Eulerian and Lagrangian

predictions of particulate two-

phase flows: a numerical study
F. Durst

Lehrstuhl fiir Strijmungsmechanik, Universitit Erlangen-Niirnberg, Egerlandstr. 13,

D-8520 Erlangen, West Germany

D. Milojevic

Boris Kidric Institute of Nuclear Sciences, 11001 Belgrade, Yugoslavia

B. SchSnung

Sonderforschungsbereich 80, Universitci’tKarlsruhe, Kaiserstr. 12, D-7500 Karlsruhe,

West Germany
(Received 1983; revised June 1983)

To predict particulate two-phase flows, two approaches are possible. One

treats the fluid phase as a continuum and the particulate second phase as
single particles. This approach, which predicts the particle trajectories in
the fluid phase as a result of forces acting on particles, is called the
Lagrangian approach. Treating the solid as some kind of continuum, and
solving the appropriate continuum equations for the fluid and particle
phases, is referred to as the Eulerian approach.
Both approaches are discussed and their basic equations for the particle
and fluid phases as well as their numerical treatment are presented.
Particular attention is given to the interactions between both phases and
their mathematical formulations. The resulting computer codes are
discussed.
The following cases are presented in detail: vertical pipe flow with
various particle concentrations; and sudden expansion in a vertical pipe
flow. The results show good agreement between both types of approach.
The Lagrangian approach has some advantages for predicting those
particulate flows in which large particle accelerations occur. It can also
handle particulate two-phase flows consisting of polydispersed particle size
distributions. The Eulerian approach seems to have advantages in all flow
cases where high particle concentrations occur and where the high void
fraction of the flow becomes a dominating flow controlling parameter.

Key words: mathematical model, two-phase flow, turbulent flow, Euler

predictions, Lagrange predictions

Multiphase flows in general and two-phase flows in particu-
lar, play an important Ir.U it many engineering applications
and in engineering research. Particulate solid-liquid, gas-
liquid or solid-gas flows belong to a wide class of two-phase
flows, usually referred to as dispersed flows. These particu-
late flows occur not only in the field of transport engineer-
ing, for example in connection with the transport of wheat,
coal or other products, but also in other fields such as
chemical engineering, mechanical engineering, energy
related fields etc. In these fields, it is well known that it is
possible, by introducing particles into a fluid, to increase
drastically the ratio of active particle surface to particle

0307-904X/84/02101-15/$03.00
o 1984 Butterworth & Co. (Publishers) Ltd. Appl. Math. Modelling, 1984, Vol. 8, April 101
Eulerian and Lagrangian predictions of particulate two-phase flows: F. Durst et al.

volume, and in this way to increase the reaction rate
between the different phases. This positive effect of
particles in fluid flows can be utilized for heat transfer and
mass transfer as well as for chemical reactions. Erosion and
corrosion problems are also heavily controlled by the
particles in solid-liquid and solid-gas particulate flows.
Last, but not least, there are many particulate flows in
nature that are of far-reaching interest, as for example the
sediment transport in rivers, and the falling of rain, snow,
hail, etc.
As long as the particles of a particulate two-phase flow
are small enough, they follow the fluid motion and the
whole flow can be treated as a homogeneous flow with
appropriately changed physical properties. This kind of
particulate two-phase flow, which is basically a single-phase
flow, will not be considered. In the present study it is
presumed that the particles are large enough for there to
exist a slip velocity between the two phases.
Most of the experimental work on particulate flows has
dealt, on the one hand, with the flow of and around single
particles in gas or liquid, and on the other hand, with the
flow of suspensions, where only integral parameters are
taken to characterize the flow properties. An example of
the latter is the measurement of the pressure loss of a
particulate flow in a pipe and its correlation with the total
mass-flow rate of the solid and fluid phase. This kind of
investigation has been forced onto studies of particulate
two-phase flows by lack of suitable measuring techniques
to yield more refined information. Only during the last
few years have new experimental techniques emerged that
permit local information on both the phases to be reliably
obtained. Laser-Doppler anemometry, for example, allows
the velocities, sizes and concentration of particles to be
locally measured as well as the fluid velocity. In this way,
local flow information can be obtained in the entire flow
domain.
Such detailed information of many relevant parameters
of particulate two-phase flow is absolutely necessary to
improve the assumptions introduced into two-phase flow
prediction procedures. Through detailed experimental
studies, it is also possible to see whether information on the
flow around single particles can be transferred to particu-
late flow systems. In addition to this, numerical models of
two-phase flow can only be developed if there exist reliable
local experimental data, against which theoretical results
can be compared.
In the last 10 years, the numerical treatment of single-
phase flow problems has reached a level which permits
calculations for laminar as well as turbulent flows. With the
increasing capacity of available computers, systems of
nonlinear, coupled, partial differential equations can be
solved for exceedingly complex flow problems. With two-
phase flows, it is necessary to solve additional transport
equations for the second phase and to handle those inter-
action terms which treat the mass, momentum and energy
exchanges between the two phases.
Particulate two-phase flows can be dealt with theoreti-
cally or numerically in two ways, and there are two
approaches to describe the transport equations for the
particulate phase. The Lagrangim approach treats the fluid
phase as a continuum and predicts the trajectory of a single
particle in the fluid flow as a result of various forces acting
on the particle. By assuming different starting positions
of the particles and following their trajectories, a solid-
fluid flow can be simulated. As the locations of the particles
are known, the mass, momentum and energy transfers of
the solid to the fluid phase and vice versa can be calculated.
Treating the particle phase as a continuum too, and
solving the appropriate equations for the fluid and the
particle phases, makes up the basic feature of the Eulerian
approach. The volume concentration of each phase has to
be introduced and the continuity equation for both phases
solved. The back-influence of the solid particles to the fluid
has then to be proportional to the void fraction of the solid
phase.
It is the aim of the present paper to compare both
approaches for some very simple flow configurations. The
chosen numerical test cases take into account the particle-
fluid interactions through the drag force, based on a single
particle motion, and the gravity effect. The influence of
fluid turbulence on the particles, and the influence of the
particles on fluid flow turbulence, are neglected. It is
assumed that there are no particle-particle interactions and
that the interactions of the particles on the wall can be
neglected.
In the following sections the basic equations and the two
solution procedures for both the approaches to predict
particulate two-phase flows are presented. As the con-
tinuous phase is treated in the same way for the two
methods, the fluid equations and their numerical solutions
will be explained first. The different treatments of the
motion of the dispersed phase and its interaction with the
fluid phase are then given, first for the Eulerian method
and then for the Lagrangian method.
Basic equations and solution procedures
The basic equations for a single-phase Newtonian fluid
are well known (see, for example, reference 1) and are
therefore only summarized here for the case of a stationary,
two-dimensional, axially symmetric laminar or turbulent
flow. With these assumptions for the flow, the continuity
equation can be written as follows:
and the momentum equations for the u and v velocity
components of the fluid phase as:
(2)
=_ap+2-a p-av
ay ay ay ( 1
+;[p(E+$)] +yt-p,
For a turbulent flow, the laminar viscosity /J is replaced by
an effective viscosity:
Peff = c1+ Pt (4)

