Chemical Engineering Science, 1970, Vol. 25, pp. 985-996. Pergamon Press. Printed in Great Britain.

A theory of constant pressure filtration

D. E. SMILES
Department of Soil Science, The University of Sydney, Sydney 2006, Australia

(Received 2 1 November 1969)

Abstract- Attention is drawn to two equivalent analyses that describe the one dimensional process
of “cake” filtration. These analyses, which arise from conventional theory of flow of fluid in porous
media, permit the prediction of density gradients in the solid material at any time, as it accumulates
on the filter membrane. and mav be used to oredict conditions that should be imposed to achieve
greatest efficiency in the’filtrationprocess.

INTRODUCTION ceptually difficult presentation and appears to be

CAKE FILTRATION is a common procedure in
laboratory, as well as industrial practice. In this
process the liquid component of a two compo-
nent (solid-liquid) system is separated from the
solid using a membrane that permits the passage
of liquid but not solid.
The flow may occur as a consequence of
gravity, but more commonly it is increased by
the imposition of a pressure difference across the
membrane.
Analyses that predict the flow of the liquid in
such systems have been based on Darcy’s Law
and reviews are presented by Scheidegger[lO]
and Dickey[3]. At the time of these reviews it
was conventional to consider the filtration pro-
cess to be subject to a steady state analysis not
materially different from that of Sperry [ 16,171.
Tiller and Cooper [ 181 appear to be the tirst to
recognise the nature of the process as a transient
flow situation properly described by a non-linear
partial differential equation. Subsequently
Tiller and Cooper[l9] and Shirato[l9] and
Shirato ef al.[ 1 l] manipulate such an equation
to predict the behaviour of various filtration
situations.
More recently, Shirato et a/.[121 have pre-
sented an analysis that recognises in principle
the importance of the movement of the solid
phase during cake filtration. Since it is based on
the analysis of Tiller and Shirato[21], however,
it has in common with the earlier work a con-
based on the misconception that Darcy’s Law
expressed in terms of their equation (1) describes
flow of Iiquid relative to fixed spatial coordinates
rather than to the solid particles.
It is the purpose of this paper to draw atten-
tion to two alternative non-steady state analyses
that are appropriate to the process of cake filtra-
tion and to identify the initial and boundary
conditions relevant to the process. The arith-
metical manipulation of one of these analyses
will be demonstrated.
It is convenient, in the first instance, to con-
sider the case where the transmission coeffi-
cientt of the membrane is great compared with
that of the accumulating cake, and to assume that
effects due to gravity are small compared to those
of the pressures imposed.
EQUATIONS OF FLOW IN CONSOLIDATING,
FILTERING SYSTEMS
In order to avoid confusion with the existing
terminology of filtration theory, an alternative
functional symbolism will be developed in this
paper. This symbolism is consistent with that of
soil physics usage [5].
The basic flow equation is Darcy’s Law:
v=- Kawax (1)
tTransmission volumetric flow rate per unit area
coefficient = pressuredifference across the membrane.

