2004 in Situ Monitoring Techniques For Concentration Polarization and Fouling Phenomena in Membrane Filtration

Advances in Colloid and Interface Science 107 (2004) 83108
In situ monitoring techniques for concentration polarization and fouling

phenomena in membrane filtration
Jim C. Chen, Qilin Li, Menachem Elimelech*
Department of Chemical Engineering, Environmental Engineering Program, Yale University, P.O. Box 208286, New Haven,
CT 05620-8286, USA
Abstract
Membrane fouling and subsequent permeate flux decline are inevitably associated with pressure-driven membrane processes.
Despite the myriad of studies on membrane fouling and related phenomenaconcentration polarization, cake formation and pore
pluggingthe fundamental mechanisms and processes involved are still not fully understood. A key to breakthroughs in
understanding of fouling phenomena is the development of novel, non-invasive, in situ quantification of physico-chemical
processes occurring during membrane filtration. State-of-the-art in situ monitoring techniques for concentration polarization, cake
formation and fouling phenomena in pressure-driven membrane filtration are critically reviewed in this paper. The review addresses
the physical principles and applications of the techniques as well as their strengths and deficiencies. Emphasis is given to
techniques relevant to fouling phenomena where particles and solutes accumulate on the membrane surface such that pore
plugging is negligible. The relevance of the techniques to specific processes and mechanisms involved in membrane fouling is
also elaborated and discussed.
2004 Elsevier B.V. All rights reserved.
Keywords: Concentration polarization; Membrane fouling; Cake formation; In situ monitoring techniques; Permeate flux decline; Membrane
filtration; Fouling mechanisms
Contents
1. Introduction ............................................................................................................................................ 84
2. Overview of concentration polarization and cake formation ................................................................................ 85
3. In situ monitoring techniques for concentration polarization ............................................................................... 86
3.1. Light deflection techniques ................................................................................................................... 86
3.1.1. Shadowgraphy ............................................................................................................................ 86
3.1.2. Refractometry ............................................................................................................................. 89
3.2. Magnetic resonance imaging (MRI) ....................................................................................................... 91
3.3. Radio isotope labeling ......................................................................................................................... 94
3.4. Electron diode array microscope ............................................................................................................ 95
3.5. Direct pressure measurements ............................................................................................................... 96
4. In situ monitoring techniques for cake formation and membrane fouling ............................................................... 96
4.1. Particle deposition and cake layer formation ............................................................................................. 96
4.1.1. Direct observation through the membrane .......................................................................................... 97
4.1.2. Direct visualization above the membrane ........................................................................................... 98
4.1.3. Laser triangulometry ..................................................................................................................... 99
4.1.4. Optical laser sensor .................................................................................................................... 100
*Corresponding author. Tel.: q1-203-432-2789; fax: q1-203-432-2881.
E-mail address: menachem.elimelech@yale.edu (M. Elimelech).
0001-8686/04/$ - see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cis.2003.10.018
84
J.C. Chen et al. / Advances in Colloid and Interface Science 107 (2004) 83108
4.1.5. Ultrasonic time-domain reflectometry (UTDR) .................................................................................

4.1.6. Electrical impedance spectroscopy .................................................................................................
4.2. Pore blockage .................................................................................................................................
4.2.1. Small-angle neutron scattering (SANS) ...........................................................................................
4.3. Characterization of cake structure ........................................................................................................
5. Concluding remarks ................................................................................................................................
Acknowledgements .....................................................................................................................................
References ................................................................................................................................................
1. Introduction
Pressure-driven membrane filtration processes have
steadily gained importance in industrial separations over
the past three decades. Numerous improvements in the
technologyfor instance, development of highly selective and permeable membranes w1,2x, efficient module
design w13x and several improvements in peripheral
technology (e.g. Ref. w4x)have spurred widespread
adaptation of this process in chemical, environmental,
pharmaceutical and biomedical applications. There are,
however, several aspects of this constantly evolving
technology that have not yet been addressed conclusively and still pose a formidable obstacle toward its wide
acceptance.
One of these important aspects is the understanding
of membrane fouling and subsequent permeate flux
decline, which is inevitably associated with membrane
processes w2,57x. Typical observations of permeate flux
over time reveal a rapid initial decline followed by a
more gradual long-term decline w8x. Traditionally, the
initial decline is attributed to concentration polarization,
a rapid buildup of solute particle concentration near the
membrane surface, while the long-term decline is attributed to various modes of membrane fouling w1,2,9x. In
ultrafiltration and microfiltration processes, fouling by
colloidal particles is usually attributed to the build-up
of a gel or a cake layer w1,2,1013x.
The emphasis in existing studies is to analyze the
long-term flux decline mechanisms, in which case the
concentration polarization ceases to be the dominant
cause of flux decline w1,2,913x, and the cake filtration
theory viewpoint is often sufficient for modeling purposes. It is, however, important to note that concentration polarization initiates cake formation w57,13x.
Hence, fundamental understanding of the dynamics of
concentration polarization can lead to a clearer microscopic view of the mechanisms of cake formation and
its subsequent growth.
Theories of concentration polarization, cake formation
and permeate flux decline have been reviewed in detail
over the recent years w1,2,5x. In general, the current
theories are based on hypotheses of assumed behavior
and are validated by a priori or post-experimental
measurements. Most experimental work on concentra-
102
103
103
103
104
106
106
106
tion polarization and fouling phenomena focused on

measurement of macroscopic, indirect, gross parameterssuch as channel-averaged permeate flux, solute
rejection and pressure dropwhich limits our ability to
elucidate the microscopic processes involved and to
validate pertinent mechanistic theories. Very few studies
have aimed to observe the microscopic processes occurring in situ during filtration. Direct and rigorous observation of processes occurring during membrane filtration
is necessary to truly substantiate current theories and
models. Novel non-invasive, in situ and rigorous quantification of processes occurring during membrane filtration is the key to breakthroughs in our understanding of
fouling phenomena.
In this review, we present the state-of-the-art of
available in situ monitoring techniques for concentration
polarization and fouling phenomena in membrane filtration. The critical review addresses strengths and deficiencies in the current techniques and points out the
relevance to our understanding of membrane fouling.
The review is organized as follows. We start with an
overview of basic principles and terminologies pertinent
to concentration polarization and cake formation in
pressure-driven membrane processes. Later, we describe
available techniques relevant to concentration polarization phenomena, namely, light deflection techniques,
magnetic resonance imaging (MRI), radio isotope labeling, electronic diode array microscope and direct pressure measurements. Following the techniques for
concentration polarization, methods and instrumentation
for observation of particle deposition, cake formation
and membrane fouling are described. The description
starts with optical techniques that monitor particle deposition onto membrane surfacesdirect observation
through a membrane and direct visualization above the
membrane. This is followed by techniques that monitor
the cake layer thickness and the dynamics of cake layer
buildup, such as optical laser sensor and ultrasonic timedomain reflectometry (UTDR). The next techniques that
we describe deal with real time characterization of pore
blockage by small-angle neutron scattering (SANS) and
cake layer structure characterization using static light
scattering (SLS), SANS, small-angle X-ray scattering
(SAXS) and local birefringence techniques. The review
ends with conclusions and implications for membrane

processes.
2. Overview of concentration polarization and cake
formation
In this section we present a short overview on the
principles of concentration polarization and cake formation. We emphasize the distinction between these two
phenomena and their underlying thermodynamic and
hydrodynamic principles. The overview is not intended
to be extensive, but rather it sets the terminologies and
physical principles relevant to concentration polarization
and cake formation, prior to describing the techniques
for in situ monitoring of these phenomena.
Reverse osmosis, nanofiltration and some ultrafiltration processes are characterized by the use of membranes
having pores that can retain solute particles ranging
from a few angstroms to several nanometers in diameter.
Solutes in this size range include small molecules,
simple ions and macromolecules. Filtration of such
solutes from their aqueous solutions generally does not
lead to formation of a cake layer on the membrane
surface. Thus, these membrane filtration processes are
traditionally explained using the concept of concentration polarization, employing thermodynamic principles
governing true solution behavior w5,14,15x.
An alternative approach for describing flux decline
during membrane separation processes is based on the
filtration theory w16,17x. Filtration or hydrodynamic
theories assume that a distinct cake (or gel) layer forms
on the membrane surface, which exerts hydrodynamic
resistance to permeate flow. Filtration theory is adequate
for modeling membrane filtration of colloidal and nonBrownian particles because the cake layer forms instantaneously on the membrane surface. Since cakes are
rarely formed during membrane filtration of very small
solutes, application of filtration theories to membrane
processes where solely concentration polarization occurs
has been debated w7,18x. Considerable debate exists
regarding the suitable theoretical framework to describe
these processes w1921x. This dichotomy often leads to
confusion regarding the domains of validity of the
thermodynamic and hydrodynamic models.
A more proper approach toward unified treatment of
membrane filtration processes is emerging over the
recent years w13,20,2224x. These new theories presume
that above a critical solute or particle concentration at
the membrane surface, a filter cake (gel) layer forms
on the membrane as a result of phase transition. When
an assemblage of solute particles in a solution is compressed by extracting the solvent molecules from the
assemblage, the solutes lose their degrees of freedom,
and the retained solution tends to become more structured w25,26x. Beyond a critical concentration and pressure, the fluid forms a separate phase where the
85
Brownian motion of the solute molecules is frozen w26x.

This phase transition point differentiates between the
true solution behavior (disordered phase) and a cake
type behavior (ordered phase) w26x. In classical studies
on ultrafiltration, the solute concentration at this point
is often termed the gel concentration w14,18,27x. Furthermore, in the mathematical formulation of the emerging modern theories, the formation of a cake is initiated
when this concentration is attained w13,20,23,24x. Below
this concentration, the permeate flux is considered to be
governed by the solution thermodynamics w13,20,23,24x.
Song and Elimelech w13x and Elimelech and Bhattacharjee w24x have quantified the thermodynamic conditions that demarcate pure concentration polarization
from cake formation. It was shown that, for hard
spherical (non-interacting) particles, a dimensionless
parameterthe so-called filtration number, NFcan be
used to determine whether a cake layer would form at
given thermodynamic conditions:
NFs
4pa3pDPyDPbymRmv.
3kT
4pa3pDPp
3kT
(1)
Here, ap is the particle radius, DP the applied pressure,

