You are on page 1of 21

DESALINATION

ELSEVIER

Desalination 141 (2001) 269-289


www elsevier com/locate/desal

Concentration polarization in ultrafiltration and reverse osmosis:


a critical review
S S Sablani, M F A Goosena, R Al-Belushi, M Wilf!
aDepartment of Bioresource and Agriculture Engineering, Sultan Qaboos University,
PO Box 34, Al-Khod, PC-123 Muscat, Sultanate of Oman
Tel +968 515289, Fax +968 513418, email shyam@qu edu om
bHydranautics Inc , Oceanside, CA, USA

Received 3 January 2001; accepted 4 June 2001

Abstract

A primary reason for flux decline during the initial period of a membrane separation process is concentration
polarization of solute at the membrane surface This can occur in conjunction with irreversible fouling of the membrane
as well as reversible gel layer formation Experimental and mathematical studies have been performed by various groups
to gain a better understanding of concentration polarization phenomena in ultraf%ra~ionand reverse osmosis This article
critically reviews published studies on concentration polarization in both systems It presents progress made in
determination of, for example, critical or limiting flux, and recommends specific models such as surface renewal, and
experimental methods such as laser-based refractometry, for quantification of the problem
Keywow%

Concentration polarization, Modeling, Reverse osmosis, Ultrafiltration, Membrane separation processes

1. Introduction
Membrane
separation processes such as
ultrafiltration
(UF) and reverse osmosis (RO)
have gained considerable importance because
they offer superior treatment at relatively modest
capital and operating costs [l] These processes
can remove a wide range of contaminants such as
suspended and dissolved solids, organic matter,
*Corresponding author
00 1 l-9 164/01/$See
PII:SOOll-9164(01)00411-8

front matter 0

2001

heavy metals, bacteria and viruses, which are


present, for example, in untreated municipal and
industrial wastewater RO, in particular, is also
becoming of increasing interest to the food
industry It was used, for instance, by Alvarez et
al [2] to study apple juice concentration
The main task in optimal design of membrane
processes is to ensure maximum permeate flow
while having maximum solute rejection, with
minimum capital and operating costs [3] This
means that it is desirable to have a membrane

Elsevier Science B V All rights reserved

270

S.S. Sablani et al. /Desalination

lifetime as long as possible [4]. Membrane lifetime and permeate fluxes are primarily affected
by the phenomena of concentration polarization
(i.e., solute build-up) and fouling (e.g., microbial
adhesion,
gel layer formation
and solute
adhesion) at the membrane surface [5]. The latter
quite often is irreversible with respect to solute
adsorption. Membrane lifetimes may be enhanced
through the use of spacers between the membranes [6].
Koltuniewicz and Noworyta [4], in an excellent paper, summarized the phenomena responsible for limiting the permeate flux during cyclic
operation (i.e., permeation followed by cleaning).
During the initial period of operation within a
cycle, concentration polarization is one of the
primary reasons for flux decline, x (Fig. 1).
Large-scale
membrane systems operate in a
cyclic mode, where clean-in-place
operation
alternates with the normal run. The figure shows
a decrease in the flux for pure water from cycle
to cycle, J,(t), due to fouling; the flux decline
within a cycle due to concentration polarization,
$t,); and the average flux under steady-state
concentration, J:. The latter also decreases from
cycle to cycle, suggesting irreversible solute
adsorption or fouling. Accumulation ofthe solute
retained on a membrane surface leads to increasing permeate flow resistance at the membrane
wall region.
Concentration polarization is considered to be
reversible and can be controlled in a membrane
module by means of velocity adjustment, pulsation, ultrasound, or an electric field. Membrane
fouling, on the other hand, is more complicated
in that it is considered as a group of physical,
chemical, and biological effects leading to
irreversible
loss of membrane permeability.
Attempts to analyze the fouling phenomena, for
example, have shown that the main factors are
adsorption of feed components,
clogging of
pores, deposition of solids on the membrane
surface accompanied
by crystallization
and
compaction of the membrane structure, chemical