985
 D. E. SMILES
in which v (cm/set) is the volumetric flux density
of liquid, $X$3x (cm/cm) is the space gradient of
total potential? of the liquid, and K (cm/set) is
the liquid conductivity of the porous material
through which flow occurs. It is crucial to the
development of the theory that it be recognised
that Eq. (1) describes the flow of water relative
to the particles of solid[8,22].
Where flow is horizontal, or where the imposed
pressure gradient is much greater than the
gravitational potential gradient, az/ax, Eq. (1)
may be written
v = - KaWax (2)
where T (cm) is the liquid potential*. In general
K is a function of q.
Two alternative approaches are now possible
when one invokes the continuity requirement to
derive a general flow equation: one,[8], con-
siders the simultaneous movement of solid and
liquid and culminates in a non-linear partial
integro-differential equation with conventional
space coordinates. Philip[8] shows how this
equation may be manipulated to calculate fluxes
of liquid and solid in space as well as movement
of liquid relative to solid.
Alternatively, an analysis, exactly equivalent
to that of Philip but based on material coordinates,
may be used. This analysis,[ 151, gives rise to a
general flow equation in the form of a diffusion
equation
in which e is the void ratio, defined by
volume of voids
- t is time (seconds); D,
e = volume of solids’
tThe Total Potential, @, of the liquid will be defined here
as the work per unit weight of liquid necessary to remove an
infinitesimal amount from the system to a reference pool of
the liquid at some defined position in the gravitational field.
The units of this uotential are cm of liquid. A full discussion
of the potential of-water in swelling porous media is presented
by Philip [9].
SThe component of potential that arises as a result of
interaction of liquid and solid[9].
986
is the di@usivity defined by
D_=&z=K,$ (4)
and the material coordinate, m, is related to the
space coordinate, x, by the relation
g= (l+e)-‘.
In m-space, unit distance encompasses unit
volume of solid per unit area of cross section.
It is interesting to note that Gibson et af.[4]
introduce, though they fail to use, such a coordi-
nate system in describing one-dimensional
consolidation of clays. Shirato et a/.[121 also
effectively define such a coordinate in their
equation (6).
THE PROCESS OF CAKE FILTRATION
The system to be considered is shown in
Fig. 1.
In this system, a two component suspension
with uniform void ratio, e,, is at some time,
t = 0, subjected to a constant pressure pi (cm
of liquid) in excess of atmospheric pressure.
Liquid then moves from the suspension through
the membrane to the vessel beneath where the
total pressure is p,, in excess of atmospheric
pressure. The solute accumulates on the mem-
brane.
It will be assumed
(i) that the void ratio, e, of the solute is a
known function of q.
(ii) that throughout the system the total stress
pi = p1 - V where pE is the liquid (hydrostatic)
pressure in the system. (This is consistent with
the civil engineering theory and soil science
theory for saturated systems [2, 131 and has been
conventionally used in filtration theory [ 181. It is
an immediate consequence of this assumption,
that at the upper surface of the filter membrane
where x = 0
90 = (PO+@-Pi
because there
Pl =po+a.
 A theory of constant pressure filtration

E pi - the imposed
pmssure

Two component
suspenaon (solid ‘liquid)
with an initially uniform
void mtio. en

Membrane that permits

the passage of liquid
but not solld
p. is the total pressure In

II the liquid beneath the membrane

Fig. 1. Idealizeddiagramof a filtrationcell.

Where 6 is the pressure increment consistent D,(e) when e(m, I) is known, is reported by
with the flux of liquid through the membrane. Smiles and Rosenthal [ 151. Alternatively, an
If now it is assumed that the transmission iterative procedure, [7], may be used to com-
coefficient of the membrane is great compared pute e(m, t) when D,(e) is known.
with that of the accumulating cake then 6 will For either procedure, the substitution A =
tend to be much less than (p,,-pi) and as a mt-1’2 is introduced, and Eqs. (3) and (7) become
consequence of assumption (i) e will tend to a respectively,
constant value e,.
(iii) that during the filtration process there is
(8)
always suspension of void ratio e, standing above
the accumulating cake.
As will be shown later, this requires that the and
rate of supply of solution to be filtered is equal e=e,;A+m
to or greater than Xt-1/2 defined by Eq. (20). (9)
Under these circumstances it is convenient e = eo; A = 0.
to define x positive upwards, whence, from
Eq. (5) Integrating (8) and noting that (9) implies that
deldh = 0 when e = e, one obtains
m = I f(l +e)-‘dx (6)
D k-t
“dA
and the relevant initial and boundary conditions
are, respectively, Equation (10) is the basis for all manipulations
of (3) when subject to (7).
e = e,; m > 0; t = 0
(7) Flux of/hid
e = e,; m = 0; t > 0. It follows from Eqs. (2) (4) and (10) that

The test of Eq. (3), and its manipulation

subject to conditions (7), for the calculation of A de
(1 la)
987
 D. E. SMILES

and that and I* = xt-1/2.

