Feiz 246
to be better satisfactory. However, there are some slight differences. These differences are
most likely caused by the coarse numerical resolution of our computations. For Re
b
= 4900,
Figure 4b shows that the peak in the distribution of the streamwise rms is located at the inner
edge of the buffer region while that of the normal and tangential components are located at
the outer edge, as it can be seen in Figures 4c,d.
Figure 4b. Rootmeansquare of the velocity fluctuations: axial velocity component, N=0.
Figure 4c. Rootmeansquare of the velocity fluctuations: radial velocity component, N=0.
Numerical Simulation of Turbulent Pipe Flow 247
Figure 4d. Rootmeansquare of the velocity fluctuations: tangential velocity component, N=0.
Figure 5a. Rootmeansquare profiles of axial velocity with Re
b
and N as parameters (0≤N≤2).
M. OuldRouis and A.A. Feiz 248
Figure 5b. Rootmeansquare profiles of velocity fluctuations for 0≤N≤18, Re=5300: <(V
z
’)
2
>
1/2
(─) ,
<(V
θ
’)
2
>
1/2
(), <(V
r
’)
2
>
1/2
(···).
The rotation of the wall has large effects on the rms, these effects being more
pronounced for the streamwise rms velocity. Similar observations have been reported by
Numerical Simulation of Turbulent Pipe Flow 249
Eggels et al. (1994a) and by Orlandi and Fatica (1997). For N=2 and N=1, the rms
distributions are almost the same (Figure 5a). Figure 5b depicts the turbulence intensities
normalized by the friction velocity for Re
b
= 5300 and higher rotation numbers. At a
moderate rotation number (N = 1), the fluctuation levels of the three velocity components
are increased in comparison to the nonrotating case, Figure 5b. From N ≥ 2, this tendency
remains for the radial and azimuthal components. On the contrary, a significant reduction
of the axial turbulence intensity is observed. The intensification of the radial and azimuthal
turbulence intensities in the core region of the rotation pipe denotes that the turbulence
there tends to be isotropic, especially at high rotation number. Near the wall (0 ≤ y ≤ 2), the
azimuthal turbulence intensity exhibits remarkable peak values with increasing N, while the
maximum of the axial fluctuations is considerably reduced (10% for N = 2 and 30% for N
= 18).
From the three rms velocities, the effect of the Reynolds number on the rms of the axial
velocity (〈V
z
'
2
〉
1/2
) is evident, and it increases with an increase in the Reynolds number, Figure
5a. Similar observations are reported by Zhang et al. (1994). For the other rms velocities,
〈V
θ
'
2
〉
1/2
and 〈V
r
'
2
〉
1/2
, this trend is observed in the core region of the flow.
Reynolds Shear Stresses
Figure 6a shows the distributions of the Reynolds stress components 〈V
r
'V
z
'〉. The
Reynolds stresses in a stationary pipe allow to check the accuracy of the simulations. The
computed values of 〈V
r
'V
z
'〉 are compared with the measurements of Eggels et al.
(1994a). The agreement between the present predictions and these results is satisfactory.
The total shear stress (the sum of the turbulent and viscous stresses) is also plotted in this
figure (solid line). As it can be seen the Reynolds shear stress dominates in the core
region of the flow since the viscous shear stress is small. When the rotation rate
increases, the decrease of the viscous stress in the core region leads to a decrease of
〈V
r
'V
z
'〉. On the other hand, the stress 〈V
r
'V
z
'〉 increases near the wall in accordance with
the increase of the viscous stresses. For Re=7400, the values of 〈V
r
'V
z
'〉 are slightly larger
than the corresponding values obtained for Re=4900, especially near the wall. The
〈V
r
'V
θ
'〉 and 〈V
θ
'V
z
'〉 stresses, which are zero in a stationary pipe flow, become nonzero
when the pipe rotates. The influence of rotation is to reduce the 〈V
r
'V
z
'〉 shear stress and
to increase the two other stresses 〈V
r
'V
θ
'〉 and 〈V
θ
'V
z
'〉 as it is shown in Figure 6b. Near
the rotating wall, the high values of 〈V
θ
'V
z
'〉 are related to the tilting of the near wall
vortical structures (Orlandi and Fatica, 1997). In the core region of the flow, the radial
oscillations of 〈V
θ
'V
z
'〉 are due to the largescale structures within this region. The
behavior of 〈V
r
'V
θ
'〉 is almost linear in the central part of the pipe. The Reynolds number
has larger effects on the component 〈V
θ
'V
z
'〉, these effects being more pronounced when
the rotation rate increases. The largest variations occur in the central region of the pipe.
At N=1, the distributions of 〈V
r
'V
z
'〉 for both Reynolds numbers are well separated, and
the 〈V
r
'V
z
'〉values are larger at Re=7400. On the other hand, the 〈V
r
'V
θ
'〉profiles appear
to be nearly independent of the Reynolds number.
M. OuldRouis and A.A. Feiz 250
Figure 6a. Distribution of the Reynolds shear stress component <V
r
’V
z
’> in wall units with Re
b
and N
as parameters.
Figure 6b. Reynolds shear stresses distributions in wall units for N=1.
Numerical Simulation of Turbulent Pipe Flow 251
Velocity and Vorticity Fields : Structures and Statistics
The instantaneous velocity and instantaneous vorticity fields are of great help in revealing
the influence of the rotation effect on turbulence coherent structures. Figure 7a depicts the
isosurface of the vorticity modulus (ω = 3) for various rotation numbers N. It clearly shows
the impact of the rotation rate on the turbulence structures:
• the more and more flat shape of the isosurface near the wall, with increasing N,
denotes a gradually reduction of the turbulence activity,
• As moving away from the wall, the augmentation in the rotation number induces a
set of turbulence structures with strong vorticity developing up to the pipe centre,
• These vortical structures are inclined and better organized with increasing N.
Figure 7a. Isosurface of the vorticity modulus (ω=3) for 0≤N≤18.
The rotation seems to have a tendency to organize the flow near the pipe wall and in the
core region, leading to a relaminarization of flow.
A plot of a crosssectional view of the streamwise vorticity field on an axialazimuthal
plane, for many rotation numbers, also exhibits the vertical structures in the pipe flow, Figure
7b. At moderate rotation rate (N = 2), the vertical structures concentrate in some regions of
the flow, while the other regions appear rather quiet. The strong longitudinal expansion of
M. OuldRouis and A.A. Feiz 252
these regions denotes presence of elongated vorticity streaks, similar to those observed in
experiments. The impact of rotation on the axial velocity fluctuation can be seen in Figure 7c.
The patterns clearly illustrate a better organization of the pipe flow when N increases, and a
more pronounced inclination of the axial velocity isosurface.
Figure 7b. Crosssectional view of the streamwise vorticity field ans axialazimuthal plane, for
0≤N≤18: N=0 (a), N=2 (b), N=6 (c), N=18 (d).
The rms values of all three vorticity components, for the four direct numerical
simulations, are plotted in Figure 7d. Once again, one can clearly see the impact of the
Coriolis force which gradually reduces the vorticity fluctuations. It is interesting to point out
the increasing anisotropy of the vorticity fluctuations. Indeed, with increasing N, the axial
vorticity fluctuations predominate those of the two other components.
Numerical Simulation of Turbulent Pipe Flow 253
Figure 7c. Isosurface of the axial velocity for 0≤N≤18.
Figure 7d. Continued on next page.
M. OuldRouis and A.A. Feiz 254
Figure 7d. Root mean square of the three vorticity components for 0≤N≤18: <(ω
z
’)
2
>
1/2
(─) , <(ω
θ
’)
2
>
1/2
(), <(ω
r
’)
2
>
1/2
(···).
HighOrder Statistics
Many experimental investigations of turbulent pipe flow at moderate Reynolds numbers
have been published (Durst et al., 1995; Den Toonder, 1995; Westerweel et al., 1996). They
used essentially Laser techniques to measure data: PIV (Particle Image Velocimetry) and
LDA (Laser Doppler Anemometers). Figure 8 compares our DNS results for higher order
statistics (skewness and flatness factors for axial velocity component) to the experimental
data of Westerweel et al. (1996). There is a satisfactory agreement between the numerical and
experimental profiles. However, the present DNS slightly overpredicts the peak of the axial
component V
z
’. Durst et al. (1995) reported similar observation when comparing their results
with many DNS of literature.
The skewness factors of the axial, radial and tangential velocity components (S
1
, S
2
and
S
3
, respectively) are compared in Figures 8c,d,e to the DNS of Kim et al. (1987), to the laser
Doppler measurements of Niederschulte et al. (1990) for channel flow, and also to the LDA
measurements of Durst et al. (1995) in the case of a stationary pipe. S
3
is also compared to the
LDV measurements of Karlsson and Johansson (1986) for a boundary layer flow.
About S
1
(Figure 8c) the trends observed in the present simulations, in the experimental
data and DNS of literature are similar, with a positive skewness near the wall and a negative
skewness factor approaching a value of –0,5 at larger y
+
in the stationary pipe. The agreement
with the experimental or numerical findings of the referenced literature is not so good, but
still reasonable. Our predictions of S
1
are somewhat overpredicted, especially near the wall.
Figure 8d shows the variations of the skewness factor of S
2
in wall region: the differences
between the present predictions and the results of literature are larger than for S
1
. These
discrepancies could be attributed to the too coarse grids used, but it should be also taken into
Numerical Simulation of Turbulent Pipe Flow 255
account that the experimental data are obscured by noises close to the wall, which lead to a
smaller skewness factor. The predicted values of S
3
, are shown in Figure 8e: for a stationary
pipe, almost zerovalues are predicted over most of the pipe diameter because of symmetry.
This is in good agreement with the data from DNS or measurements reported in the literature.
Eggels et al, 1994a.
Figure 8. Skewness (a) and Flatness (b) factors for axial velocity component, N=0 and Re
b
=5300:
lignes (present DNS), symbols.
Figure 8c. Skewness factors of the velocity fluctuations in the nearwall region: the streamwise velocity
component with Re
b
and N as parameters.
M. OuldRouis and A.A. Feiz 256
Figure 8d. Skewness factors of the velocity fluctuations in the nearwall region: the normaltothewall
velocity component with Re
b
and N as parameters.
Figure 8e. Skewness factors of the velocity fluctuations in the nearwall region: the circumferential
velocity component with Re
b
and N as parameters.
Numerical Simulation of Turbulent Pipe Flow 257
For a rotating pipe, the skewness factors S
1
and S
2
show large variations, especially in the
near wall region. However S
1
approaches the negative value of –0,5 at larger distances from
the wall (y
+
> 70), while S
2
tends towards zero. These large variations of the skewnesses with
the rotation rate are a further indication that changes of the orientation in the vortical
structures occur near the wall (Orlandi and Fatica, 1997). When the pipe rotates, the skewness
factor of the tangential velocity component, S
3
, is larger for N=1 near the wall. Its non zero
values across the flow field suggest that the reflection symmetry is broken by rotation.
In contrast to the skewness factor, the flatness factor of the axial velocity component, F
1
,
is in better agreement with the results of literature as it can be seen in Figure 9a. However,
higher flatness values are predicted at the wall region while the agreement is very good in the
core region. The behavior of the radial velocity component, F
2
, shows similar trend (Figure
9b). The agreement is good with the data reported in previous studies for y
+
> 25 (Durst et al.,
1995; Niederschulte et al., 1990; Kim et al., 1987; Karlsson and Johansson, 1986). The
flatness factor is slightly larger than 3, which is the value of flatness factor for a Gaussian
distribution. However, near the wall, there is a strong increase of the flatness factor whereas
the experimental data show a decreasing flatness factor as the wall is approached (Durst et al.,
1995; Niederschulte et al., 1990). Clearly, noticeable discrepancies between the different
results for the flatness factor of the radial velocity component are observed, but the origin of
these discrepancies is not clarified. The distribution of the flatness factor of the tangential
velocity component, F
3
, is plotted in Figure 9c. The agreement between the present simulation
for stationary pipe flow and the results of literature is fairly good.
Figure 9a. Flatness factors of the velocity fluctuations in the nearwall region: the streamwise velocity
component with Re
b
and N as parameters.
M. OuldRouis and A.A. Feiz 258
Figure 9b. Flatness factors of the velocity fluctuations in the nearwall region: the normaltothewall
velocity component with Re
b
and N as parameters.
Figure 9c. Flatness factors of the velocity fluctuations in the nearwall region: the circumferential
velocity component.
Numerical Simulation of Turbulent Pipe Flow 259
In the case of a rotating pipe, Figures 9a,c show that large variations of the flatness
factors F
1
, F
2
and F
3
occur near the rotating wall while they are almost unchanged in the
remaining part of the pipe. This is an indication of the intermittent nature of the wall region.
Durst et al. (1995) showed that the higher ordermoments and turbulence intensity are
highly interconnected, especially close to the wall. Therefore, we have examined the inter
relations between these quantities over the crosssection of the pipe. We have plotted in
Figure 10 the distribution of the turbulence intensity together with the distributions of the
skewness and flatness factors of the streamwise velocity component. It can be seen that the
increase of the turbulence intensity results from a decrease of the nonGaussian behavior of
the skewness and flatness factors. The points for the maximum intensity, zero skewness and
minimum flatness coincide. This result agrees with the measurements of Durst et al. (1995).
Figure 10. Distributions of turbulent intensity, skewness and flatness factors of the axial velocity
component.
3.2. Large Eddy Simulation
This section is devoted to Large Eddy simulation (LES) of turbulent pipe flow. The
objective of these LESs is to improve upon the description of such flow and to elucidate the
effect of higher Reynolds numbers and higher rotation numbers on the field flow. The
accuracy of the LES technique depends significantly on the ability of the subgridscale (SGS)
model. Feiz (2005) examined the performance of two turbulence models: the classical
algebraic eddyviscosity model of Smagorinsky (S.M.) and the dynamic model (D.M.) for the
range 556 ≤ Re
τ
≤ 697. Details of the corresponding grid resolutions are listed in Table 5.
M. OuldRouis and A.A. Feiz 260
The results show that the flow field and its statistics are better predicted with the dynamic
model. For example, Figure 11 compares our LES computations with the DNS data of
Wagner et al. (2001). The reader can find more details in Feiz (2005). In the this section, we
present some LES results obtained with the dynamic model. The motivation of these LES
studies is to deal with the behaviour of the turbulent pipe flow in the rotating and nonrotating
pipes for higher Reynolds numbers (Re
τ
> 1000), and to examine the effectiveness of the LES
method for predicting such turbulent flows. Table 6 gives informations on the grid resolution
for Re
τ
= 1242, using LES with the dynamic model. Table 7 lists the mean flow parameters of
the present LES, along with the DNS data of Wagner et al. (2001), for Re
b
= 10300. The
agreement between them is reasonable good.
Table 5. Grids resolutions for LES (Smagorinsky and Dynamic models) at Re
b
= 10300
and N=0
Model Re
b Re
τ
N (L
θ
, L
r
, L
z
) Grid
S. M. 556
D. M.
10300
628
0 (2π, 1, 15) 128x95x128
Table 6. Grids resolutions for LES (Dynamic model) at Re
b
= 20600 and N=0
Model Re
b Re
τ
N (L
θ
, L
r
, L
z
) Grid
D. M. 20600 1242 0 (2π, 1, 15) 128x129x256
Table 7. Meanflow parameters for Large Eddy Simulation (LES) and literature at Re
b
= 10300 and N=0
LES Wagner et al. (2001)
N
θ
128 240
N
r
95 70
N
z
128 486
RΔθ
+
max
8.89 8.36
Δr
+
min
0.11 0.64
Δr
+
max
4.03 7.68
Δz
+
14.10 6.58
Re
p
13181 13210
Re
b
10300 10300
Re
τ
628 640
U
c
/u 20.38 20.64
U
b
/u 15.94 16.09
U
c
/U
b
1.28 1.28
C
f
(x10
3
) 7.87 7.70
*/R 0.115 0.115
θ*/R 0.070 0.071
H= */θ* 1.65 1.63
Wagner et al., 2001.
Numerical Simulation of Turbulent Pipe Flow 261
Figure 11. Axial mean velocity normalized by the friction velocity as function of the wall distance for
Re
b
=10500 and N=0. Lignes (present LES): (a) Smagorinsky model, (b) dynamic model. Symboles:
Wagner et al (2001).
The rms velocity fluctuations nondimensionalized by the friction velocity are given on
Figures 12a,b,c for three Reynolds numbers Re
b
= 5300, Re
b
= 10300 and Re
b
= 20600. The
predicted fluctuations in the wallnormal, Figure 12a, azimuthal, Figure 12b, and streamwise,
Figure 12c, directions are similar to the experimental data of Westerweel et al. (1996). The
fluctuations in the streamwise direction are more intense than those in the two other
directions. The locations of the maxima for each velocity fluctuation are different: the
maxima of azimuthal and radial turbulence intensities are located in the logarithmic region,
while the maxima of the axial turbulence intensities occur in the buffer region. Furthermore,
the peak in the axial velocity fluctuation is located at a distance in wall unit y
+
≈ 13,
M. OuldRouis and A.A. Feiz 262
Figure 12. Root mean square velocity fluctuations in wall units, for N=0 and Re
b
=5300 (─), Re
b
=10300
(), Re
b
=20600 (···): (a) tangential component<(V
θ
’)
2
>
1/2
/u
τ
; (b) radial component<(V
r
’)
2
>
1/2
/u
τ
; (c)
axial component <(V
z
’)
2
>
1/2
/u
τ
Figure 13. Reynolds shear stresses distributions in wall units for N=0 and Re
b
=5300 (─), Re
b
=10300 (
), Re
b
=20600 (···).
Numerical Simulation of Turbulent Pipe Flow 263
irrespective of the Reynolds number, i.e. the Reynolds number has no effect on the peak value
of the axial velocity fluctuation. For the two other velocity components, the positions of the
corresponding peaks shift farther off from the wall with increasing Reynolds number,
suggesting that the maxima of <V
r
’V
r
’> and <V
θ
’V
θ
’> are not always located in the boundary
layers. Similar to the behaviour of the rms of velocity fluctuations in the radial and azimuthal
directions, the Reynolds shear stresses move away from the wall when Re
b
increases,
Figure 13.
Figures 14a,b depicts the distribution of the skewness and flatness factors of the axial
velocity fluctuation near the wall region of a nonrotating pipe, for many Reynolds numbers.
These distributions point out two interesting observations:
• first, the skewness and flatness coefficients of the velocity fluctuations seem to be
very sensitive to the variation of the Reynolds number, near the wall,
• in the core region, the skewness and flatness values for fully developed turbulent
flow are not equal to those of a Gaussian distribution (S = 0 and F = 3). This implies,
for the skewness of the radial velocity fluctuation (not represented here), that the
turbulent energy is convected from the wall to towards the pipe axis.
Close to the pipe axis, the probability distribution of the axial fluctuations is asymmetric.
The negative values of S
1
indicate that the negative fluctuation of V
z
’ are dominant in
probability. In the vicinity of the wall, S
1
becomes positive and is enhanced. This trend is
more apparent with increasing Re
b
. Similarly, the profiles of the flatness coefficient of the
axial velocity component increases near the wall with increasing Re
b
, suggesting that the axial
turbulence fluctuation in the wall region become more intermittent with increasing Reynolds
number.
Figure 14. Skewness (a) and Flatness (b) factors of axial velocity component for N=0 and Re
b
=5300
(─), Re
b
=10300 (), Re
b
=20600 (···).
M. OuldRouis and A.A. Feiz 264
Figure 15. Effect of Re on the skewness and flatness factors of the axial velocity fluctuation for high N
(N=6): Re
b
=5300 (─), Re
b
=10300 (), Re
b
=20600 (···).
Figure 16. Effect of N on the skewness and flatness factors of the axial velocity fluctuation for high Re
b
(Re
b
=20600): N=0(─), N=6 ().
Figures 15 a,b shows the effect of Reynolds number on the skewness and flatness factors
of the axial velocity fluctuation, for high rotation rate (N = 6). It seems that the impact of the
Reynolds number on the flatness factor is more important than that on the skewness factor.
For the highest Reynolds number (Re
b
= 20600), the influence of the rotation number on the
skewness factor is more pronounced than that on the flatness factor, Figures 16 a,b. In both
Numerical Simulation of Turbulent Pipe Flow 265
cases, the effect of Re
b
or N on the skewness and flatness coefficients are mainly restricted in
the near wall region. The present LES predictions compare reasonably well with the results of
literature. We can conclude that all phenomena in rotating and nonrotating turbulent pipe
flow can be captured directly by LES with dynamic model.
4. Conclusion
This chapter was devoted to DNS and LES of the fully developed turbulent flow in an
axially rotating pipe for various Reynolds numbers and different rotation rates. Different
statistical turbulence quantities including the mean and fluctuating velocity components,
friction coefficient, the Reynolds shear stresses and higher order statistics are obtained and
analyzed. An effort to reveal the effects of the Reynolds number and the rotation number on
the turbulent pipe flow is sketched. The validation of the present DNS and LES has been
achieved by comparing our predictions with some available results of literature. It is shown
that the computed deformation of the mean axial velocity profile agrees with the deformation
observed experimentally and that rotation has a very marked influence on the suppression of
the turbulent motion and on the drag reduction. A tendency to the relaminarization of the flow
is observed for the highest rotation number. The Reynolds number dependence of the mean
velocity profile decreases when the rotation rate increases. The friction factor decreases with
an increase in the rotation number and this tendency becomes more remarkable for larger
values of the Reynolds number.
The rms and Reynolds stresses profiles showed an encouraging level of agreement with
the measured data and DNS results of literature. From N ≥ 2, a significant reduction of the
axial turbulence intensity is observed, while the fluctuation levels of the radial and azimuthal
components are increased denoting an isotropization of the in the core region of the rotation.
The rotation also reduces the 〈V
r
'V
z
'〉 shear stress and increases the two other stresses 〈V
r
'V
θ
'〉
and 〈V
θ
'V
z
'〉. The Reynolds number has larger effects on the component 〈V
θ
'V
z
'〉, these effects
being more pronounced when the rotation rate increases. On the other hand, the 〈V
r
'V
θ
'〉
profiles appear to be nearly independent of the Reynolds number.
The overall agreement between the predicted skewness and flatness factors and the
results reported in the literature is satisfactory. For rotating pipe, the large variations of the
skewness and flatness factors of the velocity components, in the near wall region denote the
intermittent nature of the wall region and indicate that changes of the orientation in the
vortical structures occur near the wall. For the highest rotation rate, the impact of the
Reynolds number on the flatness factor is more important than that on the skewness factor.
For the highest Reynolds number, the influence of the rotation number on the skewness factor
is more pronounced than that on the flatness factor. In both cases, the effect of the Reynolds
or rotation numbers on the skewness and flatness coefficients are mainly restricted in the near
wall region. Visualizations of the instantaneous velocity and vorticity fields exhibit
turbulence structures with strong vorticity developing up to the pipe centre. These vortical
structures are inclined and better organized with increasing N.
An interesting outcome of the present investigation is to establish databases of various
turbulence statistics of the turbulent pipe flow at different Reynolds and rotation numbers.
These databases will undoubtedly helpful for evaluating and developing turbulence models.
M. OuldRouis and A.A. Feiz 266
References
[1] Durst F. , Jovanovic J. and Sender J., 1995, J. Fluid Mech., 295, p. 305.
[2] Eckelmann H., 1974, J. Fluid Mech., 65, p. 439.
[3] Eggels, J. G. M. and Nieuwstadt, F. T. M., 1993, In Proc. 9
th
Symp. on Turbulent Shear
Flows, Kyoto, Japan, p.310.
[4] Eggels, J. G. M., Unger, F., Weiss, F., M. H., Westerweel, J., Adrian, R. J., Friedrich, R.
and Nieuwstadt, F. T. M., 1994a, J. Fluid Mech., 268, p. 175.
[5] Eggels, J. G. M., Boersma, B. J. and Nieuwstadt, F. T. M., 1994b, Preprint.
[6] Feiz, A.A, 2006, Thèse de doctorat, Champs sur Marne. University of ParisEst Marne
LaVallée.
[7] Germano, M., Piomelli, U. and Cabot, W. H., 1991, Phys. Fluids A, 3, p. 1760.
[8] Imao, S. and Itoh, M., 1996, Int. J. Heat and Fluid Flow, 17, p. 444.
[9] Itoh, N., 1977, J. Fluid Mech., 82, p. 469.
[10] Karlsson R. I. and Johansson T. G., 1986, In Laser Anemometry in Fluid Mechanics III
(Ed. R. J. Adrian), Lisbon, Portugal, p. 273.
[11] Kikuyama , K., Murakami, M. and Nishibori, K., 1983, Bull. J.S.M.E., 26, p. 506.
[12] Kim J., Moin P. and Moser R., 1987, J. Fluid Mech., 177, p. 133.
[13] Kreplin H. and Eckelmann H., 1979, Phys. Fluids, 22, p. 1233.
[14] Laufer, J., 1954, NACA Report 1174.
[15] Lawn, C. J., 1971, J. Fluid Mech., 48, p. 477.
[16] Lilly, D. K., 1992, Phys. Fluids A, 4, p. 633.
[17] Loulou P., 1996, Thèse de doctorat, Stanford University, Department of Aeronautics and
Astronautics.
[18] Malin M. R. and Younis B. A., 1997, Int. Comm. Heat Mass Transfer, 24, p. 89.
[19] Mason, P. J. and Callen, N. S., 1986, J. Fluid Mech., 162, p. 439.
[20] Niederschulte N. A., Adrian R. J. and Hanratty T. J., 1990, Exps. Fluids, 9, p. 222.
[21] Nikitin, N. V., 1993, In Bulletin of APS, Vol. 38, No. 12, p. 2311.
[22] Nishibori, K., Kikuyama, K. and Murakami, M., 1987, JSME Intl J., 30, p. 255.
[23] Orlandi, P. and Fatica, M., 1997, J. Fluid Mech., 143, p. 43.
[24] Orlandi, P. and Ebstein, D., 2000, International Journal of Heat and Fluid Flow, 21, p.
499.
[25] Patera, A. T. and Orszag, S. A., 1981, J. Fluid Mech., 112, p. 467.
[26] Reich, G. and Beer, H., 1989, Intl J. Heat Mass Transfer, 32, p. 551.
[27] Tennekes, H. and Lumley, J. L., 1972, “A First Course in Turbulence”, MIT Press.
[28] Toonder Den J. M. J., 1995, Thèse de doctorat, Delft, University of Technology,
Department of Aero and Hydrodynamics.
[29] Unger, F., Eggels, J. G. M., Friedrich, R. and Nieuwstadt, F. T. M., 1993, In Proc. 9
th
Symp. on Turbulent Shear Flows, Kyoto, Japan, pages 2/1/12/1/6.
[30] Unger, F. and Friedrich, R., 1991, In Proc. 8
th
Symp. on Turbulent Shear Flows,
Munich, Germany, pp. 1931 – 1936.
[31] Wagner C., Huttl T. J. and Friedrich R., 2001, Computers & Fluids, 30, p. 581.
[32] Westerweel J., Draad A. A., Hoeven van der J. G. Th. and Oord van J., 1996,
Experiments in Fluids, 20, p. 165.
Numerical Simulation of Turbulent Pipe Flow 267
[33] Yang, Z. and McGuirk, J.J., 1999, Proc. of Turbulence and Shear Flow Phenomena,
USA, p. 863.
[34] Zhang, Y., Gandhi, A., Tomboulides, A. G. and Orszag, S. A., 1994, AGARD Conf.
Proc., 551, pp. 17.117.9.
In: Fluid Mechanics and Pipe Flow ISBN: 9781607410379
Editors: D. Matos and C. Valerio, pp. 269316 © 2009 Nova Science Publishers, Inc.
Chapter 8
PIPE FLOW ANALYSIS OF URANIUM NUCLEAR
HEATING WITH CONJUGATE HEAT TRANSFER
G.H. Yeoh
*
and M.K.M. Ho
Australian Nuclear Science and Technology Organisation (ANSTO),
PMB 1, Menai,NSW 2234, Australia
Abstract
The field of computational fluid dynamics (CFD) has evolved from an academic curiosity to a
tool of practical importance. Applications of CFD have become increasingly important in
nuclear engineering and science, where exacting standards of safety and reliability are
paramount. The newlycommissioned Open Pool Australian Lightwater (OPAL) research
reactor at the Australian Nuclear Science and Technology Organisation (ANSTO) has been
designed to irradiate uranium targets to produce molybdenum medical isotopes for diagnosis
and radiotherapy. During the irradiation process, a vast amount of power is generated which
requires efficient heat removal. The preferred method is by lightwater forced convection
cooling—essentially a study of complex pipe flows with coupled conjugate heat transfer.
Feasibility investigation on the use of computational fluid dynamics methodologies into
various pipe flow configurations for a variety of molybdenum targets and pipe geometries are
detailed in this chapter. Such an undertaking has been met with a number of significant
modeling challenges: firstly, the complexity of the geometry that needed to be modeled.
Herein, challenges in grid generation are addressed by the creation of purposebuilt body
fitted and/or unstructured meshes to map the intricacies within the geometry in order to ensure
numerical accuracy as well as computational efficiency in the solution of the predicted result.
Secondly, various parts of the irradiation rig that are required to be specified as composite
solid materials are defined to attain the correct heat transfer characteristics. Thirdly, the use of
an appropriate turbulence model is deemed to be necessary for the correct description of the
fluid and heat flow through the irradiation targets, since the heat removal is forced convection
and the flow regime is fully turbulent, which further adds to the complexity of the solution. As
complicated as the computational fluid dynamics modeling is, numerical modeling has
significantly reduced the cost and lead time in the molybdenumtarget design process, and
such an approach would not have been possible without the continual improvement of
computational power and hardware. This chapter also addresses the importance of
*
Email address: guan.yeoh@ansto.gov.au, Phone no:+6129717 3817, Fax no.:+6129717 9263. Corresponding
Author:Dr Guan Heng YEOH, B40, ANSTO, Private Mail Bag 1, Menai, NSW 2234, Australia.
G.H. Yeoh and M.K.M. Ho 270
experimental modeling to evaluate the design and numerical results of the velocity and flow
paths generated by the numerical models. Predicted results have been found to agree well with
experimental observations of pipe flows through transparent models and experimental
measurements via the Laser Doppler Velocimetry instrument.
Introduction
Australian Nuclear Science and Technology Organisation (ANSTO) is Australia’s
national nuclear research centre, and it possesses two reactors: the longserving and recently
decommissioned HighFluxAustraliaResearch (HIFAR) Reactor and the newly
commissioned Open Pool Australian Lightwater (OPAL) research reactor. The main
functions of both reactors are to produce radiopharmaceutical products for medicine, to
generate neutron beams for scientific research and to irradiate silicon ingots for
semiconductor applications.
This paper aims to examine the stateoftheart application of Computational Fluid
Dynamics to obtain sensible solutions to three separate pipe flow cooling problems of the
molybdenum99 irradiation facility, of which target cans containing uranium are irradiated
and subsequently processed for the production of medical isotopes. The diversity and
reliability of the CFD methodology will be demonstrated by the degree of detail modeled in
each study as well as by the consistency of agreement among simulation results and
experimental validation data and other numerical forms of verification comparisons. Three
case designs are examined. They are:
1. The ‘rocketcan’ design as used in the HIFAR reactor
2. The proposed ‘annular can’ replacement design for HIFAR
3. The current ‘Molybdenumplate’ design used in the new OPAL reactor
In simple terms, the study of reactor thermohydraulic characteristics is the evaluation of
a reactor coolant system’s heat removal capacity. To this regard, it is an evaluation of flow
through what is often a complex system of pipework containing various reactor components
acting as blockages. Such complex systems are difficult to evaluate for pressure drop and
flow velocity characteristics using standard correlationbased formula describing pipe flow.
To complicate matters, the heat transfer characteristics of components are closely dependent
on the unique flow characteristics of each specific pipeflow system. In the past, such unique
heat transfer characteristics had been evaluated by the use of scaled and prototype heat
transfer rigs, but this has become increasingly and prohibitively expensive. With the advent of
computers, many factors have made the direct simulation of pipe flows, by means of CFD,
feasible. The exponential increase in computational power, and everincreasing improvements
in numerical methods and graphical post processing, met with a similar decrease in
computational cost has made CFD a viable method to directly solve for such distinctive pipe
flow systems.
Pipe Flow Analysis of Uranium Nuclear Heating with Conjugate Heat Transfer 271
Background
Fluid dynamics deals with both heat transfer and fluid flow where the fluid paths can be
complicated and governing equations are nonlinear. For most reallife problems associated
with fluid dynamics, obtaining an analytical solution is difficult, if not impossible.
Experiments that simulate fluid flow using scaled models in air tunnels, water tunnels or
towing tanks are used to obtain accurate information, but these experiments can be difficult
and expensive to set up.
CFD takes the advantage of modern computing power in the application of numerical
methods to solve complex fluid dynamic problems. Retrospectively, CFD emphasizes the
resolution of the physical processes through the use of digital computers which proceeds by
first negotiating the subdivision of the domain into a number of finite, nonoverlapping sub
domains. This leads to the construction of an overlay mesh of cells covering the whole
domain. In general, the set of fundamental mathematical equations are required to be
converted into suitable algebraic forms, which are then solved via suitable numerical
techniques.
The mathematical equations governing the heat transfer and fluid flow of the pipe flow
systems within HIFAR and OPAL are those of the conservation of mass, momentum and
energy. Nevertheless, the majority of pipe flows in the reactor are turbulent in nature. Direct
Numerical Simulation (DNS) is the most accurate approach to turbulence simulation, which
directly solves the governing transport equations without undertaking any averaging or
approximation other than the numerical approximations performed on them. Through such
simulations, all of the fluid motions contained in the flow are considered to be resolved; all
significant turbulent structures are required to be adequately captured (i.e., the domain of
which the computation is carried out needs to accommodate for the smallest and largest
turbulent eddy). Alternatively, Large Eddy Simulation (LES), which essentially resolves the
large eddies exactly through the availability of mesh requirement but approximates the small
eddies, is still expensive but much less costly than DNS. The results of a DNS or LES
simulation contain very detailed information about the flow, producing an accurate realization
of the flow while encapsulating the broad range of length and time scales. DNS and LES
approaches usually require high usage of computational resources and often cannot be used as
a viable design tool in reactor design and analysis because of the enormity of the numerical
calculations and the large number of grid nodal points.
Meanwhile, a more pragmatic approach is to adopt computational procedures that can
still supply adequate information about the turbulent processes but avoid the need to predict
all of the effects associated with each and every eddy in the flow. For most engineering
purposes, especially in reactor design and analysis, information about the timeaveraged
properties of the flow (e.g., mean velocities, mean pressures, mean stresses, etc.) are
sufficient to satisfy regulatory requirements. In this sense, all details concerning the state of
the flow contained in the instantaneous fluctuations can be discarded. This process of only
obtaining mean quantities can be achieved by adopting a suitable timeaveraging operation on
the equations governing the conservation of mass, momentum and energy; these time
averaged equations are generally known as the ReynoldsAveraged NavierStokes (RANS)
equations.
G.H. Yeoh and M.K.M. Ho 272
Turbulence Modeling
The Reynoldsaveraged approach to turbulence results in the formulation of the two
equation turbulence model—the standard kε model proposed by Launder and Spalding
(1974). This model is well established, widely validated and gives rather sensible solutions to
most industrially relevant flows. As an alternative to the standard kε model, other eddy
viscosity models such as RNG kε model and reliazable kε model proposed by Yakhot et al.
(1992) and Shih et al. (1995) are possible recommendations. The improved features of these
models have been shown to be aptly applicable to predict important flow cases having flow
separation, flow reattachment and flow recovery.
For wall attached boundary layers, turbulent fluctuations are suppressed adjacent to the
wall and the viscous effects become prominent in this region known as the viscous sublayer.
The modified turbulent structure of nearwall flow generally precludes the application of the
twoequation models such as standard kε model, RNG kε model and reliazable kε model at
the nearwall region. One common approach is to adopt the socalled wallfunction method;
the nearwall region is bridged with logarithmic wall functions to avoid resolving the viscous
sublayer. Nevertheless, it is possible to totally resolve the viscous sublayer by the
application of the model standard kω model developed by Wilcox (1998) where ω is a
frequency of the large eddies of which the model has also shown to perform splendidly close
to walls in boundary layer flows. The standard kω model is nevertheless very sensitive to the
freestream conditions and unless great care is exercised, spurious results are obtained in flow
regions away from the solid walls. To overcome such problems, the SST (Shear Stress
Transport) variation of Menter’s model (1993, 1996) was developed with the aim of
combining the favorable features of the standard kε model with the standard kω model in
order that the inner region of the boundary layer is adequately resolved by the latter while the
former is employed to obtain numerical solutions in the outer part of the boundary layer. This
model works exceptionally well in handling nonequilibrium boundary layer regions such as
flow separation.
Grid Generation
The arrangement of discrete number of points throughout the flow field, normally
called a mesh, is a significant consideration in CFD. For the Molybdenum99
irradiation facility, grid generation poses the most challenging task because of its
inherent intricate geometrical details. In the three case designs considered, the
construction of suitable meshes accounts for almost the entire reactor design and
analysis. Owing to the complexity of geometry, application of structured and/or
unstructured meshes is required.
By definition, a structured mesh is a mesh containing cells having either a regularshape
element with fournodal corner points in two dimensions or a hexahedralshape element with
eightnodal corner points in three dimensions. Commonly applied in numerous CFD
investigations, it basically deals with the straightforward prescription of either an orthogonal
mesh or a bodyfitted mesh. An unstructured mesh can however be described as a mesh
Pipe Flow Analysis of Uranium Nuclear Heating with Conjugate Heat Transfer 273
overlaying with cells in the form of either a triangleshape element in two dimensions or a
tetrahedronshape in three dimensions.
