journal of

MEMBRANE
SCIENCE
Journal of Membrane Science 100 ( 1995 ) 259-272

ELSEVIER

Critical flux concept for microfiltration fouling

R.W. Field a,., D. WU a, J.A. Howell a, B.B. Gupta b
a School of Chemical Engineering, University of Bath, Bath BA2 7AY, UK
URA CNRS 858, University of Compi~gne, Compi~gne, France

Received 29 October 1993; accepted in revised form 14 October 1994

Abstract
Several constant-flux filtration experiments for yeast cell suspensions, yeast cell debris, and dodecane-water emulsion were
performed at various operating conditions in both fiat-sheet and tubular-membrane systems. The aim of the paper is two-fold.
Firstly the relationship between constant-flux behaviour and membrane fouling is discussed. In some cases constant-flux filtration
was realized at a constant transmembrane pressure which was below a critical value. In general constant-flux filtration was
obtained with moderately increasing transmembrane pressure, and this approach is shown to have some advantages over normal
constant-pressure filtration because it clearly provides for the possibility of avoiding over-fouling and so reduces the severity of
fouling. Secondly, the concept of critical flux is introduced. Whilst it has long been recognised that low-pressure microfiltration
is much more effective than high-pressure microfiltration, the emphasis in this work is upon the possible existence of a critical
flux and the desirability of starting filtration operations at a low flux. The critical-flux hypothesis is that on start-up there exists
a flux below which a decline of flux with time does not occur. Equations which may enable identification of the appropriate flux
level are included.
Keywords: Critical flux; Fouling; Membrane filtration;Dodecane-water emulsion; Yeast suspensions

1. I n t r o d u c t i o n
One of the major problems in pressure-driven
membrane processes is the reduction of the flux to far
below the theoretical capacity of the membrane. The
typical variation o f the flux with time is that of an initial
rapid decrease followed by a long and gradual flux
decline. It is well known that membrane fouling is one
of the main phenomena responsible for this. The occurrence of fouling affects the performance of the
membrane either by deposition of a layer onto the
membrane surface or by blockage or partial blockage
of the pores. This changes the effective pore size distribution.
* Corresponding author.
0376-7388/95/$09.50 1995 Elsevier Science B.V. All rights reserved
SSD10376-7388(94)00265-7

The following equation can be used to describe the

overall characteristics of flux reduction as determined
during a filtration experiment [ 1 ] :

TMP

TMP

/z(Rm+Rf)

tz(Rm+Rir+Rr)

(1)

where J is the permeation flux, T M P is transmembrane

pressure,/x is viscosity and R m is the clean membrane
hydraulic resistance. Rf accounts for both polarisation
and fouling effects on the flux and could be written as
the sum of an irreversible fouling Rir (which is not
reduced to zero when the pressure is released) and Rr
which includes reversible fouling and the polarisation
effect.
The avoidance of fouling or reduction of its severity
has been pursued for decades. However, up to now one

260

R. W. Field et al. /Journal of Membrane Science 100 (1995) 259-272

must match measures to a specific membrane process

due to the complexities of fouling phenomena. Three
major approaches can be identified [2] : hydrodynamic
(changing flow regime across the membrane surface),
surface modification (changing the surface/foulants
affinity) and regular cleaning. Now the usual way of
characterising the filtration behaviour is to perform
constant TMP filtration experiments and to study the
variation of flux with time for various operating conditions. However, as pointed out by Aimar and Howell
[ 1 ], the interpretation of the data from constant-pressure experiments often causes problems since the flux
variation produces changing conditions (of concentration, rheology, solubilities, etc.) in the boundary layer
throughout any experiment. Turker and Hubble [3]
designed a experimental set-up operating at constant
permeation flux to investigate the fouling of an ultrafiltration membrane. Aimar and Howell [ 1 ] proposed
to maintain the flux constant during a run, the TMP
then becoming a dependent variable. In such a way one
can expect the flow of material to be better controlled
since the convective flow of solute towards the
membrane is then constant throughout a run. In addition, for some chemical and biochemical processes
involving a membrane reactor, a constant flux filtration
is definitely demanded. Therefore, more studies on constant flux filtration are useful both to elucidate fouling
mechanisms and to develop processes combining reactions and separations.
In the present work, results of various constant-flux
microfiltration (MF) experiments are reported. Their
relationship to membrane fouling is then discussed. In
particular the authors introduce the concept of critical
flux together with equations which may enable identification of the appropriate flux level to minimize fouling and so improve the filtration performance of MF
membrane processes.

modes. Rep is the peak Reynolds number of the superimposed oscillatory flow. X and f are respectively
amplitude (in mm) and frequency (in Hz).
(Rep = p. Vp. D./.t- 1, where p is the density of the feed
suspension, D is the I.D. of membrane tube or the
equivalent diameter of a channel,/z is the viscosity of
the suspension and Vp = 21r.f.X is the peak velocity of
oscillatory flow.)
A summary of the experiments is given in Table 2,
and the details can be found in the related description
below.