141 (2001) 269-289

Fig. 1. Diagram of typical flux-time dependency during


cyclic operation in large-scale ultrafiltration systems [4].

interaction between membrane material and


components of the solutions, gel coacervation,
and bacterial growth. Fouling can be reduced by
proper selection of the membrane material and/or
by membrane pretreatment
using surfactants,
polymers, and enzymes. Besides these methods,
fouling can be limited by adjusting the operating
parameters such as using an intermittent mode of
operation [7] and by using all means that reduce
concentration polarization.
By the late 196Os, tangential or cross-flow
filtration systems had been developed to combat
concentration polarization and membrane fouling. Forces produced by these feed systems
sheared liquid near the surface of the membrane
and thus removed trapped particles and solutes,
which tended to keep the membranes clean for
some period of time.
In the present article only concentration
polarization phenomena were critically reviewed
in detail. The paper is intended to provide a
starting point for beginners in the field. This was
done to provide a better focus on the progress
made in recent years in terms of quantifying the
phenomenon using experimental and/or mathematical methods. Specific models and experimental methods were recommended.

271

S.S. Sablani et al. /Desalination 141 (2001) 269-289


2. Elementary

feed
flow

transport mechanisms

2.1. Gel-polarization

model

The first model proposed to explain the


effects of concentration polarization in UF was
the gel-polarization model [ 81. The basic assumption of this model is that beyond a certain value
of applied pressure, the membrane permeation
rate is limited by the presence of a gel layer
deposited
on the membrane
surface which
increases the effective membrane thickness and
so reduces its hydraulic permeability. A seconwhich is implicit in the
dary assumption,
traditional version of this model, is that the
osmotic pressures of macromolecular
solutions
are always negligible. Later, several researchers
pointed out that this assumption is not really
correct: the concentration
of macromolecules
may have an appreciable osmotic pressure; in
fact, the osmotic pressure can be of the same
order of magnitude as the applied pressures
generally used in UF [9].
In 1970, a classic paper by Michaels and
coworkers [IO] provided the first comprehensive
analysis of concentration polarization in UF and
introduced the term gel polarization. If one
assumes that the limiting resistance to flow is in
the dynamically formed secondary or gel layer, it
is possible to calculate the transport rate of water
through the membrane (flux) on the basis of the
mass transfer of membrane-retained
species
(dissolved solutes or colloidal materials) from the
membrane surface back into the bulk stream [S]
(Fig. 2). This is because the dynamic gel layer is
assumed to have a fixed gel concentration (C,)
but is free to vary in thickness or porosity
(varying permeability or resistance to flow). In
such an analysis, the solvent flux (J,) will be
independent of the pressure-driving force or the
membrane permeability
since the gel layers
resistance to flow will adjust itself until the
convective transport of retained species to the
membrane surface (J&Z) by the solvent is just
equal to the back-diffusive transport [D(dC/d.x)]

membrane
JS solute

a
.

-:

40.
_a

.A

convecltive

fiow
+
*

- back diffusiot <


i..
SOlUE!

boundary layer

Fig. 2. A schematic representation

of concentration

polarization.

(Fig. 2). Thus, at steady state:


J=-Ddc
Y

dx

(1)

where D is the diffusion coefficient for solute


transport through solvent, C is the concentration
of membrane-retained solutes or colloidal species
and dC/dx is the solute concentration gradient.
The gel-polarization model [Eq. (l)] can be
integrated since the boundary conditions are
specified; the solute concentration at the membrane surface is fixed at an upper limit (i.e.,
saturation, C,), and the bulk-stream concentration
is known (C,). Therefore,

(2)
where 6 is the thickness of the boundary layer
over which the concentration of the solute varies.
It is assumed that under conditions where the gelpolarization model holds, the flux through the
membrane is invariant with (i.e., independent of)
transmembrane pressure drop or permeability and

