RESEARCH SUMMARY· - - - - ·

Ceramic Foam for

Molten Metal Filtration
_ _ _ __ _ _ ___ L.J. Gauckler , M.M. Wae ber, C. Conti and M. J acob·Duliere

INTRODUCTION
The increasing use of aluminum alloys for sophisticated products like In this stlldy, c.1ramic foam filters
beverage cans or aircrafl components requires extremely low impurity were used for the inclustriol filtration
concentrations in t he liquid metal. The most common solid impurities are of aluminum. Results ore compared
globular oxides in magnesium containing alloys and oxide skins. Inclusion with laboratory experimer~/s which are
dispersoids are a few microns in size, and oxide skins can be several in goocl agreement with trqjectory anal·
milimeters (Figure 1). Solid impurities of a few microns may be detrimen- ysts of deep bed filtration for the early
tal to a high quality surface finish and products with high deformation slfJge of filtraiion. TM correlatioru; be·
rates and small cross sections. Therefore, efficient melt purification meth· tween structural characteristics of the
ods are needed to meet today's high quality standards, especially regarding filter media, filtration parameters and
the increasing usc of secondary aluminum. filter efficiency are giver~. In addition,
There are various methods available to remove impurities from metal the most important parameters for the
melts. They a re based on sedi mentation, notation or filtration . A good industrial use of fillers are discussed.
survey of the fundamentals of molten aluminum filtration can be found in
reference 3.
Sedimentation is performed by long time furnace treatment. Flotation is
done in separate treatment units similar to small folding furnaces of
several tons capacity. The units in general are supplied with a heating
system. An exception is the system MINT with its small volume which can
be emptied afler each cast. Some of tho units filtering solid particles are
used for single drop application, like ce•·amic foam filters.
FILTRATION
The nitration process removes part1cles by the litter medium in three
different ways, depending on size of tho particles and size of the openings
in tho filter medium. Sieving takes place on the surface of the filter
medium forming a filter cake in lhe case of particle diameters. These
diameters are of the same order of size or larger than the holes in the
fi lter medium. Woven cloth filters mainly operate in this mode. Deep bed
filtration occurs within t ho granular or foamed filter medium on its internal ...Figure 1. Sieving and cake lollralion. wavy
surface. Each internal pore surface has a probability of retaining particles lines • skins. spots • particles.
from the melt (Pigure 2). In this ease, particles are smaller than the smallest
opening, the "window size.tl

Table I. Inclusions In Aluminum

Type Form Density (g/cm') Dimen sions (J.lm)
AI,Oa PnrLiclos 3.97 0.2- 30
Skins 10 -5000
0
X MgO Particles 3.58 0.1 - 5
I Skins 10 • 5000
0
E MgAI,O, Partotles 3.6 0.1 - 5
s Skins 10 -5000
Sio, Porllclcs 2.66 0.5. 5

Salts
Chlorides Portitlcs 1.98 -2. 16 0.1 - 6
N Fluorides ...Figure 2. Deep bed filtratiOn (IWO dimensional
0 cuts of a ceramic toam. left omage schemat~) .
N Carbides o• window size. p • cell size.
AI,C. Parucles 2.36 0.5. 25
0 SiC Particles 3.22
X
I Nitrides Portitlcs
I) AIN Skins 3.26 10 60
E
s Boride.s Particle-
Ti8, Clusters 4.5 l 30 Editor't N'oc.-: This paJ!tr 11 adapc.cd from TMS pubh·
AIS. Particles 3.19 0.1. 3 ct~tton l..tAIIl MnoU S5and •u preM"nttd at the 'DIS
Annual Mfttana 1n Ne- York