tr, - t-“2 =O A de THE D,,,(e) RELATION

2 I e. (1 lb)
There is, at present, little general information
where ZJ,,is the value of u at m = 0. Also about the D,(e) relation for compressible mater-
ials although Smiles and Rosenthal[l5] deter-
Qt- l/2 = s e.
“A de (12) mined such a relation for montmorillonite in
equilibrium with a solution containing 0.25 m-
where Q(= _f,$, dt) is the cumulative outflow of equiv/l of (Na+ Ca) with a ratio of Na (JCa)-1’2
solution from the suspension at m = 0. (c.f. [S]). = 1. This relation has a maximum value of D,,,
when e + 20. The maximum is of the order of
Flux of solid 1Omgcm2/sec.
The calculation of the flux of solid is slightly Rather than use this relation (15) it is possible
more involved. It is a requirement of continuity to find D,(e) functions using Eq. (8), observed
in the two phase system that for all x q(e) relations, and K,(e) functions obtained by
the use of the Carman[ l] method.
VC-tVW=U, (13) It is suggested, however, that in practice D,(e)
should be determined experimentally for the
where PC, VW are the absolute? flux densities particular system to be filtered.
of solid and liquid respectively.
VW,however, may be written The form of*(e)
Examination of V(e) relations obtained for a
Vw=eVc+u variety of materials shows that many may be
so approximated over much of the range of e by
I/c= no-2) (14) an equation of the form
l+e’
Which, from Eqs. (1 la and b) becomes e= czlog(-*)+/3 (17)
t-l/2 e
(15) where (Yis the slope of the e (log(- ?)) graph and
Pc=2(1+e) I e0hde* p, the value of e when 9 = - 1[2,6, 11, 15, 181.
In practice, the amount of solid that has ac-
cumulated in some convenient distance from the The K,(e)function
filtering membrane is relevant. The method of This function may be “predicted” using a
calculation of this quantity will be considered number of devices [ 10,8]. The method of Kozeny
during the discussion of the calculated example modified by Carman[l], will be used here,
that follows. whence
In the example that follows a D,(e) function K, = Kge3(1 + e)-2 (18)
will be defined and the method of Philip[7] used
to compute the e(A) curve for several values of where Kz is a matching factor relating a meas-
e, and eO.The e(A) curves together with Eq. (12) ured value of hydraulic conductivity, K,, to a
will be used to calculate Qr112. In addition, measured value of e.
e(p) curves will be calculated using e(A), the
relation$ D,(e)
Combining (4), (17) and (18) it is found that
x= I f(l+e)dm (16)
tin space coordinates. -2.3 e3
0,/K; = - - (y(e--P)) (19)
fCompare with Eq. (6). a (1 + e)2 exp

988
 A theory of constant pressure filtration
D, calculated using (19) exhibits a maximum
value as does that of Smiles and Rosenthal [ 151.
The e value of the maximum depends solely on
(Y,as may be shown by differentiating (19).
Figure 2 shows D,(e)/Kz plotted as a function
of e for
(Y= - 8; p = 40 (Na-illite).
This curve will be used in the calculated
example that follows. It will be assumed that
Kz for this material takes a (reasonable) value
of 30 X lo-l2 cmlsec.
RESULTS
Figure 3 shows e(A) calculated, using an iter-
ative procedure (7) for five different combina-
tions of e, and e,,.
Figure 4 shows e(h) calculated using Fig. 3 and
Eq. (16).
Figure 5 shows Qt-1/2 calculated using Eq.
( 12) and Fig. (3)) plotted as a function of eO
1 I
I
Fig. 2. Diffusivity, D,, plotted as a function of void ratio, e.
holding e, = 15.6 (curve (I)); and of e, holding
e, = O-6 (curve (Ii)). Curve (I) represents the
situation where (pi -pO) is changed but e, is
constant, curve (II) represents the effect of
changing e, when (pi-p,,) is constant. The
apparent similarity of the two curves arises
because of the shape of the D, function; in
particular the somewhat symmetrical form about
e = 5, in this sense it is fortuitous.
It may be concluded from this diagram that
the greater the difference between e, and e,, the
greater will be Qt-‘j2 and if the filtration process
is designed to accumulate the liquid component
then the more satisfactory will be the process.
It is interesting to speculate whether it is
possible to impose conditions under which a
decrease in e, relative to constant e, could lead
to a decrease in Qt-l12 consequent on the ob-
served decrease in D, at low values of e:
consider the equation
ae
;i;-=B$$ 3(a)
I i I I 1
989
 D. E. SMILES

