Langmuir 2001, 17, 7137-7144 7137

Cake Collapse in Pressure Filtration†

D. Antelmi,‡,§ B. Cabane,*,⊥ M. Meireles,# and P. Aimar#
Equipe mixte CEA-RP, RHODIA, 93308 Aubervilliers Cedex, France;
Research School of Chemistry, Institue of Advanced Studies, Australian National University,
Acton, ACT 0200, Australia; Laboratoire PMMH, ESPCI, 10 rue Vauquelin,
75231 Paris Cedex 05, France; and CNRS UMR 5503, Laboratoire de Génie Chimique,
Université Paul Sabatier, 31062 Toulouse Cedex 4, France

Received March 26, 2001. In Final Form: July 31, 2001

Aqueous dispersions of latex particles have been aggregated by addition of Ca2+ ions and then filtered
in a pressure filtration cell. The filtration cakes have been examined through small-angle neutron scattering,
void volume fraction measurements, and permeate flux measurements. At all filtration pressures (20-400
kPa), the cakes were extensively collapsed (latex volume fractions φ ) 0.27-0.46), and the remaining
porosity had a tenuous structure (fractal dimensions df ) 1.3-1). However, the local structures remained
the same as in the aggregates of the original suspensions. The mechanisms that produce this collapse are
made of very small relative motions of the particles, which leave the local coordination of the latex particles
unchanged but allow large voids to be reduced. These motions could be inhibited by using particles with
nonspherical shapes and by increasing the friction forces that act between particle surfaces, thereby
reducing the extent of collapse and increasing the permeability of the cakes.

Introduction Higher flow rates may be achieved by aggregating the

Dispersions of fine solid particles in a liquid are
commonly made or encountered in manufacturing pro-
cesses (e.g., ceramics, pigments, adhesives, paints, phar-
maceuticals) as well as in biotechnology and biomedical
applications (biofluids, including blood and whey, fer-
mentation broths) and in oil recovery (drilling fluids). In
many cases, it is at some point necessary to separate the
particles from the liquid. This may be achieved through
a variety of processes including sedimentation, centrifu-
gation, membrane filtration, and drying. All of these
processes have some inherent inefficiency, and none of
them are totally satisfactory. Often, membrane filtration
is one of the most practical options, particularly if the
solid/liquid system cannot be heated.
One of the main factors that determine the operating
costs of membrane systems is the magnitude of permeate
flux that can be achieved. When filtering dispersions of
very fine particles, the magnitude of permeate flux can be
severely limited by membrane fouling and cake formation,
often rendering the technique economically unattractive.1
The formation of a cake results from accumulation of
particles that are driven toward the membrane by the
Stokes force, despite the interparticle forces and back-
diffusion that tend to keep them apart. If the pressure
applied to the fluid is high, the Stokes force will cause the
particles to pack densely in the cake on top of the filter
membrane. Such cakes have very small pores, and they
reduce the flow rates to very low values.2-5
†
This work used the neutron beams of ILL.
‡ Equipe mixte CEA-RP, RHODIA.
§ Australian National University.
⊥ Laboratoire PMMH, ESPCI.
# Université Paul Sabatier.
* To whom correspondence should be addressed.
(1) Schweitzer, P. A., Ed. Handbook of Separation Techniques for
Chemical Engineers, 3rd ed.; McGraw-Hill: New York, 1997.
(2) Harmant, P.; Aimar, P. AIChE J. 1996, 42, 3523.
(3) Bacchin, P.; Aimar, P.; Sanchez, V. AIChE J. 1995, 41, 368.
(4) Benkhala, Y. K.; Ould-Ris, A.; Jaffrin, M. Y.; Si-Hassen, D. J.
Membr. Sci. 1995, 98, 107.
(5) McDonogh, R. M.; Fane, A. G.; Fell, C. J. D. J. Membr. Sci. 1989,
43, 69.
particles before filtration.6,7 Indeed, the packing of irregular
aggregates in the cake leaves larger pores than a regular
packing of particles that repel each other. The magnitude
of the improvement may be estimated according to the
expressions for the flow rates. Assume that a pressure
difference ∆P is applied to the fluid that flows through
the cake (not to the cake itself; otherwise, the experiment
measures resistance to expression rather than filtration8).
Then the flow rate dV/dt across the cake and membrane
assembly is given by Darcy’s law:9
dV/dt ) A∆P/µ(Rm + Rc) (1)
where A is the area of the membrane, µ the kinematic
viscosity of the solvent, Rm the hydraulic resistance of the
bare membrane, and Rc that of the filter cake. The cake
resistance, Rc, can be used to define a specific resistance,
R, according to the mass of deposited particles, M:
Rc ) RM/A (2)
For a uniform cake, made of densely packed particles,
the specific resistance R may be calculated from the
Carman-Kozeny equation:10
R ) 180(1 - )/(FD23) (3)
where  is the pore volume fraction (on the order of 0.26
for densely packed particles), D the mean diameter of the
particles, and F their density. If the particles are small (D
in the nanometer range), the flow rates can be extremely
slow, leaving little scope for enhancement. On the other
hand, a cake made by the packing of fractal aggregates
can have a larger effective particle size. Assume for
(6) Waite, T. D.; Schäfer, A. I.; Fane, A. G.; Heuer, A. J. Colloid
Interface Sci. 1999, 212, 264.