272

S S Sablani et al /Desalination

is dependent only on the solute characteristics (D


and C,) and the boundary layer thickness (6) A
weakness is that this model is unable to describe
the whole range of flux-pressure dependency
Fluid management techniques must be directed
towards decreasing the boundary layer thickness
or, put another way, towards increasing the masstransfer coefficient, Ic, where

(3)

(4)

The validity of Eq (3) has been demonstrated


for a large number ofmacromolecular
solutes and
colloidal species The mass transfer-heat transfer
analogies well known in the chemical engineering literature make an evaluation of the masstransfer coefficient,
k, possible and provide
insight into how membrane geometry and fluid
flow conditions can be specified to optimize flux

2 2 Osmotic pressure model


The limiting flux in the osmotic pressure
model is regarded as being due to the increased
osmotic counter-pressure
produced by the high
concentration
of the rejected solute near the
membrane surface [l l] One of the osmotic
pressure models is the Spiegler-Kedem
and
solution-diffirsion model The working equations
of this model are [ 121
J, = L&Ap-o Ax)

R=

au-F)
U-OF)

(5)

141 (2001) 269-289

F = exp (-.JVa2)

(7)

with
a - (l-o)
(8)

2
PM

where R is the true rejection, Ap is the pressure


drop across the membrane, AIT is the osmotic
pressure drop across the membrane, 0 is the
reflection coefficient which represents the rejection capability of a membrane, (i e , U= 0 means
no rejection and 0 = 1 means 100% rejection),
and PM is the overall permeability coefficient
Combining the Eqs (5) to (8) with the film
theory model [Eq (2)] results in

&

= aI [l -exp(+,)]

[exp(-JJk)]

(9)

0
where a, = a/( 1-a) Using the nonlinear parameter estimation method and the experimental
data of observed rejection (R,) and the solvent
flux (J,) taken at a given pressure, feed rate and
concentration, the membrane parameters (J and
PM,and k can be estimated, simultaneously
2 3 Mass transfer coefficient
Most ofthe models used in characterization of
RO or UF membranes,
such as the threeparameter osmotic-pressure
model (SpieglerKedem model) and the (gel) concentration polarization model, make use of a mass transfer
coefficient correlation in order to calculate the
concentration at the membrane wall, C,, This
correlation has the empirical form based on the
Chilton-Colburn analogy [13,14]

(6)
(10)

where

S S Sablani et al /Desalination

where Sh is the Sherwood number, d,, the


hydraulic diameter of flow channel, A, to A,
empirical constants, Re the Reynolds number and
SC is the Schmidt number Based on the ChiltonColburn and Deissler analogies, Eq (10) can be
combined with the film theory (i e , gel layer)
equation [Eq (3)] to give

(11)
permitting calculation of C, when k is known
The above mass transfer correlations are
borrowed from non-porous smooth duct flow
Their applications
in the case of membrane
operation have been criticized by many authors
since neither porosity nor diffusivity due to
concentration polarization is taken into consideration [ 131 Gekas and Hallstrom [ 131 presented
an excellent review on the adaption of existing
Sherwood correlations to membrane operations
under turbulent duct flow Their paper included
a discussion on the factors influencing mass
transfer during RO and UF operations such as
porosity and roughness ofthe membrane wall and
change of viscosity and diffusion coefficient due
to a strong concentration gradient They noted
that many of the equations suggested in the
literature under a variety of conditions and types
of fluids in the turbulent region have the
asymptotic form (Re > 10,000 and SC > 100)
Sh- ReAScBfm

141 (2001) 269-289

273

flux The region 2,30O<Re<lO,OOO is unclear,


and no simple form of the Sherwood equation
exists for this area
The problem of estimating the mass transfer
coefficient, k, in RO is very important A variety
of methods has been employed and include direct
measurements using optical or micro-electrode
methods, indirect measurements,
in which the
true rejection is calculated by extrapolation to
infinite feed circulation, and indirect measurements, in which a concentration
polarization
model combined with a membrane transport
model is used for mass transfer coefficient
calculation [ 121