JOURNAL OF METALS • September 1985 47

 TRANSPORT MECHANISMS
There are several reasons for a particle to touch the internal macroscopio
surface of the filter. The following transport mechanisms were taken into
account by Conti 2 to calculate filter efficiencies for aluminum filtration.
Direct interception. A particle hits the filter surface following its trajectory
line. Grauity forces. Is the microscopic movement caused by the molecular
bombardment on the particles in the liquid. This phenomena is believed to
be important only for submicron particles. Inertial forces. Caused by the
apparent weight of the particle which cannot follow sudden changes of the
trajectory line and hits the internal filter surface. Hydrodynamic e((ecU;.
They are due to the velocity distribution in the filter cell. Depending on
the s hape of the particle, it rotates and translates in the flow field. In the
case of a luminum filtration with ceramic foam, the most important forces
are direct interception, inertial and hydrodynamic forces.
Particles that touch the internal filter surface are attached by Van der
Waale forccG. Po.rtic1cs, or aggregates of particles may be detached from
the filter surface by increasing shear stresses (caused by a sudden increase
of flow rate, by backwashing or pumping mode, as well as by sudden
movement of the filter media itself) .

•
Figure 3. Three-dimensional ceramic foam. STRUCTURE OF OPEN PORE CERAMIC FOAM
Information a bout the structure of the filter medium is critical in under-
standing fluid flow characteristics. Stereological methods 4 were used to
characterize the microstructures of the ceramic foams. Micrographs of a
cross-section and of the two-dimensional ceramographic sample of a three-
dimensional foam are shown in Figure 3.
The structure of the ceramic foam consists of rounded polyhedra5 with
diameter p which are connected to each other by openings (windows) of
diameter $. A roug h characterization is given by the number of pores per
0' ~ ~ ~ W U M ~ ~ U inch (ppi). The total porosity is designated f, the total internal cell surface
Mti AN WINIDOWS.lt'f' (M... S,. The distance between the centers of two cells is s.
•Figure 4. Relation between cell and window
The ceramic foam structure shows anisotropy. In the plane of the filter
plate, the cells are spherical with a mean cell size of p. Perpendicular to
size. the filter plate, the cells are elongated with a somewhat larger mean cell
size of p1 . In ceramic foam filters, the determining dimensions for permea-
..... bility and Ouid flow are the mean cell and window size. Figure 4 shows
the relation between these two parameters.
From 55 to 24 ppi, the values of t he mean cell size perpendicular to the
Oow direction increases from 0.9 to 2.2 mm. The mean window size in-
creases from 0.6 to 1.45 mm. 'fhe total open macroporosity of the ceramic
foam ranges from 75% (30 ppi) to 83% (24 ppi).
The internal surface is in contact with molten metal during filtration.
~.
w
Due to the surface tension of the molten metal, pores smaller than approxi·
ioL---------~--~------~--~ mately 0.1 mm can hardly contribute to internal surface area because
Q.2 OA 0.6 0.8 1.0 1.2 1.4 1.& 1.8 2.0 metal pressures usually applied in filtration do not exceed 500 to 1000 mm
MEAN WINDOW SI Z E fi ('""'} AI metallostatic pressure. In Figure 5, the internal surface as a function of
•Figure 5. Macroscopic internal surface of ce- the mean window size is shown. It decreases from 4.0 to 1.0 m2/m3 for 55
to 24 ppi filters.
ramic foam filters.
TRAJECTORY MODEL
Initial filter efficiencies were calculated by Conti, 2 Jacob and Conti 6
I ' using Payatakes, Tien and Turtan's method of trajectory calculations.'·" This
p~l method is well-established a nd successfully employed for water and aerosol
filtration.
The efficiency of deep bed filtration is:
convergentl cell lJ = 'lc 'P (I)
where lJc = the probability of contact between particles and internal filter
s urface, and p = retention probability of a particle in contact with the
internal surface.
In this study, the probability of retention after contact is taken to 1 (p
= 1). Therefore the filter efficiency is determined by the probability of the
particles to contact t he internal surface of the filter.
divergent j cell zone
According to the periodicity of the unit cells in the Oow direction, the
filtration bed can be divided into a series of identical unit elements with
height Hcen which is equal to the periodicity (Figure 6). It is assumed that
the concentration of particles at the beginning of each cell is uniform
perpendicular to the flow.
•Figure 6. Model ol dimensionless unit cell al· Thus, it is possible to calculate the probability of particle attachment llc by
distinguishing particles on trajectory lines coll isioning with the cell-surface
ter Conti (2). from particles wh ich simply pass through the cel l.
48 JOURNAL OF METALS • September 1985
 The retention probability of a unit cell is given by X0 , the filtration 100.------------------ -- ---,
coefficient:
1 1
X0 = - In ( - - ) (2)
H..n 1-'!c
The filter efficiency of a filter with thickness H is:
Cin-Cout
'! = - 1 _, · H
- - =-eo (3)
Cin
where C; 0 = inlet concentration of particles; and Cout = outlet concentration of
particles.
The calculation of the trl\iectory lines along which inclusions follow in
suspension requires solution of the Navier-Stokes equations, e ither by a n
analytical approach or by the method of calculating finite differences in
the velocity field.
The main portion of attachment occurs in the convergent zone of the
cell. In the divergent zone, the trajectory lines deform at high Reynolds
numbers. Therefore the probability of attachment in this area is lower.
The Reynolds number is defined by:
Yo · H.:cll
Re = (4)
v
where v = viscosity of aluminum (0.0059 cm2/s);
and v0 = velocity of
a luminum in the s mallest cross section of the structure