Q 8-

0 2 4 6 8 10 12 14 16 18
mfkx io’km. SQC%)

Fig. 3. Void ratio, e, plotted as a function of mt- 1J2for different combinations of e. and eO.

in which & is constant. The solution of 3(a) whence, using Eq. 12(a)
subject to conditions (7) is known and may be
used to calculate Qc”~ using the equation Qt-*12 = 4(Z/r)‘j2 in which

Qt-‘12 = 2(e, - e,J(&Jn)1’2. (12a) Z = leoDm(e - e,)de.

That is, the D,(e) relation necessary to cal-
culate Qt-1’2 from Eq. (12) may be replaced by a
suitable mean value 6,. This value will permit
the calculation of Qt-1/2 but not e(m, t).
Empirically, Eq. (6.1) of [7] appears to give
rise to a most satisfactory approximation to this
mean value, viz.,
&=2 e”&(e - e,) de
I e. (G - eA2
990
P.
For constant e,, Z is an increasing function of
(e,- eO) since D, is always positive, and there-
fore Qr-lj2 should also be an increasing function
of (en-e,). That is, the curves shown in Fig. 5
have general application.
Figure 6 shows Qt-‘12 plotted as a function
of (Pi-PO) for, the case when e, = 15.6. This
figure was calculated using curve (I) of Fig. 5
and Eq. (17). It is only one of a family of curves
each defined by a different value of e,. Note that
 A theory of constant pressure filtration

16

0
20 40 60 00 100 120 140 160
&2 x lo4 (cm. SOC’/~)

Fig. 4. Void ratio, e, plotted as a function of xt-1’2for different values of e, and e,.

there can be no such curves corresponding to
curve (II) of Fig. 5 because (pi --pO) defines e.
uniquely.
As with Fig. 5, it is again noted that Qt-1’2
is greatest when (pi -po) is greatest.
It should be noted for both Figs. 5 and 6 how-
ever that the conditions when (e, - e,) or (pi -po)
are greatest are not necessarily the most efficient
economically if recovery of liquid is important.
Economic efficiency could be considered only
in terms of the cost of achieving an increment
in Qt-1’2.
Accumulation of solid
By definition, the cumulative volume of solid
in any given interval, x, from the membrane is
m = At-112
991
Thus m can be determined, using Fig. 3, at
any time, t, for any e cz e * where e” is a desired
value of L In addition, we may use this relation
to determine the minimum rate of supply of solu-
tion to the cell to satisfy the requirement that
for x or m great enough e = e,: thus using Fig. 3,
Eq. (5), and choosing e* + e,, the minimum rate
of supply Xt-1’2 may be calculated using
Xt-1’2 = (1 + e,)A(e*). (20)
Alternatively, referring to Fig. 4 m may be
calculated using the relation
m = P2 o(l +e)-’ +.
I
Figure 7 shows mt-1’2 plotted as a function of
xt-1’2 for the curves shown in Fig. 4. This graph
permits the calculation of m for any x and t.
 D. E. SMILES

60-

Q
Fig. 5. QP* plotted as a function of e for e, = 0.6 -curve I; e,, = 15.6 - &rve 11.

0 40 80 120 160 xx) 240 280

(P,- @(Cl-n)

Fig. 6. QP* plotted as a function ofp, -pO for e, = 15.6.