For a bodyfitted mesh, the mesh construction of the internal region of the physical
domain can normally be achieved via two approaches. On one hand, the Cartesian coordinates
may be algebraically determined through interpolation from the boundary values. This
methodology requires no iterative procedure and it is computationally inexpensive. On the
other hand, a system of partial differential equations of the respective Cartesian coordinates
may be solved numerically with the set of boundary values as boundary conditions in order to
yield a highly smooth mesh in the physical domain. The former is commonly known as the
transfinite interpolation method and the latter is typically the elliptic grid generation method
(Smith, 1982 and Thompson, 1982).
For an unstructured mesh, triangle and tetrahedral meshing are by far the most common
forms of unstructured grid generation. In Delaunay meshing, this most commonly adopted
grid generation procedure entails the initial set of boundary nodes of the geometry to be
triangulated according to the Delaunay triangulation criterion. Here, the most important
property of a Delaunay triangulation is that it has the empty circumcircle (circumscribing
circle) property (Shewchuk, 2002). All algorithms for computing Delaunay triangulations rely
on the fast operations for detecting when a grid point is within a triangle's circumcircle and an
efficient data structure for storing triangles and edges. The most straightforward way of
computing the Delaunay triangulation is to repeatedly add one vertex at a time, then
retriangulating the affected parts thereafter. When a vertex is added, a search is done for all
triangles’ circumcircles containing the vertex. Then, those triangles are removed whose
circumcircles contain the newly inserted point. All new triangulation is then formed by
joining the new point to all boundary vertices of the cavity created by the previous removal of
intersected triangles. Delaunay triangulation techniques based on point insertion extend
naturally to three dimensions by considering the circumsphere (circumscribing sphere)
associated with a tetrahedron. More details on Delaunay triangulation and meshing can be
referred in Mavriplis (1997). Besides Delaunay method, other meshing algorithms in
unstructured grid generation include the advancing front method (Lo, 1985; Gumbert et al.,
1989; Marcum and Weatherill, 1995) and quadtree/octree method (Yerry and Shepard, 1984;
Shepard and Georges, 1991).
It should be noted that the use of hybrid grids that combine different element types such
as triangular and quadrilateral in two dimensions or tetrahedral, hexahedral, prisms and
pyramids in three dimensions can provide the maximum flexibility in matching mesh cells
with the boundary surfaces or internal solid regions where heat transfer due to conduction
needs to be resolved, and allocating cells of various element types in other parts of the
complex flow regions. Grid quality can be enhanced through the placement of quadrilateral or
hexahedral elements in resolving boundary layers near solid walls or composite materials in
solid regions within the pipe flow systems whilst triangular or tetrahedral elements are
generated for the rest of the flow domain. This generally leads to both accurate solutions and
better convergence for the numerical solution methods.
G.H. Yeoh and M.K.M. Ho 274
Outline of CFD Model
Governing Equations
The unsteady RANS equations for the CFD model that reflect the conservation of mass,
momentum and energy can be written as:
Mass
( ) ( ) ( )
0
u w
t x y z
ρ ρ ρ
ρ
∂ ∂ ∂
∂
+ + + =
∂ ∂ ∂ ∂
v
(1)
xMomentum
( ) ( ) ( ) ( )
xy
xx xz
u
u uu u wu
t x y z
u u u
S
x x y y z z x y z
ρ ρ ρ ρ
τ
τ τ
μ μ μ
∂ ∂ ∂ ∂
+ + + =
∂ ∂ ∂ ∂
′ ∂ ′ ′ ⎡ ⎤ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ⎡ ⎤ ⎡ ⎤
+ + + + + +
⎢ ⎥
⎢ ⎥ ⎢ ⎥
∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂
⎣ ⎦ ⎣ ⎦
⎣ ⎦
v
(2)
yMomentum
( ) ( ) ( ) ( )
xy yy yz
u u w
t x y z
S
x x y y z z x y z
ρ ρ ρ ρ
τ τ τ
μ μ μ
∂ ∂ ∂ ∂
+ + + =
∂ ∂ ∂ ∂
′ ′ ′ ∂ ∂ ∂
⎡ ⎤ ∂ ∂ ∂ ∂ ∂ ∂ ⎡ ⎤ ⎡ ⎤
+ + + + + +
⎢ ⎥
⎢ ⎥ ⎢ ⎥
∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂
⎣ ⎦ ⎣ ⎦
⎣ ⎦
v
v v v v
v v v
(3)
zMomentum
( ) ( ) ( ) ( )
yz
xz zz
w
w u w w ww
t x y z
w w w
S
x x y y z z x y z
ρ ρ ρ ρ
τ
τ τ
μ μ μ
∂ ∂ ∂ ∂
+ + + =
∂ ∂ ∂ ∂
′ ∂ ′ ′ ⎡ ⎤ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ⎡ ⎤ ⎡ ⎤
+ + + + + +
⎢ ⎥
⎢ ⎥ ⎢ ⎥
∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂
⎣ ⎦ ⎣ ⎦
⎣ ⎦
v
(4)
Energy
( ) ( ) ( ) ( )
x y z
y
x z
q q q
H u H H wH
p
t x y z t
q
q q T T T
x x y z z z x y z
ρ ρ ρ ρ
λ λ λ
∂ ∂ ∂ ∂
∂
+ + + = + Φ+
∂ ∂ ∂ ∂ ∂
′ ∂ ′ ′ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂
+ + + + +
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ⎣ ⎦ ⎣ ⎦ ⎣ ⎦
v
(5)
Pipe Flow Analysis of Uranium Nuclear Heating with Conjugate Heat Transfer 275
From the above equations,
( )
indicates timeaveraged quantities, t is the time, ρ is the
density, u, v and w are the respective velocity components along the Cartesian coordinate
directions x, y and z, H is the enthalpy, T is the temperature, μ is the fluid dynamic viscosity
and λ is the fluid thermal conductivity. The timeaveraged source or sink terms
u
S , S
v
and
w
S are given by
u
p u w
S
x x x y x z x
μ μ μ
′ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤
= − + + +
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
∂ ∂ ∂ ∂ ∂ ∂ ∂
⎣ ⎦ ⎣ ⎦ ⎣ ⎦
v
(6)
( )
ref
p u w
S g
y x y y y z y
μ μ μ ρ ρ
′ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ∂ ∂ ∂ ∂ ∂ ∂ ∂
= − + + + − −
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
∂ ∂ ∂ ∂ ∂ ∂ ∂
⎣ ⎦ ⎣ ⎦ ⎣ ⎦
v
v
(7)
w
p u w
S
z x z y z z z
μ μ μ
′ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤
= − + + +
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
∂ ∂ ∂ ∂ ∂ ∂ ∂
⎣ ⎦ ⎣ ⎦ ⎣ ⎦
v
(8)
From above, p′ is the modified averaged pressure defined by
2
3
p p k ρ ′ = + +
2
3 ref i i
g x μ ρ ∇⋅ − V where p is the dynamics pressure, k is the turbulent kinetic energy,
( ) , , u w ≡ V v is the velocity vector,
ref
ρ is the reference density, g
i
are the gravitational
acceleration components ( ) 0, , 0 g − and x
i
are the coordinates relative the Cartesian datum
( ) , , x y z and the effects due to the viscous stresses in the energy equation (5) are described
by the averaged dissipation function Φ. For weakly compressible flows, it is common
practice to transform the energy equation by replacing the heat flux according to the local
enthalpy gradient instead of the temperature gradient, viz.,
x
H
q
Pr x
μ ∂
=
∂
y
H
q
Pr y
μ ∂
=
∂
z
H
q
Pr z
μ
= −
∂
(9)
where Pr is the Prandtl number. Also, the pressure work term p t ∂ ∂ and the averaged
dissipation function Φ that represents the source of energy due to work done deforming the
fluid element are usually ignored in most practical applications.
On the basis of the eddy viscosity model, the Reynolds stresses defined by the stress
vector
ij
τ ′ ′ ≡ τ where i, j = x, y, z in the momentum equations (2)–(4) is given by
( )
( )
2 2
3 3
T
ij t t
k τ μ μ δ ρ δ ′ ′ − ≡ − = ∇ + ∇ − ∇⋅ − V V V τ (10)
G.H. Yeoh and M.K.M. Ho 276
where
t
μ is the turbulent eddy viscosity. The Reynolds flux in the energy equation (5)
defined by
i
q′ may also be modelled analogously to the eddy viscosity hypothesis as
t
i
t i
H
q
Pr x
μ ∂
=
∂
(11)
where
t
Pr is the turbulent Prandtl number. To satisfy dimensional requirements, at least two
scaling parameters are required to relate the Reynolds stress to the rate of deformation. In
most engineering flow problems, the complexity of turbulence precludes the use of any
simple formulae. A feasible choice is the turbulent kinetic energy k and another turbulent
quantity which is the rate of dissipation of turbulent energy ε . The local turbulent viscosity
t
μ can be obtained either from dimensional analysis or from analogy to the laminar viscosity
as
t t
v l μ ρ ∝ . On the latter definition, based on the characteristic velocity
t
v defined as k
and the characteristic length l as
3/ 2
k ε , the turbulent viscosity
t
μ can thus be ascertained
according to
2
t μ
k
C μ ρ
ε
= (12)
where C
μ
is an empirical constant. In order to evaluate the turbulent viscosity in equation
(12), the values of k and ε must be known which are generally obtained through solution of
their respective transport equations.
After a fair amount of algebra, the final forms of equations for the standard kε model
developed by Launder and Spalding (1974) for the turbulent kinetic energy k and dissipation
of turbulent energy ε can be written as:
( ) ( ) ( ) ( )
t t t
k k k
k u k k wk
t x y z
k k w
P G
x x y y z z
ρ ρ ρ ρ
μ μ μ
ρε
σ σ σ
∂ ∂ ∂ ∂
+ + + =
∂ ∂ ∂ ∂
⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ∂ ∂ ∂ ∂ ∂ ∂
+ + + + −
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
∂ ∂ ∂ ∂ ∂ ∂
⎣ ⎦ ⎣ ⎦ ⎣ ⎦
v
(13)
( ) ( ) ( ) ( )
( )
1 3 2
t t t
u w
t x y z
C P C G C
x x y y z z k
ε ε
ε ε ε
ρε ρ ε ρ ε ρ ε
μ μ μ ε ε ε ε
ρε
σ σ σ
∂ ∂ ∂ ∂
+ + + =
∂ ∂ ∂ ∂
⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ∂ ∂ ∂ ∂ ∂ ∂
+ + + + −
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
∂ ∂ ∂ ∂ ∂ ∂
⎣ ⎦ ⎣ ⎦ ⎣ ⎦
v
(14)
where P is the shear production defined by
Pipe Flow Analysis of Uranium Nuclear Heating with Conjugate Heat Transfer 277
( )
( )
( )
2
3
T
t t
P k μ ρ μ = ∇ ⋅ ∇ + ∇ − ∇⋅ + ∇⋅ V V V V V (15)
and G is the production due to the gravity, which is valid for weakly compressible flows can
be written as
t
G
ρ
μ
ρ
ρσ
= − ⋅ ∇ g (16)
where ( ) 0, , 0 g ≡ − g is the gravity vector and
3
C and
ρ
σ are normally assigned values of
unity and G in equation (16) is the imposed condition whereby it always remains positive,
i.e. ( ) max , 0 G The constants for the standard kε model have been arrived through
comprehensive data fitting for a wide range of turbulent flows (see Launder and Spalding,
1974): C
μ
= 0.09, σ
k
= 1.0, σ
ε
= 1.3,
1
C
ε
= 1.44 and
2
C
ε
= 1.92. Note that the effect of
buoyancy is included in equations (14) and (15) to account for regions in the fluid with very
low flow velocities.
For the solid regions comprising of various materials, the heatconduction equation can
be expressed as
s s s s
s ps s s s
T T T T
C S
t x x y y z z
ρ λ λ λ
⎡ ⎤ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ⎡ ⎤ ⎡ ⎤
= + + +
⎢ ⎥
⎢ ⎥ ⎢ ⎥
∂ ∂ ∂ ∂ ∂ ∂ ∂
⎣ ⎦ ⎣ ⎦
⎣ ⎦
(17)
where T
s
is the solid phase temperature and S
is the internal volumetric heat source
accounting for the total heat generated due to the irradiation process. In equation (17),
s
ρ ,
ps
C and λ
s
denote the thermophysical properties of the solid materials.
Boundary Conditions
Based on the mass flow rate, the normal velocity can be calculated and prescribed at the
inlet boundary. The temperature of the fluid is also prescribed at this boundary. At the outlet,
the static pressure is normally imposed according to fullydeveloped flow condition. For all
variables, the normal derivative at the outlet boundary is equivalent to zero.
One possible approach of overcoming the difficulty of modeling the nearwall region of a
solid wall is through the prescription of logarithmic wall functions for the standard kε model.
In order to construct these functions, the region close to the wall can usually be characterized
by considering the dimensionless velocity U
+
and wall distance y
+
with respect to the local
conditions at the wall. The dimensionless wall distance y
+
is defined as ( ) u d y
τ
ρ μ −
where very near the wall, y = d, while the dimensionless velocity U
+
can be expressed in the
G.H. Yeoh and M.K.M. Ho 278
form as U u
τ
where U is taken to represent some averaged velocity representing either of
the mixture velocity parallel to the wall, u
τ
is the wall friction phase velocity which is
defined with respect to the wall shear stress
w
τ as
w
τ ρ .
For wall distance of y
+
< 5, the boundary layer is predominantly governed by viscous
forces that produce the noslip condition; this region is subsequently referred to as the viscous
sublayer. By assuming that the shear stress is approximately constant and equivalent to the
wall shear stress
w
τ , a linear relationship between the averaged velocity and the distance
from the wall can be obtained yielding
U y
+ +
= for
0
y y
+ +
< (18)
With increasing wall distance y
+
, turbulent diffusion effects dominate outside the viscous
sublayer. A logarithmic relationship is employed:
( )
1
ln U Ey
κ
+ +
= for
0
y y
+ +
> (19)
The above relationship is often called the loglaw and the layer where the wall distance
y
+
lies between the range of 30 < y
+
< 500 is known as the loglaw layer. Values of κ
(~0.4) and E (~9.8) in equation (19) are universal constants valid for all turbulent flows past
smooth walls at high Reynolds numbers. The crossover point
0
y
+
can be ascertained by
computing the intersection between the viscous sublayer and the logarithmic region based on
the upper root of
( )
0 0
1
ln y Ey
κ
+ +
= (20)
A similar universal, nondimensional function can also be constructed to the heat transfer.
The enthalpy in the wall layer is assumed to be:
H Pry
+ +
= for
H
y y
+ +
< 21)
( )
ln
T
H
Pr
H F y
κ
+ +
= for
H
y y
+ +
> (22)
where
H
F is determined by using the empirical formula of Jayatilleke (1969):
Pipe Flow Analysis of Uranium Nuclear Heating with Conjugate Heat Transfer 279
0.75
exp 9.0 1 1 0.28exp 0.007
H
t t
Pr Pr
F E
Pr Pr
κ
⎧ ⎫ ⎡ ⎤
⎡ ⎤ ⎛ ⎞ ⎛ ⎞
⎪ ⎪
⎢ ⎥ = − + −
⎨ ⎬ ⎢ ⎥ ⎜ ⎟ ⎜ ⎟
⎢ ⎥
⎝ ⎠ ⎝ ⎠ ⎣ ⎦ ⎪ ⎪
⎣ ⎦ ⎩ ⎭
(23)
By definition, the dimensionless enthalpy H
+
is given by:
( )
0.25 0.5
w
H
H H C k
H
J
μ
ρ
+
−
= (24)
where
w
H is the value of enthalpy at the wall and the diffusion flux J
H
is equivalent to the
normal gradient of the enthalpy
( )
wall
H n ∂ ∂ perpendicular to the wall. The thickness of the
thermal conduction layer is usually different from the thickness of the viscous sublayer, and
changes from fluid to fluid. As demonstrated in equation (20), the crossover point
H
y
+
can
also be similarly computed through the intersection between the thermal conduction layer and
the logarithmic region based on the upper root of
( )
1
ln
H T H H
Pr y Pr F y
κ
+ +
= (25)
For the rest of the boundary conditions at a solid wall, all wall temperatures are
determined using the energy balance. Boundary conditions for the turbulent kinetic energy
and the dissipation have its normal derivative at the wall equal to zero and obtained through
the relation
3 4 3 2
C k
d
μ
ε
κ
= (26)
where d is the distance of the nearest grid point from the wall boundary.
Computational Procedure
The algebraic forms of the governing equations to be solved by matrix solvers can be
formed by integrating the system of equations using the finite volume method over small
control volumes. By employing the general variableφ , the generic form of the governing
equations can be written initially in the form as
( )
( )
S
t
φ φ
ρφ
ρ φ φ
∂
⎡ ⎤ +∇⋅ = ∇⋅ Γ ∇ +
⎣ ⎦
∂
V (27)
G.H. Yeoh and M.K.M. Ho 280
In order to bring forth the common features, terms that are not shared between the
equations are placed into the source term S
φ
. By setting the transport variable φ equal to 1,
u , v , w, H , k and ε and selecting appropriate values for the diffusion coefficient
φ
Γ
and source term S
φ
, special forms of each of the partial differential equations for the
continuity, momentum and energy as well as for the turbulent scalars can thus be obtained.
The cornerstone of the finite volume method is the control volume integration. In order to
numerically solve the approximate forms of equation (27), it is convenient to consider its
integral form of this generic transport equation over a finite control volume. Integration of the
equation over a threedimensional control volume ΔV yields:
( )
( )
V V V V
dV dV dV S dV
t
φ φ
ρφ
ρ φ φ
Δ Δ Δ Δ
∂
⎡ ⎤ + ∇⋅ = ∇⋅ Γ ∇ +
⎣ ⎦
∂
∫ ∫ ∫ ∫
V (28)
By applying the Gauss’ divergence theorem to the volume integral of the advection and
diffusion terms, equation (28) can now be expressed in terms of the elemental dA as
( )
( )
V A A V
dV dA dA S dV
t
φ φ
ρφ
ρ φ φ
Δ Δ Δ Δ
∂
⎡ ⎤ + ⋅ = Γ ∇ ⋅ +
⎣ ⎦
∂
∫ ∫ ∫ ∫
V n n (29)
Equation (29) needs also to be further augmented with an integration over a finite time step
Δt. By changing the order of integration in the time derivative terms,
( )
( )
t t t t
V t t A
t t t t
t A t V
dt dV dA dt
t
dA dt S dVdt
φ φ
ρφ
ρ φ
φ
+Δ +Δ
Δ Δ
+Δ +Δ
Δ Δ
⎛ ⎞ ⎛ ⎞ ∂
+ ⋅ =
⎜ ⎟ ⎜ ⎟
∂
⎝ ⎠ ⎝ ⎠
⎛ ⎞
⎡ ⎤ Γ ∇ ⋅ +
⎜ ⎟
⎣ ⎦
⎝ ⎠
∫ ∫ ∫ ∫
∫ ∫ ∫ ∫
V n
n
(30)
In essence, the finite volume method discretises the integral forms of the transport
equations directly in the physical space. If the physical domain is considered to be subdivided
into a number of finite contiguous control volumes, the resulting statements express the exact
conservation of property
k
φ from equation (21) for each of the control volumes. In a control
volume, the bounding surface areas of the element are, in general, directly linked to the
discretisation of the advection and diffusion terms. The discretised forms of these terms are:
( ) ( )
k f
f
f
A
dA A ρ φ ρ φ
Δ
⋅ ≈ ⋅ Δ
∑
∫
V n V n (31)
Pipe Flow Analysis of Uranium Nuclear Heating with Conjugate Heat Transfer 281
f
f
f
A
dA A
φ φ
φ φ
Δ
⎡ ⎤ ⎡ ⎤ Γ ∇ ⋅ ≈ Γ ∇ ⋅ Δ
⎣ ⎦ ⎣ ⎦
∑
∫
n n (32)
where the summation in equations (31) and (32) is over the number of faces of the element
and
f
A Δ is the area of the face of the control volume. The source term can be subsequently
approximated by:
V
S dV S V
φ φ
Δ
≈ Δ
∫
(33)
For the time derivative term, the commonly adopted first order accurate approximation
entails:
( ) ( ) ( )
1 n n
t t
t
dt
t t
ρφ ρφ ρφ
+
+Δ
∂ −
≈
∂ Δ
∫
(34)
where Δt is the incremental time step and the superscripts n and n + 1 denote the previous and
current time levels respectively. Equation (21) can then be iteratively solved accordingly to
the fully implicit procedure by
( ) ( )
( )
1
1
1
1
n
n n
f
f
f
n
n
f
f
f
A
t
A S V
φ φ
ρφ ρφ
ρ φ
φ
+
+
+
+
⎛ ⎞ −
+ ⋅ Δ =
⎜ ⎟
Δ
⎝ ⎠
⎛ ⎞
⎡ ⎤ Γ ∇ ⋅ Δ + Δ
⎜ ⎟
⎣ ⎦
⎝ ⎠
∑
∑
V n
n
(35)
Consider the particular control volume element in question of which point P is taken to
represent the centriod of the control volume, which is connected with the respective centroids
of other surrounding control volumes. Equation (35) can thus be expressed in terms of the
transport quantities at point P and surrounding nodal points with a suitable prescription of
normal vectors at each control volume face and dropping the superscript n +1 which by
default denotes the current time level as:
( )
1 1 1 1 1
n
P n n n n n P
P P nb nb off non u P
nb
V
a a S S S V
t
ρφ
φ φ
+ + + + +
Δ
= + + + Δ +
Δ
∑
(36)
where
( )
1
1
n
P n P
P nb P P f
nb
V
a a S V F
t
ρφ
+
+
Δ
= + Δ + +
Δ
∑ ∑
(37)
G.H. Yeoh and M.K.M. Ho 282
and the added contribution due to nonorthogonality of the mesh is given by
1 n
non
S
+
, which is
required to be ascertained especially for bodyfitted and unstructured meshes. Note that the
convective flux is given by
( )
1
1
n
n
f f
f
F A ρ φ
+
+
= ⋅ Δ V n For the sake of numerical treatment, the
source term for the control volume in equation (35) has been treated by
( )
1 1 1 n n n
u P P P
S V S S V
φ
φ
+ + +
Δ = − Δ (38)
In equation (36),
P
a is the diagonal matrix coefficient of
1 n
P
φ
+
,
1 n
f
F
+
∑
are the mass
imbalances over all faces of the control volume and
P
S is the coefficient that is extracted
from the treatment of the source term in order to further increase the diagonal dominance. The
coefficients of any neighboring nodes for any surrounding control volumes
k
nb
a in equation
(36) can be expressed by
( )
1 1
max , 0
n n
nb f f
a D F
+ +
= + − (39)
where
1 n
f
D
+
is the diffusive flux containing the diffusion coefficient
φ
Γ along with the
geometrical quantities of the particular element within the mesh system. The treatment of the
advection term which results in the form presented in equation (39) is known as upwind
differencing to guarantee diagonal dominance. In order to reduce the effect of false diffusion
caused by upwind differencing, the wellknown deferred correction approach is adopted to
treat the offdiagonal contributions
1 n
off
S
+
in equation (36) due to higher resolution
differencing schemes.
In this study, the coupled solution approach, a more robust alternative to the segregated
approach, is adopted to solve the velocity and pressure equations simultaneously. Algebraic
Multigrid accelerated Incomplete Lower Upper (ILU) factorization technique is employed to
resolve each of the discrete system of linearized algebraic equations in the form of equation
(36). The advantages of a coupled treatment over a segregated approach are: robustness,
efficiency, generality and simplicity. Nevertheless, the principal drawback is the high storage
requirements for all the nonzero matrix entries.
Results and Discussion
For the analysis of reactor thermohydraulic safety, the central focus often befalls on a
‘bounding case study’ which demarks what can be loosely regarded as the operational
envelop of the reactor cooling system. This bounding case study is a simulation of all extreme
operational conditions that may occur simultaneously during the life of the reactor. The
meaningful operands of these bounding operational conditions are represented by input
parameters (such as coolant temperatures, mass flow, etc.) of the system which become the
parametric specifications of the computer model.
Pipe Flow Analysis of Uranium Nuclear Heating with Conjugate Heat Transfer 283
However, before these inputs are specified, a computer model of the irradiation rig must
first be generated; complete with flow passages, internal structures and irradiation target(s).
Components of the model rig assembly must be built in accordance to engineering blueprints
and where possible, minor geometric simplifications are introduced into the model so that
simulations remained physically representative without having to waste computational
resources or prolong solution times over unnecessarily detailed control volumes. After the
model is generated, it is geometrically discretised by the generation of a ‘mesh’ which further
segregates each control volumes into subvolumes. The physical properties of each geometric
feature – such as the aluminium in the rig structure are then specified in the corresponding
volumes of the model.
It is standard practice to conduct a gridconvergence test before the modelled geometry is
deemed representative of the physical prototype. To perform this, the amount of control
volumes is increased in all three axes until the increase in mesh resolution does not result in
any appreciable difference in final key results, such as in the model’s maximum attainable
temperature. Finally, the results are analysed and compared to the safety limit.
This procedure for this safety analysis can be summarised in five steps:
1. Determining the bounding case scenario
2. Modelling the geometry, the physics and physical properties
3. Solving the CFD model by numerical approximation techniques
4. Checking the validity of our solution by mesh sensitivity analysis as well as by
comparisons with other simpler numerical models such as from onedimensional
simulations reactor nuclear thermohydraulic code
5. Analysing and comparing results with safety requirement
‘RocketCan’ Design as Used in the HIFAR Reactor
The first pipecooling irradiation system to be examined is the ‘rocketcan’ design used
in HIFAR. This study investigated the maximum temperature of 2.2%
235
U enriched UO
2
pellets during irradiation.
The recently decommissioned HIFAR reactor, a 10 MW nuclear research reactor at the
Australian Nuclear Science and Technology Organisation (ANSTO), produced a steady
supply of technicium99 (
99
Tc) radiopharmaceutical for domestic and international use.
Technicium99 is formed by the radioactivedecay of Molybdenum99 (
99
Mo), which itself is
a fission product of Uranium235 (
235
U).
The process of generating technicium99 started with the loading of seventeen UO
2
pellets into a thick aluminium tube shaped like a ‘rocketcan’ (Figure 2). Granulated
magnesium oxide (MgO) was packed in with the pellets to assist heat conduction and to
control pellet spacing. The rocketcans were then sealed by welding on a cap and inserted into
slots inside a long vertical ‘stringer’ assembly which looked much like a ‘cakestand’. In
turn, the stringer was inserted into a ‘liner’, a hollow tube with flow bypass inlets at its
conical tip. Finally the liner was inserted into the centre of four concentric annular fuel plates
which formed the fuel assembly (Figure 1). The purpose of the liner was to separate high
speed flow passing through the four outer concentric fuel plates from the slower flow going
through the liner cooling the cans. All components: cans, stringer and liner were fabricated
G.H. Yeoh and M.K.M. Ho 284
using aluminium on account of its unique qualities of neutron transparency, anticorrosion
and good thermal conductivity.
Figure 1. Schematic drawing of the HIFAR fuel element.
When the HIFAR reactor was operating,
235
U in fuel elements underwent fission which
absorbed and generated neutrons in a selfsustaining process. The nuclear reaction produced
large amounts of heat that was removed by upwardly flowing heavy water (deuterium oxide,
D
2
O) through the fuel element. One and a half percent of coolant flow bypassed into the
centre liner for rocketcan cooling. Also, the neutrons produced by the nuclear process in the
fuel were absorbed by the UO
2
pellets to produce molybdenum that had a short halflife of 2.7
days before betadecaying into technicium99. After irradiation, the molybdenum99 was
chemically separated and quickly packaged so that it arrived at clinics in its usable beta
decayed form as technicium.
IDENTIFICATION
NUMBER
SALVAGE COUPLING PHALANGE
FOR ASSEMBLY TO SHIELD PLUG
EMERGENCY COOLING
WEIR AND SPRAY RING
INTERMEDIATE
SECTION
COOLANT FLOW
THERMOCOUPLE TUBE
PERFORATED
EXTENSION
THERMOCOUPLE
COMB
EMERGENCY COOLING
WATER TUBE (NOT
USED IN HIFAR)
THERMOCOUPLE
TUBE
VIEW OF ARROW A
DOWEL
FUEL TUBES
OUTLINE OF
LINER
LOWER COMB
GUIDE NOSE
SPHERICAL
SEAT
SKIRT
COOLANT
FLOW
Pipe Flow Analysis of Uranium Nuclear Heating with Conjugate Heat Transfer 285
Figure 2. Schematic drawing of the inadiation rig and placement of target cans in the liner.
During irradiation, a vast amount of heat was generated in the uranium pellets which
must be evacuated. For safety and licensing purposes, it was necessary to demonstrate that the
pellet would not melt during irradiation. Thus, neutronic and thermalhydraulic analyses
using numerical methods were used to determine the pellet maximum temperature. This
method directly solved the three conservation equations of mass, momentum and energy in
the coolant flow domain which was coupled with the conduction physics of the solid domains
of the can, magnesium oxide and uranium oxide pellets. The result of this work was critical to
the licensing requirements of HIFAR as bounded by operation licence conditions (OLCs) and
evaluated safety limits.
Grid Generation of RocketCan
The complete geometry of can, stringer, liner, fuel plates and flow passages was
generated simultaneously and their respective volumes patched as separate logical entities as
shown in Figure 3. The simultaneous volume generation meant that separate component
surfaces were logically connected and no surface patching was required between components.
G.H. Yeoh and M.K.M. Ho 286
On the outset, minor geometric details were simplified in regions of little flowdynamic
consequence inorder to accelerate the time to solution convergence. Some work was needed
to organise the arrangement of each entity with respects to other parts but the effort expended
was outweighed by the benefit of not needing to specify intracomponent surface patching.
Note in Figure 3 that hexahedral bodyfitted volumes were used in the geometry to properly
model the thin conduction regions of the liner and fuel plates. The thin solid region of the
liner and stringer could only be practically constructed by using a structured mesh because of
the high thermal flux across the thin stretch of aluminium. Numerical diffusion over such a
thin area of high thermal flux given if we were using unstructured mesh would have been
exceedingly pronounced. The additional benefit of using hexahedral control volumes was in
their superiority over tetrahedral elements for the modelling of nearwall flows where the
correlationmodel of wall flowprofiles work best with bodyfitted rectangular mesh. This
was also represented an economy in the number of rectangular mesh required as compared to
tetrahedral mesh to attain the same level of solution accuracy.
The pelletstack geometry consisted of a 119.5 mm vertical cylinder of MgO with a
radius of 5 mm. The pellets modelled inside the MgO stack was 3.5mm high with a radius
4.5mm as shown in Figure 4a. In Figure 5a, the rocketcan and stringer are shown
simultaneously. Notice the relief windows at the top of the stringer which allowed the coolant
flow to pass through as was identical to the real physical prototype. Finally, five volumes of
the same model as shown in Figure 3b were stacked on each other to produce the complete
stringer assembly such as shown in Figure 9.
(a)
(b)
Figure 3. The modelled ‘block’ of multiple components is shown with the (a) rocketcan and (b) slotted
stringer highlighted.
Pipe Flow Analysis of Uranium Nuclear Heating with Conjugate Heat Transfer 287
(a)
(b)
Figure 4. The modelled stack of (a) UO2 pellets in MgO was built to fit inside (b) the rocketcan’s
internal cavity.
(a)
(b)
Figure 5. The model of (a) rocketcan inserted inside stringer and (b) multiple stringers were assembled
to produce a complete stringer assembly.
G.H. Yeoh and M.K.M. Ho 288
Specification of Material Properties
The total D
2
O mass flow through the rig was 16 kg/s with 98.5% passing through the fuel
and 1.5% passing through the liner. The material properties of aluminium and heavy water are
summarised in Table 1 and the variable conductivities of magnesium oxide and uranium
dioxide are shown in Figure 6.
Table 1. Material Properties of Aluminium and Heavy Water
MATERIAL PROPERTIES Al D
2
O
Density kg.m
3
2702 1094.92
Specific Heat Capacity J.kg
1
.K
1
903 4.12849
Thermal Conductivity W.m
1
.K
1
273 0.614259
Dynamic Viscosity μ Pa.s 0.000712
MagnesiumOxide conductivity vs. Temp.
0.2
0.4
0.6
0.8
1.0
1.2
1.4
1.6
1.8
2.0
300 3000
Temp. [ K]
T
h
e
r
m
a
l
C
o
n
d
u
c
t
i
v
i
t
y
[
W
/
(
m
.
K
)
]
(a)
UraniumOxide conductivity vs. Temp.
2
3
4
5
6
7
8
200 418 1800 3000
Temp. [ K]
T
h
e
r
m
a
l
C
o
n
d
u
c
t
i
v
i
t
y
[
W
/
(
m
.
K
)
]
(b)
Touloukian 1970.
Figure 6. Temperature dependent thermal conductivities of (a) magnesium oxide and (b) uranium
dioxide.
Computational Geometry
The domains of the simulation were specified as follows (the default soliddomain
material was aluminium):
• ‘Fuel1’, ‘Fuel2’, ‘Fuel3’, ‘Fuel4’– Solid Domain
• ‘Liner’ – Solid Domain
• ‘Stringer’ – Solid Domain
• ‘Can’ – Solid Domain
• ‘MgO’ – Solid Domain, material: Magnesium Oxide
• ‘UO
2
’ – Solid Domain, material: Uranium Oxide
• ‘Fluid’ – Fluid Domain, fluid: heavy water
Pipe Flow Analysis of Uranium Nuclear Heating with Conjugate Heat Transfer 289
In summary the CFD problem can be simplified as a list of input and outputs:
Inputs
• Geometry
• Material Properties
• Heat Source
• Boundary Conditions
Outputs
• Temperature Profile T (x,y,z)
• Wall Heat Flux q (x,y,z)
UO2 pellets producde heat at a steady rate. The heat was conducted through the MgO
into the aluminium rocket can then into the flowing heavy water D
2
O. The solutions of
interest given by CFD included: the temperature profile T (x,y,z) of the whole region and wall
heat flux q (x,y,z) over the surface of the rocket can.
The subdomain ‘Pellets’ was defined for the entire UO
2
domain as an energy source with
a volumetric power of 8.4556 x 10
8
W.m
3
. The fuel plates, the structural material
(aluminium) and the coolant (heavy water) would also produce heat but the heating from fuel
plates was not modeled because it was known that they were cooled efficiently by the faster
flowing coolant outside the liner and heating from other aluminium components was
negligible.
For the boundary conditions, there were two inlets, one at ‘INLET FUEL’ with Normal
Speed 2.58816 m.s
1
, another at ‘INLET LINER’ with Normal Speed 0.0717642m.s
1
(these
were calculated from coolant mass flow value, inlet areas and bypass ratios), both with a
static temperature of 318K (45°C) in the 16pellet simulation. In these simulations, there were
two separate flow regions.
Power Density Calculation
The power emitted from 2.2% enriched UO
2
pellets were calculated by using the Monte
Carlo NParticle (MCNP) transport code for simulating neutronic reactions. For this
safety analysis a conservative reactor power of 11MW was assumed. The 16 pellets of a can
were modeled discretely (Dia. 9mm × 3.5mm high). Total power per m
3
for 16 pellets =
3204.7 W.m
3
/ 3.79 × 10
6
m
3
(Volume of 16 Pellets) = 8.4556 × 10
8
W.m
3
. This power
density signified the source term for the conduction equation.
Turbulence Solver and Convergence Criteria
To account for the turbulent pipe flow, the standard kε model was used. Computational
predictions were deemed to be converged when the normalized residual mass was less than 1
x 10
4
.
Computational Predictions
Simulation for the heating of sixteen pellets in a rocket can had been previously modeled
by Yeoh and Storr (2000) which was interested in the rocketcan’s maximum surface heat
G.H. Yeoh and M.K.M. Ho 290
flux to determine the margin for Onset of Nucleate Boiling (ONB). The CFD model of Yeoh
& Storr’s simulation had a lesser degree of resolution and modeled the stack of pellets as a
single entity instead of discrete pellets. This was partly attributed to the more limited
computational resources available at the time. Yeoh & Storr’s CFD result’s were validated to
a degree by simple irradiation experiments where metal samples were introduced inside the
pellet stack and subsequently examined for evidence of melting in order to estimate
maximum temperatures. In this investigation calculated by CFX10, we were primarily
interested in the maximum temperature of the UO
2
pellets. Temperature contours in Figure 7a
indicate a maximum pellet temperature of 2656°C which remained below the pellet melting
temperature of 2847°C. For clarity, the four concentric annular fuel plates, liner, stringer and
can in Figure 7 are shown in silhouette. This result was consistent with the maximum
temperature of Yeoh & Storr’s largevolume pellet stack at 2277°C. The temperature was
understandably lower in Yeoh & Storr’s simulation because the same power was distributed
over a larger volumetric space.
(a)
(b)
Figure 7. Section view of temperature contours showing maximum temperatures of (a) UO2 pellets:
2656°C and (b) MgO: 2439°C.
Other maximum temperatures of interest include: the magnesium oxide at 2439°C (mp.
2800°C) in Figure 7b; the aluminium can at 307°C (mp. 660°C) in Figure 8a and the heavy
water D
2
O at 79°C (bp. 120°C at 2 atm) in Figure 8b. The temperatures were highest in the
middle of the pellets, decreasing steeply across the width of the pellet and still more steeply
across the MgO powder, reflecting its low thermal conductivity. This was in contrast with the
low thermal gradient across the aluminium rocket can, due to the high thermal conductivity of
aluminium.
Figures 9 and 10 show the wall heat flux at the surface of rocket cans, the higher fluxes
were colored blue, due to outward flux being defined as negative. There were four high heat
flux regions corresponding to the slot opening. Opposite the slot opening in Figure 10, forced
convection cooling was restricted by the stringer enclosure and as a consequence caused a
lower heatflux as indicated by the green region of the can surface. The sudden expansion of
the open slot at the side of the stringer created a highly turbulent region which assists in
Pipe Flow Analysis of Uranium Nuclear Heating with Conjugate Heat Transfer 291
forced convection cooling and this resulted in the highest thermal heat flux achieved as
indicated in Figure 10 by the dark blue region of the can.
