Vous êtes sur la page 1sur 9

Journal of Membrane Science 282 (2006) 124132

Whey protein fouling of microfiltration ceramic


membranesPressure effects
S.A. Mourouzidis-Mourouzis, A.J. Karabelas
Department of Chemical Engineering, Aristotle University of Thessaloniki and Chemical Process Engineering Research Institute,
CERTH, 6th Km Charilaou-Thermi Road, GR 570 01, Thermi, Thessaloniki, Greece
Received 25 January 2006; received in revised form 29 April 2006; accepted 7 May 2006
Available online 12 May 2006

Abstract
Permeate flux decline in dead-end microfiltration of whey protein isolate solutions is studied, using disc-type ceramic membranes of nominal
pore size 0.8 m. The tests involve five successive filtration cycles, under fixed filtration pressure, with intermediate backwashing. Flux decline
analysis and membrane resistance are employed to determine fouling mechanisms, with respect to applied pressure, in the range 2.510 psi.
Permeate and retentate concentration data complement the permeation rate measurements. Both irreversible and reversible fouling is identified for
the backwashing mode employed. Data interpretation suggests that more rapid fouling, as well as significant surface-layer compaction effects
may occur at the higher pressures employed. There is evidence that irreversible fouling effectively develops during the first cycle of the tests (of
30 min duration) and apparently does not significantly increase later on; however, reversible fouling occurs through the entire filtration series, being
more intense during the first minutes of each cycle. The effect of protein aggregates is investigated with the aid of DLS measurements and filtration
tests with prefiltered solutions; it appears that whey protein aggregates present in the solution, are responsible (almost entirely) for the observed
membrane fouling. An effective membrane cleaning procedure is proposed. Finally, the possible advantage of operating large-pore size MF (for
the process considered) at rather low pressure is suggested.
2006 Elsevier B.V. All rights reserved.
Keywords: Microfiltration; Ceramic membranes; Whey protein fouling; Pressure effects

1. Introduction
Microfiltration is a widely used process with many applications in food industry. Apart from raw beer and fruit juice
clarification, there is great interest for dairy applications; bacterial removal, selective separation of micellar casein, selective fractionation of milk fat and removal of whey fat, cheese
brine purification, are processes industrially adopted or under
development [1]. Membrane fouling is observed in such dairy
applications that involve protein solutions in both UF and MF
processes. Consequently, efforts have been made to identify the
role and effect of proteins in membrane fouling, the mechanisms
involved, as well as the influence of operating parameters [2].
However, despite progress made, there are significant gaps in
our knowledge, mainly due to many variables involved in these
filtration processes.

Corresponding author.
E-mail address: karabaj@cperi.certh.gr (A.J. Karabelas).

0376-7388/$ see front matter 2006 Elsevier B.V. All rights reserved.
doi:10.1016/j.memsci.2006.05.012

This study is motivated by the need to optimize MF-based


processes for bacterial removal from dairy fluids. Pasteurization of dairy fluids is the main process used for over 50
years, in order to produce safe consumer products. However,
in addition to the undesirable protein denaturation due to heat
effects, other drawbacks of pasteurization impacting on shelflife of these products have been identified. There is evidence,
indeed, that pasteurization does not totally destroy bacteria
spores or even thermoduric bacteria, thus reducing product shelflife; furthermore, species killed remain in the fluid as dead
cells, with their potentially active enzymes being capable of
altering organoleptic properties of the product during shelflife [1]. Although commercial milk products, characterized as
MF milk, have appeared in the market, thermal pasteurization still remains by far the most frequently applied process,
possibly due to the lack of confidence in the application of
membranes and to somewhat greater cost involved. Durable
inorganic membranes appear to be most suitable for developing a reliable and cost effective process. Ceramic membranes
have the well-known advantages of reusability, as well as the