2.2. Materials
Dodecane-water emulsion
Dodecane (Fluka, Switzerland, >98% GC) was
added to distilled water and was mixed at high shear in
a blender (Waring, USA) for 6 min. The emulsion of
0.1 wt.-% was stable during experiments.
2.3. Yeast cell suspension
Fresh bakers yeast (Saccharomyces cerevisiae) was
obtained from British Fermentation Products as compressed cake. The yeast was diluted down for use with
a solution of l0 mM phosphate buffer (pH 8.8) and 1
g l - 1 bacteriological peptone. Further details are given
elsewhere [4].
Table 1
Terminology to describe the various operating modes
Abbreviation Description
ET
B

BP
ETP

Conventional usage of tubular or flat-sheet

membrane
Baffled tube or flat sheet, no pulsing
Baffled tube or flat sheet with pulsations being
produced by the dual-beaded pump
Unbaffled system with pulsed flow

2. Experimental

2.1. Terminology
The terminology used to describe the various operating modes is defined in Table i.
Two Reynolds numbers were defined. Re is the
Reynolds number of the net forward flow in the BP
mode or cross-flow Reynolds number in the B and ET

Table 2
Summary of experimental set-up
Process fluid

Flat-sheet system

Tubular system

Yeast cell suspension

Yeast cell debris
Dodecane-water
emulsion

ET, ETP, B and BP modes ET and B modes

BP mode
ET and B modes

R.W. Field et al. / Journal of Membrane Science 100 (1995) 259-272

miD.-

261

transducers as described in Ref. [4]. Net forward flow

or steady flow was measured by a calibrated rotameter.
A temperature of 30C was used for all runs except
those with the dodecane-water emulsion. In this case
25C was chosen. Temperature control was maintained
by a coil situated in the feed tank.

Fig. 1. Helical baffle used for mode B runs with the Carbosep
membrane.

3. Results and discussion

2.4. Yeast cell debris (YCD) suspension

3.1. Flat-sheet system
Fresh bakers' yeast (S. cerevisiae) was slurried in 4
mM potassium phosphate and 150 mM sodium chloride
solution (40% wet weight suspension, i.e. 12% dry
weight) and then disrupted. Also some YCD (3% dry
weight) suspension was incubated at 29-3 IC for 3.5
h whilst it was agitated by a magnetic stirrer (this
suspension is denoted incubated YCD). Further details
are given elsewhere [5].

2.5. Filtration systems

Two filtration systems were employed in this work.
The main system consisted of a tubular membrane
(Carbosep 0.14/zm, Life Science Laboratories). For
some experiments a newly developed baffle subject to
patent protection was employed. The shape of this baffle is shown in Fig. 1. This system was only operated
under steady flow (i.e. in ET and B modes).
Reference is also made to a previously used flat-sheet
MF unit where a flat-sheet membrane was mounted in
a stainless-steel module with a perspex insert providing
ten parallel channels each 8 mm wide, 1 mm high and
84 mm long. Rectangular baffles 0.4-0.6 mm in height
were embedded in the insert and the spacing between
two baffles was 4 mm. A Ceramesh flat-sheet
membrane (Ceramesh, UK; nominal pore size 0.2/xm)
was used in this system. The membrane was a zirconiacoated metal alloy mesh. When desired pulsatile flow
was produced by a dual-head metripump. Details can
be found elsewhere [4].
For all systems, permeate flow was measured by
weighing with an electronic balance interfaced to a
computer and periodically returned to the feed vessel
so that the system may be assumed to have been operated under constant-volume conditions. The inlet and
outlet pressures were measured using absolute pressure

Fig. 2 shows some conventional flux-time data

obtained for the yeast cell suspension. There is significant flux decline with the baffled flat-sheet system,
albeit towards higher steady-state values than those
obtained in the absence of baffles. This work is covered
in detail elsewhere [4]. In retrospect the TMPs were
too high giving too higher a convective flux towards
the membrane. Subsequent work (some of which is
reported later) showed that fouling was reduced significantly at low values of TMP.
The characteristics of YCD and incubated YCD,
together with their corresponding filtration behaviour,
were described by the authors in a previous paper [ 5 ].
A limited amount of data is shown in Fig. 3. The velocity V of 0.038 m s - ~ corresponds to a net flow Re of
48. The Vp values of 0.735 and 0.38 m s - 1 correspond
to Rep = 1077 and 482, respectively. The reason for
highlighting this earlier work is the existence of constant-flux filtration at a constant TMP. For incubated
YCD the flux is invariant with respect to time provided
the TMP (and hence the corresponding flux) is sufficiently small. The steady-state fluxes for the low values
of TMP are of reasonable magnitude for YCD because
the degree of fouling is so low. The plateau of 25 1 m -2
h - ~in the flux versus time curve for the incubated YCD
at a TMP of 0.09 bar and a Vp of 0.74 m s-~ was
obtained almost immediately upon start-up; d (TMP) /
dt = 0 for all times greater than 1 min. This means that
the flux very quickly reached a steady-state value. (The
clean water flux at the operating TMP was 70 1 m - 2
h - l ). This special filtration behaviour continued, at a
lower level of flux, when TMP fell to 0.07l bar. However, after TMP was increased to 0.2 bar there was
neither a proportionate increase in flux nor was the new
value of 46 1 m - 2 h - ~ maintained.

262

R. W. Field et al. / Journal of Membrane Science 100 (1995) 259-272

6LI
,o

tb

t~

ab

2~

3b

a~

4~

50

Fig. 2. Comparisonof flux versustime behaviourfor flow in a baffledsystem (BP), pulsedflow in an unbaffledsystem (ETP) and normalflow
in an unbaffledsystem (ET) as given in Ref. 14]. Flat-sheetsystemat C=5% dry weight yeast; TMP= 1.05 bar.
50BP moOe, V = 0.038 m/e

45-

40-

35-

=,
Vp=,0,38 fft/S

'

0.20 bat, Vp=0.74 rrt~

10.07

bar, Vp,,,,0.38 m/s

II

Incui~tKI YCD

151050

Y1CD

2o

eo

loo

12o

TInm (mlnuhm)

Fig. 3. Comparisonof filtrationperformance for 3% dry weight yeast cell debris and incubatedyeast cell debris at various TMPs with a V of
0.038 m s- ~and Vpof 0.38 or 0.74 m s- 1. The baffled/pulsedmode of the fiat-sheet system was used with a Ceramesh 0.2-/xm membrane.
Details in Ref. [5].
The filtration results for the non-incubated YCD at
low TMP (i.e. low initial flux) indicate a relatively
small degree of fouling. The reasons for the existence
of the constant flux with the incubated YCD are open
to conjecture. A m o n g the several elements responsible
for it, such as the characteristics of the feed, operating
mode and TMP, a low initial flux (i.e. very low TMP)
is definitely essential.

