A Mathematical Model of Aluminum Depth Filtration with

Ceramic Foam Filters: Part I. Validation for Short-Term

Filtration
F.A. ACOSTA G. and A.H. CASTILLEJOS E.

This work presents a mathematical model to compute the efficiency of depth filtration of molten
aluminum using ceramic foam filters. In the model, the porous structure of foam filters was represented
by a unit cell that takes into account the convergent-divergent nature of the flow field. The steady,
two-dimensional. and fully developed flow field within the unit cell was obtained from the numerical
solution of the continuity and Navier-Stokes equations. The assessment of the proper assumptions
for the model was carried out by comparing the computed velocity field with that experimentally
determined for a physical model of the unit cell with scale 10:l and containing an aqueous solution
of CaC12. The measurements were done using the particle image velocimetry (PIV) technique. The
efficiency and the coefficient of initial filtration for foam filters were obtained from the determination
of the particle limiting trajectory, resulting from a force balance on a spherical inclusion. This balance
included the buoyancy and the viscous drag forces. The last force took into consideration the wall
effect on the particle motion. The values of the computed initial filtration coefficient show an excellent
agreement with the corresponding measured ones reported for laboratory and plant tests for short-
term filtration. This comparison involves several combinations of particle sizes and downward fluid
superficial velocities. This model is further extended to study long-term filtration in the second part
of the article.

I. INTRODUCTION s the amount of articles accumulatine

because it d e ~ e n d on
within the filier. The filtration periodLwhereA remains con:
THE filtration of aluminum has become widely spread stant is known as initial or short-term filtration, while that
as a final refining technique used before casting the metal.
period where the accumulation of particles starts to play an
The removal of solid inclusions improves metal fluidity,
effect on the value of the filtration coefficient is known as
resulting in greater castability.['l Furthermore, the obtained
aging or long-term filtration. The present artid: is focused
microstructure leads to enhanced mechanical properties, e.g.,
to show the results of a study on the initial filtration penod
strength and ductility are increased providing better form-
characterized by an initial filtration coefficient, Ao. For this
ability and machinability, and also, an inclusion-free metal
case, the corresponding filtration efficiency is given by the
favors a low tool wearing. Melts that have a maximum
following equation:
-
inclusion concentration of 10 ppm in volume are generally
filtered using ceramic foam filters. When these melts flow q = [l - exp (-Ao L)] X 100 [21
through a foam filter at typical melt superficial velocities
from 5 to 15 d s , depth filtration occurs.[21This mode of The previous equation results from substitution of the inte-
particle removal consists in trapping the particles in the grated Eq. [l] into the definition of filtration efficiency given
interior of the filter itself rather than on the inlet surface, as as q = (Ci - Co)lCiX 100, and it establishes the dependence
it happens with the other two modes of filtration, i.e., sieving of the filtration efficiency with the filter thickness L. How-
and cake filtration, which are outside the scope of the present ever, it should be noticed that the filtration coefficient does
work. The particle concentration, C, observed in the solidi- not depend on this thickness but on the pore size and geome-
fied metal that remains within spent filters after depth filtra- try and on the flow and particle properties as well. Equation
tion, follows an exponential profile.[3.41This profile can be [2] has been used by Mutharasan et ~ l . [ to ~ ldetermine A.
derived by integrating the following first-order rate for ceramic filters used for filtration of TiB2from aluminum
expression. in laboratory tests. They found that this coefficient decreases
sharply with the melt superficial velocity in the range from
L to - 3 mmls, but it becomes nearly constant at higher
velocity values. They offered a qualitative explanation for
for the limits C = Ci at z = 0 and C = Co at z = L, where this result saying that at high flow velocities, the inertia of
A is a filtration coefficient, with dimension of reciprocal the particles makes it easier for them to collide with the
length. In general, this coefficient changes during filtration collector wall. In a later work, Engh et a1.16] pointed out that
at high flow rates, the filtration coefficient increases with
melt velocity. Their explanation was not based on the particle
inertia, but rather on the impingement of particles against
F.A. ACOSTA G., Assistant Professor, and A.H. CASTILLEJOS E.,
Professor, are with the Centro de Investigaci6n y de Estudios Avanzados roughness elements of the collector surface. In the late 1970s,
del IPN-Unidad Saltillo, Apdo Postal 663, Saltillo, Caoh., 25000 Mtxico. introduced['] the effect of surface roughness on particle depo-
Manuscript submitted September 3, 1999. sition in order to explain the small differences in filtration

