Journal of Computational Physics 190 (2003) 550–571

www.elsevier.com/locate/jcp

A pressure-based algorithm for multi-phase flow at all speeds

a,*
F. Moukalled , M. Darwish a, B. Sekar b

a
Mechanical Engineering Department, Faculty of Engineering and Architecture, American University of Beirut,
P.O. Box 11-0236, Riad El Solh, Beirut 1107 2020, Lebanon
b
Air Force Research Laboratory, AFRL/PRTC, Wright–Patterson AFB, OH 45433-7251, USA
Received 28 February 2002; received in revised form 21 May 2003; accepted 26 May 2003

Abstract

A new finite volume-based numerical algorithm for predicting incompressible and compressible multi-phase flow
phenomena is presented. The technique is equally applicable in the subsonic, transonic, and supersonic regimes. The
method is formulated on a non-orthogonal coordinate system in collocated primitive variables. Pressure is selected as a
dependent variable in preference to density because changes in pressure are significant at all speeds as opposed to
variations in density, which become very small at low Mach numbers. The pressure equation is derived from overall
mass conservation. The performance of the new method is assessed by solving the following two-dimensional two-phase
flow problems: (i) incompressible turbulent bubbly flow in a pipe, (ii) incompressible turbulent air–particle flow in a
pipe, (iii) compressible dilute gas–solid flow over a flat plate, and (iv) compressible dusty flow in a converging diverging
nozzle. Predictions are shown to be in excellent agreement with published numerical and/or experimental data.
Ó 2003 Elsevier Science B.V. All rights reserved.

Keywords: Multi-phase flow; Pressure-based algorithm; All speed flows; Finite volume method

1. Introduction

The last two decades have witnessed a substantial transformation in the computational fluid dynamics
(CFD) industry; from a research means confined to research laboratories, CFD has emerged as an every
day engineering tool for a wide range of industries (Aeronautics, Automobile, chemical Processing, etc.).
This increasing dependence on CFD is due to a multitude of factors that have rendered practical the
simulation of large complex industrial-type problems. Some of these factors are directly related to the
maturity of several numerical aspects at the core of CFD. These include: multi-grid acceleration techniques
[1–4] with enhanced equation solvers [5,6] that have decreased the computational cost of tackling large
problems, better discretization techniques with bounded high resolution schemes [13–18] yielding more
accurate results, unstructured grid methods [7–12] that simplify the description of complex geometry, as

*
Corresponding author. Tel.: +961-3-831-432; fax: +961-1-744-462.
E-mail addresses: [email protected] (F. Moukalled), [email protected] (M. Darwish), [email protected] (B. Sekar).

0021-9991/$ - see front matter Ó 2003 Elsevier Science B.V. All rights reserved.
doi:10.1016/S0021-9991(03)00297-3
 F. Moukalled et al. / Journal of Computational Physics 190 (2003) 550–571 551

Nomenclature
ðkÞ
AP ; . . . coefficients in the discretized equation for /ðkÞ
ðkÞ
BP source term in the discretized equation for /ðkÞ
BðkÞ body force per unit volume of fluid/phase k
CqðkÞ coefficient equals to 1=RðkÞ T ðkÞ two-phase.
ðkÞ
DP ½/ðkÞ
matrix operator defined in Eq. (14)
HP ½/ðkÞ
the H operator
HP ½uðkÞ
the vector form of the HP operator
IðkÞ inter-phase momentum transfer
ðkÞD
Jf diffusion flux of /ðkÞ across cell face Ôf Õ
ðkÞC
Jf convection flux of /ðkÞ across cell face Ôf Õ
M _ ðkÞ mass source per unit volume
P pressure
ðkÞ
PrðkÞ ; Prt laminar and turbulent Prandtl number for fluid/phase k
ðkÞ
q_ heat generated per unit volume of fluid/phase k
QðkÞ general source term of fluid/phase k
rðkÞ volume fraction of fluid/phase k
Sf surface vector
t time
T ðkÞ temperature of fluid/phase k
ðkÞ ðkÞ
Uf interface flux velocity ðvf  Sf Þ of fluid/phase k
ðkÞ
u velocity vector of fluid/phase k
uðkÞ ; vðkÞ ; . . . velocity components of fluid/phase k
x; y Cartesian coordinates
ka; bk the maximum of a and b
Greeks
qðkÞ density of fluid/phase k
CðkÞ diffusion coefficient of fluid/phase k
UðkÞ dissipation term in energy equation of fluid/phase k
/ðkÞ general scalar quantity associated with fluid/phase k
DP ½/ðkÞ
the D operator
ðkÞ
lðkÞ ; lt laminar and turbulent viscosity of fluid/phase k
X cell volume
dt time step
Subscripts
f refers to control volume face ÔfÕ
P refers to the P grid point
Superscripts
C refers to convection contribution
D refers to diffusion contribution
ðkÞ refers to fluid/phase k
ðkÞ* refers to updated value at the current iteration
ðkÞ0 refers to values of fluid/phase k from the previous iteration
0
ðkÞ refers to correction field of phase/fluid k
m refers to fluid/phase m
old refers to values from the previous time step
552 F. Moukalled et al. / Journal of Computational Physics 190 (2003) 550–571

well as improved pressure–velocity (and density) coupling algorithms for fluid flow at all speeds [19–27]
resulting in better convergence behavior. The exponential increase in microprocessors power and the as-
sociated decrease in unit cost have also benefited the CFD industry. Multiprocessor systems with large
memory, set up at a fraction of the cost of the super computers of a decade ago, have pushed the limits on
the size of the problems that can be tackled.
Challenges still abound in a number of fields: increasing the robustness of the employed numerical
schemes, improving the accuracy of the used models, extending the many advances in the simulation of
single fluid flows [28–34] to multi-phase flows [35], are just a few areas of current research interest. For
multi-phase flow simulation, the basic difficulty [36] stems from the increased algorithmic complexity that
need to be addressed when dealing with multiple sets of continuity and momentum equations that are inter-
coupled (interchange momentum by inter-phase mass and momentum transfer, etc.) both spatially and
across fluids. That is, on top of the velocity–pressure coupling for each phasic continuity–momentum set,
there exist a number of inter-fluid coupling relations. This is further complicated in the simulation of su-
personic multi-phase flows, or in general when developing an all speed flow multi-phase algorithm. Despite
these complexities, successful segregated incompressible pressure-based solution algorithms have been
devised, such as the IPSA variants developed by the Spalding Group at Imperial College [37–39] and the set
of algorithms developed by the Los Alamos Scientific Laboratory (LASL) group [40–42]. For compressible
flow simulations at high Mach number, special treatment is needed to resolve the density–velocity and
density–pressure couplings. Algorithms for all-speed multi-phase flow simulation were recently presented
and new ones derived in [36] following a Pressure-based approach but none was implemented or tested.
Actually to the authorsÕ knowledge, no work dealing with a pressure-based method capable of predicting
multi-phase flow phenomena at all speeds has been reported in the literature. It is the objective of this
work to test a newly developed multi-phase pressure-based solution procedure that is equally valid at all
Reynolds and Mach number values.
In what follows the governing equations for compressible multi-phase flows are first presented and their
discretization outlined so as to lay the ground for the derivation of the pressure correction equation, which
is obtained from overall mass conservation. This class of algorithms is denoted as the mass conservation
based algorithms (MCBA) [36]. Then, a brief description of the solution procedure is given and the ca-
pability of the newly developed algorithm to predict multi-phase flow at all speeds demonstrated by pre-
senting solutions to four test problems spanning the entire subsonic to supersonic spectrum over a wide
range of physical conditions (from turbulent incompressible bubbly flows to supersonic air–particle flows).
The problems solved are: (i) turbulent incompressible bubbly flow in a pipe, (ii) turbulent incompressible
air–particle flow in a pipe, (iii) compressible dilute air–particle flow over a flat plate, and (iv) inviscid
transonic dusty flow in a converging-diverging nozzle. Furthermore, the accuracy of the method is dem-
onstrated by comparing results against published experimental and/or numerical data.