DPb the osmotic pressure difference between the
membrane surface and permeate, m the solvent viscosity,
Rm the membrane resistance, v the permeate flux, k the
Boltzmann constant and T the absolute temperature.
Note that the expression in parentheses is equal to the
pressure drop across the cake layer (if cake has formed)
or the osmotic pressure drop across the concentration
polarization layer (in the absence of cake formation).
A critical filtration number, NFc, can be identified,
above which a cake layer would form w13,24x. For hard
spherical particles and when neglecting the pressure
drop across the membrane (so-called ideal filter
assumption), it was shown that NFcf15 w13x. When
considering the pressure drop on the membrane, the
critical filtration number can be found by an iterative
technique w24x. A schematic diagram depicting concentration polarization and cake formation in crossflow
membrane filtration and their relation to the filtration
number is shown in Fig. 1.
Alternatively, when the critical filtration number is
known, Eq. (1) can be used to estimate the critical
pressure, DPc, above which a cake layer would form:
DPcfNFc
3kT
4pa3p
(2)
In this equation, DPc can be considered as the osmotic

pressure corresponding to maximum concentration of
the solute particles involved (usually a volume fraction
of 0.64 for random packing of hard spheres). Below
this critical pressure, a cake will never form in
86
Fig. 1. Schematic description of concentration polarization and cake formation over a membrane surface in crossflow filtration. (a) Below the
critical filtration number, NFc, a pure concentration polarization layer exists. (b) Above the critical filtration number, NFc, particles accumulate
and form a cake layer. Adapted from Song and Elimelech w13x.
membrane filtration. One should note the strong dependence of the critical filtration number or critical pressure
on particle size (Aap3). This explains why cake layers
would readily form for suspensions containing particles
as small as a few tens of nanometers.
3. In situ monitoring techniques for concentration
polarization
During the past few decades, various experimental
techniques have been developed for in situ monitoring
of concentration polarization in order to better understand the physico-chemical processes governing the
development of a polarized layer of solutes near a
membrane surface. Such techniques enable the testing
of theoretical models and, more importantly, provide
valuable information on the mechanisms governing the
development of concentration polarization in membrane
filtration.
3.1. Light deflection techniques
One important optical property of a solution is that
its refractive index changes with concentration. Thus,
the change in deflection when light passes through a
solution provides information about the concentration
gradient along the light pathway. Two methods that

make use of light deflection by the concentration polarization layer are shadowgraphy and refractometry.
3.1.1. Shadowgraphy
Shadowgraphy is based on the principle that light is
deflected when passing through a medium of continuously varying refractive index, with the deflection patterns determined by the gradient of the refractive index.
The extent of light deflection and the trajectory of the
light path through the medium can be calculated by raytracing algorithms where the refractive index of the
medium along the pathway is the prime governing factor.
Having measured the amount of light deflection as a
result of traversing through the medium, the refractive
index within the test section is extracted as an inverse
problem of the ray-tracing algorithm.
The refractive index of a medium is a function of a
myriad of physico-chemical parameters, one of which is
the concentration of constituents within the test section.
At given constant operating conditions, with only variations of the concentration, the change in the light
deflection and, thus, the refractive index is an indication
of the change in concentration within the test section.
Shadowgraphy thus offers a method for measurement of
concentration profiles within a medium by exploiting

this light deflection phenomenon.
Vilker et al. w6,19x utilized a shadowgraph optical
method to measure solute (bovine serum albumin)
concentration profiles in the polarization layer adjacent
to an ultrafiltration membrane during dead-end filtration.
During filtration, the induced concentration polarization
layer creates a one-dimensional refractive index gradient
above the membrane. A schematic of the refractive
index profile in the unstirred ultrafiltration cell used for
the experiment and the resulting deflection of light path
are shown in Fig. 2.
The ultrafiltration cell is constructed with two glass
windows, and the front glass window is illuminated by
a collimated beam normal to the surface. A representative light ray is shown to enter the cell at the position
Y0. Due to the refractive index gradient within the cell,
the light ray is deflected continuously towards the
membrane surface. The light ray emerging from the cell
is then imaged onto an image plane. The refractive
index is determined by measuring the light deflection
and performing numerical ray-tracing calculations. The
concentration of albumin in this study is shown to
exhibit a linear correlation to the refractive index. In the
shadowgraphy technique developed by Vilker et al. w19x,
an image of a reticle pattern was placed on the entry
glass surface and was projected through the ultrafiltration cell onto the frosted glass surface of the image
plate located in the back of the cell. The refractive index
gradient in the concentration polarization layer causes
distortion of the image that directly correlates to the
extent of light deflection.
The unstirred ultrafiltration cell used by Vilker et al.
w6,19x was fabricated from two pieces of invar metal
with glass windows of 19 mm2. The membrane was
87
placed between the top and bottom sections of the cell.

The two sections were sloped at 158 from the horizontal
plane to facilitate unhindered path length of even the
most deflected light ray near the membrane surface. The
fiducial marks were shown on the right and form a line
image reticle. Flexibility in the choice of medium was
allowable and, in general, the solvent within the test
section was water. Within the solvent, the constituents
can be proteins, macromolecules, small colloidal particles or emulsions. It is required that the refractive indices
of the solvent and constituents be different such that
appreciable deflection of light will result from the
presence of the constituent.
The optical setup consisted of a 15-mW HeNe laser
as the source of illumination that was converted into an
expanded beam of 45-mm diameter for illumination of
a 19 mm square aperture within the cell. The line image
of the reticle pattern was projected through the ultrafiltration cell onto the frosted surface of a glass fiducial
plate that was located approximately 0.25 m behind the
cell. The fiducial plate was inscribed with a 9=5 matrix
of fiducial marks whose positions relative to one another
were known to be within "0.0014 cm. The fiducial
plate and its marks served to delineate a coordinate
system for the measurement of light deflections. The
images of the distorted reticle pattern and the fiducial
marks were captured by a Graflex camera with Polaroid
Type 55 PyN sheet film. Details of the optical and
filtration experimental setup are given by Vilker et al.
w19x.
The extent of light deflection in shadowgraphy may
be correlated to the concentration within the medium by
a variety of methods. One common practice is to
calibrate the system using known concentrations of the
constituent to determine the light deflection as a function
Fig. 2. Typical light paths (solid lines) for a collimated beam normally incident on the glass window of an unstirred ultrafiltrationyoptical cell
and the refractive index profile h(y) (dashed line) generated by solute rejection at the membrane surface (ys0). The refractive index of glass is
represented by hg and that of the solution by hi. Adapted from Vilker et al. w19x
88
of concentration. The experimental measurements of

deflection can then be directly related to concentration
by the calibration. Vilker et al. w6x, however, utilized a
previously derived correlation between the concentration
of BSA and its refractive index. In this case, the
refractive index was derived from the ray-tracing algorithm and the concentration was directly computed from
the result. The concentrations were measured in the
concentration polarization layer near the membrane surface where BSA accumulate due to rejection by the
membrane.
Fig. 3 compares the theoretical predictions of BSA
concentration profiles in the polarization layer with those
experimentally determined using shadowgraph measurements. The theoretical predictions were based on solution of the convective-diffusion equation with an
osmotic pressure governed permeate flux as presented
by Vilker et al. w6x. Experimental conditions are given
in Table 1 of Vilker et al. w6x. The linear correlation
between the refractive index and the BSA concentration
is found to be valid up to 580 gyl, which sets an upper
limit for the maximum measurable concentration. The
magnitude of the deflection of light near the membrane

surface limited the concentration profile measurements
to 200 mm above the surface. Extrapolation of the
concentration profile allowed determination of the concentration values at distances less than 200 mm from
the membrane surface. Due to the high concentrations
of BSA expected near the membrane surface, such
extrapolations may be inaccurate. The measurement of
concentration was also tested against theoretical predictions that the membrane surface concentration remains
invariant with time and that the polarization layer grows
with t 1y2. The results of experimental measurements at
two elapsed times did indeed show that the surface
concentration is independent of elapsed time of experiment and that the ratio of polarization layer thickness at
the two subsequent times follows a t 1y2 dependence.
There are, however, a few limitations for the shadowgraph technique. The experimental measurements in
this study were limited to filtration conditions resulting
in concentration polarization only. This study does not
consider the polarization layer profile in the presence of
gelycake formation. The operating pressures and solu-
Fig. 3. Comparison of theoretical predictions (---) and experimental measurements () of albumin concentration profiles in the concentration
polarization layer at different applied pressures and pH. Experiment D: pH 4.5, DPs276 kPa; Experiment B: pH 4.5, DPs70 kPa; Experiment
K: pH 7.4, DPs70 kPa. The symbol C * next to each curve represents the albumin membrane surface concentration in the corresponding
experiment. After Vilker et al. w6x.
tion chemistry were adjusted to ensure the absence of

gel formation. An additional constraint is that the technique is limited to binary systems, wherein the refractive
index is directly related to a single concentration. Furthermore, although the agreement between the experimental measurements and theoretical prediction appears
to be good in Fig. 3, it should be noted that the
diffusivity of BSA molecules was used as a fitting
parameter in the theoretical model. The fitted diffusivity
values showed vast deviations from calculated values at
the specified concentrations. The deviations between
predicted and calculated diffusivities were one order of
magnitude and greater. Vilker et al. w6x attributed the
discrepancy to the uncertainties of the true diffusivity
values. Another limitation of the methods capability is
that it can only measure concentrations beyond 200 mm
above the membrane surface. The reason for this disadvantage is that the deflection of light in the test
section is bent towards the membrane surface. At high
concentrations due to polarization near the membrane
surface, the deflection may be severe to the point that
the light path is directed into the membrane surface.
The ray is then unable to emerge from the test section.
Possible remedies for this predicament include shortening the light path length across the membrane surface
to provide easier clearance or adjusting operating conditions such that the concentration near the membrane
does not cause excessive light deflection. The latter
approach, however, may not be viable, as the phenomena
of interest in membrane filtration, such as concentration
polarization and gel formation, occur at high membrane
concentrations.
3.1.2. Refractometry
Refractometry is, in principle, similar to shadowgraphy in that the refractive index gradient within the cell
causes deflection of light, which is related to the
concentration gradient in the polarized layer. Ethier and
Lin w28x utilized a differential refractometric method for
measuring the concentration profile in unstirred ultrafiltration of hyaluronic acid (HA). In this study, the
concentration was determined from calibration with
known samples. However, calibration for the concentration gradient measurement was not obtained directly due
to difficulties in producing a known concentration gradient. Instead, the calibration was performed by plotting
the spatial integral of the X-deflection data against
known concentrations of HA. The slope of this line
yielded the necessary conversion between the X-deflection data and the concentration gradient. This experimental system was improved by Gowman and Ethier
w29,30x to feature automated dual concentration and
concentration gradient measurements.
Details of the modified experimental setup are given
by Gowman and Ethier w29x. The filtration cell consisted
of two chambers that were adjacent but hydraulically
89
independent. The walls of the chambers were BK7 plate

glass to allow illumination. One of the chambers
received a reference solution that contains no HA, while
the filtration of the HA solution occurred in the other
chamber. A schematic of the experimental setup is
shown in Fig. 4. Light from a 10-mW HeNe laser was
properly collimated and filtered to produce a beam
diameter of 127 mm and a depth of field of 34.9 mm at
the center of the flow cell. The appropriate optics were
used to align the beam path parallel to the surface of
the membrane. A system of mirrors mounted on computer-controlled stepper-motor-driven translation stages
was installed to enable movement of the laser beam
vertically through the concentration polarization layer at
5-mm increments.
The light traversing through the cell is deflected in
two directions, X and Y as shown in Fig. 4. The Xdeflection results from the presence of a vertical concentration gradient within the concentration polarization
layer, while the Y-deflection is caused by the difference
in the concentrations between the reference chamber
and the polymer-containing chamber. Thus, the system
exhibits the feature of dual concentration and concentration gradient measurement capabilities. The beam
emerging from the flow cell was directed onto a quadrature position detector mounted on two orthogonal
computer-controlled stepper-motor-driven translation
stages. The quadrature detector served to measure the
amount of beam deflection at a given vertical location
within the concentration polarization layer.
Since the experimental setup was susceptible to errors
due to optical drift, proper corrections for optical drift
were made by running experiments in absence of HA
under similar experimental conditions. The HA-free
solution allowed measurements to be made under identical conditions as the actual experiments using HA. In
contrast, the calibrations could not be performed under
the same experimental conditions, as the filtration process would immediately alter the known concentration
in the filtration cell. Uncertainty in positioning the
experimental stages could also introduce errors. The
minimum possible uncertainty from positioning errors
were found to be "5 mm. Additional uncertainties
included the location of the membrane ("200 mm),
which would affect the coordinate values of measurements. The spatial resolution of measurements through
the polarization layer was 5 mm and the closest measurable distance from the membrane surface was 200
mm. The maximum allowable operating pressure was
500 mmHg. Concentrations were measurable up to 1.2%
mass HA.
Results of concentration and concentration gradient
measurements given by Gowman and Ethier w30x demonstrated that development of the concentration polarization layer with time and its general profile was
correctly depicted. The experimental measurements also
90
Fig. 4. Schematic of the automated, laser-based refractometric system for measurement of concentration and concentration gradient in the concentration polarization layer developed by Gowman and Ethier w29x. The refractive index gradient (due to the concentration gradient) in the
vertical direction causes the beam to deflect toward the membrane: X-deflection. The Y-deflection, parallel to the membrane, is due to the
refractive index difference (concentration difference) between the reference and polymer sides. Both pressure transducers are well upstream of
the flow cell and are located at the same elevation as the membrane. Translating optics are located within the dot-dashed box. Adapted from
Gowman and Ethier w29x.
showed that a decrease in the initial concentration of