METALLURGICAL AND MATERIALS TRANSACTIONS B VOLUME 31B. JUNE 2000-491

4 1
 efficiencies measured for particle sizes between 3 and 35 p n to neighboring ones through several windows; in contrast,
as opposed to that expected from the classical dependence of the proposed unit cell is formed by pores having only two
7 with d;. The interpretation of results['] of water suspen- windows, at the inlet and at the outlet, but no lateral ones.
sions with a sand bed at flow velocities between 1.3 and The reason for this simplification is based on identifying a
8.3 mm/s introduced the following: (1) an effective size of preferential flow direction through the domain under study
a spherical collector (equal to the sum of its radius plus a as it was discussed in previous w o r k ~ . [ ' ~In
, ~order
~ I to vali-
small value, RK,related to the characteristic height of the date the velocity field calculations within this system
surface roughness), and (2) an attachment probability, i.e., domain, velocity measurements were carried out as it is
the magnitude of the adhesion force between the particle described in the following section.
and the collector had a statistical distribution. The first con-
sideration tends to increase the filtration efficiency because
the collector radius becomes larger while the second favors A. Experimental Work
its decrement because particles detach once the hydrody- The experimental work was aimed to measure the velocity
namic drag force surpasses the local adhesion force. It was field inside a physical model of a unit cell. The particle
also found1'] that factor (1) played a more important role in image velocimetry (PIV) t e c h n i q ~ e I ' ~was
. ' ~ ~used for this
determining the value of the filtration coefficient of small purpose because it has the capability to acquire instantaneous
particles, while (2) was more important for large ones. This full-field velocity data in a single measurement without dis-
analysis resulted in computed filtration coefficients that turbing the flow field.
agrek closely with the corresponding measured values. It is
1. Experimental setup
convenient to point out that small differences in filtration
The physical model consisted of a set of seven hollow
efficiencies for several particle sizes have also been reported
glass spheres mutually interconnected forming a simple
for aluminum filtration in ceramic foam and packed bed cubic cell structure and immersed in a basin cup, as shown
filters.[" With respect to particle deattachment, we believe in Figure 2. The cup was further filled with individual glass
that it is still an open issue because there is no agreement
spheres, containing the model liquid, creating a flow resis-
about its importance in determining filtration efficiency. For tance to force the fluid to flow through the set of intercon-
example, several auth0rs[~.'~1 have claimed that particle deat- nected spheres. In the experimental setup, as shown in Figure
tachment i s present during aluminum filtration: however,
2, a CaCi2 soiution flowed down toward a basin cup and
they have obtained a good agreement between their calcula- into the glass model. The fluid properties and Reynolds
tions and measurements by assuming permanent particle numbers employed in these tests are shown in Table I. The
attachment. Furthermore, F r i ~ v o l dreported
l~~ that the calcu-
employed Reynolds number corresponds to typical condi-
lated surface tension force pressing an inclusion against a tions used for aluminum filtration with ceramic filters. The
filter wall is much larger than the drag forces acting on the entrained air bubbles served as tracers to measure the flow
inclusion. and that, therefore, perfect attachment occurs. He field in the midplane of the uppermost sphere of the set. It
explained the observed particle release, in industrial practice,
is important to mention that the larger bubbles floated up
by pointing out that those particles trapped but not touching while only small bubbles, -40 pm in diameter. were dragged
the wall may be re-entrained by sudden flow and pressure down into the basin cup. The measured rising terminal veloc-
variations. ity in static liquid, of these small bubbles, was -0.01 times
Depth filtration occurs for particle sizes that are several the average fluid velocity, making them ideal tracers at every
orders of magnitude smaller than the pore size, therefore, it
position except at the neighborhood of the wall, where the
is thought that particles should be transported by the flow fluid velocity is low. The motion of the tracers was observed
field and the gravity force toward the filter wall where they
using a laser sheet, 5 cm in height by 3 rnm in thickness,
may attach. This idea is reinforced by the fact that, under
emitted toward the set of glass spheres, as shown in Figure
the same interstitial melt velocity, higher filtration efficienc- 2. The image distortion caused by the refraction of light
ies and a more consistent performance are obtained when
through the sphere surface was minimized by the saturated
using filters with a small pore size, i.e., those in which
CaC12solution itself, which at (25" C) has a refraction index
['9*]

the distance from the bulk to the wall is short. The present
of 1.44, which is very close to that of glass (= 1.5) as
work reports a study of the initial filtration coefficient using
compared to that for water (= 1.33).
a particle trajectory model in which the basic ideas of Paya-
takes et al.["]are further developed. In our model, the com- 2. Particle image velocirnetry technique
plex structure of the porous medium of a ceramic foam filter The experimental arrangement for measuring the flow
is represented by a unit cell that resembles the convergent- velocity is schematically shown in Figure 2. It consisted
divergent characteristic of the flow field. The validity of basically of a charge coupled device (CCD) camera synchro-
this assumption was previously examined using permeability ni?ed to an intermittent laser head and a computer forming
measurements with water flowing through ceramic foam a PIV equipment. This technique allowed determination of
filter^.['^.'^] The obtained results agreed very well with per- velocity vectors by obtaining a picture formed by double
meability measurements. exposure images of tracer particles. This picture displayed
the tracers at sequential positions over the plane formed by
the thin sheet of laser light. A control u&t provided thk
11. FORMULATION OF THE PROBLEM necessary electronics to control the laser intensitv. An elec-
A photograph of a ceramic foam filter and the unit cell tronic device was built to synchronize the cameia with the
that represents its structure are shown in Figure 1. From the laser and generate laser pulse separations from 10 to 120
figure, it can be seen that the filter pores are interconnected ms, which were required to detect the displacement of the

492-VOLUME 31B, JUNE 2000 METALLURGICAL W MATERIALS TRANSACIlONS B

 Direction of flow

4 4 -
Fig. I -Photograph of a ceramic foam filter and a schematic representation of the corresponding unit cell.

1 i . Basin cup
2. Glass spheres
arrangement
Laser head
Laser sheet
CCD camera
Laser controller
Camera controller
Oscilloscope
Interface
. Computer

Fig. 2-Diagram of the experimental setup to measure the velocity field in the physical model of a unit cell.

Table I. Fluid Properties and Reynolds Numbers Used in of 21 V X 28 H interrogation regions to apply the autocorre-
the Physical Model of a 30 ppi Foam Filter lation method. Therefore, the analysis of each PIV frame
Density Viscosity Reynolds of the flow occuning through the sphere produced a map
Property (kg/m3) (Pa-s) Numbers containing 588 velocity vectors. Most of the velocity vectors
in each frame were the result of successful measurements,
Value 1450 0.014 13 and 67
indicating the flow structure within that region of the sphere.
However, a percentage of the measurements were erroneous
$ue to excessively large particle displacement variations,
particles moving at the low velocities found in the system i.e., large velocity gradients, andlor due to out of plane
under investigation, 0.1 to 10 mm/s. Every image was particle motion. Thus, data validation and interpolation were
acquired in digital form, using a frame processor board with canied out before the velocity information was used to inter-
a resolution of 768 V X 1024 H pixels, and stored in hard pret the physics of the fluid flow. Data validation allowed
disk. The measurements were performed over subregions removal of spurious vectors by comparing each vector with
of the domain to minimize errors associated to measuring other vectors close to it. A vector was accepted if its value,
velocities over regions with large velocity gradients. magnitude and direction, fell within some bounds, which
The image analysis was carried out using the VISIFLOW were defined by visual inspection of the vectors in the respec-
software,[16]which divided every obtained frame into a grid tive frame. The discarded vectors were replaced by new