(7) Waite, T. D. Colloids Surf. A 1999, 151, 27.
(8) Sorensen, P. B.; Moldrup, P.; Hansen, J. A. A. Chem. Eng. Sci.
1996, 51, 967.
(9) Kim, K. Y.; Chen, V.; Fane, A. G. J. Colloid Interface Sci. 1993,
155, 347.
(10) Carman, P. C. Trans. Inst. Chem. Eng. 1938, 16, 168.

10.1021/la0104471 CCC: $20.00 © 2001 American Chemical Society

Published on Web 10/05/2001
7138 Langmuir, Vol. 17, No. 22, 2001 Antelmi et al.

instance that each aggregate has a fractal dimension df

and an overall diameter ξ. The packing of such aggregates
in the cake yields an average volume fraction of particles,
φ, which is7,11,12

φ ≈ (a/ξ)3-df (4)

If the fractal dimension df is 2 or above, the corre-

sponding aggregates would be essentially impermeable
to the fluid, which will drain mostly between the ag-
gregates. Hence, for a dense packing of the aggregates,
the effective particle size put into the Carman-Kozeny
equation should be ξ, and the porosity will correspond to
the volume of voids between the aggregates, again of the
order of  ) 0.26. Taking for example df ) 2 and φ ) 0.1,
the effective particle size ξ is 10 times the size of
elementary particles, a. According to the Carman-Kozeny
equation, the flux would then be 100 times higher than
for a packing of nonaggregated particles.
In some cases, the gains in flux obtained by preaggre-
gating the particles are indeed spectacular. For instance,
the filtration of mineral particles at pH values where they
spontaneously aggregate (isoelectric point) may give flow
rates that are 10 times faster than the filtration at pH
values where they resist aggregation. Further gains may
be obtained by producing aggregates that do not pack
densely.6
In other cases, the gains obtained by preaggregating the
particles are much less. The reason is that the cake
collapses, despite the interparticle bonds that keep the
particles aggregated.13-15 The collapse of filtration cakes
is the most severe limitation to the efficiency of pressure
filtration. To overcome this limitation, one must have
detailed information on the processes by which the cakes
collapse. Specific questions are as follows: (1) Does the
whole cake collapse, or is there a thin collapsed layer next
to the membrane that limits the flux and prevents
densification of the rest of the cake? (2) What is the
structure of the compressed cake, what are the pore sizes,
and how are they connected? (3) Through what mecha-
nisms did the cake collapse, e.g., rupture of interparticle
bonds and interpenetration of aggregates, or continuous
deformation of the aggregates leading to progressive
collapse of the voids? (4) Can these mechanisms be
controlled by changing the structures and the mechanical
strengths of the aggregates?
In the present work, we addressed these questions
through a determination of the structures of filtration
cakes at various stages of compression. We used small-
angle neutron scattering (SANS) to measure the distribu-
tion of interparticle distances in the whole cake. A
description of this technique, and of its use in the
characterization of deposits, is found in our previous work
on the filtration of clay suspensions.16 To facilitate the
analysis of the results, we used a model system consisting
of monodisperse, spherical polystyrene latex particles.
These particles were preaggregated through addition of
Ca2+ ions and filtered on “Anopore” alumina membranes.
The solvent was a D2O/H2O mixture that canceled the
scattering from the membrane, thereby making it possible
(11) Cabane, B.; Dubois, M.; Lefaucheux, F.; Robert, M. C. J. Non-
Cryst. Solids 1990, 119, 121.
(12) Dubois, M.; Cabane, B. Macromolecules 1989, 22, 2526.
(13) van de Ven, T. G. M. In Colloidal Hydrodynamics; Ottewill, R.
H., Rowell, R. L., Eds.; Academic Press: London, 1989; p 351.
(14) Tiller, F. M.; Kwon, J. H. AIChE J. 1998, 44, 2159.
(15) Lee, D. J.; Ju, S. P.; Kwon, J. H.; Tiller, F. M. AIChE J. 2000,
46, 110.
(16) Pignon, F.; Magnin, A.; Piau, J. M.; Cabane, B.; Aimar, P.;
Meireles, M.; Lindner, P. J. Membr. Sci. 2000, 174, 189.
Figure 1. SEM micrographs showing the surface and cross
section of a clean Anopore membrane with a nominal pore size
of 20 nm.
to observe the undisturbed cake + membrane assembly.
We obtained scattering curves for the original aggregated
dispersion, for cakes made from this dispersion at different
filtration pressures, and also for cakes made form disper-
sions of nonaggregated particles. We also measured the
extent of cake collapse, through simple gravimetric
methods, and changes in the hydraulic resistance of these
cakes. The comparison of the microstructures determined
through SANS with the overall characteristics of the cakes
yields some of the structural information needed to answer
the questions listed above.