2. The governing equations

In multi-phase flow the various fluids/phases coexist with different concentrations at different locations in
the flow domain and move with unequal velocities. Thus, the equations governing multi-phase flows are the
conservation laws of mass, momentum, and energy for each individual fluid. For turbulent multi-phase
flow situations, an additional set of equations may be needed depending on the turbulence model used.
These equations should be supplemented by a set of auxiliary relations.
The various conservation equations are:


o rðkÞ qðkÞ

_ ðkÞ ;
þ r  rðkÞ qðkÞ uðkÞ ¼ rðkÞ M ð1Þ
ot
 F. Moukalled et al. / Journal of Computational Physics 190 (2003) 550–571 553

  i
o rðkÞ qðkÞ u ðkÞ
 
ðkÞ
 ðkÞ
þ r  rðkÞ qðkÞ uðkÞ uðkÞ ¼ r  rðkÞ lðkÞ þ lt ruðkÞ þ rðkÞ  rP þ BðkÞ þ IM ; ð2Þ
ot


o rðkÞ qðkÞ T ðkÞ

þ r  rðkÞ qðkÞ uðkÞ T ðkÞ
ot" ! #
ðkÞ     ðkÞ
ðkÞ lðkÞ lt ðkÞ rðkÞ ðkÞ ðkÞ oP
ðkÞ 
ðkÞ  ðkÞ ðkÞ I
¼r r ðkÞ
þ ðkÞ
rT þ ðkÞ
b T þ r  P u  P r  u þ U þ _
q þ EðkÞ ;
Pr Prt cP ot cP
ð3Þ
where the meanings of the various terms are as given in the nomenclature.
In addition to the above mass, momentum, and energy conservation equations (1)–(3), a geometric
conservation equation is needed for multi-phase flow. Physically, this equation is a statement indicating
that the sum of volumes occupied by the different fluids, rðkÞ , within a cell is equal to the volume of the cell
containing the fluids, and is given as
X
rðkÞ ¼ 1: ð4Þ
k

Because a static mesh is used, Eq. (4) does not include a transient term.
The effect of turbulence on interfacial mass, momentum, and energy transfer is difficult to model and is
still an active area of research. Similar to single-fluid flow, researchers have advertised several flow-de-
pendent models to describe turbulence. These models vary in complexity from simple algebraic [43] models
to state-of-the-art Reynolds-stress [44] models. In this work, the widely used two-equation k–e turbulence
model [45] with multi-phase specific modifications is adopted. The phasic conservation equations governing
the turbulence kinetic energy (k) and turbulence dissipation rate (e) for the kth fluid are given by

 !
ðkÞ
o rðkÞ qðkÞ k ðkÞ
ðkÞ ðkÞ ðkÞ ðkÞ  l t
 ðkÞ
þr r q u k ¼ r  rðkÞ ðkÞ rk ðkÞ þ rðkÞ qðkÞ GðkÞ  eðkÞ þ Ik ; ð5Þ
ot rk

 !
ðkÞ
o rðkÞ qðkÞ eðkÞ
 lt eðkÞ

þ r  rðkÞ qðkÞ uðkÞ eðkÞ ¼ r  r ðkÞ
ðkÞ
re ðkÞ
þ rðkÞ qðkÞ ðkÞ
c1e GðkÞ  c2e eðkÞ þ IeðkÞ ; ð6Þ
ot re k
ðkÞ
where Ik and IeðkÞ represent the interfacial turbulence terms. The turbulent viscosity is calculated as
 ðkÞ 2
ðkÞ ðk Þ k
lt ¼ Cl q : ð7Þ
eðkÞ
For two-phase flows, several extensions of the k–e model that are based on calculating the turbulent vis-
cosity by solving the k and e equations for the carrier or continuous phase only have been proposed in the
literature [46–51]. In a recent paper, Cokljat and Ivanov [45] presented a phase coupled k–e turbulence
model, intended for the cases where a non-dilute secondary phase is present, in which the k–e transport
equations for all phases are solved. Since the method is still not well developed, the first approach in which
only the k and e equations for the carrier phase are solved is adopted in this work.
If a typical representative variable associated with phase (k) is denoted by /ðkÞ , the above conservation
equations can be presented via the following general phasic equation:

o rðkÞ qðkÞ /ðkÞ  
þ r  rðkÞ qðkÞ uðkÞ /ðkÞ ¼ r  rðkÞ CðkÞ r/ðkÞ þ rðkÞ QðkÞ ; ð8Þ
ot
where the expression for CðkÞ and QðkÞ can be deduced from the parent equations.
554 F. Moukalled et al. / Journal of Computational Physics 190 (2003) 550–571

The presented set of differential equations has to be solved in conjunction with constraints on certain
variables represented by algebraic relations. These auxiliary relations include the equations of state and the
interfacial mass, momentum, energy, and turbulence energy transfers.
For a compressible multi-phase flow, auxiliary equations of state relating density to pressure and
temperature are needed. For the kth phase, such an equation can be written as


qðkÞ ¼ qðkÞ P ; T ðkÞ : ð9Þ

Several models have been developed for computing the interfacial mass, momentum, energy, and turbu-
lence energy transfers terms. Details regarding the closures used here are given in the results section.
In order to present a complete mathematical problem, thermodynamic relations may be needed and
initial and boundary conditions should supplement the above equations.

3. Discretization procedure

Integrating the general conservation Eq. (8) over a finite volume (Fig. 1) yields

Z Z o rðkÞ qðkÞ /ðkÞ Z Z 
dX þ r  rðkÞ qðkÞ uðkÞ /ðkÞ dX
X ot X
Z Z  Z Z
ðkÞ ðkÞ ðkÞ
¼ r  r C r/ dX þ rðkÞ QðkÞ dX; ð10Þ
X X

where X is the volume of the control cell (Fig. 1). Using the divergence theorem to transform the volume
integral into a surface integral and then replacing the surface integral by a summation of the fluxes over the
sides of the control volume, Eq. (10) is transformed to

o rðkÞ qðkÞ /ðkÞ X X 
ðkÞD ðkÞC
þ JNB þ JNB ¼ rðkÞ QðkÞ X; ð11Þ
ot NB¼e;w;n;s;t;b

Fig. 1. Control volume.

 F. Moukalled et al. / Journal of Computational Physics 190 (2003) 550–571 555

ðkÞD ðkÞC
where JNB and JNB are the diffusive and convective fluxes, respectively. The discretization of the diffusion
term is second-order accurate and follows the derivations presented in [35]. For the convective terms and
for the calculation of interface densities, the third-order SMART [13] scheme is employed and implemented
within the context of the normalized variables and space formulation (NVSF) methodology [15]. Moreover,
the integral value of the source term over the control volume is obtained by assuming the estimate of the
source at the control volume center to represent the mean value over the whole control volume. Fur-
thermore, the additional terms appearing in the momentum and energy equations, not featured in Eq. (10),
are treated explicitly and their discretization is analogous to that of the ordinary diffusion flux.
Substituting the face values by the functional relationships relating them to the neighboring node values,
Eq. (11) is transformed after some algebraic manipulations into the following discretized equation:
ðkÞ
X ðkÞ ðkÞ ðkÞ
AP /ðkÞ
p ¼ ANB /NB þ BP ; ð12Þ
NB
ðkÞ ðkÞ ðkÞ
where the coefficients AP and ANB depend on the selected scheme and BP is the source term of the dis-
cretized equation. In compact form, the above equation can be written as
h i P AðkÞ /ðkÞ þ BðkÞ
/ðkÞ
p ¼ H P / ðkÞ
¼ NB NB ðkÞNB P
: ð13Þ
AP
The discretization procedure for the momentum equation yields an algebraic equation of the form
2 3
X
 uðkÞ 0
A
uP ¼ HP uðkÞ  rðkÞ DP rP ð P Þ; where DP ¼ 4 P 5:
ðkÞ ðkÞ ðkÞ
X ð14Þ
0 vðkÞ
AP