METALLURGICAL AND MATERIALS TRANSACTIONS B VOLUME 31B. JUNE 2000193

 Developed velocity profile Table 11. Fluid, Particle, and Filter Properties Used in
at the inlet Calculations of the Filtration Coefficient

with a coordinate line in the coordinate space.[''] Addition-

ally, the plane r-z was rotated around its symmetry axis by
an azimuthal angle 4 = 0.1 radians to generate a slice of
xis of the pores. This calculation domain was preferred over that
'rnrnetry obtained with a rotation of 2 7 ~radians, because it allows
a proper representation of the trapped particles in a two-
dimensional domain. The cell axis was assumed in the verti-
cal direction, simplifying the particle force balance equation
to a two-dimensional case, in agreement with the velocity
field.
2. Assumptions, governing equations, and boundary
conditions
The initia! f!traticn ccefficients calculated in this work
correspond to TiB2 particles contained in molten aluminum,
which is filtered with a ceramic foam filter; the properties
of the system are given in Table 11. Calculations were carried
out under the following assumptions.
(a) The fluid flow is isothermal, incompressible, and with
constant viscositv. The filtration of molten metals
Fig. 3-Discretized domain employed to compute filtration efficiency of involves an initial period in which the rsrrir.:;dure of
a ceramic foam filter. the flowing metal changes with time. However, t h ~ s
period is shorter than the whole filtration period, even
for short-term filtration that may last several minutes.
Therefore, filtration conditions can be considered to
vectors estimated by interpolation from the values of neigh- involve constant temperature, density, and viscosity. Fur-
boring vectors. For obtaining an average velocity field, ten thermore, the aluminum suspension is very dilute and a
pictures were taken for every frame. and the final velocity Newtonian viscosity can be properly assumed for the
fields were averaged. The observed velocity field variations melt.
between pictures were attributed to fluctuations in the con- (b) The fluid flow is laminar. Experimental values of the
centration and distribution of the tracer particles suspended macroscopic pressure gadient and the corresponding
in the incoming flow. flow rate have allowed to identify Reynolds numbers
(NRep= U i d c / ~ vup
) to -80 for laminar flow through
B . Mathematical Formulation ceramic foam filter~.['~,'~]
(c) The fluid flow is axisymmetrical. This assumption
1. Computational domain means that the momentum transfer is bidimensional. As
The calculation domain consists of a unit cell formed by it was mentioned before. there is a preferential flow
two pores with diameter, d,, window diameter, d,,,, and direction through the unit cell, which bccurs along the
height, 2h, as shown in Figure 1. The boundaries of the
domain were chosen to reduce the possibility of having an
. axial direction; i.e., there is no flow through lateral
windows.
inflow over part of the outlet boundary, as it could be the (d) The particles are spherical, with uniform diameter, d,
case when there is a recirculating flow in the pore. Further- and do not interact among themselves. Equiaxed inclu-
more, the limiting trajectory was conveniently determined sions, commonly found in aluminum, are represented
in this domain because it would be clear, from the computed by a sphere. Any deviation of the actual shape from the
trajectories, whether particles would be trapped or not in spherical one may be considered by introducing a shape
the first pore. The boundaries of the unit cell were described factor. The present work considers aluminum suspen-
with an axisymmetrical curvilinear coordinate system r-Z, sions with a maximum particle concentration of 10 ppm,
as indicated in Figure 3. The coordinates were constructed so therefore, it is assumed that particles do not interact
that the computational boundary in physical space coincided among themselves while moving in the suspension, and