102 Appt. Math. Modelling, 1984, Vol. 8, April

 Eulerian and Lagrangian predictions of particulate two-phase flows: F. Durst et al.
and hence the present approach to compute the fluid
motion uses an eddy viscosity concept for the Reynolds
shear stress.
Furthermore, to yield a local description of the turbu-
lent momentum transport in the fluid, the turbulent
viscosity pt is described by the turbulence energy k and the
turbulence dissipation rate e, for which transport equations
have to be solved. Details of this so-called k-e model have
been given by Rodi.’ The turbulence model formulated in
reference 2 for single-phase flow has not been changed for
the two-phase flow calculations presented in this report.
Hence, the present predictions assume that the particles do
not change the turbulence structure of the fluid phase and
that there is no influence of the turbulent fluctuations on
the particles. This will be modified by the authors as soon
as they have more physical information from experimental
studies of flow-particle interactions.
For the calculation of a single-phase flow, the boundary
conditions are normally specified in the following manner:
Inlet: u, v are prescribed
au av
Exit : -_=-_=()
ax ax
(5)
Wall: u=v=o
1 every point of the flow domain is solved from a particle
Symmetry axis: E = 0, v=o
(pipe axis)
These equations for single-phase flows are also taken as a
starting point for the transport equations of the continuous
phase of the particulate two-phase flows treated in this
report. In any continuation of the present study, they have
to be changed in such a way that the presence of the
particles are taken into account in a physically correct way.
In particulate two-phase flows the particle phase
occupies a definite volume, and this has to be taken into
account in formulating the transport equations for the fluid
phase. Only the part of the incoming volume which is not
taken by the particle phase enters the continuity and
momentum equations for the fluid phase. In the Lagrangian
approach the volume of the particles is neglected, so that
the continuity equation of the fluid does not have to be
changed and the different terms in the fluid momentum
equations do not have to be weighted by the corresponding
volume concentrations. With this simplification, only the
momentum equations have to be solved for each particle
class in the Lagrangian approach. For the Eulerian method,
a volume concentration of the fluid and the particle phase
will be defined and a continuity equation solved for every
phase. In addition to this, the different terms in the
momentum equations are weighted by the volume concen-
tration.
In the equations for the continuity and momentum
transport of both phases, interaction terms occur. Since, in
the present predictions, it is assumed that there is no phase
change, only the momentum transfer between the phases
has to be considered. This transfer of momentum yields
additional terms that have to be included in the momentum
equation of the fluid phase. It is assumed here that they
will be dependent on the velocity difference between the
two phases. Of course, the momentum transfer from the
particle phase to the fluid phase has to be opposite to the
corresponding momentum transfer from the fluid to the
particles.
In the Lagrangian approach, the gain or loss of particle
momentum is calculated and included as a sink or source
term in the fluid momentum equation. The drag force
between the two phases is calculated in the Eulerian
approach and added with opposite signs to the momentum
equations of both the phases. Similarly, other terms that
describe interactions between the phases can be added to
the momentum equations.
Generally, in particulate two-phase flows, the particles
have a density different from that of the fluid. Therefore
the gravity term has to be included in the momentum
equations of the continuous phase. The resulting pressure
in the fluid equations is the static pressure, which also takes
into account buoyancy effects. As the gravity term will be
dependent on the particle concentration, its magnitude
changes with location in space.
The above-mentioned additional momentum terms in
the fluid momentum equations model the back-influence
of the particles on the fluid.
In the Lagrangian approach; Newton’s law of motion is
used to find the velocity of every particle in the flow
domain. Integration of the velocity with time gives the
particle path, so that the particle velocities and locations
can be used to get the momentum transfer terms. In the
Eulerian approach, the void fraction of the particles in
continuity equation, rather than the individual particle
locations being determined. Together with the particle
velocities, which are known as a function of space, the
corresponding transfer terms can be derived in the Eulerian
method.
The numerical solution procedure of the extended trans-
port equations of the continuous phase are the same for
both the approaches and the basic ideas are presented
below. The calculation procedure is based on the TEACH
code, which has been developed at Imperial College,
London, and which has been described in detail by Patanka
and Spalding.374
The partial differential equations for the mass and the
momentum are solved by using finite differences, derived
by integrating the differential equations over control
volumes. For this an orthogonal grid is applied, where
the two velocities and the pressure are stored in staggered
positions (Figure I ). The necessary linearization of the
nonlinear equations is performed by hybrid differences to
- u,up
v, vp
t
Figure 7 Staggered grid and control volume
Appl. Math. Modelling, 1984, Vol. 8, April 103
Eulerian and Lagrangian predictions of particulate two-phase flows: F. Durst et at.

get an implicit finite difference scheme. Because of the account the fact that both the phases are flowing through
elliptic nature of the partial differential equations, an the control volume.
iterative solution procedure is employed. We start with As there is no mass exchange between the two phases,
distributions for the velocity and pressure fields, and two separate continuity equations exist:
calculate the fluid velocities from the momentum equations.
Thereafter, the solution of a pressure correction equation
(7)
(Poisson equation), which has been derived from the
continuity equation, yields corrected values for the velocity
and for the dispersed phase:
and the pressure fields so that the continuity equation is
satisfied. With the corrected values, the momentum
equations are solved again and the whole procedure is
repeated until convergence is achieved. This was assumed
to be the case when the normalized residuals at each point, The particles will have constant density, so that the
summed over the whole calculation domain, were each continuity equation of the dispersed phase can be written
smaller than 0.005. as:
The finite difference equations are solved using a tri-
diagonal matrix algorithm. The velocities and the location
or the ‘void fraction of the particles’ are obtained by
$ (qJ + 5$ (CqJ =0

solving the corresponding equations after a specific number

of iterations of the fluid solution procedure has been Momentum equations
carried out. With the computed information on the
The momentum equation for the continuous phase has
particles, the momentum transfer terms can be calculated
to be weighted by the volume concentration 0 of the fluid.
and introduced in the fluid equations before the solution
In addition to this, the momentum transfer term, based
procedure is repeated. Details on how this has been per-
on the drag force between the two phases, has to be taken
formed for both approaches will now be shown.
into account. As this term appears in the momentum
equation of the dispersed phase with opposite sign, it will
be mentioned here and discussed in detail when the
The Eulerian approach particle momentum equations are derived.
The fluid momentum equation in the x direction (with
The basis of the Eulerian approach is the treatment of the
gravity) is:
dispersed phase as a continuum. As there are two ‘fluids’
present, the definition of a volume concentration of each
phase, the so-called void fraction, is necessary and will
be explained first. After the derivation of the additional
terms in the fluid continuity and momentum equations,
the corresponding transport equations for the particle phase
are established. Finally the numerical procedure for the
solution of the dispersed two-phase flow problem is
explained.