On the other hand, the phasic mass-conservation equation (Eq. (1)) can be either viewed as a phasic volume
fraction equation

rpðkÞ ¼ HP rðkÞ : ð15Þ

or as a phasic continuity equation to be used in deriving the pressure correction equation

 old
rpðkÞ qðkÞ
p  rpðkÞ qðkÞ
p 
_ ðkÞ ;
X þ DP rðkÞ qðkÞ uðkÞ  S ¼ rðkÞ M ð16Þ
dt
where the D operator represents the following operation:
X
DP ½H
¼ Hf : ð17Þ
f ¼NBðP Þ

4. Solution procedure

The number of equations describing an n-fluid flow situation are: n-phasic momentum equations, n
phasic volume fraction (or mass conservation) equations, a geometric conservation equation, and for the
case of a compressible flow an additional n auxiliary pressure–density relations. Moreover, the variables
involved are the n-phasic velocity vectors, the n-phasic volume fractions, the pressure field, and for a
compressible flow an additional n unknown phasic density fields. In the current work, the n momentum
equations are used to calculate the n velocity fields, n  1 volume fraction (mass conservation) equations are
556 F. Moukalled et al. / Journal of Computational Physics 190 (2003) 550–571

used to calculate n  1 volume fraction fields, and the last volume fraction field calculated using the geo-
metric conservation equation
X
rðnÞ ¼ 1  rðkÞ : ð18Þ
k6¼n

The remaining volume fraction equation can be used to calculate the pressure field that is shared by all
phases. However, instead of using this last volume fraction equation, in the class of mass conservation
based algorithms (MCBA) the global conservation equation is employed, i.e., the sum of the individual
mass conservation equations, to derive a pressure correction equation as outlined next.

4.1. The pressure correction equation

To derive the pressure correction equation, the mass conservation equations of the various phases are
added to yield the global mass conservation equation given by
8  old 9
X >
< ðkÞ ðkÞ
rp qP  rp qP ðkÞ ðkÞ >
= X
ðkÞ _ ðkÞ ¼ 0:
X þ DP rðkÞ qP uðkÞ  S ¼ rðkÞ M ð19Þ
k
>
: dt >
; k

In the predictor stage a guessed or an estimated pressure field from the previous iteration, denoted by P o , is
substituted into the momentum equations. The resulting velocity fields denoted by uðkÞ which now satisfy
the momentum equations will not, in general, satisfy the mass conservation equations. Thus, corrections are
needed in order to yield velocity and pressure fields that satisfy both equations. Denoting the corrections
0 0
for pressure, velocity, and density by P 0 , uðkÞ , and qðkÞ respectively, the corrected fields are written as
 0 0
P ¼ P ° þ P 0; uðkÞ ¼ uðkÞ þ uðkÞ ; qðkÞ ¼ qðkÞo þ qðkÞ ; ð20Þ
where the superscript ‘‘o’’ refers to values from the previous iterations. Hence the equations solved in the
predictor stage are
ðkÞ  ðkÞ
uP ¼ HP ½uðkÞ
 rðkÞo DP rP P o : ð21Þ
While the final solutions satisfy
ðkÞ ðkÞ
uP ¼ HP ½uðkÞ
 rðkÞ DP rP P : ð22Þ
Subtracting the two equation sets ((22) and (21)) from each other yields the following equation involving
the correction terms:
ðkÞ 0 ðkÞ
uP ¼ HP ½uðkÞ
 rðkÞo DP rP P 0 : ð23Þ
Moreover, the new density and velocity fields, qðkÞ and uðkÞ , will satisfy the overall mass conservation
equation if
8  old 9
X> < rpðkÞo qðkÞ >
ðkÞ
P  rpðkÞ qP  =
X þ DP rðkÞo qðkÞ uðkÞ  S ¼ 0: ð24Þ
>
k :
dt >
;

Expanding the (qðkÞ uðkÞ ) term, one gets

 0
  0
    0 0  0 0
qðkÞ þ qðkÞ uðkÞ þ uðkÞ ¼ qðkÞ uðkÞ þ qðkÞ uðkÞ þ qðkÞ uðkÞ þ qðkÞ uðkÞ : ð25Þ
 F. Moukalled et al. / Journal of Computational Physics 190 (2003) 550–571 557

Substituting Eqs. (25) and (23) into Eq. (24), rearranging, and replacing density correction by pressure
correction, the final form of the pressure correction equation is written as
XX h i  ðkÞo ðkÞ
ðkÞ° ðkÞ 0 

ðkÞo ðkÞ 0 ðkÞo ðkÞ ðkÞ 0
r Cq PP þ DP r U Cq P  DP r q r D rP  S
k
dt p
8 old 9
X> >

< rpðkÞo qðkÞ
P  r ðkÞ ðkÞ
p qP  ðkÞo ðkÞ ðkÞ =
¼ X þ DP r q U : ð26Þ
k :
> dt >
;

The corrections are then applied to the velocity, pressure, and density fields using the following equations:
ðkÞ ðkÞo ðkÞ 
uP ¼ uP  rðkÞo DP rP P 0 ; P  ¼ P ° þ P 0; qðkÞ ¼ qðkÞo þ CqðkÞ P 0 : ð27Þ

Numerical experiments using the above approach to simulate air–water flows have shown poor conser-
vation of the lighter fluid. This problem can be considerably alleviated by normalizing the individual
continuity equations, and hence the global mass conservation equation, by means of a weighting factor
such as a reference density qðkÞ (which is fluid dependent). This approach has been adopted in solving all
problems presented in this work (see [36] for details).

4.2. The MCBA–SIMPLE algorithm

The overall solution procedure is an extension of the single-phase SIMPLE algorithm into multi-phase
flows. Since the pressure correction equation is derived from overall mass conservation, it is denoted by
MCBA–SIMPLE [36]. The sequence of events in the MCBA–SIMPLE is as follows:
1. Solve the phasic momentum equations for velocities.
2. Solve the pressure correction equation based on global mass conservation.
3. Correct velocities, densities, and pressure.
4. Solve the phasic mass conservation equations for volume fractions.
5. Solve the phasic scalar equations (k; e; T ; . . .).
6. Return to the first step and repeat until convergence.

5. Results and discussion

The performance of the above-described solution procedure is assessed in this section by presenting
solutions to four two-dimensional two-phase flow problems spanning the entire subsonic to supersonic
spectrum. The first two problems deal with incompressible turbulent flows while the last two problems are
concerned with compressible flows. Computed results are compared against available experimental data
and/or numerical/theoretical values. In all problems, the first phase represents the continuous phase (de-
noted by a superscript (c)), which must be fluid, and the second phase is the disperse phase (denoted by a
superscript (d)), which may be solid particles or fluid. Unless otherwise specified the third-order SMART
scheme is used in all computations reported in this study.