494-VOLUME 318, JUNE 2000 METALLURGICAL AND MATERIALS TRANSACTIONS B

 that the rate of momentum transfer from the particles Table III. Expressions for the Coefficients a of the Top (T)
to the fluid is negligible. and Bottom (B) Neighboring Control Positions,
(e) The motion of the particles, relative to the fluid, is According to the Hybrid Scheme[2o1
laminar. The estimated free terminal velocity for the Coefficient P C -2 -21p12 P >2
largest studied particles with a diameter dp = 50 p m
and a specific gravity equal to 4.5 is u, = 2.6 rnmls in
static molten aluminum. Thus, the corresponding parti-
cle Reynolds number value (NRep= dpuJv) is smaller *P = Peclet number, D = diffusion conductance.
than one and, therefore. within the limit for creeping A = Face area of control volume, t = top face. and b =
flow around a sphere. bottom face.
(f) The particles enter the domain at a velocity equal to the
sum of its free terminal velocity plus the local fluid
velocity. This assumption is based on the short time
and distance required for a particle to reach its terminal a flow field and including the viscous drag and buoyancy
velocity; e.g., for the particle data specified in clause forces can be written as
(e), the distance and time required to reach the terminal
velocity from rest are 22 p m and 9 ms, respectively.
These values were calculated assuming an added mass
equal to 0.5 times the mass of fluid having the parti- where A is the wall-effect correction coefficient for a sphere
cle volume. falling between two parallel plates, obtained by superimpos-
(g) The added mass and the history term are neglected. ing two solutions for a single plate,[lgl and given by the
Because the particles are moving at quasi-steady state following equation:
with respect to the fluid, any change in the relative
particle-fluid velocity is rapidly transmitted to the inclu-
sion, which reaches its relative terminal velocity in a
very short time. Henceforth, the forces related to the
relative particle-fluid acceleraiion can be neglected. n.e coefficient b ir! Fq. [ 5 ] is a buoyancy factor (= 1 - plp,).
(h) Surface forces at the pore wall are neglected. It is
assumed that these short-range adhesive-surface forces
provide negligible contribution to particle transport. 111. SOLUTION PROCEDURE
These forces only determine whether the conditions for The calculation of the velocity fields for both, the physical
attachment are favorable. In this work, it is assumed model and the unit cell of a ceramic filter, were camed out
that the particles touching the pore wall adhere perma- using a BFC mesh of 20 X 160 control volumes, as depicted
nently to it. in Figure 3. The size of the control-volume elements in this
mesh was chosen to be larger than the size of tse suspended
3. Governing Equations and Boundary Conditions particles but small enough to obtain a detailed representation
The fluid velocity field was obtained from the numerical of the flow field. This mesh is also convenient for the veloc-
solution of the continuity ity calculations in the physical model because there is a
large number of velocity vectors to compare with the mea-
sured flow field. Furthermore, it was shown in a previous
and Navier-Stokes equations work[13]that mesh-independent values for the computed spe-
cific permeability are already obtained from coarser meshes.
The governing equations were solved using the control-
volume form~lation[~~1 implemented in the PHOENICS*
subjected to the boundary conditions indicated in Figure 3,
which are as follows: (1) zero flux conditions at the axis *PHOENICS is a trademark of CHAM Ltd., London.
were imposed as a result of axisymmetry; (2) nonslip and
impermeable conditions were imposed on the pore wall (i.e., The set of algebraic simultaneous linear equations
flow velocities tangential and normal to the wall were zero); resulting from discretization is expressed as
(3) a zero reference pressure was specified at the outlet;
and (4) a velocity profile was specified at the inlet surface
boundary. When modeling the physical model system, the
measured velocity profile was prescribed. On the other hand, where E , is the value of the so-called residual; gp, is the
when modeling the ceramic foam filter, a fully developed value of the p variable at control position P; g are the
velocity profile at the inlet, u = f(r) at z = 0, was specified. values of the p variable at neighboring control positions
This means that the velocity field repeats itself pore after Top, Bottom, North, and South; and a are coefficients that
pore. This assumption was verified by calculating the veloc- are defined according to the hybrid scheme and are given
ity field in several interconnected pores. The inlet boundary in Table 111. An analogous set of equations can be written
condition for the first pore was a uniform velocity profile, for the North and South neighboring positions along the y
while the computed velocity profiles at the inlet of the second direction. The set of equations in Eq. [7] was iteratively
and third pores were parabolic and mutually similar, as it solved until reaching a convergence criterion given as
will be shown in a following section.
The motion equation for a spherical body submerged in

METALLURGICAL AND MATERIALS TRANSACTIONS B VOLUME 31B,JUNE 2000495

 The-Lagrangian equation of particle motion was integrated
anaigiically and incorporated into the PHOENICS code as

- 97 [1 - erp (+)I
where 7 is a characteristic time given by the following
equation:

The particle position at time At was obtained numerically

from the velocity as

It should be noted that the Lagrangian time-step, At, was

chosen such that the particle was not allowed to jump, in
the current time-step, from a given control volume to its
neighbors: when this occurred the particle crossing the con-
trol-volume boundary was placed on the boundary itself by
reducing the time-step. This condition on At allowed a proper
( Directio: of flow j
calcul~ticnof the particle velocity because the fluid velocity
used in Eq. 191 was the proper one for the actual particle
position.
The coefficient of initial filtration was obtained from the
initial efficiency of a pore, 170, wh?ch is given as the ratio
of the trapped particles ?lass rate, Wr, to the total incoming
particles mass rate, W;, according to the following Fig. 4-Schematic representation of the limiting trajectory in the unit cell.
expression:

u, plus the terminal velocity, b.This condition is important

because, in a downward flow, light particles may float out
by buoyancy and, thus, not entering into the unit cell, spe-
The so-called limiting trajectory was used to determine cially in those locations closer to the wall, where the fluid
the trapped particles mass rate. This trajectory is the locus velocity is low. For this situation, the traditional evaluation
formed by the moticn of a particle that enters the unit cell of up, as the local fluid velocity would result in an overesti-
at position, r f , such that it is just trapped at the end of the mation of the mass particles inlet rate.
pore, as illustrated in Figure 4. Therefore, those particles The limiting trajectory was determined by a trial-and-
that entered the unit cell at locations closer to the wall, i e . , error procedure consisting of choosing an arbitrary radial
at positions, r, larger than r f , were trapped. Assuming a position for a particle to enter the domain and computing
homogeneous suspension with particle concentration cp, the its trajectory. By inspecting this trajectory, it was determined
trapped particles mass rate would be determined by integ- whether the particle just stroke the outlet boundary of the
rating the corresponding rate equation at the inlet of the unit pore. In the case that the particle did not strike, a new inlet
cell, from r = rT to r = r: = d,/2 - dp/2 as follows: position was chosen, and the procedure was repeated until
the criterion of the limiting trajectory was met. Once 770 was
determined, the initial filtration coefficient was calculated
for the length of a pore, h, as
where dA is the differential surface area vector at the inlet, &, = - 11h ln (1 - q,/100) [I41
which points outwards from the inlet surface of the unit cell.
The total particles mass rate that entered the unit cell was Notice that &, increases with an increment in 170. As a
obtained by an analogous expression but integrating from r final note, it should be stressed that the PHOENICS code
= 0 to r = r,*.It is useful to stress that most of the reports requires that the center of a particle reaches the wall to be
consider that the particle velocity at the inlet of the unit cell, trapped. For our case, a criterion for trapping of the particles
up, is equal to the local velocity of the fluid,['] however, as was established in a user accessible subroutine, which indi-
pointed out by Burganos et aL,[221 a more realistic value is cated that a particle became attached to the wall when the
obtained by considering that the particle velocity, at the inlet distance from its center to the wall was less than or equal
of the unit cell, is equal to the sum of the local fluid velocity, to its radius, R.