Experimental Section
A. Materials. Latex Dispersions. The latex particles were
made through emulsion polymerization of styrene (main com-
ponent) with acrylic acid (1%), acrylamide (1%), and styrene-
sulfonate (0.5%); sodium dodecyl sulfate (SDS) was used as an
emulsifier and persulfate as an initiator. The core was made of
polystyrene (glass transition temperature Tg ) 100 °C, density
1050 kg/m3), and the surface contained the hydrophilic monomers,
some sulfonate groups, and some SDS. The average size of the
particles was 150 nm as measured from neutron scattering and
electron microscopy. These dispersions are stabilized by elec-
trostatic repulsions originating from the ionization of the particle
surfaces.
The dispersions of latex particles used in filtration experiments
had a concentration of 0.5 wt %. Flocculated dispersions were
prepared by diluting stock latex dispersion with 0.06 M CaCl2.
At this salt concentration, the electrostatic interactions are
effectively screened, and rapid coagulation takes place, forming
large aggregates of the order of 10 µm as approximated from
scanning electron microscopy. In these flocculated dispersions
the overall latex concentration was also 0.5 wt %.
Membranes. Whatman Anopore membranes with a nominal
pore size of 20 nm and surface area of 1555 mm2 were used
throughout this study. Figure 1 shows scanning electron
microscopy (SEM) pictures of the Anopore membrane. The top
view shows that the pores are regularly spaced, with a spacing
Cake Collapse in Pressure Filtration
of about 250 nm. This repetition shows up as a peak in the SANS
scattering curves, when the membrane is not matched by the
solvent. The side view of a broken piece of membrane shows that
the pores are nearly straight. This feature is important, since it
minimizes the internal fouling of the membrane during filtration
by any particles that enter inside the pores.
B. Methods. Filtration Experiments. Ultrafiltration ex-
periments were performed in an unstirred ultrafiltration cell
(Amicon 8050) of 80 mL capacity, consisting of a cylindrical section
with two end pieces. The top piece was used to maintain the cell
at a constant pressure, obtained from a regulated N2 supply. The
bottom piece was used as a support for the membrane. Before
each experiment, 200 mL of distilled water was passed through
the membranes at 200 kPa to clean the membranes and ensure
a complete wetting of the pores. For membranes to be used in
the neutron scattering experiments, immediately prior to the
experiment, two 50 mL volumes of D2O/H2O solvent (in the ratio
75:25, see below) were passed through the membrane to fill the
pores with solvent at the contrast match composition.
The experiments were conducted with cells completely filled
with the dispersion, submitted to fluid pressures in the range
0-400 kPa at a temperature of 20 °C. During the filtration of
the dispersion, the particles accumulated on the membrane to
form a cake. The experiment was terminated, however, just before
the level of the suspension reached the surface of the cake. The
permeate flux was determined gravimetrically by weighting the
permeate on an electronic balance at regular time intervals. The
resistance of the membranes was determined by measuring the
particle-free solvent flux before each experiment. It was observed
that the membrane resistance to solvent containing 0.06 M CaCl2
was slightly lower (∼8%) than for Milli-Q water, presumably
due to electroviscous effects.17
Latex Volume Fraction of the Cakes. The latex volume
fraction of the cakes was determined in two ways. First, it was
measured directly through gravimetric analysis at the end of
filtration. Second, it was calculated using a simple mass balance
between the latex content of the initial dispersion and that in
the retentate, assuming complete retention of the particles.
Agreement between the two methods was within 5-10%.
In the direct method, the filter membranes were first weighed
in their clean, dry state before being placed in the filtration cell.
At the end of filtration, the wet filter membranes supporting the
wet cake were blotted on absorbent paper to remove excess water.
The wet membrane and deposits were then weighed before being
placed in an oven at 120 °C overnight with dry N2 constantly
circulating through the oven. The dry membranes with the dry
deposit were then reweighed. The weights of the wet and dry
cakes are then given by straightforward differences. (The amount
of water trapped in the pores of the membrane was measured
separately, for five separate membranes, and consistently found
to be around 10% of the weight of the dry membrane. This was
taken into account to calculate the mass of the wet cake on a wet
membrane.)
The weight of latex deposited on the membrane was also
calculated from the known volume fraction in the initial
dispersion and the weight of permeate that passed through the
membrane during the filtration (measured gravimetrically). This
analysis assumes that no latex passed through the membrane,
which is well justified given that the pores (20 nm) were much
smaller than the individual latex particles. The mass of latex in
the cake determined in this manner was used to check against
the gravimetric method described above.
Finally, the measured weights of wet and dry cakes were used
to calculate the latex volume fraction of the cake, Φ, using the
densities known as 1.05 g/cm3 for latex and 1 g/cm3 for water.
The volume fraction of the pores in the deposit is then
 ) 1 - Φ ) 1 - [Vlatex/(Vlatex + Vwater)] (5)
Small-Angle Neutron Scattering. Small-angle neutron
scattering is a technique that measures interference patterns
produced by the spatial variations of a quantity called the “density
of scattering length”, which is the equivalent, for neutrons, of an
(17) Huismans, I. H.; Dutré, B.; Persson, K. M.; Trägårdh, G.
Desalination 1997, 113, 95.