3.2. T u b u l a r s y s t e m

The filtration results for the 1000-ppm dodecane

emulsion are shown in Figs. 4, 5, 6, 7. Three sets of the
constant flux data and their corresponding T M P versus
time profiles are shown in Figs. 4 and 5, respectively.
The three experiments were performed in the B mode
with V at 0.84 m s - 1. TMP was gradually increased
from a low level to a higher level in order to maintain

263

R.W. Field et al. / Journal of Membrane Science 100 (1995) 259-272

250
1000ppm O o ( W ~ , B mocle,V - 0.84 m/s

200-

~(15o;

Run
_

1
m

~-Run2

~J too;
U.
Run

50

OI

6b

eo

loo

1~

laO

1~o

18o

Time (minutee)

Fig. 4. Three sets of flux data of three constant-fluxfiltrationsfor a 1000-ppmdodecane-wateremulsionand with a Carbosep 0.14-/zmtubular
membrane.The data for the three runs were acquired at a Vof 0.84 m s- ~and in the baffled mode.
the flux constant for all three runs. This means that
gradual fouling occurred. From the authors' experience
of this mode of operation the chosen flux is of crucial
importance in order to obtain the best compromise
between flux and TMP. If the starting flux was too low
as shown by run 3, the flux was easy to be maintained
but at a low level. If too high, over-fouling would occur
and it would be impossible to increase T M P sufficiently
in order to maintain the flux constant. Fig. 6 compares
a constant-flux profile with two flux profiles for constant TMPs of 0.44 and 0.70 bar. The individual data
points for the constant T M P runs were obtained at the
same frequency as for the constant flux run but are not

shown. (This comment also applies to Figs. 7 and 8.)

All three profiles were obtained in the B mode with a
V of 0.84 m s - t. When the experiment was performed
at a constant TMP of 0.70 bar, the flux decreased
sharply and rapidly to a much lower steady value of
about 70 1 m - 2 h - i. Presenting a striking contrast to
the this result, the flux for the constant-flux operating
mode was always maintained to about 140 1 m -2 h - l
by gradually increasing TMP from 0.44 to 0.58 bar. It
is interesting to note that when a constant TMP of 0.44
bar was set the flux was maintained at about 135 1 m -2
h - ~for about 15 min, then declined steadily to a steady
value of about 85 ! m - 2 h

0.8
0.7Run

0.60.5"

" o.4:
0.30.2-

,ooop p ~ , = , = ~ , s , , ~ . v = o 8, r,V,

0.10

2b

80

100

1~

140

180

180

Time (minutee)

Fig. 5. Profilesof TMP-time correspondingto the three sets of data in Fig. 4.

2O0

264

R. W. Field et al. /Journal of Membrane Science 100 (1995) 259-272

-0.6

250D=,,-

0.5

200-

0.4
,~.

i
--

0.3

OA.4I ~

100-

I-0.2

50-

0.1

lo00ppmdocleQme Bmode V os4m/s

r

20

40

60

80

100

120

140

160

180

200

1'1rile (minute)

Fig. 6. Comparisonof the performancebetweena constant-fluxfiltrationand two constant-pressurefiltrationsfor a 1000-ppmdodecane-water

emulsion.All data were acquired at a Vof 0.84 m s- l with a Carbosep 0.14-p,mtubular membranein the baffled mode. For the two constantpressure runs TMPs were 0.44 and 0.70 bar.
shown in Fig. 7 were obtained in the B mode and a
larger cross-flow velocity of 1.01 m s - ]. The constantflux experiment lasted for 7 h with TMP increasing
from a starting value of 0.32 bar to a terminal value of
0.73 bar. The flux value was between 140 and 150 1
m - 2 h - 1. The fluxes for constant TMPs of 0.52 and
0.80 bar dropped rapidly and the steady value for TMP
of 0.52 bar was about 100 1 m -2 h - ] . Therefore the
results shown in Figs. 6 and 7 suggest an operating
tactic for the commencement of filtration. It has long

A reasonable explanation for the results in Fig. 6 is

that at a high T M P such as 0.70 bar the membrane
would over-foul (for example, dodecane may be forced
to enter membrane pores) and cause a irreversible fouling resistance within a short time. Indeed the very rapid
decline of flux during the first 10 rain suggests this. On
the contrary, a lower starting TMP and a gentle increase
of TMP leads to a low rate of fouling and reduces the
degree of irreversible fouling. A further confirmation
of the above results is presented in Fig. 7. The data

0.8
"0.7

"0.6
<

-0.5 "1~

150-

lil
o

-0.40.
-0.3 ~

0.~?I~"

-0.2
looo ~

~b

16o

==,~,

I~O

B,,~,

26o

-0.1

v = 1 ol . ~

~,o

36o

4oo

0
45O

Fig. 7. Comparisonof the performancebetweenconstant-fluxfiltrationand constant-pressurefiltrationfor a 1000-ppmdodecane-wateremulsion

with a Carbosep0.14-Fmtubularmembrane.All data were acquiredat a Vof 1.01 m s- ] in the baffledmode.For the constant-pressurefiltration
the TMP was varied from 0.52 to 0.80 bar.

