Understanding Flux Decline

This item was submitted to Loughborough's Research Repository by the author.
Items in Figshare are protected by copyright, with all rights reserved, unless otherwise indicated.
Understanding flux decline in crossflow microfiltration. Part 1 - Effects of

particle and pore size
PLEASE CITE THE PUBLISHED VERSION
PUBLISHER
Elsevier / © The Institution of Chemical Engineers
VERSION
AM (Accepted Manuscript)
LICENCE
CC BY-NC-ND 4.0
REPOSITORY RECORD
Tarleton, E.S., and Richard J. Wakeman. 2009. “Understanding Flux Decline in Crossflow Microfiltration. Part
1 - Effects of Particle and Pore Size”. figshare. https://hdl.handle.net/2134/4843.
This item was submitted to Loughborough’s Institutional Repository
(https://dspace.lboro.ac.uk/) by the author and is made available under the
following Creative Commons Licence conditions.
For the full text of this licence, please go to:

http://creativecommons.org/licenses/by-nc-nd/2.5/
1
UNDERSTANDING FLUX DECLINE IN CROSSFLOW MICROFILTRATION – PART I: THE

EFFECTS OF PARTICLE/PORE SIZE
E.S. Tarleton ([email protected]) and R.J. Wakeman

Separation Processes Centre, University of Exeter, Devon, UK.
ABSTRACT
Experimental data have been obtained from a computer controlled crossflow microfiltration
apparatus. Polymer membranes of differing properties were characterised and used to filter
aqueous suspensions carrying colloidal and fine particles of known shape, size, surface charge
and chemical composition. The characterisation and experimental procedures are described and
sample flux decline curves presented. By changing the size and size distribution of the feed and
the pore size of the membrane in a systematic manner, the importance of the pore/particle size
ratio to fouling is illustrated. The data obtained explain some of the apparent anomalies which
have been presented in previous publications and highlight some important design criteria for
membrane installations.
KEYWORDS
Microfiltration; Membranes; Fouling; Characterisation
INTRODUCTION
The separation of solid/liquid systems has become an important topic in both industry and
academia. The former is driven by ever stricter legislation from governments and international
bodies to manufacture their products with a minimum of damage to the environment whilst
attempting to maintain their own competitiveness in the market place. This need for improved
efficiency (at a realistic cost) has fuelled research not only into existing separation methods but
also provided the impetus to develop new technologies. These technologies, frequently invented
and nurtured in the laboratories of academic institutions, address the growing problem of
processing feed streams, and perhaps more importantly waste streams, containing finer
particulates. The development of the pressure driven membrane processes reverse osmosis,
ultrafiltration and more recently microfiltration, promised to solve the problems inherent with other
separation techniques. Whilst reverse osmosis is now becoming the ‘standard’ method for the
desalination of sea water1, microfiltration, and to a lesser extent ultrafiltration, are still struggling to
become widely accepted as viable processes. Although some apparently successful applications
have been reported it is acknowledged that the fouling problems associated with micro- and ultra-
filtration are poorly understood and prevents the more widespread use of the technologies2,3 –
although it is recognised that ultrafiltration is now becoming more widely accepted in, for example,
the dairy industry.
Crossflow microfiltration was originally perceived as a method of dewatering aqueous suspensions,