S.A. Mourouzidis-Mourouzis, A.J. Karabelas / Journal of Membrane Science 282 (2006) 124132

capability to be cleaned using standard food industry sanitizing


agents such as NaOH, HNO3 , NaOCl and even H2 O2 if necessary.
For the removal of the bacterial content and spores from
raw or skim milk, usually ceramic membranes of pore diameter between 0.2 and 1.4 m are used, achieving permeate milk
product ranging from almost sterile to suitable for cheese making; such dairy industry applications include the Bactocatch
process or variations thereof (e.g. the one by APV) [1]. Permeate flow reduction due to fouling is always a significant obstacle
to deal with, but of equal importance are also the rejection
characteristics of the membrane filtration process in regard to
bacteria and protein content of milk [36]. Membranes with
nominal pore size 0.8 m appear to perform adequately for the
reduction of the bacterial content of skim milk, as well as for
removal of raw milk fat and for cheese brine purification [1].
However, during filtration of proteinaceous solutions (i.e. whey
or milk), it has been observed that proteins tend to foul such
membranes having pore sizes much greater than the size of constituent protein monomers or dimers. This is attributed to the
presence of protein agglomerates [7] that may have sizes up to
three orders of magnitude greater than that of monomer. Casein
micelles present in milk may be responsible to some extent
for fouling, but whey proteins are considered more important,
-lactoglobulin being the most significant one in this respect
[7,8].
In order to investigate the effect of protein fouling in a bacterial reduction filtration processes, bacteria-free protein solutions
are employed in this study first. Parallel work by the authors
in progress, using whey protein concentrate solutions as feed
to a food-grade, crossflow filtration pilot unit, utilizing tubular
ceramic membranes of nominal pore size 0.8 m, shows that
very significant flux reduction takes place; under such conditions, one might have expected smaller flux reduction (due to
possible pore narrowing and/or partial pore blockage of membranes), since whey protein normal sizes (monomers or dimers)
do not exceed 510 nm in diameter, compared to the bacteria rejecting pores of active membrane layer of size 0.8 m.
For a cost effective filtration process to be developed, performance characteristics in respect of applied trans-membrane
pressure should be established. Previous studies [912] employing organic membranes (0.10.45 m) and BSA solutions, as
well as other work [13] with -lactoglobulin solutions and
ceramic membranes (0.3 m) have already revealed that pressure has a significant effect on membrane filtration. In this
study, the performance of ceramic membranes of a larger nominal pore size (0.8 m), than employed so far, is examined
with a different feed solution. To better understand the mechanism of whey protein fouling and the effect of pressure in
microfiltration, the experiments are carried out under wellcontrolled conditions, using disc-type ceramic membranes, in
dead-end stirred-cell filtration of whey isolate protein solution. The latter is chosen in order to have a feed solution
much more pure with respect to whey proteins (compared to
WPC), as well as of better clarity, thus facilitating the use
of spectrophotometric techniques for concentration measurements.

125

2. Materials and methods


2.1. Protein solutions
Protein solutions were prepared by dissolving 4 g of whey
protein isolate powder in 1 l of distilled water with NaN3 . Difficulties were experienced in obtaining constant clean water flux,
even with freshly produced distilled water, which was attributed
to growth of bacteria in the holding tank. To cope with this problem, water for the tests was produced by adding 2 g of NaN3 in
a clean container, which was then used for the collection of
10 l of distilled water. Thus, in water with final concentration of
approximately 0.2 g/l NaN3 , no microorganisms were allowed
to grow during its production.
Whey protein isolate powder was employed (Lacprodan DI9224, Arla Foods Ingredients, Denmark). The composition of
this powder, as specified by the manufacturer, is listed in Table 1.
The product used contains whey proteins at a minimum weight
percentage of approximately 90% and small amounts of other
whey constituents. It is considered that the initial typical concentration ratio of whey proteins in milk, remains almost the
same in this whey protein isolate product, comprised of 50%
in -lactoglobulin and 20% in -lactoglobulin [14]. As this
powder is produced by a process that involves heating and spray
drying, it is likely that part of the protein content is denaturated. However, it has been reported [15] that -lactoglobulin
retains its native folding properties and the overall denaturation is probably low, under conditions used to produce isolate
powders. The protein isolate powder was dissolved in distilled
water by gently magnetically stirring the solution for 1 h in room
temperature (25 C), prior to filtration experiments. Solution
pH was 6.67 0.02. Conductivity was 5 1 S/cm for distilled
water, 285 10 S/cm for distilled water with NaN3 addition
and 410 20 S/cm for the final solution. Protein concentration was measured using a Hitachi U-3200 spectrophotometer,
through UV light absorbance at wave length 280 nm.
Dynamic light scattering measurements (DLS) were performed on the protein solution using a Brookhaven BI-9000
system. Various angles of scattered light measurements were
employed in order to better identify and size particles in the
solution. Fig. 1 provides information on the colloidal protein
size distribution, which appears to be trimodal with three groups
of particles roughly corresponding to mean apparent diameters
of 5, 50 and 500 nm. The 5 nm mean diameter may represent the size of -lactoglobulin dimers (most common form
of -lactoglobulin protein in neutral pH solution), whereas the
characteristic diameters of 50 and 500 nm are, most likely,
Table 1
Whey protein isolate powder composition, specified by manufacturer
Lacprodan DI-9224

Specifications

Protein (N 6.38) (min.)