3 Review of experimental
tration polarization

studies on concen-

It is generally accepted that concentration


polarization of solute at the membrane surface
can lead to solute adsorption, solute precipitation
and gel layer formation
In the case of RO
membranes, this all results in a decrease in the
water flux across the membrane This, in turn,
increases operating costs due to the need for
frequent
membrane
cleaning
and possibly
reduced membrane lifetime Much research has
therefore gone into trying to understand this
phenomenon in both RO and UF systems so that
its effect can be minimized
3 I Reversible adsorbed layer resistance

(12)

where m= 1 or % The dependence


of the
Sherwood number on the friction factor (for df,
was a function of the type of experiment and
suggested that the Blasius formula, which relates
f with the Re, is generally more accepted
However,
in the case of pressure-driven
membrane operations, f is found to depend on
wall roughness and porosity, and on the permeate

Nikolova and Islam [ 151 reported concentration polarization in the absence of gel layer
formation using a laboratory-scale
UF unit
equipped with a tubular membrane (Table 1) In
an excellent study, they found that the decisive
factor in flux decline was the adsorption
resistance With the development of a concentration polarization layer, the adsorbed layer
resistance at the membrane
wall increased
linearly as a function of the solute concentration

274

S S Sablani et al /Desalination

141 (2001) 269-289

Table 1
Summary ofexperimental studies on concentration polarization in ultrafiltration (UF) and reverse osmosis (RO) membrane
systems Specific papers are recommended (0) and highly recommended (00)
Reference

System

Operating parameters

Significant results

Khulbe et al
[51

UF, polyethersulphone
membranes

BSA, PEO in water

Fouling of membrane dependent on size


and shape of macromolecular solute and
operating feed pressure

Schwinge et al

UF, zigzag spacer tilled


crossflow channel

Dextran, silica, whey protein,


crossflow velocities 0 1 and
0 8 m/s, upper transition fluid
mechanics region

Spacer filled channels achieved flux


enhancement over empty channel up to
three times due to turbulence decreasing
concentration polarization Processing
costs reduced by up to 60% per unit
volume of permeate

Darton
(1999)

RO plant

High silica water, high recovery


up to 320 mg/l silica in the
concentrate, 8 years experience

It is possible to run a RO plant for an


extended period even with severe
concentration polarization problem caused
by high silica level in the feed

Nikolova and
Islam**
1151

Aqueous solution of dextran


Lab scale UF, tubular
membrane, length = Im, (MW 40,000), concentration
= 3 &50 kg/m3, temperature
diameter = 0 0125 m
= 20DC, pressure = 0 4 MPa
Re=11~10~to54~10~

Ahn et al

Nanotiltration, flat-sheet
cell, width = 0 08 m,
length = 0 21 m, height =
0 082 m

Pulp and paper waste water,


temperature = 25 C,
transmembrane pressure
= 0 5 MPa, cross flow velocity
=lOm/s

Change of filtration resistance of various


nanotiltration membranes and mechanisms
responsible for resistance were identified
In two of five membranes tested, increased
filtration resistance was attributed to
osmotic pressure build-up due to
concentration polarization

Chen et al *
[71

Laboratory-scale UF, flat


sheet perspex module,
length = 0 08 m, width =
0 027 m, length =
0 21 m, width = 0 025 m

Colloidal silica (12 nm),


concentration = 0 4 wt%,
pH = 7 5, cross flow velocity
= 1 m/s

Dynamics of transition from concentration


polarization to cake formation was
explored by imposing flux Cake
deposition was witnessed using electron
micrographs Critical flux measured
Depolarization can be increased by
crossflow, by washing and increasing pH

Gowman and
Ethiere
121,221

UF,

Biopolymer hyaluronan
dissolved in 10 mM NaCl or
phosphate buffer concentration
= 0 2 to 0 4 wt%, flow rate
= 1 15x10e5 to 1 67x10e5 ml/s

Developed an automated, laser-based


refractometer technique for measurement
of concentration and concentration
gradient in concentration polarization
layer Effect of membrane type, solvent
type and operating parameters on spatial
and temporal development was quantified
on CP layer and also on pressure drop
across layer