HA led to a thinner concentration polarization layer
with a lower steady-state concentration at the membrane
surface. In addition, they showed that a higher permeate
flux contributed to a greater membrane resistance. An
increase in ionic strength of the solution was also shown
to suppress the polarization layer thickness, while conversely increasing the membrane concentration due to
mitigation of repulsive electrostatic forces between macromolecules. However, different ionic species were used
in aqueous solutions (NaCl vs. phosphate buffer), which
may contribute to the difference in the thickness of the
polarization layer and the membrane concentration.
The experimental measurements of Gowman and
Ethier w30x were compared to the theoretical prediction
using the models of Ethier w31x, Lam and Bert w32x,
Peitzsch and Reed w33x and Johnson w34x. These theoretical models are based on Darcys law for dead-end
filtration with a zero pressure gradient within the polarized layer. The permeate flux is modeled as being
osmotic pressure-driven using empirical coefficients to
calculate the osmotic pressure. The studies of Ethier
w31x, Lam and Bert w32x, Peitzsch and Reed w33x and
Johnson w34x provided values of various parameter for

computing the osmotic pressure.
The models were found to yield different shapes of
the steady-state concentration profile from the experimental results. The membrane concentration was consistently over-predicted compared to the measured value.
Depending on which measured parameter was used as
the constraint, total mass of HA or the final thickness
of the polarization layer, the predicted final layer thickness or the total HA mass in the experiment and the
final membrane concentrations do not correspond well
to the actual values. Several reasons for the disagreement
were proposed. First, the theoretical model might not be
applicable to the experimental conditions of this study.
Assumptions for the theoretical model included no
mechanical stress between the macromolecules in the
polarized layer, which might not be valid for HA
molecules due to their large size and randomly coiling
structure. Second, the physico-chemical parameters for
the models, osmotic pressure and permeability, were not
determined under the same conditions as in the experiments of Gowman and Ethier w30x. Reasonable adjustment of the parameters resulted in better agreement
between theoretical prediction and experimental results.

Furthermore, some experimental errors also contributed
to the discrepancies. For example, the initial HA concentration loaded into an experiment might be less than
the nominal concentration due to the errors caused
during the weighing and mixing process. Also, the
membrane concentration could not be directly measured
near the membrane due to experimental constraints, and
so it was extrapolated from the concentration profile.
The extrapolation is subject to error as the concentration
near the membrane is expected to be high.
One key issue of concern that presented a source of
error was the loss of HA through the membrane. The
membranes were assumed to reject HA completely.
However, post-experimental mass conservation calculations indicated approximately 20% loss of HA. The loss
of mass was attributed to HA being entrained in
membrane pores. The loss of mass was considered a
contributing factor to the disagreement between experiment data and model predictions. Such pore blockage
may significantly alter the permeate flux and the polarization layer formation behavior and thus should be
adequately accounted.
The experimental measurements assumed the case of
concentration polarization only and did not account for
possible formation of a gel layer. Likewise, the theoretical models were derived on the basis of the concept
of an equivalence of osmotic and filtration theory w24x,
which is applicable only to instances with solely concentration polarization. However, with the high molecular weight of HA (MWs605,000 Da), the system was
susceptible to the formation of a gel layer, in which
case the filtration process becomes gel layer governed
w24x. For both the experimental measurements and the
theoretical models to be applicable, it is therefore imperative to ensure that operating conditions do not incur
gel formation. This constraint was further complicated
by the limitation of the flow cell design such that
asymmetric membranes could not be used, as they could
not be sealed to provide independent flow in the reference and polymer chambers. As a result, only tracketched membranes with a large nominal diameter of
0.015 mm were used, thus limiting the technique to
investigations of ultrafiltration of large macromolecules
such as HA. The filtration of large macromolecules is
highly susceptible to gel formation as may be realized
by calculating the filtration number from Eq. (1) in
Section 2.
3.2. Magnetic resonance imaging (MRI)
MRI is a technique that has the potential of becoming
an important tool for in situ monitoring of a variety of
chemical and particulate processes, such as concentration polarization in membrane filtration. It is an exten-
91
sively employed diagnostic tool in clinical radiology for

generating high-resolution images of internal anatomy.
Images are obtained by placing the test section in a
powerful, highly uniform, static magnetic field. Magnetized protons (hydrogen nuclei) within the sample
align and behave as magnets in this field. Radio frequency pulses are then utilized to create an oscillating
magnetic field perpendicular to the main field, from
which the nuclei absorb energy and move out of alignment with the static field, in an excited state w35,36x.
As the nuclei return from excitation to the equilibrium
state, a signal induced in the receiver coil of the
instrument by the nuclear magnetization can then be
transformed by a series of algorithms into diagnostic
images. 1H nuclear magnetic resonance (NMR) images
are generated from the distribution of mobile protons in
the sample weighted by the corresponding spinspin
(T2) and spin-lattice (T1) relaxation functions w35,36x.
Images of different components within the test section
can be obtained by varying the number and sequence of
pulsed radio frequency fields based on magnetic relaxation properties of the individual constituent. The technique allows imaging of any selected plane within a
complex sample by slicing the radio frequency through
the designated location and can image samples that are
optically opaque. Additional advantages include a spatial
resolution down to 10 mm or lower. Details of the
physical principles of NMR imaging are given elsewhere
w35,36x.
An important remark regarding the studies to be
reviewed below is that the authors of these studies
classify their work as investigations on concentration
polarization. However, their definition of concentration
polarization refers only to the general phenomena of
particle accumulation at the membrane surface. The
experimental conditions employed, such as particle size,
and the experimental observations that result, such as a
stagnant polarization layer, strongly suggest that a cake
layer of maximum packing formed in their systems.
Thus, based on our discussion in Section 2, the classical
definition of concentration polarization depicting the
development of a disordered solute layer above the
membrane surface is not suitable for these studies. Thus,
in the discussions below, the term polarization layer
refers to a cake layer of maximum packing density. In
this sense, cake formation is treated as a special case of
concentration polarization.
Yao et al. w37x utilized NMR micro-imaging techniques in a non-invasive study of flow and concentration
polarization in hollow fiber membrane filtration. The
membrane module consisted of an inner tube containing
the membrane fibers and an outer envelope, and was
fabricated from glass to facilitate micro-imaging. The
flow was fed to the shell side of the fibers in the inner
glass tube and the retentate was recirculated through the
outer envelope. The permeate was collected at one end
92
Fig. 5. Flow of doped water through a hollow fiber membrane module containing five fibers (packing density 18%). (a) NMR image obtained
with the bipolar flow encoded gradient turned off to show the disposition of the hollow fiber membranes in the module. (b) Flow velocity image
presented as a gray scale contour plot; bright areas represent regions where the flow velocity is positive (outflows), while dark areas correspond
to inflow of the feedstock around the outside of the fibers. (c) Velocity image showing the negative velocity components (flow of feedstock on
the shell side of the fiber) in the form of a stackplot. (d) Stackplot showing the positive velocity components (flow of filtrate in the inner lumens
of the hollow fiber membranes and return flow of the filtrate in the outer annular space). All images are 1-mm slice thickness, 10-mm field of
view and 256=256 pixels. The maximum flow velocities are 21.3 mmys for positive flows and 17.5 mmys for negative flows. Figures from Yao
et al. w37x.
of the inner lumens of the fibers, while the other end

was sealed. More details of the experimental setup are
given by Yao et al. w37x.
Oil emulsions of 5% vyv (Caltex Trusol DD oil in
water) doped with 0.6 mM CuSO4 with average droplet
diameter of 0.3 mm were tested in this study. With such
large droplet sizes, formation of a cake layer of maximum packing is certain, as can be inferred from the
high filtration number discussed in Section 2. The bulk
flow within the membrane module was measured by the
dynamic NMR microscopy technique w38,39x. This technique obtained a map of the distribution of velocities
over the test section by varying the amplitude of the
radio frequency excitation pulses over a range of values
and Fourier transforming the response signal with
respect to a velocity dimension w38,39x. The velocity
maps obtained by Yao et al. w37x with a module
containing five fibers are shown in Fig. 5. The flow
direction of the permeate through the lumens and the

feedstock through the outer annulus were defined positive and that of the feedstock was negative. Comparisons
of volumetric flow rates calculated from the NMR
images and those determined by direct measurement in
both positive and negative directions showed excellent
agreement. Therefore, NMR imaging proved to be an
effective technique to achieve flow velocity maps within
a membrane filtration module.
Images of oil polarization layers were also obtained
by Yao et al. w37x using chemical shift selective imaging.
The oil polarization layer was found to be the thickest
in regions of low feedstock velocity shear, such as where
the fiber was close to the inner tube wall and between
two fibers that lay close to each other. Change of
feedstock pressure showed no significant effects on the
oil polarization layer thickness because both the transmembrane pressure and the crossflow velocity increased
with increasing feed pressure. When the velocity maps

were combined with the polarization layer images, no
axial flow was found in the polarization layers. Moreover, the water velocity fell to 0 at the polarization layer
boundary instead of the membrane surface, indicating a
stationary polarization layer. Using the resistance in
series model, the resistance of the polarization layer
was found to be the dominant flux limiting factor.
Pope et al. w40x and Airey et al. w41x extended the
work of Yao et al. w37x to achieve quantitative measurements of the concentration polarization layer thickness
in membrane filtration of oilwater emulsions and
colloidal silica particles, respectively, using NMR imaging. The experimental filtration modules of Pope et al.
w40x and Airey et al. w41x were similar to that of Yao et
al. w37x, with the exception that a single sheet tubular
membrane is housed within the filtration module rather
than a bundle of hollow fibers. The flows in these
studies were from the lumen side towards the shell side.
Pope et al. w40x utilized 5% vyv Caltex Trusol DD oil
emulsions doped in CuSO4 with a resulting average
droplet diameter of 0.3 mm, while Airey et al. w41x
performed filtrations with colloidal silica suspensions
(Dupont LUDOX HS40 wt.%) with an average particle size of 12 nm. The size of the silica colloids is
significantly smaller than the membrane pore size; however, Airey et al. w41x noted that 99.9% rejection of the
particles was achieved. Therefore, it is likely that a cake
layer formed resulting in retention of the silica particles.
The use of silica particles in the work of Airey et al.
w41x also required some modification of the experimental
procedure as silica particles do not contribute to the
proton (1H) NMR signal. This limitation can be overcome because the concentration of silica particles has
been shown to alter the relaxation times (both T1 and
T2) of the solvent water protons substantially. Airey et
al. w41x exploited this property of silica particles and
obtained a series of T1 weighted NMR images during
filtration. The T1 weighted images do provide clear
details of the formation of the silica polarization layer.
Details of experimental conditions are given by Pope et
al. w40x and Airey et al. w41x.
Results of images of polarization layer development
are shown by Airey et al. w41x. Pope et al. w40x input
the results of measured polarization layer thickness into
Darcys Law to extract the specific resistance of the
polarizationycake layer. The measured polarization layer
thickness agreed with predicted behavior from Darcys
Law of increased resistance resulting from increased
applied pressure. The specific resistance of the polarization layer was calculated and revealed that significant
resistance, on the order of 145 times the clean membrane
resistance, resulted from the development of the polarizationycake layer. This calculation is subject to error as
the average measured thickness is inputted into the
model. If the oil polarization layer exhibits significant
93
non-uniformities, the use of an average thickness for the

