13h Australasian Fluid Mechanics Conference
Monash University, Melbourne, Australia
19:18 December 1998
NUMERICAL MODELLING OF TWO-PHASE FLOW
A VALIDATION EXAMPLE
Darrin W. STEPHENS and Jonathan A. HARRIS
‘School of Engineering, Mechanical Engineering
James Cook University, Townsville, Queensland, AUSTRALIA
ABSTRACT
‘This paper explores a simple test problem that
can be used for the validation of Eulerian two-
phase flow CFD models. Validation results are
presented for a commercial CFD code with multi-
phase flow capability, namely CFX-4.2, produced
by AEA Technology.
‘The results show that CPX-4.2 exhibits virtu-
ally perfect agreement with all 12 test cases over a
full range of input parameters using a very mod-
erate mesh,
INTRODUCTION
Modelling of two-phase flow using commercial
computational fluid dynamics (CFD) codes has
recently become a reality, with a number of ven-
dors now offering Eulerian multiphase flow capa-
bilities. In order to gain confidence in the accu-
racy of such codes it is essential to study simple
two-phase flow problems with known solutions.
In this way issues such as mesh requirements,
choice of solver parameters anc implementation
of user-defined subroutines may be investigated
‘thoroughly before the code is applied to more
complex problems in whieh the solution is
known. This step is particularly important in
multiphase flow as the solutions ean often be com
plex and non-intuitive
GOVERNING EQUATIONS
‘The equations governing Eulerian multiphase
fluid-fluid flows are presented in chapter 12 of
the CFX-4.2 Solver manual (Anon., 1997). Each
phase, denoted by a, interpenetrates with the
other phases and occupies a volme fraction ra
For Np phases, the sum of the volume fractions
rust equal unity, viz:
a)
‘The continuity equation for phase a is
Ltinas tinge) (2)
2 rapa) +9 (rapaite)
where ia is the velocity of phase a, pa is the den-
sity of phase a and rag characterises the mass
transfer from phase to phase a. Note that
Joa = 0. Assuming no body forces cr momen-
‘umn sources the conservation of momentum equa-
tion for phase «is,
trapatta) +ra¥a+¥- (ralpala ®a))
HV (Hal Vila + (Viéa)"))
% Ne
DV hide - ida) + DGitasiiy — tnpaita) (3)
fi 5
whore old isthe interphase drag coficint. The
imerphase mass transfer is defined as follows
ita = max(—Sa,0), thay = max(Soa,0)
where sag is the mass flow rate per unit volume
from phase a to phase 3
Thus
say 20 Tita
Sag $0 titag = Sap, thge
‘The above equations apply to the general case
of N, phases. In what follows, a simplified set
of these equations will be derived that apply to
‘one-dimensional two phase flow with mass trans-
fer. This case has a known solution and so is ideal
for development and validation of two-phase flow
numerical solutions. For example, when develop-
ing mass and momentum transfer subroutines it,
is best to solve a problem with a known solution,to ensure that the subroutines have been imple-
mented correctly.
One-dimensional two-phase flow
Consider the case of steady state, incompressible
flow in a horizontal channel of constant cross sec-
tion. Itis assumed that one-dimensional flow oc-
curs so that all quantities have uniform (ie., con-
stant) profiles at any x position, that all phases
experience the same pressure and the momentum
exchange due to drag is ignored (
‘Also, assume that only two phases are’ present
(f and g), and that there is a mass transfer rate
per unit Volume of ri from phase f to phase g
(Ge, ayy = rn). Further, let the volume fraction
of phase f be ry so that, by continuity, the vol-
ume fraction of phase g is 1 — ry. Under these
assumptions the continuity equations simplify to
ta —ryovnd “)
peressy) = )
and the momentum equations to
0 =r Z tupin= Baro) (6)
a me
iE = upin = F(rop0§) a
In the above momentum equations the shear
stress terms, such as d(rypiydu/dx)/de, have
been omitted since they can be made negligibly
small by choosing a low viscosity
“These equations may be rearranged to yield the
following set of coupled ordinary diferential equa-
Meare ine aakoowsa yp td
ey _ [ety oie] og
au, _[r+mn't] og
ae (l= rs)e
de a=") (10)
1
gpg ds (uy
If the conditions at the channel inlet are known,
these equations may be integrated along the
channel (e.g., using the ODE solver routines in
Matlab™) to yield profiles of ry, uy, p and uy.