Langmuir, Vol. 17, No. 22, 2001 7139
index of refraction.18,19 If the sample contains particles dispersed
in a solvent, the scattered intensity is proportional to the square
of the difference in density of scattering length between the
particles and the solvent. The angular distribution of scattered
neutrons is a Fourier transform of the distribution of distances
between the particles. The resulting interference pattern consists
of intensity values collected at various values of the scattering
vector Q, which is the difference between the wave vectors of
incident and scattered neutrons. For isotropic samples, this
scattering pattern can be regrouped radially to give a scattering
curve, where the intensity varies according to the magnitude Q
of the scattering vector; Q is related to the scattering angle θ and
to the wavelength λ by
Q ) (4π/λ) sin(θ/2) (6)
The high Q part of the scattering curve corresponds to
interferences between pairs of scattering centers that are quite
close to each other, e.g., on the same particle. This part of the
scattering curve is used to count how many particles contribute
to the scattering. In practice, the scattering curves are scaled to
give the same scattering intensity in this range, making it possible
to compare the scattering from different structures with the same
number of particles. The intermediate part of the scattering
curves is controlled by interference between neighboring particles.
This part of the scattering curves is used to determine what is
the coordination shell of a particle; e.g., is it surrounded by a
complete shell of neighbors at the contact distance, as in a crystal,
or does it have a less dense environment? Finally, the low Q part
of the scattering curves is controlled by interference between
remote scattering centers. In this range, samples with a uniform
particle density give no scattering because the contributions from
individual particles cancel each other through destructive
interference. Therefore, scattering in this range originates from
large-scale fluctuations in the density of particles within the
aggregates or within the cake.
SANS was considered an ideal method to probe the structure
of the filtration cakes since it allowed the cake to be left
undisturbed on the membrane. Indeed, using the contrast match
method, the solvent could be chosen to cancel the scattering from
the membrane to obtain the scattering from the cake alone.19
The composition of the solvent (a D2O/H2O mixture) that
matched, and thereby canceled, the scattering from Anopore
membranes was determined in the following way. An Anopore
membrane was loaded into the filtration unit and charged with
solvent of specific D2O/H2O composition by passing about 80 mL
of solvent through at 200 kPa two times, making sure that the
surface of the membrane was always covered with solvent. A
roughly rectangular piece of membrane was broken off and
quickly placed in a quartz cell of 1 mm thickness filled with the
appropriate solvent. The procedure was repeated for a range of
solvents with varying D2O/H2O compositions. The scattered
intensity produced by the neutron beam passing normally
through the membrane was measured for each isotopic composi-
tion. The usual plot of the square root of intensity vs isotopic
composition gave the contrast match point. This was found to be
located at 75% D2O + 25% H2O, which is significantly lower than
the contrast match point for pure alumina (92% D2O).
The accuracy of the contrast match point determined in this
way was assessed by examining the scattering from a membrane
in the region of Q that corresponds to distances between
neighboring pores of the membrane (250 nm). With solvents that
are away from the match point, the membrane scattering has a
very intense peak at the corresponding Q value. With solvents
that are at the match point, this peak is completely extinct.
Adsorption of a partial layer of latex particles on the membrane
returns a weak peak at this location.
For SANS experiments on the filtration cakes, the procedure
was essentially the same. A latex dispersion made with a solvent
of the same isotopic composition (75/25) was filtered on the
Anopore membrane in the filtration cell while the amount of
permeate was monitored as a function of time. At the completion
(18) Champeney, D. C. Fourier Transforms and Their Physical
Applications; Academic Press: New York, 1973.
(19) Jacrot, B. Rep. Prog. Phys. 1976, 39, 911.
7140 Langmuir, Vol. 17, No. 22, 2001
Figure 2. Scattering curves of the aggregated dispersion
(crosses) and of the cake made from this aggregated dispersion
(circles) filtered at 20 kPa.
of the filtration, the remaining retentate was gently swirled and
tipped out of the filtration cell. A rectangular piece of membrane,
with the cake on it, was broken off and placed in a quartz cell
with a 1 mm path length. Some permeate was used to top up the
quartz cell before closing it tightly. Great care was taken not to
disturb the cake on top of the membrane surface. The quartz
cells were then mounted on a holder for the scattering experi-
ments.
Results
Structures. In the following sections we present the
small-angle neutron scattering curves obtained with the
aggregated suspension, before filtration, and with the
membrane + cake assembly, after filtration.
Structures of the Aggregated Dispersions. The
scattering curve of the original latex dispersion, ag-
gregated by addition of Ca2+ ions, is shown in Figure 2.
At high Q, the oscillations of the scattering curve reflect
the size and shape of the latex particles. For monodisperse
spheres these oscillations would be quite strong (the
intensity would have a series of zeros), and for polydisperse
particles they are damped. The observed oscillations reflect
a weak polydispersity (standard deviation ) 10%). This
is in agreement with SEM pictures of the latex dispersion
on the membrane.
At Q values that correspond to one or two particle
diameters ((2-4) × 10-3 Å-1), the shape of the spectrum
reflects correlations between neighboring particles. A
structure where each particle was surrounded by a
complete shell of neighbors (8-12 particles) would show
a correlation peak at this location of the spectrum. The
experimental scattering curve does not have a peak or an
oscillation in this range of Q; hence, it may be concluded
that each particle is in direct contact with a reduced
number of particles. Thus, on a local scale, the aggregates
are not dense.