(a)
(b)
Figure 8. Section view of temperature contours showing maximum temperatures of (a) rocket can:
307°C and (b) D2O: 79°C.
Figure 9. Heatflux contour at the surface of the rocket cans. The view is through the opening in the
stringer. The silhouette of the stringer is made out in transparent purple.
Inside the liner, the mean flow accelerated when squeezed into tighter paths between the
can and the stringer, as indicated by yellow and orange vectors in Figure 11 and decelerated
when entering an expansion, as shown by blue vectors near the tip of the rocket can. The
deceleration was basically due to reasons of mass conservation as explained by the Bernoulli
equation. The length of an arrow in Figure 11 is proportional to the flow velocity at that point.
The long red arrows belonged to the faster flow outside the liner, with velocities of up to
4m.s
1
.
G.H. Yeoh and M.K.M. Ho 292
Figure 10. Heatflux contour at the surface of the rocket cans. The view is through the enclosed back of
the stringer.
Figure 11. Flow velocity vector plot of coolant channels inside outside the liner.
Proposed ‘Annular Can’ Replacement Design for HIFAR
Flow visualization and LDV measurements were performed to better understand the fluid
flow around the narrow spaces within the X216 irradiation rig, prototypes of annular target
cans and liner. A threedimensional computational fluid dynamics model was used to
Pipe Flow Analysis of Uranium Nuclear Heating with Conjugate Heat Transfer 293
investigate the hydraulics behavior within a HIFAR fuel element liner model. An interesting
feature of the computational model was the use of unstructured meshing, which consisted of
triangular elements and tetrahedrons within the flow space (Figure 13), to model the “scaled
up” experimental model. This present investigation focused on the evaluation of the CFD
model in its capability to predict the complex flow structures inside the liner containing the
mockup X216 rig with two targets. The reliability of the model was validated against
experimental observations and measurements.
Description of the Water Tunnel Experimental Apparatus
and Methods
Figure 12 describes the transparent model of the prototype design of the rig and annular
target cans that were placed in the water tunnel facility. In this test section, the liner was
sealed so that the flow path was only through the liner inlet holes. An orifice plate was used
to measure the flow rates within the facility. Flow visualization of the fluid flow was
performed using an Argon Ion laser light sheet, highresolution digital camera and standard
video equipment. The flow was seeded with Iriodin powder concentrations of 0.5 to 1 gram
per 3000 liters of water. Particles in the flow were illuminated while they were in the field of
the laser light sheet. With the digital images and video footage the flow field was clearly
visible.
Figure 12. Points of velocity measurements in X216 rig.
G.H. Yeoh and M.K.M. Ho 294
Figure 13. Unstructured mesh of the X216 annular target.
Velocities were measured using a twodimensional (2D) Dantec LDV system, operated in
1D mode measuring axial velocity components. Data were stored on the computer attached to
the LDV hardware. Uncertainty in measurement using the LDV equipment was determined to
be a maximum of ± 3.5% for a particular optics configuration but generally less than ±1.8%.
Points of velocity measurement are shown in Figure 12. The points measured in the plane
shown had been taken with the laser entering at the open side of the mockup rig and the
forward scatter detector viewing through the closed side of the rig. Measurements were also
taken using the LDV probe in backscatter mode, and were verified at a number of points by
using the forward scatter mode. Backscatter mode was used only for measurements in the grid
pattern between the heights of 185 mm and 360 mm, since at the other locations the beams
were sufficiently attenuated due to the additional influences of the acrylic interfaces, giving
very low data rates.
Computational Details
A threedimensional CFD program ANSYSCFX5.6 has been employed to simulate the
complex thermalhydraulics behavior in the space within a HIFAR fuel element liner model
in the water tunnel. The CFD code solved the conservation equations of mass, momentum and
energy. Turbulence of the fluid flow was accounted via a standard kε model. Buoyancy
Pipe Flow Analysis of Uranium Nuclear Heating with Conjugate Heat Transfer 295
effect was included for regions where low velocities were present. This effect was included
within the source terms of the momentum equations as well as in the turbulence models.
The unstructured meshing was adopted for the construction of the computational model
of the mockup because of the inherent intricate details of the irradiation rig and the
placements of the annular targets within the rig. The mesh was prepared in two stages. A
surface mesh of triangular mesh elements was initially generated on all the surfaces covering
the model components. The volume mesh consisted of tetrahedral elements was then
subsequently generated within the fluid flow domain from the surface mesh elements. Figure
13 shows the mesh layout of triangular surface elements around the pin and support tube nose
cone of the irradiation rig. For the entire geometrical structure that included the assembly of
two annular target cans placed within the prototype rig in the liner and mockup fuel element,
a volume mesh of 605158 tetrahedrons and a surface mesh of 42030 triangular elements were
allocated.
The governing equations were solved by matrix solution techniques formulated by
integrating the system of equations using the finite volume method over small elemental
volumes. For each elemental volume, relevant quantity (mass, momentum and turbulence)
was conserved in a discrete sense for each control volume. Here, a coupled solver, which
solved the hydrodynamic equations (for velocities and pressure) as a single system, was
employed. It has been found in the segregated approach that the strategy to first solve the
momentum equations using a guessed pressure and an equation for a pressure correction
resulted in a large number of iterations to achieve reasonable convergence. By adopting the
coupled solver, it has been established that such a coupled treatment significantly outweighed
the segregated approach in terms of robustness, efficiency, generality and simplicity. To
accelerate convergence for each of the discretised algebraic equations, the Algebraic
Multigrid solver was adopted.
Validation Against Water Tunnel Observations
and Measurements
CFD simulation of the fluid flow through the various components of the fuel element
model that included the mockup rig and annular target cans was performed. Figure 14
illustrates the computed flow distribution inside and outside of the liner nose cone.
Based on the experimental flow rate of 1.6965 kgs
1
and a base diameter of 0.3 m of the
mockup fuel element model, there observed a very low flow velocity outside of the liner
nose cone in Figure 14(b). Nevertheless, the fluid after being squeezed through the small size
bottom and side holes of the liner nose cone caused these interacting merging flows to yield a
very highly complicated flow structure consisting of multiple vortices of recirculating flows
(see Figure 14(a)). It was also evident that due to the significant acceleration of the flow
found near the liner holes, the velocities increased dramatically to a magnitude of 5.0 ms
1
and resulted in large pressure drops. Near the bottom hole of the liner nose cone, the CFD
model predicted a normal velocity of approximately of 3 ms
1
. This predicted value has been
found to be in good agreement with experimental LDV measurement of 2.6 ms
1
, which
provided confidence to the reliability of the models in the CFD computer program. As the
fluid moved vertically upwards, the flow gradually became more uniform.
G.H. Yeoh and M.K.M. Ho 296
(a)
(b)
Figure 14. (a) Velocity vectors and (b) velocity contours around perforated liner.
(a) (b)
Figure 15. Flow separation near the pin as examined by (a) CFD and (b) PIV capture.
Another important consideration for the rig and target specification was the incorporation
of a pin situated at some distance below the placement of the rig as can be seen in Figure 15.
This design feature was implemented in the liner because of safety concerns in the event of a
possible accident scenario of the rig falling through to the bottom and impacting on the liner
nose cone. The selection of pin size was an important requirement for the rig and target
specification. From the flow predictions in Figure 15(a), it could be ascertained that the pin
size chosen contributed to only minor flow disturbances in the area between the pin and the
bottom surface of the rig nose cone. It was also observed that the majority of the bulk fluid
flow was unperturbed and diverged smoothly as it approached the rig. Flow visualization
Pipe Flow Analysis of Uranium Nuclear Heating with Conjugate Heat Transfer 297
performed during experiments (see Figure 15(b)) projected a similar flow pattern, which
further confirmed the reliable predictions of the CFD model.
(a) (b)
Figure 16. CFD study of flow distribution around annular can.
Figure 16 presents the flow distribution of the fluid traveling between the inner annular
can wall and rig outer surface designated for the purpose of illustration as region 1 and the
area between the outer annular can wall and the liner inner surface designated as region 2.
The fluid flowing within these spaces was found to be rather uniform. These favorable flow
structures indicated that axial cooling in the reactor rig along the length of the target could
remove the heat effectively for the design where uranium foils are embedded in the sealed
annular targets during the irradiation process. An interesting aspect of the model predictions
through the velocity contours in Figure 16(b) showed succinctly more fluid moving vertically
upwards in region 2 than in region 1. Based on the LDV flow measurements at the discrete
locations in regions 1 and 2 in Figure 12, the experiments confirmed the CFD predictions of
the different velocities in the two regions. Velocity values of 0.44 ms
1
and 0.502 ms
1
were
measured during experiments for regions 1 and 2 respectively. The predicted velocities as
depicted by the velocity contours in Figure 16(b) demonstrated the similar trend predicted
through the CFD model where region 1 yielded lower velocities compared to the higher
velocities in region 2.
Figure 17 illustrates the LDV measurements for the unmodified and modified designs of
the can flutes affecting the axial velocity distributions for two volumetric flow rates. Changes
introduced to the flute designs through the removal of any sharp edges significantly altered
the axial velocity distributions. The profiles were flatter thereby resulting in more uniform
and lesser wake flow structures. In the same figure, comparison between predicted and
measured vertical velocities is also presented. The predicted velocity profiles through the
CFD model showed similar encouraging distributions with the measured profiles in the inner
G.H. Yeoh and M.K.M. Ho 298
and outer channels. Good qualitative agreement was achieved between the predicted and
measured velocities.
X216 Rig Channel Axial Velocities
0 degree plane
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0 50 100 150 200 250 300 350 400 450 500
Axial Distance From Stem Tip (mm)
F
l
u
i
d
V
e
l
o
c
i
t
y
m
/
s
1.61 l/s Inner Channel
1.61 l/s Outer Channel
1.61 l/s IC;RMS
1.61 l/s OC;RMS
1.68 l/s IC
1.68 l/s OC
1.68 l/s IC;RMS
1.68 l/s OC;RMS
1.61 l/s mod IC
1.61 l/s mod OC
1.61 l/s mod IC;RMS
1.61 l/s mod OC;RMS
1.61 l/s IC CFD prediction
1.61 l/s OC CFD prediction
NOTES:
1) Run @ 1.61 l/s was
performed before fin
smoothing
2) Run @ 1.61 l/s mod was
performed after fin smoothing
Figure 17. Inner and outer channel LDV velocity measurements along X216 Rig vertical axis.
Current ‘MolybdenumPlate’ Design Used in the New OPAL
Reactor
The third pipecooling irradiation system to be examined is the ‘molybdenumplate’
design used in the new Open Pool Australian Lightwater (OPAL) research reactor. This
study investigated the maximum temperature of 20%
235
U enriched UAl compound sealed in
aluminium cladding plates. It will also demonstrate the incremental development of the model
and the affect this has on the maximum temperature, as more physics and boundary
conditions were introduced. The OPAL research reactor is an openpool design constructed
by INVAP, Argentina. The core rests thirteen meters under an open pool of lightwater which
provides both cooling and radiation protection. Surrounding the core is a Heavy Water
Reflector Vessel that is physically isolated from the bulk light water coolant. Its primary
purpose is to reflect neutrons back into the core to maintain the critical (nuclear) conditions
necessary for steady fission reaction rates during normal reactor operation. The secondary
purpose of the reflector vessel is to provide a relatively large volume with high flux for
irradiation and neutron beam facilities.
A compact reactor core measuring only 0.35 m x 0.35 m x 0.62 m (width, breadth,
height) is located 13m below the reactor pool surface. The Reflector Vessel surrounding the
Core contains heavy water and accommodates facilities such as irradiation rigs and the cold
Neutron Source (see Figure 18). A number of irradiation positions penetrating the Reflector
Vessel provide facilitates for the production of radioisotopes. These positions are sealed to
Pipe Flow Analysis of Uranium Nuclear Heating with Conjugate Heat Transfer 299
prevent the H
2
O and the D
2
O from mixing. Moreover, the irradiation positions contain built
in mechanisms for cooling which draws cold H
2
O from the main pool downwards into the
irradiation targets which is then conveyed away through pipes.
Figure 18. OPAL reactor facilities (top view).
MolybdenumPlate Life Cycle
A primary aspect of the OPAL reactor is the production of Technetium99 (
99m
Tc) for the
medical supply of local and international markets. To manufacture Molybdenum99, a
product made of the compound UAl contained in an aluminium matrix (called meat) is rolled
in between two parallel plates of aluminium alloy A96061. The result is a simple, self
contained plate of aluminium measuring 230 mm x 28 mm x 1.64 mm that contains a
sandwich of UAl ‘meat’ in the centre (see Figure 19). It is of note that the isotopic content of
fission uranium is high, at an enrichment of 20%
235
U but low enough to abide by
international guidelines for LEU (Low Enriched Uranium) fuels and irradiation targets.
Assembly of the irradiation rigs are performed in a hotcell situated at pool level. Fresh,
unirradiated molybdenum plates are loaded—four at a time—into a holder known as the
‘target’ (see Figures 20 and 21). A rig stem is then inserted through the target’s hollow centre
to form an assembled irradiation ‘rig’. The rig incorporates two to three irradiation targets
along with other flow control devices like nozzles and flow restrictors (see Figure 22). Once
assembled, the rigs are transported from the hotcells into the service pool (underwater) via a
hermetically sealed elevator. These rigs could then be remotely lowered to the reflector
vessel’s bulk irradiation facilities (see Figure 18).
G.H. Yeoh and M.K.M. Ho 300
Figure 19. One irradiation target with four molybdenum plates.
Figure 20. Isometric view of inadiation target model.
Figure 21. Cross sectional view of the irradiation target with four molybdenum plates inside.
Before the reactor is brought to critical, all primary cooling and secondary cooling
systems are switched on. The rigs cooling flow draws light water from the main pool down
Pipe Flow Analysis of Uranium Nuclear Heating with Conjugate Heat Transfer 301
into the irradiation rigs before being passed onto radiation decay tanks and heat exchangers
after. The fast neutrons radiating from the fuel plates are slowed by the H
2
O core coolant and
are completely thermalised by the D
2
O between the core and the facility. Having been
slowed, these neutrons could then be captured by the Uranium235 meat resulting in fission.
A maximum number of three targets could be loaded in one irradiation rig. This
constituted the bounding case to be modeled.
Figure 22. Molybdenum plate target and rig (shown on its side but loaded vertically).
Design Characteristics
Compared to the previous method of molybdenum manufacture, the philosophy of this
purposebuilt system was for high molybdenum yield and low radioactive waste output. The
previous design of molybdenum irradiation in the recently decommissioned HIFAR reactor
(an old British design of the DIDO class), utilized 2.2% enriched UO
2
reactor power fuel
pellets packed in bulky aluminium cans. These cans were centrally inserted into
concentrically arranged annular fuel plates and differed from the new irradiation rigs that lie
outside the core itself.
Consequently, the two main changes in molybdenum targets as compared to the previous
HIFAR (High Flux Australian Research) reactor are: (1) the geometry, which has changed
from a can to a plate, (2) the enrichment, increasing from 2.2% to 20% and (3) the location of
the irradiation rig outside the core. This new configuration allows the irradiation of more
235
U
but also uses a substantially lower mass of aluminium and
238
U, the result of which is an
increase in yield and a decrease in aluminium waste.
Despite these gains in efficiency, the increase in power generated by such a high
concentration of enriched uranium poses challenging conditions for heat removal. The thin
layer of aluminium gives little impedance to the flow of heat away from the UAl centre but
at the same time, any interruption to the forced convection flow of coolant could cause the
cladding to blister or at the very worst melt, as there is very little aluminium mass to which
the very large amounts of heat produced can conduct away. Thus, the prediction of flow on
and over the molybdenum plates is crucial to ensure no isolated heat spots could develop that
would result in the blistering of the clad or in the most extreme case, the melting of clad
during irradiation (burn out).
Thus, the aim of this numerical study was to show the thermohydraulic design of the
molybdenum holder was sufficient to remove the heat generated by fission with a sizable
margin between bounding operational conditions and extreme (i.e. almost impossible)
conditions.
After a maximum of 14 days of irradiation, the rig assembly is removed from the
reflector vessel and placed in the service pool storage racks to ‘cool’. Loading and unloading
G.H. Yeoh and M.K.M. Ho 302
of irradiation rigs can be performed whilst the reactor is at full power. The removal of the rig
from its irradiation position halts the neutronirradiation heating, leaving only the nuclear
decay heat to be removed by means of natural convection. The decayheat produced by weak
nuclear reactions is much lower than the fission heat—initially at approximately 8% of the
power produced during irradiation and decreases exponentially with time.
Methods of Calculations and Model SetUp
The maximum steady state temperatures of the molybdenum99 rigs as a result of the
coupled action of nuclear heating and lightwater forced convection cooling were determined.
These values could then be used to determine two important margins: (1) the difference
between the maximum attainable temperature in the plate and the aluminium cladding blister
temperature and (2) the difference between the operational maximum heat flux and the heat
flux at the onset of nuclear boiling.
For the simulation of heat transfer during irradiation, different parts of the irradiation
assembly were separately modeled, meshed and combined by what is termed as ‘patching’.
Afterwards, material properties were explicitly specified in each domain, along with the
relevant boundary conditions. The built model incorporated nothing more than the necessary
areas under examination to optimize the use of computational resources. Other fundamental
features such as the setting of appropriate turbulence model, the solution’s massresidual
convergence criteria and the time & space discretisation schemes for the iterative solution
process were then set (Tu et al., 2008). Some features of the numerical model, such as the
modeling of buoyancy were deemed to have a negligible effect to the overall flow
characteristics and were thus not modeled.
In order to validate the CFD model, a comparison with a set of independent results was
necessary. Available INVAP results using a more simplified onedimensional code were used
as comparative data against the CFD results. In this way, we were able to evaluate the
consistency of the numerical models. For those cases which have no corresponding INVAP
study, the only method for validation was by means of grid convergence testing and by the
comparison of results using different initial values or different convergence criteria. The
approach with this numerical study was to start with the most simplified assumptions in the
CFD model and to introduce increasing complexities to examine the relative importance of
each effect. A series of CFD simulations, utilizing a threedimensional CFD program
ANSYSCFX10.0, were thus conducted to attain a final validationcase which included all the
identified bounding conditions.
Power Distribution of MolybdenumPlate Targets
To undergo irradiation, assembled rigs loaded with two or three targets are inserted
vertically inside the reflector vessel. When the reactor is at power, nuclear reactions occur in
the molybdenumplates as thermal neutrons are absorbed by uranium with the effect of
producing heat. Since the horizontal neutron flux from the core radiates uniformly in the near
vicinity of the circular reflector pool, there is little flux variation in the horizontal plane of the
irradiation rig. However, the irradiation rig is very long, so the neutron flux distribution varies
Pipe Flow Analysis of Uranium Nuclear Heating with Conjugate Heat Transfer 303
appreciably in the vertical direction, resulting in a cosine power distribution along the vertical
axis of the molybdenumrig. Using the Monte Carlo NParticle (MCNP) transport code
for simulating neutronic reactions, it was found that for a three target configuration, the
middle position of three targets generated the most power. A separate and independent
neutronic calculation conducted by INVAP (MOLY01003OEIN004, 2003) found that the
maximum total power loading was 40.2 kW for one target and 79.6 kW for two targets. Since
these results indicate a trend for proportionately lower powers for each additional target, it
was deemed that each target could generate no more than 40.2kW by itself during normal
reactor operations. However, to increase the safety margin for this analysis, it was specified
that each of the three targets would generate a power of 60 kW.
The power distribution is defined in terms of the ratio between the maximum and the
average power along the vertical position of the plates in the irradiation rig. This can be
defined either in terms of the plate surface heat flux or in terms of the volumetric power
density, with a ratio known as the Power Peaking Factor (PPF). In ref. Moly0100, the PPF
has been shown to be equal to 1.1 for a single target and 1.3 for a twotarget loading. It was
thus assumed that a PPF of 1.3 could be used for three targets. This assumption has been
validated later by an MCNP model of the three targets loaded for irradiation with a core of
fresh fuel.
Thus, the bounding case was defined as an irradiation rig loaded with three targets, each
containing four plates. The power density was defined as follows:
( ) ( ) ( ) ( )
plate
0 9
Q
15000
cos 1.3 cos 3.134453 0.413
V 5188.8 10
q PPF B z z z = − = − (40)
where
q : Power density, [W.m
3
]
PPF : Power Peaking Factor, []
Q
plate
: Total power per plate, [W]
V : Meat volume, [m
3
]
B : Adjustable constant, [m
1
]
z : Coordinate on the z axis, [m]
z
0
: offset of the centre, [m]
Under nominal conditions, the average flow velocity through most of the rig’s cross
section along the plates is 3.8 m.s
1
. It was decided to set a more conservative limit with a
reduced average flow velocity of 3 m.s
1
for this study.
Boundary Conditions and Domain Restriction
Since coolant for the molybdenumrigs is drawn from the pool, the temperature is
essentially the pool water temperature which is regulated at a steady 37°C. To establish a
safety margin of 3°C, a value of 40°C was conservatively applied for the inlet coolant
boundary condition. It is difficult to ascertain the coolant velocity profile entering the rig
G.H. Yeoh and M.K.M. Ho 304
because the initial poolwater flow conditions could not be reliably predicted. But seeing as
the purpose of this investigation was to examine the heat transfer characteristics of the rig, the
exact flow pattern entering the rig was not of primary importance. Furthermore, the reducer
expansion arrangement of the nozzle preceding the three targets had the intended effect of
straightening the flow before it passed over the molybdenumplates. Therefore the flow can
be assumed to have a developed turbulent velocity profile before entering the targets’ coolant
channels.
By these considerations, the total control volume modeled in these simulations only
extended from the rig inlet nozzle to the outlet restrictor (see Figure 22) and a fully developed
turbulent flow profile was assumed as the inlet condition. To allow the average velocity to
develop into a turbulent velocity flow profile, an artificial pipe extension of fifteen hydraulic
diameters was modeled before the inlet. Also, lest the flow after the restriction nozzle affect
the preceding flow domain, an artificial pipe of 15 hydraulic diameters was also attached to
the end of the restrictor. Finally, the outer surface of the rig was conservatively assumed to be
adiabatic to attain the highest possible temperature in the coolant, now set to be solely
responsible for the removal of heat from the rig.
Figure 23. Crosssectional view of control volumes representing the irradiation rig.
Figure 24. The control volumes grouped in their respective materials shows each separate entity more
clearly.
Pipe Flow Analysis of Uranium Nuclear Heating with Conjugate Heat Transfer 305
Since it is obvious that the rig geometry is symmetrical in two planes (see Figures 23 and
24), only onequarter of the irradiation rig needed to be modeled with symmetrical boundaries
applied to the ‘cut planes’ to limit the number of mesh elements. In truth, a oneeighth model
would have sufficed but it was decided a complete coolant channel would be modeled so that
no assumptions need be made of what was subsequently demonstrated as a symmetrical
turbulent velocity profile through the pipe.
The inlet condition was defined at mass flow rate of 0.6259 kg.s
1
, corresponding to the 3
m/s average velocity in the major channel (see Figure 25). The inlet coolant temperature was
also conservatively set at a higher temperature of 40°C, providing a 3°C margin above the
nominal 37°C pool temperature. At the outlet, a 0 Pa gauge pressure was specified and the
cylindrical outer wall was defined as adiabatic. The standard kε model was chosen for this
highly forced convective problems with automatic adjustments for kinetic and dissipation
values as calculated by local velocities.
(a)
(b)
(c)
Figure 25. Quartersections of the: (a) nozzle (b) target (c) restrictor; blue indicates flow areas.
Grid Generation
For the sake of accuracy and computational efficiency, a body fitted orthogonal mesh was
selected to model the irradiation rig. Building quadrilateral surfaces for the body fitted mesh
required more time than automatically generated grids but this resulted in the use of less
nodes and reduced the amount of meshinduced solution diffusivity when compared to
tetrahedral mesh. The occurrence of this artificial diffusion is a result of interpolation errors
arising from highly skewed mesh contact angles of tetrahedralmeshes.
In areas where large gradients were expected, the volumes were meshed with relatively
high density to capture sharp temperature changes. Similarly, grid optimization was
performed in those areas where the temperature and velocity gradients were small by a
reduction in mesh density. Within the flow domain, the change of mesh sizes was gradual to
prevent numerical instability in the solution. Whilst mesh size transitions in the solid domain
were not required to be as stringent because the harmonicmean interpolation for thermal
conduction between solid nodes is not as geometrically sensitive as the solution for
convection and diffusion in fluid nodes. As mentioned, highly skewed elements were avoided
where possible to minimize numerical diffusion. In Figure 26, the circled area in (a) indicates
the mesh is too skewed and has been remeshed in (b) so that all cell face angles stood
between 38° and 142°. This optimization improved the solution maximum temperature result
by a difference of 6°C.
G.H. Yeoh and M.K.M. Ho 306
(a)
(b)
Figure 26. Circled area remeshed from (a) to (b) to reduce the presence of skewed mesh.
Mesh refinement was checked using three different meshes. Due to limits in
computational resources, it was not possible to double the mesh seeding in every direction at
the same time. Thus, two new meshes were created, one with a doubling of mesh in the
streamwise directions and one with a doubling in the crosswise. In each case, a simulation
was run and the maximum temperatures did not vary by more than 1°C. Thus it was shown
that grid convergence had been established, with a final node element count of 432,667.
Validation Case
This simulation compared the temperatures attained in INVAP’s correlationbased one
dimensional code with our CFD model under the same flow and power conditions. The
nominal condition study conducted by INVAP evaluated a two target loading scenario with a
total power output of 79.4 kW, a cooling flow of 3.6 m.s
1
and an oxide layer thickness of 7
μm as calculated by the Argonne National Lab (ANL) model for oxide layer growth.
Since INVAP’s model calculated for two targets whilst the CFD model was built with
three, one fourplate target was completely ‘switched off’, so that only two targets produced
power as simulated in the INVAP scenario. The inlet mass flow of the CFD model was
increased from the safety scenario of 0.6258 kg.s
1
to 0.7510 kg.s
1
to correspond with the
average flow velocity increase from 3 m.s
1
(safety study) to 3.6m.s
1
(validation study).
INVAP’s input parameters of solid physical properties and oxide layer thickness were also
adopted in this validation study. A summary of settings is displayed in Tables 2 and 3.
Details of the model can have effects on the criterion maximum temperature. These
details include:
1. modeling of the oxide layer
2. modeling of contact conditions between the Plate and the Target Holder
3. power density distribution in the UAl meat
To appreciate differences between model details, a simple CFD model was first built with
the assumptions:
Pipe Flow Analysis of Uranium Nuclear Heating with Conjugate Heat Transfer 307
1. without oxide layer.
2. perfect contact between the plate and the holder
3. a homogeneous power profile distribution across the U – Al meat
Table 2. Summary of the settings for the validation case
Oxide thickness 7 μm
Number of targets 2
Total Power 79.4 kW
Inlet flow rate 0.7510 kg.s
1
Inlet temperature 37°C
Outlet gauge pressure 0 Pa
Table 3. Summary of the material properties for the validation case
MATERIAL PROPERTIES Al U2Al and Al matrix Oxide Water
Molar Mass g.mol
1
26.98 270 18.02
Density kg.m
3
2702 4625.34 997
Specific Heat Capacity J.kg
1
.K
1
903 900 4181.7
Thermal Conductivity W.m
1
.K
1
165 148 2.25 0.6069
Dynamic Viscosity μ kg.m
1
.s
1
0.0008899
Effect of Oxide Layer Modelling
During irradiation, the layer of aluminium oxide that forms on the molybdenumplate is
relatively thin at 7μm. Physically, this thin layer could not be directly represented in the
computer model because the mesh was not fine enough to virtually interpret what was
physically there. The oxide layer could not be ignored either because though small, its
reduced conductivity had a significant effect on the molybdenumplate’s maximum
temperature.
Figure 27. Stepfunction thermal conductivity profile.
G.H. Yeoh and M.K.M. Ho 308
Figure 28. Graphical representation of distance vs. temperature for realistic and modeled thermal
conductivity for aluminum and aluminum oxide.
A first attempt was made to simply prescribe a thermal conductivity profile onto the
cladding elements as shown in Figure 27. The thermal conductivity of the aluminium varies
with respects to cladding depth, with the circled regions of aluminiumoxide possessing a
much lower thermal conductivity. However, the result of this simulation was proven a failure
as the maximum temperature proved no different from a separate identical simulation run
with no oxidelayer.
To circumvent this problem without resorting to the use of more computationally
expensive mesh, the effect of the oxide layer was mathematically modeled as an integral part
of the cladding. This was done by calculating an equivalent conductivity via the formula
below which was derived from flux conservation principles (Patankar, 1980):
1
Al ox
1 f f
k
k k
−
⎛ ⎞ −
= +
⎜ ⎟
⎝ ⎠
(41)
where
ox
Al ox
= f
δ
δ δ +
. The oxide layer thickness was assumed to be 7 μm corresponding with
INVAP’s assumption. The equivalent conductivity of the cladding result was thus:
1
1 1
1 1
1 1
7
350
7 1 0.02 0.02
0.02 67.44 . .
165 . . 350 165 2.25
2.25 . .
ox
Al ox
Al
ox
m
m
f k W m K
k W m K
k W m K
δ μ
δ δ μ
−
− −
− −
− −
= ⎧
⎪
+ =
− ⎪ ⎛ ⎞
⇒ = = ⇒ = + =
⎨ ⎜ ⎟
=
⎝ ⎠
⎪
⎪
=
⎩
This formula modified the aluminium conductivity to attain the correct temperature at the
aluminium boundary but at the expense of correct temperatures inside the aluminium itself, as
graphically explained in Figure 28.
Maximum temperatures from the stepconductivity profile and the average conductivity
profile technique are displayed in detail in Table 4. The largest difference in temperature
Pipe Flow Analysis of Uranium Nuclear Heating with Conjugate Heat Transfer 309
proved to be between the nooxide and averageconductivity oxide simulations, with a
difference of 3.9°C in the UAl meat.
However, the stepwiseconductivity profile technique resulted in only a 0.5°C gain in U
Al meat maximum temperature. It is obvious from these results that the stepwise
conductivity profile technique was unsuitable. The average conductivity profile technique
was adopted as the method for simulating the effect of aluminium oxide on cladding
conductivity.
Table 4. Comparison of models
Max. Temperatures (°C)
Reference Oxide model
Oxide layer absent step average
Contact type perfect perfect perfect
Power distribution uniform uniform uniform
Surface (water side) 59.2 59.2 59.0
Surface (cladding side) 99.1 99.3 100.5
Cladding – Meat interface 100.9 101.3 104.7
Meat – centre line 102.0 102.5 105.9
Effect of Contact Surface
Since the thermal conductivity of water was much less than aluminium, a break in the
solid conduction path between the molybdenumplate and the target holder would be
detrimental for the conduction of heat away from the plate. Also, as there was a large
tolerance between the plate and the recess in which the plate is held, a water gap existed
between one side of the plate and the holder. This study examined the maximum temperature
difference caused by the presence of this water gap (see Figure 29).
During the creation of this geometry, a space between the plate and its holder was made
and patched separately to allow different contact conditions to be simulated. Since the inner
channel was slightly larger than the outer channel, slight difference in local velocities on
either side of the plate would produce a minute pressure difference that forces the plate to one
side. To model the reduced thermalconduction effects of the water gap, solids with the
thermal conductivity of water were patched in between the plate and the holder. Effectively,
the water gap acted as a low conductivity solid with no coolant advection. This was a
reasonable and conservative assumption because the very restricted flow path of the gap
would result in very small flow rates that would provide negligible amounts of forced
convection cooling. The alternative was to mesh the gap at a higher resolution to adequately
solve for the fluid advection which was in practical terms, unnecessary.
G.H. Yeoh and M.K.M. Ho 310
Figure 29. Detailed view of contact realistic conditions between the plate and its holder.
Table 5. Effect of the contact surface
Temperatures (°C)
Thermal conduction profile type: average average
Contact type perfect partial
Power distribution uniform uniform
Surface (water side) 59.0 62.1
Surface (cladding side) 100.5 103.6
Cladding  Meat interface 104.7 107.6
Meat  centre line 105.9 108.8
The difference in contact conditions only proved to be marginal as can be seen in Table
5. The reduction from ‘full contact’ to ‘half contact’ between the plate and the holder only
increased the overall maximum temperature by 3°C. This proved to be positive from an
engineer’s standpoint because it was undesirable to have heat transfer characteristics that are
too sensitive to unpredictable contact conditions. The halfcontact condition between plate
and holder was thus adopted for future simulations for its more conservative assumption.
Effect of Power Distribution
In this study, the effect of a uniform and cosine power distribution on the maximum
temperature was examined. A simpler study by INVAP using a onedimensional heat transfer
code provided a simple comparative assessment. According to MCNP calculations conducted
by ANSTO (Geoff, 2006), a power density profile distributing a total power of 79.4 kW over
eight plates with a power peaking factor (PPF) of 1.3 was calculated as:
( ) ( ) ( ) ( )
plate
0 9
79400
Q
8
cos 1.3 cos 4.97424 0.288
V 5188.8 10
q PPF B z z z = − = − (41)
Pipe Flow Analysis of Uranium Nuclear Heating with Conjugate Heat Transfer 311
where
q : Power density, [W.m
3
]
PPF : Power Peaking Factor, []
Q
plate
: Total power per plate, [W]
V : Meat volume, [m
3
]
B : Adjustable constant, [m
1
]
z : Coordinate on the z axis, [m]
z
0
: Offset of the centre, [m]
Power Density vs. Vertical Position
2 Targets
8.00E+08
1.00E+09
1.20E+09
1.40E+09
1.60E+09
1.80E+09
2.00E+09
2.20E+09
2.40E+09
2.60E+09
0 0.2 0.4 0.6
z [m]
q
[
W
.
m
^

3
]
Figure 30. Power distribution along the plates (two positions).
Table 6. Power distribution effect on temperatures
Temperatures (°C)
Validation case INVAP study
Oxide layer average average —
Contact type partial partial —
Power distribution uniform cosine —
Surface (water side) 62.1 61.9
Surface (cladding side) 103.6 114.9
105
Oxide  Cladding interface — — 109
Cladding  Meat interface 107.6 120.2 111
Meat  centre line 108.8 121.6 113
This definition allowed the use of the same maximum power density (i.e., 2.5 10
9
W.m
3
)
as the one used by INVAP to calculate the maximum meat temperature, and also the same
G.H. Yeoh and M.K.M. Ho 312
total power in two plates. It also respected the PPF of 1.3 for the two position case. Applying
the cosine power distribution to the model gave the simulation more realism than if a constant
power was applied throughout the plate. Two simulations were run to observe the effect of the
difference between a constant power and a cosine distribution power.
Using a cosineshape distribution for the power increased the maximum meat
temperature by 12.8°C as shown in Table 5. Since a cosine power distribution (see Figure 30)
is more conservatively representative, this model feature was retained for the “bounding case”
scenarios.
Conclusive Remarks for the Validation Case
Beginning from a simple model and making successive improvements with increasing
detailed features, a final validation case was attained which has been proven to be consistent
and conservative against INVAP’s onedimensional heattransfer study. This course of action
has also allowed an appreciation of the sensitivity of temperature changes with respect to
varying engineering assumptions.
The temperatures on the interface between the cladding and the oxide layer could not be
attained in the CFD model, and therefore has not been compared. The CFX commercial
software appears to quote the temperature value held at the centre of the control volume on
either side of the platewater interface, so that these results could not be directly compared
against INVAP’s study. However, the cladding side surface temperature yields valid and
comparable results to INVAP’s plate surface temperature results.
Table 7. Summary of the simulations for the validation case.
Temperatures (°C)
Validation Case INVAP study
Oxide layer absent step average average average —
Contact type perfect perfect perfect partial partial —
Power distribution uniform uniform uniform uniform cosine —
Surface (water side) 59.2 59.2 59.0 62.1 61.9
Surface (cladding side) 99.1 99.3 100.5 103.6 114.9
105
Oxide  Cladding interface — — — — — 109
Cladding  Meat interface 100.9 101.3 104.7 107.6 120.2 111
Meat  centre line 102 102.5 105.9 108.8 121.6 113
Various hypotheses were tested on the validation case. Several simulations were run
which revealed:
• The application of an average thermal conductivity over the whole cladding was
more realistic than the definition of a step function to lower the conductivity in the
oxide layer.
Pipe Flow Analysis of Uranium Nuclear Heating with Conjugate Heat Transfer 313
• The assumption of oneside contact between the plate and its holder was more
realistic than assuming full aluminium contact on both sides of the plate.
• The application of a variable power density was more realistic than a simple
definition of the average power density.
In conclusion, these results were in good agreement with INVAP’s calculations and as
such, this model could now be used to examine even more conservative flow conditions for
the purpose of safety analysis. Summary of results are presented in Table 7.
Bounding Case
Having demonstrated the consistency of the CFD model against INVAP’s verification
data, the CFD model could now be used to investigate the flow conditions of the bounding
case scenario. A summary of the material properties are detailed in Table 8.
Similar to the validation simulation, a cosine power profile (see Figure 31) for three
targets was set for this bounding simulation and a ‘high resolution’ advection scheme with
automatic timescale was adopted for the solution process inorder to accelerate solution
convergence. A summary of temperature results for oxide thicknesses of 20 μm and 10 μm
are displayed in Table 9 below:
Table 8. Summary of physical properties
MATERIAL PROPERTIES Aluminium U2Al and Al matrix oxide Water
Molar Mass g.mol
1
26.98 270 18.02
Density kg.m
3
2700 4625.34 996.2
Specific Heat Capacity J.kg
1
.K
1
903 900 4178.8
Thermal Conductivity W.m
1
.K
1
165 148 2.25 0.609
Dynamic Viscosity μ kg.m
1
.s
1
0.0008327
Power Density vs. Vertical Position
3 Targets
1.50E+09
2.00E+09
2.50E+09
3.00E+09
3.50E+09
4.00E+09
0 0.2 0.4 0.6 0.8
z [m]
q
[
W
.
m
^

3
]
Figure 31. Power distribution over 3 plates.