5.1. Problem 1: Turbulent upward bubbly flow in a pipe

The problem considered involves the prediction of radial phase distribution in turbulent upward air–
water flow in a pipe. Many experimental and numerical studies addressing this problem have appeared in
the literature [52–60]. Most of these studies have indicated that the lateral forces that most strongly affect
558 F. Moukalled et al. / Journal of Computational Physics 190 (2003) 550–571

the void distribution are the turbulent stresses and the lateral lift force. As such, in addition to the usual
drag force, the lift force is considered as part of the interfacial force terms in the momentum equations. In
the present work, the interfacial drag forces per unit volume are given by

x ðcÞ
ðdÞ CD ðcÞ ðdÞ ðcÞ

IM D ¼  IMx D ¼ 0:375 q r r Vslip uðdÞ  uðcÞ ; ð28Þ
rp

ðcÞ ðdÞ CD ðcÞ ðdÞ ðcÞ


ðIMy ÞD ¼ ðIMy ÞD ¼ 0:375 q r r Vslip vðdÞ  vðcÞ ; ð29Þ
rp
where rp is the bubble radius. The drag coefficient CD varies as a function of the bubble Reynolds and
Weber numbers defined as

rp rp2
Rep ¼ 2 V
ðcÞ slip
We ¼ 4qðcÞ Vslip ; ð30Þ
ml r

where r, the surface tension, is given a value of 0.072 N/m for air–water systems. The following correla-
tions, which take the shape of the bubble into consideration, are utilized [61,62]:
8 16
>
> CD ¼ Re for Rep < 0:49;
>
>
p
>
> CD ¼ Re20 for 0:49 < Rep < 100;
>
> 0:643
< p
6:3
CD ¼ Re0:385 for Rep  100; ð31Þ
>
>
p
>
> CD ¼ 83 for Rep  100 and We > 8;
>
>
>
>
:
CD ¼ We3
for Rep  100 and Rep > 2065:1=We2:6 :

Many investigators have considered the modeling of lift forces [61–65]. Based on their work, the following
expressions are employed for the calculation of the interfacial lift forces per unit volume:

ðdÞ 

ðIM ÞðcÞ ðdÞ ðcÞ ðdÞ
L ¼ ðIM ÞL ¼ C1 q r u  uðcÞ  r  uðcÞ ; ð32Þ

where C1 is the interfacial lift coefficient calculated from


C1 ¼ C1a 1  2:78 0:2; rðdÞ ; ð33Þ

where ha; bi denotes the minimum of a and b and C1a is an empirical constant.
Besides the drag and lift interfacial forces, the effect of bubbles on the turbulent field is very important,
as the distribution of bubbles affects the turbulence field in the liquid phase and at the same time the liquid
phaseÕs turbulence is influenced by the bubbles. In this work, turbulence is assumed to be a property of the
continuous liquid phase (c) and the turbulent kinematics viscosity of the dispersed air phase (d) is assumed
to be a function of that of the continuous phase. The turbulent viscosity of the continuous phase is
computed by solving the following modified transport equations for the turbulent kinetic energy k and its
dissipation rate e that take into account the interaction between the phases:

 " ! #
ðcÞ
o rðcÞ qðcÞ k ðcÞ
ðcÞ ðcÞ ðcÞ ðcÞ  ðcÞ ðcÞ ðcÞ mt ðcÞ


þr r q u k ¼r r q ml þ ðcÞ rk þ rðcÞ qðcÞ GðcÞ  eðcÞ
ot rk
" ! #
ðcÞ
m t
þ r  qðcÞ k ðcÞ rrðcÞ þ rðcÞ Pb ; ð34Þ
rr
 F. Moukalled et al. / Journal of Computational Physics 190 (2003) 550–571 559

 " ! #
ðcÞ
o rðcÞ qðcÞ eðcÞ
 ðcÞ m eðcÞ
þ r  rðcÞ qðcÞ uðcÞ eðcÞ ¼ r  rðcÞ qðcÞ ml þ tðcÞ reðcÞ þ rðcÞ c1e Pb
ot re k ðcÞ
" ! #
ðcÞ
ðcÞ ðcÞ eðcÞ
ðcÞ ðcÞ
 mt
þr q c1e G  c2e e þ r  qðcÞ ðcÞ
e rr ðcÞ
; ð35Þ
k ðcÞ rr

where GðcÞ is the well-known volumetric production rate of k ðcÞ by shear forces, rr the turbulent Schmidt
number for volume fractions, and Pb is the production rate of k ðcÞ by drag due to the motion of the bubbles
through the liquid and is given by

0:375Cb CD qðcÞ rðdÞ rðcÞ Vslip

2
Pb ¼ : ð36Þ
rp

In Eq. (36) Cb is an empirical constant representing the fraction of turbulence induced by bubbles that goes
into large-scale turbulence of the liquid phase. Moreover, as suggested in [60], the flux representing the
interaction between the fluctuating velocity and volume fraction is modeled via a gradient diffusion ap-
proximation and added as a source term in the continuity ðr  ðqðkÞ DðkÞ rrðkÞ ÞÞ and momentum
ðr  ðqðkÞ DðkÞ uðkÞ rrðkÞ ÞÞ equations with the diffusion coefficient D given by
ðkÞ
mt
DðkÞ ¼ : ð37Þ
rr
The turbulent kinematics eddy viscosity of the dispersed and continuous phases are related through

ðcÞ
ðdÞ mt
mt ¼ ; ð38Þ
rf

where rf is the turbulent Schmidt number for the interaction between the two phases. The above
described turbulence model is a modified version of the one described in [60] in which the turbulent
kinematics viscosities of both phases are allowed to be different in contrast to what is done in [60]. This
is accomplished through the introduction of the rf parameter. As such, different diffusion coefficients
(DðkÞ ) are used for the different phases. Results are compared to experimental data from [52,63].
Two experiments were simulated using the above-described treatment and results compared to ex-
perimental data. The two experiments differ in the Reynolds number, the bubbles diameter and the inlet
conditions. In the experiment of Seriwaza et al. [52] the Reynolds number based on superficial liquid
velocity and pipe diameter is 8  104 , the inlet superficial gas and liquid velocities are 0.077 and 1.36 m/
s, respectively, and the inlet void fraction is 5.36  102 with no slip between the incoming phases.
Moreover, the bubble diameter is taken as 3 mm [60], while the fluid properties are taken as qðcÞ ¼ 1000
ðcÞ
kg/m3 , qðdÞ ¼ 1:23 kg/m3 , and ml ¼ 106 m2 /s. In the experiment of Lahey et al. [63] the Reynolds
number is based on superficial liquid velocity and pipe diameter of 5  104 , the inlet superficial gas and
liquid velocities are 0.1 and 1.08 m/s, respectively. Both problems are solved using the same values for
all constants in the model with C1a ¼ 0:075, rf ¼ 0:5, rr ¼ 0:7, and Cb ¼ 0:05. Predicted radial profiles
of the vertical liquid velocity and void fraction presented in Figs. 2(a) and (b) using a grid of size
96  32 control volumes concur very well with measurements and compare favorably with numerical
profiles reported by Boisson and Malin [60] (Fig. 2(a)) and PHOENICS [65] (Fig. 2(b)). As shown, the
void fraction profile indicates that gas is taken away from the pipe center. This is caused by the lift
force, which drives the bubbles towards the wall.
560 F. Moukalled et al. / Journal of Computational Physics 190 (2003) 550–571

Fig. 2. Comparison of fully developed liquid velocity and void fraction profiles for turbulent bubbly upward bubbly flow in a pipe: (a)
Seriwaza et al. data; (b) Lahey et al. data.

5.2. Problem 2: Turbulent air–particle flow in a vertical pipe

In problem 2, the upward flow of a dilute gas–solid mixture in a vertical pipe is simulated. As in the
previous problem, the axi-symmetric form of the gas and particulate transport equations are employed. The
effects of interfacial virtual mass and lift forces are small and may be neglected, as reported in several
studies [66–68], and the controlling interfacial force is drag (see [69]). Denoting the continuous and dis-
persed phases by superscripts (c) and (d), respectively, the drag in the x- and y-momentum equations are
given by
 F. Moukalled et al. / Journal of Computational Physics 190 (2003) 550–571 561

x ðcÞ
ðdÞ 3 CD ðcÞ ðdÞ


IM D ¼  IMx D ¼ q r Vslip uðdÞ  uðcÞ ; ð39Þ
8 rp

ðcÞ ðdÞ 3 CD ðcÞ ðdÞ


ðIMy ÞD ¼ ðIMy ÞD ¼ q r Vslip vðdÞ  vðcÞ ; ð40Þ
8 rP
where rP represents the particleÕs radius, CD the drag coefficient computed from
8 24
< CD ¼ Rep
>

for Rep < 1;
24
CD ¼ Rep 1 þ 0:15Rep 0:687
for 1 < Rep < 1000; ð41Þ
>
:
CD ¼ 0:44 for Rep > 1000;
and Rep the Reynolds number based on the particle size as defined in Eq. (30).
As before, turbulence is assumed to be a property of the continuous gas phase (c) and the turbulent
kinematics viscosity of the dispersed particle phase (d) is assumed to be a function of that of the continuous
phase. Again, turbulence modulation due to the presence of particles is predicted using a two-phase k–e
model. Several extensions of the k–e model for carrier-phase turbulence modulation have been proposed in
the literature [46–51] and the modification of Chen and Wood [48], which introduces additional source
terms into the turbulence transport equations, is adopted here. These source terms are always negative and
act to reduce k and e. However, depending on the relative extent of reductions in k and e, the turbulent
viscosity may be either reduced or increased by the presence of particles. Thus, the turbulent viscosity is
ðcÞ
computed by solving the turbulence transport equations (Eqs. (5) and (6)) for the continuous phase with Ik
and IeðcÞ evaluated using the following relations suggested by Chen and Wood [48]:
ðcÞ k ðcÞ