496-VOLUME 3 18,JUNE 2000 METALLURGICAL AND MATERIALS TRANSACTIONS B

Fig. 5-Measured velocity field in the bottom region of the upper glass
sphere, of the expenmental arrangement, when a saturated CaC12solution,
carrying small air bubbles. flows downward for NRe = 13.

0 Vr measured 0 3000 6000 9000 12000

18 Radial position r, (pm)
16

-2 I I I I I I
0 1000 2000 3000 4000 5000 6000 7000
Radial position r, (pn)
Fig. 6-Measured velocity profile at the inlet of the physical model when
a saturated CaCI2 solution flows for NRe = 67.
-4 1.
0 2000 4000 6000 8000 10000
IV. RESULTS AND DISCUSSION Radial position r, (pm)
A. Velocity Fields in the Physical Model (b)
As mentioned previously, measurements of the velocity Fig. 7-Graphical comparison between the measured and the computed
field in a physical model were carried out to characterize velocity profiles at ( a ) the equator of the sphere, z = 1.4 cm, and (6)the
the velocity field for flow conditions relevant to aluminum outlet of the sphere, z = 2.4cm, when CaC12(,, flows for NR, = 67.
filtration in ceramic foam filters. Figure 5 shows a typical
photograph of a double exposure image corresponding to
air bubbles suspended in a saturated CaC12solution flowing condition to compute the flow field inside the glass sphere.
through the physical model. The experimentally estimated Figures 7(a) and (b) show comparisons of the measured and
velocity vectors appear superimposed to the photograph. In tne computed velocity component profiles at axial positions
the figure, the white-spot pairs show air bubbles at two equal to z = 1.5 and 2.6 cm, respectively, which correspond
positions. These pair images were processed to obtain the to the equator and to the outlet of the sphere, respectively.
velocity field indicated by the arrows. The shown flow field The good agreement between these results indicates that the
corresponds to the lower section of the glass sphere for a model is able to represent accurately the flow field in a unit
NRe = 13. At the exit of the sphere, one can notice the cell. Figure 8 shows measured velocity profiles at z = 1.5
parabolic velocity profile. cm for two studied flow conditions. It is shown that an
The measured profiles of the axial and radial velocity increment in the flow rate led to an increase in the radial
components at the inlet of the domain are plotted in Figure velocity component. It should be mentioned that a recirculat-
6 for NRe= 67. These values were used as the inlet boundary ing flow was set, in the physical model, once the NRe was

METALLURGICAL AND MATERIALS TRANSACTIONS B VOLUME 31B, JUNE 2000497

 +Vr atN,=13
-Q a-
V, atN,=13
\
\ & Vr atN,=67
- ?I -v-V, at N,, =67 h ------
-------
\
2a
V -
1-

i
b
.- 10-
->
8
P)

a Inlet of 1st pore

4 5 -
Inlet of 2nd pore
Inlet of 3rd pore

4000 6000 8000 10000 12000 14000 0 100 200 300 400 500 600
Radial position, r (pm) Radial position r, (prn)
Fig. 8-Measured velocity profiles at the equator of the sphere, 2 = 1.5 Fig. 9-Computed axial velocity profiles at the inlet of the first, second,
cm. for flow rates of a saturated CaCl?solution flowing at Reynolds number and third pores for the flow of molten aluminum through a unit cell of a
values of 13 and 67. 30 ppi foam filter for NRR, = 48.

equal to 70. This recirculating flow plays an important role

in pushing the inclusion particles toward the wall of a filter,
especially for small particles, as will be discussed in Sec-
tion B.

B . Results from the Mathematical Model

1 . Fullv developed flow
The chosen boundary condition for modeling particle tra-
jectory and filtration efficiency of ceramic foam filters was
the fully developed velocity profile as it is justified in Figure
9, which shows the computed axial velocity at the inlet of
each of three interconnected pores. The calculations corre-
spond to a typical flow condition of molten aluminum and
Reynolds number, NRer equal to 48 in a 30 ppi ceramic foam
filter, The boundary condition at the inlet of the first cell
was a uniform velocity profile, and the computed velocity
profiles at the inlet of the second and third pores were
parabolic and mutually similar, as it can be seen in the figure. 0.00 0.05 0.10 0.15 0.20 0.25
-
According to the previous result, the fully developed flow Gravity number N,
is established in most of the space of an actual filter. The
unit cell model is intended to represent the flow field that Fig. 10-Computed initial filtration coefficient in a 30 ppi foam filter as
a function of the Gravity number for three unit cell inclinations, 0, 2, and
predominates in an actual filter, i.e., that was obtained with 4 deg.
a fully developed velocity profile as inlet condition.
2. The effect of cell inclination on the Gravity number
The fluid flow calculations were based on the hypothesis irklination angles, 0, 2, and 4 deg, for molten aluminum
that the flow field is two-dimensional. This condition is filtration at a flow rate corresponding to NRe = 32 and using
valid even in those unit cells that have an inclination angle, a 30 ppi foam filter. This figure indicates that in a vertical
8, larger than zero as it was discussed by the authors in a unit cell, the initial filtration coefficient has a small value,
previous article.[121However, two-dimensional particle tra- which is practically independent of the Gravity number;
jectory calculations are strictly valid for flow through verti- however, for inclination angles larger than zero, there is a
cal cells because the radial position of the limiting trajectory, strong dependence of the initial filtration coefficient with
rl*, is symmetrical only for unit cells with t9 = 0. Figure 10 the gravity number. In a vertical unit cell, the radial compo-
shows the computed initial filtration coefficient as a function nent of the gravity force on the particle is zero; therefore,
of the Gravity number, NG ( = d ; (pp - p)gll SpU,),for three only those particles entering at low-gravity number values

498-VOLUME 31B. JUNE 2000 METALLURGICAL AND MATERIALS TRANSACTIONS B

 Particle diameter

0 2 4 6 8 10 12 14 1 6 -L
Supemcial velocity, U, (rnmls) w

Fig. 1 l-Computed and reported initial filtration coefficient and filtration

efficiencies for 30 ppi foam filters as a function of the melt velocity for
several panicle sizes.

are pushed by the flow field toward the collecting wall.