‘The mass transfer per unit volume from phase f
to phase g (1) must either be given (eg., a con-
stant) or expressed in terms of the solution vari-
ables. In this study the following equation was
used to compute ri, since this is used by CPX:
ra = Currgpy a)
Here Cy is a constant with units of inverse sec-
onds.
NUMERICAL MODELLING
In the previous section, a limiting case of two-
phase flow with a known solution was identi-
fied. ‘This limiting case is an ideal test case for
multiphase flow CFD models. ‘The case of one~
dimensional two-phase flow involves mass trans-
fer and so is particularly useful for checking that
user-defined mass transfer subroutines function
correctly. In this section a commercial code with.
‘multiphase flow capabilities is applied to solve the
test problem. The code is CFX-4.2, produced by
AEA Technology.
mt
}_—____J
Figure 1: Channel geometry used for simulations
‘The one-dimensional flow test problem was
solved using the geometry, shown in Figure 1.
The parallel plates in Figure 1 are 100 ram long
with a plate separation of 10 mm. Since CPX-4.2
roquired a third dimension, the plate width was
set to Lm. A uniform mesh of 40 cells in the x
rection and 10 cells in the y direction was used in
all cases, giving an aspect ratio of 2.5. The prop-
erties used were for water and its vapour at 300 K.
These are py = 997 ka/m®, py = 0.0256 kg/m,
hy = 855 x 10-6 Pas, and juy = 9.09 x 10-7 Pas
(incropera and De Witt, 1990). Note that these
viscosities are small enough so that the viscous
terms can be neglected in equations (6) and (7).
Other properties and further solution cetails are
given in the following sections,
One-dimensional two-phase flow
marias?™,
Equations (8) through (12) were solved nu-
merically using the fourth and fifth order
Runge-Kutta algorithm (ode45) implemented in
MATLAB (Anon., 1992). The relative error
solution tolerance was set to 10-%mT |
Feu (8) | eum) | teai’m®) | on/s) ns) |
ups] 2
1 Poor [or | anes
Tor [0s
‘Table 1
model runs.
Parameters and inlet conditions for
‘The problem requires that the values of uy, uy
and ry are specified at the channel inlet (1 = 0).
In addition, the fluid densities (py and p,) must
also be specified. A total of 12 different cases
were run, covering a range of nid densities, inlet
velocities and inlet volume fractions, as shown in
‘Table 1. Table 2 summarises the results of these
different runs. The valnes given in the last three
columns of Table 2 correspond to the values of
us, My and ry , respectively, at the outlet of the
one-dimensional channel
As an example of the output of the program,
Figure 2 shows a plot of the primary phase veloc-
ity (uj), secondary phase velocity (ug), primary
phase volume fraction (rj) and pressure versus
streamwise position along the channel for case
12. For this particular case there is not a large
change in some of the solution variables (eg. uy
changes from an inlet value of 2 to an outlet value
of 2.0179, whereas ry changes from 0.8 at the in-
Jet to 0.7928 at the outlet). These relatively small
changes provide a fairly sensitive test case for the
CED code. Some of the other cases (eg. case 9)
show substantial variation in the velocities and
volume fractions along the channel
CFKA2,
One-dimensional two-phase flow can be imple-
mented on a two-dimensional geometry by apply
ing a boundary condition of zero shear stress at
the channel walls. ‘The applied inlet. conditions,
are uniform across the channel and correspond to
caso 12 given in Table 1. Convergence testing was
applied to the continuity equation with a conver-
gence value of 110~? kg/s. No initial guess was
patched to the computational domain, Figure 2
shows a plot of the primary phase velocity (uy),
secondary phase velocity (t,), primary phase vol
ume fraction (rj) and pressure versus streamwise
position along the channel for case 12, Case 12
was the most difficult to solve requiring large un-
der relaxtion of the mass transfer term for the
solution to converge. The solution of this prob-
Jean took 3129 CPU seconds, while the o:her cases
ranged from 481 to 1456 CPU seconds running,
on an SGI Power Challenge Supercomputer using
‘one R10000 CPU.