Ik ¼ 2qðdÞ rðcÞ rðdÞ 1  e0:0825ðsp =se Þ ; ð42Þ
sp

eðcÞ
IeðcÞ ¼ 2qðdÞ rðcÞ rðdÞ ; ð43Þ
sp
where sp and se are timescales characterizing the particle response and large-scale turbulent motion, re-
spectively, and are computed from

qðdÞ rðdÞ k ðcÞ

sp ¼ Vslip ; se ¼ 0:165 ; ð44Þ
FD eðcÞ
where FD is the magnitude of the inter-phase drag force per unit volume. The turbulent kinematics eddy
viscosity of the dispersed phase is found using Eq. (38).
The model is validated against the experimental results of Tsuji et al. [66]. In their experiments, the
vertical pipe has an internal diameter of 30.5 mm. Results are replicated here for the case of an air Reynolds
number, based on the pipe diameter, of 3.3  104 and a mean air inlet velocity of 15.6 m/s using particles of
diameter 200 lm and density 1020 kg/m3 . In the computations, the mass-loading ratio at inlet is considered
to be 1 with no slip between the phases, and rf and rr are set to 5 and 1010 (i.e., the interaction terms
included for bubbly flows are neglected here), respectively. Fig. 3 shows the fully developed gas and par-
ticles mean axial velocity profiles generated using a grid of size 96  40 CV. It is evident that there is
generally a very good agreement between the predicted and experimental data with the gas velocity being
slightly over predicted and the particles velocity slightly under predicted. Moreover, close to the wall, the
model predictions indicate that the particles have higher velocities than the gas, which is in accord with the
experimental results of Tsuji et al. [66].
562 F. Moukalled et al. / Journal of Computational Physics 190 (2003) 550–571

Fig. 3. Comparison of fully developed gas and particle velocity profiles for turbulent air–particle flow in a pipe.

5.3. Problem 3: Compressible dilute air–particle subsonic flow over a flat plate

This is a well-studied problem [66–76] suitable as a benchmark test. It is known that two-phase flow
greatly changes the main features of the boundary layer over a flat plate. Typically, three regions are de-
fined in the two-phase boundary layer (Fig. 4), based on the importance of the slip velocity between the two
phases: a large-slip region close to the leading edge, a moderate-slip region further down, and a small-slip
one far downstream. The characteristic scale in this two-phase problem is the relaxation length ke [73],
defined as

2 qðdÞ rp2 u1
ke ¼ ; ð45Þ
9 lðcÞ
where qðdÞ and rp are, respectively, the density and radius of the particles, lðcÞ the viscosity of the fluid, and
u1 the free stream velocity. The three regions are defined according to the order of magnitude of the slip
parameter x ¼ x=ke . In the simulation, the viscosity of the fluid is considered to be a function of tem-
perature and varies according to [73]
0:6
T ðcÞ
lðcÞ ¼ lref ; ð46Þ
Tref
where the reference viscosity and temperature are lref ¼ 1.86  105 N s/m2 and Tref ¼ 303 K, respectively.
Even though variations in gas density are small under the conditions considered, these variations are not
neglected and the flow is treated as compressible for the continuous phase and as incompressible for the
dispersed phase. Moreover, drag is the only interfacial force retained due to its dominance over other
interfacial forces. Denoting the continuous and dispersed phases by superscripts (c) and (d), respectively,
this force is computed as [73]
 F. Moukalled et al. / Journal of Computational Physics 190 (2003) 550–571 563

Fig. 4. The three different regions within the boundary layer of dusty flow over a flat plate.

x ðcÞ
ðdÞ 9 CD ðdÞ ðcÞ
ðdÞ 

IM D ¼  IMx D ¼ r l u  uðcÞ ; ð47Þ
2 rp2

ðcÞ ðdÞ 9 CD ðdÞ ðcÞ
ðdÞ 

ðIMy ÞD ¼ ðIMy ÞD ¼ 2
r l v  vðcÞ ; ð48Þ
2 rp

where the drag coefficient is given by

1 7
CD ¼ Rep þ Re0:15 : ð49Þ
50 6 p
In the energy equation, heat transfer due to radiation is neglected and only convective heat transfer around
an isolated particle is considered. Moreover, the particles have no individual random motion, mutual
collisions, and other interactions among them. Therefore, only the process of drag and heat transfer couple
the particles with the gas. Under such conditions, the interfacial terms in the gas (continuous phase) and
particles (dispersed phase) energy equations reduce to [73]
ðcÞ
IE ¼ Qgp þ Fgp :uðdÞ ; ð50Þ

ðdÞ
IE ¼ Qgp ; ð51Þ
where

ðcÞ ðcÞ
Fg–p ¼ IMx D i þ ðIMy ÞD j; ð52Þ

1=3
Nu ¼ 2:0 þ 0:6Rep1=2 PrðcÞ ; ð53Þ

3 rðdÞ kðcÞ Nu
ðdÞ 

Qg–p ¼ T  T ðcÞ : ð54Þ
2 rp2
In the above equations, Nu is the Nusselt number, PrðcÞ the gas Prandtl number, kðcÞ the gas thermal
conductivity, T the temperature, and other parameters are as defined earlier.
In the simulation, the particle diameter is chosen to be 10 lm, the particle Reynolds number is assumed
to be equal to 10, the material density is 1766 kg/m3 , and the Prandtl number is set to 0.75. The south
boundary (wall) is treated as a no-slip wall boundary for the gas phase (i.e., both components of the gas
velocity are set to zero), while the particle phase encounters slip wall conditions (i.e., the normal fluxes are
set to zero). The gas and the particles enter the computational domain under thermal and dynamical
equilibrium conditions. A mass load ratio of 1 between the particles phase and the gas phase is used.
Results are displayed using the following dimensionless variables in order to bring all quantities to the same
order of magnitudes
564 F. Moukalled et al. / Journal of Computational Physics 190 (2003) 550–571

x y pffiffiffiffiffiffi u v pffiffiffiffiffiffi quke

x ¼ ; y ¼ Re; u ¼ ; v ¼ Re; Re ¼ : ð55Þ
ke ke u1 u1 l
Fig. 5 shows the results for the steady flow obtained on a rectangular domain with a mesh of density
104  48 CV. stretched in the y-direction. The figure provides the development of gas and particles velocity
profiles within the three regions mentioned earlier. In the near leading edge area (x ¼ 0:1), the gas velocity
is adjusted at the wall to obtain the no-slip condition as for the case of a pure gas boundary layer. The
particles have no time to adjust to the local gas motion and there is a large velocity slip between the phases.
In the transition region (x ¼ 1), significant changes in the flow properties take place. The interaction
between the phases causes the particles to slow down and the gas to accelerate as apparent in the plots. In
the far downstream region (x ¼ 5), the particles have enough time to adjust to the state of the gas motion.
The slip is very small and the solution tends to equilibrium. These results are in excellent agreement with the
numerical solutions reported by Thevand et al. [76] plotted in Fig. 5, validating the proposed methodology.