Lactose (max.)
Fat (max.)
Ash (max.)
Moisture (max.)
Total plate count (max.)

88%
0.2%
0.2%
4.5%
6.0%
1000 g1

126

S.A. Mourouzidis-Mourouzis, A.J. Karabelas / Journal of Membrane Science 282 (2006) 124132

Fig. 1. Typical data of apparent particle diameter in protein feed solution and
permeate, using dynamic light scattering (DLS measurement angle 60 ).

protein aggregates present in the solution. Data on colloidal particle sizes of permeate are also included in Fig. 1 and will be
discussed in subsequent Section 3.4. It will be noted that the
peaks in Fig. 1 designate the presence of classes of particles
but not their proportion in the distribution, since light intensity
greatly depends on the size of the particles. Solution -potential
measurements were made using a Brookhaven -sizer and found
to be 24 mV at the solution pH 6.7.
2.2. Membranes
Disc-shaped ceramic membranes were used in the experiments (DisRAM INSIDE, Tami Industries, France). These

membrane discs, of diameter 47 mm, consist of an alumina/titania/zirconia support and a ZrO2 /TiO2 active layer; a
mean pore diameter of 0.8 m, measured by porometry method,
is reported by the manufacturer. However, energy dispersive Xray spectroscopy (EDS) of a few samples (from membranes
employed in this work), showed no significant presence of Zr at
both the support and the active layer of membranes. Additionally, analyzing scanning electron microscopy (SEM) images of
these membranes, by means of suitable software (Adobe Photoshop 7.0), the mean pore diameter of the active layer was
estimated to be 0.75 m.
SEM images (Fig. 2) reveal the structure of these membranes
which are comprised mainly of irregularly sized prismatic particles, thus forming a network of highly interconnected, apparently isotropic voids (pores). The same pattern but with much
bigger particles is observed for the support layer (Fig. 2a and
c). It is expected that the morphological properties of this membrane, characterized as isotropic and asymmetric, will have a
significant influence on filtration and fouling mechanisms examined here. The thickness of the active layer is determined by
image analysis to be approximately 25 m, whereas the total
thickness of the membrane disc is 3 mm.

2.3. Filtration experiments


Filtration experiments are carried out at room temperature
(25 C), using a stirred cell apparatus. The latter is made of
a 15 cm long tubular Plexiglas section of 4 cm inner diameter.
The effective diameter of the membrane disc exposed to liq-

Fig. 2. SEM pictures: (a) membrane cross-sectional view, magnification 350; membrane top view of (b) active layer and (c) support, magnification 7000.

S.A. Mourouzidis-Mourouzis, A.J. Karabelas / Journal of Membrane Science 282 (2006) 124132

127

Fig. 3. Experimental set-up for filtration and backwashing: (1) membrane; (2) filtration cell; (3) feed solution tank; (4) clean water tank; (5) pressurized N2 ; (6)
computer; (7) balance and permeate collection container; (8) pressure transducer (measures in-cell pressure); (9) three-way valve (to direct clean water flow either
for feeding or backwashing); (10) three-way valve (for feed selection between clean water and protein solution); (11) three-way valve (to allow permeate flow or
backwash feed); (12) valve (to allow cell depressurization or overflow exit during backwashing).