with particulates typically in the range 0.05-30 μm, by utilising shear generated forces to prevent
the formation of deposits or cakes at the filtering surface. Despite the use of crossflow velocities
up to 8 m s-1, the shear rates generated at the membrane surfaces are frequently insufficient to
prevent significant fouling by particulates; the rapid decline from the relatively high membrane
clean water fluxes at the start of filtration and the subsequent low quasi-steady flux levels which
follow being the undesirable consequences. Whilst researchers have in the past attempted to
alleviate fouling through, for example the use of imposed force4 and rotational5 fields, many others
have attempted to understand the fundamental reasons for flux declines and fouling mechanisms
in crossflow filtration processes. Experimental research programmes6-12, although setting out with
Cite paper as: Tarleton E.S. and Wakeman R.J., 1993, Understanding flux decline in crossflow microfiltration: Part 1. Effects of
particle/pore size, Trans IChemE, 71(A), 399-410.
2
the best intentions, have often struggled with the need to use ill-characterised, complex feed
streams containing mixtures of molecules and particulates, either or both of which may be
membrane foulants in combination or in their own right. As a consequence the published literature
is often conflicting and confusing to the extent that problems understanding the effects of even
basic parameters such as crossflow velocity and filtration pressure remain unresolved. Theoretical
work13-21, which appears to have been particularly prevalent in the microfiltration literature recently,
has suffered as a result of the conflicting experimental data. The models produced, whilst tackling
an inherently complex problem, seem to be acclaimed successful or otherwise on their ability to
interpret flux decline data and not on the correctness of their description of the particle and deposit
dynamics. The latter has always been an assumption of the chosen model, and has been claimed
to be appropriate if interpretations of the former happen to turn out to be of the right order of
magnitude. Attempts to substantiate the assumptions have usually been by recourse to fresh
experimental observations of particles in wholly different flow and compositional environments (or
in some cases no experimental verification has been offered at all). This ‘black box’ approach to
modelling must be judged in terms of the applicability of the models to real systems. Previous
work22 has shown that even the best of the existing models23, which is perhaps surprisingly an
empirically derived relation between flux and process parameters rather than a more rigorous
model derived from fundamental principles, can only reliably predict flux performance within two
orders of magnitude of the true value.
The above considerations, taking previously published work into account, suggest that there is a
need for a coordinated experimental research programme which examines the fundamental
aspects of membrane fouling. This paper describes the results of part of such a programme were
the matrix of variables which most affect the microfiltration of aqueous suspensions are properly
delineated and investigated using a well controlled filtration apparatus. The characterisation and
experimental procedures used are described in detail and the results of the microfiltration
experiments presented through typical flux decline curves. The extensive experimental work
involved in the research programme necessitates that only the effects of particle size and
membrane pore size on flux decline are described in detail here. The effects of the other process
variables known to influence microfiltration will be related in future papers.
EXPERIMENTAL TECHNIQUES AND PROCEDURES
The experimental programme may be conveniently divided into two parts. The first, involved work
to characterise the properties of the particles, suspensions and membranes and the interfacial
interactions between them with the parameters shown in Table 1 measured using commercially
available laboratory test equipment. The information acquired from the characterisation tests was
subsequently used to aid the analysis of the microfiltration experiments and assess membrane
fouling for the range of process conditions examined.
Membrane Characterisation
The range of organic microfiltration membranes available commercially in a flat sheet form was
initially examined to identify septa with differing structures, material properties and pore size
ratings. Many different membranes were subsequently characterised, the Nuclepore
(polycarbonate, PC) and Sartorius (cellulose nitrate, CN) membranes which were available in a
range of pore size ratings from 0.2 μm to 10 μm being typical examples. The micrographs shown
in Figures 1 and 2 highlight the distinct differences in the structures of these membranes. Nuclear
track etched Nuclepore PC membranes have a structure composed of series of predominantly
cylindrical pores dispersed in what appears to be a random manner. Single, discrete pores are of
an almost uniform diameter through the membrane whilst doublets, triplets and greater
coincidences have a variable shape in both their axial and radial directions. The homogeneous
Sartorius CN membranes are produced by a casting process and thus exhibited quite different
tortuous structures composed of matrices of interconnected pores.
3
The pore size and pore size distributions of the membranes were measured with a Coulter
porometer. The instrument uses compressed air to displace a flourinated hydrocarbon wetting fluid
(trade name Porofil) from the pores of the test specimen. As the air pressure is increased the
wetting fluid is displaced from progressively smaller pores and the total air flow rate through the
membrane can be related to the pore size. Unsoiled samples of the membranes were tested in
accordance with ASTM E1294-8924 and sample results are shown in Figure 3 and Table 2. Whilst
Nuclepore PC membranes exhibit fairly narrow pore size distributions around mean values close to
the manufacturer's quoted ratings, the other membranes tested25,26 demonstrated mean pore sizes
somewhat different from the quoted ratings. For instance a 0.2 μm rated Sartorius CN membrane
has an average mean pore size of 0.51 μm and an 8 μm rated membrane of the same type has
mean and 90% pore sizes of 3.6 μm and 4.2 μm respectively where the ‘90% size’ indicates the
pore size below which 90% of the total possible flow through the membrane occurs. The range of
pore sizes found in the membrane samples widened considerably for the larger pore ratings and
more variation was observed in the 90% than the 10% pore size. This perhaps indicates the
difficulties experienced in manufacturing larger pore size membranes with ‘tight’ size distributions.
In addition to pore size and size distribution, a number of other membrane characteristics were
evaluated and the results are summarised in Table 3. Although full descriptions are not pertinent
to the present work a few points are worth emphasising. There is a clear indication that significant
differences exist in the permeability of ‘large’ and ‘small’ pore size membranes (e.g. ~102 change
between 0.2 μm and 10 μm rated Nuclepore PC membranes) though the reasons for the
sometimes large difference between measured and manufacturer quoted values are rather less
obvious. The porosities and typical thicknesses quoted for the membranes give an indication of
the primary mechanisms by which removal of solids is achieved during filtration. Track etched
Nuclepore PC membranes rely to a significant extent on a ‘sieving’ process whereas cast Sartorius
CN membranes utilise their tortuous structure to provide more depth filtration when conditions
allow.
Suspension Characterisation
A range of powders and a yeast were identified for use as challenge streams in the microfiltration
experiments. They included the minerals calcite (CaCO3), anatase (TiO2) and china clay, and are
considered to encompass the range of properties required to examine and evaluate the
parameters which most affect the microfiltration of particulate suspensions. These systems were
chosen primarily for their different size, shape and surface charge characteristics, whilst at the
same time being representative of ‘real particulates, rather than being idealised such as latex or
silica spheres. All behaved in a Newtonian manner over the region of suspension concentrations
tested.
The size and size distributions of the chosen particle systems dispersed in double distilled water
were evaluated using Malvern Auto- and Master- Sizer laser light scattering instruments and the
results are summarised in Table 4. The powders were of an Analar (or equivalent) grade and used
without further cleaning or washing in both the characterisation and microfiltration experiments.
The smallest mean size of the suspensions tested was exhibited by anatase with most of the
particulates in the region of 0.5 μm. A typical china clay suspension had a larger mean size of 4.4
μm and a correspondingly wider distribution. In contrast the unground calcite had a mean size of
24.3 μm respectively and an apparently very wide size distribution when dispersed in water. When
the calcite was wet ground in a ball mill the mean size was reduced. To obtain a range of mean
sizes the grinding time was varied over the period 0-24 h where the mean size would typically
change by an order of magnitude. The unimodal suspensions obtained through wet grinding were
also combined in the appropriate proportions to give bimodal suspensions of nominally similar
mean sizes but widely differing size distributions such as those shown in Figure 4.
4
The effect of pH on particle surface charge, and hence ζ-potential, was also determined for the
powder dispersions with a Malvern ZetaSizer. The suspension pH was altered with the same
electrolytes used in the filtration tests (HCl and NaOH) and double distilled water was used as the
dispersing medium. The rhomboidal shape calcite particles have a relatively low surface charge in
aqueous suspension with the points of zero and maximum negative ζ-potential occurring at the
pH's 8.9 and 10.6 (ζ = – 20 mV) respectively. Conversely, the tetragonal shape anatase is typical
of a high surface charge material where the zero and peak negative ζ-potentials occur at the pH's
3.9 and 9.0 (ζ = – 60 mV). The platelet shape china clay particles show a steadily increasing
negative ‘average’ ζ-potential with increasing pH (ζ = – 54 mV @ pH = 10) and an isoelectric point
in the pH region 0-2. This, however, belies the complicated distribution of surface charge on china
clay where in some aqueous solutions the ‘faces’ of a china clay particle have a predominantly
negative charge whilst the ‘edges’ exhibit a positive charge29.
Microfiltration Apparatus and Test Procedure
The equipment used to assess membrane fouling behaviour is shown schematically in Figure 5.
The unit comprised a flow circuit in which suspension of a known, essentially constant, composition
was pumped continuously through a crossflow microfilter at a predefined crossflow velocity and
trans-membrane pressure. These desired filtration conditions were maintained by two electrically
operated proportional control valves. The valves were linked to a computer via electronic control
circuitry which monitored conditions at various points within the flow circuit using suitable
transducers. When a deviation from the desired conditions was detected the control valves moved
to compensate and thus maintain a constant filtration pressure and crossflow velocity. The
temperature of the process suspension was also regulated using a plate type heat exchanger
positioned in the feed tank. In this way it was possible to vary the filtration pressure over the range
0-50 psi and the crossflow velocity 0-3 m s-1 whilst keeping the feed at a temperature of 30±2ºC.
The purpose built microfilter was constructed from plastic and stainless steel (as was the rest of
the flow circuit and ancillaries) and consisted of a supported 24 cm2 flat sheet membrane
positioned to form one side of an enclosed rectangular section. During an experiment the process
suspension flowed through the rectangular space thus formed and tangential to the membrane to
generate shear forces at the filtering surface. The filtrate produced during filtration was collected in
a suitable compartment behind the membrane and returned to the feed tank via a flowmeter and
piping.
Prior to the start of a microfiltration test the process suspension, made to a known concentration
between 0.033% v/v and 2% v/v from either a dry powder and double distilled water, from a diluted
slurry (in the case of a wet ground sample) or incubated and grown for a period of days (in the
case of yeast), was stirred in the feed tank for a suitable time to produce an homogeneous mixture.
The suspension pH was then altered if necessary using the electrolytes hydrochloric acid (HCl) or
sodium hydroxide (NaOH) and the resultant mixture stirred continuously. No other dispersants or