G.H. Yeoh and M.K.M. Ho 314
Table 9. Calculated temperatures for the bounding case
Temperatures (°C)
Oxide layer 20 μm 10 μm
Surface (water side) 92.2 92.5
Surface (cladding side) 183.1 181.1
Cladding  Meat interface 199.5 190.9
Meat  centre line 201.6 193
Thus, assuming an average flow velocity of 3 m.s
1
at 40°C, an oxide thickness of 20 μm,
and a PPF of 1.3 for a total power of 45 kW over three plates, the main results were obtained:
• Maximum water temperature: 93°C
• Maximum meat temperature: 202°C
These results showed a sufficient margin from plate melting and water boiling.
Conclusion
The cooling systems for three separate molybdenum irradiation facilities have been
solved for maximum mean temperatures using the computational fluid dynamics
methodology.
In order to accomplish this, the geometry of all three systems were first modeled and
geometrically discretised in a variety of structured and unstructured mesh. Secondly, accurate
material properties were attributed to corresponding areas of the numeric model. Uranium
volumes were then attributed Power densities as calculated by MCNP to simulate the power
produced within irradiated targets. The high velocities and thus high Reynolds numbers of all
three case studies placed the flow regime within a turbulent setting. As the study was
concerned with timeaveraged results, the RANS approach for turbulence modeling was most
suited and the simple standard kε turbulence model was selected over other models for its
applicability to fully turbulent flows. Other types of flows, such as large swirling flows or
flows with large amounts of laminarturbulent blending, would have required more
sophisticated RANS modeling like the RNG kε model, reliazable kε model and SST Menter’s
model. However, this level of sophistication was unnecessary for the solution of these pipe
flow systems and was thus not employed.
Computational results of all three case studies have been demonstrated to agree well with
independent experimental and numerical studies. The success of these studies further
confirms the robustness and versatility of CFD methods in the field of nuclear engineering
and will remain a continual feature in the field of thermohydraulics.
Pipe Flow Analysis of Uranium Nuclear Heating with Conjugate Heat Transfer 315
Acknowledgements
The authors would like to thank Dr. George Braoudakis (Head of Nuclear Analysis
Section) for providing the MCNP power input parameters in these simulations and for proof
reading this document; past internship students Mr. Tony Chung (UNSW, Sydney) and Mr.
Guillaume Bois (INSA, Lyon) for their assistance in attaining the computational simulations;
Mr. David Wassink (Water Tunnel Manager) for his experimental validation data; and,
finally, to ANSTO for allowing this work to be published.
References
[1] Durance, G. (2006), Private communication, ANSTO.
[2] Gumbert, C., Lohner, R., Parikh, P. & Pirzadeh, S. (1989), A package for unstructured
grid generation and finite element flow solvers, AIAA Paper 892175.
[3] Jayatilleke, C. L. V. (1969), The influence of Prandtl number and surface roughness on
the resistance of the laminar sublayer to momentum and heat transfer, Prog. Heat Mass
Transfer, 1, 193 321.
[4] Launder, B. E. & Spalding, D. B. (1974), The numerical computation of turbulent flows,
Comp. Meth. Appl. Mech. Eng., 3, 269289.
[5] Lo, S. H. (1985), A new mesh generation scheme for arbitrary planar domains, Int. J.
Numer. Methods Eng., 21, 14031426.
[6] Marcum, D. L. & Weatherill, N. P. (1995), Unstructured grid generation using iterative
point insertion and local reconnection, AIAA Paper 941926.
[7] Mavriplis, D. J. (1997), Unstructured grid techniques, Ann. Rev. Fluid Mech., 29, 473
514.
[8] Menter, F. R. (1993), Zonal two equation kω turbulence models for aerodynamics
flows, AIAA paper 932906.
[9] Menter, F. R. (1996), A comparison of some recent eddyviscosity turbulence models, J.
Fluids Eng., 118, 514519.
[10] MOLY01003OEIN004 (2003), Heating calculation of Molybdenum targets, INVAP
Report.
[11] Patankar, S. V. (1980), Numerical Heat Transfer and Fluid Flow, Hemisphere
Publishing Corporation, New York.
[12] Shepard, M. S. & Georges, M. K. (1991), Threedimensional mesh generation by finite
octree technique, Int. J. Numer. Methods Eng., 32, 709749.
[13] Shewchuk, J. S. (2002), Delaunay refinement algorithms for triangular mesh generation,
Computational Geometry: Theory and Applications, 22, 21–74.
[14] Shih, T.H., Liou, W. W., Shabbir, A., Yang, Z. & Zhu, J. (1995). A new kε eddy
viscosity model for high Reynolds number turbulent flows, Comp. Fluids, 24, 227238.
[15] Smith, R. E. (1982), Algebraic grid generation, Numerical Grid Generation, Thompson,
J. F. (Ed.), NorthHolland, Amsterdam, 137.
[16] Thompson, J. F. (1982), General curvilinear coordinate systems, Numerical Grid
Generation, Thompson, J. F. (Ed.), NorthHolland, Amsterdam, 130.
[17] Touloukian, Y.S., Powell, R.W., Ho, C.Y., Klemens, P.G. (1970). Thermophysical
Properties of Matter, 2.
G.H. Yeoh and M.K.M. Ho 316
[18] Tu, J. Y., Yeoh G. H. & Liu C. Q. (2008), Computational Fluid Dynamics – A Practical
approach. ButterworthHeinemann, Oxford.
[19] Wilcox, D. C. (1998), Turbulence Modelling for CFD, DCW Industries, Inc.
[20] Yakhot, V., Prszag, S. A., Tangham, S., Gatski, T. B. & Speciale, C. G. (1992),
Development of turbulence models for shear flows by a double expansion technique,
Phys Fluids A: Fluid Dynamics, 4, 15101520.
[21] Yeoh, G. H. and Storr, G. J. (2000), A threedimensional study of heat and mass transfer
within the irradiation space of the HIFAR fuel element, Advance Computational
Methods in Heat Transfer VI, Sunden, B. and Brebbia, C. A. (Eds.), WIT Press,
Southampton, 343351.
[22] Yerry, M. A. & Shepard, M. S. (1984), Automatic threedimensional mesh generation
by the modifiedoctree technique, Int. J. Numer. Methods Eng., 20, 19651990.
In: Fluid Mechanics and Pipe Flow ISBN: 9781607410379
Editors: D. Matos and C. Valerio, pp. 317342 © 2009 Nova Science Publishers, Inc.
Chapter 9
FIRST AND SECOND LAW THERMODYNAMICS
ANALYSIS OF PIPE FLOW
Ahmet Z. Sahin
*
Mechanical Engineering Department
King Fahd University of Petroleum and Minerals
Dhahran 31261, Saudi Arabia
Introduction
In a fully developed laminar flow through a pipe, the velocity profile at any cross section
is parabolic when there is no heat transfer. But when a considerable heat transfer occurs and
the thermophysical properties of the fluid vary with temperature, the velocity profile is
distorted. If the thermophysical properties of the fluids in a heat exchanger vary substantially
with temperature, the velocity and the temperature profiles become interrelated and, thus, the
heat transfer is affected. Viscosity of a fluid is one of the properties which are most sensitive
to temperature. In the majority of cases, viscosity becomes the only property which may have
considerable effect on the heat transfer and temperature variation and dependence of other
thermophysical properties to temperature is often negligible. The viscosity of the liquids
decreases with increasing temperature, while the reverse trend is observed in gases (Kreith
and Bohn, 1993). Heat transfer and pressure drop characteristics are affected significantly
with variations in the fluid viscosity. When the temperature is increased from 20 to 80
o
C, the
viscosity of engine oil decreases 24 times, the viscosity of water decreases 2.7 times and the
viscosity of air increases 1.4 times. Therefore, selection of the type of fluid and the range of
operating temperatures are very important in the design and performance calculations of a
heat exchanger.
In the process of designing a heat exchanger, there are two main considerations. These
are the heat transfer rates between the fluids and the pumping power requirement to overcome
the fluid friction and move the fluids through the heat exchanger. Although the effect of
viscous dissipation is negligible for lowvelocity gas flows, it is important for highvelocity
*
Email address: azsahin@gmail.com
Ahmet Z. Sahin 318
gas flows and liquids even at very moderate velocities (Shah, 1981; Kays and Crawford,
1993). Heat transfer rates and pumping power requirements can become comparable
especially for gastogas heat exchangers in which considerably large surface areas are
required when compared with liquid to liquid heat exchangers such as condensers and
evaporators. In addition, the mechanical energy spent as pumping power to overcome the
fluid friction is worth 4 to 10 times as high as its equivalent heat (Kays and London, 1984).
Therefore, viscous friction which is the primary responsible cause for the pressure drop and
pumping power requirements is an important consideration in heat exchanger design.
On the other hand, efficient utilization of energy is a primary objective in designing a
thermodynamic system. This can be achieved by the minimization of entropy generation in
processes of the thermodynamic system. The irreversibilities associated with fluid flow
through a pipe are usually related to heat transfer and viscous friction. Various mechanisms
and design features contribute to the irreversibility terms (Bejan, 1988). There may exist an
optimal thermodynamic design which minimizes the amount of entropy generation. For a
given system, a set of thermodynamic parameters which optimizes the operating conditions
may be obtained.
The irreversibility associated with viscous friction is directly proportional to the viscosity
of the fluid in laminar flow. Therefore, it is necessary to investigate the effect of a change of
viscosity during a heating process for an accurate determination of entropy generation and of
the required pumping power.
Heat transfer and fluid pumping power in a piping system are strongly dependent upon
the type of fluid flowing through the system. It is important to know the fluid properties and
their dependence to temperature for a heat exchanger analysis. As a result of heat transfer, the
temperature changes in the direction of flow and the fluid properties are affected. The
temperature variation across the individual flow passages influences the velocity and
temperature profiles, and thereby influences the friction factor and the convective heat
transfer coefficient. If the thermophysical properties of the fluids in a piping system vary
substantially with temperature, the velocity and the temperature profiles become interrelated
and, thus, the heat transfer is affected. Viscosity of a fluid is one of the properties which are
most sensitive to temperature. Therefore, selection of the type of fluid and the range of
operating temperatures are very important in the design and performance calculations of a
piping system.
On the other hand, the exergy losses associated with fluid flow through a pipe are usually
related to heat transfer and viscous friction. Dependence of the thermophysical properties on
the temperature affects not only the viscous friction and pressure drop, but also the heat
transfer. This implies that the exergy losses associated with the heat transfer and the viscous
friction are also affected. The contributions of various mechanisms and design features on the
different irreversibility terms often compete with one another. Therefore, an optimal
thermodynamic design which minimizes the amount of entropy generation may exist. In other
words, a set of thermodynamic parameters which optimize operating conditions could be
obtained for a given thermodynamic system.
In this chapter, the entropy generation for during fluid flow in a pipe is investigated. The
temperature dependence of the viscosity is taken into consideration in the analysis. Laminar
and turbulent flow cases are treated separately. Two types of thermal boundary conditions are
considered; uniform heat flux and constant wall temperature. In addition, various cross
sectional pipe geometries were compared from the point of view of entropy generation and
First and Second Law Thermodynamics Analysis of Pipe Flow 319
pumping power requirement in order to determine the possible optimum pipe geometry which
minimizes the exergy losses.
Pipe Subjected to Uniform Wall Heat Flux
We consider the smooth pipe with constant cross section shown in Figure 1. A constant
heat flux q′′ is imposed on its surface. An viscous fluid with mass flow rate m and inlet
temperature
0
T enters the pipe of length L . Density ρ , thermal conductivity k , and
specific heat
p
C of the fluid are assumed to be constant. Heat transfer to the bulk of the fluid
occurs at the inner surface with an average heattransfer coefficient h , which is a function of
temperature dependent viscosity.
The effect of viscosity on the average heat transfer coefficient is given by Kays and
Crawford (1993)
. .
. .
n
b
c p
c p w
h Nu
h Nu
μ
μ
⎛ ⎞
= =
⎜ ⎟
⎝ ⎠
(1)
where the exponent n is equal to 0.14 for laminar flow. In the case of turbulent flow the
exponent n is equal to 0.11 for heating and 0.25 for cooling.
For fully developed laminar flow, constant property heat transfer coefficient
. . c p
h is
given by Incropera and DeWitt (1996).
D
k
Nu
D
k
h p c
p c
11
48
. .
. .
= = (2)
Figure 1. Sketch of constant cross sectional area pipe subjected to uniform wall heat flux.
Ahmet Z. Sahin 320
And for fully developed turbulent flow, constant property heat transfer coefficient
. . c p
h is
given by
( ) ( )
( )
. .
. .
2/ 3
/ 8 Re 1000 Pr
1 12.7 Pr 1 / 8
c p
c p
f
k k
h Nu
D D
f
−
= =
+ −
(3)
where the Reynolds number is
Re
b
UD ρ
μ
= . (4)
The average Darcy friction factor, f , for a smooth pipe is also considered to be a
function of temperature dependent viscosity and is given by Kays and Crawford (1993)
m
w
b
p c
f
f
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
=
μ
μ
. .
for liquids (5a)
m
w
b
p c
T
T
f
f
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
=
. .
for gases (5b)
where the exponent 14 . 0 − = m for laminar flow and 25 . 0 − = m for turbulent flow case.
The friction factor for constant properties for laminar flow is given by Incropera and
DeWitt (1996) as
Re
64
. .
=
p c
f (6)
and for turbulent flow as
[ ]
2
. .
0.79ln(Re) 1.64
c p
f
−
= − (7)
To account for the variation of the bulk temperature along the pipe length,
b
μ and
therefore Re and Pr, in Eqs. (3) – (7), are related to the bulk fluid temperature halfway
between the inlet and outlet of the pipe, as suggested by Kreith and Bohn (1993). Since the
temperature variation along the pipe is initially unknown and depends on h , a trial and error
procedure is followed to determine both h and f .
First and Second Law Thermodynamics Analysis of Pipe Flow 321
Total Heat Transfer Rate
The rate of heat transfer to the fluid inside the control volume shown in Figure 1 is
p
Q mC dT q Ddx δ π ′′ = =
(8)
where
2
/ 4 m U D ρ π = . In writing Eq. (8), the pipe is assumed to have a circular cross
section. However, the analysis is not affected by assuming crosssectional areas other than
circular. Integrating Eq. (8), the bulktemperature variation of the fluid along the pipe is
obtained as
0
4
p
q
T T x
UDC ρ
′′
− = . (9)
The temperature variation along the pipe is linear when the viscous dissipation and axial
conduction effects are neglected. The temperature gradient for the fluid in this case depends
mainly on the magnitude of the heat flux.
For a constant heat flux q′′ and average heat transfer coefficient h evaluated at the bulk
temperature halfway between inlet and outlet of pipe, the temperature difference between the
wall surface and bulk of the fluid is given as
w
q
T T
h
′′
− = . (10)
Eq. (9) may be rewritten as
4
St
x
D
θ = , (11)
where θ is the dimensionless temperature defined as
0
/
T T
q h
θ
−
=
′′
(12)
and the Stanton number is
p
h
St
UC ρ
= . (13)
Ahmet Z. Sahin 322
Total Entropy Generation
The entropy generation within the control volume of Figure 1 can be written as (Bejan,
1996)
gen
w
Q
dS mds
T
δ
= −
, (14)
where the entropy change is
p
dT dP
ds C
T T ρ
= − . (15)
Substituting Eq. (8) into Eq. (14),
1
w
gen p
w p
T T
dS mC dT dP
TT C T ρ
⎡ ⎤
−
= −
⎢ ⎥
⎣ ⎦
. (16)
The pressure drop is (Kreith and Bohn, 1993)
2
2
f U
dP dx
D
ρ
= − , (17)
where f is the Darcy friction factor.
Integrating Eq. (16) along the pipe length L , using Eqs. (9) and (17), the total entropy
generation becomes (Sahin, 1999 and Sahin, 2002)
( ) ( )
( )
3
1 4 1 1
ln ln 1 4
1 4 8
gen p
St f U
S mC St
St q
τλ τ ρ
τλ
τ τλ
⎧ ⎫ + + ⎡ ⎤
⎪ ⎪
= + +
⎨ ⎬
⎢ ⎥
′′ + +
⎪ ⎪ ⎣ ⎦ ⎩ ⎭
, (18)
where the dimensionless wall heat flux is
0
/ q h
T
τ
′′
=
(19)
and the dimensionless length of the pipe is
L
D
λ = . (20)
First and Second Law Thermodynamics Analysis of Pipe Flow 323
Eq. (18) can be written in a dimensionless form as
0
/ T Q
S
gen
= ψ (21)
where the total heat transfer is
( )
0
4
p
Q mC T Stτλ =
(22)
Thus, the entropy generation per unit heat transfer rate becomes
⎭
⎬
⎫
⎩
⎨
⎧
+ +
⎥
⎦
⎤
⎢
⎣
⎡
+ +
+ +
= ) 4 1 ln(
8
1
4 1
) 1 )( 4 1 (
ln
4
1
τλ
τλ τ
τ τλ
τλ
ψ St
St
Ec
f
St
St
St
(23)
where the Eckert number is defined as
2 2
( ) ( / )
p w p
U U
Ec
C T T C q h
= =
′′ −
. (24)
Two dimensionless groups arise naturally in Eq. (23), namely,
1
Stλ Π = (25)
and
2
Ec
f
St
Π = (26)
Thus, Eq. (23) becomes in compact form
⎭
⎬
⎫
⎩
⎨
⎧
Π + Π +
⎥
⎦
⎤
⎢
⎣
⎡
Π + +
+ Π +
Π
= ) 4 1 ln(
8
1
4 1
) 1 )( 4 1 (
ln
4
1
1 2
1
1
1
τ
τ τ
τ τ
τ
ψ (27)
or
( )
⎥
⎦
⎤
⎢
⎣
⎡
Π +
Π + +
+
Π
=
Π +
2
8
1
1
1
1 1
4 1
4 1
1
ln
4
1
τ
τ τ
τ
τ
ψ . (28)
The first and second terms in Eq. (27) are related to entropy generation due to heat
transfer and due to viscous friction respectively. Eq. (27) contains three nondimensional
parameters, namely, τ ,
1
Π and
2
Π . Among these parameters, τ represents the heat flux
imposed on the wall of the pipe q′′ and
1
Π represents the pipe length L . Once the type of
the fluid and the mass flow rate are fixed, the parameter
2
Π can be calculated on the basis of
Ahmet Z. Sahin 324
temperature analysis. Thus, τ and
1
Π are the two design parameters that can be varied for
determining the effects of pipe length and/or wall heat flux on the entropy generation.
For small values of the wall heat flux ( 1 τ << ), Eq. (27) is reduced to
) 4 1 ln(
32
1
1
1
2
Π +
Π
Π
= τ
τ
ψ . (29)
For low viscosity,
2
/ 8 1 Π << and Eq. (28) shows that entropy generation becomes
( )
⎥
⎦
⎤
⎢
⎣
⎡
Π + +
+ Π +
Π
=
1
1
1
4 1
1 ) 4 1 (
ln
4
1
τ τ
τ τ
τ
ψ (30)
Amount of heat transfer is the primary concern in a heat exchanger design. However, a
considerable amount of exergy is destroyed during the heat exchange process. For efficient
utilization of energy the entropy generation needs to be minimized. Thus, exergy destruction
per unit amount of heat transfer is considered to be a suitable quantity in dealing with the
second law analysis of a pipe flow. In the present analysis this ratio is represented by the
dimensionless entropy generation, ψ , defined as the entropy generation per unit heat transfer
rate for specified pipe inlet temperature.
Since
1
Π represents the length of the pipe, the dimensionless entropy generation per unit
heat transfer rate to the pipe ψ decreases with the pipe length. The dimensionless entropy
generation ψ approaches a common value for short pipes with a limit
2
0
8
1
1
lim
1
Π +
+
=
→ Π
τ
τ
ψ . (31)
For the case of constant viscosity, ψ decreases with a slope of
⎥
⎦
⎤
⎢
⎣
⎡
Π +
+
− − =
Π ∂
∂
→ Π
2 2
1
0
8
1
) 1 (
1
1 2 lim
1 τ
τ
ψ
. (32)
ψ decreases and approaches zero asymptotically (and therefore the total heat transfer) as
pipe length increases.
Since τ represents the amount of heat flux, the dimensionless entropy generation ψ ,
defined on the basis of total heat transfer rate to the pipe, goes to infinity when 0 τ = . ψ
decreases sharply as τ increases and then starts increasing.
A second dimensionless entropy generation may be defined, on the basis of unit heat
capacity rate of fluid in the pipe, as
First and Second Law Thermodynamics Analysis of Pipe Flow 325
) / ( T Q
S
C m
S
gen
p
gen
Δ
= =
φ , (33)
where T Δ is the increase
0
( ) T L T − of the bulk temperature of the fluid in the pipe. Since
/
p
Q T mC Δ =
is constant for a fixed mass flow rate, the total entropy generation (φ ),
which increases with an increase in pipe length, can be given as
ψ τ φ
1
4 Π = . (34)
For the case of constant viscosity, φ starts to increase with a rate of
⎥
⎦
⎤
⎢
⎣
⎡
Π +
+
=
Π ∂
∂
→ Π
2
1
0
8
1
1
4 lim
1 τ
τ
τ
φ
(35)
and approaches to infinity as pipe length increases.
It should be noted that
2
Π is inversely proportional to τ and therefore the propipe
2
τΠ
is constant. Thus the value of φ is finite for 0 τ =
) (
2
1
lim
2 1
0
Π Π =
→
τ φ
τ
(36)
φ decreases with a slope of
) ( lim
2
2
1
0
Π Π − =
∂
∂
→
τ
τ
φ
τ
. (37)
Pumping Power to Heat Transfer Rate Ratio
The ratio of pumping power to heat transfer rate is
2
4
r
D PU
P
Q
π Δ
=
. (38)
Using Eqs. (8) and (17), the pumping power to heat transfer ratio
r
P is
2
1
8
r
P = Π . (39)
Ahmet Z. Sahin 326
Recalling that the product
2
τΠ is invariable with varying τ , it can be concluded from
Eq. (39) that the pumping power ratio is inversely proportional to τ .
Effect of Pipe Cross Section
The dimensionless entropy generation given in Eq. (28) can be expressed in terms of Re
number as (Sahin, 1998)
⎥
⎥
⎥
⎦
⎤
⎢
⎢
⎢
⎣
⎡
⎟
⎠
⎞
⎜
⎝
⎛
+
+ +
+
=
+
2
2
Re 1
1
1
1
Re
1
Re
1
1
ln
Re
C
C
C
C
τ
τ
τ
τ
τ
ψ (40)
where
1 1
Re 4 Π = C and
2
2
2
Re 8
Π
= C .
For any given pipe geometry and stream of constant thermophysical property fluid,
dimensionless numbers C
1
and C
2
are constants. Therefore, the dimensionless total entropy
generation as given in Eq. (40) is a function of Re and τ only.
Values of Nu and Re) ( f for fully developed laminar flow are given in Shah and
London (1978) for a variety of pipe geometries. The hydraulic diameter D
H
for any given pipe
geometry can be calculated using the relation:
p
A
D
c
H
4 = . (41)
Table 1 gives the hydraulic diameters for some common pipe geometries.
Similarly, the modified dimensionless entropy generation becomes
ψ
τ
φ
Re
1
C
= (42)
and the pumping power to heat transfer ratio is obtained to be
2
2
Re C P
r
= (43)
As the temperature difference between the fluid inlet and the surface of the pipe, τ , is
increased, the entropy generation increases, however, the pumping power requirement per
unit heat transfer decreases. In the limit when τ approaches zero, the contribution of heat
transfer to the entropy generation disappears and
First and Second Law Thermodynamics Analysis of Pipe Flow 327
Re lim
2 1
0
τ ψ
τ
C C =
→
(44)
which indicates a linear increase in entropy generation that is solely due to the viscous
friction as expected.
Table 1. Hydraulic diameters of some common pipe geometries.
Pipe Geometry Hydraulic diameter (D
H
)
Circular
c H
A D
π
2
=
Square
c H
A D =
Triangle
c H
A D
4
3
3
2
=
Rectangle*
c H
A
X
Y
X
Y
D
⎟
⎟
⎟
⎟
⎠
⎞
⎜
⎜
⎜
⎜
⎝
⎛
+
=
1
1
2
Sinusoidal* **
c H
A
X
Y
X
Y
E
X
Y
X
Y
D
1
2
2
2
1
2
,
1
1
2 2
4
−
⎥
⎥
⎥
⎥
⎥
⎦
⎤
⎢
⎢
⎢
⎢
⎢
⎣
⎡
+
⎟
⎟
⎟
⎟
⎟
⎠
⎞
⎜
⎜
⎜
⎜
⎜
⎝
⎛
⎟
⎠
⎞
⎜
⎝
⎛
+
⎟
⎠
⎞
⎜
⎝
⎛
⎟
⎠
⎞
⎜
⎝
⎛
+ =
π
π
π
π
π π
* Y/X is the aspect ratio and
**
∫
− =
σ
σ σ σ
0
2 2
sin 1 ) , ( d k k E
is the elliptic integral of second kind (Tuma, 1979)
Figure 2 shows the variation of ψ with Re for five selected pipe geometries. The cross
section
c
A and
0
T were constant. The sinusoidal pipe geometry gives the lowest
dimensionless entropy generation. As Re is increased, the circular geometry becomes the best
choice. The total entropy generation tends to decrease initially and then increases with Re. ψ
includes terms form heat transfer and viscous friction. As Re is increased, the heat transfer
contribution decreases and that of viscous friction increases. Thus, for low Re number flow,
those geometries with higher surface areas are favorable. Circular geometry is the best choice
for high Re numbers. Square geometry appears to be good choice after the sinusoidal and
circular ones. ψ is found to be considerably high for triangular and rectangular pipes.
Therefore, they are inferior choices. Rectangular pipes would yield the lowest ψ when the
Ahmet Z. Sahin 328
aspect ratio becomes 1 (square geometry). For rectangular pipes with lower aspect ratio, the
entropy generation is higher and there may not even exist a minimum.
As 0 Re → , the effect of geometry disappears and the dimensionless entropy generation
becomes a function of ) (τ , i.e.,
( ) τ ψ + =
→
1 ln lim
0 Re
(45)
Comparison of pipe geometries using φ may be appropriate when the total heattransfer
rate is important as shown in Figure 3. Favorable pipe geometry appears to be circular but
sinusoidal and square geometries yield comparable results. Triangular and rectangular pipe
geometries are inferior choices because there is no optimum Re with minimum entropy
generation.
The pumping power required to overcome viscous friction is shown in Figure 4. The
circular geometry is superior to all other geometries. The results are similar to those for the
modified dimensionless entropy generation.
Figure 2. Dimensionless entropy generation ψ vs Re for various pipe geometries;
2 6
m 10 4
−
× =
c
A
and
2
W/m 500 = ′ ′ q .
First and Second Law Thermodynamics Analysis of Pipe Flow 329
Figure 3. Modified dimensionless entropy generation φ vs Re for various pipe geometries;
2 6
m 10 4
−
× =
c
A and
2
W/m 500 = ′ ′ q .
Figure 4. Pumping power to heattransfer ratio
r
P ve Re for various pipe geometries;
2 6
m 10 4
−
× =
c
A and
2
W/m 500 = ′ ′ q .
Ahmet Z. Sahin 330
Pipe Subjected to Constant Wall Temperature
We consider the smooth pipe with constant cross section shown in Figure 5. The surface
temperature T
w
is kept constant. An viscous fluid with mass flow rate m and inlet
temperature
0
T enters the pipe of length L . Density ρ , thermal conductivity k , and
specific heat
p
C of the fluid are assumed to be constant. Heat transfer to the bulk of the fluid
occurs at the inner surface with an average heattransfer coefficient h , which is a function of
temperature dependent viscosity.
The effect of viscosity on the average heat transfer coefficient is given by Shah and
London (1978)
. .
. .
n
b
c p
c p w
h Nu
h Nu
μ
μ
⎛ ⎞
= =
⎜ ⎟
⎝ ⎠
(46)
where the exponent n is equal to 0.14 for laminar flow. In the case of turbulent flow the
exponent n is equal to 0.11 for heating and 0.25 for cooling.
For fully developed laminar flow, constant property heat transfer coefficient
. . c p
h is
given by Incropera and DeWitt (1996)
D
k
Nu
D
k
h p c
p c
66 . 3 . .
. .
= = . (47)
Constant property heat transfer coefficient
. . c p
h for fully developed turbulent flow is
given by Eqn. (3). Correlations for the average Darcy friction factor, f , for a smooth pipe are
given in Eqns. (5) – (7) for laminar and turbulent flow cases.
Temperature Distribution
The rate of heat transfer to the fluid inside the control volume shown in Figure 1 is
dx T T D h dT C m Q
w p
) ( − = = π δ
(48)
where
2
/ 4 m U D ρ π =
Integrating Eq. (48), the bulk temperature variation of the fluid along the pipe can be
obtained as:
⎥
⎥
⎦
⎤
⎢
⎢
⎣
⎡
− − − = x
DC U
h
T T T T
p
o w w
ρ
4
exp ) ( (49)
First and Second Law Thermodynamics Analysis of Pipe Flow 331
Figure 5. Sketch of constant cross sectional pipe subjected to constant wall temperature.
The temperature variation along the pipe approaches the pipe wall temperature
exponentially assuming a uniform heat transfer coefficient evaluated at the bulk temperature
halfway between the inlet and outlet of the pipe.
Eq. (49) can be rewritten as:
⎟
⎠
⎞
⎜
⎝
⎛
− = x
D
St
4 exp θ (50)
where θ is the dimensionless temperature defined as,
w o
w
T T
T T
−
−
= θ
and St is the Stanton number defined as
p
h
St
UC ρ
= .
Total Entropy Generation
The total entropy generation within the control volume in Figure 1 can be written as:
gen
w
Q
dS mds
T
δ
= −
(51)
where the entropy change is,
Ahmet Z. Sahin 332
p
dT dP
ds C
T T ρ
= −
Substituting Eq. (48) in Eq. (51), the total entropy generation becomes
1
w
gen p
w p
T T
dS mC dT dP
TT C T ρ
⎡ ⎤
−
= −
⎢ ⎥
⎣ ⎦
(52)
The pressure drop in Eq. (52) is given by Bejan (1988)
2
2
f U
dP dx
D
ρ
= − (53)
Integrating Eq. (52) along the pipe length, L, using Eqs. (49) and (53), the total entropy
generation is obtained as (Sahin, 1998 and Sahin 2000)
⎭
⎬
⎫
⎩
⎨
⎧
⎥
⎦
⎤
⎢
⎣
⎡
−
−
+ − −
⎥
⎦
⎤
⎢
⎣
⎡
−
−
=
−
−
τ
τ
τ
τ
τ
λ
λ
λ
1
ln
8
1
) 1 (
1
1
ln
4
4
4 St
St
St
p gen
e
St
Ec
f e
e
C m S
(54)
where τ is the dimensionless temperature difference
w
o w
T
T T −
= τ ,
λ is the dimensionless length of pipe
D
L
= λ ,
and Ec is the Eckert number defined as
w p
T C
U
Ec
2
= .
The total rate of heat transfer to the fluid is obtained by integrating Eq. (48) along the
pipe length and can be written as:
). 1 )( (
4 λ St
o w p
e T T C m Q
−
− − =
(55)
First and Second Law Thermodynamics Analysis of Pipe Flow 333
Now defining a dimensionless entropy generation as:
) /(
o w
gen
T T Q
S
−
=
ψ
Eq. (54) can be written as
⎭
⎬
⎫
⎩
⎨
⎧
⎥
⎦
⎤
⎢
⎣
⎡
−
−
+ − −
⎥
⎦
⎤
⎢
⎣
⎡
−
−
−
=
−
−
−
τ
τ
τ
τ
τ
ψ
λ
λ
λ
λ
1
ln
8
1
) 1 (
1
1
ln
1
1
4
4
4
4
St
St
St
St
e
St
Ec
f e
e
e
(56)
Therefore, two dimensionless groups naturally arise in Eq. (56) as:
1
Stλ Π = (57)
and
2
Ec
f
St
Π = (58)
Thus Eq. (56) can be written in a compact form for the constant viscosity assumption as:
⎭
⎬
⎫
⎩
⎨
⎧
⎥
⎦
⎤
⎢
⎣
⎡
−
−
Π + − −
⎥
⎦
⎤
⎢
⎣
⎡
−
−
−
=
Π
Π −
Π −
Π −
τ
τ
τ
τ
τ
ψ
1
ln
8
1
) 1 (
1
1
ln
1
1
1
1
1
1
4
2
4
4
4
e
e
e
e
(59)
which is a function of three nondimensional parameters, namely τ ,
1
Π and
2
Π . Among
these parameters, τ represents the fluid inlet temperature T
o
and
1
Π represents the pipe
length L. Once the type of the fluid and the mass flow rate are fixed, the parameter
2
Π can
be calculated based on temperature analysis. Thus, τ and
1
Π are the two design parameters
that can be varied for determining the effects of pipe length and/or inlet fluid temperature on
the entropy generation.
Since
1
Π represents the length of the pipe, the dimensionless entropy generation defined
on the basis of total heat transfer rate to the pipe, ψ , decreases initially and then starts
increasing along the pipe length. The rate of increase in entropy generation approaches a
constant value as the total heat transfer rate to the fluid approaches its maximum value of
) (
max o w p
T T C m Q − =
(60)
For long pipes where 1
1
4
<<
Π −
e and τ >>
Π
1
4
e , it can be shown from Eq. (59) that the
entropy generation increases linearly with the slope
Ahmet Z. Sahin 334
2
2
1
Π
=
Π d
dψ
for the case of constant viscosity.
Since τ represents the difference between the temperature of the pipe surface and that of
the inlet fluid, the dimensionless entropy generation defined on the basis of total heat transfer
rate to the pipe, ψ , increases as τ increases due to the increase in the gap between the bulk
and wall temperatures. For small values of τ , the total entropy generation is due to the
viscous friction; and in the limit when 0 = τ , the total dimensionless entropy change using
Eq. (59) becomes:
1
4
2 1
0
1 2
1
lim
Π −
→
−
Π Π
=
e
ψ
τ
. (61)
It should be noted that the dimensionless entropy generation, ψ , in the above analysis is
a function of the total heat transfer rate, Q
, which in turn depends on the length of the pipe
and inlet fluid temperature. However, a modified dimensionless entropy generation can be
defined on the basis of unit heat capacity rate of fluid in the pipe as:
) /( T Q
S
C m
S
gen
p
gen
Δ
= =
φ (62)
where T Δ is the increase of the bulk temperature of the fluid in the pipe,
o
T L T T − = Δ ) ( .
Noting that
p
C m T Q
= Δ / is constant for fixed mass flow rate and
ψ φ ) 1 (
1
4Π −
− = e . (63)
φ , indicating the total entropy generation along the pipe, is expected to increase with the
increase in pipe length.
φ and ψ differ only for small values of
1
Π . For large values of
1
Π , ψ φ = as can be
seen from Eq. (63).
The modified dimensionless entropy generation defined, based on the unit heat capacity
rate through the pipe, φ , shows similar behavior to that of ψ . This was expected, since φ
and ψ are related through a constant factor of ( )
1
4
1
Π −
− e as given in Eq. (63).
First and Second Law Thermodynamics Analysis of Pipe Flow 335
Pumping Power to Heat Transfer Rate Ratio
The pumping power to heat transfer rate ratio is
2
4
r
D PU
P
Q
π Δ
=
. (64)
Using Eqs. (50) and (53), the pumping power to heat transfer rate ratio,
r
P is obtained as
) 1 ( 2
1
1
4
2 1
Π −
−
Π Π
=
e
P
r
τ
. (65)
Due to an increase in the bulk temperature and a decrease in viscosity, the pumping
power to heat transfer rate ratio may decrease initially and then increase, as the total heat
transfer rate to the fluid decreases as the bulk temperature approaches the wall temperature as
ψ is increased. For large values of ψ and constant viscosity, it can be shown from Eq. (65)
that the pumping ratio, Pr, increases linearly with the slope
τ 2
2
1
Π
=
Π d
dP
r
.
For small values of
1
Π the effect of the viscosity variation is small and in the limit,
τ 8
lim
2
0
1
Π
=
→ Π
r
P (66)
which is clearly a function of the fluid viscosity.
Effect of Pipe Cross Section
The dimensionless entropy generation as given in Eqn. (59) can be rewritten as function
of Re as (Sahin, 1998)
⎭
⎬
⎫
⎩
⎨
⎧
⎥
⎦
⎤
⎢
⎣
⎡
−
−
+ − −
⎥
⎦
⎤
⎢
⎣
⎡
−
−
−
=
−
−
−
τ
τ
τ
τ
τ
ψ
1
ln Re ) 1 (
1
1
ln
1
1
Re /
2
2
Re /
Re /
Re /
1
1
1
1
C
C
C
C
e
C e
e
e
(67)
where
1 1
Re 4 Π = C and
2
2
2
Re 8
Π
= C .
Ahmet Z. Sahin 336
For any given pipe geometry and stream of constant thermophysical property fluid,
dimensionless numbers C
1
and C
2
are constants. Therefore, the dimensionless total entropy
generation as given in Eq. (67) is a function of Re and τ only.
Values of Nu and Re) ( f for fully developed laminar flow are given in Shah and
London (1978) for a variety of pipe geometries. The hydraulic diameter D
H
for any given pipe
geometry can be calculated using the relation:
p
A
D
c
H
4 = .