uid is 40 mm, corresponding to an active membrane area of


approximately 12.5 cm2 . This cell has a magnetic rod rotating
above (but not in contact with) the membrane surface in order
to homogenize the solution. A schematic diagram of the experimental set-up is shown in Fig. 3. The feed solution is placed
in a Plexiglas container, which is connected to the top of the
cell and continuously stirred. Pressurized N2 is used to feed the
solution at the desired pressure into the cell. The pressure is measured by an Omega PX181 transducer. Filtrate is collected and
weighed using a Mettler-Toledo PB3001-S balance connected
to a personal computer.
Pressure and permeate mass are monitored and recorded,
using a personal computer, by means of ADVANTEC Genie software. Membrane backwashing is effected using distilled water
with NaN3 from another container, by manually redirecting the
N2 pressure to that container and by depressurizing the cell with
a set of valves. The water flow is then reversed, entering the cell
from the bottom.
The system is almost instantly pressurized to the desired
value. Prior to each protein filtration series, clean water (distilled water plus NaN3 ) permeate flow is measured in order to
determine the degree of membrane cleaning. A series of five
successive protein filtration cycles (of 30 min duration each) is
performed, at constant pressure; in-between each of these cycles,
a backwash is carried out, for 5 min at 15 psi pressure. After each
cycle, the protein concentration of the collected filtrate, as well
as the volume and concentration of solution remaining in the
cell are measured. The filtration cell is emptied and fresh protein solution is fed, prior to each filtration cycle.
The choice of cleaning procedure was made based on published work, dealing with WPC foulants on ceramic membranes
[16]. This work provides evidence that 50 C, zero transmembrane pressure and alkali/acid treatment sequence, is the most
effective cleaning procedure. Also based on recent work regarding use of ultrasound in membrane cleaning [17], the following

novel combined cleaning procedure is proposed. After a series


of filtration cycles, the membrane is immersed in an ultrasonic
bath, heated at 50 C, in the presence of 20 g/l NaOH for 30 min
and then it is backwashed with distilled waterNaN3 in the
cell, under 10 psi transmembrane pressure, until a neutral pH is
achieved. The same procedure is followed again in the presence
of 5 ml/l HNO3 for 15 min and backwashing until neutrality.
Before each series of filtration cycles, the permeation flux of
clean water (distilled water with the addition of NaN3 ), is measured and found to be fairly constant (varying between 170 and
190 lmh/psi g). These clean water runs are necessary to determine the state of cleanliness of the membrane prior to each
experiment and to obtain a representative reference flux (clean
water flux).
3. Results and discussion
3.1. Initial ux
Using permeate volume data, the flux J is determined as usual
by
J=

dV
dtA

(1)

where A is the active membrane area. Fig. 4 provides an overview


of a series of filtration cycles, with intermediate backwashing.
In Fig. 5 the maxima and minima of each cycle are plotted to
facilitate data interpretation. The initial permeate flux value after
each backwash cycle, compared to initial flux of the first cycle,
provides a measure of membrane irreversible fouling. Indeed,
the difference between the initial permeate flux at the beginning
of a series of filtration cycles (i.e. before any fouling takes place)
and the initial permeate flux of each subsequent filtration cycle
after a backwash, represents the flux reduction due to irreversible
fouling. This type of fouling is apparently related to the protein

128

S.A. Mourouzidis-Mourouzis, A.J. Karabelas / Journal of Membrane Science 282 (2006) 124132

eral other series of tests, made under the same conditions, despite
some minor scatter of peak values, show exactly the same trends
displayed in Fig. 4.
3.2. Final ux

Fig. 4. Normalized permeate flux during filtration of whey protein isolate solution (4 g/l), with intermediate backwashing, at P = 2.5, 5, 7.5 and 10 psi. Filtration steps are 30 min long, with backwashing period 5 min long, at P = 15 psi.

mass deposited on and/or within the membrane, which is not


removed during backwash. The data of Figs. 4 and 5 show consistently an almost constant value of irreversible fouling at each
pressure, meaning that, during the first filtration cycle of each
series, irreversible fouling of the membrane takes place nearly to
its full extent; i.e., apparently only little irreversible fouling may
take place after the first filtration cycle. Furthermore, it appears
that the higher the imposed pressure, the greater the effect and
the rate of attaining irreversible fouling.
It will be pointed out that only one series of filtration cycles,
with intermediate backwashing, is presented here (Fig. 5). Sev-