Those particles exceeding a critical gravity number value
tend to follow vertical straight trajectories avoiding collision
with the wall. The situation in a tilted unit cell is rather Fig. 12-Computed trajectories of TiBz particles with (a) 15-pm and ( 6 )
different because the gravity force has a nonzero component 5 0 - p n diameter, suspended in molten aluminum that flows at superficial
in the radial direction. This component increases with the velocities rllf 3 = 16) and ! 5 mm/s (NR, = 80) through a unit cell of
Gravity number allowing a strong dependence of the filtra- a 30 ppi filter.
tion coefficient to be obtained with this dimensionless num-
ber. An inclination angle, 0, equal to 4 deg was chosen for
the trajectory calculation to obtain a dependence of the initial between 15 and 20 and 30 and 35 pm, respectively. Finally,
filtration coefficient proportional to INc\'.', as it has been the plant measurements reported by Tian and Guthrierglfor
found for depth filtrati~n.[~.~]These results mean that actual the filtration of 40 to 60 p m alumina particles are well
particle trajectories are closely simulated when the fluid represented by the computations carried out for a particle
flows in a quasi-vertical cell assuming an axisymmetrical diameter of 50 pm.
particle trajectory. The predicted dependence of the filtration coefficient with
3. Results of the computed initial filtration coeficient particle size and fluid velocity agrees with the corresponding
Figure 11 shows the computed initial filtration coefficient, measured behavior and explains the relatively small differ-
using 0 = 4 deg, as a function of the melt superficial velocity ence in filtration efficiency observed for particles with differ-
for the filtration of TiB2 particles in aluminum using 30 ppi ent size. For example, a closer inspection to Keegan et
foam filters. The curves shown in the figure belong to parti- al.[8] measurements shows that the ratio of the filtration
cles of different diameter, and the spot values indicate coefficients for particles within the size range of 55 to 60
reported measurements. The figure also shows the corres- p m to that for particles within the range of 15 to 20 p n is
ponding filtration efficiencies for filters of 2.5 and 5 cm equal to 2.2. This ratio is almost four times lower than
thickness. From the figure, it is seen that the computed the value expected from the generally accepted quadratic
filtration coefficient decreases with the melt superficial dependence of the filtration coefficient with particle diame-
velocity until reaching a minimum value at a critical melt ter. This low ratio agrees with the behavior shown in Figure
velocity. At higher flow velocities, the filtration efficiency 11, where at high superficial velocities the values of the
increases with an increment in the superficial velocity. The filtration coefficient, for particles with different size, become
value of the critical velocity at which the minimum filtration closer to each other. Therefore, the computed behavior of
coefficient is reached depends on the particle size. For exam- the filtration coefficient is consistent with these observations
ple, the critical velocities for particle sizes of 8 and 50 p m and with those reported in the chemical engineering litera-
are -3 and 13 mm/s, respectively. The computed filtration turef7]where the small differences in the values of the filtra-
coefficients agree with experimental values reported by sev- tion coefficient, for large and small particles, are explained
eral authors; for example, the classical laboratory results by introducing a surface roughness, RK,on the wall collector,
reported by Mutharasan et ~ l . [ for~ ] the depth filtration of as it was described previously.
TiB2particles using filters of 30 ppi include a velocity range 4. Mechanism of particle trapping at different flow
up to nearly 8 mm/s. These experimental results are clearly rates
within the range of the computed filtration coefficients for The mechanism explaining the increment in the filtration
particles between 8 and 15 pm. The calculated coefficients coefficient of small particles with the increase in the flow
for particle sizes of 20 and 30 pm agree with plant filtration rate is depicted in Figure 12(a). This figure shows computed
efficiencies obtained by Keegan et a1.r81 for particle sizes trajectories for a 15 p m particle of TiB2suspended in molten

METALLURGICAL AND MATERIALS TRANSAmIONS B VOLUME 31B.JUNE 2000499

 Fig. 14-Computed trajectories of TiB2particles of 8 and 20 p m in diame-
ter, suspended in molten aluminum that flows at a superficial velocity of
15 mmls (Nu== 80) through a unit cell of a 30 ppi filter.