Similarly,
ψ φ ) 1 (
Re /
1
C
e
−
− = (68)
and
) 1 (
Re
Re /
2 1
1
C r
e
C C
P
−
−
=
τ
(69)
For very low Re number flow, it is interesting to note that the importance of geometry
disappears. In the limit as
τ
τ
ψ +
⎟
⎠
⎞
⎜
⎝
⎛
+
=
→
1
1
ln lim
0 Re
(70)
As the temperature difference between the fluid inlet and the surface of the pipe, τ , is
increased, the entropy generation increases, however, the pumping power requirement per
unit heat transfer decreases. In the limit when τ approaches zero, the contribution of heat
transfer to the entropy generation disappears and
Re lim
2 1
0
C C =
→
ψ
τ
(71)
which indicates a linear increase in entropy generation that is solely due to the viscous
friction as expected.
Figure 6 shows the variation of dimensionless entropy generation, ψ , with Re number
for five selected pipe geometries; namely circular, square, equilateral triangle, rectangular
(aspect ratio of 1/2) and sinusoidal (aspect ratio 2 / 3 ). The cross sectional area of each
pipe, A
c
, and the inlet fluid temperature, T
o
, are kept constant. For low Re numbers,
sinusoidal pipe geometry gives lowest entropy generation of all the pipe geometries
considered. However, as the Re number is increased, circular pipe geometry becomes the best
choice. In general, the total entropy generation tends to decrease initially and then increases
while the Re number is increased.
First and Second Law Thermodynamics Analysis of Pipe Flow 337
The dimensionless total entropy generation, ψ , includes contributions due to heat
transfer and viscous friction. As the Re number is increased, the contribution due to heat
transfer decreases and that of viscous friction increases. Therefore, for low Re number flow
for which the viscous frictional contribution can be neglected, those pipe geometries with
higher surface areas are favorable. This is because of the fact that the total heat transfer is a
function of the surface area for fixed cross sectional area and length of pipe. On the other
hand, circular pipe geometry is the best choice for high Re numbers where viscous
frictional contribution to the entropy generation becomes dominant. Sinusoidal and square
pipe geometries appear to be good choices after the circular pipe for the high Re number
region. The dimensionless entropy generation is found to be high for triangular and
rectangular pipes almost for the entire Re number region. Therefore, they are the worst
choices from the point of view of entropy generation. This conclusion is obtained only if
the above comparison of entropy generation is made by keeping the flow cross sections of
the pipes constant. Comparison of pipe geometries with same equivalent diameter, on the
other hand, leads to the results in which triangular and square pipes seem to be better
choices than the circular pipe geometry. This may be misleading. Rectangular pipes would
yield the best results when the aspect ratio becomes 1, which corresponds to square
geometry, and the square geometry is almost nowhere better than the circular one. For
lower aspect ratio rectangular pipes there may not even exist a minimum entropy
generation.
For very low Re number flow, it is interesting to note that the importance of geometry
disappears and all the curves approach to the same value as Re number approaches zero. This
limit of dimensionless entropy generation is a function of the temperature difference between
the inlet fluid stream and surface of the pipe, τ , as
τ
τ
ψ +
⎟
⎠
⎞
⎜
⎝
⎛
+
=
→
1
1
ln lim
0 Re
(72)
The modified dimensionless entropy generation, φ , versus Re number is given in Figure
7. The most favorable pipe geometry appears to be the sinusoidal type in this case, for the
whole range of Re numbers. The circular geometry is the next favorable geometry as far as
the modified dimensionless entropy generation is concerned; however, the differences in
entropy generation for the geometries selected are not significant. It should be noted that,
there is no optimum Re number to provide a minimum modified entropy generation, since the
total heat transfer increases as Re number is increased.
Figure 8 shows the pumping power requirement to overcome the viscous friction. Clearly
the circular geometry is superior to all other geometries. Sinusoidal geometry appears to be a
bad choice from the point of view of pumping power to heat transfer ratio. Triangular pipe is
the worst choice in this case.
Ahmet Z. Sahin 338
Figure 6. Dimensionless entropy generation ψ versus Re number for various pipe geometries,
2 7
m 10 4
−
× =
c
A and 01 . 0 = τ .
Figure 7. Modified dimensionless entropy generation φ versus Re number for various pipe
geometries m 10 4× =
c
A and 01 . 0 = τ .
First and Second Law Thermodynamics Analysis of Pipe Flow 339
Figure 8. Pumping power to heat transfer ratio
r
P versus Re number for various pipe geometries
2 7
m 10 4
−
× =
c
A and 01 . 0 = τ .
Appendix: Viscosity Dependence on Temperature
Experimentally it is known that the viscosity of liquids vary considerably with
temperature. Around room temperature (293 K), for instance, a 1% change in temperature
produces a 7% change in viscosity of water and approximately a 26% change in viscosity of
glycerol (Sherman, 1990).
As a first approximation, the variation of viscosity with temperature can assumed to be
linear,
( ) ( )
ref ref
T b T T μ μ = − −
where b is a fluid dependent dimensional constant and
ref
T is a reference temperature (=
293 K). This would be a reasonable approximation if the bulk temperature variation is small.
However, for highly viscous liquids the variation of viscosity with temperature is exponential
and a more accurate empirical correlation of liquid viscosity with the temperature is given by
Sherman (1990) as
1 1
( ) exp
n
ref
ref ref
T
T B
T T T
μ μ
⎡ ⎤ ⎛ ⎞ ⎛ ⎞
= −
⎢ ⎥ ⎜ ⎟ ⎜ ⎟
⎜ ⎟ ⎜ ⎟
⎢ ⎥
⎝ ⎠ ⎝ ⎠ ⎣ ⎦
Ahmet Z. Sahin 340
where n and B are fluid dependent constant parameters. In the present work, both the linear
and the exponential viscosity models were used. In addition, a constant viscosity model (in
which the numerical value of viscosity at a reference bulk temperature is taken as constant) is
also included in order to see the significance of the variation of viscosity on the results.
Nomenclature
p
C Specific heat (J/kg K)
D Diameter (m)
Ec Eckert number,
( )
2
/ /
p
U C q h
⎡ ⎤ ′′
⎣ ⎦
f Average Darcy friction factor
h Average heat transfer coefficient (W/m
2
K)
k Thermal conductivity (W/m K)
L Length of the pipe (m)
m Mass flow rate (kg/s)
Nu Average Nusselt number, / hD k
P Pressure (N/m
2
)
Pr Prandtl number
r
P Pumping power to heat transfer rate ratio
q′′ Heat flux (W/m
2
)
Q
Total heat transfer rate (W)
Re Reynolds number, /
b
UD ρ μ
s Entropy (J/kg K)
gen
S
Entropy generation (W/K)
St Stanton number,
( )
/
p
h UC ρ
T Temperature (K)
0
T Inlet fluid temperature (K)
ref
T Reference temperature (=293 K)
w
T Wall temperature of the pipe (K)
U Fluid bulk velocity (m/s)
x Axial distance (m)
Greek Symbols
T Δ Increase of fluid temperature (K)
μ Viscosity (N s/m
2
)
First and Second Law Thermodynamics Analysis of Pipe Flow 341
b
μ Viscosity of fluid at bulk temperature (N s/m
2
)
ref
μ Viscosity of fluid at reference temperature (N s/m
2
)
w
μ Viscosity of fluid at wall temperature (N s/m
2
)
λ Dimensionless axial distance, / L D
1
Π Modified Stanton number, Stλ
2
Π Dimensionless group, / fEc St
ψ Dimensionless entropy generation,
( )
0
/ /
gen
S Q T
φ Modified dimensionless entropy generation,
( )
/ /
gen
S Q T Δ
ρ Density (kg/m
3
)
τ Dimensionless wall heat flux,
( )
0
/ / q h T ′′
θ Dimensionless temperature, ( ) ( )
0
/ / T T q h ′′ −
. . c p Constant properties
References
[1] Bejan, A., Advanced Engineering Thermodynamics, John Wiley and Sons, New York
(1988).
[2] Bejan, A., Entropy Generation Minimization, CRC Press, New York (1996).
[3] Incropera, F.P. and DeWitt, D.P., Introduction to Heat Transfer, Third Ed., John Wiley,
New York (1996).
[4] Kays, W. M. and Crawford, M. E., Convective Heat and Mass Transfer, Third Ed.
McGraw Hill, New York (1993).
[5] Kays, W. M. and London, A. L., Compact Heat Exchangers, 3rd Ed., McGraw Hill,
New York (1984).
[6] Kreith, F. and Bohn, M.S., Principles of Heat Transfer, fifth ed., West Publ. Co., New
York (1993).
[7] Shah, R. K., Compact Heat Exchanger Design Procedures, in: Heat Exchangers,
ThermalHydraulic Fundamentals and Design, Ed: Kakac, S., Bergles, A.E., and
Mayinger, F., McGraw Hill, New York (1981).
[8] Shah, R.K., and London, A.L., Laminar Flow Forced Convection in Ducts, Academic
Press, New York (1978).
[9] Sahin, A.Z., Irreversibilities in Various Duct Geometries with Constant Wall Heat Flux
and Laminar Flow, Energy  The International Journal, vol. 23, no. 6, pp. 465 473,
(1998).
[10] Sahin, A.Z., A Second Law Comparison for Optimum Shape of Duct Subjected to
Constant Wall Temperature and Laminar Flow, Heat and Mass Transfer, vol. 33, pp.
425430, 1998.
Ahmet Z. Sahin 342
[11] Sahin, A.Z., Second Law Analysis of Laminar Viscous Flow Through a Duct Subjected
to Constant Wall Temperature, ASME Journal of Heat Transfer, vol. 120, no. 1, pp. 76
83, (1998).
[12] Sahin, A.Z., The Effect of Variable Viscosity on the Entropy Generation and Pumping
Power in a Laminar Fluid Flow through a Duct Subjected to Constant Heat Flux, Heat
and Mass Transfer, vol. 35, pp. 499506, (1999).
[13] Sahin, A.Z., Entropy Generation in Turbulent Viscous Flow Through a Smooth Duct
Subjected to Constant Wall Temperature, International Journal of Heat and Mass
Transfer, vol. 43, no. 8, pp. 14691478, (2000).
[14] Sahin, A.Z., Entropy Generation and Pumping Power in a Turbulent Fluid Flow through
a Smooth Pipe Subjected to Constant Heat Flux, Exergy, an International Journal, vol.
2, no. 4, pp. 314321, (2002).
[15] Sherman, F.S., Viscous Flow, McGraw Hill Book Co., New York (1990).
[16] Tuma, J.J., Engineering Mathematics Handbook, Second Ed., McGraw Hill Book. Co.,
New York (1979).
In: Fluid Mechanics and Pipe Flow ISBN: 9781607410379
Editors: D. Matos and C. Valerio, pp. 343363 © 2009 Nova Science Publishers, Inc.
Chapter 10
SINGLEPHASE INCOMPRESSIBLE FLUID FLOW
IN MINI AND MICROCHANNELS
Lixin Cheng
*
School of Engineering, University of Aberdeen, King’s College, Aberdeen,
AB24 3UE, Scotland, the UK
Abstract
This chapter aims to present a stateoftheart review on singlephase incompressible fluid
flow in mini and microchannels. First, classification of mini and microchannels is
discussed. Then, conventional theories on laminar, laminar to turbulent transition and
turbulent fluid flow in macrochannels (conventional channels) are summarized. Next, a brief
review of the available studies on singlephase incompressible fluid flow in mini and micro
channels is presented. Some experimental results on single phase laminar, laminar to turbulent
transition and turbulent flows are presented. The deviations from the conventional friction
factor correlations for singlephase incompressible fluid flow in mini and microchannels are
discussed. The effect factors on mini and microchannel singlephase fluid flow are analyzed.
Especially, the surface roughness effect is focused on. According to this review, the future
research needs have been identified. So far, no systematic agreed knowledge of singlephase
fluid flow in mini and microchannels has yet been achieved. Therefore, efforts should be
made to contribute to systematic theories for microscale fluid flow through very careful
experiments.
Keywords: Singlephase flow; Minichannel; Microchannel; Incompressible fluid flow;
Pressure drop; Friction factor; Surface roughness.
1. Introduction
Miniaturization has recently become the key word in many advanced technologies as well
as traditional industries. The miniaturized systems are being progressively applied in
commercial sectors such as the electronics, chemical, pharmaceutical, and medical industries,
*
Email address: lixincheng@hotmail.com
Lixin Cheng 344
as well as in the military sector, for biological and chemical warfare defense, as an example.
With the emergence of microscale thermal, fluidic, and chemical systems, the development
of ultracompact heat exchangers, miniature and micro pumps, miniature compressors, micro
turbines, microreactors, micro thermal systems for distributed power production has become
an important agenda of many researchers, research institutions, and funding agencies.
Miniaturization is essential not simply from the standpoint of producing compact systems, but
for the challenge of fluid flow with yet smaller and smaller channel sizes, as is the case for
cooling of electronic devices, microfluidic components and sensors. In addition, it has been
shown that proper miniaturization by use of mini and microchannels can result in higher
system efficiency. However, fluid flow phenomena in mini and microchannels are quite
different from those in conventional size channels. Therefore, it is of great significance to
understand these fundamental aspects.
Due to the significant differences of fluid flow phenomena in mini and microchannels
as compared to conventional channels or macroscale channels, one very important issue
should be clarified about the distinction between mini and microscale channels and macro
scale channels. However, a universal agreement is not yet clearly established in the literature.
Furthermore, although research in mini and microchannels has been greatly increasing due
to the rapid growth applications which require fluid flows in the tiny channels, the available
studies reveal contradictory results in singlephase friction factors, the transition for laminar
and turbulent flows and the effect factors. There are still significant discrepancies between the
experimental results obtained by different researchers.
At first, one very important issue should be the clarification of the distinction between
microscale channels and macroscale channels when studying fluid flow in mini and micro
channels. However, a universal agreement is not yet clearly established in the literature.
Instead, there are various definitions on this issue [16]. Here, just to show one example,
based on engineering practice and application areas such as refrigeration industry in the small
tonnage units, compact evaporators employed in automotive, aerospace, air separation and
cryogenic industries, cooling elements in the field of microelectronics and microelectro
mechanicalsystems (MEMS), Kandlikar et al. [6] defined the following ranges of hydraulic
diameters d
h
which are attributed to different classifications:
Conventional channels: d
h
> 3 mm.
Minichannels: 200 μ m < d
h
≤ 3 mm.
Microchannels: 10 μ m < d
h
≤ 200μ m.
Transitional microchannels: 1 μ m < d
h
≤ 10μ m.
Transitional nanohannels: 0.1 μ m < d
h
≤ 1μ m.
Nanochannels: d
h
≤ 0.1 μ m.
In the case of noncircular channels, it is recommended that the minimum channel
dimension, for example, the short side of a rectangular crosssection should be used in place
of the diameter d.
In the available studies on fluid flow in mini and microchannels, some researchers have
concluded that the conventional theories work for mini and microchannels while others have
showed that the conventional theories do not work well. Therefore, these controversies should
be clarified. As the channel size becomes smaller, some of the conventional theories for
(bulk) fluid, energy and mass transport need to be revisited for validation. There are two
SinglePhase Incompressible Fluid Flo in Mini and Microchannels 345
fundamental elements responsible for departure from the conventional theories at mini and
microscale. For example, differences in modeling fluid flow in mini and microchannels
may arise as a result of (i) a change in the fundamental process, such as a deviation from the
continuum assumption for fluid flow, or an increase influence of some additional forces, such
as electrokinetic forces, etc. (ii) uncertainty regarding the applicability of empirical factors
derived from experiments conducted at larger scales, such as entrance and exit loss
coefficients for fluid flow in pipes, etc., (iii) uncertainty in measurements at mini and micro
scale, including geometrical dimensions and operating parameters.
In this chapter, the available studies of singlephase incompressible fluid flow in mini
and microchannels are reviewed to address these relevant important issues.
2. SinglePhase Frictional Pressure Drop Methods
in Macrochannels
2.1. Laminar and Turbulent Flow
The flow of a fluid in a pipe may be laminar or turbulent flow, or in between transitional
flow. Figure 1 shows the x component of the fluid velocity as a function of time t at a point A in
the flow for laminar, transitional and turbulent flows in a pipe. For laminar flow, there is only
one component of velocity. For turbulent flow, the predominant component of the velocity is
also along the pipe, but it is unsteady (random) and accompanied by random components
normal to the pipe axis. For transitional flow, both laminar and turbulent features occur.
V
A
A Q
t
Turbulent
Transitional
Laminar
x
Figure 1. Time dependence of fluid velocity at a point.
Lixin Cheng 346
For pipe flow, the most important dimensionless parameter is the Reynolds number Re,
the ratio of the inertia to viscous effects in the flow, which is defined as
Re
Vd ρ
μ
= (1)
where V is the average velocity in the pipe. The flow in a pipe is laminar, transitional or
turbulent provided the Reynolds number Re is small enough, intermediate or large enough. It
should be pointed out here that the Reynolds number ranges for which laminar, transitional or
turbulent pipe flows are obtained cannot be precisely given. The actual transition from
laminar to turbulent pipe flow may take place at various Reynolds numbers, depending on
how much the flow is distributed by vibrations of the pipe, roughness of the entrance region,
and the like. For general engineering purposes (i.e. without undue precautions to eliminate
such disturbances), the following values are appropriate: the flow in a round pipe is laminar if
the Reynolds number Re is less than approximately 2100 and the flow in is turbulent if the
Reynolds number Re is greater than approximately 4000. For the Reynolds numbers between
these two limits, the flow may switch between laminar and turbulent conditions in an
apparently random fashion (transitional flow).
2.2. Entrance Region (Developing Flow) and Fully Developed Flow
The region of flow near where the fluid enters a pipe is termed the entrance region
(developing flow) and is illustrated in Figure 2. The fluid typically enters the pipe with a
nearly uniform velocity profile at section 1. As the fluid moves through the pipe, viscous
effects cause it to stick to the pipe wall (the noslip boundary condition). Thus, a boundary
layer in which viscous effects are important is produced along the pipe wall such that the
initial velocity profile changes with distance along the pipe, x, until the fluid reaches the end
of the entrance length, section 2, beyond which the velocity profile does not vary with x. The
boundary layer has grown in thickness to completely fill the pipe. Viscous effects are of
considerable importance within the boundary layer. For fluid outside the boundary layer
(within the inviscid core surrounding the centerline from 1 to 2), viscous effects are
negligible.
The shape of the velocity profile in the pipe depends on whether the flow is laminar or
turbulent, as does the length of the entrance region, l
e
. As with many other properties of pipe
flow, the dimensionless entrance length l
e
/d, correlates quite well with the Reynolds number
Re. Typical entrance lengths are given by
0.06Re
e
l
d
= for laminar flow (2)
1/ 6
4.4Re
e
l
d
= for turbulent flow (3)
Calculation of the velocity profile and pressure distribution within the entrance region
(developing flow) is quite complex. However, once the fluid reaches the end of the entrance
region, section 2 in Figure 2, the flow is simpler to describe because the velocity is a function
SinglePhase Incompressible Fluid Flo in Mini and Microchannels 347
of only the distance from the pipe centerline, r, and independent of x. The flow after section 2
is termed fully developed flow.
Entrance region flow l
e
Fully developed flow
Inviscid core
Boundary layer
1 2
x
d
r
V
Figure 2. Entrance region, developing flow and fully developed flow in a pipe.
2.3. SinglePhase Friction Pressure Drop Methods
The nature of the pipe flow is strongly dependent on weather the flow is laminar or
turbulent. This is a direct consequence of the differences in the nature of the shear stress in
laminar and turbulent flows. The shear tress in laminar flow is a direct result of momentum
transfer among the randomly moving molecules (a microscopic phenomenon). The shear
stress in turbulent flow is largely a result of momentum transfer among the randomly moving,
finitesized bundles of fluid particles (a macroscopic phenomenon). The net result is that the
physical properties of the shear stress are quite different for laminar than for turbulent flow.
The friction pressure drop for both laminar and turbulent flow can be expressed as
2
2
l V
p f
d
ρ
Δ = (4)
For fully developed laminar flow in a circular pipe, the friction factor f is simply as
64
Re
f = (5)
and the value of f is independent of the relative roughness
d
ε
.
For turbulent flow, the dependence of the friction factor on the Reynolds number Re is
much more complex than that given by Eq. (5) for laminar flow. For fully developed
turbulent flow and transition from laminar to turbulent flow. The Moody chart [7] shown in
Figure 3 provides the friction factor f, which can be expressed as
Re, f
d
ε
φ
⎛ ⎞
=
⎜ ⎟
⎝ ⎠
(6)
Figure 3. Friction factor as a function of the Reynolds number and relative roughness for circular pipesthe Moody chart [7].
SinglePhase Incompressible Fluid Flo in Mini and Microchannels 349
For turbulent flow, the friction factor f is a function of the Reynolds number Re and
relative roughness
d
ε
. The results are obtained from an exhaustive set of experiments and
usually presented in terms of a curvefitting formulae or the equivalent graphical form.
In commercially available pipes, the roughness is not as uniform and well defined as in
the artificially roughened pipes. However, it is possible to obtain a measure of the effective
relative roughness of typical pipes and thus to obtain the friction factor. Typical roughness
values for various pipe surfaces are also given in Figure 3. It should be noted that the values
of
d
ε
do not necessarily correspond to the actual values obtained by a microscopic
determination of the average height of the roughness of the surface. They do, however,
provide the correct correlation for friction factor f. The following characteristics are observed
from Figure 3 for laminar flow, friction factor is calculated by Eq. (2), which is independent
of relative roughness. For very large Reynolds number Re, the friction factor is a function of
the relative roughness
d
ε
as
f
d
ε
φ
⎛ ⎞
=
⎜ ⎟
⎝ ⎠
(7)
which is independent of the Reynolds number Re. For such flows, commonly termed
completely turbulent flow (or wholly turbulent flow), the laminar sublayer is so thin (its
thickness decreases with increasing the Reynolds number Re) that the surface roughness
completely dominates the character of the flow near the wall. Hence, the pressure drop
required is a result of an inertiadominated turbulent shear stress rather than the viscosity
dominated laminar shear stress normally found in the viscous sublayer.
For flows with moderate values of Re, the friction factor is indeed dependent on both the
Reynolds number Re ad relative roughness
d
ε
. Flow in the range of 2100< Re <4000 is a
result of the fact that the flow in this transition range may be laminar or turbulent (or an
unsteady mix of both) depending on the specific circumstances involved.
Note that even for smooth pipes (ε = 0), the friction factor is not zero. That is, there is a
head loss in any pipe, no mater how smooth the surface is made. This is a result of the noslip
boundary condition that requires any fluid to stick to any solid surface it flows over. There is
always some microscopic surface roughness that produces the noslip behavior (and thus the
friction factor f does not equal 0) on the molecular level, even when the roughness is
considerably less that the viscous sublayer thickness. Such pipes are called hydraulically
smooth, and for turbulent flow in this flow, the Blasius [8] equation is used for friction factor
as
0.25
0.3164
Re
f = (8)
where 4000< R e < 100,000.
Lixin Cheng 350
The Moody chart shown in Figure 3 is universally valid for all steady fully developed
incompressible pipe flows. The following equation from the Colebrook and White [9] is valid
for the entire nonlaminar range of the Moody chart, as
1 / 2.51
2log
3.7 Re
d
f f
ε
⎛ ⎞
= − + ⎜ ⎟
⎜ ⎟
⎝ ⎠
(9)
In fact, the Moody chart is a graphical representation of this equation, which is an
empirical fit of the pipe flow pressure drop data. A difficulty with its use is that it is implicit
in the dependence of f. that is, for given conditions (Re and
d
ε
), it is not possible to solve f
without some sort of iterative scheme.
For easytouse, Swamee and Jain [10] proposed an explicit ColebrookWhite equation as
follows:
2
0.9
0.25
/ 5.74
log
3.7 Re
f
d ε
=
⎡ ⎤ ⎛ ⎞
+
⎜ ⎟ ⎢ ⎥
⎝ ⎠ ⎣ ⎦
(10)
which matches the ColebrookWhite equation within 1% for 10
6
< ε/d < 10
2
and 5000 < Re
< 10
8
.
Churchill [11] proposed a more complicated expression for all flow regimes and all the
relative roughnesses, which agrees well with the Moody diagram:
( )
1/12
1.5
16
12 16
0.9
8 1 37530
8 2.457ln
Re Re
7 / Re 2.7 /
f
d ε
−
⎧ ⎫
⎡ ⎤
⎛ ⎞
⎪ ⎪ ⎛ ⎞ ⎛ ⎞
⎢ ⎥
⎜ ⎟ = + +
⎨ ⎬
⎜ ⎟ ⎜ ⎟
⎜ ⎟ ⎢ ⎥
⎝ ⎠ ⎝ ⎠ +
⎪ ⎪
⎝ ⎠
⎣ ⎦
⎩ ⎭
(11)
For noncircular ducts: the hydraulic diameter d
h
is used. The hydraulic diameter of a non
circular enclosure is four times the cross sectional area divided by the wetted perimeter. The
wetted perimeter is the edge of the cross sectional area that is in direct contact with the flow
medium, as
4
h
wetted
A
d
P
= (12)
Given the complexities of viscous sublayers, turbulence shear stress and surface
roughness, etc., the use of the hydraulic diameter method introduces some limitations. For
one, the use of the hydraulic diameter d
h
is only valid if the ration of the duct’s height to
width is less than about 3 or 4 [8]. Use the hydraulic diameter method to estimate head loss in
laminar flow can also introduce large errors. This is due to friction from the action of
viscosity throughout the whole body of the flow. It is independent of surface roughness and is
not associated with the region close to the boundary walls. Although the details of the flows
SinglePhase Incompressible Fluid Flo in Mini and Microchannels 351
in noncircular conduits depend on the exact crosssectional shape, many round pipe results
can be carried over, with slight modification, to flow in conduits of other shapes.
Practical, easytouse results can be obtained as follows: regardless of the crosssectional
shape, there are no inertia effects in fully developed laminar pipe flow. Thus, the friction
factor can be written as
Re
h
C
f = (13)
where the constant C is the Poiseuille number which depends on the particular shape of the
duct, and Re
h
is the Reynolds number based on the hydraulic diameter d
h
. The hydraulic
diameter is also used in the definition of the friction factor and the relative roughness. Typical
values of C for concentric annulus and rectangle channels are given in Table 1 along with the
hydraulic diameter. Note that the value of C is relatively insensitive to the shape of the
conduit. Unless the cross section is very “thin” in some sense, the value of C is not too
different from its circular pipe value, C = 64 as shown in Eq. (5). Shah and London [12]
provide the following correlation for a rectangular channel with short side a and long side b,
and a channel aspect ratio a
c
= a/b.
( )
2 3 4 5
Re 24 1 1.3553 1.9467 1.7012 0.9564 0.2537
c c c c c
C f a a a a a = = − + − + − (14)
Table 1. Friction factors for laminar flow in noncircular ducts [8]
Shape
Hydraulic
diameter
Parameter
Re
h
C f =
d
1
/d
2
0.0001 71.8
0.01 80.1
0.1 89.4
0.6 95.6
Concentric annulus
2 1 h
d d d = −
1.0 96
a/b
0 96
0.05 89.9
0.10 84.7
0.25 72.9
0.50 62.2
0.75 57.9
Rectangle
a
b
2
h
ab
d
a b
=
+
1.00 56.9
Lixin Cheng 352
Once the friction factor is obtained, the calculations for noncircular conduits are identical
to those for circular pipes.
Calculations for fully developed turbulent flow in ducts of noncircular cross section are
usually carried out by the Moody chart data for circular pipes with the diameter replaced by
the hydraulic diameter and the Reynolds number based on the hydraulic diameter. Such
calculations are usually accurate to within about 15%.
3. SinglePhase Frictional Pressure Drop Methods
in Mini and MicroChannels
To explore the fundamental physical mechanisms of fluid flow in mini and micro
channels, many effects, including the size effect, the surface roughness effect, viscous effect,
electrostatic force effect in the channel wall, surface geometry and the measurement errors,
etc. should be taken into account. The conventional theories for macrochannels are the
fundamental to investigation of singlephase fluid flow in mini and microchannels. So far, a
number of studies have been performed to study the hydrodynamic characteristics of
incompressible fluid flow in mini and microchannels. However, divergences of conclusions
still exist in quite a few fundamental understandings of the microscale fluid flow phenomena.
In this section, some recent studies of incompressible fluid flow and in mini and micro
channels are reviewed and the main conclusions from these studies are summarized.
There is big contradictory regarding the applicability of the conventional theories to
microscale fluid flow phenomena. Several good reviews on fluid flow in mini and micro
channels were presented by Kandliar [6], Celata et al. [13] Guo and Li [14], Sobhan and
Garimella [15] and Morini [16]. There are large deviations among the available studies in the
literature. Little agreement among the experimental results of singlephase pressure drops in
the laminar and turbulent flow regimes by different researchers has been reached. Although
some experimental results agree with the conventional theories, most of the experimental
results deviate from the conventional theories, with both under and overpredictions
obtained. Especially, the experimental results also show different trends of variation relative
to the conventional predictions in the laminar and turbulent flow regimes. In addition, quite
different results on the transition from laminar to turbulent regimes have been obtained by
different investigators. Some reported that the transition occurred at lower Reynolds numbers
while others reported that the transition agreed with the conventional theory. In general, the
fundamental aspects of fluid flow in mini and microchannels are not well understood so far.
Here only some representative studies in the literature are reviewed below to show these
differences.
Mala and Li [17] conducted experiments of water flowing in microchannels with
diameters from 50 to 254 μm. Their lengthtodiameter ratio was from 1200 to 5000 and the
materials were fused silica and stainless steel. Their experimental results showed significant
higher values than those predicted by the conventional theory. The difference increased as the
microtube diameter decreased. The effect is caused by the surface roughness or an early
transition from laminar to turbulent flow regime. They found an early transition from laminar
to turbulent flow in the Reynolds number from 300 to 900, while the flow changed to fully
developed turbulent flow at the Reynolds number larger than 1000 to 1500.
SinglePhase Incompressible Fluid Flo in Mini and Microchannels 353
Yu et al. [18] conducted experiments on nitrogen gas and water flowing in micro
channels with diameter 19, 52 and 102 μm. The test Reynolds number ranged from 250 to
20,000. Their experimental results showed that the friction factor was lower than the
conventional theory for fully developed laminar flow while the transition from laminar to
turbulent flow occurred in the Reynolds number range 2000 to 6000.
Xu et al. [19] conducted experiments on water flowing in microchannels with hydraulic
diameter from 50 to 300 um and their test Reynolds number was from 50 to 1500. Their
experimental results showed that the transition to from laminar to turbulent flow region did
not occur in their test Reynolds number range. They found that the flow characteristics
deviated from conventional theory when the channel dimensions were below 100 μm. Their
friction factor was smaller than the predicted results by the conventional theory. Later on, Xu
et al. [20] reported a similar study of water flow in microchannels with hydraulic diameter
from 30 to 344 and the Reynolds number from 20 to 4000. Their experimental results showed
that the experimental friction factors agreed with the conventional theories. Although the
channel sizes were similar using the same test fluid, their experimental results are quite
different.
Judy [21] conducted experiments of water, hexane and isopropanol flowing in fused
silica capillary tubes with diameters from 20 to 150 μm. Their experimental results showed
that when the tube diameter was lower than 100 μm, the friction factor deviated significantly
from the conventional theory. This is in consistence with that of Xu et al. [19]. The deviation
is independent of the Reynolds number and depended on the tube diameter. However, later
on, they did very accurate experiments on fluid flow in microchannels with diameters from
15 to 150 μm with the same fluids and the similar test conditions used in their previous study
[22]. Their experimental results showed that the friction factors were in good agreement with
the conventional theory. The value of the Poiseuille number C was found close to 64,
considering the experimental error and it was independent of the Reynolds number for Re <
2000.
Gao et al. [23] conducted experiments on water flowing in microchannels with diameters
from 200 to 1923 μm. Their experimental results were in good agreement with the
conventional theories. Figure 4 shows their results on the influence of the Reynolds number
and channel height on the Poiseuille number. Both the Shah and London [12] and the Balsius
[8] correlations suitably predicted their experimental data for laminar and turbulent regimes.
No scale effects were found in their experiments for flow hydrodynamics. Similar
conclusions were obtained by Caney et al. [24], Grimella and Singal [25], Qi et al. [26], Lee
and Lee [27], Warrier et al. [28] and Qu and Mudawar [29] in their singlephase friction
pressure drop experiments in microchannels.
Celata et al. [30] conducted experiments of R114 liquid flowing in 6 parallel stainless
steel microtubes with a diameter of 130 μm and a tube length of 90 mm for both laminar and
turbulent flows (Reynolds number from 100 to 8000). The relative roughness of the test tubes
was 2.65%. Their experimental results showed that the singlephase pressure drop in laminar
flow agreed with the conventional theory. The laminar to turbulent flow transition was in the
Reynolds number range of 1880 to 2480. They noted that the high relative roughness played
an important role in the laminar to turbulent transition. Figure 5 shows their test results of
friction factor in both laminar, transitional and turbulent flows compared to the Blasius
Lixin Cheng 354
equation and the Colebrook correlation. The Blasuis correlation underpredicts their data while
the Colebrook correlation overpredicted their data.
Figure 4. Influence of Reynolds number and channel height on Poiseuille number: + e = 1 mm, × e =
0.7 mm, ● e = 0.5 mm, e = 0.4 mm, ▲e = 0.3 mm, ◊ e = 0.2 mm, □ e = 0.1 mm, – – – Blasius law
[23].
Figure 5. Comparison of experimental friction factors to the Colebrook and the Blasius correlations
[30].
SinglePhase Incompressible Fluid Flo in Mini and Microchannels 355
Wibel and Ehrhard [31] conducted experiments on the laminar to turbulent transition of
deionized water flowing in rectangular stainless steel microchannels with a hydraulic
diameter of about 133 μm. Three aspect ratios of 1:1, 1:2 and 1:5 were studied. Their results
showed that the laminar to turbulent transition was in the Reynolds number range of 1900
2200, which agreed with the conventional theory.
Li et al. [32] conducted experiments on singlephase friction pressure drops of deionized
water flowing in smooth fused silica microtubes and rough stainless steel microtubes with
diameters of 50 to 100 μm and 373 to 1570 μm, respectively. Their test Reynolds number was
from 20 to 2400. For the stainless steel tubes, the surface relative roughnesses of 2.4%, 1.4%
and 0.95% were tested. Figure 6 shows their experimental friction factors for the fused silica
tubes compared to the conventional theory and Figure 7 shows those for stainless steel tubes.
Their experimental friction factors were well predicted by the conventional theory for the
smooth silica tubes while for the roughness stainless steel tubes, their experimental friction
factors were higher than the predictions by the conventional theory and increased with the
relative roughness. For deionized water flowing through stainless steel microtubes with a
relative surface roughness less than about 1.5%, the friction factors also agreed with the
conventional theory. When the relative surface roughness was larger than 1.5%, the
experimental friction factor of the microtube flow showed an appreciable deviation from the
conventional theory.
Figure 6. Comparison of experimental friction factors to the conventional theory for the fused silica
tubes [32].
In general, the available experimental results on microscale fluid flow in the literature
have demonstrated that there are big disagreements among the studies on the single phase
friction pressure drops in terms of the friction factors for laminar and turbulent flows and the
laminar to turbulent flow transition. It seems that conventional theory works for relative large
channels while it does not work well for relative smaller channels. So far, no systematic
agreed theories on microscale singlephase fluid flows have yet been achieved. Quite big
Lixin Cheng 356
contradictory exists in the available studies. Some researchers reported that the friction
factors for laminar fully developed flow were lower than the conventional theory, some
reported that the friction factors for laminar fully developed flow were higher than the
conventional values while some reported that the friction factor could be predicted by the
conventional theory. Some reported that the Poiseuille number for laminar fully developed
flow depends on the Reynolds number while some reported that it did not. In fact, the friction
factor depends on the material of the microchannel walls (metals, semiconductors and so
on) and the test fluid (polar fluid or not), thus evidencing the importance of electroosmotic
phenomena to microscales. Furthermore, the dependence of the friction factor on the relative
roughness of the microchannel wall also exits in laminar flow regime. Regarding the
laminartoturbulent flow transition in microchannels, no agreement on the transitional
Reynolds number has been reached so far.
Figure 7. Comparison of experimental friction factors to the conventional theory for the stainless steel
tubes [32].
It should be pointed out that no study on the entrance length (developing flow) in mini
and microchannels is available in the literature so far. This poses a question: whether the
correlation for entrance length in macrochannels can be used for mini and microchannels?
No doubt, it is very important to identify the developing and the fully developed flow
correctly in mini and microchannels. Simply using the criteria for macrochannels might
results in very big errors in the microscale fluid flow research.
SinglePhase Incompressible Fluid Flo in Mini and Microchannels 357
4. Channel Size Effect on Single Phase Incompressible Fluid Flow
in Mini and Microchannels
The effects of the channel size are of significance in fluid flow. These can be the effect of
experimental uncertainties, the effect of surface roughness and the like. Furthermore, the
effect of surface electrostatic charges may also be significant but beyond the scope of this
review. The accurate measurements become more difficult in mini and microchannels than
in macrochannels. For example, how can we accurately measure the very small flow rate in
mini and microchannels? For example, Celeta et al. [30] used 6 parallel stainless steel
microtubes in their experiments as shown by their test section in Figure 8. This test section
provided an easy method to measure the flow rate in the microtubes. However, it poses
another question: how could the flow rate be uniformly distributed in each submicrotube? In
fact, the accuracy of the flow rate measurements can greatly affect the pressure drop results.
Furthermore, for pressure drop measurements, how could we implement more accurate
measurements for the mini and microchannels? All these issues should be carefully
considered in the experimental facility and test section design [33].
Figure 8. Test section consisting of 6 parallel microtubes in the experimental study of Celeta et al.
[30].