Vo = Np · 'IT • (cbl2)2 (S)

where v, = apparent macroscopic melt velocity through the porous structure;
and N P = number of unit cells per filter thickness. OL-~~~~llL--~~~~
The filter efficiencies range from 75 to 95% for 20 I-'m particles at low 101 1 10
melt velocities and decrease to the range of 15 to 35% for high melt MElT VELOCITY Vs (cm/s)
velocities and s mall particles as shown in Figure 7.
Decreasing efficiency with increasing melt velocity is due to the in- •
Figure 7. Calculated initial filter efficiencies as
creased turbulence in the convergent zone of the pores, and the diminished a function of melt velocity.
sedimentation possibi lity at higher melt velocities.
Filter efficiency improves with increasing particle size. {n Figure 8,
examples are shown for particles with densities of 4.5 g/cm3 and melt ...,. 100
X
55Pt)i
.o ...

~
- 80
velocity of 1.0 cm/s. More than 80% of all particles greater than 40 J.Lm are 30 ...
removed from the melt. Even for particles of 20 ,.m diameter, efficiencies

~
>
were obtained ranging from 30 to 65% for finer filter (55 ppi).2 ·6 ~eo
w
If the particles are heavier than the liquid, the effect of sedimentation ;;;

9
;;:
becomes more important. In consequence, the particle retention in the ~
w
•o
convergent part of the cell is favored. Efliciency increases with increasing
specific gravity of the particles. This behavior is more pronounced for coarse 20
filters (30 ppi) than for fine filters (55 ppi).
oL-~~~~uL--~~~~~
1 10
LABORATORY EXPERIMENTS
PARTICLE DIAME TER ilf9 (,..,..)
The initial filter efficiencies of different filter types were experimentally
proven using titanium diboride part icles as tracers in aluminum melt. •
Figure 8. Calculated initial filter efficiencies as
The filter is placed in the bottom of a cast iron crucible inside a resistance a function of particle dimension. H = 5 em, p.,
furnace. Aluminum of 99.85% purity with titanium diboride from Kawecki = 4.5 g/cm3 , p f = 2.36 g/cm3 v, = 1.0 cm/s.
Beryl co (master alloy AI-Ti5%/B1 %) were used . The TiB 2 particles range
from 1 to 10 ,.m in diameter. The concentration is detected by spectrographic 60 6 H = 5cm
techniques us ing boron and titanium s ignals.
Three analyses were used in each run to determine the ingoing and '' Qp
Ot =
s 4 . ~ g/cnTl

outgoing tracer concentration. The average values were used to calculate '' dp• Spm
2.36 g/cm3