Continuity equations
- flpg - 0.75 o$cD $2Re,(u - up) (10)
The Eulerian approach is based on the assumption that
there are two continua in two-phase flow. At one point
in the flow domain and at a specific time, however, only The fluid momentum equation in they direction is:
one fluid can be present. As with a numerical procedure,
it is not possible to resolve every point in time or space;
it is necessary to average over a specific time and space.
Definition of a volume concentration gives the fraction of a
control volume which is occupied by the first or the second
phase.
The void-fraction of the dispersed phase is:

2p/l av v
As stationary flow problems only are being dealt with, it is +_ -_- -0.75@3c~~Rep(v-Vn)(ll)
assumed that this variable is not changing with time and an r ( ay r1
integration with time is thus not necessary. In order to get
a physically reasonable meaning for the void-fraction, the The starting point for the momentum equations of the
ratio of the control volume and the particle volume should dispersed phase is Newton’s law of motion for a particle in
be much larger than unity; otherwise the concept that the a fluid, which is, for the x direction:
dispersed phase can be treated as a continuum is not
justified. The control volume over which the void-fraction
(12)
is averaged is the same as that for the numerical control
volume approach. The transport of mass or momentum has
now to be weighted by this void-fraction, to take into Changing the total by the substantial derivative and

104 Appl. Math. Modelling, 1984, Vol. 8, April

 Eulerian and Lagrangian predictions of particulate two-phase flows: F. Durst et al.

dividing by the particle volume results in: ap a av

( 1
P
=-_a-+2_ (y-2

Introduction of a particle Reynolds number and multi-

plication by the void-fraction gives:
The use of the continuity equation for the dispersed phase
results in the momentum equation for a suspension of
particles flowing with a slip velocity Au in a fluid:
$(appui)+i
fj (ar~~v~)=~olc~$R~~A~
To obtain consistent momentum equations for a flowing
suspension, the following have to be taken into account:
(1) The pressure variation, which can occur in the flowing
fluid, acts also on the dispersed phase.
(2) The gravity force on the particles: as gravity is
also introduced into the fluid momentum equation,
buoyancy effects result from the coupling of the two
phases by the pressure.
(3) The momentum transfer term has to be symmetric
with respect to the volume concentration of both the
phases, as two continua are being dealt with. For this
reason, instead of the particle void-fraction, the product of
both the volume concentrations has to be.used.
(4) The second derivatives in the momentum equations of
a Newtonian fluid are simulating the molecular momentum
transport in a laminar flow. The additional momentum
transport, resulting from turbulent fluctuations, is modelled
very often with the eddy viscosity concept. This turbulent
momentum transport, which is derived from the Reynolds
equations, also has a physical meaning in the momentum
equations of the dispersed phase. For this reason, the
second derivatives also will be taken into account in the
dispersed phase for simulation of the particle turbulent
fluctuations. As the modelling of diffusion phenomena with
the Lagrangian approach is not possible in such a way, the
influence of the turbulence on the particles has not been
considered here (u = 103’).
Finally, the momentum equations for the dispersed
phase can be written in the following way. The dispersed
phase momentum equation in the x direction is:
- cUP,g + 0.75 @co $Re,(u - Up)
(13)
+;of;!
+“;),
+yp)
+ 0.75 ff&, $ Ren(v -vp) (17)
(14)
From the momentum equations for the two phases, it
can be seen that, for cr -+ 0, the fluid momentum equation
transforms to a single-phase flow equation and the particle
momentum equation approaches the equation of motion
for a single particle.
The drag coefficient in the momentum transfer term,
(15) which is based on the motion of a single particle, is
expressed by the relationship:
24
CD ~~(1 -t 0.15 Rei6*‘) (18)
P
which approximates quite well to the standard drag curve
for a sphere. The experimental data for this correlation
have been obtained by measuring the drag coefficient of a
single sphere in a uniform flow. The question as to how
far it is possible to extend such results also for flow con-
ditions where turbulence, non-uniform fluid velocity or
multiparticle motion have roles will not be discussed here.
Besides the drag term introduced, additional forces,
acting on the particles, are possible. A non-uniform fluid
velocity causes migration of particles’ in the same way as
the result of particle spinning.6 With increased concentra-
tion, there will be some diffusion-like interaction of the
particles. These effects are not taken into account, as the
influence of turbulence on particles and vice versa will be
neglected.
The system of six partial differential equations can now
be solved together with the equation:
p=1-ff (19)
to obtain the seven independent variables u, v, up, vp, OL,
/? and p). As already described, the solution of the mass and
momentum equation of the continuous phase gives the
fluid velocities and the pressure. The particle velocities
are obtained from the particle momentum equation and the
particle void-fraction by solving the particle continuity
equation.
The fluid boundary conditions are the same as those
described already. For inlet and exit, the particle boundary
conditions are the same as for the fluid phase.
As no second derivatives occur in the particle transport
equations (a = 103’) and up = 0 is assumed for the sym-
metry axis and wall, no conditions are necessary for up and
01for those boundaries; so we have:
inlet: up, vp, a! prescribed
au av aa
exit: P=P=-~0 (20)
ax ax ax
(16)

Solution procedure
and the dispersed phase momentum equation in the _Y The numerical solution procedure is based on the same
direction is: ideas as for single-phase flow. Instead of three, six partial
differential equations have to be discretized and solved with

; @PPUPVP)
+ s
(cuppv;)
an implicit difference scheme. The location of the void-
fraction in the staggered grid is at the same nodes as is the