flocculants were added. The membrane chosen to achieve the separation was cut to size from a
larger sheet and mounted in the microfilter pre-wetted with double distilled water. The test was
performed at the desired filtration settings with the relevant data permanently stored by the
computer. Each test was run for a maximum of two hours or until the membrane fouled to such an
extent that the flow rate could not be recorded satisfactorily.
The matrix of properties shown in Table 5 were investigated for the range of membranes and feed
suspensions. The sample data shown in Figures 6-12 highlight the effects of particle size, particle
size distribution and membrane pore size on flux decline. Many more similar data were
accumulated that confirm the effects shown.
MICROFILTRATION TEST RESULTS
5
Whilst the parameters shown in Table 5 affected the microfiltration of the suspensions to an extent
dependent on their relative magnitudes, the tests showed almost without exception the same
general characteristics. When the filtration test commenced a sharp fall in filtration rate from the
clean water flux was observed. The rapid fouling of the membrane during this period can be
attributed to the accumulation of particulate matter at or near the filtering surface (i.e. due to
particle ‘polarisation’ or to cake formation), the particulates being drawn toward the filtering surface
by the convective flow of filtrate through the septum. After the rapid initial fouling the rate of flux
decline progressively lessened until a near equilibrium filtration rate was observed in most
experiments. Here the shearing action generated by the crossflowing stream seemingly prevented
further significant deposition of matter at the fouling layer/membrane surfaces.
The Formation of Fouling Layers
When a pseudo-equilibrium flux state is reached in microfiltration it has been suggested that the
particles forming the fouling layer(s) are possibly in a state of ‘dynamic equilibrium’ whereby
particles are leaving and joining the layer(s) at the same rate20. Romero and Davis20 attributed this
to a phenomenon which they called ‘shear induced hydrodynamic self diffusion’, which formed the
basis of much of the theoretical work published in recent years. It is hypothesised that a
concentrated layer, part of which may be mobile and part stationary depending on the flow
conditions in the flow volume above the membrane. The formation of dynamic membranes has
been observed30 in a complimentary experimental programme using a high speed, high
magnification video system capable of operation at recording speeds up to 400 frames per second
and optical resolutions of ~2 μm. Layer build-up was examined for a range of process conditions
whilst monitoring flux performance. When larger particle feeds such as unground calcite were
filtered the majority of the particulates comprising the fouling layers were deposited within a few
seconds of the start of filtration; the resultant, predominantly stationary, layer having a surface
unevenness between one and three particle diameters. At the ‘rough’ surface of the deposit
particles flowing with the free stream (close to the surface) were observed to roll and jostle
between sites along the surface to become permanently lodged or occasionally move back into the
free stream. When finer particle feeds such as ground calcite or anatase were filtered a stationary
layer was again rapidly established after filtration was initiated, however, here no saltation motion
was observed and particles within three (particle) diameters of the deposit surface could be seen to
undergo an erratic rolling motion. In conjunction with this erratic motion a moving ‘front’ of particles
could sometimes be observed to travel at a low velocity in the direction of the crossflow over the
stationary fouling layer already deposited on the membrane. Thus, it would appear that for the
process conditions typically found in ‘real’ microfiltration systems, shear induced hydrodynamic self
diffusion only tenuously describes the phenomena which are observed in practice.
Fouling by particulates appears to be due to two apparently independent mechanisms which occur
simultaneously. The first accounts for the initial sharp decline in permeate flux and is irreversible
for all practical purposes, and due to the rapid deposition and capture of the finer particles from the
suspension and their subsequent penetration into the pores of the membrane. This is a stochastic
process, dependent on localised hydrodynamic conditions close to the pore entrances. The
second mechanism is largely reversible and causes further particulate layer(s) to form above the
membrane surface; the so called ‘dynamic membrane’.
From hereon we shall refer to the first mechanism where particle penetration occurs and which is
largely irreversible as 'fouling by particulates', and to the second which is reversible as 'cake
formation'. The nature of the 'cake' formed in crossflow filtration processes is not to be confused
with that which results from conventional deadend filtration. Unlike a conventional cake which has
its thickness limited by the available pressure and whose limiting thickness is approached as the
available hydraulic pressure is dissipated as a cumulative drag at the surface of the particles as the
fluid passes through the cake, one formed in microfiltration (or ultrafiltration) has its thickness
limited by the shear forces at the free surface. When the sum of the cohesive forces between the
particles and the drag acting on the particles at the free surface, causing the particles to remain in
6
the cake, is equal to the shear forces tending to remove particles from the cake, the cake can be
said to have a pseudo constant thickness. Hence, in deadend filtration the filtrate flux tends to
zero as the limiting thickness of the cake is reached, but in crossflow filtration this flux condition is
frequently not reached and a finite flux can be associated with the limiting thickness. When a gel,
rather than a cake, is formed at the surface of the membrane, then a flux close to zero can be
found when the limiting cake thickness is reached (this is more common in ultrafiltration than it is in
microfiltration).
Effects of Particle Size
Figures 6 and 7 show the typical effects on flux decline of changing the median particle size of a
calcite suspension by wet grinding prior to filtration. As a general rule a reduced particle size in the
feed resulted in lower overall flux levels being recorded. However, the situation is complicated by
a number of factors. With a smaller 50% particle size there was a tendency for the filtrate flux to
be reduced more rapidly at the start of filtration and a pseudo-equilibrium flux to be established
sooner. Thus, the fluxes recorded at longer filtration times for ‘large’ and ‘small’ particle systems
were often similar in magnitude, particularly at higher crossflows (and possibly higher
concentrations) where the flux performance did not necessarily follow the simple ‘smaller size
gives lower flux’ relationship. Crossflow velocity has been shown to affect the filtration of
particulate suspensions in different ways31. For feed suspensions containing a higher proportion of
larger particles, a raised crossflow velocity caused the filtrate flux to fall to lower values – this can
be observed by comparing Figure 6 and 7. Here, the seemingly preferential deposition of finer
particle species at the septum surfaces8,32 introduces a classification effect and the layer(s)
responsible for membrane fouling can have a resistance considerably higher than that which might
be expected from a simplistic approach. As the crossflow velocity is raised the fouling layer is
thought to become composed of progressively finer particle species which form higher resistance
deposits leading to lower filtration rates. With smaller particle systems a raised crossflow has the
opposite effect of improving the filtration rate. At intermediate particle sizes there can exist
situations where increasing the crossflow over the range 0-3 m s-1 does not significantly affect
filtrate flux rates. A higher crossflow velocity will tend to move the fluxes recorded for different
particle size feeds closer together and may thus result in the flux recorded for the smaller particle
size system sometimes being above that measured for the larger (e.g. Figure 7).
Effects of Particle Size Distribution
It seems likely that the fouling layers formed during the filtration of different particle size systems
will be of a different structure and composition. In conventional deadend filtration, where the bulk
suspension flow is normal to the filtering medium, smaller particle systems usually form cakes with
a higher resistance to filtration than large particle systems. For crossflow microfiltration, however,
the situation is not as clear due to the particle classification phenomena described previously.
Such an effect in combination with the lack of true cake formation makes identifying the particles
responsible for the formation and structure of the cake layer more difficult to quantify. Although
previous attempts have been made to measure the size distributions in cake layers8,32, perhaps a
more reliable insight can be gained by testing the filtration performance of suspensions with the
same 50% size but differing size distributions. The data shown in Figure 8 for low concentration
and crossflow were acquired using calcite suspensions with a nominal 50% size of 10 μm (see
Table 4). The flux decline curves for mixes 1 and 2 and mixes 3 and 4 are essentially identical and
exhibit flux levels below that recorded for the experiment performed using the suspension
designated ‘grind 2’. What is interesting to note here is that mixes 1 and 2 are both partly
composed of a suspension with a 50% size of 5.2 μm (grind 3) whilst mixes 3 and 4 both have a
component from a suspension with a 50% size of 2.6 μm (grind 4). These results support the
hypothesis that it is the smaller particles in the feed which are predominantly responsible for the
formation of the fouling layer in or on the membrane.
7
It therefore follows that some size fraction less than the 50% percentile should be used to
characterise the feed as this is more indicative of the fouling potential of the feed. When either the
crossflow velocity or suspension concentration is raised the flux decline curves for the different size
distribution feeds tend to come together at longer filtration times and near identical flux declines
are observed for the suspensions shown in Figure 8 when filtering with a crossflow of 2.3 m s-1 and
a suspension concentration of 0.33% v/v. When the magnitude of either of these two operating
parameters is raised the flux performance achieved is essentially ‘de-sensitised’ to the influences
of particle size distribution and the similar fluxes achieved after a short initial differential period
suggest that the fouling layers have similar properties and are perhaps composed of particles with
similar sizes and size ranges.
The influence of particle fines is perhaps most clearly illustrated in Figure 9. The three sets of data
on the graph show the flux decline curves for calcite suspensions exhibiting 50% sizes of 2.6 μm,
9.9 μm and 24.3 μm respectively. The suspension with a 50% size of 9.9 μm was composed of a
mix of suspensions with median sizes of 2.6 μm and 24.3 μm respectively. The near identical flux
performance between the 2.6 μm and 9.9 μm median size suspensions confirms that only a certain
proportion of the feed, that is the fines, determine the fouling rates. This result was confirmed with
a sequence of tests using mixtures of anatase and calcite suspensions were even the addition of
small amounts of anatase powder to otherwise pure calcite suspensions resulted in significantly
lower flux rates.
Effects of Membrane Pore Size
Membrane pore size was found to affect filtration performance to an extent dependent on a
number of factors. Figure 10 shows the typical effects of filtering a feed suspension with a majority
of particulates significantly larger than the pore sizes present in the membrane septum. Here, 0.2
μm and 10 μm rated Nuclepore PC membranes were used to filter unground calcite suspensions
having 50% sizes in excess of 20 μm. It is immediately obvious that for the given experimental
conditions the fluxes recorded for both membranes are very similar with perhaps only the initial
stages of filtration showing any discernible difference. The result is perhaps surprising when one
considers the permeabilities of the two membranes (see Table 3). The characterisation
experiments showed that there are approximately two orders of magnitude difference in the
permeabilities of 10 μm and 0.2 μm rated Nuclepore PC membranes with intermediate pore size
ratings somewhere between. Thus, in a simplistic sense it would be reasonable to assume that the
use of more permeable membranes would produce improved filtration rates whilst allowing
macromolecules and maybe some fines to pass through the membrane with the filtrate. The fact