the filter efliciency 1J for 2 inch filters according to equation 6: ' (calculated)
(TiB2l;n - (TiB2lout I'" 40
1) = (6)
(TiB2);,. >
()
Detailed results of single runs for 40 and 30 ppi filters can be found in z
reference 2. w
In Figure 9, the filter efficiencies arc given as a function of melt 0
velocity. The •·esults from laboratory experiments are represented by the -=~ 20
shaded areas. Dotted lines are the calculated efficiencies fo•· 8 J.Lm TiB 2
particles. The experimental results are in good agreement with the calcu-
lated values. Efficiencies increase for decreasing melt velocities. Efficiencies
are higher for finer filters (55 ppi) with high internal surface and lower for
coarser filters (30 ppi). O·L-~~~LU~--~~LLLUW
10-1 1 10
ALUMINUM CAST HOUSE EXPERIENCE M ELT VELOCITY v 5 (cmls)

Ceramic filters are best used in-line between furnace and cast.ing machine.
Filters are preheated prior to casting. The metal level difference 6h is used
•
Figure 9. EJ<petimenlal and calculared (dolled
lines) efficiencies or ceramic roam fillers for
as the driving force to obtain fluid now through the filter. different melt velocities.

JOURNAL OF METALS • September 1985 49

 Prior t.o filtration, the filter has t.o be impregnated with aluminum by a
metal pressure h 0 higher than that used during filtration. After
' impregnation, the head loss h; is the metal level difference between ingoing
and outgoing metal, which is needed for a constant flow through the
porous medium.
During constant rate filtration, the metal level increases by the differ-
ence llh per time At according t.o the deposition of impurities on, or in the
filter plate.9 This metal level difference increases almost linearly with time
during phasc 11 of filtration. During this phase, impurities mainly deposit
inside the filter, but also on the ftlter forming a filter cake which is
compressed in phase m. Then, the ftlter clogs, and the metal level differ-
0 .,.. · -· ._, lO ,,.. ... u .s u
wtlriOOW Ill« , 4- ) ence tl.h increases drastically with time. At the beginning of phase III, metal
4
filtration should be terminated.
Figure 10. Metal prime head for different fil·
ters versus window size. Metal Prime Head hp
For d ifferent filter types, the metal prime head hp is given in Figure 10.
Metal prime head is mainly determined by the wondow size of the filters.

!!
E
u
2.0
- ---- u..,., limit - - - -7 (. It can be expressed by:
4 • 'YLS
p = -p. g. hp =-$- - cos 0 (7)

.. l5
is where p is the pressure; p is the density of the melt (2.4 glcm 3); h.P is the

/
0 metal prime head in mm; -y LS is the surface energy solid/liquid (AI2UafAI =
..
g
> lO ~
860 dynlcm); and 8 the wetting angle of aluminum on alumina <~ 17SO).
Using these values, the metal prime head is expressed by:
~ I
"':1! / U,.t
R~ r•nt•.)Oppf hp (mm) = 146 (mm2) (
$ mm
) (8)
0.5
. Calculated and experimental values are in good agreement.
w
N 12"
;;; 17 '
2 • e 8 10 Wo rking Metal Head h0
• 8 12 16 20
....
a:
w e
20'
;;: 23' 8
12
18
18
2.
2•
32
:l')

40
I
At the beginning of filtration after impregnation, the working metal
head is determined by the specific now resistnncc of the filter, the filter
size, and the flow rate. Recommended flow rates and filter types a re given
FLOW RATE ( t ontlh) in ~'igure 11. Working metal heads for the recommended range are typical-
4
ly 5 to 50 mm.
Figure 11 . Flow rates tor different ceramic Bes ides the mentioned parameters, the impurity level of the first metal
foam filters. approaching the filter also determines the working metal level. Precleaning
of this first metal by removing large oxide s kins avoids clogging of filter
surface. This res ults in low working metal heads and extended filter
SEL£E 30 pp, 1r A10y6100 CA>M9-&l lifetime.
Tot11 nutnOtr OlfJr<IOet CAl 27 C8)27 Total drop sire (e.g. the critical time when the filter starts to clog) is
TOWl riU'TOef OllrG.tllonl CAl 00 (8) 23
AY'W8Qtt tad' 01 ~ {,.m) CAl 13$ (8) l t t determined by the impurity level of the melt. The increase of working
EFFK::IEHCY n • 71.3 ,_, metal level versus time can be measured and used to characterize both the
quality of the melt and the lifetime of the filter.
-• ,.
~
Filter Efficiency
~ 14
! ,, Filter efficiencies in cast house operation were determined by analysing
~. the type and amount of impurities in chilled metol samples as described in
i J.. reference I. Differences of impurity concentrations, before and after t he