Appl. Math. Modelling, 1984, Vol. 8, April 105

Eulerian and Lagrangian predictions of particulate two-phase flows: F. Durst et al.
x - direction
r-------1
A--
I
-t-
41 JI
L-_-_-_-J
aw 9 aE
Figure 2 Upwind-downwind scheme for flow in x direction
pressure. The particle concentration is obtained from the
particle continuity equation, which is discretized by the
upwind-downwind scheme, shown in Figure 2 for a one-
dimensional flow. The value of the void-fraction at an
interface of the control volume is taken from the nearest
upstream-located grid point.
It is well known (see, for example, reference 7) that
when we use upwind schemes, artificial diffusion problems
can occur, especially in separated flows. Since in the follow-
ing predictions no turbulent diffusion is taken into account
in the particle continuity equation, this effect becomes
even more important in the predictions of separated flows,
which are presented later. For this reason, in the final
version of separated flow calculations a different
discretization, only of the particle continuity equation,
has been used.
A quadratic upstream interpolation scheme proposed by
Leonard*-‘e uses a third-order upstream bias in the con-
vective terms. It has been shown by Leonard that numerical
diffusion problems can be avoided by using this scheme.
For this reason it has been introduced for the predictions
for separated flows in a later section, with slight modifica-
tions proposed by Pereira,” in order to assure stability
also for those cases where the diagonal dominance of the
tridiagonal matrix is not achieved.
The two phases are coupled by the void-fraction and the
momentum transfer term. Depending on the void-fraction,
many more iterations are therefore sometimes necessary
to obtain a converged solution than for a single-phase
flow problem, and under-relaxation factors have to be used.
The Lagrangian approach
The Lagrangian treatment of the particulate phase combined
with the Eulerian approach for the fluid was first proposed
by Migdal and Agosta. l2 They showed that the particles
could be treated as sources of mass, momentum and energy
in the fluid flow field. The elliptic form of the equation
for the conservation of the general variable + in axisym-
metrical coordinates is as follows:
+ s, + s;
Equation (21) contains the additional source term
S$ which represents the net efflux of $Jinto the fluid
phase owing to the particle-fluid interaction. This source
term is calculated by solving the Lagrangian equation for
the corresponding particle variable ~9’. This is the essence
of the so-called PSI cell (particle-source in cell) approach,
which was used by Sharma l3 for the computation of gas-
particle flows in orifices and venturis, by Crowe et al. I4
for gas-droplet flows, by Sharma et aZ.I5 for the computa-
tion of swirling gas-particle flows in pulverized coal
106 Appl. Math. Modelling, 1984, Vol. 8, April
classifiers, and by Lockwood et ~1.‘~for the prediction of
pulverized coal-fired furnaces.
Inspection of equation (21) shows that the solution
procedure for the fluid can be the same as for the single-
phase flow or for the above Eulerian treatment of the
particulate phase. We may neglect mass transfer and use
time-averaged variables for the turbulent flow; then the
equations for the conservation of mass and axial/radial
momenta are discretized and solved according to the
TEACH numerical solution procedure. The standard two-
equation k-e turbulence model, together with the isotropic
eddy viscosity hypothesis, is employed for the prediction
of the turbulence transport.
Calculation of the particle trajectories by solution of
the equations of motion of the particles, as well as the
calculation of the particle source terms, is performed in
the same iteration loop after the solution of the fluid
flow equations, in order to handle the interaction between
the fluid and the particles.
Velocities and trajectories of particles
According to the TEACH numerical solution code for
single-phase flow, the flow domain is subdivided into a
finite number of control volumes (cells) forming an
orthogonal grid for the numerical calculation. The particles
are introduced at a finite number of starting locations with
a finite number of particle sizes at each starting loca-
tion. If the total mass flow rate of particles is tip, if the
mass fraction of particles starting at a location j is 4, and
if the mass fraction of particles of class i with diameter
di is f;:, the number flow rate of spherical particles of
class i at starting location j is:
(22)
The equation of motion for the particle of class i starting
at location j is:
+myg ( l-- 1 (23)
- \ Pp,i/
where rnf is the mass of a single particle.
The influence of the Reynolds number (Rep =
pd ( V- Vi?l/p) on the drag coefficient of spherical particles
is satisfactorily described by:l’
co = E (1 + 0.15 Z2e>687) (24)
P
over a wide range of Reynolds number (approximately
1 <Rep < 103).
(21) For the calculation of the particle trajectories, the
method of Sharma13 is used. If the fluid velocity is con-
stant during short time intervals At, equation (23) is
solved analytically to give:
V$=V-(V-V&)exp(-:)+gr (l-e)
x [I -exp (-:)I (25)
where V$, is the particle velocity at the beginning of the
 Eulerian and Lagrangian predictions of particulate two-phase flows: F. Durst et al.
time increment At, and r is the particle characteristic
defined by :
4~p, i di’
7=
31.1CDKen
The particle position after the time interval At is determined
from :
Note that equation (23) can be also solved using a
Runge-Kutta integration, as was suggested by Lockwood
et al. l6
Particle momentum source term
The transfer of momentum between the phases is due
to the drag force. If we consider only one class of particles
and only one starting location, and assume a constant
particle diameter along the trajectory, the particle momen-
tum source term for any arbitrary control volume can be
written as :
tin
where tin and tout are the times when the particle enters
and leaves the control volume. However, the calculation of
the source terms is more efficient if we multiply equation
(23) by at dt and integrate it over the residence time of
the particle in the control volume Atn = tout - tin:
S$ = fi$(Vfi,out- VJ,,in)-$g l -p
( )
PP, i
X Ctout- tin)
The mass flow rate of the particles along the trajectory
is:
+=mq’.@
The momentum source term resulting from all particle
trajectories crossing the control volume is obtained from:
%en = c c s$
i, cell j, cell
Inspection of equation (29) shows that the particle
momentum source term is composed of two parts: the
change of momentum of the particles, and the influence
of all external forces acting on the particles. In our case,
all other forces acting on particles, except gravity, are
neglected. Some other authors using the Lagrangian
approach, such as Sharma,13 Crowe et al. l4 and Sharma
et al. l5 define the particle source term as the net efflux
of particle momentum represented by the first term on the
right-hand side of equation (29). This can lead to an
erroneous solution of the equations for the fluid flow,
because the interaction of particles and the fluid is realized
through the drag force and not only through the change
of the particle momentum. When particles reach their
settling velocities their momentum remains constant, but
the exchange of momentum with the surrounding fluid
owing to the gravity force still exists. It has to be noted
time here that the second term on the right-hand side of equation
(29) is very often much larger than the first one.
(26) Void-fraction
The void-fraction, which is used in the Eulerian
approach, has to be calculated also by the Lagrangian
method, to compare the results of both the approaches.
For the arbitrary control volume V and a single particle
trajectory, the void-fraction is given by:
(27)
ti$(tOrrt - tin)
OLij= (32)
Pp,i v
For all the particle trajectories crossing the control
volume, the following equation is obtained:
(33)
The Lagrangian approach requires a relatively large
number of starting locations for the particles in order to
obtain a reasonably smooth void-fraction profile. This is
due to the Eulerian definition of the void-fraction, which
assumes that particles can be treated as a continuum.
Solution procedure
For the calculation of the fluid-particle elliptic flow
field, a solution scheme modified from the TEACH
program was used. The calculations for the particle veloci-
ties, trajectories and the source terms were incorporated
into the iteration loop after solving the equations for the
fluid. Large scatterings of the particle source terms during
the iterative calculations were damped by under-relaxation.
The solutions were assumed to have converged when the
normalized sums of all residuals for the axial and radial
momentum equations and for the continuity equation
were less than 5 x 1Oe3. The same convergence criterion was
(29) also used for the Eulerian approach.
In the case of a liquid-solid two-phase flow, the back-
influence of the particles on the fluid is small and the
convergence is reached as fast as for single-phase flow.
However, in the case of a gas-solid flow, especially when
(30) the particle loading is high, the back-influence can be so
large that the solution diverges. This can be explained by
the coupling of the Lagrangian with the Eulerian part of
the calculation, which is done by means of the particle
source terms. The particles travel along a finite number
(31)
of trajectories, leaving behind a field of source terms for
the fluid equations. This source term field must be con-
tinuous in a flow domain in order to obtain a converged
solution of the fluid equations. Hence, a sufficiently large
number of starting locations of the particles is necessary.
According to the present authors’ experience, it is almost
impossible to obtain a fully converged solution of the fluid
equations with less than two starting locations for each
grid node at the cross-section where particles are introduced
Most of the calculations in this paper were done with ten
particle starting locations per grid node in order to ensure
convergence and to obtain a relatively continuous void-
fraction field.
In the Lagrangian approach presented here, the influence
of the void-fraction on the fluid velocities was not taken
into account. Generally, this could be done, because the
void-fraction field is calculated from known particle
trajectories. It should be noted, however, that a rough
void-fraction field can cause convergence problems.
Appl. Math. Modelling, 1984, Vol. 8, April 107
Eulerian and Lagrangian predictions of particulate two-phase flows: F. Dust et al.

