Вы находитесь на странице: 1из 10

Numerical calculations of two-phase flow in a

liquid bath with bottom gas injection:


The central plume
Malcolm R. Davidson

CSIRO Divisim of Mine& and Process Engineering, Lucas Heights Research


Laboratories, Lucas Heights, N.S. W., Arrstralia

The two-phase flow resulting from the bottom injection of gus into u liquid bath, constrained to be
axisymmetric, is modelled by governing eyuutions that incorporute the virtuul mass and purticle lift
forces and u diffusive interfuciul force. The equations ure solved numerically by using the transient,
two-fluid-jlow program K-FIX. Both the bath and the top space ure included in the culculation domain,
but attention is focused on the central plume. Inclusion of terms describing the virtual muss ,force
together with a force due to microscopic bulk pressure d$firences ure found to ensure formal stubility
of a simplified case for all admissible parameter values; this ulso en.sures numerical stability in pructice.
The particle lift force, together with a diff&sive interfuciul force, is shonpn to ticcount for the observed
spreading of the plume, the vertical variation in the centerline voidfruction, und the mugnitude of the
bubble rise velocity in the upper half of the plume: however, although the void,fraction is udequately
predicted uwuy from the centerline, the off-center bubble velocity rrmuins poorly determined despite
only small errors in the culculated gas jlonj rate ut d@erant heights.

Keywords: two-phase flow, gas injection, numerical model, lift force, diffusive force

systems ranged from being acceptable to qualitative


Introduction
only. More recently, fully two-phase numerical flow
High-speed injection of gas into liquid metals or slags models have been presented’-I2 in which the gas and
is widely used in smelting and refining processes to liquid phases are regarded as two interpenetrating con-
promote mixing and chemical reaction. A typical ex- tinua, each with their own set of governing equations
ample, considered in this paper, is that of gas injection coupled together by phase interaction terms. Such
through an orifice in the base of the containing vessel. models offer the greatest potential, since they are more
The plume of rising gas bubbles stirs the liquid contents fundamentally based than the single-fluid models and
by setting up recirculatory flow patterns. Theoretical do not require a prior knowledge of the gas distribution.
modelling of this two-phase flow and the corresponding Cross et a!.’ assumed a flat top boundary surface,
gas distribution within it is necessary to understand thus excluding the gas top space from the calculation
the underlying physical processes and is important for domain, and obtained favorable agreement with liquid
design purposes. velocity data of Grevet et a!.4 Koh et a!.Y extended the
The first attempt to mode! the flow of gas-stirred mode! to include the top space in the calculation, thus
liquids assumed a single-phase liquid flow driven by a allowing the free surface to deform, and simulated an
central gas core of known width and velocity.’ Sub- experiment of Castillejos and Brimacombe”; however,
sequently, models were developed that treated the gas- no agreement was obtained with the corresponding
liquid mixture as a single, variable-density fluid in which plume void fraction and bubble velocity data. Durst et
the void fraction distribution is either prescribed2-s or a!.x obtained accurate flow predictions for extremely
determined by a separate transport equation for the low gas flow rates (IO-‘-IO- cm3/s); however, best
mass fraction of liquid.” Agreement between numeri- comparisons with velocity and void fraction data at gas
ca!!y predicted and experimental velocities in gas-water flow rates (lo’- IO’ cm3/s) characteristic of the above
experiments have been made by Schwarz and co-work-
ers. They too ignored the top space but found it nec-
Address reprint requests to Dr. Davidson at CSIRO Division of essary to include interphase diffusion terms in the mass
Mineral and Process Engineering, Lucas Heights Research Labo-
ratories, Lucas Heights, N.S.W., 2232. Australia.
balance equations to obtain the correct spreading of
the plume.
Received 30 November 1988; accepted 8 September 1989 Although there are arguments to support the inclu-

0 1990 Butterworth Publishers Appl. Math. Modelling, 1990, Vol. 14, February 67
Two-phase flow in a liquid with bottom gas injection: M. R. Davidson