Careful experimental operation should also be performed for experiments of fluid flow in
mini and microchannels. As already mentioned in section 3, the studies for the similar test
conditions with the same fluids in [19] and [20] showed quite different results. The same case
Lixin Cheng 358
is for the studies in [21] and [22]. These two examples have shown the importance of very
careful experimental operations in the study of fluid flow in mini and microchannels. To
perform accurate measurements, very careful test facility design and experimental operation
are necessary. Steinke et al. [33] developed an experimental facility for the investigation of
singlephase liquid flow in microchannels with a variety of geometries considering the
microscale effects. Furthermore, careful experimental uncertainties should be analyzed [13,
34].
The channel geometry evaluation is also very important because the channel dimensions
have a major effect on the friction factor calculation. Figure 9 shows the SEM image of the
rectangular microchannel geometry tested by Steinke and Kandlikar [34]. It can be seen how
a rectangle channel looks like in microscope, which indicates that the channel shape actually
becomes significance for microchannels. Figures 10 and 11 show the SEM images of the
circular microchannel geometries measured by Li et al. [32]. Obviously, a very little error in
the crosssection area measurement may cause very big errors in the flow rate calculation and
thus in the friction factor calculation.
Figure 9. Actual cross section of the tested rectangular microchannel by Steinke and Kandlikar [34]: a =
194 μm, b = 244 μm, L = 10 μm, θ= 85 degrees and d
h
= 227μm.
The surface roughness plays a very important role in the study of microscale fluid flow
[6, 3032, 3538]. For fluid flow in rough microchannels, a number of the available studies
considering the surface relative roughness effect have shown that the friction factors in
microchannels are higher than those in macrochannels, and the surface roughness also leads
to the earlier transition from laminar to turbulent. However, some studies have shown that the
SinglePhase Incompressible Fluid Flo in Mini and Microchannels 359
d
h
= 1570 μmd
h
= 624.4μmd
h
= 373 μm
Figure 10. The SEM images of crosssection and inner surface of three stainless steel test tubes in the
experimental study of Li et al. [32].
Cross section
[
Surface roughness
Figure 11. The SEM images of crosssection and inner surface of a fused silica test tube in the
experimental study of Li et al. [32].
relative roughness does not affect the friction factor. Thus, one issue is still open to discuss as
what is the relative roughness limitation below or beyond which the channel can be regarded
as smooth or rough. For example, to investigate the role of the surface relative roughness
effect on pressure drop characteristics in microtubes, Kandlikar et al. [35] conducted
experiments of water flowing in two microtubes with diameters of 1067 and 620 μm. For the
1067 μm diameter tube, the effects of the relative roughness from 0.178% to 0.3% on
Lixin Cheng 360
pressure drop are insignificant and the tube can be considered smooth. For the 620 μm
diameter tube with the same relative roughness values, the pressure drops are increased. In
this case, the tube having a relative roughness 0.3% cannot be considered as smooth.
Although several studies on the relative surface roughness effects are available [3538], no
general agreed conclusions have been obtained so far. Therefore, it is worthwhile to
understand under what condition the surface relative roughness effect can be ignored or
cannot be ignored for the fluid flow in mini and microchannels.
5. Conclusion
The conventional fluid dynamic theories developed for macrochannels are not always
applicable to fluid flow in mini and microchannels. The available studies on microscale
fluid flow in the literature reveal quite a few contradictory and there are still quite big
discrepancies among the experimental results by different researchers. Various causes for
such deviations are analyzed in this review. Furthermore, the effects of channel size on
microscale fluid flow are analyzed. The surface roughness is a very important factor but still
not well investigated. So far, no systematic agreed knowledge of microscale fluid flow has yet
been achieved. Therefore, efforts should be made to achieve complete theories on fluid flow
in mini and microchannels.
Nomenclature
A crosssectional area of flow channel, m
2
a length of rectangle, m
a
c
channel aspect ratio
b width of rectangle, m
C constant in Eq. (13)
d internal tube diameter, m
e thickness of channel base, m
f friction factor
l length, m
P
wetted
wetted perimeter, m
Re Reynolds number
r radical direction
V mean velocity, m/s
x xaxis
Greek symbols
Δp pressure drop, Pa
ε surface roughness, m
μ dynamic viscosity, Ns/m
2
SinglePhase Incompressible Fluid Flo in Mini and Microchannels 361
ρ density, kg/m
3
surface tension, N/m
Subscripts
crit critical
e entrance
h hydraulic
References
[1] S. G. Kandlikar, Fundamental issues related to flow boiling in minichannels and
microchannels, Exp. Therm. Fluid Sci., 26 (2002) 389407.
[2] L. Cheng, and D. Mewes, Review of twophase flow and flow boiling of mixtures in
small and mini channels, Int. J. Multiphase Flow, 32 (2006) 183207.
[3] P.A. Kew, and K. Cornwell, Correlations for the prediction of boiling heat transfer in
smalldiameter channels, Appl. Therm. Eng., 17 (1997) 705715.
[4] S.S. Mehendale, A.M. Jacobi, and R.K. Ahah, Fluid flow and heat transfer at micro and
mesoscales with application to heat exchanger design, ASME Appl. Mech. Rev., 53
(2000) 175193.
[5] L. Cheng, G. Ribatski, and J.R. Thome, Gasliquid twophase flow patterns and flow
pattern maps: fundamentals and applications, ASME Appl. Mech. Rev., 61 (2008)
050802105080228.
[6] S.G. Kandlikar, S. Garimella, D. Li, S. Colin, and R. King Michael, Heat Transfer and
Fluid Flow in Minichannels and Microchannels, Elsevier Science & Technology, UK,
2005.
[7] N.E. Moody, Friction factors for pipe flow, Trans. ASME, (1944) 671684.
[8] R.M. Olson, Essentials of Engineering Fluid Mechanics, 4
th
Ed., Harper & Row, New
York, 1980.
[9] C.F. Colebrook, and C.M. White, Experiments with fluid friction in roughened pipes,
Proc Roy Soc. (A), 161 (1937) 367.
[10] P.K. Swamee, and A.K. Jain, Explicit equations for pipeflow problems, J. Hydraulic
Div., 102 (1976) 657664.
[11] S.W. Churchill, Friction factor equation spans all fluid flow regimes, Chemical Eng., 7
(1977) 9192.
[12] R.K. Shah, and A.L. London, Laminar Flow Forced Convection in Ducts, Academic
Press Inc., New York, 1978.
[13] G.P. Celata, Heat Transfer and Fluid Flow in Microchannels, Begell house Inc, New
York, 2004.
[14] Z.Y. Guo and Z.X. Li, size effect on singlephhase channel flow and heat transfer at
microscale, Int. J. Heat Fluid Flow, 24 (2003) 284298.
[15] C.B. Sobhan, and S.V.A. Garimella, Comparative analysis of studies on heat transfer
and fluid flow in microchannels, Microscale Thermophysical Eng., 5 (2001), 293311.
Lixin Cheng 362
[16] G.L. Morini, Singlephase convective heat transfer in microchannels: a review of
experimental results, Int. J. Therm. Sci. 43 (2004) 631651.
[17] G.M. Mala, and D. Li,, Flow characteristics of water in microtubes, Int. J. Heat fluid
Flow, 20 (1999) 142148.
[18] D. Yu, R. Warrington, R. Barron, and T. Ameel, An experimental and theoretical
investigation of fluid flow and heat transfer in microtubes, ASME/JSME Thermal
Engineering Conference, Vol. 1, ASME, 1995.
[19] B. Xu, K.T. Ooi, N.T. Wong, C.Y. Li, and W.K. Choi, Liquid flow in microchannels,
Proceedings of the 5the ASME/JSME Joint Thermal Engineering Conference, San
Diego, California, March 1519, 1999.
[20] B. Xu, K.T. Ooi, N.T. Wong, and W.K. Choi, Experimental method to calculate the
friction factor for liquid flow in microchannels, Int. Comm. Heat Mass Transfer, 27
(2000) 11651176.
[21] J. Judy, D. Maynes, and B.W. Webb, Liquid flow pressure drop in microtubes,
Proceedings of the International Conference on Heat Transfer and Transport Phenomena
in Microscale, Banff, Canada, October 1520, 2000.
[22] J. Judy, D. Maynes, and B.W. Webb, Characterization of frictional pressure drop for
liquid flows through microchannels, Int. J. Heat Mass Transfer, 45 (2002) 34773489.
[23] P. Gao, S. L. Person, and M. FavreMarinet, Scale effects on hydrodynamics and heat
transfer in twodimensional mini and microchannels, Int. J. Therm. Sci., 41 (2002) 1017
1027.
[24] N. Caney, P. Marty, and J. Bigot, Friction losses and heat transfer of inglephase flow in
a minichannel, Appl. Therm. Eng., 27 (2007) 17151721.
[25] S.V. Garimella, and V. Singhal, Singlephase flow and heat transport and pumping
considerations in micro heat sinks, Heat Transfer Eng., 25(1) (2004) 1525.
[26] S.L. Qi, P. Zhang, R.Z. Wang, and L.X. Xu, Singlephase pressure drop and heat
transfer characteristics of turbulent liquid nitrogen flow in microtubes, Int. J. Heat Mass
Transfer, 50 (2007) 19932001.
[27] H.J. Lee, and S.Y. Lee, Heat transfer correlation for boiling flows in small rectangular
horizontal channels with low aspect ratios, Int. J. Multiphase Flow, 27 (2001) 2043
2062.
[28] G.R. Warrier, V.K. Dhir, and L.A. Momoda, Heat transfer and pressure drop in narrow
rectangular channels, Exp. Therm. Fluid Sci., 26 (2002) 5364.
[29] W. Qu, I. Mudawar, experimental and numerical study of pressure drop and heat transfer
in a singlephase microchannel heat sink, Int. J. Heat Mass Transfer, 45 (2002) 2549
2565.
[30] G.P. Celata, M. Cumo, M. Guglielmi, and G. Zummo, Experimental investigation of
hydraulics and single phase heat transfer in 0.130 mm capillary tube, Microscale
Thermophysical Eng., 6 (2002) 8597.
[31] W. Wibel, and P. Ehrhard, Experiments on the laminar/turbulent transition of liquid
flows in rectangular micochannels, Heat Transfer Eng., 30(12) (2009) 7077.
[32] Z. Li, Y.L. He, G.H. Tang and W.Q Tao, Experimental and numerical studies of liquid
flow and heat transfer in microtubes, Int. J. Heat Mass Transfer, 50 (2007) 34473460.
[33] M.E. Steinke, S.G. Kandlikar, J.H. Magerlein, E.G. Colgan, and A.D. Raisanen,
Development of an experimental facility for investigating singlephase liquid flow in
microchannels, Heat Transfer Eng., 27(4) 2006 4152.
SinglePhase Incompressible Fluid Flo in Mini and Microchannels 363
[34] M.E. Steinke, and S.G. Kandlikar, Singlephase liquid friction factors in microchannels,
Int. J. Therm. Sci., 45 (2006) 10731083.
[35] S.G. Kandlikar, D. Joshi, and S. Tian, Effect of channel roughness on heat transfer and
fluid flow characteristics at low Reynolds numbers in small diameter tubes, in:
Proceedings of 35th National Heat Transfer Conference, Anaheim, CA, USA, 2001,
Paper: 12134.
[36] D. Gloss and H. Herwig, Microchanel roughness effects: a close up view, Heat Transfer
Eng., 30(12) (2009) 6269.
[37] P. Young, T.P. Brackbill, and S.G. Kandlikar, Comparison of roughness parameters for
various microchannels surface in singlephase flow applications, Heat Transfer Eng.,
30(12) (2009) 78690.
[38] Z.X. Li, D.X. Du, and Z.Y. Guo, Experimental study on flow characteristics of liquid in
circular microtubes, Microscale Thermophysical Eng. 7, (2003) 253265.
In: Fluid Mechanics and Pipe Flow ISBN: 9781607410379
Editors: D. Matos and C. Valerio, pp. 365378 © 2009 Nova Science Publishers, Inc.
Chapter 11
EXPERIMENTAL STUDY OF PULSATING TURBULENT
FLOW THROUGH A DIVERGENT TUBE
Masaru Sumida
*
School of Engineering, Kinki University
1 Takaya Umenobe, Higashi−Hiroshima, 739−2116 JAPAN
Abstract
An experimental investigation was conducted of pulsating turbulent flow in a conically
divergent tube with a total divergence angle of 12°. The experiments were carried out under
the conditions of Womersley numbers of α =10∼40, mean Reynolds number of Re
ta
=20000
and oscillatory Reynolds number of Re
os
=10000 (the flow rate ratio of η = 0.5). Time
dependent wall static pressure and axial velocity were measured at several longitudinal
stations and the distributions were illustrated for representative phases within one cycle. The
rise between the pressures at the inlet and the exit of the divergent tube does not become too
large when the flow rate increases, it being moderately high in the decelerative phase. The
profiles of the phaseaveraged velocity and the turbulence intensity in the cross section are
very different from those for steady flow. Also, they show complex changes along the tube
axis in both the accelerative and decelerative phases.
Keywords: divergent tube, pulsating flow, turbulent flow, pressure distribution, velocity
distribution.
1. Introduction
Divergent tubes are an important pipeline component and are widely used as diffusers to
convert kinetic energy into pressure energy and as devices to connect two tubes of different
diameters in pipework systems and fluid machinery. Therefore, there have been a number of
studies, e.g., Singh and Azad [1, 2], Gan and Riffat [3], and Xu et al. [4], devoted to the flow
in divergent tubes to date. That is, much attention has been given to the tube geometry, which
*
Email address: sumida@hiro.kindai.ac.jp. Tel: +81−82−434−7000, Fax: +81−82−434−7011
Masaru Sumida 366
affects pressure recovery efficiency and losses; the relationship between performance and the
optimum geometry has been investigated. The flow in divergent tubes is highly unstable and
then, recently, computational fluid dynamics has been applied to the prediction of mean and
turbulent flow characteristics (Gan and Riffat [3], and Xu et al. [4]). However, all these
studies concerned steady flow.
On the other hand, unsteady flow has become important in connection with problems
concerning the starting and stopping or undesirable accidents of pumps and blowers, because
fluid machines are becoming better and fluid transport is becoming diversified and faster.
Nevertheless, there have been few studies on unsteady flow in divergent tubes. Mizuno and
Ohashi [5], and Mochizuki et al. [6] conducted experiments for a twodimensional diffuser.
The former group oscillated one plane, whereas the latter group used periodic flows of a wake
generated by a cylinder into the diffuser entrance. Thus their studies were aimed at grasping
the flow features, including unsteady separation, and/or establishing a method of controlling
the flow. Unfortunately, the periodically volumecycled, unsteady flow has never been
treated, to the author’s knowledge.
The purpose of the present study is to treat the problem of volumecycled, pulsating
turbulent flow through a conical divergent tube with a total divergence angle of 2θ
=12°.
First, the distribution of wall static pressure is measured for the various pulsation frequencies.
Subsequently, periodical changes of profiles, such as the phaseaveraged velocity and
turbulent intensity, are examined under a representative flow condition. Furthermore,
knowledge of their characteristics is obtained through comparison with those in the case of
steady flow.
2. Experimental Apparatus and Measurement Method
2.1. Experimental Apparatus
A schematic diagram of the experimental apparatus is shown in Figure 1. The working
fluid is air. The system consists of a pulsatingflow generator, a test tube and measuring
devices. Moreover, the pulsatingflow generator is composed of a steady flow and an
oscillating flow. The steady flow, i.e., timemean flow, was supplied, through a surge tank, by
a suction blower, which ensured that the flow rate was independent of the superimposed
oscillation and the length of the test tube. On the other hand, the volumecycled oscillating
flow was introduced by a piston, with a diameter of 300 mm and a stroke length adjustable
from 0 to 1000 mm, controlled by a personal computer.
The test tube had a total divergence angle 2θ of 12° and an area ratio m of 6.25. Here the
ratio m was (d
2
/d
1
)
2
, d
1
(=2a
1
=80 mm) and d
2
(=200 mm) being the diameters at the inlet and
the exit, respectively, of the divergent tube. These dimensions are shown in Figure 2, together
with the coordinate system. The divergent tube was constructed from 5 accurately machined,
transparent acrylic blocks connected via a slip ring. The ring, furthermore, had static pressure
holes, 0.8 mm in diameter and spaced 90° apart, and a small hole for inserting a hotwire
probe. Straight transparent glass tubes with lengths of 3700 mm (=46.3d
1
) and 4200 mm
(21d
2
) were connected to the inlet and the exit of the divergent tube, respectively.
Experimental Study of Pulsating Turbulent Flow through a Divergent Tube 367
Figure 1. Schematic diagram of experimental apparatus – 1) upstream tube, 2) divergent tube, 3)
downstream tube, 4) oscillating flow generator, 5) surge tank, 6) blower.
Figure 2. Coordinate system and dimensions of test tube.
2.2. Measurement Procedure and Data Acquisition
The wall static pressure was measured, using a diffusivetype semiconductor pressure
transducer (Toyoda MFG, DD1020.1F), at 11 stations between z = −22.1d
1
in the upstream
straight tube and z = 21.9d
1
in the downstream one, where z is the length measured along the
tube axis from the inlet of the divergent tube. Velocity measurements were performed with a
hotwire anemometer, and the velocity w in the axial direction was obtained at 8 stations
including sections of z/d
1
= 2 and 9.6.
The voltage output from the DC amplifier of the pressure transducer and from the
anemometer was sampled, with the personal computer, synchronously with a timemarker
signal that indicates the position of the piston. The data was processed as follows. For a
periodically unsteady turbulent flow, any instantaneous quantity, e.g., the axial velocity w in
the pulsation cycle, can be written as
w = W + w
t
. (1)
Masaru Sumida 368
Here, W is the phaseaveraged velocity and w
t
is the fluctuating one. The data at each
measuring point was collected, in equal time intervals, for 200 to 500 pulsation cycles. The
recorded data were ensemble phaseaveraged to obtain W at each phase. Moreover, the
turbulence intensity w’ was obtained as the square root of the ensemble phase average of w
t
2
.
This procedure was also applied to the phaseaveraged value P of the wall static pressure. It
was confirmed beforehand that averaging over 200 cycles has no influence on the values of
the quantities. The scatter in the results was less than 4 and 6% for the phaseaveraged values
and the turbulence intensity, respectively. In the hotwire measurement, it is difficult to
determine accurately the direction of flow at the position and time when the ratio of the radial
to the axial component of the velocity is rather large. For such a case, errors of a limited
extent are not avoidable. Furthermore, it was also checked that the flow properties are
symmetric with respect to the tube axis.
Additionally, in order to gain insight into the pulsating flow features, the visualization
experiment using water was executed for a divergent tube with d
1
=22 mm. In the experiment,
the flow in the horizontal plane including the tube was rendered visible by a solid tracer
method using polystyrene particles of about 0.2 mm diameter.
3. Results and Discussion
3.1. Experimental Conditions
Pulsating flow in divergent tubes is governed by five nondimensional parameters: the
total divergence angle 2θ, the area ratio m, the Womersley number α, the mean Reynolds
number Re
ta
and the oscillatory Reynolds number Re
os
(or the flow rate ratio η). Here, the
former two and the remainder are related to the geometry of the divergent tubes and the fluid
motion, respectively. The Womersley number is defined as α = a
1
(ω/ν)
1/2
, ω being the
angular frequency of pulsation and ν the kinematic viscosity of the fluid. The mean Reynolds
number is expressed as Re
ta
=d
1
W
a1,ta
/ν and the oscillatory Reynolds number as Re
os
=d
1
W
a1,os
/ν. Here, W
a1
is the crosssectional average velocity in the upstream tube and the
subscripts ta and os indicate the timemean and amplitude values, respectively. Finally, the
flow rate ratio is given by η =W
a1,os
/W
a1,ta
(=Re
os
/Re
ta
). The experiments were performed
systematically under the conditions of α =10~40, Re
ta
=20000 and Re
os
= 10000 (η = 0.5).
These conditions were chosen referring to works on pulsating straighttube flows (for
example, Iguch et al. [7]).
In order to check the flow entering the divergent tube, preliminary measurements were
conducted for the axial velocity at z/d
1
= 2 and the pressure gradient in the upstream tube. In
the former, integrating the measured velocities over the cross section, a timedependent flow
rate Q was calculated. The result is shown in Figure 3. In the figure, the solid line denotes the
first fundamental component developed by the Fourier series, in which the second and higher
components are less than 3% compared with the first one and are considerably small. This
confirms that the prescribed flow is realized, and also indicates that a sinusoidal flow rate is
achieved in the form
Q = Q
ta
+ Q
os
sinΘ , (2)
Experimental Study of Pulsating Turbulent Flow through a Divergent Tube 369
Θ (= ωt) being the phase angle and t the time.
Figure 3. Flow rate variation.
Furthermore, the results obtained for the varying pressure gradient in the upstream tube
were in good agreement with the previous ones obtained by Ohmi and Iguchi [8]. For
example, the phase difference Φp between the variation of the pressure gradient and the flow
rate is shown in Figure 4. Therefore, it is reconfirmed that a fully developed pulsating
turbulent flow enters the divergent tube.
Figure 4. Phase difference between axial pressure gradient in the upstream tube and flow rate variation.
Ohmi and Iguchi [8] showed that the pulsating turbulent flow in straight tubes is
classified roughly into three regimes depending on a characteristic number of α/Re
ta
3/8
. The
flow conditions taken up in this study are in the α/Re
ta
3/8
region of 0.2 to 1, which is the
intermediate regime in their classification. Thus we can say that the entry flow conditions into
the divergent tube in the present experiment are of higher interest in this research field.
Masaru Sumida 370
3.2. Wall Static Pressure
3.2.1. Distribution of Wall Static Pressure and Its Variation with Time
In divergent tubes, for steady flow, the kinetic energy of the flow converts into the
pressure energy and then the static pressure rises in the downstream direction. On the other
hand, for pulsating flow with a periodic change of the flow rate, the drop and rise of the
pressure in the longitudinal direction are needed to drive the fluid acceleratively and
deceleratively, respectively, with time. Hence the amplitude of the pressure variation is
predicted to be proportional to the pulsation frequency ω. Consequently, the pressure in the
divergent tube shows a complex distribution with time.
The wall static pressure P is shown in Figure 5 for the various Womersley numbers. The
pressure P is expressed in the form of the pressure coefficient C
p
, which is defined as C
p
= (P
P
ref
) / (ρ W
a1,ta
2
/2), P
ref
being the pressure at the station of z/d
1
= 2 in the upstream tube and
ρ the density of the fluid. In the figure, the broken lines indicate the timeaveraged values and
the chain line the result for steady flow at the same Reynolds number as Re
ta
. In the upstream
tube, C
p
changes in the phase leads of Φ
p
, shown in Figure 4, with flow rate variation. The
timeaveraged C
p
is slightly larger than that for the steady flow at Re=2000. On the other
hand, the pressure in the divergent tube rises with an increase of z/d
1
, except for the part of
the period for moderate and high α. That is, the variation of C
p
becomes larger with an
increase of α. Furthermore, the phase with the largest value of C
p
changes from the Θ ≈ 90°
(α=10) of the maximum flow rate to the Θ ≈ 150° (α=40) of the middle of the decelerative
phase. Conversely, the phase with the minimum value of C
p
shifts from the Θ ≈ 270° (α=10)
of the smallest flow rate to the Θ ≈ 330° (α=40) of the first half of the accelerative phase. At
these phases with the lowest C
p
, for the low α (α=10), there is little change of C
p
in the
longitudinal direction. However, for the moderate and high α (α=20, 40), C
p
takes negative
values, and the pressure in the divergent tube shows a favorable gradient.
(A)
Figure 5. Continued on next page.
Experimental Study of Pulsating Turbulent Flow through a Divergent Tube 371
(B)
(C)
Figure 5. Longitudinal distribution of wall static pressure. (○: Θ=0°, ●: Θ= 90°, Δ: Θ= 180°, ▲:
Θ=270°, ⎯  ⎯ : timeaveraged, ⎯ ⎯ ⎯ : steady flow at Re=20000). (a) α=10, (b) α=20, (c) α=40
We examine the pressure rise ΔP
L
in the length L, i.e., between the inlet and the exit of
the divergent tube. The representative result is shown in Figure 6, in which ΔP
L
is
nondimensionalized by the dynamic pressure based on W
a1,ta
in the upstream tube. In the
figure, the waveform of ΔP
L
is developed with the Fourier series denoted by the solid line.
The broken line denotes the result that is theoretically obtained using Bernoulli’s theorem for
a quasisteady flow. It is expressed as
ΔP
L
= (1 + η sinΘ)
2
(1− m
2
) . (3)
Moreover, the symbol ← indicates the pressure rise ΔP
L,s
for steady flow, at Re=20000,
with the same crosssectional average velocity as W
a1,ta
. The pressure rise ΔP
L
in the pulsating
flow changes almost sinusoidally. However, it lags behind the variation of the flow rate. The
phase difference Φ
ΔP
between the fundamental waveform of ΔP
L
and the pulsating flow rate
Q becomes large with an increase in the Womersley number. Incidentally, Φ
ΔP
changes from 
Masaru Sumida 372
5° to 60° when α increases from 10 to 40. On the other hand, ΔP
L
is lower than the
theoretical value for quasisteady flow. Furthermore, ΔP
L
takes approximately zero values for
phase in the range of 230 ~ 340° with a small flow rate.
Figure 6. Difference between pressures at the inlet and exit of a divergent tube (← : steady flow at
Re=20000).
As has been stated above, for pulsating flow in a divergent tube, the pressure at the exit
of the divergent tube rises when the crosssectional average velocity is large and is in a
decelerative phase. That is, ΔP
L
becomes large from the latter half of the accelerative phase to
the middle of the decelerative phase (Θ ≈ 50 ~ 180°), as seen in Figure 6. By contrast, C
p
exhibits a small change in the axial direction from the ending of the decelerative phase to the
first half of the accelerative phase (Θ ≈ 230 ~ 330°). This is because the kinetic energy to be
converted to pressure energy is low and because, to accelerate the fluid in the axial direction,
the downstream pressure must be lowed. Therefore, it can be understood that the pressure
distribution at the beginning of the accelerative phase shows the favorable larger gradient for
higher Womersley number where the fluid is strongly accelerated in the streamwise direction.
3.3. Axial Velocity
3.3.1. Changes in Centerline Velocity with Time and along the Tube Axis
In this section, we discuss the axial velocity. To start, we will elucidate the outline of the
flow features by focusing on the axial velocity on the tube axis. Figure 7 shows the
waveforms of instantaneous axial velocity w, together with those of its phaseaveraged one W
and its turbulence intensity w’. For comparison with those on the tube axis, the results of the
measurement at the radial position r/R
z
= 0.75 near the tube wall are also shown in Figure 7.
Moreover, the changes in W
c
and w
c
’ along the tube axis are illustrated in Figure 8. The
subscript c indicates the values on the tube axis and the broken lines indicate the time
averaged values.
Experimental Study of Pulsating Turbulent Flow through a Divergent Tube 373
Figure 7. Velocity waveforms. (A) Instantaneous velocity w (B)Phaseaveraged velocity W (C)
Turbulence intensity w’.
The phaseaveraged velocity W exhibits an almost sinusoidal change in time, as displayed
in Figure 7(b). However, since the flow extends with increasing z/d
1
in both the divergent
tube and the downstream straight tube, W
c
in these sections is reduced (see Figure 8(a)). In
particular, the degree of reduction in the divergent section becomes smaller and larger for the
middle, i.e., Θ = 180 and 0°, of the decelerative and the accelerative phase, respectively.
Thus, the phase angle at which W
c
shows a peak is discernibly delayed compared with the
flow rate variation as z/d
1
increases. This lag is equivalent to the time at which the fluid
flowing into the divergent tube at the maximum flow rate (Θ = 90°) reaches each section. As
a result, the time lag at the divergence exit (z/d
1
= 7.1) amounts to approximately onetenth of
one cycle. On the other hand, W near the wall (r/R
z
= 0.75) in the divergent tube changes
almost in synchronous phase with the flow rate. That is, W takes a maximum value at about Θ
= 90°. In the case of steady flow, W
c
/W
a1
at the Reynolds numbers of 10000 and 30000,
corresponding to the minimum and maximum flow rates in pulsating flow, respectively,
attenuates axially in the same manner as the flow at Re = 20000. Hence, the Reynolds number
has little effect on W
c
/W
a1
(the illustration is omitted here). It can be inferred from the above
discussion that the differences in the change in W
c
along the tube axis in phase are attributed
to an unsteady effect.
The turbulence intensity w’
c
becomes large, at any phase, with an increase in z/d
1
in the
first half of the divergent tube. It shows the peak magnitude in the second half of the
decelerative phase, as seen in Figures 7(c) and 8(b). Furthermore, w’
c
near the divergence exit
in the first half of the decelerative phase is about 2.6 times the value at the divergence inlet.
Consequently, the variation of w’
c
over one cycle becomes larger.
Masaru Sumida 374
(A)
(B)
Figure 8. Change in W
c
and w’
c
along the tube axis (○: Θ=0°, ●: Θ= 90°, Δ: Θ= 180°, ▲: Θ=270°, ⎯
⎯ ⎯ : timeaveraged). (a) Phaseaveraged velocity W
c
(b) Turbulence intensity w’
c.
On the other hand, w’ near the wall (r/R
z
=0.75) in the upstream tube is twice as large as
w’
c
, but the phase difference between w’ and W
a
is small. In the section immediately behind
the divergence inlet, the turbulence intensity is high in the phase with the high flow rate.
However, the change of w’ gradually becomes similar to that of waveform w’
c
with an
increase in z/d
1
.
3.3.2. Distributions of PhaseAveraged Velocity and Turbulence Intensity
Figure 9 shows distributions of W and w’ at four representative phases. The broken lines
in the figure denote results for steady flow at the Reynolds number of Re
ta
. Moreover,
sketches of the stream are displayed in Figure 10, which is based on, with moderate
Experimental Study of Pulsating Turbulent Flow through a Divergent Tube 375
exaggeration, the observation of the water flow. The painted parts enclosed by the broken
lines show the main current of the flow.
Figure 9. Distributions of W and w’ in upper and lower figures, respectively (○: Θ=0°, ●: Θ= 90°, Δ:
Θ= 180°, ▲: Θ=270°, ⎯ ⎯ ⎯ : steady flow at Re=20000).
In pulsating flow, the phase difference among fluid motions in the divergent tube
becomes larger as z/d
1
increases. Consequently, the difference between the flow states of the
accelerative and decelerative phases becomes considerable. Therefore, the phaseaveraged
velocity and turbulence intensity exhibit complicated distributions.
In the upstream tube (Figure 9; z/d
1
=−2), the periodic change of the axial velocity leads
slightly near the tube wall, whereas it is delayed in the central part of the cross section.
Nevertheless, on the whole, the velocity at each phase shows a profile similar to the steady
one with a simple 1/7thpower law. In the divergent tube, the pressure rise in the first half of
the accelerative phase is smaller than the theoretical one derived for quasisteady flow, as
shown in Figure 5(b). Moreover, there is little pressure difference in the divergent tube. As a
result, the main current in the central part of the cross section reaches the exit plane without
extending too much towards the tube wall (Figure 10(b); Θ ≈ 0°). Meanwhile, near the inlet, a
local pressure drop occurs, as seen in Figure 5(b). This accelerates the fluid in the
neighborhood of the wall, and the phase of the velocity variation is more advanced (z/d
1
=
1.4). When the flow rate increases and becomes maximum, eddies generated in the shear
layer surrounding the main current grow rapidly, accompanying the acceleration of the flow,
and become massive vortices. These vortices strongly interact with each other near the
divergence exit. Following this, the main stream suddenly begins to meander in the central
part of the cross section, resulting in a rapid increase in w’ (Θ = 90°; Figure 10(b)). By the
middle of the decelerative phase (Θ = 180°), the pressure at the divergence exit is, as before,
about twice as high as that in the case of steady flow at the same flow rate, because the
Masaru Sumida 376
pressure there decelerates the fluid flow (Figures 5(b) and 6). Thus the stream becomes
unstable. As a result, the massive vortices in the section downstream of the divergence exit
collapse and start to decay. In addition, the main current is mixed. At the end of the
decelerative phase (Θ = 270°), the turbulence intensity in the section downstream of the
divergence exit is considerably attenuated and part of the main current retreats to the first half
of the divergent tube (Figure 10(b)). The fluid particle in the vicinity of the wall is carried,
locally and momentarily, upstream along the wall. Nevertheless, flow separation and
backward flow are indistinct, and the fluid almost always flows downstream.
(A)
(B)
Figure 10. Sketches of flow pattern obtained by visualization experiment. (A) Steady flow at Re=20000
(B) Pulsating flow (α=20, Reta=20000, η=0.5).
Experimental Study of Pulsating Turbulent Flow through a Divergent Tube 377
From the above description, in the phase at which the flow rate increases, the degree of
acceleration is high in the central part of the cross section. As a result, W near the divergence
exit (z/d
1
= 5.5) at the maximum flow rate (Θ = 90°) shows a profile with convexity near the
tube axis, as seen in Figure 9. In the decelerative phase, on the other hand, the velocity is
reduced almost uniformly throughout the cross section because of the stronger action of
turbulent mixing in the section further downstream.
The distribution of the turbulence intensity differs appreciably from that in a steady flow,
as seen in Figure 9. That is, the turbulence intensity is high in the phase from the end of the
increase to the first half of the decrease of the flow rate. This is because, during this phase,
the shearing layer between the main current and the tube wall rolls up and becomes massive
vortices. On the other hand, w’ is low in the first half of the accelerative phase when these
vortices are almost decayed. To put it concretely, when the flow rate is large, the turbulence
intensity immediately behind the divergence inlet (z/d
1
= 0 ∼1.4) takes large values near the
wall. Downstream, the region with high w’ extends radially, accompanied by the formation of
a shearing layer. Furthermore, the maximum w’/W
a1,ta
distribution in this section occurs in the
first half of the decelerative phase and becomes about 18% at z/d
1
=2.7 from the 14% in the
upstream straight tube. Downstream, in the middle section of the divergent tube, at z/d
1
= 2.7
~ 5.5, w’ becomes maximum in the vicinity of the inflection point of the W distribution when
a fluid flowing at high speed is passed through the section. Subsequently, the location of
maximum w’ shifts toward the tube axis. In this phase (Θ ≈ 180°), w’ is about 20 percent
higher than that of steady flow with Re
ta
. The maximum value corresponds to 20% of W
a1,ta
.
In the section downstream of the divergence exit, the distribution of the turbulence intensity is
uniformalized in the cross section (z/d
1
= 9.6).
5. Conclusion
Experimental investigations were performed for pulsating turbulent flow in a divergent
tube and the flow field was examined. The principal findings of this study are summarized as
follows.
(1) The axial distributions of the pressure coefficient C
p
in the divergent tube are high in
the phase from the latter half of acceleration to the middle of deceleration. In
contrast, they are low in other phases, namely, from the ending of the deceleration to
the first half of the acceleration.
(2) The timemean value of the pressure rise ΔP
L
between the inlet and the exit of the
divergent tube is larger than that in the steady flow with the Re = Re
ta
. The variation
of ΔP
L
and the phase lag Φ
ΔP
behind the flow rate depend on and increase with α
2
.
(3) The phaseaveraged velocity W shows profile swelling in the central part of the cross
section when the flow rate increases. The phase at which the value becomes
maximum in each plane is delayed from Θ = 90° at the maximum flow rate with an
increase in z/d
1
. However, as the flow rate decreases, the W profile becomes flat
downstream of the divergent tube.
(4) The turbulence intensity w’ for a high flow rate is maximum at the position on the
radius near the inflection point of the W profile. As the phase proceeds, the region
Masaru Sumida 378
with a large w’ extends towards the wall and the tube axis, and the w’ profile
becomes level at the plane of the divergence exit.
Acknowledgements
The author would like to thank Mr. J. Morita of Tokuyama Corporation for his assistance.
References
[1] R. K. Singh and R. S. Azad, Exp. Thermal Fluid Sci. 1995, vol. 10, 397413.
[2] R. K. Singh and R. S. Azad, Exp. Thermal Fluid Sci. 1995, vol. 11, 190203.
[3] G. Gan and S. B. Riffat, Appl. Energy 1996, vol. 542, 181195.
[4] D. Xu, M. A. Leschziner, B. C. Khoo and C. Shu, Comput. Fluids 1997, vol. 264,417
423.
[5] Mizuno and H. Ohashi, Trans. Jpn. Soc. Mech. Eng. Ser. B 1984, vol. 50453, 1223
1230 (in Japanese).
[6] O. Mochizuki, M. Kiya, Y. Shima and T. Saito, Trans. Jpn. Soc. Mech. Eng. Ser. B
1997, vol. 63605, 5461 (in Japanese).
[7] M. Iguchi, M. Ohmi and S. Tanaka, Bull. JSME 1985, vol. 28246, 29152922.
[8] M. Ohmi and M. Iguchi, Bull. JSME 1980, vol. 23186, 20292036.
In: Fluid Mechanics and Pipe Flow
Editors: D. Matos and C. Valerio, pp. 379397
ISBN 9781607410379
c _2009 Nova Science Publishers, Inc.
Chapter 12
SOLUTION OF AN AIRFOIL DESIGN INVERSE
PROBLEM FOR A VISCOUS FLOW USING A
CONTRACTIVE OPERATOR
Jan Šimák and Jaroslav Pelant
Aeronautical Research and Test Institute (VZLÚ), Prague, Czech Republic
Abstract
This chapter deals with a numerical method for a solution of an airfoil design in
verse problem. The presented method is intended for a design of an airfoil based on
a prescribed pressure distribution along a mean camber line, especially for modifying
existing airfoils. The main idea of this method is a coupling of a direct and approxi
mate inverse operator. The goal is to ﬁnd a pseudodistribution corresponding to the
desired airfoil with respect to the approximate inversion. This is done in an iterative
way. The direct operator represents a solution of a ﬂow around an airfoil, described
by a system of the NavierStokes equations in the case of a laminar ﬂow and by the
k−ω model in the case of a turbulent ﬂow. There is a relative freedom of choosing the
model describing the ﬂow. The system of PDEs is solved by an implicit ﬁnite volume
method. The approximate inverse operator is based on a thin airfoil theory for a poten
tial ﬂow, equipped with some corrections according to the model used. The airfoil is
constructed using a mean camber line and a thickness function. The so far developed
method has several restrictions. It is applicable to a subsonic pressure distribution sat
isfying a certain condition for the position of a stagnation point. Numerical results are
presented.