Table 1 Specification of the test cases

No. Pipe Pipe Fluid Fluid Reynolds Inlet inlet Particle Particle Particle Notes
diameter length density viscosity number void- particle loading density diameter
D Re fraction to fluid ratio d
PP
h) tin velocity y (kg/m31 (pm)
ratio

Pipe flows
1 0.2 10.0 1.0 1.6X10-' 3.525X104 5X1O-4 1.0 0.5 1000 100 upward
flow
2 0.2 10.0 1.0 1.6X10-' 5x104 2x lo-" 0.2 1.0 2500 100 upward
flow
3 0.1 15.0 1.0 1.6X1O-5 5x104 Non-uniform 1 .O 2.0 2000 100 upward
2.04 X lo-” flow
for y Q 0.7 R
4 0.1 5.0 1000 l.ox1o-3 5x104 0.1 1.0 0.1 100 upward
flow
5 0.2 1.0 1000 1.ox1o-3 8X104 0.2 0.1 0.025 500 upward
flow
6 0.1 5.0 10 1.6X10-' 5x104 Non-uniform 1 .O 2.0 Mixture: upward
2.04 X lo-” 100/300 flow
for y Q 0.7 R 50% mass
fraction each
size

Separated flows (sudden pipe expansions)

7 large0.2 4.0 1.0 1.6X1O-5 2.5X lo4 - - - - - single-
small 0.1 phase
8 large 0.2 4.0 1.0 1.6X10-’ 2.5 X lo4 4x10- 1 .o 1.0 2500 100 downward
small 0.1 flow
9 large 0.2 4.0 1.0 1.6X1O-4 2.5 X IO4 1O-3 1 .o 1 .o 1000 100 downward
small 0.1 flow
10 large 0.2 4.0 1.0 1.6X10-’ 2.5X IO4 2x1o-3 1 .o 1 .o 500 100 downward
small 0.1 flow
-

Predictions of dispersed flows lated void-fraction profile at the distance 50 pipe diameters
The computer codes based on the Eulerian and Lagrangian from the inlet. The Eulerian approach gives a smooth and
equations have been used for the prediction of various almost uniform void-fraction profile, while the result of
two-dimensional particulate flows. Two simple flow con- the Iagrangian calculations depends on the number of
figurations were chosen for the comparison of the two starting locations. For 28 starting locations (two per grid
approaches: node in the inlet cross-section of the pipe), the scatter of
the points is very large. If the number of starting locations
(1) vertical pipe flow is increased to 140, the calculated points are very close to
(2) sudden expansion in a vertical pipe flow. the Eulerian solution. Because of this finding, all other
The specifications for the computer predictions carried out runs were carried out with 10 starting locations per grid
to compare the results are given in Tabk 1. node in the inlet cross-section of the pipe.
Run 2 represents a test case in which both programs
were employed to compute the steep acceleration of
Pipe j70 ws
Applications of the two computer codes to various flow
fields yielded those parameters that are most influential
on predictions of two-phase flows, for both the Eulerian
and Lagrangian approaches:
(1) choice of the number of particle starting locations in 0
t
the Lagrangian approach 0.6- 0
. . ’ b
(2) large particle accelerations or decelerations
0% OOaO .
(3) non-uniform void-fraction distributions in the Eulerian
approach
(4) high void-fractions
(5) polydispersed particle size distributions. -Euler
In Table 1 runs l-6 are used to show the advantages
and disadvantages of the two approaches with respect to
the problems stated above. Run 1 was performed with both
computer codes in order to see the influence of the number
of particle starting locations on the Lagrangian calculation
0.2

01
0
I ’
o
.

’
0.2
Lagrange NSL = 28
Lcgronge NSL = 140

’
04
I I I
0.6
I I
0.8
I I
1.0
of the void-fraction in a turbulent upward pipe flow with a Y/R

uniform void-fraction at the inlet. Figure 3 shows the calcu- Figure 3 Concentration of particles at x/D = 50

108 Appl. Math. Modelling, 1984, Vol. 8, April

 Eulerian and Lagrangian predictions of particulate two-phase flows: F. Durst et al.
particles in a turbulent upward pipe flow into which
particles were added with an inlet velocity which was only
20 per cent of the fluid velocity. The Eulerian approach
employs the same grid for the particles and for the fluid
while very short time increments are used for the calcula-
tion of the particle velocities and trajectories in the
Lagrangian approach. Because of these chosen conditions
for the computations, Figure 4 shows large discrepancies - _.--.-

of the centre-line particle velocities obtained from the q fro-

E
Lagrangian (circles) and the Eulerian (continuous line)
approaches. If a 10 times smaller x grid node spacing is P -x- Smgle phase flow
used, the result of the Eulerian calculation (dashed line) - FluId velocity
Euler I
- - - Particle velocity
becomes practically the same as the one obtained from the

1
!
0 FluId velocity
Lagrangian approach.
Run 3 was performed to compute the influence of a
. Particle velocky
I Lagrange

x
non-uniform particle concentration in a turbulent upward
pipe flow. In the central part of the pipe inlet cross-section I
(70% of the diameter) the particle void-fraction was set to t
be 2.04 x 1Od3, while in the outer part there are no particles 01 ’ ’ ’ ’ ’ ’ ’ ’ ’ ’
0 0.2 0.4 0.6 06 10
present. According to Figure 5, the two approaches predict Y/R

the same particle spreading and the same void-fraction Figure 6 Fluid and particle velocities at x/D = 150
profile at a cross-section which is located at x = 1SOD. A
comparison of the fluid and particle velocity profiles at
this cross-section (Figure 6) also shows that the two
approaches predict the same back-influence of the particles
on the fluid. The fluid velocity protile is completely
different from the one obtained for single-phase flow
calculations. Those represent a special case, i.e. (Y= 0, in
the present computations.