sion of interphase mass diffusion terms in a two-fluid Brimacombe13 (who measured the gas fraction and gas
model, such terms do not occur in existing two-fluid velocity in the plume) in a calculation covering both
model equations derived by averaging the microscopic the liquid and gas top space regions. The predicted
flow equations (see, for example, Refs. 14-16). In these steady void fraction and gas velocity distributions are
models, interpenetration of the phases occurs solely compared with the corresponding time-averaged ex-
by the advection of each phase separately according perimental data. Prediction of bath circulation using
to its own velocity field and is determined by the cou- this model is considered elsewhere.23
pling of phase interaction effects with the advection
and diffusion of momentum; only in the corresponding
mixture formulation of two-phase flow do terms appear
The model
in the continuity equations that are interpreted as dif-
fusion of the dispersed phase relative to the mixture. We consider isothermal, axially symmetric flow in a
An interaction force that may account for the ob- cylindrical vessel (shown schematically in Figure I) of
served spreading of the plume during bottom gas in- radius CIand height H having an initial depth h of liquid
jection, and that is not included in current models of and a centrally located orifice of radius r0 at the bottom
gas-agitated flows, is particle lift. This is an inertial through which gas is injected. Although the assumption
effect, which has been observed and analyzed for of axial symmetry is expected to be valid for the ex-
Poiseuiffe flow in tubes, whereby a spherical particle periment simulated here, it is not valid in general be-
travelling more quickly (or slowly) than the local fluid cause of the presence of surface waves and a preces-
velocity migrates away from (or toward) the faster- sion of the plume about the centerline. (Castillejos and
moving fluid at the centerline (see Ref. 17, p. 229). A Brimacombe mounted an annular baffle near the bath
similar lift force has also been predicted theoretically surface to dampen such motions.)
in unbounded simple shear flow (see the review by The equations governing the transient motion of each
Lawler and Lur8). Further, Drew and Laheyr5 and phase have the following general form,r6 assuming that
DrewI have presented generalized expressions for the each phase is incompressible and no phase change oc-
phase interaction force (including lift), and Drew and curs:
Laheyi9 have shown that inclusion of the lift force
together with the drag and their form of the virtual 2 + V*(Ly& = 0 (continuity) (1)
mass force is necessary for the combination to reduce
to that calculated for a single sphere. A familiar ex-
a~kpkuk
ample, which may well illustrate the action of lift, is ___ + v~(~&p&u&u&)
the suspension of a sphere in a vertical jet; since the at
sphere is either stationary or moving more slowly than = -~&VP& - ffkpkgf + (Pki - P~IV~I,
the fluid, any tendency for it to topple will be countered + Mx + v-(a& (momentum) (2)
by a restoring force toward the faster-moving fluid at
the center. Here p, cy, II, p, 7, g, and 2 denote density, volume
An inter-facial force associated with bubble disper- fraction, velocity vector, average pressure, viscous
sion is also of likely importance in the plume, where stress tensor (including turbulent stresses), gravita-
gradients in void fraction can be significant. Such a tional acceleration, and the upward vertical unit vec-
diffusive force, proportional to the gradient of void
fraction, has been proposed by Drew and Lahey.‘O
More recently, Lee and Weisler2’ described the phys-
ical mechanisms of particle dispersion in turbulent flow;
GAS TOP SPACE

1
the instantaneous drag on a fluid eddy is greatest, and
hence the amplitude of the oscillation is least, when
the eddy is moving in the direction of increasing par-
ticle concentration. The uneven amplitude of such os-
cillations results in the ratchetlike motion of entrained
particles away from the region of high particle con- H
centration. In general, the degree of particle dispersion
will depend on the concentration gradient, the fre-
quency and amplitude of oscillation, and the instan-
taneous drag force, which in turn depends on the par-
ticle/fluid properties.
In this paper the modelling of two-phase flow in a 1
liquid bath resulting from centered gas injection at the
bottom is reconsidered, with attention focused on the
plume, and the effect of including the particle lift and -220
diffusive inter-facial force is investigated. The transient,
two-phase fluid flow computer program K-FIX22 is Figure 1. Schematic representation of the liquid bath and the
used to simulate an experiment of Castillejos and central plume

68 Appl. Math. Modelling, 1990,Vol. 14, February


Two-phase flow in a liquid with bottom gas injection: M. R. Davidson

tor, respectively. Subscript k refers to either the gas function of void fraction, bubble diameter, fluid prop-
(k = g) or liquid (k = 1) phase, and erties, and relative velocity. The second term describes
the virtual mass force that arises because the accel-
cryx+ cY/= I (3) eration of gas bubbles must overcome the inertia of
The pressure pki represents the average pressure of displaced liquid. The corresponding acceleration artm
phase k at microscopic phase interfaces, and MA con- is given in equation (10) in the frame-indifferent (math-
sists of phase interaction terms (for example, drag, ematically objective) form derived by Drew et al.‘x The
virtual mass, particle lift, diffusive force), where last term in equation (9) represents the lift force that
arises from the relative motion of the phases in cross-
M,= -M, (4) stream velocity gradients. (Note that Drew and Lahey’”
The term (pAi - pn)Vak in equation (2) has also been recently redefined the virtual mass and lift terms, keep-
derived by Pauchon and Banerjee,‘4 who noted that ing the sum of these terms unchanged; however, their
earlier partitioning is used in equation (9).) Here C,,,
Pxi - Pr = 0 (3
is the virtual mass coefficient, A is a parameter, and L
is a good approximation for gas bubbles and showed is the lift coefficient. Also, the lift term is written in
that terms of gradients of the mixture velocity (u,,) rather
than the liquid velocity; this is discussed later. For an
P/i - Pl = -&dug - U/I2 (6) isolated, rigid spherical bubble,”
with 5 = $ for potential flow around an isolated sphere
c,,,,, = a A=2 L = - 2C, ,,@I (12)
of constant radius. A similar expression has also been
proposed by Stuhmiller.25 Since pli - pl should ap- corresponding to the limit of low void fraction. Note
proach zero at high void fraction, we set that this expression for L differs by a factor of 2 from
that given in earlier work ‘h,2hbased on less convincing
I$= t( 1 - (Y&J (7) arguments. We now consider each term comprising M,
Assuming that equations (5) and (6) apply and neglect- and the viscous stress term in equation (2) in more
ing surface tension (relative to the much larger dynamic detail.
pressures) so that
Pai = P/i (8)
determines all but one of the four unknown pressures, Drug coefjcirnt
the remaining pressure forming part of the solution of
Following Cook and Harlow,2y we set
equations (1) and (2).
Problems with numerical stability have plagued the
calculation of two-phase flo~s,~~ and in recent years, (13)
considerable effort has been devoted to the accurate
modelling of phase interaction effects as a means of where C,) is the drag coefficient for a single bubble
stabilizing numerical calculations. Apparently unim- and rh denotes bubble radius. Typically, large bubbles
portant physical effects can become important at short generated at the orifice break up to form a stable dis-
wavelengths, thus preventing their rapid growth and tribution of bubble sizes, bubbles at the edge of the
the consequent onset of instability.” Here we assume plume being smaller than those at the center.“’ How-
the following form for M,, based on the work of Drew ever, in the present work we ignore the distribution of
and Laheyrs and the review by Drew’? bubble sizes and factors such as bubble breakup and
coalescence by assuming a uniform value of Y,,.Equa-
M, = K u/ - u, - Drf(Vap + ax C,-,dvb,,, tion (I 3) is simply the expression for K for spherical
*I: ) bubbles at low void fraction multiplied by (I - CQ,)to
give the correct limit at high void fraction. This form
+ a,L(u, - u,1.; mm + (Vu,,JT) (9) for K gave better overall results than the alternative
of using the detailed mixture-based correlations of Ishii
where and Zuber” for steadily rising interacting bubbles.
From Ishii and Zuber” the drag coefficient of a
ar:m= ($+n,Cu,) - (%+u,.~u,)
single bubble in a liquid, corresponding to the various
flow regimes, is
+ (1 - h)(u, - u,).V(u, - u,) (IO)
Cnl = g(l + 0.1Re0.75)
and
(viscous regime) (14a)
nl?l = Lygug+ a/u/ (1 I) 2112
a-4 o.25
The first term in equation (9) represents the interfacial C,,, = - Re 7
3 ( P/c- )
drag force in which -D, V~,/(Y, denotes the appar-
(distorted particle regime) (14b)
ent mean drift velocity of liquid relative to the gas
phase due to turbulent dispersion and where K is a CL)3= B (churn turbulent regime) (14c)