‘Comparison with MATLAB™ solution.
Assuming that the numerical MATLAB™ solu-
tions given in Table 2 are at least accurate to the
number of significant figures shown, then a com-
parison between the results given by CFX-4.2 to
those given by MATLAB?™! can be made,
i i
i E
i
Figure 2: Comparison of MATLAB™ and CEX-
4.2 results for run 12, Inlet conditions and param-
eters are given in Table 1. Solid lines and sym.
bols correspond to MATLAB™ computations.
and CFX-4.2 results, respectively.
‘The results from CFX-4.2, shown in fig
ure 2, show near perfect agreement with the
MATLAB™ predictions, Additionally, the so-
lution values at the channel outlet. (given in Ta-
ble 2) also show near perfect agreement with the
MATLAB™ predictions, over the full range of
test cases. At worst, there are differences in the
second decimal place of the solution, which is con-
sidered excellent for the relatively coarse mesh
used.
Figure 2 shows that as mass is transferred from
phase f to phase g the velocity of phase f do-
creases slightly, while the velocity of phase g in-
creases drastically. The continuity equations (4)
and (5) for phase g and phase f respectively, allow
for the explanation of the results shown, Since
the density of phase g is small compared to phase
f, phase g must accelerate to satisfy continuity.
From the conservation of momentum, equationsaA} PACD) 7)
Run [CPX [MPLAB] CPX [MPLAB] CPX [MAB
TL 2] 2 | 2 | 2 [over] orore
[Waa [RT| TRS | TRAD | 07455 0.70
3 [anes | 3.968 [2.2070 | 22071 [o-7a10| 0-790
[sre] ore | 1191s [1.1915 [0.7794] 0.7704
[Losi] 198i | s.819 | SBAST [0.7755 O.77oR
t
7
5
7
2 [2 | 2 [2 [o.aiso} o41o1
TaB1S [LOST | T1se | TARR |0.4305 [0-135
Pe ows 2 1100 [2 2017 | 22017 [0.5955 | 0.5055.
$3292] 35004 [1F.8019] 15-8019 [ 0.2410] 0.2497
10” | 20535 | 205K [TT-BOOT [11.1005 | 0.7782 0.7787
Ti] 1.965 | 1.965 [1.9819 | 1.98109 [o.re2t| o.7a20
12 [0rro] 2017s [se.07s1 | 0507 [0.7928] 0.728
Table 2: Comparison of MATLAB?™ and CFX-
4.2 results at the channel outlet.
(6) and (7), this acceleration of phase g causes
a decrease in the pressure along the length of
the channel as shown in figure 2. It should be
noted that the solutions given by MATLAB?
and CFX-4.2 for all 12 cases do not contain inter-
phase drag, This omission of the interphase drag
is indicated by the large increase in the velocity
of phase g. It would be expected that such a large
increae in velocity would be normally retarded by
the interphase drag,
CONCLUSIONS
A simplified two-phase flow problem has been
given that can be used for validation of mul-
liphase flow CFD codes. ‘The case of one-
dimensional two-phase flow with mass transfer re
sults in a system of coupled ODEs which can be
integrated with high accuracy.
‘The performance of a commercial CFD code
CFX-4.2, was evaluated by applying it to solve
the test problem. The results show that CFX-4.2
exhibits virtually perfect agreement with all 12
‘est cases over a range of input parameters using
a very moderate mesh
REFERENCES
ANON., CFX-4.2 Solver Manual, CFX Inter-
national, Harwell, Didcot, Oxfordshire, 1997.
ANON., MATLAB Reference Guide, The Math-
Works Inc, MA, USA, MathWorks, 1992.
INCROPERA, F.P. and DE WITT, DP,
Fundamentals of Heat and Mass Transfer, Wiley,
Singapore, 1990.