that little difference in filtration rates or filtrate clarity were observed with unground calcite
suspensions and various pore size membranes suggests that here it was the fouling layer formed
at the membrane surface during filtration which played the most significant role in determining flux
performance.
When suspensions of finer particulates were filtered using different pore size membranes further
effects were observed (see Figure 11). Whilst in some cases virtually no difference in flux
performance could be detected (in particular for china clay suspensions where the situation is
complicated by particle shape effects) the majority of tests indicated that increasing the membrane
pore size caused a reduced filtration performance both in terms of flux levels and visual quality of
filtrate. Here it seems that particulates were fouling the membrane pores internally, as well as at
the pore throats, to an extent dependent on the overlap between the particle and pore sizes
present in the feed stream and membrane respectively. The filtrate quality usually improved with
filtration time as the deposit on or in the membrane gradually formed and the clarity was often
further enhanced at higher crossflows. In the case of these finer suspensions it seems that
membrane fouling was influenced to a greater extent by the mechanical interactions between the
internal surfaces of the membranes and the particles forming the fouling layers, rather than merely
the formation of surface cake layer.
8
Although the effects of membrane pore size described were observed in the vast majority of tests a
couple of the experiment sequences gave rather curious, but pertinent, results. Figure 12 shows
that for the given experimental conditions a 5 μm rated membrane gave a flux significantly higher
than the 0.2 μm, 1 μm and 10 μm rated membranes. The filtrate produced during the test with the
5 μm membrane also remained cloudy throughout the course of the filtration whilst the other three
tests all produced visually clear filtrates; in some cases after a few minutes filtration. That these
data could be repeated using different batches of membranes (i.e. result could not be explained by
faulty samples or the presence of pin-holes in the membrane) is curious, however, similar types of
behaviour were observed for a sequence of data obtained from experiments with yeast and have
been occasionally reported in ultrafiltration literature33. It would seem that for certain combinations
of process conditions and pore to particle size ratios some membranes will not produce a high
rejection when comparisons between the particle and pore sizes present in the challenge stream
and membrane respectively suggest they should.
DISCUSSION
A recent review of crossflow microfiltration of cell suspensions has been presented by Lojkine et
al.34. It is confirmed by that review that particle size effects and particle-pore interactions have not
been the subject of systematic experimental studies taking into account all the main variables in
the process, but what observations have been made are from investigations of a limited range of
parameters. Nonetheless, some important observations have been made, and it is noted that in
general the mean size of the particles found in the cake layer is less than that of the feed
particles6,32,36-38; this is confirmed by the present work. Fischer et al.36, filtering calcium carbonate
slurries, also found that the size of the particles in the cake layer decreased as the crossflow
velocity increased, leading to a reduction of the permeate flux with increased crossflow velocity
even though the cake thickness was decreased. Contradictory results also exist in the literature.
For example, Sims et al.39 found that flux increased when the particle size of Aspergillus niger
suspensions was reduced from 20-850 μm to 20-150 μm. It is conceivable that this is due to a
reduction of the packing density in the cake layer, as it is well known that a wider size distribution
will lead to a tighter packing density40 and a higher cake resistance41. It has also been noted that
permeate flux increases with particle size when filtering kaolin slurries42.
The extent of particulate fouling in crossflow microfiltration is related to a matrix of feed stream,
membrane and process parameters. The microfiltration experiments showed that among the more
important of these are the particle size of the feed and the membrane pore size. When the particle
size which is representative of the fouling potential of the feed stream is large in comparison with a
‘representative’ membrane pore size, cake formation by particulates is the predominant
mechanism causing flux decline. When the converse is true, however, and the pore size is large in
comparison to the particle size, pore penetration by particulates occurs to a degree dependent on
the overlap between the size distributions of the solids in the feed and the pores in the membrane.
Fouling in this latter instance occurs through a combination of surface and internal pore blockage.
The fouling and cake layer formed during crossflow filtration has previously been recognised as
being an important factor in determining flux decline35. It has been proposed that three distinct
classes of cake (referred to as the dynamic membrane by Murkes et al.35) can be identified, with
the ratio of a mean particle size in the feed to the mean pore size in the membrane determining the
type of ‘dynamic membrane’ formed (Table 6). The classification reproduced in Table 6 is quite
self-evident when the particle and pore size variations are both monosized and have no variance.
However, in reality the distinction between classifications is not clear-cut as both particle and pore
sizes possess associated distribution densities. This was in fact the case in the present work.
Nonetheless, the following general observations can be made: (i) the formation of the cake layer
on the membrane occurs rapidly when filtration is initiated, and (ii) once formed the cake acts as
the filtering medium which rejects the remainder of the dispersed phase(s). The degree to which
any of Murkes35 proposed dynamic membrane types forms in a test is unclear, however, it is
9
reasonable to suppose that the filtrations performed with finer particle suspensions involved a
significant degree of internal pore blocking (i.e. fouling) by particulates, while experiments with
coarser feed streams may have shown more pore bridging and plugging phenomena. Although
these terms say nothing quantitative (or even qualitative) about the static or dynamic nature of any
cake or fouling layer, similar effects of pore and particle size could be demonstrated with a wide
range of particle systems and membrane types used in this work. This indicates that it is not
membrane structure that is of prime importance in crossflow microfiltration but rather the
interaction between the larger membrane pores and the particle fines in the feed. The relation
between these two parameters can determine the degree of internal, largely irreversible,
membrane fouling and pore penetration found in any given filtration and points to the need for a
clearer definition of the term ‘particle fines’. In the context of crossflow microfiltration, particle fines
may be considered to be those particulates forming the fouling layer(s) which have sizes similar to
or smaller than those of the pores in the filtering surface. The extent to which these particulates
influence the flux performance and filtrate clarity is dependent on their number relative to the
number of pore entrances in the membrane. When more concentrated suspensions are filtered,
there are likely to be more particles (on average) challenging each pore entrance with the result
that most deposition will occur at the pore throats near the membrane surface in contact with the
feed stream. If more dilute suspensions are considered there will be a lower number of
particulates approaching each pore with the consequence that particles have an enhanced
probability of entering the pores to cause internal blockage.
McDonough et al.43 carried out experiments with colloidal silica (mean sizes 0.024 μm and 0.046
μm) and polystyrene lattices (0.19 μm and 0.58 μm), and found that filtrate (or permeate) flux first
decreased and then increased with particle size (the minimum flux was obtained with 0.046 μm
spheres). They attributed this behaviour to electrical and double layer interactions – these
parameters also formed part of this study and will be discussed in Part 2 of this series of papers.
The hypothesis advanced in the preceding paragraphs is equally as applicable to McDonough’s
data as it is to the data presented here, and when combined with electrokinetic arguments may
provide a more complete explanation of his observations.
The shear induced classification of particulates at the membrane surface means that the median
particle size of the feed stream is a poor indicator of fouling potential. It would be more appropriate
to specify some other size percentile as a representative measure and it is proposed here that the
10% percentile of the cumulative undersize distribution of the particle sizes in the feed should be
used. The microfiltration experiments pointed to this. The 10 percentile particle size, together with
the 90 percentile of the cumulative undersize number distribution of the pore sizes in the
membrane, are considered to provide more reasonable estimates of fouling potential. By utilising
such an argument, the data presented in Figures 6-12 (and other similar collections of data) can be
illustrated in a way which emphasises the importance of the relationship between pore and particle
size in microfiltration. For a data sequence, where for instance membrane pore size is varied and
all other experimental conditions remain constant, it is possible to construct a plot similar to Figure
13. Here, the ratio of the 90% pore to 10% particle sizes are plotted against filtrate flux after 2 h
filtration at three different crossflow velocities. When the pore to particle size ratio << 1, particles
which are relatively large compared to the pores in the septum form predominantly cake layers on
the membrane surface. As the ratio approaches one challenging particles will tend to lodge in the
pore throats and flux levels will fall. For ratios >> 1 particles approaching the pores are more likely
to be trapped internally such that irreversible fouling and lower flux rates ensue.
The selection of a membrane for a given duty must be performed with care. It would seem prudent
to attempt to exclude particulates from the internal pore passages of the membrane to enhance
both flux performance and rejection. When the membrane is selected such that most of the
particles in the feed stream are large in comparison to the membrane pores, cake formation will
predominate and provided the membrane permeability is not too low the structure of this deposit,
and not the membrane permeability, will control the flux levels achieved. If the membrane is
chosen such that there is a significant overlap in the larger membrane pore and finer particle sizes
10
then more severe fouling is likely to occur. Furthermore, should membrane regeneration be
required and cleaning procedures such as backflushing are employed, the removal of particulates
from membrane pores will prove more troublesome than the removal of surface deposits.
CONCLUSIONS
The descriptions given in this paper highlight some of the major trends observed in the
characterisation and microfiltration experiments. Although the general effects of particle and pore
size and their inter-relations are given in Table 7, these statements only summarise part of an
integrated series of projects relating to membrane filtration, aimed at providing fundamental
understanding and technical solutions to the fouling problems associated with particulates. The
remainder of the results concerning particulate fouling will be published in future pares and it is
envisaged that, together with a more appropriate model, a new approach to the design and sizing
of crossflow microfilters will be produced. Few of the data considered in this paper are
independent of each other in any ’real’ filtration problem and the task of developing a global model
to account for the interactions of all the basic parameters (as well as any chemical changes at the
membrane/feed interface(s)) is daunting. Until there is a fundamental understanding of such
interactions and their contributions to the fouling process, no complete picture of membrane
filtration will be obtained.
REFERENCES
1. K. Wangnick, 1991, 1990 Worldwide desalting plants inventory, the development of the
desalination market, Proc. 12th International Conference on Desalination and Water Re-use,
pp.19-37 (Vol. 1), Valetta, Malta, April.
2. T.. Finnigan, R. Shackelton and P. Skudder, 1987, Filtration of beer and recovery of extract
from brewery tank bottoms using ceramic microfiltration, Proc. FILTECH Conference, pp.533-
543, The Filtration Society, Utrecht, Holland, September.
3. V. Milisic and R. Ben Aim, 1986, Developing a better understanding of cross-flow