R. W. Field et al. / Journal of Membrane Science 100 (1995) 259-272

250-

265

0.45

1000 pprn Oodec.a/~, E'f mode, V = 0.11 m/~

i

i 0.:35
!

* ~

~t.-~-t

-~1+" .~--~/~

0.3

io25

0.2

loo

rl

I--

0.15

i1

0.1

50

0,32_ bar

~)

4()

I~)

"
0.3.2 bar
-

130

100

120

0.05

, -140

J-O
160

Time (minutee)
Fig. 8. Comparison of the performance between constant-flux filtration and constant-pressure filtration for a 1000-ppm dodecane-water emulsion
and with a Carbosep 0.14-/xm tubular membrane. All data were obtained at a V of 0.71 m s t and in the conventional mode. For the constantpressure filtration, the T M P was varied between 0.32 and 0.80 bar.

been recognised that low-pressure MF is much more

effective than high-pressure MF but the emphasis here
is the desirability of starting filtration operations at a
moderate flux and then maintaining it by gradually
increasing the pressure.
Since the data in Figs. 4, 5, 6 and 7 were obtained in
the B mode, one may ask whether the tactics work
without the baffle. All the data in Fig, 8 were obtained
in the ET mode and with a V of 0.71 m s - 1. For the
constant-flux experiment TMP was increased from a
starting value of 0.25 bar to a terminal value of 0.4 bar.
For the constant-pressure experiment, the fluxes were
separately measured at TMPs of 0.32, 0.55 and 0.80
bar. For constant TMP of 0.32 bar the flux remained
constant for the first 15 min and then declined to about
80 1 m - 2 h - 1. However, for the constant-flux run, the
flux was increased slowly over the first 5 min and then
maintained at about 95 ! m - 2 h - i with TMP increasing
slowly to 0.32 bar. Although the fluxes are lower than
what would be obtained with the new baffle the constant-flux tactic still has an advantage over the conventional constant-pressure approach. Towards the end of
the constant-pressure run the TMP was increased
beyond that necessary to maintain a flux of 95 1 m - 2
h - ' but the new and slightly higher value of flux was
not maintained. It seems that the flux value of 95 1 m 2
h - ~ was critical.
The final experiments in this work were conducted
with yeast cell suspension (5% dry weight) using the

B mode. Two sets of constant-flux data were obtained

at V o f 0.84 and 0.76 m s - '. (For reasons of clarity the
individual data points are not shown in Fig. 9 for the
latter experiment but the frequency of data sampling
was the same as for the other experiment.) For the case
of V of 0.76 m s - ', the constant-flux behaviour was
maintained at about 40 1 m -2 h-~, which was about
80% of the clean water flux corresponding to a TMP
of 0.10 bar. The corresponding TMP was always at the
level of about 0.10 bar except for the first 4 min when
TMP was at 0.075 bar. When an initial TMP of 0.16
bar was used in the B mode together with a higher
cross-flow velocity of 0.84 m s - ' it was only possible
to maintain constant flux by gradually increasing TMP
to over 0.60 bar. Thus the earlier flux of 40 1 m - 2 h '
(corresponding to the TMP of about 0.10 bar) is critical. The constant-flux filtration rate for the 0.16--0.6
bar run had a modestly higher value of about 57 1 m 2
h - ~ but the final overall resistance is over four times
greater.
It might be thought that at the moderate flux of 40 1
m - 2 h - ' and with the enhanced mass transfer being
generated by the new baffle, there might be only a
"flowing cake" layer and a very small (or nil) "stagnant cake" layer on the membrane surface [6]. The
difference between the process flux and the corresponding clean water flux might then be attributed to this
flowing cake layer which only forms a reversible

266

R. 14/, Field et al. /Journal of Membrane Science 100 (1995) 259-272

"0.7

140-

a %o'wyearn.,B mode
-0.6

120-

100-

-0.5

I V = 0.84 rt~

-0.4 ~.

<

Is

-0.3 I-

r.

-0.2
V = 0.7e nVs

-0.1

OJ

20

40

60

80

100

120

140

160

Time (minutee)
Fig. 9. Comparison of two constant-flux filtrations for 5% dry weight yeast cell suspension and a Carbosep 0.14-/zm tubular membrane. The
two sets of data were obtained at a V of 0.76 and 0.84 m s- t, respectively, using the baffled mode.
membrane resistance. However, the water flux obtained
after rinsing the membrane at the end o f the experiment
indicated that none o f the fouling material had been
removed. The pre-experiment water flux was restored
by cleaning. This indicates that the small increase in
resistance during the 401 m - 2 h - l constant-flux experiment was not due to a flowing cake. The water flux
obtained after rinsing the membrane at the end of the
57 1 m - 2 h - i constant-flux experiment, during which
period the overall resistance increased four-fold, indi-

cated that the resistance was halved by the simple action

of rinsing. Thus, in this case it is believed that a stagnant
cake layer was formed. Following the recent work of
Hodgson et al. [7], the increase in resistance could be
due to the nature o f the extra-cellular material in the
fouling layer and not to a layer of yeast cells alone. The
critical flux would seem to correspond to the flux which
just avoids cake deposition.
Fig. l 0 provides further support for the suggested
operating tactic. Some data for constant-pressure filtra-

160-

-0.7

140.