The difference of permeate flux values at the beginning and at


the end of each filtration cycle (i.e. the maximum and minimum
values) after the first one, is a measure of reversible fouling of
the membrane, since irreversible fouling does not significantly
increase after the first filtration cycle. The data of Figs. 4 and 5
show that there is a nearly constant degree of reversible fouling for each pressure, throughout all filtration cycles. It is also
observed in Fig. 4 that there is a sharp flux reduction in the first
few minutes of each filtration cycle (especially after the first
one), which is indicative of rather rapid reversible fouling taking
place initially. Moreover, the normalized final (nearly constant)
permeate flux of Fig. 5, indicates the pressure dependence of
the extent of reversible fouling; i.e. with increasing pressure a
greater reduction of J/J0 is observed due to fouling. The effect
of pressure on fouling is also shown, more clearly in Fig. 6a and
b, where J/J0 data are plotted for the first and second filtration
cycle, respectively; it is evident that the higher the pressure, the
greater the effect of fouling. This observation provides an indication of a surface protein layer formation, possibly resulting
in a more compact and/or thicker layer at increased pressure,
which could then lead to reduced flux. More data to clarify this
issue follow.
3.3. Protein rejection and deposition
Rejection can be determined as follows:
R=1

Fig. 5. Initial and final permeate normalized flux values, for each cycle, during
filtration of whey isolate protein solution (4 g/l), with intermediate backwashing, at P = 2.5, 5, 7.5 and 10 psi. Filtration cycles are 30 min long each, with
backwashing period 5 min long, at P = 15 psi.

Cp
Cf

(2)

where Cp and Cf are measured protein concentrations of permeate and feed solution, respectively. The protein amount deposited
on and/or within the membrane is estimated using simple mass
balance calculations; by measuring volume and concentration
of feed, permeate, and retentate (remaining in the cell above the
membrane), one can estimate the protein rejection, as well as the
amount of protein that is trapped within, and/or deposited on,
the membrane. The data in Fig. 7, suggest that after the first filtration cycle, there is a significant increase of rejection (especially
for pressures above 2.5 psi), which may attain an asymptotic
value. This trend seems to be in accord with the previous observations about irreversible fouling taking place practically only
during the first cycle, as well as the almost constant degree
of reversible fouling in successive cycles, which results in a
fairly constant total membrane resistance Rm (accounting for
irreversible, plus reversible fouling, plus clean membrane resistance), at the end of each filtration cycle following the first one.
Fig. 8 shows that the amount of protein which is deposited
on and/or within the membrane is significantly lower for successive filtration cycles, compared to the first one. There is
some scatter in these data, which is attributed less to common
errors of measurements (of concentration or mass) and more

S.A. Mourouzidis-Mourouzis, A.J. Karabelas / Journal of Membrane Science 282 (2006) 124132

129

Fig. 8. Estimated amount of proteins deposited in/on the membrane at the end
of each filtration step.

observed that the amount of deposited protein at each cycle, is


roughly half that of the first one, at each pressure tested; thus,
one may conclude that the amount of protein responsible for
irreversible fouling is comparable to that responsible for the
reversible one, at the present operating conditions.
However, the effect of deposited amount causing reversible
fouling appears to be greater at higher pressures, which along
with the actual protein rejection (Fig. 7), seems to provide
another indication for the presence of a surface fouling layer,
being subject to compaction effects by increased pressure.
3.4. Effect of aggregates on the ltration process

Fig. 6. Normalized permeate flux during: (a) the first cycle filtration and (b) the
second cycle filtration of whey protein isolate solution (4 g/l) for 30 min.

to inaccuracies introduced by the experimental procedures; e.g.


some non-negligible and possibly variable dilution of permeate solution, due to water trapped in the permeate tubing from
the preceding back-washing step. Despite this scatter, it may be

Fig. 7. Estimated protein rejection at each filtration cycle.

Useful information is obtained regarding species causing


fouling, if the permeate from the aforementioned tests is
employed as feed solution in another filtration cycle. Data not
presented here due to space limitations, show that flux decline
is practically eliminated over a significant period of time, if the
prefiltered solution is filtered again through a clean membrane
disc. This prefiltered solution is also examined by DLS to identify the characteristic sizes of remaining particles. The results,
included in Fig. 1, show that the class of large aggregates of the
original solution is apparently completely removed during the
prefiltration step, leaving in the solution two particle classes of
approximate characteristic sizes 1.5 and 38 nm. This observation is consistent with previous work of Kelly and Zydney
[7], who report that protein aggregates are responsible for the
initial flux decline observed and that their removal essentially
eliminates protein fouling. It must be noted that the correlation function obtained in the measurement with permeate is not
very satisfactory; thus these DLS data are considered helpful
in identifying the dominant particle classes present, rather than
their accurate size distribution. The reason for this inaccuracy
may be attributed to the increased polydispersity factor after the
prefiltration process, possibly due to shear induced breakage of
large aggregates originally present in solution.
Kelly and Zydney [7], also report that flux reduction during microfiltration of protein solutions by means of a 0.22 m
polyvinylidene fluoride membrane, is much greater in both rate
and extent for -lactoglobulin compared to other milk proteins.

