Separation and Purification Technology 165 (2016) 114122

Contents lists available at ScienceDirect

Separation and Purification Technology

journal homepage: www.elsevier.com/locate/seppur

The permeability prediction of beds of poly-disperse spheres with

applicability to the cake filtration
Sebastian Osterroth a, Caleb Preston b, Bojan Markicevic b,, Oleg Iliev a, Mark Hurwitz c
a
Fraunhofer ITWM & University of Kaiserslautern, 67663 Kaiserslautern, Germany
b
Pall Corp., 3669 State Route 281, Cortland, NY 13045, United States
c
Robert Frederick Smith School of Chemical and Biomolecular Engineering, 171 Kimball Hall, Cornell University, Ithaca, NY 14853, United States

a r t i c l e i n f o a b s t r a c t

Article history: The interactions between the cake and depth filtration of poly-disperse spherical particles is examined by
Received 23 December 2015 comparing experimentally measured cake permeability to the permeability predictions of analytical
Received in revised form 9 March 2016 models. In the experiments, the influence of the cake forming history on the cake permeability is inves-
Accepted 28 March 2016
tigated, where the parameters as (i) different filtration materials, (ii) test suspension flow rates, and (iii)
Available online 29 March 2016
particle concentrations in the suspension are varied. The permeability models are given as product of pre-
set constant, porosity function and square of characteristic particle size. For the poly-disperse porous
Keywords:
media, the characteristic particle size has to account for the distribution of particle sizes which is typi-
Cake filtration
Poly-disperse particles distribution
cally accomplished through the use of various moments of the distribution. Clearly, the size distribution
Effective/average particle diameter function of particles forming the cake has to be utilized which is obtained after correcting the original
Cake permeability prediction distribution function of particles used in the test suspension for the particles which pass through the
(experimental/models) cake. This implies that the particles have to be counted after the test suspension passes the filtering mate-
rial. Following this framework, a set of experiments is carried out to determine the permeability of poly-
disperse cake. For each experiment, the permeability is also evaluated analytically using four different
long-established models in combination with different averages for the particle diameter of the poly-
disperse particle sample, trying to identify an averaging rule for which the analytical predictions are most
close to the experimental results.
2016 Elsevier B.V. All rights reserved.

1. Introduction
Filtration is widely used to separate particles from gases or liq-
uids in numerous industries including petroleum, food, pharma-
ceutical and automotive, to name just a few. In the majority of
these applications, in addition to particle entrapment within the
porous filter medium itself, a layer of particles, called a cake, is
formed at the upstream surface of the filtering medium and con-
tributes substantially to the flow resistance of the filter. Prediction
of cake permeability is clearly an important aspect of filtration pro-
cess design. There are two major difficulties with predicting cake
permeability. First, in the majority of applications, the particles
comprising the cake are poly-disperse in size requiring determina-
tion of an average, effective, particle size if a mono-disperse model
is to be employed. Second, in many applications, the cake is formed
in the second stage of the filtration process, with the first being
depth filtration. Finally, some particles pass all the way through
Corresponding author.
E-mail address: bojan_markicevic@pall.com (B. Markicevic).
the filter and cake, so that the cake permeability is due to a particle
size distribution different from that of the upstream suspension.
For slow flows, the fluid velocity is proportional to the pressure
gradient resulting from viscous interaction with the surfaces of the
medium. That is, the flow obeys Darcys law. The proportionality
constant is the ratio of the permeability to the fluid viscosity [1].
In the groundwater sciences the expression hydraulic conductivity
[2,3] and in the cake filtration community the expression (specific)
cake resistance [4,5] are commonly used to denote, respectively,
the permeability or its inverse. Many different expressions for
the permeability exist in the literature. Common to all is that the
permeability is expressed as a product of a preset constant, specific
geometrical function and particle size. The best known example is
perhaps the Carman-Kozeny relation [6]. The geometrical function
in the permeability models is calculated from either equivalent
medium consisting of parallel tubes further corrected for
non-circular cross section and the fluid flow tortuous paths
[1,4,710], or by representing the porous medium as a periodic
arrays of mono-disperse spheres [11,12]. The preset constant in
the permeability models varies stipulating that there are additional