Appl. Math. Modelling, 1990, Vol. 14, February 69


Two-phase flow in a liquid with bottom gas injection: M. R. Davidson
where title lift by showing that the combined expressions for
these two forces, with parameters given by equation
C* = C,’ CD, 5 C,’ (I5a) (12), reduce to the correct result for a single sphere.
= CIX CD’ 5 CLX5 C,, (I5b) In this paper we set the lift coefficient to be
= CD3 CD3 5 CD2 (15c) L = -p,(l - (Yg) (18)
Here we have used Ishii and Zuber’s expression for consistent with equation (12) at low void fraction. The
relative velocity to write equation (14b) in terms of factor 1 - (Yeensures that L, like C,,, is zero when
dimensionless groups, g denotes the gas-liquid inter- only gas is present.
facial tension, pI denotes the liquid viscosity, and
Re = 2~,,&~ - u#p, (16) Viscous stress
is the bubble Reynolds number. Note that, as the Rey- One possible choice for the turbulent stresses T/, is
nolds number increases, the drag coefficient initially that they have the familiar Newtonian form
decreases according to the standard drag curve for rigid
Tk = pkE(vuk + (VUk)? (19)
spherical particles (equation (14a)). It then increases
(equation (14b)), as shape changes in irregularly mov- where &E is the eddy viscosity of phase k. However,
ing bubbles increase the effective bubble cross-section, in general, there are additional terms’” that account for
before reaching a limiting value (equation (14~)). an induced eddy pressure, bubble expansion or con-
traction, and nonviscous distribution effects associated
Virtual mass force und particle lif with the averaging process. Although similar types of
Virtual mass terms have a stabilizing effect on the such terms recur in the literature,‘“.35 there is no agree-
numerical calculation of two-phase flows by altering ment on their precise form. Such additional terms are
the eigenvalues of the equation system; however, since not included in the present work. In view of the un-
these terms are important only at high frequencies and certainty surrounding the proper modelling of the tur-
during rapid acceleration, they do not significantly af- bulent stresses in two-phase flow, we simplify the vis-
fect the final results in most cases.‘h,28 For example, cous terms even further by setting
Lahey et al.‘2 found that the inclusion of virtual mass V’(akTk) = V.tffk~kEVUk) (20)
terms reduced the computation time of a one-dimen-
sional steady-state calculation by a factor of 40. Con- for two-dimensional, axisymmetric flow. This corre-
sistent with this, inclusion of virtual mass terms in the sponds to using equation (19) but ignoring those terms
present work is found to be necessary to ensure sta- that become zero for constant (Y,#kEand v.uk = 0. We
bility of the initial value problem. Stability will be dis- further assume that the turbulent kinematic viscosities
cussed in more detail later. (&E/PA, k = g, f) are equal and uniform throughout
As was indicated earlier, the virtual mass coefficient the bath and are given by the mean value estimatkc by
c L’m= 4 for a single spherical bubble, corresponding Sahai and Guthrie36 as
to the limit of low void fraction. However, this value 5.5 x 10-3h (‘1 -2ad)gB)“3
of C,, ignores the effect of (nonspherical) bubble shape.
It also ignores details of the interaction between bub-
bles. For identical spherical bubbles, van Wijngaarden33 where < is a mean plume void fraction. Although he
showed that C,, < & when one bubble travels in the more sophisticated k-E turbulence model has been used
wake of another, whereas C,,, > t when two bubbles in this application with some success,‘.” similar (and
move side by side. He also showed that for a random only slightly less accurate) results are obtained by us-
distribution of bubbles, C,, > t and increases with (Ye ing the simpler constant viscosity model.”
for (Ye << 1. However, for (Ye = I (gas only), C,,
should be zero. We therefore set Diffusive interfacial force
C”,, = C,(l - “J C,Li (17) The diffusive force on the gas phase is assumed to
have the form
Drew et al.28 conclude that the parameter A in their
objective form of the virtual acceleration (equation (IO)) -&FVa,
satisfies 0 5 A 5 2, since the values 0 and 2 correspond (21)
Luw
to the limiting cases of a single droplet (high void frac-
tion) and single bubble (low void fraction), respec- which is the product of the drag function K and an
tively. In practice, results are found to be insensitive apparent mean drift velocity of the liquid relative to
to the value of h.16 Cook and Harlow34 derive a dif- the gas phase. Here Deff is an effective bubble disper-
ferent objective formulation of the virtual mass force sion coefficient. The form of (21) is similar to that
that involves only one parameter (analogous to C,,), assumed by Lee. 37 Expression (21) becomes
but this is not considered here.
Recently, Drew and Lahey19 confirmed the principle - 5CDC-UIP,(UR rb
D,,V