At small Q values, the scattering curve reflects large-
scale correlations in the distribution of matter within an
aggregate. In this range of distances (150-1500 nm) the
intensity follows a power law I(Q) ≈ Q-2.45. This power
law (df ) 2.45) indicates that the aggregates have a “bushy”
fractal structure, somewhat more dense than the struc-
tures of aggregates made by reaction limited cluster
aggregation (df ) 2.2).7 The high value of the exponent
indicates that some reordering has taken place after
aggregation.
The overall size of these aggregates may be estimated
from the fact that they could barely be seem by eye; this
indicates a size on the order of 10 µm. A confirmation of
Antelmi et al.
this size was obtained from their sedimentation rate.
Indeed, in H2O, these aggregates were seen to slowly
sediment, due to their small density increment (∆F ) 50
kg/m3). For fractal aggregates of size R, the gravitational
force is calculated from the number of particles per
aggregate, N ) (R/a)df, and the volume per particle, V )
(4π/3)a3
Fg ) (R/a)df(4π/3)a3∆F (7)
The frictional force is calculated from the aggregate
radius, R, the aggregate velocity, u, and the viscosity of
water, η ) 10-3 N m-2 s:
Fu ) 6πηRu (8)
In steady-state sedimentation, these forces are equal.
For a sedimentation velocity u ) 10 mm/h, this yields
sizes slightly in excess of 10 µm.
Structures of Cakes Made at 20 kPa. The scattering
curve of a wet filtration cake is presented in Figure 2. At
high Q values, corresponding to intraparticle distances,
the spectrum shows the usual oscillations and Q-4 decay
due to intraparticle interferences, as in the original
dispersion. Consequently, the intensity scales of the cake
and of the dispersion were shifted so that the spectra
superposed in this range of Q, corresponding to scattering
by identical numbers of particles.
At intermediate Q values, comparable to distances
between neighboring particles, the spectra are also similar.
Accordingly, the packing of particles in the cake does not
modify the organization of particle-particle contacts.
At small Q values, which correspond to large-scale
fluctuations of the particle distribution, the intensity
scattered by the cake is far below that scattered by the
dispersion. Accordingly, the destructive interference is
stronger in the cake than in the dispersion. This is caused,
of course, by the denser packing of particles in the cake,
which limits the magnitude of fluctuations in the particle
density. This is a well-known effect for random packing
of hard spheres20 or dense packing of fractal aggregates.
At high volume fractions, the effect is so strong that the
total scattered intensity is predicted to decrease when
more particles are added (Figure 3). In this regime, it is
easier (and equivalent, through Babinet’s theorem) to
describe the scattering as produced by the voids that
separate the particles. Accordingly, the power law observed
at low Q for the intensity scattered by the cake reflects
directly the fractal structure of the voids. (It also reflects
the fractal structure of the aggregates, but then interfer-
ence between neighboring aggregates must be taken into
account.) Then, the value of the exponent, which is -1.33,
implies that these pores are tenuous.
Structures of Cakes Made at Pressures 120 and
400 kPa. Figure 4 shows the scattering curves of cakes
made by filtering, at different pressures, identical amounts
of aggregated latex dispersions. At high Q values, in the
Q range where the scattering reflects interference within
a particle, the raw spectra are identical, as expected for
samples containing the same amount of latex particles.
At intermediate Q values, in the range corresponding to
distances between neighboring particles, the spectra are
also similar. The absence of a scattering peak indicates
that the particles still do not have complete coordination
shells. Finally, at small Q values, all spectra are power
laws of Q, with exponents that depend on the applied
pressure.
(20) Ashcroft, N. W.; Lekner, J. Phys. Rev. 1966, 145, 83.
Cake Collapse in Pressure Filtration
Figure 3. Intensity scattered by a hard sphere liquid,
calculated through the Percus-Yevik equation.20 At volume
fractions in excess of φ ) 0.12, interparticle interferences depress
the scattered intensity so much that the total intensity decreases
upon adding more particles into the liquid.
Figure 4. Scattering curves of cakes made from aggregated
dispersions filtered at 20 (highest curve), 100, 200, and 400
kPa (lowest curve).
The evolution of the spectra is interesting: the ap-
plication of higher pressures produced cakes that gave a
weaker scattering, with an exponent of lower magnitude.
Since the volume fraction of particles in the cake had
increased, a weaker intensity must result from stronger
destructive interference between particles. As indicated
above, in this regime of strong interparticle interference,
it is easier and equivalent to describe the intensity in
terms of the voids that separate the particles. Accordingly,
the variation of the exponent, from -1.33 to -1, indicates
that the voids become still more tenuous as the pressure
is increased.
Structures of Cakes Made from Nonaggregated
Dispersions. A remarkable feature of the cakes made
from aggregated dispersions is that they retain a set of
tenuous voids. Thus, they do not collapse to a fully dense,
ordered structure, within the range of applied pressures
studied here. To confirm that this set of tenuous voids is
a memory of the structure of the original aggregates, we
also made filtration cakes deposited from nonaggregated
dispersions.