http://dx.doi.org/10.1016/j.seppur.2016.03.046
1383-5866/ 2016 Elsevier B.V. All rights reserved.
 S. Osterroth et al. / Separation and Purification Technology 165 (2016) 114122
morphological parameters that should be taken into account in
predicting the permeability. A broad variability of the Kozeny con-
stant is summarized in Tien and Ramarao [13], where it is argued
that same porosity functionality is used for porous media which
are morphologically very different, i.e. spherical, cylindrical, fibers
or consolidated media of different types.
Although porous media are mainly heterogeneous, an attempt
can be made to generalize models developed for mono-disperse
media to poly-disperse ones, by exploiting averaged, effective size
scales (diameters), Dm, and Dmn, based on the particle size distribu-
tion function moments, Mm, and ratios of moments, Mmn (see next
section for notations and explanations of the moments). Poly-
disperse models using the three means, harmonic [14], geometric
[15,16], and arithmetic [17], are also reported in the literature.
Models using many varieties of moment ratios are available. Nota-
bly in the filtration literature, all the ranges of average particle
diameters from D21 up to D43 have been used, i.e., D21 in [18],
D31 in [19], D43 in [20], and the Sauter diameter, D32 in
[4,5,14,2123]. Endo and Alonso [19] formulated a model for a
log-normal distribution also including a shape factor for non-
spherical particles and a so-called void function. Their result is
similar to using D31 with log-normally distributed sizes in the
Kozeny-Carman formula. Additionally, in some studies the cut off
diameter of 10% smallest particles [3], and the median particle size
diameter [15] are used to calculate the poly-disperse medium per-
meability. The different permeability models and particle size
averages which give the best agreement to the experimental data
are used in these comparisons. It can be seen that there is a large
variability in average particle size used, implying that in poly-
disperse media, besides the permeability model constant, the par-
ticle size distribution affects the permeability value.
In this study, poly-disperse filter cake permeability is measured
experimentally and compared to the predictions of mono-disperse
permeability models. In the models, the particle diameter is
replaced by an average, effective particle diameter found from
the particle size distribution function. For a broad particle size dis-
tribution, calculated average particle diameters can vary as much
as two orders of magnitude, which produce even higher differences
in calculated permeability as it is a quadratic function of the
selected particle size average. Due to the fact that some of the
smallest particles pass through the filter, the particle size distribu-
tion function of suspended particles differs to some extent from
the distribution function in the filtration cake; the same applies
for the averaged diameters. Hence, the goal of this investigation
is twofold: (i) to perform a detailed experimental study of the
influence of the history of the cake formation on its permeability
(e.g., cake formed on top of various porous filter media, cake
formed from higher and lower suspension concentration, etc.),
and (ii) to use a detailed set of experimental data for poly-
disperse media to determine a proper averaging rule (a moment)
which, for the available permeability models, provides the best
fit to the experimental results.
2. Permeability model
The slow flow through a porous medium is described by Darcys
law, where the permeability quantifies how easily fluid can flow
through a porous medium. The permeability is an intrinsic material
property of the porous medium depending only on the medium
geometry [23] including porosity, tortuosity and particles sizes
and their distribution functions [2426], and not on the nature of
the fluid. In predicting the permeability, probably the most used
empirical model is the Kozeny-Carman formula. It is based on a
combination of Hagen-Poiseuille and Darcy law for steady, laminar
115
and incompressible flow through a bundle of circular capillary
tubes [17]. The famous formulation is given as [6]:
/3 2
K dp 1
kK 1  /2
in which / is the porosity, dp is the particle diameter and kK is the
Kozeny constant, an empirical constant which is often reported hav-
ing different values greatly underlining the uniqueness of Eq. (1)
and suggesting that the medium morphology should be taken into
account in the permeability predictions [13]. This becomes clearer
once looking onto an alternate form of Eq. (1) which is given by
Panda and Lake [17]:
/3
K 2
2s1  /2 S2V
in which s is the tortuosity, and SV is the specific surface area equal
to the ratio of the particle surface to the particle volume, i.e. equal
to SV = 6/dp for a spherical particle of diameter dp. Eq. (2) has been
modified in many different ways, one being by setting different val-
ues for tortuosity [17], or by introducing additional parameters as
threshold porosity to account for close packing when void spaces
start losing connectivity. Another permeability model for spherical
particles has been formulated by Happel [27], where:
2
dp 3  4:5u1=3 4:5u5=3  3u2
K 3
18u 3 2u5=3
with u = 1/ being the fraction of solid phase.
In Eq. (2), the specific area, SV, is defined for the mono-disperse
spherical particles which is directly related to the particle diame-
ter. Similarly, the specific area can be defined for the poly-
disperse sample, where from the particle number distribution
function, the distribution moment of order m and m is an integer,
Mm, is calculated from known weights, wi, and particle diameters,
dp,i. The weight of each particle size is calculated from the number
of particles, ni, and total number of particles in the sample, N:
X m ni
Mm wi dp;i ; where wi 4
i
N
(i.e. for m = 2, M2 is arithmetic average of particles surface area).
The ratio of two moments is defined as Mmn = Mm/Mn, and from the
moments, the average particle diameter is calculated from
Dm = Mm1/m and Dmn = Mmn1/(mn). The specific area (expressed per
unit volume) can be calculated from the distribution moments as
follows:
P 2
A i wi dp;i 1
SV 6P 3
6 5
V w d
i i p;i
D 32
So SV is proportional to the reciprocal of the Sauter diameter and
can be used directly in the Kozeny equation, Eq. (2). This general-
ization shows at least two limitations, the first being the value of
the tortuosity, which in the poly-disperse sample, may be changed
by small particles nesting between larger ones. The second limita-
tion is caused due to the flow dissipation in heterogeneous med-
ium producing an effective medium which is different from one
that consists of particles of equal sizes. Thus, other average particle
diameters can fit better in the Kozeny-Carman formula for the per-
meability of poly-disperse media. A similar reasoning can be
applied to the Happel permeability model, as well as two addi-
tional empirical models for poly-disperse media, proposed by
Rumpf and Gupte [28] and Garcia et al. [29] using the Sauter and
harmonic mean diameters respectively:
/5:5 2 2
K RG D and K G 0:11/5:6 dh : 6
5:6 32
116 S. Osterroth et al. / Separation and Purification Technology 165 (2016) 114122

Clearly a consensus on the most appropriate average particle

diameter has not yet been achieved.