ffY
of objectivity and added further support for their gen-
eralized formulation of the virtual mass force and par- upon substitution of equation (13). The effective bub-

70 Appl. Math. Modelling, 1990, Vol. 14, February


Two-phase flow in a liquid with bottom gas injection: M. R. Davidson
ble dispersion coefficient Oeff is set equal to the (as- The program K-FIX is based on an implicit multi-
sumed) uniform turbulent kinematic viscosity. tield solution method,4? which reduces to the ICE
technique4’ for single-phase flow. The method uses a
Well-posedness and stability staggered mesh with pressure and void fraction defined
It is well known that the simple one-dimensional at the center of computational cells and with normal
equal pressure model, with the right side of equation velocities at cell walls defined at their midpoints. Do-
(2) consisting only of the pressure gradient term, has nor cell differencing of the advection terms is used,
complex characteristics so that the initial value prob- and at each time step the pressure is adjusted to ensure
lem is ill-posed.” The same is true of many more- conservation of mass by a point relaxation technique.
detailed two-phase flow models.3” Despite this, appar- The boundary conditions chosen in the present work
ently stable numerical solutions of such systems can are now described. At solid boundaries, wall functions
often be obtained; reasons for this have been discussed are often used to model the near-wall shear stress to
by Lyczkowski et al.,3H Stewart,39 and more recently avoid resolving the boundary layers. However, if the
Stewart and Wendroff” in a review. However, a soundly computational cell dimensions are much greater than
based two-phase flow model should be fundamentally the boundary layer thickness (large Reynolds number
stable, and, to test this, it is only practical to consider limit), then a free slip boundary condition is appro-
one-dimensional flow. priate. In this paper we assume free slip of both gas
Jones and Prosperettl m40.41 have derived linear sta- and liquid at solid boundaries; this should at least be
bility conditions for a general class of one-dimensional acceptable in a simulation aimed at predicting the flow
two-phase models by considering perturbations to the in the plume. The normal velocity of each phase and
uniform steady flow solution. (However, note that the the normal gradient of the void fraction are also set to
first occurrence of k* in condition (24) of their 1987 zero at solid boundaries. At the top boundary of the
paper is missing a harmonic mean density factor.) An computation domain (z = H) we assume that V, = 0
expected consequence is that when viscous terms are (that is, liquid is prevented from passing through),
omitted, reality of the solution characteristics is a nec- aV,/az = 0, and &,I& = 0. The boundary condition
essary (but not sufficient) condition for stability. It is at the orifice is complicated by the formation and de-
straightforward but tedious to show that the one- tachment of bubbles there and the coarseness of the
dimensional form of the model used in the present work radial grid spacing (Ar) relative to the orifice radius;
has real characteristics only and is also stable for all these matters are discussed below.
values of A 5 2 when the drag coefficient is indepen- A substantial body of literature exists on bubble
dent of velocity and the turbulent kinematic viscosity formation at a submerged orifice under a variety of
is uniform. Conditions for stability, when the lshii and conditions (Ref. 17, p. 323). Recently, Engh and
Zuber drag coefficient (equations (14) and (15)) is used, Ni1mani44 made an experimental and theoretical study
depend on the flow variables and are not considered of bubble formation during high-velocity gas discharge
explicitly here. for which gas momentum causes elongation of the de-
The lift terms in equation (9) do not enter into the veloping bubbles. The cycle of bubble formation was
stability analysis because, from equation (I), the mix- observed to begin with the growth and detachment of
ture velocity is independent of position in one-dimen- a large bubble followed by the development of smaller
sional flow. However if a lift term, based on liquid bubbles, which rapidly coalesce into the preceding large
rather than mixture velocity, is used, the one-dimen- bubble before they can develop independently. They
sional problem is stable only when the virtual mass predicted that the bubble frequency f is given by
fraction C, is large enough. The notion that the lift
force depends on the mixture velocity gradients at in- (23)
termediate void fractions is appealing. Of course, in
the limit of low void fraction, II,,, and II! are approxi- independent of flow rate Q (constant frequency re-
mately equal, and the correct expression for a single gime). They also plotted the diameter ratio (d,ld,) of
bubble is obtained.
the major and minor axes of the assumed ellipsoidal
bubble as a function of the dimensionless group
Numerical considerations
N = P,~Q
Equations (1) and (2) for the gas and liquid phases,
incorporating equations (3)-(I I), (13)-(18), and (20)-(22),
pm%
are solved numerically for the velocities (Us = (U,, V,), which can be used to determine d, and d,, since the
k = g, 0, pressure, and void fraction on a uniform bubble volume is given by
mesh in cylindrical coordinates (r, z) with the origin
-e = $jd;
centered at the orifice. The K-FIX computer code”
for calculating transient two-fluid flow was used after f 6 (25)
adaption to include the terms in equation (2) involving If we assume that the large bubble detaches at a time
the void fraction gradient, virtual mass, and particle equal to two thirds of the total cycle time (estimated
lift; the virtual mass force was incorporated by follow- from Figure 3 in Ref. 44), then the bubble rise velocity
ing Cook and Harlow.29 (V,,) at the orifice is