1. Introduction
The method described in this chapter is assumed for an airfoil design corresponding to
a given pressure distribution. It is an extension of a method for a potential ﬂow and later for
an inviscid compressible ﬂow [1], [2] and a laminar viscous ﬂow [3]. It is useful in the cases
where a speciﬁc distribution is desired. The method is based on the use of an approximate
inversion and the solution of the ﬂow around an airfoil. Since the ﬂow is assumed turbulent,
a model of turbulence is used to improve the quality of the solution. In the following text
the detailed description of the method is given.
380 Jan Šimák and Jaroslav Pelant
2. Description of the Method
The presented method, as most of the others inverse methods, is based on the idea of
an approximate inversion. This approximate inverse operator, in the following text denoted
by L, is accompanied by a socalled direct operator, denoted by P. This direct operator
represents a solution of the ﬂow around and airfoil. The pressure distributions f
up
and f
lo
on the upper and lower part of the airfoil are prescribed along the mean camber line above
the chord line, in the direction from the leading edge to the trailing edge. The chord line is
of length b. A new function f on the interval ¸−b, b) is deﬁned by the following relations
f(x) = f
up
(−x), for x ∈ ¸−b, 0),
f(x) = f
lo
(x), for x ∈ ¸0, b).
Then the inverse problem can be deﬁned as the following:
Find a function u : ¸−b, b) →R such that
PL(u) = f. (2.1)
The function u will be referred to as a pseudodistribution.
The equation (2.1) is transformed to a contractive operator whose ﬁxed point is searched
for. This yields a sequence of pseudodistributions
¦u
k
¦
∞
k=0
, u
k+1
= u
k
+ α(f −PLu
k
) . (2.2)
If this sequence converges, the limit is the solution of (2.1). In order to ensure that, a
suitable parameter α ∈ (0, 1) is chosen. The numerical results indicate that the values less
then one are usually sufﬁcient. In many examples the value was set to α = 0.6.
3. Inverse Operator
The approximate inverse operator L, described in this section, is a mapping between a
velocity distribution f and a curve ψ representing an airfoil. Since the given distribution is
a pressure, it is necessary to transform it to a velocity distribution. The pressure is assumed
to be constant across the boundary layer in the normal direction to the airfoil. From that
reason, the velocity on the edge of a boundary layer is computed from the pressure on the
airfoil. The transformation is derived from relations for the pressure, density, Mach number
and speed of sound,
p = p
0
_
1 +
γ −1
2
M
2
_
−γ/(γ−1)
,
ρ = ρ
0
_
1 +
γ −1
2
M
2
_
−1/(γ−1)
,
c
2
= γp/ρ.
The derived formula for the velocity related to the velocity in the free stream is
_
v(x)
v
∞
_
2
=
2/M
2
∞
+ γ −1
γ −1
(p
0
/p(x))
(γ−1)/γ
−1
(p
0
/p(x))
(γ−1)/γ
, (3.3)
Solution of an Airfoil Design Inverse Problem... 381
where p
0
is the pressure at zero velocity, M
∞
is the Mach number in the free stream and γ
is the Poisson adiabatic constant.
The velocity distribution, obtained by the above mentioned transformation (or is simply
given), is on an upper and lower side of the airfoil above its chord. This is a straight line
connecting the leading and trailing edge of the airfoil. The derivation of the operator will
be in two steps. At ﬁrst, a function describing a mean camber line is derived. This is a line
in the middle of an airfoil, its end points are the same as the end points of the chord. The
second step is construction of a thickness function. It is the distance between points on a
surface and points on a mean camber line. Putting the function describing the mean camber
line and the function describing the thickness together we get the operator L.
The prescribed distribution has to satisfy a condition, that the stagnation point on the
leading edge is located at the beginning of the chord line. Otherwise the method can have
troubles near the stagnation point. The required position is ensured by the determination of
the appropriate angle of attack.
3.1. Construction of the Mean Camber Line
The derivation of the function describing the mean camber line is based on the theory
of thin airfoils. In this theory, the airfoil thickness is neglected and the airfoil shape is
simpliﬁed to a line. From this reason we assume that the mean camber line represents an
airfoil in the following text.
At ﬁrst, the origin of the coordinate system is set at the beginning of the chord line on
the leading edge. The xaxis is set so that the chord line, which length is b, lies on the axis.
Let us consider the direction of the ﬂow v
∞
parallel to the xaxis or with a small angle of
attack α
∞
. The main idea is to consider a system of vortices on the airfoil surface. This
system determines a circulation along the airfoil.
A vortex with an intensity Γ generates a velocity according to the relation
2vπr = Γ,
where v is the size of a velocity at some point of the plane and its direction is perpendicular
to a line connecting this point and the point of the vortex. The symbol r denotes the distance
of these two points.
According to this, if a system of vortices is assumed on the interval ¸0, b) on the xaxis,
then there will be a velocity generated by this system in the direction of the yaxis at the
point x ∈ ¸0, b), given by the formula
ˆ v
y
(x) =
1
2π
(PV)
_
b
0
dΓ(ξ)
ξ −x
. (3.4)
The symbol (PV) means a principal value of an integral, deﬁned as a limit
(PV)
_
b
a
f(x) dx = lim
ǫ→0
__
c−ǫ
a
f(x) dx +
_
b
c+ǫ
f(x) dx
_
,
where c is a point at which the function f(x) has a singularity.
382 Jan Šimák and Jaroslav Pelant
Let us denote the coordinates of the mean camber line by [x, s(x)], x ∈ ¸0, b). Since the
both end points of the mean camber line lie on the xaxis, the condition s(0) = s(b) = 0
must be satisﬁed. Thus, although the formula (3.4) is valid only for points on the xaxis, it
can be also used for points on the mean camber line causing only a small error, especially
if the camber is small.
The vortices on the chord line causing the circulation are unknown but they are deter
mined by the given velocity distribution. The relation between them is quite easy to derive.
Assume a closed curve whose image lies over the mean camber line (Figure 1). The cir
v
up
v
lo
A
A
′
B
B
′
dξ
y
Figure 1. Derivation of the circulation along the mean camber line.
culation along a closed curve is deﬁned as a curvilinear integral of the tangent component
of the velocity along this curve. The ﬂow on the upper and lower side has the velocities
equal to v
up
, v
lo
, the velocities across the mean camber line are zero because it is assumed
impermeable. So it holds for the elementary circulation
dΓ(ξ) =
_
v
up
(ξ) −v
lo
(ξ)
_
dξ,
where dξ is the length of the segments AB and A
′
B
′
.
Next, the velocity of the free stream is denoted by v
∞
, its components are
v
x
∞
= v
∞
cos α
∞
, v
y
∞
= v
∞
sin α
∞
,
where α
∞
is the angle of attack. By v(x) =
_
v
x
(x), v
y
(x)
_
is denoted the velocity on the
mean camber line, which is the free stream velocity disturbed by the vortices. Let α be the
angle between the vector v(x) and the xaxis. Then the following must be true:
tan α =
v
y
(x)
v
x
(x)
= tanα
∞
+
ˆ v
y
(x)
v
x
∞
. (3.5)
Since the mean camber line represents an airfoil, the vector of velocity v(x) must be
tangential to this line. This yields the relation
tan α = s
′
. (3.6)
Introducing a speciﬁc circulation
γ(ξ) =
dΓ(ξ)
dξ
and using relations (3.4) and (3.6) in (3.5), the differential equation for the mean camber
line is obtained,
s
′
(x) = tanα
∞
+
1
2πv
x
∞
(PV)
_
b
0
γ(ξ)
ξ −x
dξ. x ∈ ¸0, b) , (3.7)
Solution of an Airfoil Design Inverse Problem... 383
The equation is completed with boundary conditions s(0) = s(b) = 0.
In the following let us assume the velocity v
∞
= 1 (this can be achieved using a suitable
normalization). It is also assumed that v
x
∞
˙ =v
∞
= 1, that means v
y
∞
is small compared to
v
x
∞
. The equation (3.7) can be rewritten as
s
′
(x) = a −c (PV)
_
b
0
γ(ξ)
x −ξ
dξ,
where c = 1/(2π) and a = tan α
∞
. The angle α
∞
is left as a parameter in order to satisfy
the boundary conditions. The solution to the differential equation is expressed as
s(x) = ax −c (PV)
_
x
0
_
(PV)
_
b
0
γ(ξ)
t −ξ
dξ
_
dt,
using the boundary condition for x = 0.
Changing the order of integration and setting the parameter a to ensure the boundary
condition for x = b is satisﬁed, the ﬁnal results are obtained. The function s(x) and its
derivative s
′
(x) are given by
s
′
(x) =
1
2πb
(PV)
_
b
0
γ(ξ) ln
¸
¸
¸
¸
b −ξ
ξ
¸
¸
¸
¸
dξ −
1
2π
(PV)
_
b
0
γ(ξ)
x −ξ
dξ, (3.8)
s(x) =
x
2πb
(PV)
_
b
0
γ(ξ) ln
¸
¸
¸
¸
b −ξ
ξ
¸
¸
¸
¸
dξ −
1
2π
(PV)
_
b
0
γ(ξ) ln
¸
¸
¸
¸
x −ξ
ξ
¸
¸
¸
¸
dξ, (3.9)
where γ(ξ) = v
h
(ξ) −v
d
(ξ).
3.2. Construction of the Thickness Function
During the derivation of the mean camber line, an airfoil with zero thickness was as
sumed. By contrast, in this part of the text an airfoil with zero camber is assumed. In the
previous part, a system of vortices was assumed in order to derive the mean camber line
function s(x). This time, a system of sources is assumed on the chord in order to derive the
thickness function. This system generates a velocity potential ϕ(x, y) at some point of the
xyplane. The potential is given by the relation
ϕ(x, y) =
1
2π
_
b
0
q(ξ) ln
_
(x −ξ)
2
+ y
2
dξ, (3.10)
where q(ξ) describes the intensity of the sources.
As will be shown in the following text, it is possible to derive a relation between q(x)
and the airfoil thickness t(x). From this reason a ﬂow parallel to the xaxis is considered.
According to the symmetry of the airfoil it is sufﬁcient to work only with the upper half
of the airfoil. By ˆ v = (ˆ v
x
, ˆ v
y
) is denoted a velocity vector produced by the distribution of
sources on the chord. The relation between the thickness and the source intensity is based
on the ﬂow rate along the chord (see Figure 2). The difference between velocities at points
x and x+dx is dˆ v
x
and the difference between the thickness at these points is dt. Thus it is
necessary (since the ﬂow is considered incompressible) to add or remove some mass using
sources in order to achieve balance.
384 Jan Šimák and Jaroslav Pelant
0
b
v
∞
+ ˆ v
x
v
∞
+ ˆ v
x
+ dˆ v
x
t
t + dt
x
dx
x + dx
Figure 2. Illustration for derivation of the thickness function.
Mathematically formulated, the balance is
(v
∞
+ ˆ v
x
) t +
1
2
q dx = (v
∞
+ ˆ v
x
+ dˆ v
x
) (t + dt) .
After some rearrangement and omitting the term dˆ v
x
dt, the relation is
1
2
q(x) =
d
dx
_
v
∞
+ ˆ v
x
(x)
_
t(x).
Under the assumption the airfoil is thin enough, it is true [ˆ v
x
[ ≪ v
∞
inside the interval
(0, b). Hence, the relation can be simpliﬁed by neglecting the term ˆ v
x
and the result can be
written as
q(x) = 2v
∞
t
′
(x). (3.11)
Differentiation of the equation (3.10) yields
ˆ v
x
(x) =
∂ϕ(x, 0)
∂x
=
1
2π
(PV)
_
b
0
q(ξ)
1
x −ξ
dξ
and putting here for q(ξ) results in
ˆ v
x
(x) =
v
∞
π
(PV)
_
b
0
t
′
(ξ)
dξ
x −ξ
. (3.12)
The velocity on the chord is assumed as an average of velocities on the upper and lower side
of the airfoil. Thus the velocity increment ˆ v
x
can be expressed as ˆ v
x
= (v
h
+ v
d
)/2 −v
∞
.
Assuming again the free stream velocity v
∞
= 1, a new function v
p
(x) can be deﬁned,
v
p
(x) =
v
h
(x) + v
d
(x)
2
−1. (3.13)
Substituting this into (3.12) yields
v
p
(x) =
1
π
(PV)
_
b
0
t
′
(ξ)
dξ
x −ξ
. (3.14)
The unknown in this equation is the derivative of the thickness t
′
(x) and it is possible to
ﬁnd an analytical solution. This integral equation is a special case of a more general integral
equation of the form
αϕ(x) +
β
πi
_
L
ϕ(ξ)
ξ −x
dξ = f(x), x ∈ R,
Solution of an Airfoil Design Inverse Problem... 385
where
_
L
denotes a complexvalued curvilinear integral along a smooth curve L. In our
case α = 0, β = −πi and the curve L is an identity on the interval ¸0, b). According to [4],
the solution of our problem can be written in the form
t
′
(x) = −
1
π
(PV)
_
b
0
v
p
(ξ)
x −ξ
¸
ξ(b −ξ)
x(b −x)
dξ +
C
_
x(b −x)
.
After integration with respect to x, the function t(x) is determined by
t(x) =−
1
π
(PV)
_
b
0
_
v
p
(ξ)
_
ξ(b −ξ) (PV)
_
x
0
dy
_
y(b −y)(y −ξ)
_
dξ+
+ C arctan
_
x −b/2
_
x(b −x)
_
−c,
where c and C are two constants determined later. The second integral can be written in the
form
(PV)
_
x
0
dy
_
y(b −y)(y −ξ)
= −
1
_
ξ(b −ξ)
ln
¸
¸
¸
¸
¸
¸
¸
1 +
_
ξ
b−ξ
_
b−x
x
1 −
_
ξ
b−ξ
_
b−x
x
¸
¸
¸
¸
¸
¸
¸
.
Finally we obtain the equation describing the airfoil thickness
t(x) =
1
π
(PV)
_
b
0
v
p
(ξ) ln
¸
¸
¸
¸
¸
¸
¸
1 +
_
ξ
b−ξ
_
b−x
x
1 −
_
ξ
b−ξ
_
b−x
x
¸
¸
¸
¸
¸
¸
¸
dξ + C arctan
_
x −b/2
_
x(b −x)
_
+ c.
(3.15)
This equation contains two parameters C and c. If we set x = 0 or x = b, the above men
tioned integral equals zero. The second term containing the parameter C is a monotonous
function which equals −Cπ/2 in the case of x = 0 or is equal to Cπ/2 in the case of
x = b. Since we are interested in an airfoil which is closed on both sides of the chord, we
choose C = c = 0. This choice is fully natural because it agrees with the situation where
the function v
p
is set to zero. In this case the velocity on the airfoil equals the free stream
velocity and from this reason the corresponding airfoil has zero thickness. Under this con
dition the mapping between the velocity v
p
(x) and the thickness function t(x) is unique. It
is also obvious that we can obtain an airfoil with prescribed thickness of the trailing edge
by a suitable choice of C and c.
The resulting equation is
t(x) =
1
π
(PV)
_
b
0
v
p
(ξ) ln
¸
¸
¸
¸
¸
¸
¸
1 +
_
ξ
b−ξ
_
b−x
x
1 −
_
ξ
b−ξ
_
b−x
x
¸
¸
¸
¸
¸
¸
¸
dξ. (3.16)
At the end, there is necessary to mention one drawback. The airfoil thickness should be
positive, of course. But the procedure mentioned above should result in a negative thick
ness, depending on the input velocity distribution. Since the input is the socalled pseudo
distribution, it is necessary to include some control mechanism into the implementation.
386 Jan Šimák and Jaroslav Pelant
3.3. Construction of the Airfoil
(−s
′
(x), 1)
(1, s
′
(x))
s(x)
x
t(x)
t(x)
x
y
0 b
Figure 3. Construction of the airfoil.
Construction of the airfoil is based on the mean camber line s(x) and the thickness
function t(x). Both functions are dependent on the velocity distributions f = ¦v
up
, v
lo
¦.
The airfoil is expressed as ψ(x) =
_
ψ
1
(x), ψ
2
(x)
_
, where x ∈ ¸0, b). Since the distance
between points on the surface and on the mean camber line is given by t(x) (Figure 3), the
coordinates of the airfoil are expressed in the form
ψ
1
(x) = x ∓t(x)
s
′
(x)
_
1 + s
′
2
(x)
, (3.17)
ψ
2
(x) = s(x) ±t(x)
1
_
1 + s
′
2
(x)
. (3.18)
The upper sign is meant for the upper part of the airfoil and the bottom sign for the lower
part. It is easy to see that the following is true,
ψ
1
(0) = 0, ψ
1
(b) = b,
ψ
2
(0) = 0, ψ
2
(b) = 0.
Thus we get a closed airfoil over the chord ¸0, b). The derived approximate inverse operator
is the mapping L(f) = L(v
up
, v
lo
) = (ψ
1
, ψ
2
).
According to the formulation of the inversion, it is necessary to deal with continuous
functions. In order to have an airfoil with an acceptable geometry, the functions s(x) and
t(x) must be continuous and smooth. This is achieved by assuming a subsonic ﬂow with
continuous pseudodistributions u
k
for all k. Thus the distributions obtained in each iter
ation as a solution of the ﬂow problem must be continuous. From the numerical point of
view, this restriction is not so strong. The prescribed distribution must be continuous but
the intermediate distributions can have discontinuities with small jumps which are smeared
by the iterative process and by the integral formulation of the functions s(x) and t(x).
Solution of an Airfoil Design Inverse Problem... 387
3.4. Numerical Realization
The evaluation of the functions ψ
1
(x), ψ
2
(x) includes evaluations of integrals, which
are done by a suitable numerical quadrature. The integrands are functions in two variables x
and y and have singularities for x = y. From this reason it is necessary to avoid this points.
The quadrature used in this method is the ChebyschevGauss quadrature. The integrals are
discretized in the way of
_
b
0
f(x, y) dy =
_
b
0
f(x, y)
_
y(b −y)
_
y(b −y)
dy =
N/2
k=1
w
k
f(x, y
k
)
_
y
k
(b −y
k
)+R
N
, (3.19)
where x ∈ ¸0, b), N is an even number, w
k
= 2π/N are quadrature coefﬁcients,
R
N
=
2π
2
N
N!
∂
N
_
f(x, η)
_
η(b −η)
_
∂y
N
is an error of the quadrature and y
k
= (b + bˆ x
k
)/2 are nodes. In this case ˆ x
k
= cos
_
(2k −
1)π/N
_
are roots of the Chebyschev polynomial
T
N/2
(x) = cos
_
N
2
arccos x
_
.
Finally, a sequence of points x
i
is deﬁned,
x
i
=
b
2
_
1 + cos
iπ
N
_
, i = 0, 1, . . . , N.
In the formula (3.19) is set
x = x
i
, i = 0, 2, 4, . . . , N
and the desired quadrature formula
_
b
0
f(x
i
, y) dy ≈
2π
N
N
2
k=1
f(x
i
, x
2k−1
)
_
x
2k−1
(b −x
2k−1
), i = 0, 2, 4 . . . , N (3.20)
is obtained.
The distribution of nodes using the Chebyschev polynomial has a favourable property.
The density of nodes is higher near the ends of the chord. From this reason the shape of
the airfoil is expressed more precisely. The velocity is needed to be known at points x
i
,
i = 1, 3, . . . , N − 1 and the resulting airfoil is evaluated at points x
i
, i = 0, 2, 4, . . . , N.
The number of evaluated points along one side is N/2 +1 and the total number of different
points of the airfoil is 2(N/2 + 1) −2 = N.
3.5. Viscous Correction
The approximate inversion described so far can deal with a pressure distribution regard
less of viscosity. However, the inverse operator is based on the fact that an airfoil can be
388 Jan Šimák and Jaroslav Pelant
represented by streamlines. In the case of an inviscid ﬂow, streamlines are attached to the
upper and lower surface and so the airfoil contours can be replaced by them. But this is
not true in the case of a viscous ﬂow. Streamlines are inﬂuenced by a boundary layer and
thus the obtained shape is different from the physical shape. The airfoil is thicker and the
camber is smaller. Although the inverse operator has some capability to overcome this fact,
in the cases with strong inﬂuence of the boundary layer it is not enough. From this reason
it is useful to introduce a correction to eliminate this problem. This correction is based on
the displacement thickness corresponding to the pressure and the airfoil geometry.
Using the mentioned inverse operator, the mean camber line function and the thickness
function are computed. Now the correction is applied. First, the thickness is reduced by
the displacement thickness. Second, the mean camber line is corrected. Since the geometry
of the airfoil is closed and the effective shape is open, it is not possible to simply subtract
the displacement thickness form the thickness function. The remedy is to start from the
deﬁnition of the displacement thickness and derive the correction.
The displacement thickness is deﬁned as the following integral,
δ
∗
=
_
δ
0
_
1 −
ρu
ρ
e
u
e
_
dy, (3.21)
where the subscript e denotes the values on the edge of the boundary layer, u is the velocity
across the boundary layer and δ is the boundary layer thickness. Since u/u
e
≈ 1 for
y ≥ δ (in this case u
e
means a velocity outside the boundary layer) and the inverse operator
assumes an incompressible ﬂow, the formula can be rewritten
δ
∗
≈ d −
_
d
0
u
u
e
dy, for d ≥ δ.
Denoting f
′
x
(y) = u
x
(y)/u
e
(x) (the subscript x denotes the dependency on the location
on the airfoil), the relation for the corrected airfoil thickness is
t
vis
(x) = t(x) −δ
∗
(x) = t(x) −d + f
x
(d), for d ≥ δ.
That means t
vis
= f
x
(t) for t ≥ δ. Assuming the velocity proﬁle u nonnegative, we can
extend this relation for all t nonnegative. Thus the corrected airfoil thickness is determined
by
t
vis
= f
x
(t) (3.22)
providing that the velocity proﬁle on the airfoil is known.
The velocity proﬁle is obtained by the Pohlhausen’s method [5], which was derived for
laminar boundary layers. The approximation of the velocity by the polynomial of fourth
order is assumed
u
x
(y)
u
e
(x)
= 2
y
δ(x)
−2
_
y
δ(x)
_
3
+
_
y
δ(x)
_
4
+
du
e
(x)
dx
δ(x)
6ν
y
_
1 −
y
δ(x)
_
3
. (3.23)
The velocity described by this polynomial satisﬁes the boundary conditions
u
x
(0) = 0, u
x
(δ) = u
e
(x),
∂
2
u
x
(0)
∂y
2
= −
u
e
(x)
ν
du
e
(x)
dx
,
∂u
x
(δ)
∂y
= 0,
∂
2
u
x
(δ)
∂y
2
= 0. (3.24)
Solution of an Airfoil Design Inverse Problem... 389
This representation makes sense for a parameter Λ = (δ
2
/ν)(du
e
/dx) from an interval
¸−12, 12). If Λ > 12, the velocity proﬁle overshoots the value of the velocity u
e
(x). If
Λ < −12, a separation occurs and the velocity proﬁle is not nonnegative.
The boundary layer thickness δ is obtained from the von Kármán’s momentum equation
and relations for the boundary layer parameters:
τ
wall
ρ
=
du
e
dx
(2θ + δ
∗
) u
e
+ u
2
e
dθ
dx
, (3.25)
δ
∗
= δ
_
3
10
−
1
120
du
e
dx
δ
2
ν
_
, (3.26)
θ = δ
_
37
315
−
1
945
du
e
dx
δ
2
ν
−
1
9072
_
du
e
dx
δ
2
ν
_
2
_
, (3.27)
τ
wall
=
µu
e
ρ
_
2 +
1
6
du
e
dx
δ
2
ν
_
. (3.28)
The new thickness t
vis
is computed using the formula (3.22). After that, the mean
camber line is modiﬁed to eliminate viscous effects. To do so, a correction function
∆(x) =
δ
∗
up
−δ
∗
lo
2
is subtracted from the mean camber line function s(x). The function s(x) is then trans
formed to satisfy s(0) = s(b) = 0.
3.6. Finding the Mean Camber Line of a General Airfoil
As was mentioned earlier, the pressure distribution is given along a mean camber line.
The inverse operator evaluates the function describing this line, so the mean camber line of
the designed airfoil is known. But if the given pressure distribution is based on a known
airfoil, it is useful to have a way how to get its mean camber line.
Assume the airfoil coordinates are known. The process of the airfoil construction
(3.17)–(3.18) can be rewritten as
ψ
up
_
x −t(x)
s
′
(x)
_
1 + s
′2
(x)
_
= s(x) + t(x)
1
_
1 + s
′2
(x)
,
ψ
lo
_
x + t(x)
s
′
(x)
_
1 + s
′2
(x)
_
= s(x) −t(x)
1
_
1 + s
′2
(x)
, x ∈ ¸0, b) (3.29)
under the assumptions s(x) ∈ C
1
¸0, b), t(x) ∈ C ¸0, b). The symbols ψ
up
(˜ x), ψ
lo
(˜ x)
denote the ycoordinates of the upper and lower part of the airfoil (corresponding to x
coordinate ˜ x). Linearization of the left hand sides and rearrangement leads to a differential
equation for an unknown function s(x),
s
′
(x)
_
ψ
up
(x)ψ
′
lo
(x) + ψ
′
up
(x)ψ
lo
(x) −s(x)
_
ψ
′
up
(x) + ψ
′
lo
(x)
_
_
= 2s(x) −ψ
up
(x) −ψ
lo
(x), x ∈ (0, b) . (3.30)
390 Jan Šimák and Jaroslav Pelant
This equation is equipped with boundary conditions
s(0) = 0, s(b) = 0.
The equation was derived under the assumptions that s
′
(x)ψ
′
up
(x) ,= −1 and
s
′
(x)ψ
′
lo
(x) ,= −1. In other words, the tangential vectors of the airfoil shape and the
mean camber line are not perpendicular to each other. This is true with an admissible ge
ometry. The expression enclosed by parentheses in (3.30) can be zero at some points from
the interval ¸0, b). Thus the derivative s
′
(x) at these points is undeﬁned by (3.30). How
ever, the equation then says s(x) = (ψ
up
(x) + ψ
lo
(x))/2, which corresponds to the idea
of a mean camber line. If the airfoil is closed and the curves representing upper and lower
parts are smooth, then there exists at least one point x ∈ ¸0, b) such that ψ
′
up
(x) = ψ
′
lo
(x).
At this point, either s
′
(x)ψ
′
up
(x) = −1 or the term in parentheses in (3.30) is zero. From
that reason it is necessary to be careful when solving the differential equation and utilize
both boundary conditions. Generally, the existence and uniqueness of the solution is not
guaranteed, but in the common cases the solution is unique.
4. Direct Operator
This operator represents the solution of the ﬂow problem. Depending on the model of
ﬂow used, its formulation can vary. In this case, where the viscous compressible ﬂow is
assumed, the model is described by the system of the NavierStokes equations.
4.1. Mathematical Formulation
The NavierStokes equations are given by
∂ρ
∂t
+
2
j=1
∂(ρv
j
)
∂x
j
= 0, (4.31)
∂(ρv
i
)
∂t
+
2
j=1
∂(ρv
i
v
j
+ p δ
ij
)
∂x
j
=
2
j=1
∂τ
ij
∂x
j
, i = 1, 2, (4.32)
∂E
∂t
+
2
j=1
∂
_
(E + p)v
j
_
∂x
j
=
2
j=1
∂
∂x
j
_
τ
j1
v
1
+ τ
j2
v
2
+
_
µ
P
r
+
µ
T
P
rT
_
γ
∂e
∂x
j
_
.
(4.33)
Since most of the ﬂow in a real situation is turbulent, the laminar model seems insufﬁ
cient. To improve the quality of the predicted ﬂow and also the stability of the method, a
model of turbulence is included (hence the term with the subscript T in (4.33)). In the case
described in this chapter the k − ω turbulence model is used [6], [7]. New variables de
scribing the turbulence properties are introduced, a turbulent kinetic energy k and a speciﬁc
Solution of an Airfoil Design Inverse Problem... 391
turbulent dissipation ω. These two variables are linked together by equations
∂ρk
∂t
+
2
j=1
∂ρkv
j
∂x
j
=
2
j=1
∂
∂x
j
_
(µ + σ
k
µ
T
)
∂k
∂x
j
_
+ P
k
−β
∗
ρωk, (4.34)
∂ρω
∂t
+
2
j=1
∂ρωv
j
∂x
j
=
2
j=1
∂
∂x
j
_
(µ + σ
ω
µ
T
)
∂ω
∂x
j
_
+ P
ω
−βρω
2
+ C
D
. (4.35)
For the simplicity, the system can be rewritten into the vector form
∂w
∂t
+
2
j=1
∂F
j
(w)
∂x
j
=
2
j=1
∂G
j
(w, ∇w)
∂x
j
+S(w, ∇w) . (4.36)
By µ
T
an eddy viscosity coefﬁcient is denoted. This coefﬁcient is given by the formula
µ
T
=
ρk
ω
.
The stress tensor in the N.S. equations is given by relations
τ
11
= (µ + µ
T
)
_
4
3
∂v
1
∂x
1
−
2
3
∂v
2
∂x
2
_
−
2ρk
3
,
τ
22
= (µ + µ
T
)
_
−
2
3
∂v
1
∂x
1
+
4
3
∂v
2
∂x
2
_
−
2ρk
3
,
τ
12
= τ
21
= (µ + µ
T
)
_
∂v
1
∂x
2
+
∂v
2
∂x
1
_
.
The production of turbulence P
k
on the right hand side of the eq. (4.34) and the production
of dissipation P
ω
in eq. (4.35) are expressed as
P
k
= τ
11
∂v
1
∂x
1
+ τ
12
_
∂v
1
∂x
2
+
∂v
2
∂x
1
_
+ τ
22
∂v
2
∂x
2
,
P
ω
= α
ω
ω
P
k
k
,
where τ
ij
= τ
ij
for µ = 0. Finally, the crossdiffusion term C
D
is given by the relation
C
D
= σ
D
ρ
ω
max
_
∂k
∂x
1
∂ω
∂x
1
+
∂k
∂x
2
∂ω
∂x
2
, 0
_
.
The turbulence model is closed by parameters β
∗
= 0.09, β = 5β
∗
/6, α
ω
= β/β
∗
−
σ
ω
κ
2
/
√
β
∗
(where κ = 0.41 is the von Kármán constant), σ
k
= 2/3, σ
ω
= 0.5 a σ
D
=
0.5. This choice of parameters resolves the dependence of the k − ω model on the free
stream values [7]. In the standard Wilcox model without cross diffusion the parameters are
β
∗
= 0.09, β = 5β
∗
/6, α
ω
= β/β
∗
−σ
ω
κ
2
/
√
β
∗
, σ
k
= 0.5, σ
ω
= 0.5 a σ
D
= 0.
If the turbulent kinetic energy k is set to zero, the turbulence model has no inﬂuence
upon the N.S. equations and the laminar model can be solved.
392 Jan Šimák and Jaroslav Pelant
4.2. Numerical Treatment
In the numerical method, a dimensionless system of equations is solved. The vari
ables p, ρ, v
1
, v
2
, k and ω are normalized with respect to the critical values of density ρ
∗
,
velocity c
∗
and pressure p
∗
and to the characteristic length l
∗
. The critical values are the
values corresponding to a unit Mach number. The dimensionless system has the same form
as (4.31)–(4.35). The relations between the original and normalized variables are the fol
lowing:
ˆ v
1
=
v
1
c
∗
, ˆ v
2
=
v
2
c
∗
, ˆ ρ =
ρ
ρ
∗
, ˆ p =
p
ρ
∗
c
2
∗
,
ˆ
k =
k
c
2
∗
, ˆ ω =
ωl
∗
c
∗
, ˆ µ
T
=
µ
T
ρ
∗
c
∗
l
∗
. (4.37)
The problem is solved by an implicit ﬁnite volume method. Details about this method
and the numerical solution of a ﬂow in general can be found in many textbooks, for ex
ample [8]. Since the coupling between the equations describing the ﬂow and the equations
describing the turbulence is only by the viscous terms, it is possible to solve the problem
in two parts [6]. When the ﬂow variables p, ρ, v
1
, v
2
in the time t
k+1
are computed using
(4.31)–(4.33), the turbulent variables k and ω are assumed to be constant in time with values
corresponding to the time t
k
. On the contrary, when computing the turbulent variables k,
ω in a time t
k+1
using (4.34) and (4.35), the ﬂow variables p, ρ, v
1
, v
2
are assumed to be
constant in the time t
k
. That means the solution of the problem consists of two systems,
1. (v
k+1
1
, v
k+1
2
, ρ
k+1
, p
k+1
) = NS(v
k+1
1
, v
k+1
2
, ρ
k+1
, p
k+1
, k
k
, ω
k
),
2. (k
k+1
, ω
k+1
) = Turb(v
k
1
, v
k
2
, ρ
k
, p
k
, k
k+1
, ω
k+1
).
These systems of equations can be solved independently of each other.
The solution of the system of equations (4.31)(4.33) is similar to the way howa laminar
problem is solved. The equations are identical with the pure N.S. equations except the
stress tensor τ
ij
in viscous terms and the heat ﬂux, which depend on k and ω. Since k and ω
are taken as parameters in the system, it is quite easy to modify the existing laminar solver.
By w
h
will be denoted a vector of 6dimensional blocks w
i
of the values of an approx
imate solution on ﬁnite volumes D
i
∈ T
h
. For w
h
∈ R
n
the vector Φ(w
h
) consists of
6dimensional blocks Φ
i
(w
h
) given by
Φ
i
(w
h
) =
1
[D
i
[
j∈S(i)
_
2
s=1
n
s
F
s,h
(w
h
; i, j) [Γ
ij
[ −
2
s=1
n
s
G
s,h
(w
h
; i, j) [Γ
ij
[
_
−S
h
(w
h
; i, j) , (4.38)
where [D
i
[ denotes the cell area, [Γ
ij
[ denotes the length of the edge between D
i
and D
j
,
n
ij
= (n
1
, n
2
) denotes the outer normal to D
i
. Functions F
s,h
(w
h
; i, j), G
s,h
(w
h
; i, j)
and S
h
(w
h
; i, j) are approximations of F(w), G(w, ∇w) and S(w, ∇w) on the grid T
h
.
In order to have a higher order method we apply the Van Leer κscheme or the Van Albada
limiter inside the functions F
s,h
. By S(i) is denoted a set of indices of neighbouring ele
ments and by the symbol τ will be denoted the time step. Thus the implicit ﬁnite volume
Solution of an Airfoil Design Inverse Problem... 393
scheme in a cell D
i
can be written as
w
k+1
i
= w
k
i
−τ
k
Φ
i
(w
k+1
h
). (4.39)
The nonlinear equation above is linearized by the Newton method. The arising system
of linear algebraic equations is solved by the GMRES method (using software SPARSKIT2
[9]). The convective terms F
i
are evaluated using the OsherSolomon numerical ﬂux in the
case of the ﬂow part and by the Vijayasundaram numerical ﬂux in the turbulent part. The
numerical evaluation of a gradient on the edge V
2
V
3
on a boundary (Fig. 4) is done by the
following formulae (the index denotes the value in the corresponding vertex)
∂f
∂x
1
¸
¸
¸
¸
V
2
V
3
≈
−(V
3,y
−V
2,y
) (f
1
−f
wall
)
[(V
2,x
−V
1,x
)(V
3,y
−V
1,y
) −(V
3,x
−V
1,x
)(V
2,y
−V
1,y
)[
,
∂f
∂x
2
¸
¸
¸
¸
V
2
V
3
≈
(V
3,x
−V
2,x
) (f
1
−f
wall
)
[(V
2,x
−V
1,x
)(V
3,y
−V
1,y
) −(V
3,x
−V
1,x
)(V
2,y
−V
1,y
)[
. (4.40)
If the edge is inside the domain, another scheme according to Fig. 5 is used,
[V
1,x
, V
1,y
]
[V
2,x
, V
2,y
] [V
3,x
, V
3,y
]
Figure 4. Scheme for a derivative on a wall.
[V
1,x
, V
1,y
]
[V
2,x
, V
2,y
]
[V
3,x
, V
3,y
]
[V
4,x
, V
4,y
]
Figure 5. Scheme for a derivative inside the domain.
∂f
∂x
1
¸
¸
¸
¸
V
1
V
3
=
(f
3
−f
1
)(V
4,y
−V
2,y
) −(f
4
−f
2
)(V
3,y
−V
1,y
)
[(V
3,x
−V
1,x
)(V
4,y
−V
2,y
) −(V
4,x
−V
2,x
)(V
3,y
−V
1,y
)[
,
∂f
∂x
2
¸
¸
¸
¸
V
1
V
3
= −
(f
3
−f
1
)(V
4,x
−V
2,x
) −(f
4
−f
2
)(V
3,x
−V
1,x
)
[(V
3,x
−V
1,x
)(V
4,y
−V
2,y
) −(V
4,x
−V
2,x
)(V
3,y
−V
1,y
)[
. (4.41)
The points V
2
and V
4
are centres of corresponding cells, f
2
and f
4
are values in these cells.
The values f
1
and f
3
are obtained as an arithmetic mean of values of the four neighboring
cells.