3. Experimental method

The poly-disperse particle size distribution function of the par-

ticles composing the filtration cake and the cake permeability are
determined experimentally in a round filtration flow cell as shown
in Fig. 1 with an active area of Ac = 0.0396 m2. In the test, particles
are deposited at constant flow rate of a suspension onto a filter
material, and the time changes of the pressure difference, Dp,
across the filtering material are recorded. Initially, Dp increases
due to the particle in-depth entrapment in the medium, and later,
the filtration cake starts to form. The number of particles of differ-
ent sizes in the suspension is counted using two laser counters
mounted up and downstream of the filter medium. In the counter,
the light of specific wave length is shed onto suspension and the
light shadow caused by sufficiently large passing particle is identi-
fied at the light detector. The laser counters used do not detect par-
ticles below one micron in size. In one minute intervals, the
cumulative number of each of sixteen different preset particle sizes
are measured throughout the experiment. From this data, the
change in particle size distribution in upstream and downstream
suspension, and the beta ratio are calculated for each time interval.
The beta ratio for particles of size, dp, is defined as the ratio of the
number of particles upstream to the number of particles of that
size downstream, b = nup/ndo.
A suspension of PTI ISO 12103-1, A2 Fine Test Dust (Powder
Technology Inc.) spherical particles is used to form a filtration cake
over three different filtration media. The particle aggregation in the
suspension is prevented by preparing the suspension in a three-
step process, where initially, high-concentration suspension is
homogenized, then mixed using a high-shear dispersing mixer at
elevated temperature, and after dissolving to the final concentra-
tion, the suspension is mixed at lower speed for a very long time Fig. 2. Particle size distribution function of fine test dust used in filtration: (a)
to stabilize. The number and volume particle size distribution number, and (b) volume distribution.
functions are shown in Fig. 2. In terms of number, a large fraction
of the particles are of submicron size even though these particles
account for a small fraction of the total mass (volume). The filtra-
tion media are chosen to differ in their ability to capture smaller
particles, while all three materials essentially block completely
the larger particles. The filter media are a few micron grades,
fibrous porous materials with infinitely long fibers randomly ori-
ented in-plane direction building the medium of final thickness.
All three media are highly porous (/  0.8), and permeable
(Km = 14  1012 m2). In addition to the barrier properties of the
filtration material, the particle concentration in the suspension
and its flow rate are varied in the experiments. The inlet concentra-
tion is varied over nearly an order of magnitude from 1.8 g/m3 to
15.6 g/m3 and flow rate is set from 2 L/min to 6 L/min (superficial
velocity is in order of 1 mm/s). For the water based suspension, the
suspension flow rate is set such that the particle Reynolds number
based on the particle Sauter diameter is small, Rep < 0.01. In all
cases, the particle mass flow rate is in a range that allows each
experiment to be completed in no less than one half hour and no
more than five hours. Each experiment is stopped once the pres-
sure difference across the medium and filtration cake reaches
Dp  5.6 atm.
In the experiments, the pressure difference across the medium
and the particle beta ratios are measured. A typical pressure curve
is shown in Fig. 3(a), where the pressure difference is initially low
due to the low flow resistance through the medium. As the flow
passages inside the filter get blocked by particles, the pressure dif-
ference across the medium increases. After a transition period, a
Fig. 1. Filtration experimental set up used to measure the pressure difference and cake starts to build up on the surface of the medium. Besides the
particle composition in the filter cake. compressibility of the cake, for poly-disperse particles and very
 S. Osterroth et al. / Separation and Purification Technology 165 (2016) 114122
Fig. 3. Experimental results for changing the flow resistance and particle retention
over the porous medium as a function of time: (a) pressure difference, and (b) beta
ratios.
thin cake, the cake composition - particle size wise - can vary
through the cake depth, as the entrapment of small particles by
cake increases gradually as cake grows in depth. Once the cake
thickness is large enough to entrap the smallest particles (or
majority of them), the cake of uniform composition continues to
grow for non-compressible materials. As the flow rate is constant
and for non-compressible cake thick enough to capture small par-
ticles, the pressure difference is a linear function of time. It turns
out that we were able to capture this linear region in the experi-
ments as shown in inset in Fig. 3(a), where the correlation coeffi-
cient between pressure difference and time exceeds 0.99 for
Dp > 1.4 atm.
The beta ratio number of upstream to the number of down-
stream particles for each particle size is nearly constant for a
clean filtering material, and increases once the flow paths inside
the medium get more restrictive and the cake starts forming as
shown in Fig. 3(b). After an initial increase, the beta ratio decreases
for a short period of time, reaching a local minimum, and increase
steadily in time thereafter. Once the cake is thick enough, it
entraps small particles and the beta ratios are large, around 10 as
can be seen in Fig. 3(b). Further entrapment of small particles
and increase in beta ratios produces small changes in porosity,
i.e. around 1.5%. This change in porosity is caused by small parti-
cles, where they account for around 15% of overall volume of par-
ticles. At the beginning of the cake growing stage, already 90% of
small particles is captured (b = 10), leaving only additional 10% to
be captured in the cake growing phase (thus around 1.5% change
117
in porosity). The contribution of small particles to the resistance
of flow paths through the cake is twofold, as at least a fraction of
particles settles at the crevices of larger particles, thus having
fewer particles directly blocking the flow paths and reducing the
cake permeability. It is not clear which mechanism is prevailing,
but from the linear pressure profile during the cake growth stage,
no significant changes of the permeability are observed due to the
additional capture of small particles. Similarly, from the linear
pressure profile even though the beta ratios change during the cake
growth stage (times from 70 to 90 min in Fig. 3), the average beta
ratios can be used. For b = 10 and increasing, changes in effective-
ness, 11/b, are small. Knowing the beta ratios and the particle size
distribution in the upstream suspension, f(dp), the particle size dis-
tribution in the filtration cake, fc(dp), is calculated from:
 