microfiltration, Filt. and Sep., 23(1), 28.
4. R.J. Wakeman and E.S. Tarleton, 1991, An experimental study of electroacoustic

microfiltration, Trans. IChemE, 69(A), 386.
5. R.G. Holdich and G.M. Zhang, 1992, Crossflow microfiltration incorporating rotational fluid flow,
Trans. IChemE, 70(A), 527.
6. R.J. Baker, A.G. Fane, C.J.D. Fell and B.H. Yoo, 1985, Factors affecting flux in crossflow
filtration, Desalination, 53, 81.
7. A.G. Fane, 1986, Ultrafiltration: Factors influencing flux and rejection, in Progress in Filtration
and Separation (Vol. 4), R.J. Wakeman (Ed.), pp.101-179, Elsevier, Amsterdam.
8. E. Fischer and J. Raasch, 1986, Model tests of the particle deposition at the filter medium in
cross-flow filtration, Proc. 4th World Filtration Congress, pp.11.11-11.17, KVIV, Ostend,
Belgium, April.
9. W. B. Bedwell, S.F. Yates, I.M. Brubaker and S. Uban, 1988, Crossflow microfiltration - Fouling
mechanism studies, Sep. Sci. Tech., 23(6-7), 531.
11
10. P. Aimar, J.A. Howell and N.M. Turner, 1989, Effects of concentration boundary layer
development on the flux limitations in ultrafiltration, Chem. Eng. Res. Des., 67(3), 255.
11. I.W. Cumming, R.G Holdich and M. Freund, 1991, The filtration of slurries using crossflow,
Proc. FILTECH Conference, pp.41-50, The Filtration Society, Karlsruhe, Germany, October.
12. S. Ilias and R. Govind, 1993, A study on concentration polarisation in ultrafiltration, Sep. Sci.
Tech., 28(1-3), 361.
13. A.L. Zydney and C.K. Colton, 1986, A concentration polarisation model for the filtrate flux in
crossflow microfiltration of particulate suspensions, Chem. Eng. Commun., 47, 1.
14. R.H. Davis and D.T. Leighton, 1987, Shear-induced transport of a particle layer along a porous
wall, Chem. Eng. Sci., 42(2), 275.
15. C.A. Romero and R.H. Davis, 1988, Global model of crossflow microfiltration based on
hydrodynamic particle diffusion, J. Mem. Sci., 39, 157.
16. R. Rautenbach and G. Schock, 1988, Ultrafiltration of macromolecular solutions and cross-flow
microfiltration of colloidal suspensions: A contribution to permeate flux calculations, J. Mem.
Sci., 36, 231.
17. R.E. Lebrun, C.R. Bouchard, A.L. Rollin, T. Matsuura and S. Sourirajan, 1989, Computer
simulation of membrane separation processes, Chem. Eng. Sci., 44(2), 313.
18. D.A. Drew, 1990, Fouling layer growth and distribution at the interface of pressure-driven
membranes, Chem. Eng. Sci., 45(5), 1223.
19. R.H. Davis and J.D. Sherwood, 1990, A similarity solution for steady-state crossflow
microfiltration, Chem. Eng. Sci., 45(11), 3202.
20. C.A. Romero and R.H. Davis, 1991, Experimental verification of the shear-induced
hydrodynamic model of crossflow microfiltration, J. Mem. Sci., 62, 249.
21. S. Kimura, 1992, Transport properties in membrane separation processes, J. Chem. Eng.
Japan, 25(5), 469.
22. C. Smith, R.J. Wakeman and E.S. Tarleton, 1991, The relevance of fouling models to crossflow
microfiltration, Proc. FILTECH conference, pp.51-64, The Filtration Society, Karlsruhe,
Germany, October.
23. M.R. Hoogland, C.J.D. Fell, A.G. Fane and D.A.R. Jones, 1990, The optimum design of
crossflow filtration elements for mineral processing, Proc. 5th World Filtration Congress, pp.604-
610, Sociètè Française de Filtration, Nice, France.
24. American Standard Test Method, E1294, 1989.
25. R.J. Wakeman and E.S. Tarleton, 1992, Membrane characterisation: The need for a standard,
Process Industry J., July, 16.
26. E.S. Tarleton and R.J. Wakeman, 1991, Factors affecting the fouling of membranes in
crossflow microfiltration, Proc. FILTECH Conference, pp.29-40, The Filtration Society,
Karlsruhe, Germany, October.
12
27. J.A. Ibanez, J. Forte, A. Hernandez and F. Tejerina, 1988, Streaming potential and
phenomenological coefficients in Nuclepore membranes, J. Mem. Sci., 36, 45.
28. W.R. Bowen and R.A. Clark, 1984, Electro-osmosis at microporous membranes and the
determination of zeta-potential, J. Colloid Interface Sci., 97(2), 401.
29. H. van-Olphen, 1963, An Introduction to Clay Colloid Chemistry, Wiley, New York.
30. R.J. Wakeman and C. Smith, 1992, Visualisation of particulate foulant layer formation in
crossflow filtration processes, Proc. European Symposium on Solid/Liquid Separation, Paper
L2, GVC-VDI, Cologne, November.
31. R.J. Wakeman and E.S. Tarleton, 1991, Colloidal fouling of microfiltration membranes during
the treatment of aqueous feed streams, Desalination, 83, 35.
32. W-M. Lu and S-C. Ju, 1989, Selective particle deposition in crossflow filtration, Sep. Sci. Tech.,
24(7-8), 517, 1989.
33. J.F. Scamehorn and J.H. Harwell, 1988, Surfactants in chemical/process engineering, in
Surfactant Science Series Vol. 28, D.T. Wasan, M.E. Ginn and D.O. Shah (Eds.), Marcel
Dekker, New York.
34. M.J. Lokjine, R.W. Field and J.A. Howell, 1992, Crossflow filtration of cell suspensions: A
review of models with emphasis on particle size effects, Trans. IChemE, 70(C), 149.
35. J. Murkes and C.G. Carlsson, 1988, Crossflow Filtration, pp.11-18, Wiley, Chichester.
36. E. Fischer and Raasch J., 1985, Crossflow filtration, German Chemical Eng., 8, 211.
37. B. Reismeier, K.H. Kroner and M.R. Kula, 1989, Tangential filtration of microbial suspensions:
Filtration resistance and model development, J. Biotechnol., 12, 153.
38. A. Rushton and G.S. Zhang, 1988, Rotary microporous filtration, Desalination, 70, 379.
39. K.A. Sims and M. Cheryan, 1986, Crossflow microfiltration of Aspergillus niger fermentation
broth, Biotechnol. Bioeng. Symp., 17, 495.
40. R.J. Wakeman, 1974, Packing densities of particles with log-normal size distributions, Powder
Technol., 11, 297.
41. R.J. Wakeman, M.N. Sabri and E.S. Tarleton, 1991, Factors affecting the formation and
properties of wet compacts, Powder Technol., 65(1-3), 283.
42. S. Vigneswaren and J.R. Paudey, 1988, Assessing the suitability of membranes for
microfiltration – An empirical approach, Filt. and Sep., 25(4), 253.
43. R.M. McDonough, K. Welsch, A.G. Fane and C.J.D. Fell, 1988, Flux and rejection in the
ultrafiltration of colloids, Desalination, 70, 251.
ACKNOWLEDGEMENTS
The authors wish to acknowledge receipt of a research award from BP International and helpful
technical discussions with the staff at the BP Research Centre, and the award of a grant from the
13
Science and Engineering Research Council under the auspices of the Specially Promoted
Programme in Particle Technology to fund a part of the work presented in this paper.
14
FIGURES AND TABLES
Figure 1: The topography and cross-section of a 1 μm rated Nuclepore PC membrane.
Figure 2: The topography of a 1.2 μm rated Sartorius CN membrane.
15
Figure 3: Pore size distributions of the Nuclepore PC membranes.
Figure 4: Size distributions of the calcite suspensions with a mean particle size of 10.1±0.18 μm.
16
Figure 5: Schematic representation of the microfiltration flow circuit.
Figure 6: Effect of particle size on flux decline for calcite suspensions at lower crossflow velocity.
17
Figure 7: Effect of particle size on flux decline for calcite suspensions at higher crossflow velocity.
Figure 8: Effect of particle size distribution on flux decline for calcite suspensions at low crossflow
and concentration.
18
Figure 9: Effect of particle fines on flux decline for calcite suspensions.
Figure 10: Effect of membrane pore size on flux decline for unground calcite suspensions.
19
Figure 11: Effect of membrane pore size on flux decline for anatase suspensions.
Figure 12: An effect of membrane pore size on flux decline for china clay suspensions.
20
Figure 13: Influence of pore/particle size ratio on filtrate flux for calcite (grind 4) suspensions and
Nuclepore PC membranes.
21
Particle/process stream Membrane