The filtration was performed using a nonaggregated
latex dispersion, with NaCl added rather than CaCl2. In
Langmuir, Vol. 17, No. 22, 2001 7141
Figure 5. Scattering curves of a cake made from an aggregated
dispersion filtered at 400 kPa (triangles) and of a cake made
from a nonaggregated dispersion, filtered at 200 kPa and
washed by a CaCl2 solution in order to lock the structure (stars).
these conditions, the latex particles do not adhere upon
mutual contact, and normally, the cake that forms on the
membrane redisperses spontaneously as soon as the flow
is stopped. To counteract the redispersion, a small amount
of CaCl2 solution was fed into the filtration cell under the
same pressure difference at the end of the filtration. The
Ca2+ ions formed bonds between the particle surfaces in
the cake and locked them into their positions, preventing
redispersion. Although this technique does not give an
ideal representation of the structure that a cake comprised
of repelling particles would have, it does provide a useful
contrast to the cakes formed from preaggregated particles.
Once again, the intensity scales have been adjusted to
give the same intensities at high Q, corresponding to the
same numbers of particles.
The spectrum of the cake made from nonaggregated
particles shows a peak corresponding to the distance
between neighboring particles, indicating that each
particle is surrounded by a complete shell of neighbors
(Figure 5). At lower Q values, the intensity given by cake
is depressed. This depression is caused by strong destruc-
tive interference at scales beyond the distances between
first neighbors, indicating that the cake is homogeneous
at those scales.
The spectrum of the cake made from an aggregated
dispersion differs by the excess scattering at low Q, caused
by heterogeneities (voids) with sizes ranging from 1 to 10
particle diameters.
At this stage, the microstructures of the filter cakes
may be described as follows. Filter cakes made by packing
repelling particles have a homogeneous structure (few
large-scale fluctuations of the particle density) with a
strong short-range order, as expected. Filter cakes made
by packing preaggregated particles show large-scale
fluctuations of the particle density. In comparison with
the density fluctuations of the original aggregates, the
density fluctuations of the filter cakes are strongly
attenuated. These density fluctuations are spatially self-
similar, with fractal dimensions in the range 1-1.33,
characteristic of tenuous objects such as pores with low
connectivity.
Overall Characteristics of Cakes. In the following
sections we present the overall characteristics of cakes
formed at different pressures in terms of void fraction (or
latex volume fraction) and hydraulic resistance. These
overall characteristics were evaluated independently. The
7142 Langmuir, Vol. 17, No. 22, 2001
Figure 6. Overall collapse of filter cakes. Vertical scale:
average volume fraction of latex particles in the filter cake,
from gravimetric measurements. Horizontal scale: pressure
applied to the supernatant fluid.
cake void fraction was evaluated both from gravimetric
experiments and from a mass balance, as explained in
“Methods”. The cake hydraulic resistance was evaluated
from filtration experiments, also in two independent ways.
Void volume fraction and cake hydraulic resistance
measurements indicate a strong evolution of the cake with
the applied pressure.
Latex Volume Fraction. The values of the latex
volume fraction φ are presented in Figure 6. They show
two successive stages for the packing of latex aggregates
(flocs) in the cake under the effect of applied pressure:
(i) Application of moderate pressures (0-20 kPa) is
enough to pack loose flocs into a fairly dense cake (φ )
0.27). This is achieved through collapse of the spaces
between flocs and also of the largest voids within the flocs.
In this regime, the dispersion is still highly compressible.
(ii) Application of much higher pressures (up to 400
kPa) produces a (modest) further densification of the cake
(up to φ ) 0.46). This is achieved through further collapse
of voids in the cake. Since the increase in volume fraction
achieved at this stage is comparatively small, the cake
may be considered as practically consolidated. This is the
point where the compressive yield stress of the material
diverges with further increases in volume fraction.21,22
These volume fractions are averages over the whole
volume of the cake. Of course, the local solid pressures
and volume fractions must vary according to the depth z
within the cake. However, it is obvious from the variation
shown in Figure 6 that cakes made at high pressures must
be collapsed over most of the cake height. Consider, for
instance, a cake made at a total pressure difference ∆P
) 400 kPa. At each depth in this cake, the pressure exerted
on the particles is the sum of hydrodynamic forces exerted
by the permeating fluid on all particles located above this
level. At the bottom of the cake, this pressure is the highest,
and it equals the applied pressure, i.e., 400 kPa, leading
to full collapse at the bottom of the cake. Strong collapse
must also occur at other depths in the cake, as long as
these layers support the forces transmitted by a sub-
stantial fraction of the cake mass. The only part of the
cake that is submitted to pressures less than 20 kPa is the
top layer, which, according to the pressure gradient, may
contain no more than 5% of the cake mass. According to
this pressure, this top layer may have a volume fraction
on the order of 0.27. Therefore, the cake is collapsed over
(21) Channel, G. M.; Zukoski, C. F. AIChE J. 1997, 43, 1700.
(22) Channel, G. M.; Miller, K. T.; Zukoski, C. F. AIChE J. 2000, 46,
72.
Antelmi et al.
Figure 7. Specific resistance of cakes made at different
pressures, calculated from the final permeation rate and final
weight of the cake (filled diamonds), or from the variation of
the permeation rate during the course of filtration (circles).
most of its height. This is in agreement with the results
of cake dissection experiments, performed by Meeten:23
filtration cakes made from polystyrene latex had a uniform
volume fraction profile at φ ) 0.55, whereas filtration cakes
made from kaolinite had a continuous gradient of volume
fraction through the cake.