4.ok I

-Euler 0 20 40 60 00 100
10
- - Euler wth 10 ., smaller grid Spacing X/R

0 Lagrange
Figure 7 Pressure loss for high void-fractions

01 ’ ’ ’ ’ ’ ’ ’ ’ ’ ’ I
0 2 4 6 8 10
X/R
Run 4 is a test case for a particulate two-phase flow with
Fjgure 4 Particle velocity along the axis a high particle void-fraction. However, a special case was
chosen for the predictions with the same density for
flurds and particles. According to Figure 7, less pressure
4.0-
drop is computed when the Eulerian approach is used.
This difference can be explained by the fact that, in the
Eulerian approach, the shear stress term is multiplied by
30- the fluid volume fraction fl= (1 - CY),in order to get con-
sistent momentum equations. This results in a decreasing
pressure loss for an increasing void-fraction of the particles,
at least in those cases where the shear stress is dominant for
the pressure drop (Ap = 0). However, in the Lagrangian
I approach, the local void-fraction information computed
- -- Inlet void fraction
l.O- - Euler
in the program is not transferred to the fluid momentum
0 Lagrange equation and, because of this, the pressure drop is the
same for the single-phase and two-phase pipe flows.
In order to obtain information on the performance of
I I I I , I I ^ I
0 the two computer codes under conditions of high particle
0 0.2 04 0.6 0.8 1.0
Y/R loadings, additional test runs were carried out for high
Figure 5 Concentration of particles at x/D = 150 particle void-fraction two-phase flows. These differed from

Appl. Math. Modelling, 1984, Vol. 8, April 109

Eulerian and Lagrangian predictions of particulate two-phase flows: F. Durst et al.

Run 4 owing to density differences between the particle

material and the fluid. Under these conditions, the authors
were not able to obtain converged solutions using the
computer code for the Lagrangian approach. The computer
program based on the Eulerian approach provided con-
verged solutions after only a few iterations. This indicates
an advantage of the Eulerian approach for predictions of
particulate two-phase flows under high particle loadings in
the presence of particles with density differences between
particle and fluid materials.
The back-influence of steep particle acceleration and
changing particle void-fraction was tested in Run 5. The
flow rates of the fluid and of the particles were the same
in both approaches. Since with the Eulerian method the
void-fraction of both the phases has to be considered for
the calculation of the flow rates, this resulted in a higher
inlet fluid velocity in the Eulerian approach. The influence
of the void-fraction on the fluid velocity was not taken into
account in the Lagrangian approach and this is the explana-
tion for the differences during the acceleration period of - Lagmnge

the particles shown in Figure 8. -‘-Euler (v&h ovemged diameter )

Run 6 is a test case for both programs when applied to
flows with polydispersed particle size distributions; a

601 ’ ’ ’ ’ ’ ’ ’ ’ ’ ’
0 0.2 04 0.6 06 10
Y/R

Figure 70 Fluid velocity distribution at x/D = 30

non-uniform void-fraction distribution like the one in Run

3 was chosen at the inlet cross-section.
Euler The size distribution is made up of two sizes with
-u diameters of 100 pm and 300 pm. Since with the Eulerian
--_-_”
P method a set of three partial differential equations has to
-.-_a
Lagrange be solved for each particle diameter, those predictions are
0 u therefore very expensive; an averaged diameter is used to
0
UP simulate the back-influence of the particles on the fluid.
x a
Test calculations have been performed with a mean length
diameter, a mean volume diameter and a mean surface
diameter (for the definitions of different kinds of diameters,
see reference 18, p. 11). The mean length diameter, which
I I I I I I I , I I I , J
0 0.02 004 006 0.06 0.1
corresponds to the arithmetic mean, was regarded as the
x(m) best to simulate the polydispersed size distribution. For
Figure 8 Velocities and concentration at the centre-line of the pipe particles of different densities, however, a different mean
diameter should be used, to simulate the various density
classes.
Figures 9 and 10 show the variation of the fluid velocity
along the centre-line of the pipe and in a cross-section at
x/D = 30, respectively. It can be seen that the back-
influence of the mixture of two particle sizes on the fluid
velocity obtained from the Lagrangian approach is satis-
factorily predicted by using single-sized particles with an
arithmetic average diameter of 200 pm in the Eulerian
approach; at least, in this way, the agreement between the
results of both approaches is satisfactory. The centre-line
distributions of the velocities of the particles are given in
Figure 11. Here the particle velocities of the two particle
-.-Euler (with averaged diameter ) size classes from the Lagrangian approach are shown
together with that of the averaged 200 pm particle in the
t Eulerian method. The dots and crosses show the results of
the Eulerian predictions, where only one particle size
(100 pm or 300 pm) was present. The difference from the
6.01 ’ ’ ’ ’ ’ ’ ’ ’ ’ ’ ’ I 1
0 1.0 20 30 4.0 Lagrangian results can be explained by the different fluid
x(m) velocities of the mono- and polydispersed suspensions. The
Figure9 Fluid velocity at the pipe axis corresponding pressure distribution is given in Figure 12.

110 Appl. Math. Modelling, 1984, Vol. 8, April

 Eulerian and Lagrangian predictions of particulate two-phase flows: F. Durst et al.

1 tational schemes and, for a number of reasons, they were

expected to show up most by a comparison of the spread-
ing of the void-fraction distribution after the sudden
expansions. If streamline curvature was not present, the
spreading of the particle concentration was shown to be
Ti computed equally well by both computational schemes
‘; (see test case 3).
-
P
60- The Reynolds numbers of the oncoming flow, i.e.
--I
-- Lagrange Re = uD/v, was identical for all the calculations.
-*- Euler
The inlet void-fraction and the density of the particles
.-.
5o . )R were varied so that the loading y was constant for all the
Corresponding monodupersed flew
x
I
(Euler) predictions, as can be seen in Table 1, which also contains
other flow specifications. For these, computations were
4.01 ’ ’ ’ ’ ’ ’ ’ ’ ’ ’ ’ 11
0 10 20 30 40 performed as described below.
x (ml Figures 14-18 show the predicted data of the two
Figure 71 Particle velocities at the centre-line of the pipe computational approaches for a single-phase flow and for
particulate two-phase flows with particle densities of
pP = 2500 kg/m3 and pP = 1000 kg/m3 (Runs 7-9). In
I Figure 14 the friction coefficient rw/pu2 L shown. It can
be seen that the reattachment of the single-phase flow is
at x&J = 4.4 and that the corresponding value for the
two-phase flow is about twice as large. The reason for this
is the higher inlet moment of the particulate flow; the
particles have the same inlet velocity as the fluid but higher
density (p,/p = 2500 and 1000 respectively). A parameter
study of this reattachment length shows that with increasing
loading the reattachment length is increasing; this is quite
different from a single-phase flow where, at least in the
turbulent range, the reattachment length approaches a
constant value, if the Reynolds number is increased.
The different shear stress coefficient for single-phase
and two-phase flow at the exit of the calculation domain
x,(m)
can be explained by the shape of the fluid velocity pro-
files. As the particles are concentrated at the centre of the
Figure 12 Pressure loss
pipe and are more dense than the fluid, there is a down-
ward force on the fluid, which causes a higher fluid velocity
at the pipe axis for the particulate flow.

-HH

10 IR Single phase
Euler
-
Lagrange

1
Pp+= 2500 --- o

9 c \:I I++= 1000 -.- x

r
‘I
- I*
5- $

I?