~
f .~ filter usage, for efficiency characterization as s hown in Figure 12.
! • The data in Figure 12 are based on 54 specimens of 100 cm 2 each from
2 •
,n L. ...L. 1'" 1 1
10 different casts using flow rates of 13 tonslh and drop size of 21 tons.
Inclusions smaller than 20 11m were not counted. Total number of inclusions
taken in specimens before the filter were 80, after the filter 23. About 80%
of the inclusions were identified as oxide particles a nd skins. The rest were
4 salts and borides. A total efficiency of 71% was obtained.
Fogure 12. S.ze dostnbutions of inclusions (a)
black ~ne • beiO<e. (b) white 6ne = a«er.
References
1. K . Buxmann , P . Purrer . ..Zum Problem der crauen Z.ealen auf anodis th oxidierten
Alum•n•umprodukten•. Mnt:tll, $4 ( $ ) (1980) ~· 222.-228".
2. C. Cclnt1, "'Contnbuuon a l'erude de Ja filtrat.on profondoe ckla m«aux'", 'l'beiil. Faeulte polyt«:hnJque
ABOUT THE AUTHORS - - - - - · de I'Unl\'ert:ete de M o~ Jklgium, ()983).
3. C.R E!tkert, R E Milltr, D. Apelian and R. MutharAN.n, ..Molten Alum•num Filtration: F\andamen·
L.J. Gauckler and M. M. Waeber are currently tab • nd Model.11'", pp 1281 -1304 ln L ight M e-krif 1984, J .l'. Mc.{'.etr, eel.: Conference Proceedings.. The
working at the Research and Development Mtt.allurgu•:.t~l Society or Al:.tE.
• · P. t:. Underwood, "Quantitath·c SU!reolog,y··, p. 274. Addi.aon-We1ley, ~l a11811.Chuseu... USA. (1970).
Departmem of Swiss Aluminum Limited in 5. J.lt DAwton, J .B. S hortall, ""The Miaostrueture or Rigid l'olyurethane r oom•"· Journol of Malf'ri<Jis
Switzerland. S(!UUI" · 17 (1982) pp. 220·2'l4.
6. M. Jtioob, C. Conti. "Etude des probabilitctl initialea d'ttrrct d'inelutiont dtenl de. tilt.res pour
C. Conti and M. Jacob·Dullere are currently mct11ux", Internal rcpon., Alusu is~~e Research & Oe\'CIOI)mtnt, Neuhl'luAen, Swil:r.e rland: r aeulw
polyt«hnlquc de Mona. BeiKium <1984).
working at Facult8 Polytechnique de Mons in 7. A.C. P.ayatake~~o, C. 1'ien, R.M. Turian, "'A New Model ror OrMuler Poi"'UU Media", A1.Cit.E.
Belgium. Journal , 19 (I ) <1973) _pp 68-76.
8. A C. PayMakea.. C . Tlen. R M. 'f\man. "1'r~tory Cakulation of PJrtic:te Oepoention in O«p Bed
l'' iltration". A I.CII.E. Jou.nt41, 20 (S) 0974) pp. 889-905.
9. K.J. lvu, '"Deep Bed filtration .., pp. 234.-301 In Solld•LU/Uld &poro.tron. 1... Savarovsfty, eel.•
ButterworthJ. London (198U.

50 JOURNAL OF METALS • September 1985