992
 A theory of constant pressure filtration
From Fig. 7 it may be seen that for any x, m is
proportional to P while for any t, m increases
as x increases. In addition, m increases as e,
increases for constant e,; and for constant e,,
m increases as e,, decreases.
In practice m is not a very useful concept for
comparing filtration procedures that aim to
accumulate solid because it includes the solid
component originally present in the length of
column considered. For example, in Fig. 7,
curve III would appear to represent the most
satisfactory conditions for accumulation of solid.
This arises however because this treatment had
the smallest value of e,, and hence the greatest
initial amount of solid. It is therefore more
16
14
0 20
Fig. 7. mt-“2 plotted as a function of xt- “* for the different combinations of
C.Es Vd. 25 NO. 6-F
useful to consider M, the “cumulative increase”
in solid in a particular length. Referring to Fig.
4, we may define M by
Mt-1'2=I :((l+e)-‘-(l+e.)-‘)dCc.
Figure 8 shows Mt-1'2 plotted as a function of
~f-~‘~for the curves shown in Fig. 4.
In terms of M, the most satisfactory filtration
conditions are those appropriate to curve I.
Examining this set of curves further it will be
noted that curves with a common value of e,
and different values of e, (I, II, III) are more
similar than those with common e, and different
values of eO (I, IV, V). This phenomenon arises
40 60 80 100 120 140
xi’+ x 10’ (cm. sot-‘* 1
e. and e, used in Fig. 3.
993
 D. E. SMILES

4%
xt x 10" (cm s&%
Fig. 8. IMP’* plotted as a function of xt- I’*for the different combinations of e,
and e,, used in Fig. 3.

because the flow of liquid throughout the accu-
mulating pad is dominated by the properties of
the pad near x = 0 where e = e, and these pro-
perties are relatively insensitive to change in
e,,. Further discussion of the phenomenon and
its consequences is presented in[ 141.
CONCLUSION
Two. alternative but equivalent analyses of
cake filtration are available. One is based on a
space coordinate system[8], the other[ 151, is
based on material coordinates. Both recognise
that Darcy’s Law describes flow of water
relative to the solid particles and not relative to
fixed space and both take account of mass flow
of the solid component.
Using either of these treatments together with
measurable behavioural properties of the system
994
to be filtered, one can calculate (i) the fluxes of
water and of solid, (ii) the cumulative amounts
of water leaving the system and of solid within
any distance from the filter membrane, and (iii)
the distribution in space and time of water and
solid. Simple manipulation permits a comparison
of effects of different imposed filtration condi-
tions.
It appears from the worked example, based on
Eq. (3) and the D, function shown in Fig. 2,
that the greatest yield of both solid and liquid
occurs for any system when e, = e, is greatest.
Finally, it is worth emphasizing that the
analysis is based on behavioural properties of
the two component systems and that these pro-
perties should be determined empirically. There
is not at present information sufficient to justify
the use in prucfice of relations such as those
 A theory of constant pressure filtration

given by Eqs. (17) or (18), particularly when the originally present in the same interval of
empirical determination of D,(e) presents few X

problems. PI3 atmospheric pressure, cm

Pi imposed gas pressure, cm
Acknowledgments-The author wishes to acknowledge pt liquid pressure, cm
helpful criticism by Professor N. Collis-George of the Q cumulative outflow of water, cm2 cmm2
Department of Soil Science, University of Sydney, and
Dr. J. R. Philip, Division of Plant Industry, C.S.I.R.O., t time, set
and financial assistance from the Australian Research Grants to t at imposition of pi, set
Committee. V volumetric flux density of water relative to
solid, cm set-’
NOTATION no v at x and m = 0, cm set-’
D KcW/& = moisture diffusivity, cm2 set-l V, absolute flux density of solid, cm set-l
& moisture diffusivity in the material coordi- V, absolute flux density of water, cm set-l
nate system, cm2 set-’ X space coordinate, cm
volume of voids Z gravitational component of Q,, cm
e void ratio =
volume of solids
O;t=O tireek symbols
en valueofewhenx>
e. vahreofewhenx=O;t> 0 a the slope of the e vs. log (-‘u) curve
p the value of e in the e vs log (-9) curve
I
Iey
D,(e - e,)de when* =- 1 cm
K hydraulic conductivity defined by Eq. (I), A mt-1’2, cm set-112
cm set-l p xt-1’2, cm set-112
L K/( 1 + e), cm set-l @ total potential of liquid, cm
m material coordinate defined by Eq. (5), @=*+z
cm * liquid potential component of @, cm
M cumulative amount of solid relative to that q. ?atmandx=O