5.4. Problem 4: Inviscid transonic dusty flow in a converging–diverging nozzle

As a final test for the newly suggested numerical procedure, dilute two-phase transonic flow in an axi-
symmetric converging-diverging rocket nozzle is considered. Several researchers have analyzed the problem
and data is available for comparison [77–86]. In most of the reported studies, a shorter diverging section, in
comparison with the one considered here, has been used when predicting the two-phase flow. Two-phase
results for the long configuration have only been reported by Chang et al. [81]. The flow is assumed to be
inviscid and the single-phase results are used as an initial guess for solving the two-phase problem. The
physical configuration (Fig. 6) is the one described in [81]. The viscosity of the fluid, which is solely used in
the calculation of the interfacial drag force, varies with the temperature according to SutherlandÕs law for
air

Fig. 5. Comparison of fully developed gas and particle velocity profiles inside the boundary layer at different axial locations for dilute
two-phase flow over a flat plate.
 F. Moukalled et al. / Journal of Computational Physics 190 (2003) 550–571 565

Fig. 6. Physical domain for the dusty gas flow in a converging–diverging nozzle.

pffiffiffiffiffiffiffiffi
ðcÞ 6 T ðcÞ T ðcÞ
l ¼ 1:458  10 : ð56Þ
T ðcÞ þ 110:4
The coupling between gas and particle phases is through the interfacial momentum and energy terms. The
force exerted on a single particle moving through a gas is given as [82]


fx ¼ 6prp fD lðcÞ uðdÞ  uðcÞ ; ð57Þ


fy ¼ 6prp fD lðcÞ vðdÞ  vðcÞ ; ð58Þ

so that for N particles in a unit volume the effective drag force is

x ðcÞ
ðdÞ 9 rðdÞ

IM D ¼  IMx D ¼ fD lðcÞ uðdÞ  uðcÞ ; ð59Þ
2 rp2

ðcÞ ðdÞ 9 rðdÞ


ðIMy ÞD ¼ ðIMy ÞD ¼ 2
fD lðcÞ vðdÞ  vðcÞ ; ð60Þ
2 rp

where fD is the ratio of the drag coefficient CD to the stokes drag CD0 ¼ 24/Rep and is given by [81]

0:0175Rep
fD ¼ 1 þ 0:15Re0:687
p þ ; Rep < 3  105 : ð61Þ
1 þ 4:25  104 Rep1:16

The heat transferred from gas to particle phase per unit volume is given as [82]
3 rðdÞ ðcÞ
ðdÞ 
Qg–p ¼ k Nu T  T ðcÞ ; ð62Þ
2 rp
where kðcÞ is the thermal conductivity of the gas and Nu, the Nusselt number, is written as [82]
Nu ¼ 2 þ 0:459Re0:55 0:33
p Prc : ð63Þ
566 F. Moukalled et al. / Journal of Computational Physics 190 (2003) 550–571

Fig. 7. (a,b) Volume fraction contours and (c,d) particle velocity vectors for dusty gas flow in a converging–diverging nozzle.

The gas–particle inter-phase energy term is given by

ðcÞ 9 rðdÞ ðcÞ

ðdÞ ðcÞ
 9 rðdÞ

ðcÞ ðdÞ ðcÞ
 3 rðdÞ ðcÞ
ðdÞ 
IE ¼ fD l u  u u d þ f D l v  v v d þ k Nu T  T ðcÞ ; ð64Þ
2 rp2 2 rp2 2 rp

ðdÞ 3 rðdÞ ðcÞ
ðcÞ 

IE ¼ k Nu T  T ðdÞ ; ð65Þ
2 rp
where the first two terms on the right-hand side of Eq. (64) represent the energy exchange due to mo-
mentum transfer.
 F. Moukalled et al. / Journal of Computational Physics 190 (2003) 550–571 567

Fig. 8. Comparison of one-phase and two-phase gas Mach number distributions along the (a) wall and (b) centerline of the dusty flow
in a converging–diverging nozzle problem.

The physical quantities employed are similar to those used in [81]. The gas stagnation temperature and
pressure at inlet to the nozzle are 555 K and 10.34  105 N/m2 , respectively. The specific heat for the gas and
particles are 1.07  103 J/kg K and 1.38  103 J/kg K, respectively, and the particle density is 4004.62 kg/m3 .
With a zero inflow velocity angle, the fluid is accelerated from subsonic to supersonic speed in the nozzle.
The inlet velocity and temperature of the particles are presumed to be the same as those of the gas phase.
Results for two particle sizes of radii 1 and 10 lm with the same mass fraction / ¼ 0:3 are presented using a
568 F. Moukalled et al. / Journal of Computational Physics 190 (2003) 550–571

grid of size 188  80 CV. Figs. 7(a) and (b) show the particle volume fraction contours while Figs. 7(c) and
(d) display the velocity distribution. For the flow with particles of radius 1 lm, a sharp change in particle
density is obtained near the upper wall downstream of the throat, and the particle density decreases to a
small value. With the large particle flow (10 lm), however, a much larger particle-free zone appears due to
the inability of the heavier particles to turn around the throat corner. These findings are in excellent
agreement with published results reported in [81] and others using different methodologies. In addition the
contour lines are similar to those reported by Chang et al. [81]. A quantitative comparison of current
predictions with published experimental and numerical data is presented in Fig. 8 through gas Mach
number distributions along the wall (Fig. 8(a)) and centerline (Fig. 8(b)) of the nozzle for the one-phase and
two-phase flow situations with particles of radii 10 lm. As can be seen, the one-phase predictions fall on top
of experimental data reported in [84–86]. Along the centerline of the nozzle, current predictions are of
quality better than those obtained by Chang et al. [81]. Since the nozzle contour has a rapid contraction
followed by a throat with a small radius of curvature, the flow near the throat wall is overturned and
inclined to the downstream wall. A weak shock is thus formed to turn the flow parallel to the wall. This
results in a sudden drop in the Mach number value and as depicted in Fig. 8(b), this sudden drop is
correctly envisaged by the solution algorithm with the value after the shock being slightly over predicted.
Due to the unavailability of two-phase flow data, predictions are compared against the numerical results
reported in [81]. As displayed in Figs. 8(a) and (b), both solutions are in good agreement with each other
indicating once more the correctness of the calculation procedures. The lower gas Mach number in the two-
phase flow is caused by the heavier particles (qðdÞ  qðcÞ ), which reduce the gas velocity. Moreover, owing to
the particle-free zone, the Mach number difference between the one- and two-phase flows along the wall is
smaller than that at the centerline.

6. Closing remarks

A new finite volume-based numerical procedure for the calculation of multi-phase flows at all speeds was
presented. The virtues of the method were demonstrated by solving four two-phase flow problems spanning
the entire subsonic to supersonic spectrum over a wide range of physical conditions: turbulent bubbly flow
in a pipe, turbulent air–particle flow in a pipe, subsonic compressible air–particle flow over a flat plate, and
transonic dusty flow in a converging diverging nozzle. Results generated were compared against experi-
mental and/or numerical simulation data where available. The accuracy of the predicted quantities, which
was shown to be similar or better than that obtained with special purpose methods, was a clear demon-
stration of the effectiveness of the new method as a tool for modeling multi-phase flows at all speeds.

Acknowledgements

The financial support provided by the European Office of Aerospace Research and Development
(EOARD) (SPC00-4071) is gratefully acknowledged.