-0.6

120-

-0.5

<~ 100-0.4
-0.3 iI,-

6o-

-0.2

40-

0.38 ~

201
0 - - -

-0.1

B% O'WymM, B m~kl, V = o,e4 m~

20

40

80
Time

8()
100
(mlnutee)

120

140

ld

Fig. 10. Comparison of the performance between a constant-flux filtration and a constant-pressure filtration for 5% dry weight yeast cell
suspension and with a Carbosep 0.14-~m tubular membrane. All data were obtained at a V of 0.84 m s- ~and in the baffled mode. For the
constant-pressurefiltration, the TMP was varied between 0.38 and 1.04 bar as shown.

R. W. Field et al. / Journal of Membrane Science 100 (1995) 259-272

267

250
Meuured ~ Q-I~O mVmin, TMP-0.4 ~

and 30 C

~I0"

15010050-

8
1'0
1;~
1'4
Number of exp~mente

16

18

20

Fig. l 1. Illustration of the membrane condition at the end of experiments involving yeast cell suspensions in the tubular system with and without
baffles. No baffle was used for experiments numbered 1, 2, 7 and 8. The helical baffle as shown in Fig. 1 was u ~ d for experiments 3~5 and 1118. A simple symmetrically located rod insert was used for experiments 9 and 10.

tion were obtained in the B mode at a Vof 0.84 m s - i .

Several TMP levels, namely 0.38, 0.50, 0.83 and 1.04
bar, were used. Compared with constant-flux filtration,
the flux declined rapidly and no improvement in the
flux was observed by increasing TMP to 1.04 bar. Actually the flux at highest TMP was even lower than the
corresponding flux of the constant-flux filtration terminating at 0.6 bar since the high starting TMP and the
following higher TMPs caused over-fouling. The magnitude of the fouling resistances (i.e. the size of the R f
values) increase with initial TMP; Rr was lowest for
the 40 1 m -2 h -1 constant-flux experiment, significantly higher ( and increasing) during the 571 m - 2 h - l
constant-flux run and even higher during the constantpressure runs which started at a flux of 1101 m - 2 h - 1.
The final overall resistances were in the ratio of 1:4:7.
Finally, it should be pointed out that the avoidance
of over-fouling resulted in another advantage; besides
improving the filtration rates, the membrane was more
easily cleaned. Fig. 11 gives some data on the water
fluxes achieved after rinsing and after cleaning for a set
period of time. The eighteen experiments were for the
yeast cell suspension. The fluxes achieved are obviously dependent upon the various conditions used during each experiment but clearly there is a significant
difference when the spiral baffle is used. Experiments
16-18 correspond to the three runs discussed in the
previous paragraph.

4. Critical flux concept and implications

4.1. Concept of critical flux

From the experiments that show that one can operate
MF at a constant flux with no increase in TMP so long
as that pressure is low we deduce that under these
circumstances fouling is slight or negligible. On the
other hand as soon as TMP is increased fouling is
observed and the flux declines towards its previous
value or even below it. Certainly once the lower TMP
is restored the flux is lower because of the additional
fouling layer on the membrane surface.
The constancy of the plateau values for conditions
of low and moderate flux in, for example, Figs. 3 and
9 suggests that there was a critical flux for these particular conditions. For the incubated YCD in the fiat-sheet
system the value was 25 I m -2 h-1 whilst for the 5%
dry weight yeast in the tubular system the value was
40 1 m - 2 h ~ for the conditions used. A hypothesis is
now proposed.
The critical flux hypothesis for MF is that on
start-up there exists a flux below which a decline of
flux with time does not occur; above it fouling is
observed. This flux is the critical flux and its value
depends on the hydrodynamics and probably other
variables.
The strong form of the hypothesis is that a flux exists
which is equivalent to the corresponding clean water

268

R. W. Field et al. / Journal of Membrane Science 100 (1995) 259-272

flux at the same TMP. The weak form is that on startup a constant flux is rapidly established and maintained.
The determination of critical flux is difficult to determine a priori because surface interactions of the species
involved can vary significantly from system to system.
Nevertheless it is suggested that the concept may be of
value, as a guide, when optimising a system. Theoretical calculations for idealised particles have been made
[ 8 ]. For small particles of the order of 0.1/~m StokeEinstein diffusion away from the surface will be important and calculations suggest a critical flux of 1-10/zm
s-1 depending on the significance of charge effects.
For particles of the order of 1/zm charge is less important and shear-induced diffusion alone, for non-interacting particles, should give a critical flux of the order
of 10/zm s - 1 (equates to 36 1 m - 2 h - ~). This is close
to the values of 25 and 401 m -2 h - ~determined experimentally for incubated YCD and 5% dry weight yeast.
The critical flux should not be confused with the
limiting flux found in ultrafiltration which occurs independently of fouling and is due to concentration polarisation. This phenomenon induces both osmotic effects
and reduced mass transfer (due to increases in the viscosity at the membrane surface). The physical phenomena governing the critical flux may be in doubt but
evidence for the existence of critical-flux effects is now
quite extensive. Two experimental observations will be
mentioned. Firstly, Benkahla et al. [9] have observed
that as TMP is increased the flux increases linearly and
provided a critical value of flux is not exceeded the
behaviour is reversible, i.e. pressure can be reduced
and the same fluxes are again observed. However, if
the critical flux value is exceeded then reducing TMP
does not restore the original flux but a lower one. The
cycling of pressure then leads to a hysteresis effect. The
value of the critical flux below which no hysteresis is
observed can be increased by increasing the cross-flow
velocity. Secondly, the work of Hodgson et al. [ 10] is
also relevant; non-intrusive observation of the surface
of a 0.02- and a 0.2-/zm Anopore membrane filtering
0.1 wt.-% yeast indicated that at low flux cells were not
observed to approach the surface.
Now the rate of convection of material towards a
membrane surface under low TMPs is much lower than
that which occurs during the initial period of conventional operation. The rate of back-diffusion (or tangential removal) of rejected material will depend upon its
concentration, material properties and the fluid