pore are mutually equivalent, as obtained from the fully

developed boundary condition previously described. In Fig-
Fig. 13-Computed fields for aluminum that flows at Nuc of (a) 16, (b)
32, ( c ) 64, and (d) 80 through a unit cell of a 30 ppi filter.
ure 13(a), there are no recirculatory flows, and the velocity
gradient in the radial direction is relatively snialler than
that obtained in Figure 13(b), which does not exhibit any
aluminum flowing at two superficial velocities, 3 mm/s (NRe recirculatory flow either. Figure 13(c) shows incipient recir-
= 16) and 15 mm/s (NRe= 80); in both cases, the particle culatory flows and a steeper velocity gradient in the radial
trajectories start at a common position at the inlet of the direction than those shown in the previous cases. Figure
domain. It is seen that at 3 d s , the particle follows a 13(d), at the highest Reynolds number condition, presents
trajectory that takes it to the lower pore where it is trapped. also recirculatory flow and the steepest velocity gradient in
In contrast, at the high melt velocity of 15 mm/s, a similar the radial direction. These recirculatory flows pushed the
particle collides with the wall of the upper pore. This result small particles towards the wall, favoring their trapping. In
means that the increment of the melt velocity improved the this regard, Figure 14 compares the computed trajectories
filtration efficiency of the upper pore,* and, therefore, of of two particles with diameters of 8 and 20 pm under an
identical velocity field at NRe = 80. From this figure, it is
*Notice that, in order to trap a particle in the upper pore at 3 W s , it
should enter the domain at some position closer to the wall, i.e., at position
seen that the small particle was pushed upwards by the
r f , larger than for the case of melt velocity equal to 15 W s . From Eqs. recirculatory flow and was trapped at a higher position with
[12] through [14]. it can be seen that this larger value in r,* means a lower respect to the large particle. Because this recirculatory flow
value of the filtration coefficient. Therefore, the modification of a particle was also present in the upper pore, the filtration efficiency,
Irajectory, resulting in the trapping of the particle at lower positions of the obtained from the analysis in this pore, was higher for the
domain, will be interpreted in the article as a decrement in the filtration
coefficient and efficiency. smaller particle than for the large one, as it can be seen in
Figure 11.
the whole filter. The computed trajectories for a 50 pm
particle (Figure 12(b)) indicate that the increment in the
melt superficial velocity from 3 to 15 mm/s decreased the
V. SUMMARY AND CONCLUSIONS
filtration efficiency of the upper pore for this particle size. A two-dimensional mathematical model has been devel-
The flow pattern computed at high flow velocities is the oped to quantitatively represent the fluid flow and the filtra-
reason for the behavior previously described for the particle tion efficiency obtained during depth filtration of molten
trajectories. Figure 13 shows velocity fields computed at aluminum with ceramic foam filters. The model is based on
several Reynolds number values: (a) 16, (b) 32, (c) 64, and the calculation of the particle trajectory obtained from the
(d) 80. Notice that the flow field in the upper and the lower motion equation for a submerged body in a fluid flowing

500-VOLUME 31B,JUNE 2000 METALLURGICAL AND MATWALS TRANSAC'ONS B

 within a unit cell. The filtration efficiency obtained in this diffusion conductance at the bottom and top con-
representative unit cell is used to evaluate the overall effi- trol-volume faces and equal to pb/(zp - zb)
ciency of the filter. and ~ r / ( ~ - bzp), respectively
The velocity field within the chosen unit cell was charac- absolute value of the residual
terized in a physical model. using the PIV technique, for vector function defining velocity profile along
typical flow conditions for aluminum filtration in foam fil- the radial direction ( d s )
ters. The corresponding computed flow field was determined acceleration of gravity (9.8 d s 2 )
using the measured velocity profile at the inlet of the pore distance between successive windows (mm)
as a boundary condition. The good agreement between these filter thickness (cm)
results indicates that the model properly represented the distances from the particle center to each one of
flow field. the plates that bound it (mm)
The assumption of an axisyrnmetric particle trajectory was half-length of separation between the plates that
justfied on the basis that a unit cell inclination angle equal to bound a particle (mm); in our case, it is con-
4 deg was sufficient to obtain the strong dependence of the sidered to be equal to half the pore cord length
filtration coefficient with the gravity number V(Y IN^^'.^. This measured horizontally at particle position z
result means that actual inclusion trajectories describe. as an mass of the suspended particle (kg)
average, quasi-vertical trajectories within the filter. gravity number defined as d i (p, - p)gl18pUl
The computed initial filtration coefficient decreased with Reynolds numbers for the fluid, NRe = U,d,lv,
melt velocity until reaching a minimum value at some critical and for the particle, NRe., = Udjplv
melt velocity. At superficial velocities above this value, the Eulerian pressure in the fluid (Pa)
filtration coefficient increases with the increase in melt reference pressure in the fluid at the outlet of
velocity. The critical velocity value depends on the particle the unit cell (Pa)
size; for example for particle sizes of 8 and 50 p m the Peclet number evaluated at the bottom and top
critical values are around 3 and 13 rnmls, respectively. The control-volume faces and equal to PU:,~ (z, -
increment in the filtration coefficient is a consequence of zb)/,ub and pu,, (z, - z )lpr, respectively
the formation of a recirculatory flow pattern that pushes the voluminic flow rate (cm3"Is)
particles toward the wall. Large particles are less sensitive coordinate distances in the radial and axial direc-
to this recirculatory flow. tions, respectively (mm)
maximum radial position at which a particle can
enter the unit cell (mm)
ACKNOWLEDGMENTS inlet radial position of the particle that follows
the limiting trajectory (rnm)
The authors express gratitude to The National Council particle position and particle initial position vec-
of Science and Technology of MCxico (CONACyT) for tors (mm), respectively.
the financial support to this work and for the scholarship radius of the suspended particle (pm)
(Catedra Patrimonial) granted to FAAG as a graduate stu- radius of the collector grain m d dwrage length
dent at CINVESTAV. Also, we thank Mrs. Noemi Neira of roughness elements (mm),respectively
for locating and supplying the technical information for time and Lagrangian time-step(s)
this work. Eulerian velocity vector of the fluid ( c d s )
interstitial melt velocity (Urn/&)(cmls)
melt superficial velocity (QlA) (cmls)
LIST OF SYMBOLS Lagrangian particle velocity and particle termi-
nal velocity ( c d s ) , respectively
a; coefficients corresponding to the north, south, velocity of the particle in a previous time ( c d s )
top, and bottom neighbors in the control vol- radial and axial components of the fluid velocity
ume discretization method, Eq. [7]. ( c d s ) , respectively
as a~9
coefficients corresponding to the bottom (or radial component of the fluid velocity at the
downstream) and top (upstream) neighbors bottom and top control volume faces,
(Table 111), respectively respectively
A fluid flow area or downstream control volume mass flow rates of trapped particles and particles
face area (m') entering the pore (g/s), respectively
Ab?Ar areas of the bottom (or downstream) and top (or size of the control volume in the x and y direc-
upstream) control volume faces (m2), respec- tions, respectively (m)
tively, and equal to AxAy axial positions of the bottom face, the control
b buoyancy factor (1 - pip,) point, and the top face of the control volume,
C Eulerian particle concentration in the melt that respectively (m)
flows through a filter (g/cm3) main flow direction of the fluid through the filter
Ci Co particle concentration in the melt at the inlet and
at the outlet of the filter, respectively Greek letters
c~ particle concentration in the melt before filtra- E filter porosity (pct)
tion (ppm and g/cm3) 4 azimuthal angle, radians
d,,d,, d, particle, window, and pore diameters (mm), cPi, &, values of the dependent 4 variable at neighbor
respectively position i and control position P,respectively.