References

[1] A. Brandt, Multi-level adaptive solutions to boundary-value problems, Math. Comp. 31 (1977) 333–390.
[2] C.M. Rhie, A pressure based Navier–Stokes solver using the multigrid method, AIAA Paper 86-0207, 1986.
[3] P. Wesseling, An Introduction to Multigrid Methods, Wiley, Baffins Lane, Chichester, West Sussex PO19 1UD, UK, 1995.
[4] E. Dick, Multigrid formulation of polynomial flux-difference splitting for steady Euler equations, J. Comput. Phys. 91 (1990) 161–
173.
 F. Moukalled et al. / Journal of Computational Physics 190 (2003) 550–571 569

[5] D. Kershaw, The incomplete Cholesky-conjugate gradient method for the iterative solution of systems of linear equations,
J. Comput. Phys. 26 (1978) 43–65.
[6] H.L. Stone, Iterative solution of implicit approximations of multidimensional partial differential equations, SIAM J. Numer.
Anal. 5 (3) (1968) 530–558.
[7] V. Venkatakrishnan, D.J. Mavriplis, Implicit method for the computation of unsteady flows on unstructured grids, AIAA Paper
95-1705, 1995.
[8] V. Venkatakrishnan, Perspective on unstructured grid solvers, AIAA J. 34 (3) (1996) 533–547.
[9] D.L. Whitaker, Three dimensional unstructured grid Euler computations using a fully-implicit, upwind method, AIAA Paper 93-
3357, 1993.
[10] D. Oan, J.C. Cheng, Upwind finite volume Navier–Stokes computations on unstructured triangular meshes, AIAA J. 31 (9) (1993)
1618–1625.
[11] C.R. Mitchell, Improved reconstructuion schemes for the Navier–Stokes equations on unstructured grids, AIAA Paper 94-0642,
1994.
[12] T.J. Barth, D.C. Jespersen, The design and application of upwind schemes on unstructured meshes, AIAA Paper 89-0366, January
1989.
[13] P.H. Gaskell, A.K.C. Lau, Curvature compensated convective transport: SMART, a new boundedness preserving transport
algorithm, Int. J. Numer. Methods Fluids 8 (1988) 617–641.
[14] B.P. Leonard, Locally modified quick scheme for highly convective 2-D and 3-D flows, in: C. Taylor, K. Morgan (Eds.),
Numerical Methods in Laminar and Turbulent Flows, 15, Pineridge Press, Swansea, UK, 1987, pp. 35–47.
[15] M.S. Darwish, F. Moukalled, Normalized variable and space formulation methodology for high-resolution schemes, Numer. Heat
Transfer, Part B 26 (1994) 79–96.
[16] F. Moukalled, M.S. Darwish, A new family of streamline-based very high resolution schemes, Numer. Heat Transfer 32 (3) (1997)
299–320.
[17] M. Darwish, F. Moukalled, An efficient very high-resolution scheme based on an adaptive-scheme strategy, Numer. Heat
Transfer, Part B 34 (1998) 191–213.
[18] F. Moukalled, M. Darwish, New family of adaptive very high resolution schemes, Numer. Heat Transfer, Part B 34 (1998) 215–
239.
[19] C.H. Marchi, C.R. Maliska, A non-orthogonal finite-volume methods for the solution of all speed flows using co-located
variables, Numer. Heat Transfer, Part B 26 (1994) 293–311.
[20] I. Demirdzic, Z. Lilek, M. Peric, A collocated finite volume method for predicting flows at all speeds, Int. J. Numer. Methods
Fluids 16 (1993) 1029–1050.
[21] F.S. Lien, M.A. Leschziner, A general non-orthogonal collocated finite volume algorithm for turbulent flow at all speeds
incorporating second-moment turbulence-transport closure, Part 1: computational implementation, Comput. Methods Appl.
Mech. Eng. 114 (1994) 123–148.
[22] E.S. Politis, K.C. Giannakoglou, A pressure-based algorithm for high-speed turbomachinery flows, Int. J. Numer. Methods Fluids
25 (1997) 63–80.
[23] K.H. Chen, R.H. Pletcher, Primitive variable, strongly implicit calculation procedure for viscous flows at all speeds, AIAA J. 29
(8) (1991) 1241–1249.
[24] M. Darbandi, G.E. Shneider, Momentum variable procedure for solving compressible and incompressible flows, AIAA J. 35 (12)
(1997) 1801–1805.
[25] M. Darbandi, G.E. Schneider, Use of a flow analogy in solving compressible and incompressible flows, AIAA Paper 97-2359,
January 1997.
[26] K.C. Karki, A calculation procedure for viscous flows at all speeds in complex geometries, Ph.D. Thesis, University of Minnesota,
June 1986.
[27] F. Moukalled, M. Darwish, A high-resolution pressure-based algorithm for fluid flow at all speeds, J. Comput. Phys. 168 (1)
(2001) 101–133.
[28] S.V. Patankar, D.B. Spalding, A calculation procedure for heat, mass and momentum transfer in three dimensional parabolic
flows, Int. J. Heat Mass Transfer 15 (1972) 1787.
[29] S.V. Patankar, Numerical Heat Transfer and Fluid Flow, Hemisphere, Washington DC, 1981.
[30] R.I. Issa, Solution of the implicit discretized fluid flow equations by operator splitting, Mechanical Engineering Report, FS/82/15,
Imperial College, London, 1982.
[31] J.P. Van Doormaal, G.D. Raithby, Enhancement of the SIMPLE method for predicting incompressible fluid flows, Numer. Heat
Transfer 7 (1984) 147–163.
[32] J.P. Van Doormaal, G.D. Raithby, An evaluation of the segregated approach for predicting incompressible fluid flows, ASME
Paper 85-HT-9, Presented at the National Heat Transfer Conference, Denver, CO, August 4–7, 1985.
[33] S. Acharya, F. Moukalled, Improvements to incompressible flow calculation on a non-staggered curvilinear grid, Numer. Heat
Transfer, Part B 15 (1989) 131–152.
570 F. Moukalled et al. / Journal of Computational Physics 190 (2003) 550–571