Appl. Math. Modelling, 1990, Vol. 14, February 71


Two-phase flaw in a liquid with bottom gas injection: M. R. Davidson

V, = $fd, (26) From Figure 2 the spread of the plume is well pre-
dicted at all heights; the centerline void fraction is also
Since the orifice diameter has only an indirect effect
predicted within IO%, very close agreement occurring
on the bubble velocity at the orifice through equations
at heights exceeding 10 cm. However, for the 28 x 48
(23)-(26), we assume that the gas enters the bath with
grid the shape of the calculated void fraction profile
the bubble velocity V, through an apparent orifice hav-
progressively deviates from the observed Gaussian shape
ing a diameter equal to the bubble diameter d, and that
with increasing height as the profile maximum moves
the void fraction (Ye,immediately behind the orifice is
away from the centerline. (Offset void profile maxima
given by
are expected when the plume precesses, but this does
Q not occur here.) Coincidently, predicted vertical bub-
% = 0.25V,rd; ble velocities agree well with the corresponding ex-
perimental centerline values in the upper half of the
with the vertical liquid velocity at the orifice set to plume. However, they are too low away from the cen-
zero. Substituting equations (25) and (26) into (27) gives terline (Figure 3); this is attributed to the overpre-
a0 = 4/9. The above approach ensures that the correct diction of off-center void fractions, since the steady-
gas volume and momentum enter the bath and is similar state volume flux of gas through the plume must be
to the approach taken by Durst et al.* for very slow constant. Errors in the volumetric gas flow rate, cal-
bubbling. Note that the calculated void fraction im- culated at each height shown in Figures 2 and 3 by
mediately above the orifice depends on the total gas integrating the product of the void fraction and vertical
flux and the local flow and will usually be greater than gas velocity over the bath cross-section, were found
a,, . to range from 0.02% to 4%. Refinement of the mesh
We simulate an air-water experiment of Castillejos in a calculation that omits the gas top space (to mini-
and Brimacomber3 for which Q = 876 Ncm3/s, a = mize computation time) shows that the solution pre-
25 cm, h = 40 cm, and 2r,, = 0.635 cm, and we set sented here for the 28 x 48 grid is grid independent.
H = 60 cm. The actual flow rate Q,) at operating con- When the coarser 14 x 24 grid is used, rather than the
ditions (assumed to be atmospheric pressure plus the 28 x 48 grid, the central dimple in the calculated void
hydrostatic head at 20°C) is 905.15 cm’/s. Calculations fraction distribution disappears, so the void fraction
are performed by using 14 x 12, 14 x 24, and 28 x profiles are almost geometrically similar and Gaussian,
48 internal grids. In the first two cases, Ar = 1.7857
cm and is approximately equal to id,. We set the virtual
mass parameters C, = 1 and A = I .8 and the bubble
radius rb = 0.35 cm, corresponding to the mean bubble
radius found experimentally at this flow rate.‘j
Initially, the bath is assumed to be stagnant with the
gas jet being turned on at time t = 0+ . We set the
time step At = 10d4 s for the first 100 steps and At =
10mi s thereafter. Time stepping is continued until there
is no apparent change in the void fraction and velocity
patterns over a 0.2-s time period. At each time step
after the initial few the internal pressure iteration is

r---lT-l
z=14cm z=19cm
continued until equation (1) is satisfied everywhere to 08
?3
within 10P3; this ensures that the calculated difference 5 0.6
between the overall gas flow rates in and out is ac- d
y 0.4
ceptably small (O-5%). However, in the first few time D
steps the error in the pressure iteration is relaxed to 9 0.2
lo-‘-lo-* to reduce the computational effort associ- 00
ated with rapid changes as the gas jet is turned on.

Comparison between theory and experiment


Calculated radial profiles of void fraction and bubble
rise velocity at various heights above the orifice are
compared with the corresponding experimental data of
Castillejos and Brimacomber3 in Figures 2 and 3. The
0.0
velocity data used in Figure 3 corresponds to flow from -60 00 60 -6.0 0.0 60 -60 00 60
a smaller orifice diameter (0.41 cm) than is simulated RADIUS (cm)
here. However, measured bubble velocities for the two
orifice sizes coincide at both the centerline and the Figure 2. Void fraction profiles at various heights above the
edge of the plume at heights exceeding 10 cm; we orifice, calculated by using a 28 x 48 grid (solid curves) and a
14 x 24 grid (dashed curves), compared with the corresponding
therefore assume that the radial distribution of such data of Castillejos and Brimacombe.‘3 Data on the left of the
velocities will also coincide at these heights. centerline are taken to be the mirror image of those on the right