130

S.A. Mourouzidis-Mourouzis, A.J. Karabelas / Journal of Membrane Science 282 (2006) 124132

This suggests what is already known for fouling of plate heat


exchangers in the milk industry [8], i.e. that the main foulant
during milk processing is -lactoglobulin apparently due to
its very active free sulfhydryl group. Also well known is the
increased self-association tendency of -lactoglobulin, which
leads to dimers and octamers, as well as aggregates of much
greater number of primary species, which are apparently responsible for the observed aggregates in DLS results of this study.
3.5. Fouling mechanism
Total membrane resistance can be determined from the following expression:
Rm =

P
J

(3)

where P is the filtration pressure, the solution viscosity and


J is the permeate flux. To interpret the data in d2 t/dV2 versus
dt/dV graphs, the following equations are used for evaluation of
the respective quantities, in terms of flux J:
dt
1
=
dV
JA

(4)

1 dJ
d t
= 3 2
dV 2
J A dt

(5)

According to Tracey and Davis [18], membrane resistance


versus filtration time is an increasing function with an increasing
slope if a pore blockage or pore constriction fouling mechanism
is dominant; a decreasing slope is observed if cake filtration
dominates (due to building-up of external protein surface layer).
The present data for the first cycle, shown in Fig. 9a, suggest
that the fouling mechanism is not a simple one, possibly dominated by pore blocking initially and by a sort of cake filtration
later on. One also observes that the time at which this change
of mechanism takes place (i.e. the change from increasing to
decreasing slope-inflection point) is longer if pressure is higher.
Additional evidence regarding the fouling mechanism may be
gained from the data of Fig. 9b, indicating that during the second filtration cycle (and in subsequent cycles) the pore blockage
or pore constriction mechanism may not be significant; indeed,
the decreasing slope of those curves suggests that some kind of
surface-layer (cake) formation may take place.
The present data for the first filtration cycle evaluated in terms
of d2 t/dV2 versus dt/dV (Fig. 10), despite the scatter, suggest
that there is a transition taking place (from pore blocking to
cake formation mechanism) at somewhat lower fluxes (higher
dt/dV values) for higher pressures, indicated by the maxima of
d2 t/dV2 . Furthermore, a pseudo-steady state is likely occurring
earlier at lower pressures, possibly due to surface-layer resistance affected less by pressure, in contrast to higher applied
pressures which may lead to higher effective compression of
this layer, thus further lowering flux before a pseudo-steady flux
is achieved. These observations are based on previous work by
Ho and Zydney [11], who employ d2 t/dV2 versus dt/dV graphs to
identify dominant fouling mechanisms in the filtration process.
According to that approach, the large reduction on rate of flux

Fig. 9. Total membrane resistance versus time during first (a) and second (b)
filtration cycle of whey protein solution (4 g/l) using a 0.8 m disc membrane,
at P = 2.5, 5, 7.5, and 10 psi.

(dJ/dt) that occurs during the transition from pore blockage to


cake filtration mechanism, is reflected in the decreasing value
of d2 t/dV2 . Thus, the respective line should keep rising up to the
point where the cake filtration mechanism sets in. Furthermore,
if a steady flux is achieved (i.e. constant permeate flux) the line
would sharply drop to become parallel to the y-axis, and when
cake growth proceeds a zero slope is expected. In contrast to the
above work [11], where 0.2 polycarbonate track-etched membranes and BSA solutions were employed, this study shows that
cake filtration onset is likely taking place at somewhat lower (or
possibly the same) flux with increasing pressure, rather than at

Fig. 10. Flux decline analysis of first filtration cycle at P = 2.5, 5, 7.5, and
10 psi.