oil layer can result in underestimation of the specific
resistance. However, it does demonstrate an insightful
application for the measured polarization layer thickness.
Furthermore, the measured polarization layer thickness
was inputted into mass transfer concentration polarization models, and the results showed good agreement
between observed flux behavior and that predicted by
theory when Brownian diffusion is taken as the dominant
back-transport mechanism. The observation technique
employed in this study thus shows the added benefit of
deciphering the presence and form of the back-transport
mechanisms. From the results of Airey et al. w41x, the
acquired images confirmed expectations of a growing
polarizationycake layer with time that reached a steady
thickness. The polarization layer also increased in the
downstream direction of the membrane module. The
measurements of Pope et al. w40x and Airey et al. w41x
demonstrated the flexibility of NMR techniques to image
both oil emulsions and colloidal suspensions. The polarizationycake layer was clearly discernible and its thickness was readily measurable.
Beyond the polarization layer thickness, an additional
useful application of NMR techniques is that it enables
the measurement of colloidal suspension concentrations.
Airey et al. w41x showed that both the T1 and T2
relaxation times of the colloidal silica suspension are
strong linear functions of the concentration of particles.
Therefore, a calibration curve of known concentrations
of silica particle suspensions and their relaxation times
can be obtained from NMR measurements. The concentration of a given suspension is then readily determined
from the calibration curve. However, the applications in
these studies only yielded measurements of the polarization layer thickness and did not attempt to measure
the concentration and concentration gradient profiles
within the polarization layer. Indications are that a
concentration corresponding to maximum packing is
present within the investigated polarization layer.
Airey et al. w41x and Yao et al. w42x utilized dynamic
NMR microscopy and chemical selective NMR flow
imaging, respectively, to investigate the fluidity of the
polarization layer. Chemical selective NMR imaging
signifies a mode of operation with a designated sequence
of radio frequency pulse excitations. Details of the
principles of chemical selective NMR are given elsewhere w35,36x. Airey et al. w41x performed filtration of
colloidal silica particles (12 nm diameter), while Yao et
al. w42x studied polarization effects of 20% vyv oil
water emulsions of 300-nm-diameter droplets. The
experimental filtration module is again the same as that
described in Airey et al. w41x. The results of Airey et
al. w41x indicate that the motion of particles within the
polarization layer is clearly discernible up to approximately half of the layer thickness. Under the operating
conditions of DPs75 kPa and a crossflow Reynolds
94
Fig. 6. Gamma radiation signal profiles measured from experiments with tracer (I125 ) labeled proteins using membranes of different pore sizes.
Signal corrected to original activity of sample. Experimental conditions: Reynolds numbers10, permeate fluxs10 lym2 h, total BSA concentrations1.1 gyl. Figure from McDonogh et al. w44x.
number of 215, it is observed that the velocity in the

boundary separating the polarization layer and the bulk
is 4 mmys and falls to 0 approximately half-way through
the developed polarization layer (f500 mm total thickness). However, the results of Yao et al. w42x for
filtration of oil emulsions show that the polarization
layer is effectively stationary with "7 mmys error under
experimental conditions of DPs70 kPa and varying
Reynolds number from 100 to 1000. This difference
shows that the fluidity of the polarization layer is
strongly influenced by the nature and composition of
the feedstock being filtered. However, both results of a
zero velocity within the polarization layer confirm that
a stagnant deposit layer has been formed.
3.3. Radio isotope labeling
Radio-isotope-labeled macromolecules were first used
to study concentration polarization by McDonogh et al.
w43,44x. The accumulation of radio-labeled macromolecules on the membrane surface due to concentration
polarization was detected as an increase in voltage by a
scintillation detector. The detected signal is converted
to a corresponding mass of the macromolecule above
the membrane, so that the development of concentration
polarization can be monitored.
The filtration cell consisted of a rectangular flat sheet
module with a polycarbonate top and an aluminum
bottom. The upper half of the cell was constructed in
such a way to minimize loss of scintillation signal. The
scintillation detector was mounted directly above the
filtration cell. BSA was the model macromolecule used
in filtration. Filtration was first carried with I125 labeled
BSA and the formation of the polarization layer was
monitored until constant values of scintillation, Stot, were
reached. Then, unlabeled BSA was filtered for 1 h and
a buffer solution alone for an additional hour. Following

filtration, the membrane was subjected to mechanical or
hydraulic cleaning.
The total signal, Stot, is the sum of signals by proteins
(i) in the channel (bulk solution), (ii) in the dynamic
layer, (iii) fixed to the membrane but removable by
cleaning and (iv) fixed to the membrane and not
removed by cleaning:
StotsSchqSdynqSweakqSstrong
(3)
The signal Sch was determined by filling the channel

with radio-labeled BSA without filtration, and the
amount fixed to the membrane, SstrongqSweak, was determined by the signal after the 1-h filtration with the
buffer solution. Scintillation measured after the
membrane cleaning was attributed to Sstrong. Knowing
these contributions allows the determination of Sdyn.
The correlation between scintillation signal and extent
of polarization was determined from calibration with
known masses of protein in the cell without filtration.
The known quantity of protein does not undergo filtration and subsequent concentration polarization, but simply is stagnant within the cell for measurement by the
scintillation detector. A range of membranes were used
in these studies: polysulfone membranes with pore size
0.01 mm and cellulose-acetate membranes with pore
size ranging from 0.1 to 1.0 mm. The membranes with
pore sizes near 1 mm did not effectively retain protein
and exhibited poor rejection. Details of the experimental
setup are given by McDonogh et al. w43,44x.
The results of McDonogh et al. w44x are given in Fig.
6. An initial jump in the detected signal was observed
at the beginning of filtration as the cell volume was
filled with labeled species. Subsequently, the signal
increased at a much slower rate as the protein accumu-
95
Fig. 7. Observed time-dependent concentration polarization profiles. (a) No crossflow; (b) crossflow at Reynolds number of 10. Experimental
conditions: cuprophan membrane, 2 gyl dextran blue in water, DPs1 bar. Figure from McDonogh et al. w44x.
lated above the membrane due to concentration polarization. Such behavior is indicative of concentration
polarization without cake layer formation.
The radio isotope labeling technique described by
McDonogh et al. w44x did not provide quantitative
measures of the polarization layer thickness or the
concentration profile within it. It provided limited information regarding the extent of polarization in terms of
total mass accumulated above and within the membrane.
3.4. Electron diode array microscope
McDonogh et al. w43,44x presented a method for
observing concentration polarization in ultra- and microfiltration of BSA and dextran blue solutions by use of
an electron diode array microscope (EDAM). A collimated, near infrared light parallel to the membrane
surface, but perpendicular to the flow direction, was
channeled through the filtration cell to a photodetector
by microscopic lenses. Owing to absorption of the
incident light, the concentration gradient in the polarization layer resulted in an intensity pattern on the
detector, which was monitored by an oscilloscope. The
optical setup of the system allows sampling within a 2mm zone above the membrane surface with a resolution
of 2 mm. Details of the experimental setup are given by
McDonogh et al. w44x.
The system requires calibrations to determine the
distance from the membrane surface and to translate the
intensity signal to concentration. The distance from the
membrane can be determined by introducing a machined
position mark on the channel wall and measuring the
signal when placing an object of known dimension
above the position mark. McDonogh et al. w44x used a
hair as the reference object; such an approach may not
provide an accurate calibration of the system. Calibration
of solute concentration can be achieved from signals

detected when the cell is filled with solute of known
concentrations but with no pressure applied. However,
when pressure is applied for filtration, there is an
apparent deflection in the membrane peak possibly due
to a downward shift of the membrane position by 510
mm. This variability was not controllable and moreover
showed time dependence. As such, measurements were
only considered up to 20 mm above the original location
of the membrane surface. One alternative is to use the
signal immediately after the introduction of pressure,
but the filtration process will instantaneously alter the
solute concentration in the cell. Thus, for such a technique, a better calibration method should be developed.
Results of McDonogh et al. w44x are given in Fig. 7,
where the growth of the concentration polarization layer
of dextran blue is demonstrated. The concentration
within the layer increased and approached a quasi-steady
state with time. The results also depict the influence of
crossflow. The quasi-steady state was reached more
quickly and the thickness of the polarization layer was
smaller in the presence of a tangential flow. As expected,
the applied pressure was shown to increase the
membrane concentration and the polarization layer
thickness. Furthermore, the concentration polarization
layer thickness and the membrane concentration were
both shown to increase with increasing feed concentration. Comparisons of filtration behavior of BSA and
dextran blue solutions indicated that under the same
operating conditions, dextran blue exhibits a markedly
greater extent of concentration polarization as reflected
by the membrane concentration. This difference is attributable to the higher diffusivity of BSA that functions to
mitigate concentration polarization effects by aiding
back-transport of the protein. The observations therefore
confirm the expected effects of diffusivities on concen-
96
tration polarization. McDonogh et al. w44x did not,

however, present validation of the experimental results
with theoretical predictions. And so, although the experimental observations are in agreement with expected
qualitative trends, accurate quantitative confirmation was
not available.
McDonogh et al. w44x identified the possible sources
of error in their experiments to be from uncertainties in
the applied pressure, membrane position, solution concentration, flow conditions, concentration calibration,
optical setup, data plotting and data reading. The estimated total error was 23%. However, the lack of
significant scatter of the experimental data suggests that
the error was overestimated.
An important attribute of the EDAM technique is that
the concentrations within the polarization layer are
measurable as close as 20 mm from the membrane
surface. This method thus exhibits a vastly superior
measurable distance above the membrane over shadowgraphy and refractometry, which only allow for measurements beyond 200 mm from the surface.
3.5. Direct pressure measurements
In addition to solute concentration, knowing the pressure distribution across the concentration polarization
layer can shed light on the physico-chemical transport
processes governing concentration polarization. A
dichotomy exists regarding the nature of the pressure
profile with opposing views of a constant pressure
profile and a non-zero pressure gradient. Detailed discussions of the nature of the pressure profile within the
polarization layer are given by Elimelech and Bhattacharjee w24x and Peppin and Elliott w45x. Seeking for
experimental verification of the proposed theories,
Zhang and Ethier w46x developed a system that can
directly measure both pressure and concentration profiles
of the polarization layer during unstirred ultrafiltration.
The experimental apparatus is similar to that of Ethier
and Lin w28x with the addition of a miniature pressure
sensor on the inner wall of the polymer-containing
chamber. Concentration profiles of HA were measured
using the refractometric technique described previously.
The pressure sensor was situated to a suture thread that
facilitated upward and downward movements for measurement at various locations within the cell. The spatial
resolution was 0.2 mm. Details of the experimental
setup are given by Zhang and Ethier w46x.
Pressure measurements were made within the polarization layer once the system reached steady state as
signaled by a constant feed pressure and concentration
profile. A slight pressure drop of 7.9 mmHg within the
solute polarization layer was found as compared to 400
mmHg pressure drop across the membrane, with the
highest gradient within 2 mm above the membrane
surface. Thus, it is clear that the pressure gradient within
the polarization layer was insignificant, yet appreciably

different from zero. The concentration polarization
developed under operating conditions used in this study
was not as severe as typically would be encountered in
ultrafiltration. With higher permeate flux and the concomitant increase in membrane concentration, the pressure drop across the polarization layer is expected to be
higher. Zhang and Ethier w46x commented that the nonzero pressure gradient measured in their study contradicts previous results from Kim et al. w47x of a constant
pressure profile, but is in agreement with theoretical
predictions of Peppin and Elliott w45x. However, they
explained that the BSA solution used by Kim et al. w47x
exhibits a much more compact structure than HA, the
solute used in their study. The comparison is therefore
not under equal terms.
One disadvantage of this technique is that the pressure
sensor disturbs the concentration polarization layer. The
entry of the pressure sensor into the polarized layer
causes constriction to the flow in that vicinity. This
deficiency was detected when pressure measurements
were taken at a fixed location with elapsed time. Results
indicated a time-dependent pressure profile at a fixed
location under steady state conditions. Such transient
behavior when steady state conditions are expected does
pose as a source of inaccuracy to the measurement.
Also, the measurements obtained may not be representative of an unperturbed system. The intrusive nature of
the technique also requires a long duration for experiments to allow for steady state to be reached upon
displacing the pressure sensor to its designated location.
The lengthy time scale of experiments (f10 h for one
traverse) introduced additional sources of error including
drift in the pressure measurement caused by fluctuations
in atmospheric pressure. In the study by Zhang and
Ethier w46x, the local atmospheric pressure was continually monitored to account for any fluctuations. The
intrusive nature of the technique hampers the experimental
process
with
requiring
cumbersome
accommodations.
4. In situ monitoring techniques for cake formation
and membrane fouling
Membrane fouling is generally associated with a cake
or gel formation on the membrane surface, or the
plugging of membrane pores by macromolecules, colloids or particulate matter. At present, the mechanisms
of cake formation and fouling are relatively poorly
understood. Thus, in situ measurements of fouling and
direct observation of cake layer formation are of paramount importance in our efforts to understand the
fundamental processes governing membrane fouling.
4.1. Particle deposition and cake layer formation
The techniques described in this section were developed to visualize or monitor the depositionyaccumula-
97
Fig. 8. Particle deposition images taken during filtration of 11.9-mm diameter latex particles. Images from Li et al. w48x. Crossflow velocity was
fixed at 0.375 mys, while permeate fluxes were as follows: (a) 35 lym2 h, (b) 45 lym2 h, (c) 51 lym2 h and (d) 51 lym2 h.
tion of particles onto a membrane surface and the