1
f c dp 1  f dp 7
bav dp
where bav(dp) is the time average of beta ratio for particle of size dp.
In the laser particle counting method, only the particles
1 lm < dp < 20 lm are counted. The number of particles with
diameter dp larger than a specific, ds,i cut off diameter (e.g.
12 lm) is counted and equal to n12(dp > 12 lm). Then, new (e.g.
10 lm) cut off diameter is set, ds,j, and number of particles larger
than ds,j is equal to n10(dp > 10 lm). For one interval (e.g. from 10
to 12 lm), the number of particles, n1012(10 lm < dp < 12 lm), is
equal to, n10(dp > 10 lm)-n12(dp > 12 lm). The range
1 lm < dp < 20 lm is divided into sixteen intervals (e.g. from
1 lm to 2 lm, n12, from 2 lm to 3 lm, n23, . . .) and number of
particles for all intervals are counted by laser for one minute. For
each interval, k, containing number of particles, nk, the particles
of different sizes in the range (ds,i, ds,j) are replaced by the same
number of spherical particles of same size which have the interval
particle diameter, dp,k, (e.g. from 10 to 12 lm dp,k can be equal to
10.93 lm) calculated from the following expression:
dp;k dp;i adp;j  dp;i 8
and a is an independent variable between 0 and 1 that needs to be
determined. Assuming a is known, the mass from the laser particle
count, mp,l, is calculated for the number of particles measured
upstream of the porous medium as:
pX
16
mp;l qp
3
nk dp;k 9
6 k1
In Eq. (9), only the particles larger than 1 lm and smaller than
20 lm are accounted for. In order to obtain the mass of all particles
in suspension flowing through laser measuring chamber, the laser
particle count mass, mp,l, needs to be corrected for this uncounted
mass of small and large particles. From the volume distribution
curve in Fig. 2(b), the fractions of small, dp < 1 lm, and large,
dp > 20 lm, particles are equal to 12% and 20%, respectively. The
mass from the laser particle count, mp,l, is measured for the sam-
pling time, ts, and for the flow rate through the laser measuring
chamber, qlas, which means that mp,l needs to be multiplied by
(t/ts)  (q/qlas) obtaining the particle mass, mp.
The particle mass, mp, is calculated for the total duration of the
experiment assuming that the value of variable a in Eq. (8) is
known. The value of a can be found by comparing the particle
mass, mp, to the gravimetric mass, mg, which is measured from
upstream concentration and particle suspension flow rate, cin and
q, such that for an experiment duration time, t, the total mass of
the particles in the upstream suspension is equal to mg = qcint. As
shown in Fig. 4 in which the ratio mp/mg for all investigated cases
is plotted, selecting a = 0.5 reduces the variation between gravi-
metric mass, mg - calculated from nominal flow rate and particle
concentration, and particle mass, mp - calculated from the laser
118 S. Osterroth et al. / Separation and Purification Technology 165 (2016) 114122
Fig. 4. Comparison between the gravimetric and particle mass count methods for
constant a = 0.5 used in calculating the particle mass.
particle count to within 15%. For eleven out of seventeen experi-
mental points, the agreement is within 10%. Finally, it should be
noted that the value a = 0.5 is used for all calculations of the parti-
cle mass, mp, in all seventeen different experiments with each
experiment having sixteen measuring particle intervals.
4. Results and discussion
The permeability of the poly-disperse porous medium is mea-
sured experimentally and predicted analytically. In order to inves-
tigate the influence of the particle packing, three fibrous filtering
media with different inter-fibril distance are selected. One of the
selected filter materials had a porosity that varied monotonically
through the thickness, with more coarse side toward the upstream
side. Three materials should produce different entrapment (size
wise) of the particles within the medium thickness and its surface
which could lead to cake forming of different particle populations.
Assuming particle concentration might affect particle packing in
the cake, suspensions with different particle concentration are
used. Different suspension flow rates are also tested to determine
whether the residence time of the particles within the cake could
produce different particle size distributions within it. In total, sev-
enteen different cases are investigated, and for each case, the pres-
sure and beta ratios as in Fig. 3 are measured. The permeability and
characteristic particle diameters are then calculated from these
experimental data.
Under constant flow rate condition, the permeability can be
determined by measuring how the pressure difference across the
medium and cake layer changes with time. After a certain time, a
cake starts to build up on the surface of the filter medium, and
for non-deformable and cake which captures a majority of small
particles from the test suspension, the pressure difference is a lin-
ear function of time. From the linear data as shown in Fig. 3(a), the
slope Dt(Dp)/Dt is computed from a linear regression line (Dt refers
to the pressure difference in time and Dp is pressure difference
across medium). Using this value and casting it into the Darcys
law, the permeability is calculated from
Dt
K luc u 10
Dt Dp
where uc denotes the cake growth velocity and is related to the
superficial suspension velocity, u, through the particle mass concen-
tration at the inlet, cin, density of the particles, qs, and the cake
porosity, /. Thus, the growth rate is obtained as:
cin u
uc 11
qs 1  /
The experimental determination of cake porosity was carried
out with a dry dust test in which a known volume is filled with
the test dust and the total mass is measured. From this mass and
the known density of each particle, the volume fraction of particles
and the porosity are determined. The porosity of the dry dust is
found to be / = 0.56.
In order to calculate the permeability from Eqs. (10) and (11),
the time gradient of pressure, suspension particle concentration
and cake porosity need to be known. The pressure gradient with
respect to time is calculated from the results given in inset in
Fig. 3(a). As can be seen from the particle size distribution function
in Fig. 2(b) and beta ratios in Fig. 3(b), the particle concentration,
cin, in upstream suspension and cake can differ. On the other hand,
some fraction of small particles is held at thicker cake. Hence, in
the initial considerations, the concentration of particles in the cake
is considered similar to the upstream concentration. The most
uncertain quantity is the cake porosity as it is measured for dry
dust with only shaking the sample to pack the particles more den-
sely. In the experiments, the dust is suspended and there may be
some influence of the exerted pressure on the particle packing in
the sample.
Since there is an uncertainty for the cake porosity in Eq. (11),
the permeability experimental values are calculated for a range
of porosities. The initial value is set equal to dry dust porosity, /
= 0.56, which can be seen as an upper bound obtained for very
low compression pressure. In the experiments, the cake is formed
up to 5.4 atm and to allow for possibly higher packing of sus-