Hydraulic Resistance of the Cakes. The specific
resistance of the cakes was determined from the flow
rate measured at the end of the filtration (hydraulic
resistance of the cake + membrane assembly), the flow
rate measured before filtration (membrane resistance),
and the mass of cake deposited on the membrane,
according to eqs 1 and 2.
These resistances are rather high (1013-1014 m/kg),
compared with those of cakes made of other colloidal
particles. For instance, the specific resistances measured
for aggregated hematite suspensions submitted to similar
pressures are in the range 1012-1013 m/kg.6 Since the
individual particles have about the same sizes, the higher
values found in the present case must reflect a stronger
collapse of the cakes.
The variation of the specific resistance with pressure
(Figure 7) follows the law
R ) R0Pn (9)
where the exponent n is 0.5 for both sets of data, and the
prefactor R0 is 5 × 1013 m/kg. This increase in specific
resistance must result from a corresponding increase in
volume fraction throughout the cake. Here, a 4-fold
increase in resistance results from a rise in average volume
fraction from φ ) 0.27 to φ ) 0.46. Thus, the variation of
resistance with volume fraction must be quite fast in this
range. This is indeed expected for nearly dense packings
of colloidal particles, as shown by eq 3.
The flow rates measured during the formation of the
cake can also be used to calculate the specific resistance
of these intermediate states. A common observation in
pressure filtration experiments where the membrane
resistance is comparatively small is that the volume of
permeate grows as a square root function of time.24 There
is a pedestrian explanation for this behavior, which
assumes that particles are deposited on the cake at a rate
determined by the flow of permeate (i.e., there is no
gravitational sedimentation and no diffusion) and that
(23) Meeten, G. H. Chem. Eng. Sci. 1993, 13, 2391.
(24) Langman, K. A.; White, L. R. AIChE J. 1995, 41, 1687.
Cake Collapse in Pressure Filtration
Figure 8. Unstirred filtration of 0.5 wt % dispersions of latex
aggregates in a solvent of 75% D2O + 25% H2O at different
applied pressures plotted according to the cake filtration model
(eq 11). The linear behavior indicates that the cake grows at
constant specific resistance.
the cake grows at a constant specific resistance which
solely arises from Darcyan flow of the fluid relative to the
particles in the immobilized bed. Indeed, the mass of
deposited latex, M, used in eq 2, is related to the volume
of permeate, V(t), and to the mass concentration W of latex
in the dispersion:
M(t) ) V(t)W (10)
Consequently, the filtration data can be expressed in
a way that determines the average specific resistance R:
dt ) (µRm/A∆P) dV + (µRW/A2∆P)V(t) dV (11)
If the specific resistance R remains constant during the
growth of the cake, this can be integrated to give a filtration
law with only two parameters, the initial flux, Q0, and the
specific resistance, R:
t/V(t) ) (1/Q0) + (µRW/2A2∆P)V(t) (12)
Thus, if the conditions stated above are met, then t/V(t)
must be a linear function of V(t).9 Figure 8 presents the
filtration results, plotted in this way, for suspensions of
latex aggregates in the D2O/H2O mixtures. In this case,
there was no sedimentation, and the usual linear behavior
was observed. The specific resistance, calculated from the
slope of these lines, is close to that determined from the
permeation rate of the final cake. This suggests that the
cake grows at a nearly constant specific resistance.
At this stage, the macroscopic features of the filter cakes
made with preaggregated latex particles may be sum-
marized as follows. All cakes are fairly dense, even those
made at rather low pressures. This dense packing produces
a high resistance to permeation. Thus, the gain in
permeability obtained from preaggregating the particles
is almost totally lost through the mechanisms that lead
to cake collapse.
Discussion
The aim of this work was to shed some light on the
mechanisms by which filter cakes may collapse under
applied pressure. The term “collapse”, in the present
context, is defined as the processes by which a dispersion
of bushy aggregates is compressed into a dense cake. The
term “mechanisms” is defined as the set of particle motions
Langmuir, Vol. 17, No. 22, 2001 7143
that occur under the effect of applied pressure and that
produce this collapse. The link between the observed
collapse and the mechanisms that make it possible will
be examined in four steps: (a) What is the extent of
collapse? (b) At what spatial scale does collapse take place?
(c) Which type of particle motions make it possible? (d)
Which properties of the particles may favor or inhibit these
motions?
What Is the Extent of Collapse? There is no doubt
that collapse, or some type of densification, has occurred.
The initial dispersion contained bushy aggregates that
had a fractal exponent df ) 2.45 and overall dimensions
ξ in excess of 10 µm. Simple packing of these aggregates,
as in a sedimentation experiment, would give a volume
fraction φ ≈ (a/ξ)3-df ) 0.1. In the filtration experiment,
the applied pressure caused these aggregates to pack at
a higher volume fraction. At the lowest pressure (20 kPa)
the overall volume fraction of latex in the cake was already
φ ≈ 0.27, and at the highest pressure (400 kPa) it went
up to φ ) 0.46. Accordingly, these cakes are of the
“supercompactible” type. The distributions of volume
fraction of solids throughout such cakes have been
measured by Meeten23 and discussed by Tiller et al. and
Lee et al.25,26
At What Spatial Scales Does the Collapse Take
Place? The macroscopic collapse of the cakes must
originate from structural rearrangements that take place
at shorter spatial scales. Changes in the local structure
(i.e., the coordination shell of each latex particle) show up
in the scattering curves at Q values that match the
distances between nearest neighbors (3 × 10-3 < Q < 6
× 10-3 Å-1). In this range, cakes made from nonaggregated
dispersions show a diffraction peak (Figure 5), indicating
that each particle has a complete coordination shell and
therefore that the local structure is dense and regular.