72 Appl. Math. Modelling, 1990, Vol. 14, February


Two-phase flow in a liquid with bottom gas injection: M. R. Davidson

and predicted off-center void fractions are almost un-


changed (Figure 2). Also unchanged are the bubble rise
velocity profiles at heights exceeding 10 cm (Figure 3).
Efforts to produce a Gaussian void fraction profile on
the 28 x 48 grid by increasing the eddy viscosity failed
to maintain the excellent agreement with the data on
the centerline. In particular, increasing the eddy vis-
240 0 ,‘I.,, ‘,I I * cosity by a factor of about 10 eliminated the dimpling
z=24cm of the void fraction profile but produced centerline
140.0
values of void fraction that were too high and bubble
40.0
velocities that were everywhere too low.
Figure 4 shows that ignoring the particle lift and
diffusive forces results in centerline void fractions that
are unrealistically high. It also shows that, as the mesh
is refined, the calculated variation in centerline void
fraction with height above the orifice approaches the
corresponding data of Castillejos and Brimacombe more
closely.
In Figure 5 the dimensionless half-value radius of
240 .O the plume r,&lQY2 is plotted against height mea-
140 .o sured in orifice diameters. At a given height, yIjz de-
40 .o fines the radius at which the void fraction falls to half
its centerline value and is a useful measure of plume
240 .O
width when the centerline value is also the profile max-
140.0 imum. The dashed line represents a correlation derived
40 .o by Castillejos and Brimacombe from their data. It is
240 .O clear that when the particle lift and diffusive forces are
ignored, no spreading of the plume occurs. When these
140.0
forces are included, calculated values of rl12 are in
40.0
-80 -4.0 0.0 4.0 8.0 acceptable agreement with the data over most of the
r(cm1 bath height. Major percentage differences occur close
to the orifice because of the steepness of the void pro-
Velocity profiles of vertical gas velocity at various files there. Since the void fraction profiles for the
heights above the orifice, calculated by using a 28 x 48 grid
(solid curves) and a 14 x 24 grid (dashed curves), compared
28 x 48 grid have maxima that do not occur on the
with data of Castillejos and BrimacombeT3 for a smaller orifice centerline, and hence are not Gaussian or even geo-
diameter but with other conditions identical (see text for justi- metrically similar as is observed, the corresponding
fication). Data on the left of the centerline are taken to be the r,,? curve does not fit the data as well as that for the
mirror image of those on the right

0.8
5
E
a 0.6
E
0

5 0.4 or diff. force)


S
z
2 0.2
IO-'!
E
IO’ IO2
0.01 I I I t I I I J z12r,
0.0 10.0 20.0 30.0 40.0
z (cm) Figure 5. Dimensionless half-value radius of the plume as a
function of height measured in orifice diameters, predicted by
Figure 4. Centerline void fraction as a function of height above using different grid sizes (solid curves) and compared with the
the orifice, calculated by using different grid sizes and compared experimental correlation (dashed line) of Castillejos and
with the data of Castillejos and Brimacombe13 Brimacombe13

Appl. Math. Modelling, 1990, Vol. 14, February 73


Two-phase flow in a liquid with bottom gas injection: M. R. Davidson
coarser 14 x 24 grid, which does achieve geometrically
similar void profiles by numerical smearing.
Centerline bubble rise velocities as a function of
height above the orifice are shown in Figure 6. When
the particle lift and diffusive forces are ignored, the
calculated gas velocity is unrealistically high in the
upper half of the bath, whereas satisfactory agreement
with the data is achieved there when these forces are
included. In all cases the predicted gas velocity is too
small in the lower half of the bath, although results
based on the 28 x 48 grid are better than those derived
by using coarser grids. The reasons for this discrep-
ancy are not clear. Factors that may contribute include
discretization errors near the orifice, errors in mod-
elling the drag force under forced flow conditions with
bubble breakup, and the arbitrary choice of a simple
linear dependence of the lift coefficient on void frac-
tion.
In view of the present incomplete resolution of the
-
void profile shapes, there seems little point in display-
ing the time-varying void contour map of the advancing (b)
plume, and we confine ourselves to steady-state re- Figure 7. Steady-state void fraction contours predicted using
sults. However, even the calculated steady state only (a) a 28 x 48 grid and (b) a 14 x 24 grid
models the time average of the presumably steady cyclic
behavior of the plume that occurs in practice. Figure
7 compares the steady-state void fraction contours pre-
dicted by the 28 x 48 and 14 x 24 grids, showing the tions are formally stable under simplified conditions
dimpling of the central contour pattern for the fine grid and are necessary to ensure numerical stability for the
and the expected qualitative shape for the coarser grid. present calculations. It is suggested that already stable
calculations (based on some other algorithm) will be
accelerated by the inclusion of the above terms.
Conclusion
It is shown that the particle lift and diffusive inter-
The bottom injection air-water experiment of Castil- facial forces can account for the observed spreading
lejos and Brimacombe13 is simulated by using an adapted of the plume, the vertical variation in the centerline
form of the two-phase fluid flow program K-FIX to void fraction, and the magnitude of the bubble rise
model flow in both the bath and the top space. Gov- velocity in the upper half of the bath. Physical param-
erning equations are chosen that include terms ac- eters such as particle lift are chosen a priori, based on
counting for the pressure difference between micro- results for a single bubble, and no adjustment is made
scopic phase interfaces and the bulk phases, a diffusive to obtain agreement with the data. However, although
interfacial force, and the virtual mass and particle lift coarse grid calculations give void profiles having the
forces in the mathematically objective form given by expected centerline maximum, fine grid calculations
Drew and Lahey. I9 The above pressure and virtual produce off-center void fraction maxima but neverthe-
mass terms together are found to ensure that the equa- less give better agreement with the centerline void data.