S.A. Mourouzidis-Mourouzis, A.J. Karabelas / Journal of Membrane Science 282 (2006) 124132

higher flux with increasing pressure. This observation seems to


lead to the suggestion (for the membranes used), that the transition from dominant pore blocking to dominant cake formation
is likely to occur at roughly the same permeate flux, regardless
of applied pressure.
There is another observation based on the data of
Figs. 6a and 10, which for 2.5 psi show a pronounced difference
of flux decline pattern, during the first few minutes of the filtration experiments (marked by an arrow in Fig. 6a); indeed, the
2.5 psi experiment is quite different in terms of flux decline rate,
which is significantly smaller. This initial flux decline pattern
is considered [12,19,20] typical for isotropic membranes with
interconnected pore structure, as the one employed in this study.
However, higher pressure apparently causes this pattern to disappear and to revert to typical flux decline curves observed in the
case of rather uniform, elongated pores, as those of track etched
membranes. This trend may be attributed to very fast blocking of lateral permeation passages, leading to a pseudo-straight
through pore modification; the same trend is also observed in
the 2.5 psi experiments after the first cycle (Fig. 6b). Thus, irreversible fouling may be associated with pore blockage at all
pressures, taking place faster at higher pressures.
4. Concluding remarks
Experimental evidence suggests that there is a fast initial
pore blocking effect, which causes irreversible fouling, that cannot be eliminated by the applied backwashing; furthermore, it
seems that it takes place during the first few minutes of filtration
and does not increase significantly later on. This effect is more
pronounced at higher filtration pressures. The same trend with
respect to pressure is observed for reversible fouling; relatively
loosely bound proteins inside the pores and protein layer formation on the membrane surface may be the cause for this. It is
clear that both irreversible and reversible fouling significantly
contribute to flux reduction; however, it appears that the relative
influence of irreversible fouling, compared to that of reversible
fouling, tends to increase with increasing pressure (Fig. 5).
Aggregates of measured diameter less than the mean pore
size, can infiltrate the internal membrane structure, possibly
blocking some passages. Taking into consideration the pore size
distribution (which is expected to be relatively wide for ceramic
membranes) it appears that aggregates existing in the solution
and even larger aggregates formed in the membrane by bridging,
which can have sizes up to 1 m [21], can effectively block the
pores.
It is reasonable to assume that protein aggregates can be
firmly packed inside membrane pores, when pressure is relatively high, since flux and convective flow is initially high.
Judging from the higher deposited protein values at lower pressures and the fact that surface protein layer compaction should
be less (thus leading to relatively loose packing), a layer formed
under lower pressure may lead to increased thickness, compared to that at higher pressure. Keeping in mind the previous
observation that reversible fouling is more pronounced at higher
pressures, it may be concluded that cake/layer thickness or the
amount of proteins forming it, may not be the important fac-