dynamics of cake layer buildup as fouling progresses.
4.1.1. Direct observation through the membrane
The most straightforward method to directly observe
particle deposition is visualization by an optical microscope. One of the first applications of this technique
was developed by Li et al. w48x and was termed direct
observation through the membrane (DOTM). In this
method, a microscope objective was positioned at the
permeate side of a transparent membrane and particle
deposition was observed in real time by the microscope.
A key to this technique is the use of an appropriate
membrane. Li et al. w48x used an Anopore (Whatman,
UK) anodized aluminum membrane with high porosity
(60%) and straight through pores. When wet, the
membrane is transparent and thus facilitates observation
of the membrane surface from the permeate side. The
membrane module was fabricated from Perspex, which
allows the transmission of light from the feed side.
Images of particle deposition were recorded by a video
camera. Details of the experimental setup are given by
Li et al. w48x.
Examples of images acquired by Li et al. w48x using
11.9-mm diameter latex particles are given in Fig. 8.
Using the DOTM technique, the authors were able to
estimate the critical flux, the lowest permeate flux for
the onset of cake formation. Above the critical flux,

rapid particle deposition was observed. In contrast, very
few particles were found when the flux was below the
critical value. The deposition of particles was found to
be reversible upon reduction of flux or increase of
crossflow shear. The observations also showed that the
motion of particles near the membrane surface subsequent to deposition was dependent upon the crossflow
velocity, the particle size and the particle size distribution. Adjustment of operating variables such as flux and
crossflow could induce particle motion along the
membrane surface. The DOTM technique can also yield
the size distribution of the deposited particles. Comparison of particle size distributions of deposited particles
under various operating conditions showed that under
high crossflow velocity the deposited particles were
smaller than those deposited under low crossflow conditions. This observation supports the theory that crossflow shear forces deter the deposition of large particles
and thus the cake layer that formed under high crossflow
conditions has a compact structure consisting of small
particles. Furthermore, results from this study are also
in qualitative agreement with theoretical model predictions that the critical flux increases with increasing
crossflow velocity and particle size.
A simplified way to explain the critical flux is to
assume that, at the critical flux, particle convection by
98
Fig. 9. Schematic diagram of the direct visual observation system used by Kang et al. w50x. See description in text.
the permeate flow is balanced by the particle backtransport away from the membrane. Assessment of the
critical flux can therefore be used to ascertain the nature
of back-transport mechanisms. Li et al. w49x utilized the
DOTM technique discussed above to compare experimentally measured critical flux values with predictions
by models that incorporate back-transport mechanisms.
Back-transport mechanisms employed by the models
were either inertial lift or shear-induced diffusion. The
inertial lift model was found to give much lower critical
flux values than those measured by DOTM, whereas the
shear-induced diffusion model showed better agreement
with experimental measurements. The model calculations, however, were quite simplified as colloidal interactions and the two-dimensional nature of the flow in
crossflow filtration were not considered.
4.1.2. Direct visualization above the membrane
The major drawbacks of the DOTM technique
described above are the need to use a transparent
membrane and the positioning of the microscope objective below the membrane at the permeate side. The first
requirement is a major disadvantage as it confines the
use of the technique to only a limited number of
inorganic porous membranes, mostly microfiltration
membranes. The second limitation does not allow the
observation of particle accumulation beyond a monolayer as observation through the deposited particle layer is
not readily attained. To overcome these constraints,
Kang et al. w50x and Mores and Davis w51x constructed
a rectangular membrane channel where the microscope
objective is mounted above the membrane to view
particle deposition from the feed side.
Kang et al. w50x constructed a rectangular crossflow
membrane cell from polycarbonate and glass. The
dimensions of the flow channel were 1 mm (height) by

25.4 mm (width) by 76.2 mm (length). The cell was
mounted on the stage of a phase contrast microscope
(Olympus, BX-51) to allow direct observation of particle deposition by light and fluorescence microscopy
(Fig. 9). A CCD camera (5= magnification) was
mounted on the microscope and images were downloaded in real time to a laboratory PC for post-processing
and image analysis.
In order to operate at common applied pressures for
MF and UF membranes, two interchangeable parallel
plates were fused to the top and bottom cell plates and
reinforced as needed. The membrane sample was placed
between the two plates with a permeate spacer beneath
it. Applied pressure was maintained constant by connecting the flow cell to a pressure vessel (the feed tank)
through a closed line loop. Peristaltic pumps were used
for both retentate and permeate flows, which allowed
for accurate control of the crossflow and permeation
velocities. The retentate and permeate flows were
merged and sent back to the feed tank. Such a configuration enabled extremely stable feed pressure and flow
rates within the crossflow filtration cell.
Kang et al. w50x used a polyacrylonitrile UF
membrane and a polysulfone MF membrane to observe
in real time the deposition of microbial and synthetic
particles. The deposition of synthetic particlesfluorescent blue carboxylate-modified latexwere observed by
fluorescence microscopy. The microbial particlesSaccharomyces cerevisiae and Pesudomonas capacia
cellswere stained and the deposition was observed by
light microscopy. Using this setup, Kang et al. w50x
investigated the influence of various physico-chemical
parameters (ionic strength, permeate flux and crossflow
rate) on the rate of particle deposition on the membrane
99
Fig. 10. Steps of image analysis to determine the area covered by cells deposited on a membrane. Images were taken from Kang et al. w50x.
surface. In addition, the system was used to study

patterns of particle deposition onto the membrane.
Images obtained from the CCD camera were transferred to an image processing program (NIH image J).
Each pixel in the image holds 256 steps of the integer
gray-level value (Fig. 10a), which describes the intensity
of the light (0sblack, 255swhite) in the gray-scale
image. After selecting the adherent cell found at the
same position in consecutive images, the contrast and
brightness were adjusted to obtain a clear image of the
cells as shown in Fig. 10b. The distribution of the graylevel was then analyzed to determine the threshold value
for conversion of gray-scale images to black and white
images.
A somewhat similar direct visualization technique
was used by Mores and Davis w51x to investigate
cleaning of fouled microfiltration membranes by backwashing and backpulsing. In their study, the foulant was
a dyed yeast (Saccharomyces cerevisiae), and the MF
membranes used in the crossflow experiments were a
polymeric cellulose-acetate membrane and an Anopore
anodized-aluminum disk membrane. Two different methods of cleaning were carried outbackwashing and
single backpulses of duration 0.2 or 0.5 s. The direct
visualization results showed that the yeast particles
deposited less uniformly on the surface of the celluloseacetate membrane than that on the inorganic aluminum
membrane. Foulant was shown to be removed by reverse
flow in clumps. The time constant for foulant removal
and the efficiency of membrane cleaning were determined from the obtained photomicrographs. The results
of Mores and Davis w51x provided valuable insights into
the mechanisms of MF membrane cleaning, an area that
surprisingly received little attention so far.
4.1.3. Laser triangulometry
In laser triangulometry, a laser beam is directed onto
a surface of varying elevation, such as a developing
particle cake layer. The reflection of the laser beam off
the surface is captured by a CCD camera. As the
elevation of the surface changes, the location where the

reflected beam is imaged by the CCD array will shift.
Measurement of the shift can then be directly related to
the displacement of the elevation of the surface. During
membrane filtration, laser triangulometry can be used to
track the growth of a particle cake layer and to measure
the thickness of the cake layer.
Altmann and Ripperger w52x studied the development
of a particle cake layer on a membrane surface using
laser triangulometry. The filtration system investigated
was a crossflow microfiltration unit. Details on the
experimental setup are given in the original article. A
window on the top of the membrane unit allows entry
of the laser beam from the laser triangulometer as well
as in-line observation of particle deposition. Diatomaceous earth with variable particle size distributions and
monodisperse silica particles of variable sizes in the
range of 0.241.5 mm diameter were filtered in independent experiments. The feed volume fraction of particles in all experiments was 0.3%.
Results of experimental observations by Altmann and
Ripperger w52x are shown in Fig. 11. The data indicate
that a particle cake layer formed rapidly at the initial
stage of filtration and gradually progressed towards a
steady thickness. The formation of the cake layer
occurred concurrently with a decrease in the permeate
flux. The resistance of the cake layer is therefore evident.
The thickness of the cake layer, however, increased at a
faster rate than the decrease in permeate flux during the
initial stage of filtration. Furthermore, after a steady
state cake thickness was reached, the permeation flux
continued to decline. Therefore, a direct link between
cake layer thickness and rate of flux decline was
ambiguous and additional factors such as changes in
cake structure might be contributing to the observed
flux behavior.
Incremental computations of the cake layer resistance
during the filtration process revealed that the top portion
of the cake layer exerts a considerably higher resistance
than the bottom portion. Hence, it indicates that the
100
Fig. 11. Comparison of theoretical calculations and experimental results of changes in cake layer thickness and filtration rate with time. The
conditions used in the microfiltration experiments are (a) top figure: applied pressure of 1.0 bar and a crossflow velocity of 1.5 mys; (b) bottom
figure: applied pressure of 1.5 bar and a crossflow velocity of 1.5 mys. Figures from Altmann and Ripperger w52x.
cake structure varied during the filtration process, with

smaller particles depositing in greater proportions at the
top of the cake layer leading to the increased resistance.
The experimental observations in Fig. 11 were compared with theoretical predictions of cake growth.
Details of the theoretical model for cake growth are
given by Altmann and Ripperger w52x. Results showed
good agreement between experimental data and model
predictions. However, it should be noted that the cake
porosity was assumed to be 0.4 in the theoretical model,
which may not be representative of the actual cake
layer. Pressure relaxation experiments were also performed to investigate the reversibility of cake layer
formation. Upon pressure relaxation, the deposited cake
layer persisted on the membrane and did not reentrain
into the bulk. This finding is consistent with theoretical
model predictions that the adhesive forces among particles in the cake layer exceeded the drag force from the
crossflow under the operating conditions used.
4.1.4. Optical laser sensor