REFERENCES
Ul CARMAN P. C., Trans. lnstn them. Engrs 1938 16 168.
PI CRONEY D. and COLEMAN J. D., Pore Pressure andSuction in Soils. Butterworth 1960.
131 DICKEY G. D., Filtration. Reinhold 1961.
[41 GIBSON R. E., ENGLAND G. L. and HUSSEY M. J. L., Geotechnique 1967 17 261.
[51 Int. Sot. of Soil Science, Bull. No. 22, 1963.
161 MITCHELL J. K., In Interparticle Forces in Clay-Water-Electrolyte Systems. Commonwealth Scientific and Indus-
trial Research Organization, Melbourne 1960.
171 PHILIP J. R., Trans. Faraday Sot. 1955 6 885.
PI PHILIP J. R.,Aust.J. Soil Res. 1968 6 249.
[91 PHILIP J. R.,Aust. .I. Soil Res. 1969 7 99.
[lOI SCHEIDEGGER A. E., The Physics of Flow Through Porous Media. Macmillan 1957.
1111 SHIRATO M., SAMBUICHI M. and OKAMURA S.,A.I.Ch.E.d 1963 9 599.
[I21 SHIRATO M., SAMBUICHI M., HIROO K. and ARAGAKI T.,A.I.Ch.E.J11969 15 405.
SMILES D. E., Aust. J. Soil Res. 1969 7 91.
:::; SMILES D. E. and POULOS H. G.,Aust.J. Soil Res. 1969 7 285.
[15] SMILES D. E. and ROSENTHAL M. J.,Aust.J. Soil. Res. 1968 6 237.
1161 SPERRY D. R.,Metall. them. Eng. 1916 15 198.
1171 SPERRY D. R., Metall. chem.Engng 1917 16 161.
[181 TILLER F. M., InSolid-LiquidSeparation (Edited by POOLE J. B. and DOYLE D.). H.M.S.O., London 1966.
[19] TILLER F. M. andCOOPER H. R.,A.I.Ch.EJll9606 595.
[20] TILLER F. M. and COOPER H.A.I.Ch.E.Jl 19606 445.
[21] TILLER F. M. and SHIRATO M.,A.I.Ch.E.J11964 10 61.
[22] ZASLAVSKY D., Soil Science 1964 98 3 17.
 D. E. SMILES

RhmC- L’attention est centr6e sur deux analyses dquivalentes qui decrivent le pro&de de tiltrage
“cake” unidimensionnel. Ces analyses, qui proviennent de la theorie traditionnelle de I’ecoulement
fluide dans un milieu poreux, permettent la prevision de gradients de densite dans la mat&e solide, a
tout moment, au fur et a mesure qu’elle s’accumule sur la membrane filtrante, et peuvent &re utilisees
pour prevoir les conditions qui devraient &tre imposees en vue d’obtenir une plus grande efficacite du
processus de filtrage.

Zusammenfassung- Es wird auf zwei iiquivalente Analysen zur Beschreibung des eindimensionalen
Vorganges der “Kuchen” Filtration aufmerksam gemacht. Diese Analysen, die sich aus der herkiimm-
lichen Theorie der Fliissigkeitsstromung in pordsen Medien ergeben, gestatten die Voraussage der
jeweiligen Dichtegradienten im Festmaterial im Laufe seiner Ansammlung auf der Filtermembran,
und konnen dazu verwendet werden Bedingungen vorauszusagen, die auferlegt werden miissten urn
griisste Wirksamkeit im Filtrationsprozess zu erzielen.

996