131

tor here, but rather that cake compaction and porosity play the
most signicant role. The relatively small differences in protein
rejection cannot be safely used for conclusions, although the
observed somewhat greater rejection at higher pressures tends
to support the above arguments.
The cleaning procedure used was quite effective, resulting in
clean water flux recovery, after each series of tests, in the range of
90100%. It seems that ultrasounds (directed backwards towards
the surface), combined with typical cleaning agents at 50 C
temperature, are effective in unbinding blocking proteins.
The above observations, combined with data showing the
effect of pressure on permeation rate (Figs. 4 and 5), suggest that
due to the fouling mechanism, it is preferable to employ relatively low pressure, at which a protein surface-layer apparently
grows relatively early but may suffer less compaction, compared
to relatively high pressures at which the opposite trends seem to
prevail. It will be added that filtration cycles of longer duration
appear to be needed, in order to determine the possible effect
of compaction and growth of such a surface layer on permeate flux, as there are no other published relevant data regarding
whey protein isolate solution filtration, in ceramic membranes
of relatively large pore size (i.e. above 0.5 m).
Acknowledgement
The authors wish to thank Arla Foods Ingredients, Denmark
for donating the whey protein isolate powder for the tests.
References
[1] L.V. Saboya, J.-L. Maubois, Current developments of microfiltration technology in the dairy industry, Lait 80 (2000) 541553.
[2] A.D. Marshall, P.A. Munro, G. Tragardh, The effect of protein fouling in
microfiltration and ultrafiltration on permeate flux, protein retention and
selectivity: a literature review, Desalination 91 (1993) 65108.
[3] N. Olesen, F. Jensen, Microfiltration. The influence of operation parameters
on the process, Milchwissenschaft 44 (8) (1989) 476479.
[4] M.N. Madec, S. Mejean, J.L. Maubois, Retention of Listeria and
Salmonella cells contaminating skim milk by tangential membrane microfiltration (Bactocatch process), Lait 72 (1992) 327332.
[5] A. Guerra, G. Jonsson, A. Rasmussen, E. waagner Nielsen, D. Edelsten,
Low cross-flow velocity microfiltration of skim milk for removal of bacterial spores, Int. Dairy J. 7 (1997) 849861.
[6] F. Beolchini, F. Veglio, D. Barba, Microfiltration of bovine and ovine milk
for the reduction of microbial content in a tubular membrane: a preliminary
investigation, Desalination 161 (2004) 251258.
[7] S.T. Kelly, A.L. Zydney, Protein fouling during microfiltration: comparative behavior of different model proteins, Biotechnol. Bioeng. 55 (1) (1997)
91100.
[8] T. Nylander, Protein adsorption in relation to solution association and
aggregation, in: Biopolymers at Interfaces, Marcel Dekker, New York,
1998.
[9] C. Herrero, P. Pradanos, J.I. Calvo, F. Tejerina, A. Hernadez, Flux decline in
protein microfiltration: influence of operative parameters, J. Colloid Interface Sci. 187 (1997) 344351.
[10] C. Velasco, M. Ouammou, J.I. Calvo, A. Hernadez, Protein fouling in
microfiltration: deposition mechanism as a function of pressure for different pH, J. Membr. Sci. 266 (2003) 148152.
[11] C.-C. Ho, A.L. Zydney, A combined pore blockage and cake filtration
model for protein fouling during microfiltration, J. Colloid Interface Sci.
232 (2000) 389399.

132

S.A. Mourouzidis-Mourouzis, A.J. Karabelas / Journal of Membrane Science 282 (2006) 124132

[12] C.-C. Ho, A.L. Zydney, Protein fouling of asymmetric and composite
microfiltration membranes, Ind. Eng. Chem. Res. 40 (2001) 14121421.
[13] J.M.K. Timmer, H.C. van der Horst, J.-P. Labbe, Cross-flow microfiltration of -lactoglobulin solutions and the influence of silicates on the flow
resistance, J. Membr. Sci. 136 (1997) 4156.
[14] P. Walstra, R. Jenness, Dairy Chemistry and Physics, Wiley, New York,
1984.
[15] C. Holt, D. McPhail, T. Nylander, J. Otte, R.H. Ipsen, R. Bauer, L. Ogendal,
K. Olieman, K.G. de Kruif, J. Leonil, D. Molle, G. Henry, J.-L. Maubois,
M.D. Perez, P. Puyol, M. Calvo, S.M. Bury, G. Kontopidis, I. McNae, L.
Sawyer, L. Ragona, L. Zetta, H. Molinari, B. Klarenbeek, M.J. Jonkman, J.
Moulin, D. Chatterton, Some physico-chemical properties of nine commercial or semi-commercial whey protein concentrates, isolates and fractions,
Int. J. Food Sci. Technol. 34 (1999) 587601.
[16] M. Bartlett, M.R. Bird, J.A. Howell, An experimental study for the development of a qualitative membrane cleaning model, J. Membr. Sci. 105 (1995)
147157.

[17] M.O. Lamminen, H.W. Walker, L.K. Weavers, Mechanisms and factors
influencing the ultrasonic cleaning of particle-fouled ceramic membranes,
J. Membr. Sci. 237 (2004) 213223.
[18] E.M. Tracey, R.H. Davis, Protein fouling of track-etched polycarbonate
micro filtration membranes, J. Colloid Interface Sci. 167 (1994) 104
116.
[19] C.-C. Ho, A.L. Zydney, Effect of membrane morphology on the initial
rate of protein fouling during microfiltration, J. Membr. Sci. 155 (1999)
261275.
[20] C.-C. Ho, A.L. Zydney, Theoretical analysis of the effect of membrane
morphology on fouling during microfiltration, Sep. Sci. Technol. 34 (1999)
2461.
[21] W. Youravong, A.S. Grandison, M.J. Lewis, Effect of hydrodynamic and
physicochemical changes on critical flux of milk protein suspensions, J.
Dairy Res. 69 (2002) 443455.

Vous aimerez peut-être aussi