[61

[I61

flat sheet flow cell

With development of concentration


polarization layer, adsorbed layer
resistance increases and flux declines

S S Sablani et al /Desalination

141 (2001) 269-289

275

Reference

System

Operating parameters

Significant results

Konieczny and
Bodzek

UF, tubular (two


membranes in parallel),
length = 122 m,
diameter=O016m

Latex waste water, concentration


= l-13%, temperature = 25C
pressure = 0 05 to 0 35 MPa,
velocity = 1 0 to 3 5 m/s

Determined suitability of polyacrylonitrile


and polysulfone membranes for UF of latex
waste water Filtration proceeds in gellayer regime Permeate flux linearly
dependent on cross-flow velocity,
exponentially dependent on latex
concentration, and independent of
transmembrane pressure Developed
Dittus-Boelter type relationship between
Sherwood, Schimdt and Reynolds numbers

MF, tubular, length =


40 cm, diameter =
0 0052 m

Oil-water emulsion,
concentration = S%(v/v),
pressure = 65 kPa, feed flow
= 1 62 l/h

Concentration polarization layer thickness


measured using NMR micro-imaging
Brownian diffusion main mechanism
controlling build-up of oil and polarization
layer

WI

Pope et al

[231

Soltys et al

UF, hollow fiber, length Dextran (4, T40, TlO),


concentration = 0 2 to 1 0 g/l,
diameter of fiber = 203 pressure = 50-200 mm of Hg
to 2300 urn

Produced dual-skinned membranes that


exhibited a clear directional selectivity and
sieving coefficients were different in two
flow directions When flow passes through
more porous skin first, then there is
significant concentration polarization
inside membrane matrix adjacent to tight
skin layer resulting in large osmotic
pressure across tight skin layer, reducing
effective transmembrane pressure and
causing reduction in permeate flux This
effect is absent when flow passes through
tighter skin first

WI

= 0 22 m, internal

Jonsson and
Jijnsson
[411

UF, flat sheet cell,


diameter = 0 257 m,
area=005m2

Negatively charged colloidal


hydrophilic silica (12 nm,
density = 2450 kg/m3),
concentration = 4 g/l,
temperature = 25C, pressure
= 0 1 MPa, rotor blade speed
= O-1000 rpm

Demonstrated flux behavior of stable and


unstable colloidal dispersion Basics of
colloidal chemistry was used to illustrate
mechanisms underlying difference in
performance between two types of
colloidal dispersion With stable
dispersions flux reduction is reversible,
with unstable colloidal dispersion it is
irreversible

Jijnsson
u71

UF, flat sheet cell,


diameter = 0 257 m,
area=005m2

Negatively charged colloidal


hydrophilic silica, octanoic acid
and sodium octanoate,
concentration = 0 4-2 O%,
temperature = 25C, pressure
= 0 1-O 3 MPa, rotor blade
speed = O-1000 pm

Distinguished between CP and fouling


using turbulent-promoting module
Differentiated between fouling
mechanisms: deposition of material at
membrane surface, interactions in
membrane matrix (solute-solute, solutemembrane)

Cherkasov et al
[LoI

UF, stirred cell


diameter = 0 047 m

Model protein mixture, Dextran,


room temperature, pressure
= up to 0 3 MPa

Showed that selective properties of UF


membranes do not depend on solute and
membrane hydrophilicities but on intensity
of concentration polarization

276

S S Sablani et al /Desalination

141 (2001) 269-289

Reference

System

Operating parameters

Significant results

Gekas and Olund


v71

UF, flat sheet, length =


0 1 m,width=O02m

Dextran (TlO) solution,


concentration = 0 5%,
temperature = 25C pressure
= 0 5 MPa, velocity =
l-5 m/s, Re = 4,000-20,000