pended particles under pressure during cake formation, lower
porosities of / = 0.50 and / = 0.44, are also used in the permeabil-
ity predictions. Figure 5 shows the permeability results for three
different cake porosities. Each of seventeen different measure-
ments are shown, where different symbols are used to distinguish
between three filter materials and flow cases with low and high
particle concentration. The average permeability is calculated
(solid line), and the permeability interval 30% is shown by dashed
lines. From Eqs. (10) and (11), it should be noted here that the
porosity affects the permeability value for a factor 1/(1/), but
not its variation. All permeability values show a low variability
except in a few experiments, where moderate to significant devia-
tion is found, probably due to an error during the experimental
procedure. The results are biased toward higher values due to
two high experimental values. Still, grouping experiments by filter
media type or particle concentration shows that the final packing
of the particles in cake is similar as implied from the comparable
permeability in all experiment. This implies that once the cake is
formed, it is a prevailing barrier for smaller particles to pass
through.
4.1. Permeability predictions
Having the permeability of the layer of poly-disperse particles
measured experimentally, the applicability of permeability expres-
sions in Eqs. (1), (3) and (6) can be addressed. All four permeability
expressions are function of model constant, porosity and particle
effective diameter. It has been already explained that there is an
uncertainty in the porosity value, and therefore, three distinct
porosities as in experimental measurements are used in predicting
the permeability. If in the experiments, the porosity is measured
accurately, and for mono-disperse media, the Kozeny constant
can be determined as particle diameter is known (same applies
for preset constant in other three models). For the poly-disperse
media, there are two unknowns in the permeability model: model
constant and effective particle diameter. Their lumped value, i.e.
dp2/kK in Kozeny-Carman formula, can be found from known poros-
ity and permeability, but to determine the effective particle diam-
eter, the Kozeny constant has to be known. We set the value of the
 S. Osterroth et al. / Separation and Purification Technology 165 (2016) 114122
Fig. 5. Experimentally measured cake permeability for three different estimated
values of the cake porosity and its variations: (a) / = 0.56, (b) / = 0.50, and
(c) / = 0.44.
Kozeny constant equal to kK = 180 as this value is often cited [13]
for near spherical, unconsolidated granular media and is in an
excellent agreement with the Happel model.
The role of the particle distribution in predicting the permeabil-
ity varies, depending on the effective particle diameter. If perme-
ability is a function of the lower moments of the particle size
distribution, the particle sizes with high population influence the
119
low moments most. For such cake, porosity and permeability are
weak functions of the amount of small particles passing through
the cake, while the effective particle diameter is a strong function
of fraction of the particles passed. For the particle size distribution
of the dust used in our experiments, the sub-micron particles are
predominant and their correct count would be important in calcu-
lating the average particle diameter to be used in the permeability
expression. Alternatively, if the permeability depends on the
higher moments, the contribution of the smaller particles dimin-
ishes as the higher moments are influenced mainly by larger
particles.
In calculating the permeability, it is assumed that all particles in
the suspension are captured in the cake and the original test dust
particle size distribution function (see Fig. 2) is used to calculate
the particle diameter characteristic values. The calculated diame-
ters from the distribution function in Fig. 2(a) are listed in Table 1
(see later in the text), where in addition to the average diameters,
the cut off diameters of smallest 10% and 50% particles, d10 and d50,
as well as harmonic and geometric means, dh and dg, are included
(note that D10 is arithmetic mean diameter). The choice of charac-
teristic diameter is significant. For example, although D32 is
approximately 15 times dh, the predicted permeabilities differ by
more than two orders in magnitude depending on which of two
diameter is used. The permeabilities calculated from the Kozeny-
Carman formula and the other three models are plotted in Fig. 6
for three different porosities, where the characteristic diameters
are given in increasing order on the horizontal axis. The permeabil-
ity mean and its 30% interval are also depicted in Fig. 6 with solid
and dashed lines, respectively. It can be seen from Fig. 6 how the
predicted cake permeability changes with the selected diameter
(note that dh and d50 are very close for our distribution function).
In the models, the best agreement to the permeability are given
by (i) dh and d50 for / = 0.56, (ii) dh, d50, dg and D10 for / = 0.50
and (iii) dg, D10 and D20 for / = 0.44, where the comparison reveals
effective diameter shift toward higher distribution moments as
porosity decreases.
4.2. Particle size distribution
The particle diameters used in Fig. 6 are calculated based on the
original distribution curves in Fig. 2, which implies that all parti-
cles are entrapped in the cake. The analysis in Fig. 6 also shows that
in order to predict the permeability, lower average particle diame-
ters need to be used. Thus, the distribution function for particles in
the cake needs to be corrected [see Eq. (7)], as experimental results
in Fig. 3(b) show that a fraction of the smaller particles pass
through the filter. Due to the passing small particles, (i) average
diameters of particles increase, (ii) effective inlet concentration
for determining cake growth rate by Eq. (11) is smaller than the
particle concentration on the inlet, cin, and (iii) porosity may differ
from one estimated by dry testing. All three parameters have a lim-
ited influence on the experimental permeability once cake is thick
enough and hold small particles as argued earlier. However, the
particles which pass through the filter cause the average diameter
to increase and shift the characteristic average particle size used in
models predicting the cake permeability towards even lower
moments.
In order to determine the fraction of particles of each size form-
ing the cake from Eq. (7), the beta ratios, b, or filter efficiencies,
11/b, have to be known. The beta ratio of particles between 1 lm
and 20 lm are measured experimentally, and it can be assumed
that particles larger than 20 lm are completely captured at the
filter. To set the beta ratios for submicron particles, which are
not measured experimentally, bounds are estimated. For a lower
bound, we consider that the beta ratio linearly decreases with
reducing particle diameter as smaller particles pass more easily
120 S. Osterroth et al. / Separation and Purification Technology 165 (2016) 114122