14~12(without lift or diff. force)

0.0 10.0 20 .o 30.0 40.0


2 (cm1
Figure 6. Centerline vertical gas velocities as a function of height above the orifice, calculated by using different grid sizes and
compared with the data of Castillejos and Brimacombe’3 (squares on the centerline; circles at the plume edge)

74 Appl. Math. Modelling, 1990, Vol. 14, February


Two-phase flow in a liquid with bottom gas injection: h4. R. Davidson

Notation References
a radius of the bath I Szekely, J., Wang, H. J. and Kiser, K. M. Flow pattern ve-
locity and turbulent energy measurements and predictions in
a“m virtual acceleration a water model of an argon-stirred ladle. Metall. Trans. B 1976,
CL) drag coefficient 7B, 287-295
Ct?? virtual mass constant in equation (17) 2 Deb Roy.T., Mazumdar, A. K. and Spalding, D. 9. Numerical
c urn virtual mass coefficient prediction of recirculation flows with free convection encoun-
tered in gas-agitated reactors. Appl. Math. Mode//kg 1978, 2,
4, d, major and minor axes of assumed ellip-
146-150
soidal bubbles at the orifice 3 Sahai. Y. and Guthrie, R. I. L. Hydrodynamics of gas stirred
effective bubble dispersion coefficient melts. Part II: Axisymmetric flows. Metall. Trans. B 1982,13B,
F bubble frequency at the orifice 203-2 I I
4 Grevet, J. H., Szekely, J. and El-Kaddah. N. An experimental
F? gravitational acceleration
and theoretical study of gas bubble driven circulation systems.
h initial depth of liquid Internat. .I. Heat Mass Transfer 1982. 25, 487-497
H height of liquid plus top space 5 Mazumdar, D. and Guthrie, R. I. L. Hydrodynamic modelling
K drag function of some gas injection procedures in ladle metallurgy operations.
L lift coefficient Metal/. Trans. B 1985, 16B, 83-90
interfacial forces on phase k 6 McKelliget, J. W., Cross, M. and Gibson, R. D. A turbulent
Mk fluid flow model of gas agitated reactors. Appl. Math. Mod-
N dimensionless group defined by equation el/inn 1982, 6, 469-480
(22) 7 Cross, M., Markatos, N. C. and Aldham, C. Gas injection in
Pk, Pki bulk pressure and pressure at micro- ladle processing. CONTROL ‘84: Proceedings of‘ the First In-
scopic interfaces, respectively, in ternaiional Symposium on Automutic Control in Mineral Pro-
cessinP and Process Metalluwv. AIME-SME, 1984, PP. 291-297
phase k 8 Durst:‘F., Schonung, B., Sel&ger, K. and Winter, .G. Bubble-
Pn7Pni bulk pressure and pressure at micro- driven liquid flows. J. Fluid Mech. 1986, 170, 53-82
scopic interfaces, respectively, in the 9 Koh, P. T. L., Markatos, N. C. and Cross, M. Numerical
gas phase simulation of gas-stirred liquid baths with a free surface. Phys-
ice-Chemical Hydrodynumics 1987, 9, 197-207
Pl? P/i bulk pressure and pressure at micro-
10 Schwarz, M. P.. Wright, J. K. and Baldock, B. R. Measure-
scopic interfaces, respectively, in the ment and simulation of the flow in an iron bath stirred with
liquid phase bottom-injected nitrogen. Muthematical Modelling of Mute-
Q, Qo volumetric gas flow rate at STP and at riuls Processing Operations. The Metallurgical Society, War-
operating conditions, respectively rendale. PA, U.S.A., 1987. pp. 565-579
II Schwarz. M. P. and Turner, W. J. Applicability of the standard
r, Ar radius and radial grid spacing, respec- k-6 turbulence model to gas-stirred baths. Appl. Muth. Mod-
tively elling 1988, 12, 273-279
rb mean bubble radius 12 Schwarz, M. P. Two and three dimensional models of a gas-
r, orifice radius stirred bath of molten pig iron. PHOENICS J. 1989, 1, 282-
half-value radius of the plume 310
rI/2
13 Castillejos, A. H. and Brimacombe, J. K. Structure of turbulent
Re bubble Reynolds number defined by gas-liquid plumes in vertically injected jets. SCANINJECT IV:
equation (16) Proceedings of the Fourth International Conference on lnjec-
time and time step, respectively tion Metallurgy, Lulea. Sweden. 1986, pp. 16:l-16:34
velocity of phase k 14 Ishii, M. Thermo-Jluid Dynumic Theory of Two-Phase Flow.
Eyrolles. Paris, 1975
velocity of gas and liquid phases, and I5 Drew, D. A. and Lahey, R. T. Application of general consti-
mixture, respectively tutive principles to the derivation of multidimensional two-
u,, vk radial and vertical velocity components, phase flow equations. Internut. J. Multiphase Flow 1979, 5,
respectively, of phase k 243-264
bubble rise velocity at the orifice 16 Drew, D. A. Mathematical modelling of two-phase flow. Ann.
Rev. Fluid Mech. 1983. 15, 261-291
height above the orifice and vertical grid 17 Clift, R., Grace, J. R. and Weber, M. E. Bubbles, Drops, und
spacing, respectively Particles. Academic, London, 1978
upward vertical unit vector 18 Lawler, M. T. and Lu, P. The role of lift in the radial migration
volume fraction of phase k of particles in a pipe flow. Advances in Solid-Liquid Flow in
Pipes and its Application, ed. I. Zandi. Pergamon, New York,
volume fraction of gas and liquid phases, 1971, pp. 39-57
respectively I9 Drew, D. A. and Lahey, R. T. The virtual mass and lift force
mean plume void fraction on a sphere in rotating and straining inviscid flow. Internat. J.
apparent void fraction behind the orifice Multiphase Flow 1987, 13, 113-121
parameter in equation (10) 20 Drew. D. A. and Lahey, R. T. Interfacial dissipation in two-
phase flow. Basic Mechanisms in Two-Phase Flow and Heat
liquid viscosity Transfer. ASME, Chicago, 1980, pp. 47-51
eddy viscosity of phase k 21 Lee. S. L. and Wiesler, M. A. Theory on transverse migration
gas-liquid interfacial tension of particles in a turbulent two-phase suspension flow due to
density of phase k turbulent diffusion-l. Internat. J. Multiphase Flow 1987, 13,
99-111
density of gas and liquid phases, respec-
22 Rivard, W. C. and Torrey, M. D. K-FIX: A computer program
tively for transient, two-dimensional, two-fluid flow. Rept. No.
viscous stress tensor LA-NUREG-6623, Los Alamos Scientific Laboratory, 1977
given by equation (7) 23 Davidson, M. R. Numerical simulation of two-phase flow aris-