0
\i
\x
_ I!
- c
ib
Figure 73 Test geometry for separated flows
0 ‘k”“““” 10 3fl

Separated flows
As a test case of particulate two-phase flow in a sudden
expansion, the downward flow in a vertical pipe was
chosen. In this flow, streamline curvature occurs and this
influences the particle motion. Because of this, differences I
are to be expected between the results of the two compu- Figure 14 Shear-stress coefficient with x-distance

Appl. Math. Modelling, 1984, Vol. 8, April 111

Eulerian and Lagrangian predictions of particulate two-phase flows: F. Durst et al.

As Figure 14 shows, the agreement between the pre-

dicted friction coefficient obtained with both approaches ---PP*=2500
is quite good for pp = 2500 kg/m3. The Lagrangian method,
-.-.-PP*= 1000
however, predicts for the case pP = 1000 kg/m3 a smaller
shear stress coefficient than does the Eulerian approach
over most of the downstream flow region. As is shown
below, the cause of this is a ‘flatter’ fluid velocity profile,
as indicated in Figures 15 and 16, where the computed
fluid velocity profiles are presented for two x locations.
For both computational approaches, there is a close agree-
ment at x/D = 1.13, but the profiles differ for pP/p = 1000
at x/D = 24. Looking at the void-fraction distribution
of the particles, which are shown in Figures 17 and 18,
this disagreement in Figure 16 can be explained. Here the
normalized void-fraction distribution from the Eulerian
method is plotted together with the location of the particle 0 0 25 05 0.75 10

nearest to the wall, from the Lagrangian approach. Y/R

In Figure 17 it can be seen that the steep decrease in Figure 17 Normalized void-fraction at x/D = 1.02
the void-fraction resulting from the Eulerian approach is
at about the same y location as the particle nearest to the
wall which has been predicted by the Lagrangian approach. I 1
With decreasing density, the particles go nearer to the wall,
as they increasingly follow the fluid motion. As Figure 18
shows, this effect is predicted by both the methods. The
Eulerian approach, however, predicts a more distributed
particle phase, which results in a flatter fluid velocity
profile.
The differences described above between the results of
the two approaches should increase with decreasing particle

-.-
Y/R

Figure 18 Normalized void-fraction at x/D = 27.3

YJT
\2
E
Single phase -
;
1 Pp” =2500
z,OOO-.-
--- 0
x
100

I
0
---=T --
8 ___---------x2
--=-=-
-1 ’ I I I I
0 0.25 0.5 0.75 10
Y/R 6
Figure 75 Fluid velocity at x/D = 1 .I3

I 1

1 I I I I I I 1 I I
0’
\;; 0 500 1000 1500 2OQO 2500
5 LO- BP (kg/t+)
2 Euler Lagmnge Figure 79 Reattachment length
Single phase -

PP*=2500 --- o

r 1000 -.- x
density, in such a way that the predicted results of the two
01 I I I 1 approaches differ more and more. This is indicated in
0 0.25 0.5 0 75 1.0 Figure 19, which shows the reattachment length for
Y/R different particle densities, where the increasing disagree-
Figure 16 Fluid velocity at x/D = 24 ment between the two methods with decreasing particle

112 &WI. Math. Modelling, 1984, Vol. 8, April

 Eulerian and Lagrangian predictions of particulate two-phase flows: F. Durst et al.
density can be seen clearly. As already mentioned, the
larger spreading of the particles with decreasing particle
density is caused by the smaller particle relaxation time.
The difference between the results of the two prediction
schemes is caused, however, by the numerical diffusion
which occurs by solving the particle continuity equation
in the Eulerian approach. As there are no physical diffusion
terms in the particle continuity equation, the effect of
numerical diffusion is significant, if upwing differences are
employed in the computational volume. If the flow method
geometry and the fluid velocities are kept constant, the
numerical diffusion is significant, if upwind differences are - - - Lagrange
employed in the computational volume. If the flow method X Euler ( Leonard scheme)
increasing the number of grid points for the Eulerian ?
,
calculations.
With decreasing grid spacing, the spreading of the
;x
\x
I
I
particles reduced and the particle concentration distribution 1 I I I \ XI
approached that predicted by the Lagrangian method. As a 0 0.25 0.5 0.75 1.0
consequence of this, the velocity profile predicted with the Y/R
Figure 21 Void-fraction distribution at x/D = 27.3 for pp =
Eulerian approach and that predicted with the Lagrangian
500 kg/m3
approach also become closer with increasing grid numbers.
In Figure 19, the reattachment length, predicted by the
Eulerian approach with a grid of 16 x 44 points, is also
included and the agreement between both the prediction
methods is quite satisfactory. distributions are shown far downstream of the reattach-
The problems caused by the numerical diffusion, occurr- ment point for the most critical test case (10) of pp = 500
ing with the Eulerian method, are due to the upwind kg/m3. The different profiles correspond to the results
differences, used in solving the particle continuity equation. obtained by the Eulerian upwind, the Leonard Eulerian
Since it is assumed that numerical diffusion is not so and the Lagrangian prediction methods.
important in the momentum equations for the fluid and the Figure 20 shows that a more parabolic velocity profile
particles (with respect to physical diffusion), it is not is achieved with the Eulerian approach by using the Leonard
worth while solving all six partial differential equations scheme. The reason for this lies in the large difference
with a much finer grid. Another approach would be to between the void-fraction distributions predicted by the
employ, at least for the particle continuity equation, a two numerical schemes. This can be seen in Figure 21,
higher order solution procedure, with which artificial where the three concentration profiles are plotted. The
diffusion can be reduced. void-fraction distribution, calculated by the Lagrangian
The Leonard quadratic upstream schemes-r0 mentioned approach, has to be smoothed because of the scattering
earlier has been incorporated into the subroutine for already explained.
solving the particle continuity equation, and the calculations The above comparison of separated particulate flow
have been performed with the same grid number (16 x 16) as predictions has shown that there is a problem of
in the Lagrangian predictions. In Figure 19, the correspond- numerical diffusion involved in solving the continuity
ing reattachment lengths are also shown for these calcula- equation of the particles. A numerical scheme, at least
tions. The slight but constant differences between the for the particle continuity equation, has to be used with
Eulerian and Lagrangian results can be explained by the which artificial diffusion can be avoided. In realistic calcu-
fact that the Leonard scheme is not conservative, and in the lations, where turbulent diffusion terms must be included
Eulerian predictions a loss of mass of about 8% was obtained. in the particle continuity equation, this problem will be not
In Figures 20 and 21, fluid-velocity and void-fraction so important.