Table 1
Characteristic particle diameters, in lm, for test dust used in the experiments with the particle size distribution function equal the original distribution and modified distributions
linear approximation (Pop-1) and constant approximation (Pop-2).

d10 dh d50 dg D10 D20 D21 D30 D31 D32 D43

Orig. 0.097 0.179 0.183 0.211 0.267 0.389 0.567 0.703 1.141 2.296 11.36
Pop-1 0.125 0.256 0.288 0.326 0.445 0.665 0.995 1.140 1.817 3.320 12.74
Pop-2 0.097 0.179 0.183 0.211 0.268 0.394 0.580 0.730 1.207 2.512 12.25

Fig. 7. Efficiency curves for the two submicron bounds and corresponding cake
particle size distribution functions: (a) efficiency, and (b) cake distribution function.

Fig. 6. Theoretically computed permeability values with the mean and corre-
sponding 30% range of variation from the experimental results and influence of
porosity on characteristic particle diameter for: (a) / = 0.56, (b) / = 0.50, and (c) /
= 0.44.
through the cake. The upper bound is estimated as the beta ratio
for all submicron particles is equal to the 1 lm particle beta ratio.
Thus, we assume that the beta ratio for submicron particles either
(a) decreases linearly with particle size, the linear approximation,
or (b) remains constant for all submicron particles, the constant
approximation. The filter efficiency defined as a fraction of
upstream particles kept onto the filter as a filter cake, 11/b, is
more intuitive and filter efficiency calculated from both approxi-
mations for beta ratios is depicted in Fig. 7(a).
Adopting these two assumptions, the new distribution function
for particle sizes building the cake can be calculated as shown in
Fig. 7(b). Compared to the original distribution, the constant
approximation is very similar, while for the linear approximation,
the particle number distribution deviates significantly. From the
new distribution functions of the particles building the cake, parti-
cle size distribution moments are found and characteristic particle
diameters are tabulated in Table 1. As expected, the diameter
 S. Osterroth et al. / Separation and Purification Technology 165 (2016) 114122
Fig. 8. Permeability comparison and corresponding characteristic particle size
diameter for the linear approximation beta ratio decreases linearly for submicron
particles for: (a) / = 0.56, (b) / = 0.50, and (c) / = 0.44.
averages for the linear approximation increase considerably
compared to their values calculated from the original particle size
distribution as small particles pass through the filtering material.
For the constant approximation, distributions remain similar as
the majority of small particles are captured on the cake.
The shift of particle average diameters toward larger values in
the two approximations requires that, in order to predict the per-
meability correctly from the analytical models, even lower
moments have to be used for new distributions (some particles
pass through) compared to the original distribution (all particles
121
are held on the filter). Furthermore, the cake growth velocity deter-
mined by Eq. (11) is reduced because the particle concentration is
smaller than cin. Since the suspension velocity and pressure gradi-
ent in Eq. (10) remain the same (experiments are carried out for
preset flow rate and pressure gradient is measured experimen-
tally), the permeability determined from the experimental data is
smaller once corrected for the passing particles. This change in per-
meability should be small as both cake porosity and its growing
velocity are influenced by larger particles, but still might cause that
even lower moments need to be used in predicting the permeabil-
ity analytically.
Figures 8 and 9 summarize the permeability results for the two
approximations with effective particle size diameters calculated
from new distribution curves (see Fig. 7) and the experimental per-
meabilities are recalculated accounting for the changes in the cake
growth velocity due to reduced particle concentration held by the
filter. The experimental permeability remains similar for two
approximations [e.g. values in Fig. 8(a) versus Fig. 9(a)]. For a
porosity range from / = 0.44 to / = 0.56, it can be seen how the
effective average diameter used to calculate permeability increases
as porosity decreases. Thus, (i) for high porosities / = 0.56 and