Hamachi and Mietton-Peuchot w53,54x developed an
optical laser sensor method for investigating the thickness of the cake layer during microfiltration of bentonite
suspensions. The principle of this technique is that the
formation of the deposit layer will absorb light from a
bypassing laser beam. The variation of the signal intensity detected after the laser beam has traversed through
the cake layer will correspond to the deposit thickness.
A calibration procedure with known values of cake layer
thickness is performed to extract the correlation.
The experimental apparatus consists of one tubular
membrane module with two optical windows through
which the laser beam traverses tangentially over the
membrane surface. Upon passing through the membrane
module, the light source is captured by a photomultiplier
that converts the intensity of light into an equivalent
voltage. Operating conditions during filtration were
crossflow velocities from 0.05 to 0.5 mys and transmem-
101
Fig. 12. Cake layer thickness as a function of cumulative permeate volume. (a) Effect of crossflow velocity (applied pressures110 kPa, particle
concentrations0.375 gyl). (b) Effect of particle concentration (applied pressures110 kPa, crossflow velocitys0.055 mys). Figures from Hamachi
and Mietton-Peuchot w53x.
brane pressures of 110210 kPa. Bentonite particles

with average size 2.45 mm were filtered. Details of the
experimental setup are given by Hamachi and MiettonPeuchot w53,54x.
Calibration was conducted using a micrometric screw
device that probes the gap distance between the cake
surface and the membrane surface. The measured gap
distance was then correlated to its corresponding signal
intensity from the optical laser sensor. Details of the
calibration procedure can be found in Hamachi and
Mietton-Peuchot w54x.
Examples of experimental measurements of cake layer
thickness with respect to variation in operating conditions are shown in Fig. 12. As expected, the cake
thickness was shown to increase with increasing transmembrane pressure, increasing bulk concentration and
decreasing crossflow velocity. Thicker cake layers were
formed with more concentrated bulk suspensions for the
same volume of filtrate. Further analysis was performed
by using Darcys law to calculate cake layer resistance.
The specific cake layer resistance of the deposit was
shown to be higher when filtering more dilute suspensions. One proposed reason for this observation is that
dilute particle suspensions contain fewer large particles
and thus form less permeable deposits. At a given
deposit thickness, the resistance of the cake layer was

also found to increase with increasing transmembrane
pressure, which was attributed to cake compaction. The
CarmanKozeny equation was utilized to calculate the
cake layer porosity. The porosity was shown to increase
with increasing cake thickness as larger particles begin
to deposit during the development of the deposit layer.
The porosity also was shown to decrease with increasing
transmembrane pressure, which, as described earlier, is
attributed to cake compaction. It is to be noted that the
cake layer structure depicted by Hamachi and MiettonPeuchot w53x is in disagreement with the conclusion by
Altmann and Ripperger w52x, which indicated a decrease
in porosity with the development of the cake layer,
using laser triangulometry (Section 4.1.3).
The maximum concentration that can be filtered in
the studies described above was 375 mgyl, above which
the turbidity of the suspension obstructs the photodetection. However, this limitation can be extended with use
of a more powerful laser. A HeNe laser with power
rating of 0.2 mW was employed in these experiments.
In addition, the maximum cake thickness that could be
measured was 30 mm. This restriction is due to the
dimensions of the membrane channel and the choice of
calibration procedure. However, this restriction can be
102
circumvented with adjustment of the experimental

apparatus.
4.1.5. Ultrasonic time-domain reflectometry (UTDR)
UTDR is a technique that uses sound waves to
measure the location of a moving or stationary interface
and can provide information on the physical characteristics of the media through which the waves travel.
When ultrasonic waves encounter an interface between
two media, energy will generally be partitioned such
that a reflected wave occurs, which can be detected by
an ultrasonic transducer. The amplitude of the reflected
wave relative to the incident wave is determined by the
acoustic impedance difference between the two media
as well as the topography of the interface. The time
interval between the initiation of the incident wave and
the detection of the reflected wave can be used to
measure the distance between interfaces. Detailed theories of UTDR are given by Ensminger w55x.
Mairal et al. w56x used the UTDR technique for in
situ measurement of membrane fouling during reverse
osmosis of calcium sulfate solutions. Their experimental
apparatus consisted of a stainless steel rectangular RO
cell with dimensions 60 cm (L)=6 cm (W)=0.2 cm
(H). Six ultrasonic measurement ports were situated on
the top plate where ultrasonic transducers were mounted.
The axial velocity used ranged from 0 to 4.6 cmys and
the transmembrane pressure was 4.14 MPa. During the
membrane filtration, the incident wave reflects off both
the membrane and fouling layer surfaces. Therefore, the
difference in signal delay between the two reflected
waves is a measure of the cake fouling layer thickness.
The results of the filtration experiments by Mairal et
al. w56x showed that the development of the fouling
layer was insufficient to produce two separate reflected
waves from which the cake thickness could be determined. However, the development of the cake layer was
shown to be closely related to the decline of the
amplitude of the reflected wave. The amplitude of the
reflected wave was thus taken to be the indicator and
gauge of cake formation. The attenuation of the amplitude of the reflected wave does correlate well with the
measured flux decline. Therefore, detection of the amplitude of the reflected wave does provide indication of
the formation of the cake layer. The actual quantitative
thickness of the cake layer, however, could not be
obtained from the amplitude measurements. The amplitude measurements provided only qualitative characterization of the cake formation process that may not be
sufficiently specific.
The major drawback of the study is that the desired
method of measuring the cake thickness by creating two
reflected waves and measuring the difference in their
time delay of signals could not be achieved. The
amplitude measurement was taken as a substitute technique to solve the predicament; however, it is neither
quantitative nor qualitatively specific. It is also likely

that the choice of the foulant, calcium sulfate, did not
result in classical cake layer but rather scale on the
membrane surface with limited thickness. Mairal et al.
w56x stated that further investigation is necessary to fully
document the quantitative capabilities of this technique.
Reinsch et al. w57x applied UTDR to investigate the
effects of membrane compaction on flux decline during
gas separation. As compaction proceeds upon the application of pressure, the change in the location of the
membrane surface leads to shifts in the arrival time of
the reflected wave. The displacement of the membrane
surface due to compaction can then be calculated from
the measured time shift. The experimental setup in this
study was in principle similar to that used by Mairal et
al. w56x, except that the medium was gas.
The use of time-domain reflectometry by Reinsch et
al. w57x achieved the desired measurement of changes
in arrival time of the reflected signal without having to
resort to amplitude attenuation approaches as a substitute. The resolution of the method was "18 mm for
detection of the change in location of the membrane
surface. This method shows promising outlook for aiding
previously mentioned techniques where the location of
the membrane varies during experimentation and is a
source of error. The extension of the medium from gas
to liquid is required. However, liquid phases generally
are better suited for propagation of mechanical waves
with less amplitude attenuation than that of gas phases.
The resolution of the method does incur a limitation
that the observed strain could not be specifically attributed to the membrane skin layer or the porous substructure. Such definition of the location of strain would
require resolution on the scale of nanometers. This may
be attainable with more sophisticated hardware
components.
Recent developments in ultrasonic reflectometry are
beginning to approach the sought-after measurable difference in arrival times for the reflected waves in a
liquid medium. Li and Sanderson w58x present an UTDR
technique that effectively detects separate signals from
the surfaces of the membrane and particle deposit in the
filtration of a liquid suspension. From the difference in
arrival times, the thickness of the deposit layer can be
readily determined by a simple relationship between the
speed of the waves and their traversed displacement in
the time interval. The experimental setup is similar to
that of Mairal et al. w56x with the exception that
relatively large kaolin particles of size 2 mm were used
in microfiltration (Pall nylon membranes of pore size
0.2 mm and applied pressure of 100 kPa). This may
then point to a key element for successful UTDR,
namely, an adequate particle size and correspondingly
thick cake layer. The possibility of obtaining a detectable
signal difference appears to be influenced by the particle
size of the deposit with larger particles allowing for
improved measurements. However, Li and Sanderson do

not analyze the effects of particle size and cake layer
thickness in their experiments.
The experimental results of Li and Sanderson w58x
show definitively that the formation of a deposit layer
has been captured by the UTDR technique. Comparisons
of the UTDR measurements with post-experimental
SEM images affirm that the realized signal differences
do reflect the formation of an appreciable deposit layer.
However, Li and Sanderson do not proceed from their
qualitative findings and perform quantitative computations of the deposit layer thickness and rate of cake
growth. Rather, they only validate the qualitative correctness of their method but not its quantitative accuracy.
The contribution of Li and Sanderson offers promising
potential for the UTDR technique from which more
rigorous applications should be added.
4.1.6. Electrical impedance spectroscopy
Chilcott et al. w59x and Gaedt et al. w60x proposed the
possible use of electrical impedance spectroscopy (EIS)
for characterizing membrane properties and investigating
membrane fouling. This was an attempt to extend earlier
work where this technique was applied to probe biological membranes and interfaces w61x. Unlike typical EIS
methods, which require electrodes in the feed and
permeate solutions, an alternating current is injected
directly into the membrane via external electrical contacts with the edges of the membrane. A metal layer
sputtered onto the membrane surface is used to enhance
conduction properties. The flow of the current across
the membrane surface leads to dispersion of the current
into the bulk solution and the membrane pores. This
dispersion phenomenon is characterized by the capacitance and conductance of various components of the
system, such as the membrane material and the bulk
solution, and possible polarization or fouling layer. The
dispersion of the current changes as foulants accumulate
on the membrane surface leading to alteration of the
capacitive and conductive properties of the membrane
interfacial region. Measuring the changes in the capacitance dispersion of the system therefore becomes a
means of monitoring in situ accumulation of particulates
that could potentially foul the membrane. Characterization of the dispersion behavior of the clean membrane
could also yield information pertaining to membrane
properties, such as porosity. The system was modeled
by the MaxwellWagner equation that characterizes the
impedance as a function of the capacitance and conductance of the components of the system. Details of the
experimental setup and the mathematical formulation of
the analysis are given by Chilcott et al. w59x and Gaedt
et al. w60x.
The method introduced in these studies, however, was
not applied to measurement of fouling behavior in
membrane filtration. It was presented only as a technique
103
with the potential of achieving such measurements. The

actual performance of the method was cited to be
included in subsequent publications. One major limitation of the proposed method is that it requires the
surface of the membrane to be coated with thin metal
films. The coating of the membrane surface not only
departs from a true representation of the system but may
also occlude membrane pores and alter the experimental
conditions. Thus, while the technique may be used to
evaluate properties of fouled membranes, its application
for in situ observation of the dynamics of fouling
behavior is questionable.
4.2. Pore blockage
Although most fouling studies focus on foulant deposition on the membrane surface and the ensuing dynamics of the foulant (deposit) layer, internal fouling caused
by blockage of membrane pores is important for microfiltration and ultrafiltration, at least at the initial stages
of filtration. More importantly, fouling by pore blockage
is often irreversible and very difficult to clean. An
accurate description of the foulant layer structure and
identification of the foulant location in the membrane
system, i.e. in membrane pores or on the membrane
surface, is thus very important for understanding fouling
mechanisms, developing low fouling membranes or
creating favorable hydrodynamic conditions to minimize
fouling.
4.2.1. Small-angle neutron scattering (SANS)
Su et al. w6264x utilized SANS to carry out in situ
studies of membrane pore blockage during ultrafiltration
of protein solutions by alumina membranes. SANS is
an established technique for investigating the structure
and distribution of aggregates in suspensions as well as
the surface layer structure in some porous media. Details
on the basic principles of SANS are given by Su et al.
w6264x. For porous media, the neutron scattering intensity can be related to the media structure, including
volume fraction of the pores, relative location of the
pores and geometrical structure of the pores. Therefore,
changes in membrane pore structure caused by foulant
deposition within membrane pores can be monitored by
detecting changes in SANS intensity with time.
A schematic of the experimental setup used by Su et
al. w62x to study protein fouling using SANS is shown
in Fig. 13. The membrane cell consists of two quartz
windows that are transparent to neutron beams. A
neutron beam is passed perpendicularly through the
membrane and captured by a detector. The membrane
module was operated in crossflow mode at a flow rate
of 10 mlymin, corresponding to a Reynolds number of
24. The pressure within the module was maintained at
0.5 bar. BSA solutions in buffered D2O at concentrations
ranging from 0.1 to 1 gycm3 were filtered. More details
104
Fig. 13. Schematic diagram of (a) the filtration system and (b) the traverse of the neutron scattering beam through the filtration cell. Under
operational and scattering conditions, the alumina membrane is fully immersed in D2O solution. Adapted from Su et al. w62x.
of the experimental conditions are given by Su et al.