Conclusions
Since only a simple drag force interaction has been intro-
duced between the phases in both programs, and other
interaction effects have been neglected, the present study
has been more concerned with the numerics of the two
approaches than with the actual physics.
From a consideration of the equations employed to
predict particulate two-phase flows it is apparent that the
two approaches are very similar; the main difference is in
the treatment of the particle phase. This can easily be
shown by comparing how the momentum transfer terms
are handled in both computational schemes. In the program
based on the Eulerian approach, the momentum equation
of the dispersed phase is added to the continuous phase
I I I
0 025 05 0 75 10 momentum equation. In this way the introduced drag
Y/R term cancels, and instead of this two additional terms
Figure 20 Fluid velocity profiles at x/D = 24 for pp = 500 kg/m” are added. They are the mathematical description of the

Appl. Math. Modelling, 1984, Vol. 8, April 113

Eulerian and Lagrangian predictions of particulate two-phase flows: F. Durst et al.

back-influence of the particle motion onto the fluid, and 6 Rubinow, S. I. and Keller, J. B. ‘The transverse force on a
stem from the acceleration of the particles and the gravity spinning sphere moving in a viscous fluid’, J. Fluid Mech. 196 1,
11,447
influence. Considering the particle acceleration term, for 7 Roache, P. J. ‘On artificial viscosity’, J. Computational Phys.
example, and integrating this term over a control volume, as 1972,10,169
is done in the Eulerian approach, we obtain the momentum 8 Leonard, B. P. ‘A stable and accurate convective modelling
transfer by the difference in the inlet and outlet particle procedure based on quadratic upstream interpolation’, Comp.
Meth. Appl. Mech. Engng 1979, 19,59
momentum, quite the same as in the Lagrangian approach.
9 Leonard, B. P. ‘A stable, accurate, economical and compre-
It is for this reason that the back-influence is treated hensible algorithm for the Navier-Stokes and scalar transport
basically in the same manner in both computational equations’, 2nd International Conference on Numerical Methods
schemes. This clearly indicates that differences in the in Laminar and Turbulent Flow, Venice, 1981
results obtained cannot result from the physics of the 10 Leonard, B. P. ‘A survey of finite differences with upwinding
for numerical modelling of the incompressible convective-
interactions between fluid and particles, but rather are diffusion equation’, in ‘Recent advances in numerical methods
caused by the numerical treatments of the various terms. in fluids, 2’, Pentech Press, London, 1981
The results of the computations show that both 11 Pereira, J. C. ‘A comparison between hybrid and quadratic
approaches have their advantages, generally speaking. The upstream differencing for low and high Reynolds number
Lagrangian approach gives more detailed information on elliptic flows’, Institut fi.ir Hydromechanik, University of
Karlsruhe, to be published
the particle phase yielding the trajectories and the resi- 12 Migdal, D. and Agosta, D. V. ‘A source flow model for
dence times of the particles in certain flow regions. Owing continuum gas-particle flow’, Trans. AWE, J. Appl. Mech.
to the very small time increments used for the calculation 1967,34E, 860
of the particle velocities and trajectories, the Lagrangian 13 Sharma, M. P. ‘Numerical and experimental study of gas-particle
flows in orifices and venturis: application to flowmeter design’,
approach gives better predictions of the flow with high
Ph.D. Thesis, Washington State University, 1977
particle velocity gradients and is less subject to discretiza- 14 Crowe, C. T., Sharma. M. P. and Stock. D. E. ‘The particle
tion errors of the particles, even if a coarse grid for the source-inceIl (PSI-cell) model for gas droplet flows’, Trans.
fluid phase is used. A well-known advantage of the ASME, J. Fluids Engng 1977,325
15 Sharma, M. P., Cornelius, D. K., Rice, J. G. and Dougan, D. R.
Lagrangian approach, that the computer storage require-
‘Numerical computation of swirling g&-particle flows: applica-
ment does not increase with the number of particle size tion to pulverized coal classifiers’. ASME Winter Annual
groups l6 has been confirmed by the present study. The Meeting, Nov. 1980, Paper No. 8i)-WA/HT-31
back&luence of polydispersed particles on the fluid can 16 Lockwood, F. C., Salooja, A. P. and Syed, S. A. ‘A prediction
be satisfactorily predicted if the average diameter is used in method for coal-fired furnaces’, Cornbust. Flame 1980, 38, 1
17 Cliff, R. and Gauvin, W. H. ‘Motion of entrained particles in
the Eulerian approach; but to obtain more complete in- gas streams’, Canad. J. Chem. Eng. 1971,49,439
formation about the particulate phase, the equation of 18 Soo, S. L. ‘Fluid dynamics of multiphase systems’, Blaisdell
motion of each particle size has to be solved, as in the Publishing Co., 1967
Lagrangian approach. This would be very time-consuming
and the Lagrangian approach is therefore recommended for
polydispersed distributions. No tation
Convergence problems of the Lagrangian approach, in
cases where the particle loadings are high, are easily over- axial coordinate
come by the Eulerian approach, which is superior for radial coordinate
suspension flows with high void-fractions. The problem of radial coordinate
numerical diffusion in separated flows can be avoided by density of fluid
choosing a suitable higher order difference scheme. The axial velocity of fluid
main advantage of the Eulerian method is in its simulation radial velocity of fluid
of turbulent mixing processes of the particles. In a real pressure
flow the particle concentration distribution (as in test molecular dynamic viscosity of fluid
run 6) would spread according to the capacity of the effective viscosity @,ff = p + pt)
particles to follow the turbulent fluctuations. This would I4 turbulent viscosity
influence the momentum and energy transfer. To predict k turbulent energy
this, the transfers have to be modelled for the prediction E dissipation
of turbulent particulate flows. Such modelling attempts (2 void fraction of dispersed phase (CX*= cr/cyO)
require, however, numerically satisfactory computational V control volume
codes for particulate two-phase flows. volume occupied by dispersed phase
7 void fraction of fluid phase
PP density of particles (p,* = pp/p)
UP axial particle velocity
References VP
radial particle velocity
1 Hughes, W. F. and Gaylord, E. W. ‘Basic equations of engineering g,g acceleration due to gravity
science’, Schaum’s Outline Series, McGraw-Hill, New York, cD drag coefficient
1964 d particle diameter
2 Rodi, W. ‘Turbulence models and their application in hydraulics’, particle Reynolds number
IAHR-Section, Experimental and Mathematical Fluid Dynamics, Rep
mp mass of a particle
1980
3 Patankar, S. V. and Spalding, D. B. ‘A calculation procedure t time coordinate
for heat, mass and momentum transfer in three-dimensional Au velocity difference (AU = u -up)
parabolic flows’, Int J. Heat Mass Transfer 1972, 15, 1787 u constant for turbulent mixing of particles
4 Patankar, S. V. ‘Numerical heat transfer and fluid flow’,
McGraw-Hill, New York, 1980
rizp total mass flow rate of particles
5 Saffman, P. G. ‘The lift on a small sphere in a slow shear flow’, 4 general variable
J. Fluid Mech. 1965,22, 385 r exchange coefficient

114 Appl. Math. Modelling, 1984, Vol. 8, April

 Eulerian and Lagrangian predictions of particulate two-phase flows: F. Durst et al.

S@ source term At time interval

S8 source term representing interaction of particulate 7 particle characteristic time
phase x particle position
mass fraction of particles, starting at location j D pipe diameter
j mass fraction of particles with diameter di 7 loading of particles
i$ number flow rate of spherical particles with diameter
di at starting location j Indices
V fluid velocity i diameter class di
Vp particle velocity i starting location
AP projection area of a particle (A = nd2/4) 0 time t = 0

Appl. Math. Modelling, 1984, Vol. 8, April 115