394 Jan Šimák and Jaroslav Pelant
4.3. Boundary Conditions
In our problem, three types of boundary conditions occur: a condition on a wall, a
condition at an inlet boundary and a condition at an outlet boundary. Due to the viscosity,
the zero velocity on the wall is prescribed, further the zero turbulent kinetic energy and
a static temperature are prescribed. The value of the speciﬁc turbulent dissipation ω is
obtained by the formula
ω
wall
=
120µ
ρy
2
c
,
where y
c
is the distance between the wall and the centre of a cell in the ﬁrst row. At the
inlet part of the boundary, the velocity vector (v
1
, v
2
), the density ρ, turbulent energy k
and dissipation ω are prescribed. At the outlet part of the boundary, three variables are
prescribed, the static pressure p, turbulent energy k and dissipation ω. The other variables
are evaluated from values inside the domain. The values of k and ω on the boundary are
values of the free stream and are given in the form of a turbulent intensity I and a viscosity
ratio Re
T
= µ
T
/µ. The turbulent intensity is deﬁned as
I =
_
2
3
k
v
∞
. (4.42)
Following this, the relations for k
∞
and ω
∞
are obtained,
k
∞
=
2
3
(v
∞
I)
2
, (4.43)
ω
∞
=
ρk
µ
_
µ
T
µ
_
−1
. (4.44)
4.4. Mesh Deformation
During the inverse method iterations, the direct operator is applied a number of time.
This is of course true, a new airfoil shape is designed in each iteration. Moreover, there
is a need to ﬁnd an appropriate angle of attack, which ensures the required position of the
stagnation point on the leading edge. This is done by the rotation of the airfoil. This all
results in changes of the computational domain and of the mesh, of course. In order to
remove the dependency on the mesh generator used, the actual mesh is deformed to ﬁt the
new geometry.
The deformation is based on the linear elasticity model which is described in many
textbooks. The grid cells are stretched or shrinked and moved to ﬁt the new domain. No
grid points are created or deleted, the neighbouring cells are the same in the new grid as in
the original grid.
The linear elasticity model is described by the equation
div σ = f in Ω, (4.45)
where σ is a stress tensor and f is some body force. The symbol Ω denotes the domain, its
boundary will be denoted by Γ. The stress tensor is expressed using a strain tensor ǫ as
σ = λTr(ǫ)I + 2µǫ
Solution of an Airfoil Design Inverse Problem... 395
and the components of the strain tensor can be expressed using a displacement function u
as
ǫ
ij
=
1
2
_
∂u
i
∂x
j
+
∂u
j
∂x
i
_
.
The symbols λ and µ are the Lamé constants which describe the physical properties of the
solid. They can be expressed using the Young’s modulus E and Poisson’s ratio ν,
λ =
νE
(1 + ν)(1 −2ν)
, µ =
E
2(1 + ν)
.
Substituting the above mentioned relations into (4.45), the following problem for the
displacement is obtained:
Find an unknown function u : Ω →R
2
such that
(λ + µ)∇(div u) + µ∆u = f in Ω,
u = u
D
on Γ. (4.46)
The function u
D
is the displacement on the boundary, which is known. This problem
is reformulated into the weak sense. Thus the problem for the mesh deformation can be
formulated in the form:
Find a function u ∈ H
1
(Ω)
2
such that u−u
∗
∈ H
1
0
(Ω)
2
, where u
∗
represents Dirichlet
boundary conditions (that means u
∗
∈ H
1
(Ω)
2
, u
∗
[
Γ
= u
D
) and the function u satisﬁes
the equation
−µ
_
Ω
∇u: ∇ϕ
T
dx −(λ + µ)
_
Ω
div u div ϕdx =
_
Ω
f ϕdx (4.47)
for all ϕ ∈ H
1
0
(Ω)
2
. The colon operator is deﬁned by the relation
A: B =
2
i=1
2
j=1
a
ij
b
ji
, A, B ∈ R
2×2
.
The weak problem described above can be solved by a ﬁnite element method. The
domain is discretized by a triangular mesh with nodal points, which are the same as in the
mesh for the ﬂow problem. This leads to the fact, that the solution, which represents the
movement of the nodal points of the original mesh, is evaluated at the appropriate points.
The parameter E is set proportional to the reciprocal values of the cell volumes. This
ensures that the most deformation is carried out on large cells instead of the small ones.
5. Numerical Examples
5.1. Example 1
In this example the given pressure distribution is obtained as a result of a ﬂowaround the
NACA4412 airfoil. The inlet Mach number is M
∞
= 0.6, angle of attack α
∞
= 1.57
◦
and
the Reynolds number Re = 6 10
6
. The ﬂow is described by the use of the turbulence k−ω
model. The resulting airfoil is compared to the original one and thus shown the correctness
of the method. The results are in Figure 6. The relative error of pressure distribution
(measured in L
2
norm) after 40 iterations is 8.32 10
−4
.
396 Jan Šimák and Jaroslav Pelant
X
P
0 0.2 0.4 0.6 0.8 1
0.4
0.6
0.8
1
1.2
1.4
Y 0
0.2
0.4
0.6
0.8
X


e


0 0.2 0.4 0.6 0.8 1
0
0.0001
0.0002
0.0003
X
p

f
0 0.2 0.4 0.6 0.8 1
0.005
0
0.005
0.01
Figure 6. Example 1. Pressure distribution and resulting airfoil shape, error of the resulting
airfoil measured as a normψ
result
−ψ
NACA4412

e
, difference between the prescribed and
resulting pressure distribution on the chord (values are normalized).
5.2. Example 2
In this example a laminar ﬂow with low Reynolds number is computed. The starting
pressure distribution is computed on the NACA3210 airfoil with parameters Re = 1000,
M
∞
= 0.6, α
∞
= 4.56
◦
. At ﬁrst, the problem is computed without any viscous correction
and then the mentioned correction based on the Pohlhausen’s method is used. From the
results is evident, that for very low Reynolds numbers the correction is necessary. The
comparisons are in Figure. 7.
X
P
0 0.2 0.4 0.6 0.8 1
0.6
0.8
1
1.2
1.4
Y 0
0.2
0.4
0.6
0.8
X
y
0 0.2 0.4 0.6 0.8 1
0
0.2
X
p

f
0 0.2 0.4 0.6 0.8 1
0.02
0.01
0
0.01
0.02
Figure 7. Example 2. Pressure distributions and resulting airfoil shapes (solid  correction,
dashed  without correction), comparison of the airfoils with the NACA3210 (dotted), dif
ference between the prescribed and resulting pressure distribution along the chord (values
are normalized).
Solution of an Airfoil Design Inverse Problem... 397
6. Conclusion
A numerical method for a solution of an inverse problem of ﬂow around an airfoil was
described. The advantage of this method is the weak dependency on the model describing
the ﬂow. It is easy to modify the method by assuming a different model. The correction due
to viscosity is necessary only with low Reynolds numbers. The presented method has still
some drawbacks. These are the dependency of the angle of attack on the prescribed distri
bution and the applicability on the subsonic regimes only. The method can be improved in
the future.
Acknowledgment
This work was supported by the Grant MSM 0001066902 of the Ministry of Education,
Youth and Sports of the Czech Republic.
References
[1] Pelant, J. Inverse Problem for Twodimensional Flow around a Proﬁle, Report No.
Z–69; VZLÚ, Prague, 1998.
[2] Šimák J.; Pelant J. A contractive operator solution of an airfoil design inverse prob
lem; PAMM Vol. 7, No. 1 (ICIAM07), pp. 2100023–2100024.
[3] Šimák, J.; Pelant, J. Solution of an Airfoil Design Problem With Respect to a Given
Pressure Distribution for a Viscous Laminar Flow, Report No. R–4186; VZLÚ,
Prague, 2007.
[4] Michlin, S. G. Integral Equations; Pergamon Press, Oxford, 1964.
[5] Schlichting, H. BoundaryLayer Theory; McGrawHill, New York, 1979.
[6] Wilcox, D. C. Turbulence Modeling for CFD; DCW Industries Inc., 1998; 2nd ed.
[7] Kok, J. C. Resolving the Dependence on Freestream Values for the k −ω Turbulence
Model; AIAA Journal 2000, vol. 38, No. 7, pp. 1292–1295.
[8] Feistauer, M.; Felcman, J.; Straškraba, I. Mathematical and Computational Methods
for Compressible Flow; Clarendon Press, Oxford, 2003.
[9] Saad, Y. Iterative Methods for Sparse Linear Systems; SIAM, 2003; 2nd ed.
In: Fluid Mechanics and Pipe Flow
Editors: D. Matos and C. Valerio, pp. 399440
ISBN 9781607410379
c 2009 Nova Science Publishers, Inc.
Chapter 13
SOME FREE BOUNDARY PROBLEMS IN POTENTIAL
FLOW REGIME USING THE LEVEL SET METHOD
M. Garzon
1
, N. BobilloAres
1
and J.A. Sethian
2
1
Dept. de Matem´ aticas, Univ. of Oviedo, Spain
2
Dept. of Mathematics, University of California, Berkeley, and
Mathematics Department, Lawrence Berkeley National Laboratory.
Abstract
Recent advances in the ﬁeld of ﬂuid mechanics with moving fronts are linked to the
use of Level Set Methods, a versatile mathematical technique to follow free boundaries
which undergo topological changes. A challenging class of problems in this context
are those related to the solution of a partial differential equation posed on a moving
domain, in which the boundary condition for the PDE solver has to be obtained from
a partial differential equation deﬁned on the front. This is the case of potential ﬂow
models with moving boundaries. Moreover, the ﬂuid front may carry some material
substance which diffuses in the front and is advected by the front velocity, as for ex
ample the use of surfactants to lower surface tension. We present a Level Set based
methodology to embed this partial differential equations deﬁned on the front in a com
plete Eulerian framework, fully avoiding the tracking of ﬂuid particles and its known
limitations. To show the advantages of this approach in the ﬁeld of Fluid Mechanics
we present in this work one particular application: the numerical approximation of a
potential ﬂow model to simulate the evolution and breaking of a solitary wave propa
gating over a slopping bottom and compare the level set based algorithm with previous
front tracking models.
1. Introduction
In this chapter we present a class of problems in the ﬁeld of ﬂuid mechanics that can
be modeled using the potential ﬂow assumptions, that is, inviscid and incompressible ﬂuids
moving under an irrotational velocity ﬁeld. While these are signiﬁcant assumptions, in the
presence of moving boundaries, the resulting equations is a non linear partial differential
equation, which adds considerable complexity to the computational problem. In the litera
ture this model is often called the fully non linear potential ﬂow model (FNPFM). Several
400 M. Garzon, N. BobilloAres and J.A. Sethian
interesting and rather complicated phenomenon are described using the FNPFM, as for ex
ample, HelleShaw ﬂows, jet evolution and drop formation, sprays and electrosprays, wave
propagation and breaking mechanisms, etc, see [21], [22], [30], [13].
Level Set Methods (LSM) [31], [33], [34] [37] are widely used in ﬂuid mechanics, as
well as other ﬁelds such as medical imaging, semiconductor manufacturing, ink jet print
ing, and seismology. The LSM is a powerful mathematical tool to move interfaces, once the
velocity is known. In many physical problems, the interface velocity is obtained by solving
the partial differential equations system used to model the ﬂuid/ﬂuids ﬂow. The LSM is
based on embedding the moving front as the zero level set of one higher dimensional func
tion. By doing so, the problem can be formulated in a complete Eulerian description and
topological changes of the free surface are automatically included. The equation for the
motion of the level set function is an initial value hyperbolic partial differential equation,
which can be easily approximated using upwind ﬁnite differences schemes.
Recently, the LSM has been extended to formulate problems involving the transport
and diffusion of material quantities, see [3]. In [3] model equations and algorithms are
presented together with the corresponding test examples and convergence studies. This
led to the realization that the nonlinear boundary conditions in potential ﬂow problems
could also be embedded using level set based methods. As a result, the FNPFM can also be
formulated with an Eulerian description with the associated computational advantages. Two
difﬁcult problems that have been already approximated using this novel algorithm are wave
breaking over sloping beaches [16], [17] and the Rayleigh taylor instability of a water jet
[20]. Moreover, related to drop formation and wave breaking, it has been recently reported
in the literature [46], [45] that the presence of surfactants on the ﬂuid surface affects the
ﬂow patterns. The models described in this chapter are the groundwork for solving these
complex problems.
This chapter is organized as follows: in section 2. we have made an effort to obtain dy
namic equations valid for any spatial coordinate system. To do so, we derive the equations
using only objects deﬁned in an intrinsic way (i.e., independent of any coordinate system).
At the same time, in accordance with the level set perspective, we have avoided as much
as possible, the “ material description” (Lagrangian coordinates). Geometric quantities are
deﬁned using the level sets and tensor ﬁelds in the space. In section 3. a brief description
of the Levels Set Method is given using this intrinsic approach. Section 4. is devoted to
describe two particular potential ﬂow models, the ﬁrst one related to drop formation in the
presence of surfactants, which combines all the models derived in section 2.. The wave
breaking problem is modeled in 2D, code development in 3D is underway. In section 5., we
present the numerical approximation and algorithm for the wave breaking problem. Numer
ical results and accuracy tests are also presented in section 6.. Precise deﬁnitions of certain
needed geometrical tools, throughout used in this chapter, are shown in Appendix III.
2. Some Physical Models
Here, we discuss the derivations of ﬂuid problems and their corresponding reformula
tion using the Level Set Method (LSM) techniques. The brief derivation of known physical
laws is used also as a pretext to introduce some preliminary concepts and notation.
Some Free Boundary Problems in Potential Flow Regime... 401
2.1. Kinematic Relationships
Reference conﬁguration. The conﬁguration of a continuous medium at certain time t
is known when the position of each particle is speciﬁed. We name Ω
t
the space region
occupied by the continuous medium at that time.
Kinematics require the movement description of each particle. To this aim, we must:
i. Label the particles.
ii. Specify the movement of each particle.
The ﬁrst step is done considering the conﬁguration at an arbitrary instant t
0
(reference
conﬁguration). Particles are marked by the point P
0
∈ Ω
t
0
they occupy. Points in Ω
t
0
are
good labels because they are in a 1 to 1 correspondence with the particles (“particles can not
penetrate each other”). In what follows we will abbreviate the phrase “particle with label
P
0
” by “particle P
0
”.
Once all the particles are labeled, it is now possible to undertake the second step. Let
P
0
∈ Ω
t
0
be a particle. Its position P at instant t is given by the function:
P = R(P
0
, t), P ∈ Ω
t
, P
0
∈ Ω
t
0
. (1)
According to the reference conﬁguration deﬁnition, we have:
R(P
0
, t
0
) = P
0
. (2)
The mapping R
t
, R
t
(P
0
) := R(P
0
, t) = P, must be invertible:
P
0
= R
−1
t
(P) ∈ Ω
t
0
, P ∈ Ω
t
. (3)
Lagrangian/Eulerian descriptions. Any tensor ﬁeld w may be described in two ways,
using (1):
w = w(P, t) = w(R(P
0
, t), t) = w
0
(P
0
, t). (4)
Function w(P, t) corresponds to the so called Eulerian description and w
0
(P
0
, t) corre
sponds to the Lagrangian description. As a consequence any tensorial ﬁeld w admits two
time partial derivatives. The “spatial” derivative, corresponding to the Eulerian description:
∂
t
w :=
d
dǫ
w(P, t +ǫ)
¸
¸
¸
¸
ǫ=0
, (5)
measures the variation rate with time of w from a ﬁxed point in the space. The “convective”
derivative, corresponding to the Lagrangian description:
D
t
w :=
d
dǫ
w
0
(P
0
, t +ǫ)
¸
¸
¸
¸
ǫ=0
, (6)
gives the variation rate of w following the particle P
0
.
402 M. Garzon, N. BobilloAres and J.A. Sethian
Velocity ﬁeld The velocity u = u(P
0
, t) of the particle P
0
is obtained using the convec
tive derivative (“following the particle”) of the position P = R(P
0
, t):
u = D
t
P, P = R(P
0
, t). (7)
Obviously, u admits both descriptions:
u = u(P, t) = u
0
(P
0
, t), P = R(P
0
, t). (8)
Given an arbitrary tensor ﬁeld w, its spatial and convective derivatives are related using
the calculus chain rule and the deﬁnition (7):
D
t
w = ∂
t
w +∂
u
w. (9)
Here, ∂
u
w designates the directional derivative of w along u (see Appendix III). The ac
celeration of particle P
0
is obtained by the convective derivative of the velocity ﬁeld. Using
(9), we have:
D
t
u = ∂
t
u +u · ∇u. (10)
Transport of a vector due to a moving medium. A ﬂuid particle is located at point
1
P
at time t. After a time ∆t, the same particle is at point R(P, t + ∆t). Clearly, the function
R must verify that R(P, t + 0) = P. A nearby particle at same time t is located at P +ǫa,
and at t + ∆t is at point R(P + ǫa, t + ∆t). We have again P + ǫa = R(P + ǫa, t + 0).
The vector ǫa that connects both particles varies as they move. Denote by D
t
ǫa its rate of
change with time:
D
t
ǫa = lim
∆t→0
1
∆t
[(R(P +ǫa, t + ∆t) −R(P, t + ∆t)) −(R(P +ǫa, t) −R(P, t))]
= lim
∆t→0
R(P +ǫa, t + ∆t) −R(P +ǫa, t)
∆t
− lim
∆t→0
R(P, t + ∆t) −R(P, t)
∆t
.
The ﬁrst term of the right hand side of previous equation is, by deﬁnition, the particle
velocity at P + ǫa, u(P + ǫa, t), and the second term the particle velocity at P, u(P, t).
Thus we have:
D
t
ǫa = u(P +ǫa, t) −u(P, t).
Letting ǫ → 0, we obtain the rate of change with time of an inﬁnitesimal vector dragged by
the medium:
D
t
a := lim
ǫ→0
1
ǫ
D
t
ǫa = lim
ǫ→0
u(P +ǫa, t) −u(P, t)
ǫ
=
d
dǫ
u(P +ǫa, t)
¸
¸
¸
¸
ǫ=0
= ∂
a
u. (11)
We denote ∂
a
the operator that performs the directional derivative along the vector a (see
Appendix III).
1
For this calculation we use here the conﬁguration at t as the reference conﬁguration.
Some Free Boundary Problems in Potential Flow Regime... 403
Fluid volume change as it is transported by the velocity ﬁeld. Let a, b and c be three
small vectors with origin at point P. The volume of the parallelepiped spanned by vectors
a, b, c is given by the trilinear alternate form
δV = [a, b, c] = a · b ×c.
The rate of change of this volume, when particles located on its vertices move, is given by
D
t
δV , and thus we have
D
t
δV = D
t
[a, b, c] = [D
t
a, b, c] + [a, D
t
b, c] + [a, b, D
t
c].
Using now (11) we get
D
t
δV = [∂
a
u, b, c] + [a, ∂
b
u, c] + [a, b, ∂
c
u],
which is also a trilinear alternate form. As in the tridimensional space all these forms are
proportional, we can set
D
t
[a, b, c] = (div u)[a, b, c], (12)
which gives us an intrinsic deﬁnition of the divergence of the ﬁeld u. If the continuous
medium is incompressible, the volume δV does not change, D
t
δV = 0, and we arrive at
the incompressibility condition
div u = 0. (13)
2.2. Dynamic Relationships
Conservation of mass. Denote by ρ = ̺(P, t) the volumetric mass density of the contin
uous medium at point P and at time t. The rate of change of the mass in a small volume
δV dragged by the velocity ﬁeld is, using deﬁnition (12),
D
t
(ρδV ) = (D
t
ρ)δV +ρD
t
δV = (D
t
ρ +ρ div u)δV.
The mass conservation law is thus
D
t
ρ +ρ div u = 0. (14)
Applying general formula (9) to ρ, we have D
t
ρ = ∂
t
ρ + ∂
u
ρ. In the case of an homoge
neous and incompressible medium with uniform initial density ρ
0
, using equations (14) and
(13), we have D
t
ρ = 0 which gives ̺(P, t) = ρ
0
.
Conservation of the momentum associated with a small piece of continuous medium.
From Newton’s second law applied to a ﬂuid volume V we get the relation
D
t
_
V
u ρdV =
_
V
g ρdV +
_
∂V
τ(ds). (15)
The term in left hand side of this equation is the rate of change with time of the momentum
associated with volume V when dragged by the continuous medium. The ﬁrst term in the
right hand side corresponds to the volumetric forces inside V , generated by a vector ﬁeld
404 M. Garzon, N. BobilloAres and J.A. Sethian
per unit mass g, usually the gravitational ﬁeld. The second term represents the “contact”
forces applied by the rest of the medium over the part in V . The Cauchy’s tensor τ is a linear
operator ﬁeld that is obtained from speciﬁc relationships which depend on the material, the
so called constitutive relations. We are interested in inviscid ﬂuids which verify the Pascal’s
law:
τ(ds) = −pds,
where p is the pressure scalar ﬁeld. Green’s formula,
_
∂V
−p ds =
_
V
−∇p dV,
shows that contact forces may be computed as a kind of volume forces with density −∇p.
For a small volume δV dragged by the ﬂuid, equation (15) can be written:
D
t
(u ρδV ) = (g ρ −∇p)δV. (16)
Due to the mass conservation law, D
t
(ρδV ) = 0, equation (16) leads to the Euler equation:
D
t
u = ∂
t
u +∂
u
u = g −
1
ρ
∇p. (17)
If g is a uniform ﬁeld it comes from the gradient of a potential function:
g = −∇U(P), U(P) = −g · (P −O),
where P −O is the position vector of the point P.
2.3. Potential Flow
Assuming an irrotational ﬂow regime, curl u = 0, there exists an scalar ﬁeld φ such
that
u = ∇φ. (18)
Outside of the ﬂuid domain, and separated by a free boundary, there is a gas at pressure p
a
that is assumed to be constant. This means that, within the gas, the time needed to restore
the equilibrium is very small compared with the time evolution of the ﬂuid. Therefore, at
the ﬂuid free boundary, the boundary condition is just:
p = p
a
. (19)
Using the vectorial relationship ∇u
2
/2 = ∂
u
u +u ×(curl u), and relations (18) and (13)
we have
∇
_
∂
t
φ +
1
2
u
2
+
p
ρ
+U
_
= 0.
Performing the ﬁrst integration,
∂
t
φ +
1
2
u
2
+
p
ρ
+U = C(t),
Some Free Boundary Problems in Potential Flow Regime... 405
where C(t) is an arbitrary function of time, which can be chosen in such a way that the
previous relation can be written:
∂
t
φ +
1
2
u
2
+
p −p
a
ρ
+U = 0.
Now using the obvious relation ∂
t
φ +u
2
= ∂
t
φ +∂
u
φ = D
t
φ, we ﬁnally obtain
D
t
φ −
1
2
u
2
+
p −p
a
ρ
+U = 0. (20)
2.4. Advection
On the surface of a continuous medium with a known movement, a certain substance is
distributed, which will be named as “charge”. This is adhered to the ﬂuid particles and it
is transported by them. In this way a set of particles will always carry the same amount of
“charge”. This phenomenon is called advection.
The continuous medium surface is implicitly described as the zero level set of a certain
scalar function Ψ = ψ(P, t):
Γ
t
= {Qψ(Q, t) = 0}. (21)
Vectors a tangent to the surface are characterized by the condition
∂
a
Ψ = a · ∇Ψ = 0;
thus, the tangent vectorial plane at each point of the surface is given by the normal unit
vector
2
n =
∇Ψ
∇Ψ
.
The function ψ by itself does not specify the particle movement on the surface, just its
shape. We need to add the information about how these particles move, e.g., specifying the
velocity ﬁeld on the surface
Q ∈ Γ
t
, u = u(Q, t).
A small vector a connecting two nearby particles on the surface and dragged by them as
they move, has a rate of change given by (11),
D
t
a = ∂
a
u. (22)
Note that a is a tangent vector, n · a = 0.
Surface areas. Using the normal vector to the surface, n, a 2–form to calculate surface
areas can be constructed:
3
ω(a, b) := [n, a, b] = n · a ×b,
2
Ψ must increase from the interior to the exterior of the surface to get n outwards.
3
The surface area deﬁnition is not made using the Gram determinant of two tangent vectors, because this
procedure involves a particular parametrization of the surface. Instead, a 2form is deﬁned from the volume
form in space (“the parallelogram area is the volume of a rectangular prism of unit height”).
406 M. Garzon, N. BobilloAres and J.A. Sethian
where ω(a, b) is the area spanned by tangent vectors a, b, and [n, a, b] the volume form in
the 3D space.
As the tangent vectors a, b are transported by the surface movement, the parallelogram
area associated to them changes. The rate of change with time is easily obtained:
D
t
ω(a, b) = (D
t
n) · a ×b +n · (D
t
a) ×b +n · a ×D
t
b.
First term of the right hand side of previous equation is zero since a ×b is a normal vector
and D
t
n is tangent: indeed, as n
2
= 1, we have D
t
n
2
= 2n · D
t
n = 0.
Using (22) we have
D
t
ω(a, b) = ∂
a
u · b ×n +∂
b
u · n ×a.
This expression is bilinear and alternate with respect the tangent vectors. It must be, at
each point on the surface, proportional to the 2–form ω. We denote by Div u, “surface
divergence”, the proportionality coefﬁcient:
∂
a
u · b ×n +∂
b
u · n ×a := (Div u) ω(a, b) (23)
This deﬁnition of Div u does not depend upon the choice of tangent vectors a and b. In
Appendix I, the expression for the surface divergence of an arbitrary vector ﬁeld w using
rectangular coordinates is shown.
Advection law. Now, let be
σ = σ(Q, t), Q ∈ Γ
t
the “charge” surface density. The “charge” δq carried by a small parallelogram, spanned by
two small tangent vectors (a, b), of area ω(a, b) is
δq = σ ω(a, b).
As the “charge” is conserved, the advection law is
D
t
δq = 0.
Now, by deﬁnition (23), we have
D
t
(σω) = (D
t
σ)ω +σD
t
ω = (D
t
σ +σ Div u)ω. (24)
Hence we arrive to the intrinsic equation for the advection phenomena:
D
t
σ +σ Div u = 0.
Some Free Boundary Problems in Potential Flow Regime... 407
2.5. AdvectionDiffusion
Next, we are going to assume that the “charge” diffuses along particles on the surface
according to the Fick’s law:
4
j = −α∇σ,
where α is the diffusion coefﬁcient, j is the “charge” ﬂux and ∇σ is the “charge” surface
density gradient. As σ is only deﬁned on the surface Γ
t
, ∇σ is only deﬁned for tangent
vectors:
∇σ · a := ∂
a
σ, a tangent vector.
On the surface Γ
t
let us consider a surface region S, bounded by a curve ∂S. Let be ννν the
unit vector ﬁeld tangent to Γ
t
and orthogonal to the curve ∂S at each point. The “charge”
that leaves the surface per unit time is the outward ﬂux through the boundary ∂S:
_
δS
j · ννν dl = −
_
∂S
j · n ×dl =
_
∂S
n ×j · dl.
Applying now Stokes’ theorem, we have
_
∂S
n ×j · dl =
_
S
A ω(d
1
P, d
2
P). (25)
The 2form of the surface integral is obtained using the intrinsic formula
A ω(a, b) = ∂
a
(n ×j · b) −∂
b
(n ×j · a). (26)
We interpret A ω(a, b) as the “charge” per unit time that, by diffusion, leaves the small
parallelogram spanned by the tangent vectors (a, b).
Now it is straightforward to set the condition for the advectiondiffusion mechanism
_
_
_
“charge” rate of change
within the tangent
parallelogram (a, b)
_
_
_
= −
_
_
_
“charge” that leaves
the parallelogram
by diffusion
_
_
_
,
that is
D
t
(σω(a, b)) = −A ω(a, b).
In Appendix II the following expression for A is obtained:
A = Div j −(Div n) j · n.
Hence, using (24) we arrive at the general equation for the advectiondiffusion model:
_
D
t
σ +σ Div u = −Div j + (Div n) j · n
j = −α ∇σ
or
D
t
σ +σ Div u = α Div ∇σ −α(Div n) ∇σ · n (27)
4
Fick’s diffusion law applies when the “charge” particles move randomly without any preferential direction
(Brownian movement).
408 M. Garzon, N. BobilloAres and J.A. Sethian
The Cartesian expressions
5
for Div u, Div ∇σ and Div n are:
Div n = (δ
ij
−n
i
n
j
)∂
j
n
i
=
1
∇Ψ
3
_
(∇Ψ)
2
∂
i
∂
i
Ψ−∂
j
Ψ∂
i
Ψ∂
j
∂
i
Ψ
¸
, (28)
Div ∇σ =
1
∇Ψ
2
_
(∇Ψ)
2
∂
i
∂
i
σ −∂
i
Ψ∂
j
Ψ∂
i
∂
j
σ
¸
, (29)
Div u =
1
∇Ψ
2
_
(∇Ψ)
2
∂
i
u
i
−∂
i
Ψ∂
j
Ψ∂
i
u
j
¸
. (30)
Expanding the implicit summands, we obtain the following expressions for the 3D space
(i, j = 1, 2, 3):
Div n =
1
∇Ψ
3
_
(∂
1
Ψ)
2
(∂
2
2
Ψ +∂
2
3
Ψ) + (∂
2
Ψ)
2
(∂
2
1
Ψ +∂
2
3
Ψ)+
+(∂
3
Ψ)
2
(∂
2
1
Ψ +∂
2
2
Ψ) −2∂
1
Ψ∂
2
Ψ∂
1
∂
2
Ψ−
−2∂
1
Ψ∂
3
Ψ∂
1
∂
3
Ψ−2∂
2
Ψ∂
3
Ψ∂
2
∂
3
Ψ
¸
, (31)
Div ∇σ =
1
∇Ψ
2
_
(∂
1
Ψ)
2
(∂
2
2
σ +∂
2
3
σ) + (∂
2
Ψ)
2
(∂
2
1
σ +∂
2
3
σ)+
+(∂
3
Ψ)
2
(∂
2
1
σ +∂
2
2
σ) −2∂
1
Ψ∂
2
Ψ∂
1
∂
2
σ −
−2∂
1
Ψ∂
3
Ψ∂
1
∂
3
σ −2∂
2
Ψ∂
3
Ψ∂
2
∂
3
σ
¸
, (32)
Div u =
1
∇Ψ
2
_
(∂
1
Ψ)
2
(∂
2
u
2
+∂
3
u
3
) + (∂
2
Ψ)
2
(∂
1
u
1
+∂
3
u
3
)+
+(∂
3
Ψ)
2
(∂
1
u
1
+∂
2
u
2
) −∂
1
Ψ∂
2
Ψ(∂
1
u
2
+∂
2
u
1
) −
−∂
1
Ψ∂
3
Ψ(∂
1
u
3
+∂
3
u
1
) −∂
2
Ψ∂
3
Ψ(∂
2
u
3
+∂
3
u
2
)
¸
. (33)
To obtain the formulas for the plane we assume axial symmetry in the direction 3:
∂
3
Ψ = 0, ∂
2
3
Ψ = 0, ∂
3
σ = 0, u
3
= 0, ∂
3
u
i
= 0,
∇Ψ
2
= (∂
1
Ψ)
2
+ (∂
2
Ψ)
2
.
Inserting these values in (31), (32) and (33) we get:
Div n =
1
∇Ψ
3
_
(∂
1
Ψ)
2
(∂
2
2
Ψ) + (∂
2
Ψ)
2
(∂
2
1
Ψ) −2∂
1
Ψ∂
2
Ψ∂
1
∂
2
Ψ
¸
, (34)
Div ∇σ =
1
∇Ψ
2
_
(∂
1
Ψ)
2
(∂
2
2
σ) + (∂
2
Ψ)
2
(∂
2
1
σ) −2∂
1
Ψ∂
2
Ψ∂
1
∂
2
σ
¸
, (35)
Div u =
1
∇Ψ
2
_
(∂
1
Ψ)
2
(∂
2
u
2
) + (∂
2
Ψ)
2
(∂
1
u
1
)−
−2∂
1
Ψ∂
2
Ψ(∂
1
u
2
+∂
2
u
1
)
¸
. (36)
5
In the following expressions we use only subscripts because orthonormal bases coincides with their corre
sponding reciprocal ones. Then, the position of the indices becomes irrelevant.
Some Free Boundary Problems in Potential Flow Regime... 409
3. The Level Set Method
The Level Set method is a mathematical tool developed by Osher and Sethian [31] to
followinterfaces which move with a given velocity ﬁeld. The key idea is to viewthe moving
front as the zero level set of one higher dimensional function called the level set function.
One main advantage of this approach comes when the moving boundary changes topology,
and thus a simple connected domain splits into separated disconnected domains.
Let be Γ
t
the set of points lying in the surface boundary at time t. This surface is deﬁned
through the zero level set of the scalar ﬁeld Ψ = ψ(P, t):
Γ
t
= {Qψ(Q, t) = 0}. (37)
To identify the ﬂuid particles, the conﬁguration at t
0
(reference conﬁguration) is used:
Γ
t
0
= {Q
0
ψ(Q
0
, t
0
) = 0}. (38)
The particle movement is speciﬁed through the function
Q = R(Q
0
, t), (39)
which gives the position Q ∈ Γ
t
of the ﬂuid particle Q
0
∈ Γ
t
0
. The particle Q
0
velocity is
calculated using the convective derivative D
t
(“following the particle”):
u = D
t
Q =
d
dǫ
R(Q
0
, t +ǫ)
¸
¸
¸
¸
ǫ=0
. (40)
According to deﬁnition (37), we have ψ(R(Q
0
, t), t) = 0. Deriving with respect to time
and applying the chain rule, we obtain
∂
t
Ψ +u · ∇Ψ = 0. (41)
which has to be completed with the value of the level set function at time t = 0, usually set
to be the signed distance function to the initial front,
Ψ(P, 0) = s(P)d(P),
being d(P) the distance from the point P to the surface at the initial conﬁguration Γ
0
,
s(P) = −1 if P ∈ Ω
0
and s(P) = +1 if P / ∈ Ω
0
.
Now, if we take a ﬁxed 3D domain Ω
D
that contains the free surface for all times, we
can deﬁne the initial value problem for the level set function Ψ posed on Ω
D
:
∂
t
Ψ +u · ∇Ψ = 0 in Ω
D
(42)
Ψ(P, 0) = s(P)d(P) in Ω
D
(43)
A graphical interpretation of the level set function evolution is depicted in ﬁgure 1 Equation
(42) moves all the level set of Ψ, not just the zero level set, and in many physical applica
tions the front velocity is just deﬁned for points lying on the free boundary. Therefore for
this equation to be valid on the whole domain we have to extend the velocity u off the front.
There exist several extension procedures which will be brieﬂy commented below.
410 M. Garzon, N. BobilloAres and J.A. Sethian
G
0
y(P, 0)
y(P, t)
G
t
t
Figure 1. Evolution of the level set function.
3.1. Extension of Functions Deﬁned on the Front
Let us consider a classical result from functional analysis: suppose a domain Ω bounded
by a closed surface ∂Ω. If for k ≥ 1 the surface ∂Ω ∈ C
k
, then for all functions F(x) ∈
C
k
(∂Ω) there exists a function F
ext
(x) ∈ C
k
(
¯
Ω) such that F
ext

∂Ω
= F(x).
In practice, there are several ways to extend any magnitude F deﬁned on the front onto
Ω
D
. As shown in [10] for the numerical stability of the level set equation it is convenient
to preserve Ψ as a signed distance function, which is characterized by the property ∇Ψ =
1. One way is to perform reinitializations of the level set function at chosen times. If
this is done periodically, it will smooth the level set function. However, done too often,
especially using poor reinitialization techniques, spurious mass loss/gain will occur. Thus,
it is important to perform reinitialization both sparingly and accurately. For the potential
ﬂow problems presented in this chapter we follow the strategy introduced in [2]. The idea is
to extrapolate F given at the front along its gradient. Mathematically the extended variable
F
ext
is the solution of
∇F
ext
· ∇Ψ = 0. (44)
It is straightforward to show that this choice maintains the signed distance function for
the level sets of Ψ for all times. For the numerical approximation we proceed as follows:
given a level set function Ψ at time n, namely Ψ
n
, one ﬁrst obtains a distance function
ˆ
Ψ
n
around the zero level set. Simultaneous with this construction, the extended quantity F
ext
is obtained satisfying Eq. (44). For a complete explanation of this extension method see
[2].
4. Examples of Potential Flow Models with Moving Boundaries
In this section the governing equations of two interesting physical problems will be for
mulated using a level set framework. First, drop formation is a complex 3D phenomena
driven mainly by capillary forces, which can be modeled using the potential ﬂow assump
Some Free Boundary Problems in Potential Flow Regime... 411
tions. It is wellknown that the presence of surface surfactants lowers the surface tension
affecting drops shape. Secondly, propagation and wave breaking over sloping beaches can
also be modeled with the potential ﬂow equations, which are valid until the jet of the wave
impinges against the ﬂat water surface. In this case we can formulate the equations in 2D
taking a vertical section of the beach, which facilitates the algorithm and code development.
The wave numerical simulations will be presented in section 6..
4.1. Governing Equations for Surface Tension Driven Flows with Material
AdvectionDiffusion
Let Ω
t
be the 3Dclosed ﬂuid domain surrounded by air and Γ
t
the free surface boundary
at time t. Suppose that initially a certain amount of surfactants, which are assumed to be
insoluble in water, are uniformly distributed on the surface (see Figure 2).
Figure 2. A ﬂuid volume with surface surfactants.
For an incompressible and inviscid ﬂuid, the governing equations are the Euler equa
tions (17). On the free boundary the following partial differential equations apply:
• The advectiondiffusion equation for the surfactant is (27):
D
t
σ +σDiv u = α(Div∇σ −κ ∇σ · n) on Γ
t
,
where σ is the surface density of the surfactant, u is the free boundary velocity, α is
the surface diffusion coefﬁcient, κ = Div n =
1
R
1
+
1
R
2
and R
1
, R
2
the principal
radii of curvature of Γ
t
at each point.
• Continuity of the stress tensor between water and air leads to the balance of the
surface tension forces, p = p
a
+γ(
1
R
1
+
1
R
2
), where γ is the surface tension coefﬁcient
412 M. Garzon, N. BobilloAres and J.A. Sethian
that may depend on the surfactant concentration σ. Thus Eq. (20) becomes
∂