METALLURGICAL AND MATERIALS TRANSACTIONS B VOLUME 31B,JUNE 2000-501

 filtration efficiency of the filter defined by (Ci 5. R. Mutharasan, D. Apelian, and C. Romanowski: JOM, 1981, Dec.,
pp. 12-17.
- C,)/Ci x loo (pct) 6. T.A. Engh, B. Rasch, and E. Bathen: Proc. 115rh TMS Lighr Merals
filtration efficiency of a pore determined from Annual Meering. New Orleans, LA, Mar. 2-6.1986, TMS, Warrendale.
Eq. [I21 (pet) PA, 1986. pp. 829-36.
filtration coefficient and initial filtration coeffi- 7. Handbookof Flu& in Motion, N.P. Cherernisinoffand Ramesh Gupta.
cient (m- ') eds., Ann Arbor Science, Ann Arbor, MI, 1983, pp. 969-1000.
8. N.J. Keegan, W. Schneider, H.P. Krug, and V. Dopp: Proc. 126rh TMS .
wall-effect correction coefficient for a sphere Lighr Metals Annual Meering, Orlando, FL,Feb. 9-13, 1997, TMS,
falling between two parallel plates and Warrendale, PA. pp. 973-82.
defined in Eq. 6 9. C. Tian and R.I.L. Guthrie: Merall. Marer: Trans. B, 1995, vol. 26B.
dynamic and kinematic melt viscosities, (Pa-s pp. 537-46.
and m2/s), respectively. 10. J.P. Desmoulins, H. d'Hondt, J.M. Hicter, and P. Nener: Proc. 118rh
TMS Lighr Merals Annual Meering. Las Vegas, NV, Feb. 27-Mar. 3.
melt viscosity at the bottom and top faces of the 1989, TMS, Warrendale, PA, 1989, pp. 757-67.
control volume (Pa-s), respectively 11. A.C. Payatakes, C. lien, and R.M. Turian: AIChE J. 1974, vol. 2 q s l
constant, 3.141592654 ' pp. 889-99.
angle between the unit cell axis and the main 12. F.A. Acosta G., A.H. Castillejos E., J.M. Alrnanza R., and A. Flores
V.: Merall. Marer: Trans. B, 1995, vol. 26B, pp. 159-172.
flow direction, (rad and deg) 13. F.A. Acosta G.: Ph.D. Thesis CINVESTAV, Coah., MCxico. 1999, pp.
density of the fluid and particle, respectively 69-74.
(kg/m3) 14. Y.A. Hassan. T.K. Blanchat, C.H. Seeley Jr., and R.E. Canaan: lnr. J.
characteristic time given by Eq. [lo] (s) Mulriphase Flow, 1992, vol. 18(3), pp. 371-95.
15. C.C. Landreth and R.J. Adrian: Proc. 4rh lnr. Sym.9. of Applications of
Laser Anemomerry to Fluid Mechanics, Lisbon, Portugal, July 11-14.
REFERENCES 1988, Springer-Verlag. Lisbon, Portugal, pp. 484-97.
16. VISIFLOWa software. Rev. 7, AEA Technology, Oxfordshire,
1. Merals Handbook, vol. 15, Casting. ASM INTERNATIONAL, Metals U.K., 1996.
Park, OH, 1988, pp. 489-93. 17. C.A.J. Fletcher: Computational Techniques for Fluid Dynamics.
2. L.S. Aubrey and J.E. Dore: Proc. I22nd TMS Lighr Metals Annual Springer-Verlag. Berlin. 1977, vol. 2, pp. 47-48.
Meeting, Denver, CO, Feb. 21-25,1993. TMS. Warrendale, PA, 1993, 18. C. Xan and R.I.L. Guthrie: Proc. 122nd TMS Light Metals Annual
pp. 1009-20. Meeting, Denver, CO, 1993, TMS, Warrendale, PA, 1993, pp. 1003-07.
3. F. Frisvold: Ph.D. Thesis. The Norwegian Institute of Technology, 19. H. Brenner: Chem. Eng. Sci.. 1961, vol. 16(3-4). pp. 242-5 1.
Trondheirn, Norway. 1990, pp. 98- 142. 20. S.V. Patankar: Numerical Hear Transfer and Fluid Flow. McGmw-
4. A.H. Castillejos-E., F.A. Acosta-G., and J.M. Alrnanza-R.: Proc. 121sr Hill. New York, NY. 1980, pp. 88-90.
TMS Lighr Merals Annual Meeting. San Diego. CA. Mar. 1-5. 1992. 21. H.I. Rosten and D.B. Spalding: PHOENICS, Software ver. 2.1.3.
TMS, Warrendale, PA. 1992, pp. 1 113-22. CHAM Ltd., London, 1996.
22. V.N. Burganos, C.A. Paraskeva, and A.C. Payatakes: AIChE J., 1995.
vol. 41(2), pp. 272-85.

METALLURGICAL AND MATWALS TRANSACfI( 'NS