Appl. Math. Modelling, 1990, Vol. 14, February 75


Two-phase flow in a liquid with bottom gas injection: IV. R. Davidson
ing from bottom injection of gas into liquid. CTAC 89, Pro- 33 van Wijngaarden, L. Hydrodynamic interaction between gas
ceedings of the International Conference on Computational bubbles in liuuid. J. Fluid Mech. 1976. 77. 27-44
Techniques and Applications, ed. W. L. Hogarth and B. L. 34 Cook, T. L.-and Harlow, F. H. Virtual mass in multiphase
Noye, Brisbane, Australia, July 10-12, 1989 flow. Internat. J. Multiphase Flow 1984, 10, 691-696
24 Pauchon, C. and Banerjee, S. Interphase momentum interac- 35 Biesheuvel, A. and van Wiineaarden. L. Two-phase flow equa-
tion effects in the averaged multifield model. Part 1: Void prop- tions for a dilute dispersion\f gas bubbles in liquid. J. Fluid
agation in bubbly flows. Internat. J. Multiphase Flow 1986, 12, Mech. 1984,148,301-318
559-573 36 Sahai, Y. and Guthrie, R. I. L. Effective viscosity models for
25 Stuhmiller, J. H. The influence of interfacial pressure forces gas stirred ladles. Metall. Trans. B 1982, 13B, 125-127
on the character of two-phase flow model equations. Internut. 37 Lee, S. L. A unified theory of particle transport in a turbulent
J. Multiphase Flow 1977, 3, 551-560 dilute two-phase suspension flow-II. Internut. J. Multiphase
26 Stewart, H. B. and Wendroff, B. Two-phase flow: Models and Flow 1987, 13, 137-144
methods. J. Comput. Phys. 1984, 56, 363-409 38 Lyczkowski, R. W. Characteristics and stability analyses of
27 Ramshaw, J. D. and Trapp, J. A. Characteristics, stability, and transient one-dimensional two-phase flow equations and their
short-wavelength phenomena in two-phase flow equation sys- finite difference approximations. Nucleur Sci. Engrg. 1978,66,
tems. Nuclear Sci. Engrg. 1978, 66, 93-102 378-396
28 Drew, D. A., Cheng, L. and Lahey, R. T. The analysis of 39 Stewart, H. B. Stability of two-phase flow calculations using
virtual mass effects in two-phase flow. Internut. J. Multiphuse two-fluid models. J. Comnut. Phvs. 1979. 33. 259-270
Flow 1979, 5, 233-242 40 Jones, A. V. and Prosperetti, A. On the suitability of first-
29 Cook, T. L. and Harlow, F. H. VORT: A computer code for order differential models for two-phase flow prediction. Inter-
bubbly two-phase flow. Rept. No. LA-10021-MS. Los Alamos nut. J. Multiphase Flow 1985, 11, 133-148
National Laboratory, 1983 41 Prosperetti, A. and Jones, A. V. The linear stability of general
30 Tacke, K. H., Schubert, H. G., Weber. D. J. and Schwerdt- two-phase flow models--II. Internut. J. Multiphnse Flow 1987,
feger, K. Characteristics of round vertical gas bubble jets. Me- 13, 161-171
tal!. Truns. 3 1985, 16B, 263-275 42 Harlow, F. H. and Amsden. A. A. Numerical calculation of
31 Ishii, M. and Zuber, N. Drag coefficient and relative velocity multiphase fluid flow. J. Comput. Phys. 1975, 17, 19-52
in bubbly. droplet or particulate flows. AIChE J. 1979, 25, 43 Hat-low, F. H. and Amsden, A. A. A numerical fluid dynamics
843-855 calculation method for all flow speeds. J. Comput. Phys. 1971,
32 Lahey, R. T., Cheng, L. Y., Drew, D. A. and Flaherty, J. E. 8, 197-213
The effect of virtual mass on the numerical stability of accel- 44 Engh. T. A. and Nilmani, M. Bubbling at high flow rates in
erating two-phase flows. Internat. J. Multiphase Now 1980,6, inviscid and viscous liquids (slags). Metall. Truns. B 1988. 19B,
281-294 83-94

76 Appl. Math. Modelling, 1990, Vol. 14, February

Вам также может понравиться