Measured concentration polarization,


retention coefficient, water permeability
and reflection coefficient for several
membranes Used Sherwood correlations
based on various turbulence models:
classical film theory, surface renewal
theory and high flux correction Solute
concentration at membrane/water interface
predicted Could not determine which mass
transfer model is most correct due to too
many parameters and difficulty to measure
wall concentration experimentally

Chudacek and
Fane [32]

UF, flat sheet-stirred


cell, membrane area =
154cm2

Dextran = 0 4-2%, silica sol


= l-10%, BSA = 0 2-2 wt%,
temperature = 3OC, pressure
= 30-200 kPa, stirring speed
= O-l 000 rpm

Used modified constant pressure filtration


theory to model CP and analyzed
experimental data of specific filtration
resistance Polarization was complete
within 10-50 s Polarization layer
estimated to be several micrometers thick
(equivalent to several hundred
monolayers) Irreversible pore plugging
with BSA, not accounted in model

at the wall They described


following relationship

the flux by the

(13)

where AP is the hydraulic pressure difference


across membrane, C, is the concentration at the
membrane surface and AX(W) is the corresponding osmotic pressure, R, is the membrane
resistance, kc,,, is the adsorbed layer resistance,
and u is the fluid viscosity In a key finding, they
showed that the adsorption resistance was of the
same order of magnitude as the membrane
resistance Surprisingly, the osmotic pressure was
negligible in comparison to the applied transmembrane pressure This was all in the absence
of gel layer formation and membrane fouling
The significance of this study is that it showed
that the adsorbed solute layer at the membrane

surface is the primary cause of flux decline and


not the higher osmotic pressure at the membrane
surface We can speculate that if we can decrease
solute adsorption by, for example, changing the
membrane material or chemically modifying it,
we may be able to decrease flux decline
Ahn et al [ 161 investigated the mechanisms
involved in the reduction of the permeate flux in
nanofiltration ofwastewater from paper regeneration The observed total filtration resistance was
attributed and quantified as the intrinsic membrane resistance for pure water, the resistance
due to concentration
polarization,
and the
resistance due to membrane fouling The paper,
although containing useful experimental data on
the relative contribution of each of these resistances at different recoveries, did not attempt to
include an analysis of osmotic pressure on the
flux, as was done by Nikolova and Islam [ 151,
nor did it consider the transition from reversible
to irreversible adsorption

S S Sablani et al /Desalination 141 (2001) 269-289


3 2 Transition j?om reversible
irreversible fouling

adsorption

to

The solute adsorption described by Nikolova


and Islam [15] in the previous section is
reversible
The transition from this type of
adsorption to irreversible fouling is crucial in
determining the strategy for improved membrane
performance and for understanding the threshold
values for which optimal flux and rejection can
be maintained In a very thorough study, Chen et
al [7] reported on the dynamic transition from
concentration polarization to cake (i e , gel layer)
formation for membrane filtration of colloidal
silica Once a critical flux, Jcrit,was exceeded, the
colloids
in the polarized
layer formed a
consolidated cake structure that was slow to
depolarize and which reduced the flux This
study showed that by controlling the flux below
Jcrit, the polarization layer may form and solute
adsorption may occur, but it is reversible and
responds quickly to any changes in convection
This paper is a very valuable source of information for membrane plant operators who need
to maximize the flux while at the same time want
to reduce the frequency of membrane cleaning
Jbnsson [ 171 also studied the transition from
reversible solute adsorption to irreversible fouling She even went one step further and showed
the difference between solute deposition at the
membrane surface and in the membrane matrix
itself It was shown that a flux decrease that can
not be restored by increasing the shear rate may
be completely recovered by a temporary interruption of the permeate flow The question to ask
here is if the flux can be reversed completely
after repeated cycles or if there is actually a slow
decrease in flux with each cycle, as is seen in
Fig 1 [4] For a more detailed study of the
mechanisms underlying the differences in filtration characteristics between stable and unstable
colloidal dispersions, see Jijnsson and Jijnsson