However, cakes made from preaggregated suspensions do
not produce this peak (Figure 5); instead, their scattering
curves are identical, in this Q range, with that of the
original aggregated suspension (Figure 2). Therefore, these
cakes have retained the same local structures, with
incomplete coordination shells, as the original aggregates.
Thus, the processes that caused the collapse must have
taken place at larger spatial scales.
At large spatial scales, the collapse is evident. Indeed,
at Q values corresponding to large groups of particles (Q
< 3 × 10-3), the intensities scattered by the cakes are
systematically lower than that of the original dispersion
(Figure 2). Moreover, the application of higher pressures
makes them lower still (Figure 4). This depression of the
intensity reflects the suppression of fluctuations of the
particle concentration, such as alternations of lumps and
voids. The remaining voids are tenuous, as indicated by
the fractal exponent df, which is 1.3 at lower pressures
and 1 at the highest pressure.
This interpretation states that the large-scale porosity
is collapsed throughout the cake. At this point, it is
instructive to examine the alternative interpretation,
according to which the shape of the scattering curves would
be related to a spatial gradient in the cake. Assume that
the bottom part of the cake were fully collapsed, with a
dense ordered structure similar to that of the cake made
from a nonaggregated dispersion (Figure 5). Assume also
that the top part of the cake had resisted the collapse and
retained the same aggregate structure as the original
suspension. Then the scattering curve of the whole cake
(25) Tiller, F. M.; Kwon, J. H. AIChE J. 1999, 44, 2159.
(26) Lee, D. J.; Ju, S. P.; Kwon, J. H.; Tiller, F. M. AIChE J. 2000,
46, 110.
7144 Langmuir, Vol. 17, No. 22, 2001
would be a linear combination of both scattering curves.
This combination would show a steep decay at very low
Q values (Q-2.45 power law in Figure 2) and a peak at the
nearest-neighbor distance (scattering curve of the cake
made from nonaggregated particles, shown in Figure 5).
None of these features are observed in the experimental
scattering curves of the cakes. Thus, the conclusion is
that, throughout the cake, the local structure has remained
the same as in the original suspension, while the large-
scale porosity has collapsed and left only tenuous voids.
What Types of Particle Motions Make This Col-
lapse Possible? The deformations that cause the collapse
must result from relative motions of particles that are in
contact. These motions must have small amplitudes, since
the local structures remain essentially unchanged. Thus,
particles that are in contact are displaced or rotated by
small amounts, and these relative motions accumulate to
produce large-amplitude continuous motions of structural
elements such as clusters or voids.
Why Does the Cake Collapse in This Way? A
continuous deformation, such as that postulated in the
mechanism described above, requires that the particles
glide or rotate around each other. The main factor that
favors this type of deformation is the fact that the latex
particles are spherical. Indeed, simple rotation of one
particle around another may be enough to bend a branch
of an aggregate. Only friction forces that tend to maintain
the established interparticle contacts oppose these rota-
tions.
In the latex aggregates that we have examined, the
friction forces originate from Ca2+ ions that bind the
particle surfaces together. Since the aggregates collapse
under moderate pressures, the yield stress of these bonds
must be low compared to the forces applied to particle
surfaces. More specifically, the yield stress must be low
compared to forces applied to large voids (these are
collapsed efficiently), but not compared to forces applied
to the smallest voids (since some porosity remains in the
cakes).
Antelmi et al.
How Can This Collapse Be Controlled? In practice,
two main factors should limit the collapse of filter cakes.
First, aggregates of nonspherical particles, containing
facets and edges, would resist collapse better, because
the relative displacements of particles would require shear
of interparticle contacts instead of pure rotations. Alter-
natively, collapse and densification will be limited if the
friction forces between particles are high, as dictated by
the strength of interparticle bonds. Enhancing either of
these two effects should stabilize a cake to pressure
increases.
Conclusions
Some aggregated suspensions, when filtered, form
“supercompactible” cakes. The mechanism by which such
cakes collapse and consolidate involves very small relative
displacements of particles that are in contact with each
other. These motions leave the local coordination of each
particle unchanged; however, the continuous deformations
that result from the accumulation of such motions cause
all large voids to collapse. The remaining voids are
tenuous, and the flux of permeate though the consolidated
cakes is reduced to very low values.
The suspensions that form these cakes are made of
spherical particles, held together by physical forces. The
collapse of these cakes is easy, because the friction forces
that oppose relative motions are low. Control of this
collapse may be achieved by increasing the friction forces
that act on interparticle contacts or by using nonspherical
particles. This would provide a way to maintain a low
hydraulic resistance in the cakes.
Acknowledgment. We thank O. Aguerre-Chariol for
the scanning electron micrographs and S. Egelhaf and P.
Lindner for help and useful suggestions with the neutron
scattering experiments.
LA0104471