/ = 0.50, the value of d10 and dh are lower and upper bounds, while
(ii) for / = 0.44, it appears that dh and d50 are proper measures. As
already stated, the linear approximation of beta ratio is highly
stringent for the filter efficiency for the submicron particles. If this
assumption is relaxed as in the constant approximation of beta
ratio, the effective particle size which should be used in calculating
the permeability increases. As in the linear approximation, average
diameter increases as porosity decreases, where it goes from as low
as dh for high porosity to dg and D10 and even D20 for low / = 0.44. It
should be mentioned that the Sauter diameter, D32, is too large for
calculating the permeability of our poly-disperse sample. For test
dust used in this experiment, D32 differs more than an order in
magnitude compared to the lower moments used in this analysis,
and hence, would result in a discrepancy of about two orders of
magnitude in the permeability prediction. Finally, this discrepancy
would be much smaller for a narrower particle size distribution
function.
5. Conclusions
The permeability of the poly-disperse porous medium mea-
sured experimentally and predicted from the analytical models
are compared. It has been shown that the type of filtering material,
particle suspension concentration and suspension flow rate do not
affect the permeability of the cake. On the other hand, cake poros-
ity and particles size distribution function in the cake have a detri-
mental role and the uncertainties in predicting those two
parameters can lead to larger errors in the permeability predic-
tions. This is observed in the literature, where the cake permeabil-
ity is modeled with a broad range of effective particle diameters. In
the present work, the lack of beta ratios for sub-micron particles
required consideration of two limiting cases to estimate particles
average diameter in the filter cake. Although the data does not
establish a clear choice, the results indicate that the harmonic
mean is a reasonable approximation for the used sample. The liter-
ature results [5] suggest that the finer particles are most important
for the characterization of filtration, which is true if lower distribu-
tion moments, e.g. harmonic, are used in predicting the permeabil-
ity. The often cited Sauter diameter overestimates the permeability
values measured in our experiments. This might be caused by the
fact that the particle size distribution of the test dust used in the
experiments is very broad, while the use of the Sauter diameter
is only recommended for narrow distributions [23]. Finally, for
porosity around 50%, the predictions of all four analytical models
122 S. Osterroth et al. / Separation and Purification Technology 165 (2016) 114122
Fig. 9. Permeability comparison and corresponding characteristic particle size
diameter for the constant approximation beta ratio is constant for submicron
particles for: (a) / = 0.56, (b) / = 0.50, and (c) / = 0.44.
are close and the choice of a suitable average diameter is more
important than the choice of model in predicting the permeability.
Acknowledgements
The authors thank to Ralf Kirsch from ITWM and Bruce Wilson
from Pall Corp for useful suggestions and discussions on the under-
standing of experimental results and their interpretation.
Appendix A. Supplementary material
Supplementary data associated with this article can be found, in
the online version, at http://dx.doi.org/10.1016/j.seppur.2016.03.
046.
References
[1] J. Bear, Dynamics of Fluids in Porous Media, Courier Corporation, 2013.
[2] J. Bear, Hydraulics of Groundwater, Courier Corporation, 2012.
[3] J. Rosas, O. Lopez, T.M. Missimer, K.M. Coulibaly, A.H.A. Dehwah, K. Sesler, L.R.
Lujan, D. Mantilla, Determination of hydraulic conductivity from grain-size
distribution for different depositional environments, Groundwater 52 (2014)
399413, http://dx.doi.org/10.1111/gwat.12078.
[4] J. Olivier, J. Vaxelaire, E. Vorobiev, Modelling of cake filtration: an overview,
Sep. Sci. Technol. 42 (8) (2007) 16671700, http://dx.doi.org/10.1080/
01496390701242186.
[5] R. Wakeman, The influence of particle properties on filtration, Sep. Purif.
Technol. 58 (2) (2007) 234241, http://dx.doi.org/10.1016/j.
seppur.2007.03.018. ISSN 1383-5866.