P81

3 3 Effect of membrane
on solute adsorption

277
material and pore size

Cherkasov et al [ 191 presented an analysis of


membrane selectivity from the standpoint of
concentration polarization and adsorption phenomena The results of their study showed that
hydrophobic
membranes
attracted a thicker
irreversible adsorption layer than hydrophilic
membranes The layer thickness was determined

by the intensity of concentration polarization


Fig 3)
An effort was also made to produce and characterize new polymeric hollow-fiber membranes
that would have significant directional selectivity
[20] These membranes were designed to have
selective skin layers on both the inner and outer
surfaces of a porous (annular) matrix Transport
characteristics of these dual-skinned membranes
were determined by independently controlling the
pore size ofthe two skin layers Results indicated
that concentration polarization on the inner skin
layer significantly increased the solute sieving
coefficient when flow occurred through the more
open skin layer first, but was largely absent when

Fig 3 Gel layer formation on the surface of an ultrafiltration membrane from(I) hydrophobic and (II) hydrophilic material C, solute concentration; C, < C, < C,
1 adsorption layer, 2 gel-polarization layer, 3 membrane
material [ 191

278

S S Sablani et al /Desalination 141 (2001) 269-289

the flow was in the other direction This type of


research will help to minimize the problem of
back diffusion which may occur in hemodialysis
and membrane reactors
3 4 Measurement of concentration polarization
Gowman and Ethier [21,22] developed an
automated laser-based refractometric technique
to measure the solute concentration polarization
gradient during dead-end filtration of a biopolymer solution The observed trend of concentration polarization data for the biopolymer
hyaluronan, however, did not agree well with the
current theory This difference was attributed to
the uncertainty in the physico-chemical properties of the polymer Nevertheless, this is a good
paper which attempts to reconcile theory with
experimental data The refractometric technique
may be useful to other researchers working in
quantification of concentration polarization
A nuclear magnetic resonance technique was
employed by Pope et al [23] for quantitative
measurement of the concentration polarization
layer thickness during cross-flow filtration of an
oil-water emulsion With this technique, a series
of images was acquired during formation of the
polarization layer, prior to the establishment of
steady-state conditions The technique provided
visualization of the concentration polarization
layer and non-invasive measurement of its
thickness using chemical shift selective microimaging The specific resistance ofthe layers was
estimated from the average oil layer thickness
and concentration at a given time, using the
images and the filtrate flux measured gravimetrically
In summary, the work of Nikolova and Islam
[ 151,Gowman and Ethier [21,22], and Pope et al
[23] helps to give better quantitative insight into
the relative contributions to flux decline of the
adsorbed layer resistance, and the concentration
polarization layer gradient and thickness

3 5 EfSect ofjluid mechanics


Rapid fluid flow at a membrane surface will
reduce the effects of concentration polarization in
membrane systems Let us therefore take a closer
look at the combined effects of fluid mechanics
and other process parameters in these systems
Controlled centrifugal instabilities (called Dean
vortices), resulting from flow around a curved
channel, were used by Mallubhotla and Belfort
[24] to reduce both concentration polarization
and the tendency towards membrane fouling
These vortices enhanced back-migration through
convective flow away from the membranesolution interface and allowed for increased
membrane permeation rates As part of an effort
to further understand the opportunities and
limitations of using Dean vortices in membrane
separations, they reported on a series of experiments with colloids in which the performance of
runs in the presence of vortices was compared to
that in the absence of vortices For example, the
flux decreased with increasing ionic strength,
passed through a minimum and then increased to
a constant value They explained these surprising
observations by the different behavior of
aggregated colloids in solution and within the
deposited cake layer on the membrane surface
Experiments by McDonough et al [25] with
monodisperse silica colloid shed further light on
this process They showed that the limiting fluxes
in membrane filtration were dependent on the
zeta potential (i e , charge) of the colloid
Belfort [26], in an extensive review,
summarized the advantages and limitations of
different membrane permeators He reported on
theoretical studies of flow in an annulus with
porous walls and on experiments showing the
effect of transmembrane flux on the friction
coefficient of a flowing fluid inside a porous tube
with suction The performance of several
commercial modules was assessed in light of the
theory He reported that porous tubular systems
were the most efficient at the capture of particles

S.S. Sablani et al. /Desalination

from dilute suspensions due possibly to different


fluid mechanics compared to that of conventional
membrane systems.