[34] C.R. Maliska, G.D. Raithby, Calculating 3-D fluid flows using non-orthogonal grid, Proc. Third Int. Conf. on Numerical
Methods in Laminar and Turbulent Flows, Seattle, 1983, pp. 656–666.
[35] F. Moukalled, M. Darwish, A unified formulation of the segregated class of algorithms for fluid flow at all speeds, Numer. Heat
Transfer, Part B 37 (1) (2000) 103–139.
[36] M. Darwish, F. Moukalled, B. Sekar, A unified formulation of the segregated class of algorithms for multi-fluid flow at all speeds,
Numer. Heat Transfer, Part B 40 (2) (2001) 99–137.
[37] D.B. Spalding, The Calculation of free-convection phenomena in gas-liquid mixtures, Report HTS/76/11, Mech. Eng. Imperial
College, London, 1976.
[38] D.B. Spalding, Numerical computation of multi-phase fluid flow and heat transfer, in: C. Taylor, K. Morgan (Eds.), Recent
Advances in Numerical Methods in Fluid, 1, 1980, pp. 139–167.
[39] D.B. Spalding, A general purpose computer program for multi-dimensional, one and two phase flow, Report HTS/81/1, Mech.
Eng. Imperial College, London, 1981.
[40] W.W. Rivard, M.D. Torrey, KFIX: A program for transient two dimensional two fluid flow, Report LA-NUREG-6623, 1978.
[41] A.A. Amsden, F.H. Harlow, KACHINA: An Eulerian computer program for multifield flows, Report LA-NUREG-5680, 1975.
[42] A.A. Amsden, F.H. Harlow, KTIFA two-fluid computer program for down comer flow dynamics, Report LA-NUREG-6994,
1977.
[43] B.S. Baldwin, H. Lomax, Thin Layer approximation and algebraic model for separated turbulent flows, AIAA Paper 78-257,
1978.
[44] F. Sotiropoulos, V.C. Patel, Application of Reynolds-stress transport models to stern and wake flow, J. Ship Res. 39 (1995) 263.
[45] D. Cokljat, V.A. Ivanov, F.J. Srasola, S.A. Vasquez, Multiphase k–e models for unstructured meshes, ASME 2000 Fluids
Engineering Division Summer Meeting, June 11–15, 2000, Boston, MA, USA.
[46] F. Pourahmadi, J.A.C. Humphrey, Modeling solid–fluid turbulent flows with application to predicting erosive wear, Int. J. Phys.
Chem. Hydro. 4 (1983) 191–219.
[47] S.E. Elghobashi, T.W. Abou-Arab, A two-equation turbulence model for two-phase flows, Phys. Fluids 26 (4) (1983) 931–938.
[48] C.P. Chen, P.E. Wood, Turbulence closure modeling of the dilute gas–particle axisymmetric jet, AIChE J. 32 (1) (1986) 163–
166.
[49] M. Lopez de Bertodano, S.J. Lee, R.T. Lahey Jr., D.A. Drew, The prediction of two-phase turbulence and phase distribution
phenomena using a Reynolds stress model, ASME J. Fluids Eng. 112 (1990) 107–113.
[50] M. Lopez de Bertodano, R.T. Lahey Jr., O.C. Jones, Development of a k–e model for bubbly two-phase flow, ASME J. Fluids
Eng. 116 (1994) 128–134.
[51] M. Lopez de Bertodano, R.T. Lahey Jr., O.C. Jones, Phase distribution in bubbly two-phase flow in vertical ducts, Int.
J. Multiphase flow 20 (5) (1994) 805–818.
[52] A. Serizawa, I. Kataoka, I. Michiyoshi, Phase distribution in bubbly flow, Data set No. 24, Proc. Second Int. Workshop on Two-
Phase Flow Fundamentals, Rensselaer Polytechnic Institute, Troy, NY, 1986.
[53] S.K. Wang, S.-j. Lee, O.C. Jones Jr., R.T. Lahey Jr., 3-D turbulence structure and phase distribution measurements in bubbly
two-phase flows, Int. J. Multiphase Flow 13 (3) (1987).
[54] S.P. Antal, R.T. Lahey, J.E. Flaherty, Analysis of phase distribution in fully developed laminar bubbly two-phase flows, Int.
J. Multiphase Flow 17 (5) (1991) 635–652.
[55] Y. Sato, M. Sadatomi, K. Sekoguchi, Momentum and heat transfer in two-phase bubble flow, in: International Journal of
Multiphase Flow, I. Theory, pp. 167–177. II. A Comparison Between Experiment and Theoretical Calculations, 1981, pp. 179–
190.
[56] M. Lopez de Bertodano, S.-J. Lee, R.T. Lahey Jr., D.A. Drew, The prediction of two-phase turbulence and phase distribution
phenomena using a Reynolds stress model, J. Fluids Eng. 112 (1990) 107–113.
[57] M. Lopez de Bertodano, R.T. Lahey Jr., O.C. Jones, Development of a k–e model for bubbly two-phase flow, J. Fluids Eng. 116
(1994) 128–134.
[58] M. Lopezde Bertodano, R.T. Lahey Jr., O.C. Jones, Phase distribution in bubbly two-phase flow in vertical ducts, Int.
J. Multiphase Flow 20 (5) (1994) 805–818.
[59] V.E. Nakoryakov, O.N. Kashinsky, V.V. Randin, L.S. Timkin, Gas–liquid bubbly flow in vertical pipes, J. Fluids Eng. 118 (1996)
377–382.
[60] N. Boisson, M.R. Malin, Numerical prediction of two-phase flow in bubble columns, Int. J. Numer. Methods Fluids 23 (1996)
1289–1310.
[61] B. Huang, Modelisation Numerique DÕecoulements Diphasiques a Bulles dans les Reacteurs Chimiques, Ph.D. Thesis, Universite
Claude Bernard, Lyon, 1989.
[62] K.O. Peterson, Etude Experimentale et Numerique des Ecoulements Diphasiques dans les Reacteurs Chimiques, Ph.D. Thesis,
Universite Claude Bernard, Lyon, 1992.
[63] R.T. Lahey, M. Lopez de Bertodano, O.C. Jones, Phase distribution in complex geometry ducts, Nuclear Eng. Design 141 (1993)
177.
 F. Moukalled et al. / Journal of Computational Physics 190 (2003) 550–571 571

[64] D.A. Drew, T.J. Lahey Jr., The virtual mass and lift force on a sphere in rotating and straining inviscid flow, Int. J. Multiphase
Flow 13 (1) (1987) 113.
[65] PHOENICS: ‘‘http://www.cham.co.uk/phoenics/d_polis/d_applic/d_flows/laheya.htm’’.
[66] Y. Tsuji, Y. Morikawa, H. Shiomi, LDV measurements of an air–solid two-phase flow in a vertival pipe, J. Fluid Mech. 139 (1984)
417–434.
[67] A. Adeniji-Fashola, C.P. Chen, Modeling of confined turbulent fluid-particle flows using Eulerian and Lagrangian schemes, Int.
J. Heat Mass Transfer 33 (1990) 691–701.
[68] S. Naik, I.G. Bryden, Prediction of turbulent gas–solids flow in curved ducts using the Eulerian–Lagrangian method, Int.
J. Numer. Methods Fluids 31 (1999) 579–600.
[69] F.H. Harlow, A.A. Amsden, Numerical calculation of multiphase fluid flow, J. Comput. Phys. 17 (1975) 19–52.
[70] A.N. Osiptsov, Structure of the laminar boundary layer of a disperse medium on a flat plate, Fluid Dynamics 15 (1980) 512–517.
[71] S. Prabha, A.C. Jain, On the use of compatibility conditions in the solution of gas particulate boundary layer equations, Appl. Sci.
Res. 36 (1980) 81–91.
[72] R.E. Sgleton, The incompressible gas solid particle flows over a semi-infinite flat plate, Z. Angew. Math. Phys. 19 (1965) 545.
[73] B.Y. Wang, I.I. Glass, Compressible laminar boundary layer flows of a dusty gas over a semi-infinite flat plate, J. Fluid Mech. 186
(1988) 223–241.
[74] S.L. Soo, Boundary layer motion of a gas–solids suspension, in: Proc. Symp. Interaction Between Fluids and Particles, Inst.
Chem. Eng., 1962, pp. 50–63.
[75] A.J. Chamkha, J.J.R. Peddieson, Boundary layer flow of a particulate suspension past a flat plate, Int. J. Multiphase Flow 17
(1991) 805–808.
[76] N. Thevand, E. Daniel, J.C. Loraud, On high-resolution schemes for solving unsteady compressible two-phase dilute viscous
flows, Int. J. Numer. Methods Fluids 31 (1999) 681–702.
[77] I.S. Chang, One and two-phase nozzle flows, AIAA J. 18 (1980) 1455–1461.
[78] R. Ishii, K. Kawasaki, Limiting particle streamline in the flow of a gas–particle mixture through an axially symmetric nozzle, Phys.
Fluids 25 (6) (1982) 959–966.
[79] R. Ishii, Y. Umeda, K. Kawasaki, Nozzle flows of gas–particle mixtures, Phys. Fluids 30 (3) (1987) 752–760.
[80] C.J. Hwang, G.C. Chang, Numerical study of gas–particle flow in a solid rocket nozzle, AIAA J. 26 (6) (1988) 682–689.
[81] H.T. Chang, L.W. Hourng, L.E. Chien, Application of flux-vector-splitting scheme to a dilute gas–particle JPL nozzle flow, Int.
J. Numer. Methods Fluids 22 (1996) 921–935.
[82] R.C. Mehta, T. Jayachandran, A fast algorithm to solve viscous two-phase flow in an axisymmetric rocket nozzle, Int. J. Numer.
Methods Fluids 26 (1998) 501–517.
[83] O. Igra, I. Elperin, G. Ben-Dor, Dusty gas flow in a converging–diverging nozzle nozzle, J. Fluids Eng. 121 (1999) 908–913.
[84] L.H. Back, R.F. Cuffel, Detection of oblique shocks in a conical nozzle with a circular-arc throat, AIAA J. 4 (1966) 2219–2221.
[85] L.H. Back, P.F. Massier, R.F. Cuffel, Flow phenomena and convective heat transfer in a conical supersonic nozzle, J. Spacecraft 4
(1967) 1040–1047.
[86] R.F. Cuffel, L.H. Back, P.F. Massier, Transonic flowfield in a supersonic nozzle with small throat radius of curvature, AIAA J. 7
(1969) 1364–1366.