mechanics. It is conceivable that this could match the

convection rate of rejected material towards the
membrane surface. For example, the results presented
in Fig. 9 provide convincing evidence that for the yeast
cell suspension and operating, used a flux of 57 1 m - 2
h - ~ is excessive whilst one of 40 1 m - 2 h - 1 is not.
Now for a single given fouling mechanism a relationship can be written relating flux, rate of flux decline
and back-diffusion. Relevant equations have been
derived. They are a development of those of Hermia's
work [ 11 ] and are given in the Appendix. A unifying
equation for cross-flow MF is presented. This demonstrates that mathematically one can anticipate that there
are fluxes below which no decline of flux with time
Occurs.

4.2. Consequences of the concept of critical flux

Consider an experiment in which the initial flux is
greater than the critical flux for all components of a
suspension, i.e. the mode of operation that can be considered to be the normal mode of start-up. All particles,
irrespective of size, in the vicinity of the membrane will
experience a net force towards the membrane. Now
back-flux mechanisms are generally considered to be a
function of particle size and for most proposed mechanisms [ 13] the back-flux increases with particle size.
Thus, if the critical flux is higher for larger particles,
then, as flux declined, one would observe that (a) initially deposition would not favour any particular size
of particle, but (b) as the critical flux of each size of
particle was progressively passed the percentage of
larger particles would decline (because the critical flux
would no longer be exceeded) and the nature of the
cake would become finer and finer. Such changes in
cake size distributions have been observed but not fully
explained (see for example Ref. [ 14] which cites five
other references). A shear classification mechanism
has been one explanation offered [15]. These observations may also be interpreted in the light of the critical-flux hypothesis. This not only suggests why in
general it has been observed that the mean size of particles found in the cake layer is less than that of the
feed particles, but indicates that operation below the
critical flux of all particles should lead to a cake with a
different structure or possibly no cake at all. Measurement of the critical flux for a range of particle properties
is highly desirable.

R. W. Field et al. /Journal of Membrane Science 100 (1995) 259-272

Finally the concept of critical flux immediately suggests that there may be advantages in operating below
it so as to minimise cleaning. In fact this has been
exploited by Kubota. Ishida et al. [ 16] reported on their
membrane-activated sludge process in which membranes are submerged in the aeration tank and operated
at constant flux. Under normal conditions flux is maintained at 0.5 m 3 m - 2 per day with a steady pressure of
about 30 kPa. Means to cope with storm overloads is
important and tests at 70 kPa showed that the flux would
initially increase to 1.05 m 3 m -2 per day but decline
gradually to 0.94 m 3 m - 2 per day due to fouling. Thus,
the new increased flux can be said to have been greater
than the critical flux for this system. On returning to
the original flux, which was below the critical flux, the
pressure required was one third higher than that originally required but the extent of fouling gradually diminished and after 4 h the pressure was about 35 kPa. The
advantages of operating in such a simple mode over
extended periods in unmanned systems is clear. Lastly
a second Japanese paper presented at ICOM '93 will
be mentioned. Whilst Yamamura et al. [ 17] attribute
the performance of the Toray "superfiltration
membrane" to low absorption of fouling materials and
the pore size range of 0.01-0.03/.tm, the operation at
moderate fluxes is also considered, by us, to be highly
relevant. The corresponding TMP was around 50 kPa.

5. Conclusions
For three different systems constant-flux filtration
was realized by gradually increasing TMP from an
appropriate starting value. Under certain circumstances
a constant flux can be achieved at a constant TMP as
shown in the filtration of yeast cell debris and 5% dry
weight yeast suspension. The critical and generally low
value of TMP below which constant-flux filtration can
be realized depends upon the hydrodynamics. Below
this critical TMP, there will be little or even no irreversible surface fouling.
Fundamentally the process is best understood in
terms of a critical flux; the value of TMP is not of
physical significance per se but is crucial in that for a
given membrane resistance it is related to flux and is
often the controlled parameter. By correctly selecting
the initial TMP the rate of fouling is greatly reduced
because a critical flux is not exceeded. Thus ideally a

269

constant-flux, rather than a constant-pressure, operating mode is to be preferred.

A mathematical analysis of the classic constant pressure blocking filtration laws, with due allowance for a
cross-flow removal mechanism, has been completed.
This demonstrates that provided certain terms take
finite values fluxes should exist at which there is no
decline of flux with time.

Appendix

I. Incorporation of cross-flow removal mechanisms

into classic constant-pressure blocking filtration laws
Hermia [ 11 ] presented a physical model to derive
the so-called "intermediate blocking" law, which had
previously been considered totally empirical. Also he
determined a characteristic form of the blocking filtration laws for constant-pressure filtration. He presented
the result in the form

d2t

k( dt ~n

t791

(A1)

where V is volume of filtrate collected in time t, and k

and n are constants depending upon the mechanism
involved.
As analyses of membrane filtration are normally performed in terms of flux it is noted that Eq. A1 can be
presented in an alternative form. As dV/dt=AJ it follows that

dZt

d( l/AJ)
dV

dV 2 =

1 dJ

1 dJdt

AJ 2 dV

AJ 2 dt dV

1 dJ
A2J 3

dt

Thus Eq. A2 can be rewritten in a physically more

meaningful form as

1 dJ k
A2J 3

dt

(A2)

For dead-end filtration and a complete blocking

mechanism n = 2, thus Eq. A2 is immediately seen to
reduce to - dJ/dt = kJ. This and the other three mechanisms explored by Hermia will be re-examined, in
turn, with a cross-flow mechanism being added where
appropriate.