Particle size Pore size
Particle size distribution Pore size distribution
Particle shape Pore shape
pH, ζ-potential relationship Asymmetry
Rheology Permeability
Compression characteristics Wettability
Table 1: Parameters evaluated in the characterisation experiments.
Membrane type Rating† 90% size variation 50% size variation 10% size variation
(μm) (μm) (μm) (μm)
Nuclepore PC 10 11.70 (+3.11/-0.73) 10.26 (+0.65/-0.27) 8.99 (+0.33/-0.29)
Nuclepore PC 5 6.00 (+0.12/-0.12) 5.56 (+0.10/-0.11) 5.03 (+0.12/-0.10)
Nuclepore PC 1 1.32 (+0.02/-0.03) 1.26 (+0.02/-0.04) 1.17 (+0.02/-0.03)
Nuclepore PC 0.2 0.364 (+0.007/-0.008) 0.351 (+0.008/-0.007) 0.323 (+0.008/-0.006)
Sartorius CN 8 4.15 3.63 3.24
Sartorius CN 5 3.51 3.15 2.82
Sartorius CN 1.2 1.66 (+0.05/-0.03) 1.45 (+0.03/-0.02) 1.33 (+0.04/-0.04)
Sartorius CN 0.2 0.58 (+0.05/-0.03) 0.51 (+0.04/-0.02) 0.46 (+0.02/-0.02)
†
manufacturers designation
Table 2: Average pore size and variations in ten samples taken from 293 mm membrane circles.
Property Nuclepore PC Sartorius CN