w6264x.
The results of the studies described above revealed
that adsorption of foulant proteins on the wall of
membrane pores caused greater diffusion of the neutrons
from their intended path than that in the case of the
bare alumina wall. The result was then a faster decay in
scattering intensity for the blocked membrane pores.
Therefore, the variation of SANS intensity was used to
monitor pore blockage by the foulant protein. By using
a simple model, the authors were able to estimate the
thickness of the adsorbed layer in membrane pores and
the percentage of the pore surface area that is covered
by the foulant. More importantly, it was found that cake
layer of BSA formed at the front surface of the
membrane had no effect on the measured neutron
scattering intensity. This finding indicates that the
change in measured scattering intensity is only attribut-
able to pore blockage and that SANS can quantitatively

describe foulant deposit in membrane pores. It also
suggests that combination of SANS measurement with
permeate flux data can provide information on fouling
mechanisms and ascertain the location of foulants, since
flux decline is caused by fouling both at the front
membrane surface and within the membrane pores.
4.3. Characterization of cake structure
Pignon et al. w65x presented static light scattering
(SLS), SANS and local birefringence techniques to
quantify the inner structure of deposited colloid fouling
layers on the membrane surface. In SLS and SANS,
incident light or neutrons were directed through the cake
layer resulting in scattering through the deposit. The
scattering intensity was then related to the fractal dimension of the deposit. In local birefringence measurements,
Fig. 14. Schematic diagram of the local birefringence measuring apparatus. Adapted from Pignon et al. w65x.
a laser beam traverses through the deposited cake layer

and the intensity of the transmitted light was analyzed
by lock-in amplifiers to determine the optical anisotropy
of the deposit. The optical anisotropy was related to the
local birefringence and was used to interpret the orientation of the particles within the deposit. The principles
of SLS, SANS and local birefringence techniques are
described in Pignon et al. w65x.
The filtration cell used by Pignon et al. w65x consists
of two bottomless rectangular quartz cells located above
and below the membrane. The unit was operated in
frontal (dead-end) filtration mode. The incident light or
neutron beam traverses through the deposit layer above
the membrane surface and is captured by a detector. In
the local birefringence setup, additional optical devices
are required. A schematic of the birefringence setup is
shown in Fig. 14. The incident beam is polarized and
then crosses a photo-elastic modulator (PEM) set at 458
to the polarizer. The PEM consists of a birefringent
blade-optic and piezoelectric plate, a quarter wave plate
and a converging lens. Upon transmission through the
sample, the beam traverses through a circular polarizer
consisting of a quarter wave plate and a straight polarizer
at 458 to each other. Finally, the transmitted beam is
captured by a detector.
Suspensions of polydisperse disc-shaped laponite clay
particles with a diameter of approximately 30 nm and a
thickness of 1 nm were used in the membrane filtration
experiments. A stabilizing chemical agent (peptizer)
(tetrasodium pyrophosphate) is added in some experimental runs to alter the rheology of the suspension. The
peptizer has been shown to convert the suspension from
a gel state to a liquid suspension (sol). Details of the
experimental setup are given by Pignon et al. w65x.
In SLS and SANS measurements, the scattering
behaviors display a power law relationship from which
the fractal dimension can be derived. The fractal dimension provides useful insight into the structure of the
colloid deposit layer above the membrane. The deposit
105
of the clay suspension without peptizer exhibited a

fractal dimension of 2 according to the slope of the
intensity profile. With the addition of the peptizer, the
deposit exhibited a fractal dimension of 1.3. A more
compact structure of deposit as indicated by a higher
fractal dimension was obtained when filtering a wellstructured gel suspension. Integration of the intensity
profile along two directions y and z reveals that the
intensity decays much less along the y direction than
the z. Such anisotropy implies that the short axes of the
clay discs are oriented along the y direction. Therefore,
the orientation of particles within the deposit may be
derived from SANS and SLS measurements. Fitting the
intensity data to Guiniers law also yields the halfdimension of the clay discs, which is its thickness. In
this case, the fitting produced a value of 1.21 nm for
the thickness of the clay platelets, in good agreement
with the nominal value of 1 nm.
For both cases, with and without the peptizer, considerable birefringence was found near the membrane,
which decreased with increasing distance from the
membrane surface and became 0 at a certain distance
that is roughly equal to the thickness of the colloid
deposit layer. This signals that the deposit is more
anisotropic near the membrane surface. The birefringence profile of the suspension without peptizer fluctuated, indicating a random structure for the deposit. The
results show that a thicker deposit layer was formed
when filtering suspensions containing no peptizer. This
is expected as the peptizer induces a more compact and
ordered structure within the deposit.
Pignon et al. w66x have advanced their previously
described observation techniques w65x with a combination of small angle X-ray scattering (SAXS) and magnetization application to simultaneously characterize the
structure of particle deposits during ultrafiltration,
manipulate the particle orientations and enhance filtration performance. The general approach remains the
same as the case of SANS, with SAXS used in its place.
The innovation of this latest method is the application
of a magnetic field to the filtration cell that functions to
effectively alter the orientation of particles within the
deposits.
The underlying theory of this technique relies on the
capability of the applied magnetic field to create a
directional anisotropy in the deposit such that particles
within the deposit become predominantly oriented in a
preferred direction. This process facilitates formation of
deposits of higher compactness that translates to
increased concentration of particles in the deposit. Concurrently, the compactness is of an orderly nature and
so flow channels for the permeate are less tortuous than
the less compact but random deposits. This results in an
increase in permeability even for the more concentrated,
compact deposit. Control of particle orientation induces
the surprising, antithetical coincidence of a more con-
106
centrated particle deposit while gaining improved filtration performance and lower deposit specific resistance.
The observation technique by SAXS competently captures these phenomena and specifically quantifies them
in terms of scattering intensities and an anisotropy
parameter. Thus, SAXS not only proves to be capable
of quantifying the thickness and compactness of particle
deposits, but also extends its application to attaining the
deposit anisotropy and exploiting this property to
achieve optimized membrane operations.
The experimental setup is similar to the previous
work of Pignon et al. w65x with the exception of X-ray
being the probing beam. The filtration cell is made from
polycarbonate, and its dimensions are 4=1=38 mm3.
Amicon Millipore asymmetric polysulfone organic ultrafiltration membranes, with an average pore size of 50
nm, were used in the filtration process with montmorillonite particles of radius 500 nm. The montmorillonite
particles are saturated with Fe2q by ion exchange to
enhance their responsiveness to a magnetic field. Details
of the conditioning of the particles with Fe2q are given
by Pignon et al. w66x.
The filtration cell is flanked by the magnetic setup
consisting of two stacks of five magnetic elements
mounted on a motorized jaw. The flow is positioned at
the center of the generated magnetic fields. The magnetic setup is capable of delivering field strength of up
to 1.43 T over 1 cm3. Scattering measurements by
SAXS and ultra-SAXS (USAX) are performed. The
data acquisition apparatus consists of a two-dimensional
detector and a Bonse-Hart camera for SAXS and USAX,
respectively.
The observation technique of Pignon et al. w66x does
impose certain restrictive conditions. First, the particles
must be preconditioned with Fe2q to be responsive to
the magnetic field. Experiments conducted with Fe2q
free particles show greatly reduced effects from the
magnetic field. The ion-exchange process with Fe2q
causes the experiments to deviate from real conditions
of interest. However, the preconditioning with Fe2q may
also be regarded as a form of proposed pretreatment of
particles to enhance filtration performance. Second, the
magnetic field can only generate a preferred directional
anisotropy for non-spherical particles, for instance ellipsoids. The re-orientation of particles would not exhibit
a preferable direction of alignment in cases where they
are spherically shaped.
5. Concluding remarks
Laboratory-scale in situ techniques for probing concentration polarization and fouling phenomena in pressure-driven membrane filtration can provide valuable
insights into the mechanisms controlling these phenomena. Such techniques can provide detailed information
on changes in local (i.e. along a filtration channel) and
transverse (i.e. perpendicular to the membrane surface)

properties of the solute polarization layer or cake. The
obtained information will enable assessment and refinement of current models and theories for concentration
polarization and cake formation. A mechanistic understanding of fouling phenomena will lead to refinement
of strategies to control and minimize fouling in
membrane systems as well as development of chemical
and physical cleaning techniques for fouled membranes.
Despite the enormous potential of in situ monitoring
techniques to advance our quantitative understanding of
membrane filtration, it is quite surprising that the available investigations are quite limited in scope. The
reviewed studies provided only qualitative or perhaps
semi-quantitative information on concentration polarization, cake formation and fouling phenomena. Several
of the techniques need refinements to improve accuracy
and resolution. The latter is quite important as many
transport processes within the concentration polarization
layer vary over a short length scale (on the order of a
few micrometers) near the membrane surface. The latest
advancements in the fields of nanotechnology, microfluidics, optics, spectroscopy and sensors offer vast
opportunities for development of robust in situ monitoring techniques in membrane filtration.
Acknowledgments
The work was supported by the National Science
Foundation (Grant BES 0114527).
References
w1x L.J. Zeman, A.L. Zydney, Microfiltration and Ultrafiltration:
Principles and Applications, Marcel Dekker, New York, 1996.
w2x G. Belfort, R.H. Davis, A.L. Zydney, The behavior of suspensions and macromolecular solutions in crossflow microfiltration, J. Membrane Sci. 96 (1994) 158.
w3x M.G. van der Waal, I.G. Racz, Mass transfer in corrugatedplate membrane modules. I. Hyperfiltration experiments, J.
Membrane Sci. 40 (1989) 243260.
w4x R. van Reis, S. Gadam, L.N. Frautschy, et al., High performance tangential flow filtration, Biotechnol. Bioeng. 56 (1997)
7182.
w5x G.B. van den Berg, C.A. Smolders, Flux decline in ultrafiltration processes, Desalination 77 (1990) 101133.
w6x V.L. Vilker, C.K. Colton, K.A. Smith, Concentration polarization in protein ultrafiltration. Part II: theoretical and experimental study of albumin ultrafiltered in an unstirred batch cell,
AIChE. J. 27 (1981) 637645.
w7x J.G. Wijmans, S. Nakao, C.A. Smolders, Flux limitation in
ultrafiltration-osmotic pressure model and gel layer model, J.
Membrane Sci. 20 (1984) 115124.
w8x E.S. Tarleton, R.J. Wakeman, Understanding flux decline in
crossflow microfiltration. 1. Effects of particle and pore size,
Chem. Eng. Res. Design 71 (A4) (1993) 399410.
w9x L.F. Song, Flux decline in crossflow microfiltration and ultrafiltration-mechanisms and modeling of membrane fouling, J.
Membrane Sci. 139 (1998) 183200.
w10x R.S. Faibish, M. Elimelech, Y. Cohen, Effect of interparticle
electrostatic double layer interactions on permeate flux decline
in crossflow membrane filtration of colloidal suspensionsan
experimental investigation, J. Colloid Interface Sci. 204 (1998)
7786.
w11x C.A. Romero, R.H. Davis, Global-model of cross-flow microfiltration based on hydrodynamic particle diffusion, J.
Membrane Sci. 39 (1988) 157185.
w12x C.A. Romero, R.H. Davis, Transient model of cross-flow
microfiltration, Chem. Eng. Sci. 45 (1990) 1325.
w13x L.F. Song, M. Elimelech, Theory of concentration polarization
in crossflow filtration, J. Chem. Soc. Faraday Trans. 91 (1995)
33893398.
w14x M. Mulder, Basic Principles of Membrane Technology, second
ed., Kluwer Academic, Dordrecht, The Netherlands, 1996.
w15x M.C. Porter, Concentration polarization with membrane ultrafiltration, Ind. Eng. Chem. Prod. Res. Dev. 11 (3) (1972)
234248.
w16x W.B. Russel, The Dynamics of Colloidal Systems, University
of Wisconsin Press, Madison, WI, 1987.
w17x M.R. Mackley, N.E. Sherman, Cross-flow cake filtration mechanisms and kinetics, Chem. Eng. Sci. 47 (1992) 30673084.
w18x D.R. Trettin, M.R. Doshi, Pressure independent ultrafiltration
is it gel limited or osmotic pressure limited? In: A.F. Turbak
(Ed.), Synthetic Membranes: Hyper and Ultrafiltration Uses,
vol II, P 373, ACS Symp. Ser. 154, 1981.
w19x V.L. Vilker, C.K. Colton, K.A. Smith, Concentration polarization in protein ultrafiltration. Part I: An optical shadowgraph
technique for measuring concentration profiles near a solution
membrane interface, AIChE. J. 27 (1981) 632637.
w20x S. Bhattacharjee, A. Sharma, P.K. Bhattacharya, A unified
model for flux prediction during batch cell ultrafiltration, J.
Membrane Sci. 111 (1996) 243258.
w21x W.R. Bowen, F. Jenner, Theoretical descriptions of membrane
filtration of colloids and fine particles: an assessment and
review, Adv. Colloid Interface Sci. 56 (1995) 141200.
w22x D.N. Petsev, V.M. Starov, I.B. Ivanov, Concentrated dispersions
of charged colloidal particles: sedimentation, ultrafiltration and
diffusion, Colloid. Surf. A 81 (1993) 6581.
w23x S. Bhattacharjee, A.S. Kim, M. Elimelech, Concentration
polarization of interacting solute particles in crossflow
membrane filtration, J. Colloid Interface Sci. 212 (1999)
8199.
w24x M. Elimelech, S. Bhattacharjee, A novel approach for modeling
concentration polarization in crossflow membrane filtration
based on equivalence of osmotic pressure model and filtration
theory, J. Membrane Sci. 145 (1998) 223241.
w25x J. Happel, H. Brenner, Low Reynolds Number Hydrodynamics,
Kluwer, Dordrecht, 1991.
w26x W.B. Russel, D.A. Saville, W.R. Schowalter, Colloidal Dispersions, Cambridge University Press, Cambridge, 1989.
w27x A.A. Kozinski, E.N. Lightfoo, Protein ultrafiltration: a general
example of boundary-layer filtration, AIChE. J. 18 (1972)
10301038.
w28x C.R. Ethier, D.C. Lin, Refractometric measurement of polarized
layer structure: studies of hyaluronic acid ultrafiltration, J.
Membrane Sci. 68 (1992) 249261.
w29x L.M. Gowman, C.R. Ethier, Concentration and concentration
gradient measurements in an ultrafiltration concentration polarization layer. Part I: A laser-based refractometric experimental
technique, J. Membrane Sci. 131 (1997) 95105.
w30x L.M. Gowman, C.R. Ethier, Concentration and concentration
gradient measurements in an ultrafiltration concentration polarization layer. Part II: Application to hyaluronan, J. Membrane
Sci. 131 (1997) 107123.
107
w31x C.R. Ethier, The hydrodynamic resistance of hyaluronic acid:

estimates from sedimentation studies, Biorheology 23 (1986)
99113.
w32x L.S. Lam, J.L. Bert, Hydraulic flow conductivity of hyaluronic
acid solutions: effects of concentration and molecular weight,
Biorheology 27 (1990) 789795.
w33x R.M. Peitzsch, W.F. Reed, High osmotic stress behavior of
hyaluronate and heparin, Biopolymers 32 (1992) 219238.
w34x M. Johnson, Transport through the aqueous outflow system of
the eye, Ph.D. thesis, Massachusetts Institute of Technology,
Cambridge, MA, 1987.
w35x S.W. Young, Magnetic Resonance Imaging: Basic Principles,
second ed., Raven Press, NewYork, 1988.
w36x M.T. Vlaadingerbroek, J.A. den Boer, Magnetic Resonance
Imaging: Theory and Practice, Second Review, Springer, New
York, 1999.
w37x S. Yao, M. Costello, A.G. Fane, J.M. Pope, Non-invasive
observation of flow profiles and polarisation layers in hollow
fibre membrane filtration modules using NMR micro-imaging,
J. Membrane Sci. 99 (1995) 207216.
w38x P.T. Callaghan, Y. Xia, Veloctiy and diffusion imaging in
dynamic NMR microscopy, J. Magn. Reson. 91 (1991)
326352.
w39x P.T. Callaghan, C.D. Eccles, Y. Xia, NMR microscopy of
dynamic displacements: k-space and q-space imaging, J. Phys.
Part E. Sci. Intrum. 21 (1991) 820822.
w40x J.M. Pope, S. Yao, A.G. Fane, Quantitative measurements of
the concentration polarisation layer thickness in membrane
filtration of oilwater emulsion using NMR micro-imaging, J.
Membrane Sci. 118 (1996) 247257.
w41x D. Airey, S. Yao, J. Wu, V. Chen, A.G. Fane, J.M. Pope, An
investigation of concentration polarization phenomena in
membrane filtration of colloidal silica suspensions by NMR
micro-imaging, J. Membrane Sci. 145 (1998) 145158.
w42x S. Yao, A.G. Fane, J.M. Pope, An investigation of the fluidity
of concentration polarisation layers in crossflow membrane
filtration of an oilwater emulsion using chemical shift selective flow imaging, Magn. Reson. Imaging 15 (2) (1997)
235242.
w43x R.M. McDonogh, H. Bauser, N. Stroh, d.H. Chmiel, Separation
efficiency of membrane in biotechnology: an experimental and
mathematical study of flux control, Chem. Eng. Sci. 47 (1)
(1992) 271279.
w44x R.M. McDonogh, H. Bauser, N. Stroh, U. Grauschopf, Experimental in situ measurement of concentration polarisation
during ultra- and micro-filtration of bovine serum albumin and
dextran blue solutions, J. Membrane Sci. 104 (1995) 5163.
w45x S.S.L. Peppin, J.A.W. Elliott, Non-equilibrium thermodynamics
of concentration polarization, Adv. Colloid Interf. Sci. 92
(2001) 172.
w46x W. Zhang, C.R. Ethier, Direct pressure measurements in a
hyaluronan ultrafiltration concentration polarization layer, Colloid. Surface. A: Physicochem. Eng. Aspects 180 (2001)
6373.
w47x A. Kim, C.H. Wang, M. Johnson, R. Kamm, The specific
hydraulic conductivity of bovine serum-albumin, Biorhelogy
28 (5) (1991) 401419.
w48x H. Li, A.G. Fane, H.G.L. Coster, S. Vigneswaran, Direct
observation of particle deposition on the membrane surface
during crossflow microfiltration, J. Membrane Sci. 149 (1998)
8397.
w49x H. Li, A.G. Fane, H.G.L. Coster, S. Vigneswaran, An assessment of depolarisation models of crossflow microfiltration by
direct observation through the membrane, J. Membrane Sci.
172 (2000) 135147.
108
w50x S. Kang, E.M.V. Hoek, M.A. Deshusses, M. Matsumoto, A

novel quantification method for the initial deposition of microorganisms onto a membrane surface, Water Sci. Technol., in
press.
w51x W. Mores, R.H. Davis, Direct visual observation of yeast
deposition and removal during microfiltration, J. Membrane
Sci. 189 (2001) 217230.
w52x J. Altmann, S. Ripperger, Particle deposition and layer formation at the crossflow microfiltration, J. Membrane Sci. 124
(1997) 119128.
w53x M. Hamachi, M. Mietton-Peuchot, Experimental investigation
of cake characteristics in crossflow microfiltration, Chem. Eng.
Sci. 54 (1999) 40234030.
w54x M. Hamachi, M. Mietton-Peuchot, Cake thickness measurement with an optical laser sensor, Chem. Eng. Res. Des. 79
(A2) (2001) 151155.
w55x D. Ensminger, Ultrasonics, Marcel Dekker, New York, 1988.
w56x A.P. Mairal, A.R. Greenberg, W.B. Krantz, L.J. Bond, Realtime measurement of inorganic fouling of RO desalination
membranes using ultrasonic time-domain reflectometry, J.
Membrane Sci. 159 (1999) 185196.
w57x V.E. Reinsch, A.R. Greenberg, S.S. Kelley, R. Peterson, L.J.
Bond, A new technique for the simultaneous, real-time measurement of membrane compaction and performance during
exposure to high-pressure gas, J. Membrane Sci. 171 (2000)
217228.
w58x J. Li, R.D. Sanderson, In situ measurement of particle deposition and its removal in microfiltration by ultrasonic timedomain reflectometry, Desalination 146 (2002) 169175.
w59x T.C. Chilcott, M. Chan, L. Gaedt, T. Nantawisarakul, A.G.

Fane, H.G.L. Coster, Electrical impedance spectroscopy characterisation of conducting membranes I. Theory, J. Membrane
Sci. 195 (2002) 153167.
w60x L. Gaedt, T.C. Chilcott, M. Chan, T. Nantawisarakul, A.G.
Fane, H.G.L. Coster, Electrical impedance spectroscopy characterisation of conducting membranes II. Experimental, J.
Membrane Sci. 195 (2002) 169180.
w61x H.G.L. Coster, T.C. Chilcott, A.C.F. Coster, Impedance spectroscopy of interfaces, membranes and ultrastructures, Bioelectochem. Bioenerg. 40 (2) (1996) 7998.
w62x T.J. Su, J.R. Lu, Z.F. Cui, R.K. Thomas, R.K. Heenan,
Application of small angle neutron scattering to the in situ
study of protein fouling on ceramic membranes, Langmuir 14
(1998) 55175520.
w63x T.J. Su, J.R. Lu, Z.F. Cui, B.J. Bellhouse, R.K. Thomas, R.K.
Heenan, Identification of the location of protein fouling on
ceramic membranes under dynamic filtration conditions, J.
Membrane Sci. 163 (1999) 265275.
w64x T.J. Su, J.R. Lu, Z.F. Cui, R.K. Thomas, Fouling of ceramic
membranes by albumins under dynamic filtration conditions,
J. Membrane Sci. 173 (2000) 167178.
w65x F. Pignon, A. Magnin, J.M. Piau, et al., Structural characterisation of deposits formed during frontal filtration, J. Membrane
Sci. 174 (2000) 189204.
w66x F. Pignon, A. Alemdar, A. Magnin, T. Narayanan, Small-angle
X-ray scattering studies of Fe-montmorillonite deposits during
ultrafiltration in a magnetic field, Langmuir 19 (2003)
86388645.

2004 in Situ Monitoring Techniques For Concentration Polarization and Fouling Phenomena in Membrane Filtration

Uploaded by

Document Information

Original Title

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

2004 in Situ Monitoring Techniques For Concentration Polarization and Fouling Phenomena in Membrane Filtration

Uploaded by

Copyright:

Available Formats

Advances in Colloid and Interface Science 107 (2004) 83108

In situ monitoring techniques for concentration polarization and fouling

4.1.5. Ultrasonic time-domain reflectometry (UTDR) .................................................................................

tion polarization and fouling phenomena focused on

ends with conclusions and implications for membrane

Brownian motion of the solute molecules is frozen w26x.

Here, ap is the particle radius, DP the applied pressure,

In this equation, DPc can be considered as the osmotic

gradient along the light pathway. Two methods that

concentration profiles within a medium by exploiting

placed between the top and bottom sections of the cell.

of concentration. The experimental measurements of

magnitude of the deflection of light near the membrane

tion chemistry were adjusted to ensure the absence of

independent. The walls of the chambers were BK7 plate

showed that a decrease in the initial concentration of

Johnson w34x provided values of various parameter for

between theoretical prediction and experimental results.

sively employed diagnostic tool in clinical radiology for

of the inner lumens of the fibers, while the other end

direction of the permeate through the lumens and the

with increasing feed pressure. When the velocity maps

non-uniformities, the use of an average thickness for the

number of 215, it is observed that the velocity in the

a buffer solution alone for an additional hour. Following

The signal Sch was determined by filling the channel

of solute concentration can be achieved from signals

tration polarization. McDonogh et al. w44x did not,

the polarization layer was insignificant, yet appreciably

tion of particles onto a membrane surface and the

the onset of cake formation. Above the critical flux,

dimensions of the flow channel were 1 mm (height) by

surface. In addition, the system was used to study

elevation of the surface changes, the location where the

cake structure varied during the filtration process, with

4.1.4. Optical laser sensor

brane pressures of 110210 kPa. Bentonite particles

deposit thickness, the resistance of the cake layer was

circumvented with adjustment of the experimental

quantitative nor qualitatively specific. It is also likely

improved measurements. However, Li and Sanderson do

with the potential of achieving such measurements. The

of the experimental conditions are given by Su et al.

able to pore blockage and that SANS can quantitatively

a laser beam traverses through the deposited cake layer

of the clay suspension without peptizer exhibited a

transverse (i.e. perpendicular to the membrane surface)

w31x C.R. Ethier, The hydrodynamic resistance of hyaluronic acid:

w50x S. Kang, E.M.V. Hoek, M.A. Deshusses, M. Matsumoto, A

w59x T.C. Chilcott, M. Chan, L. Gaedt, T. Nantawisarakul, A.G.

You might also like