270

R. W. Field et al. / Journal of Membrane Science 100 (1995) 259-272

" C o m p l e t e " blocking filtration law

Normally with this mechanism it is assumed that

each particle reaching the membrane participates in
blocking by means of pore sealing. A term representing
a removal rate of particles from the pore mouths will
be added in. The velocity through the unblocked pores
is unaffected, thus the fractional reduction in flux is
equal to the fractional reduction in open area, i.e.
dJ.

da.

-~lJo=-~lao

(A3)

The area at time t is, in the absence of a removal

term, given by
a=ao-crV

(A4)

Combining Eqs. A3 and A4, remembering that

a filter resistance, which is assumed to remain unchanged, and a cake resistance. For dead-end filtration of
incompressible particles the cake thickness is proportional to the volume filtered and the overall resistance
is
R = Ro + am = Ro + ak V/A

where a = specific resistance of cake (m kg- 1),

m ---mass of cake per unit area, and kc = cake filtration
constant relating m to V (it is a function of certain
physical properties) (kg m - 3)
In classic filtration the equations are solved to show
that this mechanism yields a linear relationship between
V and t/V. An equivalent analysis yields an equation
which might be of greater interest for the analysis of
membrane performance

d V / d t = A J and that ao = eoA, one obtains

dJ

- - = - ( crAJ / ao) Jo = - ( crJo/ eo) J

dt

(A5)

If the cross-flow leads to a constant rate of particle

removal from the mouths of the pores, the equation for
the decrease in open area is modified
da

- ~ = erAJ - Bao
dt

(A6)

(A9)

l=~ofJdt
ak '

Jo

(A10)

Eq. A10 is for dead-end filtration. A term allowing

for cake erosion can be added to Eq. A9 and a crossflow microfiltration (CFMF) cake filtration equation
can be obtained analytically. For a time t, Eq. A9 can
be written as
t

where B is a constant related to the membrane porosity

and rate of removal of particles per unit area. Hence

R = Ro + a(k~V/A - f S d t )

(All)

dJ
dt

- - = - ( trJo/ eo)J + BJo

(A7)

The implication of Eq. A7 is that when J is equal or

less than Beo/cr, there will be no flux decline. Writing
Beo/tr asj~, integration of Eq. A7 yields
J = ( J o - j ) e x p ( - (O'Joi~)t) +Jc

(A8)

A simple form of this equation has been used by

others for principally empirical reasons (e.g. Ref.
[ 12] ). However, its use is not recommended unless
the fouling mechanism is described by the above equations. In the context of this paperjc can be viewed as a
critical flux; i f J is less than j there is no flux decline.
Cake filtration law

The mechanism which differs most greatly from the

complete blocking mechanism is the one of cake filtration. In this case the overall resistance is composed of

where S is the rate of erosion of cake per unit area in

kg m-2 s -I"
Assuming S to be invariant with t, it can rapidly be
shown that
t

Jo

Jdt-JoRo

(A12)

Thus
l dJ

j2 dt

akJ

aS

JoRo JoRo

(A13)

Writing G = ak/(JoRo) andjs = S/k one obtains

ldJ
--~ dt= G(J-js)
This can be solved to give

(A14)

R.W. Fieldet al. / Journalof MembraneScience 100 (1995)259-272

J Jo-Js~
G t = ~1 [ l n ( -j s k kJo

(1

1~]

(A15)

Whilst Eq. A15 is useful in the analysis of flux

decline curves, the main point to note in the context of
this paper is that Js can be considered to be a critical
flux; from Eq. A14 it is seen that dJ/dt = 0 when J =Js.

271

The term Ks' is defined by Hermia [ 11 ] and is the

value taken by k in Eq. A2 when n = 3/2, this being the
characteristic value for this form of fouling. This form
of fouling will not be mediated by back-diffusion from
the membrane surface and will not be considered further.

2. Summary
Intermediate blocking filtration law
Each particle does not necessarily block a pore; the
probability of landing upon particles already on the
surface is taken into account. The physical model leads
to
1

....
J Jo

trt

(A16)

For dead-end filtration the rate of flux decline is

given by the following equation which is obtained from
Eq. Al6:
ldJ
....
J dt

trJ

(A17)

The addition of a back flux factor B' gives

ldJ
....
J dt

o-J-B'

(A18)

Writingji = o-B' and solving one obtains an implicit

equation for J
~rt = 1 [lnJj-~ ji j J j i ]
Ji

(A19)

This equation (and also Eq. A8) should prove useful

in the analysis of the initial phase of flux decline. The
main point to note in the context of this paper is thatji
can be considered to be a critical flux. The previous
value Js is the limiting value if cake laydown is to be
avoided, whilst j~ is the limiting value if intermediate
blocking is to be avoided.

Standard blocking filtration law

In deriving this law it is assumed that the pore volume decreases due to particle deposits within the pores.
In classic filtration the equations are solved to show
that fouling of this type yields a linear relationship
between t and t/V. An equivalent analysis in terms of
flux generates
1/jo.5 = 1/jo.5 + (Ks,/2)A'St

(A20)

The appendix started by giving Hermia's unifying

equation for dead-end filtration. It is now appropriate
to generate a unifying equation for CFMF. Eqs. A7,
A14 and A18 were all of the form

-~tJ~- Z=k(J-J )

(A21)

For cake filtration n = 0, for complete blocking n = 2,

whilst for the intermediate mechanism n = 1. Thus Eq.
A21 is characteristic of CFMF with the constant k and
index n taking different values depending upon the
fouling mechanism.
In each case the term J* can be considered to be a
critical flux which should not be exceeded if fouling is
to be avoided. This analysis therefore suggests that,
provided the terms B, B' and S take finite values, fluxes
should exist at which there is no decline of flux with
time.