Ratings (μm) 0.2-10 0.2-8
Pore size variation with pH not tested† none for pH's 2.3-11.5
Measured permeability (m2) 2.7-110 x10-16 5-120 x10-15
Stated permeability‡ (m2) 4.8-600 x10-16 4.7-180 x10-15
-2 5
Pore density (pores cm ) 3000-1x10 not applicable
ζ-potential -ve for pH's 2-11 27 -ve for pH's 2-12 28
Typical porosity (%) 10 75
Typical thickness (μm) 10 135
Wettability (contact angle) hydrophilic* (53º) hydrophilic (31º)
†
due to design limitation of porometer, ‡by manufacturer for water permeation
*
known to be coated during manufacture
Table 3: Miscellaneous membrane characterisation data.
22
Solids Suspension pretreatment 10% size 50% size 90% size

(and designated name) (μm) (μm) (μm)
calcite unground 11.03 24.31 45.59
calcite wet ground 20 mins (grind 1) 5.17 17.10 36.35
calcite wet ground 1.5 h (grind 2) 1.86 9.97 23.22
calcite wet ground 5 h (grind 3) 1.01 5.23 13.12
calcite wet ground 24 h (grind 4) 0.77 2.52 5.68
calcite unground + grind 3 (mix 1) 2.49 10.15 31.53
calcite grind 1 + grind 3 (mix 2) 1.96 10.16 28.99
calcite unground + grind 4 (mix 3) 1.31 9.93 38.32
calcite grind 1 + grind 4 (mix 4) 1.28 10.29 31.90
anatase - 0.20 0.54 1.49
china clay - 0.79 4.42 14.24
Table 4: Particle size data for powders dispersed in double distilled water.
Feed stream Membrane Process parameters