[6] P.C. Carman, Fluid flow through granular beds, Trans. Inst. Chem. Eng. 15 (1937)
150166. Reprinted in Chemical Engineering Research and Design, 75 (1997)
S32S48(Supplement), http://dx.doi.org/10.1016/s0263-8762(97)80003-2.
[7] A. Bejan, A.D. Kraus, Heat Transfer Handbook, 1, John Wiley & Sons, 2003.
[8] F.A. Dullien, Porous Media: Fluid Transport and Pore Structure, Academic Press,
1991.
[9] A.S. Kim, K.D. Stolzenbach, The permeability of synthetic fractal aggregates
with realistic three-dimensional structure, J. Colloid Interface Sci. 253 (2)
(2002) 315328, http://dx.doi.org/10.1006/jcis.2002.8525. ISSN 0021-9797.
[10] P. Xu, B. Yu, Developing a new form of permeability and Kozeny-Carman
constant for homogeneous porous media by means of fractal geometry, Adv.
Water Resour. 31 (1) (2008) 7481, http://dx.doi.org/10.1016/j.
advwatres.2007.06.003. ISSN 0309-1708.
[11] A.S. Sangani, A. Acrivos, Slow flow through a periodic array of spheres, Int. J.
Multiphase Flow 8 (4) (1982) 343360, http://dx.doi.org/10.1016/0301-9322
(82)90047-7. ISSN 0301-9322.
[12] A.A. Zick, G.M. Homsy, Stokes flow through periodic arrays of spheres, J. Fluid
Mech. 115 (1982) 1326, http://dx.doi.org/10.1017/S0022112082000627.
[13] C. Tien, B.V. Ramarao, Can filter cake porosity be estimated based on the
Kozeny-Carman equation?, Powder Technol 237 (2013) 233240, http://dx.
doi.org/10.1016/j.powtec.2012.09.031. ISSN 0032-5910.
[14] W.C. Yang, Handbook of Fluidization and Fluid-Particle Systems, CRC Press,
2003.
[15] P.J. Kamann, R.W. Ritzi, D.F. Dominic, C.M. Conrad, Porosity and permeability
in sediment mixtures, Ground Water 45 (2007) 429438, http://dx.doi.org/
10.1111/j.1745-6584.2007.00313.x.
[16] L.B. Porter, R.W. Ritzi, L.J. Mastera, D.F. Dominic, B. Ghanbarian-Alavijeh, The
Kozeny-Carman equation with a percolation threshold, Ground Water 51
(2013) 9299, http://dx.doi.org/10.1111/j.1745-6584.2012.00930.x.
[17] M.N. Panda, L.W. Lake, Estimation of single-phase permeability from
parameters of particle-size distribution, AAPG Bull. 78 (7) (1994) 10281039.
[18] M.J. MacDonald, C.F. Chu, P.P. Guilloit, K.M. Ng, A generalized Blake-Kozeny
equation for multisized spherical particles, AIChE J. 37 (1991) 15831588,
http://dx.doi.org/10.1002/aic.690371016.
[19] Y. Endo, M. Alonso, Physical meaning of specific cake resistance and effects of
cake properties in compressible cake filtration, Filtr. Sep. 38 (7) (2001) 4246,
http://dx.doi.org/10.1016/S0015-1882(01)80447-X. ISSN 0015-1882.
[20] A.C.B. Neiva, L. Goldstein Jr., A procedure for calculating pressure drop during
the build-up of dust filter cakes, Chem. Eng. Process. 42 (6) (2003) 495501,
http://dx.doi.org/10.1016/S0255-2701(02)00066-1. ISSN 0255-2701.
[21] M.L. Aguiar, J.R. Coury, Cake formation in fabric filtration of gases, Ind. Eng.
Chem. Res. 35 (10) (1996) 36733679, http://dx.doi.org/10.1021/ie960042p.
[22] Y.H. Cheng, C.J. Tsai, Factors influencing pressure drop through a dust cake
during filtration, Aerosol Sci. Technol. 29 (4) (1998) 315328, http://dx.doi.
org/10.1080/02786829808965572.
[23] D.A. Nield, A. Bejan, Convection in Porous Media, Springer Science & Business
Media, 2006.
[24] Y. Endo, D.-R. Chen, D.Y.H. Pui, Effects of particle polydispersity and shape
factor during dust cake loading on air filters, Powder Technol. 98 (3) (1998)
241249, http://dx.doi.org/10.1016/S0032-5910(98)00063-1. ISSN 0032-5910.
[25] J.A. Sorrentino, From particle collective characteristics to cake permeability:
the use of the pore-particle shape factor, Part. Part. Syst. Charact. 24 (2) (2007)
97100, http://dx.doi.org/10.1002/ppsc.200601047.
[26] M. Stie, Mechanische Verfahrenstechnik 1, Springer-Verlag, 2013.
[27] J. Happel, Viscous flow in multiparticle systems: slow motion of fluids relative
to beds of spherical particles, AIChE J. 4 (1958) 197201, http://dx.doi.org/
10.1002/aic.690040214.
[28] H.C.H. Rumpf, A.R. Gupte, Einflsse der Porositt und Korngrenverteilung im
Widerstandsgesetz der Porenstrmung, Chem. Ing. Tech. 43 (1971) 367375,
http://dx.doi.org/10.1002/cite.330430610.
[29] X. Garcia, L.T. Akanji, M.J. Blunt, S.K. Matthai, J.P. Latham, Numerical study of
the effects of particle shape and polydispersity on permeability, Phys. Rev. E 80
(2) (2009), http://dx.doi.org/10.1103/PhysRevE.80.021304. 021304.