3. Symbols used in appendix

Symbols in main text defined at point of use.
membrane surface area (m 2)
open area
(m 2)
constant in Eq. A6
( s - ~)
constant in Eq. A18
(s-t)
parameter defined in
( s m - 2)
text
J
volumetric flux
(m s- l )
Jr
critical flux in modified (m s - 1)
complete blocking law
Ji
critical flux in modified (m s - ~)
intermediate blocking
law
Js
critical flux in modified (m s - l )
cake filtration law
Ks' Hermia's parameter for
standard blocking
filtration
A
a
B
B'
G

272

R.W. Field et al. /Journal of Membrane Science 100 (1995) 259-272

defined by Eq. A 1

V
a

(units depend on
mechanism)
cake filtration constant (kg m-3)
mass of cake per unit (kg m-2)
area
index defined by Eq.
(units depend on
A1
mechanism)
hydraulic resistance
(m - t )
rate of cake erosion per ( k g m - 2 s -1)
unit area
filtrate volume
(m 3)
specific resistance of
(m kg- 1)

cake
membrane surface

k
m
n
R
S

porosity
blocked area per unit ( m - ~)
volume of filtrate
Subscript
0
condition of time t = 0
tr

Acknowledgement
One of us (RWF) is grateful to the Royal Academy
of Engineering and Shell Research for financial support
and the award of a Senior Research Fellowship.

References
[ 1 ] P. Aimar and J.A. Howell, Effects of concentration boundary
layer development on the flux limitations in ultrafiltration,
Chem. Eng. Res. Des., 67 (1989) 255-261.
[2] J.A. Howell, Fouling and process design, Plenary Lecture, in
P. Aimar and P. Aptel (Eds.), Euromembrane 92, Pads, 1992
(Recents Progres en Genie des Procedes, Voi. 6, No. 22),
Groupe Francais de Genie des Procedes, Nancy, 1992, pp. 195207.
[ 3 ] M. Turker and J. Hubble, Membrane fouling in a constant flux
ultraliltration ceil, J. Membrane Sci., 34 (1987) 267-281.

[4] J.A. Howell, R.W. Field and D. Wu, Yeast cell microfiltration:
flux enhancement in baffled and pulsatile flow systems, J.
Membrane Sci., 80 (1993) 59-71.
[5l D. Wu, J.A. Howell and R.W. Field, Pulsatile flow filtration of
yeast cell debris: influence of pre-incubation on performance,
Biotechnol. Bioeng., 41 (1993) 998-1002.
[6] C.A. Romero and R.H. Davis, Global model of crossflow
microfiltration based on hydrodynamic particle diffusion, J.
Membrane Sci., 39(1988) 157-185.
[7] P H. Hodgson, G.L. Leslie, R.P. Schneider, A.G. Fane, C.J.D.
Fell and K.C. Marshall, Cake resistance and solute rejection in
bacterial microfiitration: the role of the extracellular matrix, J.
Membrane Sci., 79 (1993) 35-53.
[8] P. Bacchin, P. Aimar and V. Sanchez, A model for colloidal
fouling of membranes, AIChE J., in press.
[9] Y.K. Beakahla, A. Ould-Dfis, M.Y. Jaffrin, Hydrodynamic
anti-fouling mechanism in crossflow microfiltration, Poster
Presentation at ICOM '93, Aug. 30-Sept. 3, 1993, Heidelberg.
[ 10] P.H. Hodgson, V.L. Pillay and A.G. Fane, Visual studies of
crossfow microfiltration with inorganic membranes: resistance
of biomass and particulate cakes, 6th World Filtration
Congress, Nagoya, 1993, pp. 607--610.
[11] J. Hermia, Constant pressure blocking filtration laws.
Application to power-law non-newtonian fuids, Trans. Ind.
Chem. Eng., 60 (1982) 183-187.
[ 12] A. Koltuniewicz, Predicting permeate flux in ultrafiltration on
the basis of the surface renewal concept, J. Membrane Sci., 68
(1992) 107-118.
[13] M.H. Lojkine, R.W. Field and J.A. Howell, Crossflow
microfiltration of cell suspensions: a review of models with
emphasis on particle size effects, Trans. Ind. Chem. Eng., 70C
(1992).
[ 14] E.S. Tarleton and R.J. Wakerman, Understanding flux decline
in crossflow microfiltration. Part 1. Effects of particle and pore
size, Trans. Ind. Chem.Eng., 71A (1993).
[ 15] R.J. Baker, A.G. Fane, C.J.D. Fell and B.H. Yoo, Factors
affecting flux in crossflow filtration, Desalination, 53 (1985)
81.
[ 16] H. lshida, Y. Yamada, M. Tsuboi and S. Matsumura, Kubota
submerged membrane activated sludge process. Its application
into activated sludge process with high concentrations of
MLSS, Presentation at ICOM'93, Aug. 30~Sept. 3, 1993,
Heidelberg.
[ 17] H. Yamamura, Y. Fujii and T. Uemura, Advanced feed outside
hollow fibre superfiltration membrane with excellent
characteristics to remove the fouling components, Presentation
at ICOM'93, Aug. 30-Sept. 3, 1993, Heidelberg.