Particle size Pore size Solids concentration
Particle size distribution Pore shape Crossflow velocity
Particle shape Pore asymmetry Filtration pressure
pH (surface charge) Surface charge
Polymer type
Table 5: Matrix of properties examined in the microfiltration experiments.
Particle (x) and pore Comments

(dp) sizes
x > dp A ‘dynamic membrane’ forms on the membrane surface with
little pore penetration by particulates (designated Class 1).
x < dp Pore penetration by particulates occurs here to cause internal
fouling of the membrane. As the foulant layers accumulate so
the rejection of the membrane improves (Class 2).
x ≈ dp With similar particle and pore sizes the ‘dynamic membrane’ is
formed both in the pores due to penetration and around the
pore throats (Class 3).
Table 6: The proposed classification of dynamic membrane types35.
23
Property Comment
Particle size At smaller particle sizes filtrate fluxes were generally lower
and equilibrium was established more rapidly. The presence
of even a small percentage of fines significantly lowered flux
rates. At higher crossflows and longer filtration times similar
fluxes were often observed for a range of suspensions with
differing median size.
Size distribution Influence most pronounced at low crossflow and
concentration where feeds containing the greater proportion
of fines gave the lower filtration rates. At higher crossflows
and concentrations, where the effective number of particles
challenging each pore was increased, the effects on flux
performance were negligible.
Membrane pore size Little influence on flux or rejection when the majority of the
particles in the feed were significantly larger than the pores in
membrane. Filtrate quality and flux levels often worse when a
significant proportion of the particles in the feed were close to
or smaller than the membrane pores. If the pore sizes in the
membrane were much larger than the particles in the feed
stream, flux rates improved to higher levels although solids
rejection was very poor. In a minority of tests poor rejection
was observed when comparisons of pore/particle size
suggested that filtrate clarity should be good.
Table 7: Summary of the influence of particle/pore size in microfiltration.

Understanding Flux Decline

Uploaded by

Copyright:

Available Formats

Understanding Flux Decline

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Understanding Flux Decline

Uploaded by

Copyright:

Available Formats

This item was submitted to Loughborough's Research Repository by the author.

Understanding flux decline in crossflow microfiltration. Part 1 - Effects of

Elsevier / © The Institution of Chemical Engineers

For the full text of this licence, please go to:

UNDERSTANDING FLUX DECLINE IN CROSSFLOW MICROFILTRATION – PART I: THE

E.S. Tarleton ([email protected]) and R.J. Wakeman

Microfiltration; Membranes; Fouling; Characterisation

Crossflow microfiltration was originally perceived as a method of dewatering aqueous suspensions,

EXPERIMENTAL TECHNIQUES AND PROCEDURES

Microfiltration Apparatus and Test Procedure

MICROFILTRATION TEST RESULTS

The Formation of Fouling Layers

Effects of Particle Size

Effects of Particle Size Distribution

Effects of Membrane Pore Size

3. V. Milisic and R. Ben Aim, 1986, Developing a better understanding of cross-flow

4. R.J. Wakeman and E.S. Tarleton, 1991, An experimental study of electroacoustic

24. American Standard Test Method, E1294, 1989.

FIGURES AND TABLES

Figure 1: The topography and cross-section of a 1 μm rated Nuclepore PC membrane.

Figure 2: The topography of a 1.2 μm rated Sartorius CN membrane.

Figure 3: Pore size distributions of the Nuclepore PC membranes.

Figure 5: Schematic representation of the microfiltration flow circuit.

Figure 9: Effect of particle fines on flux decline for calcite suspensions.

Particle/process stream Membrane

Table 1: Parameters evaluated in the characterisation experiments.

Property Nuclepore PC Sartorius CN

Table 3: Miscellaneous membrane characterisation data.

Solids Suspension pretreatment 10% size 50% size 90% size

Feed stream Membrane Process parameters

Table 5: Matrix of properties examined in the microfiltration experiments.

Particle (x) and pore Comments

Table 6: The proposed classification of dynamic membrane types35.

Table 7: Summary of the influence of particle/pore size in microfiltration.

You might also like