4. Review of semi-empirical

modeling studies

Mathematical modeling in conjunction with


experimental research is a powerful tool in the
study of phenomena such as concentration
polarization. A valid and accurate model allows
significant time to be saved by permitting the
computer to simulate a variety of experimental
conditions.
4. I. Comparison of models

Bader and Veenstra [14] tested the applicability of three concentration polarization models,
all based on the Chilton-Colburn analogy. The
models consisted the original film theory (i.e., gel
layer) model, the Sherwood correlation model
and a modified film theory model. They used UF
data reported by Gekas and Olund [27]. Their
analysis showed that the modified film theory
model was more capable of correlating and
qualitatively predicting the rejection trend of UF
data than either the original film theory model or
the Sherwood correlation model. The observed
solute rejection, R,, was described by

hr[T]

+lnv=K(

5)

+lnQ

(14)

where v is the solvent velocity through the membrane, u is the solvent velocity parallel to the
membrane surface, K is the film diffusion
parameter and K,,, is the membrane-diffusion
mass transfer coefficient. The better prediction of
solute rejection by this model was due to the
inclusion of the convective and diffusion
mechanisms into the equation in the form of their
mass transfer coefficients (i.e., K, and K,

141 (2001) 269-289

279

respectively). This equation is one of the


recommended models.
The prediction of membrane performance
requires a knowledge of transport parameters
such as the mass transfer coefficient. Voros et al.
[28] estimated the transport parameters by fitting
model predictions to experimental results. The
mass transfer coefficient on the high-pressure
side of the membrane was found to vary with the
bulk feed flow rate. They concluded that the feed
flow rate should never be omitted in process
engineering calculations.
4.2. Flux decline and relative contributions
A combined experimental and modeling study
was carried out by Konieczny and Bodzek [29],
which demonstrated that permeate flux during UF
of latex waste waters did not depend on the
compactness ofthe membrane and the membrane
forming polymer because the main resistance was
in the gel layer.
Dal-Cin et al. [30] developed a series resistance model to quantify the relative contributions
of adsorption, pore plugging and concentration
polarization to flux decline during UF of a pulp
mill effluent. They proposed a relative flux loss
ratio as an alternative measure to the conventional resistance model which was found to be a
misleading indicator of the flux loss. Using
experimental and simulated flux data, the series
resistance model was shown to under-predict
fouling due to adsorption and to over-predict
concentration polarization. This appears to be a
major disadvantage and would make the model of
limited use to others.
As mentioned in the introduction, Koltuniewicz and Noworyta [4] modeled the flux decline
as a result of the development of a concentration
polarization layer based on the surface renewal
theory developed by Danckwerts [3 11. This is a
highly recommended paper. The surface renewal
model is more realistic than the commonly used
film model since mass transfer at the membrane

280

S.S. Sablani et al. /Desalination

boundary layer is random in nature due to membrane roughness. Specifically, the membrane is
not covered by a uniform concentration polarization layer, as was assumed in the film model,
but rather by a mosaic of small surface elements
with different ages, and therefore, different
permeate flow resistances. Any element can be
swept away randomly
by a hydrodynamic
impulse, and then a new element starts building
up a layer of retained solute at the same place on
the membrane surface. They showed that the
decrease in flux with respect to time, <t,), due to
the development ofthe concentration polarization
layer, is given by the following equation which
also takes into account the rate of surface
renewal, s:

(tp)=(lo-J*)5

;f;;+
J'

(1%

where A is the rate of flux decline, JOis the initial


value of the flux, J* is the flux observed after
infinite time and t,, is the time of permeation.

s=A

Jlti-J

(16)
Jo-4im

where Jlimis the limiting flux which is similar to


critical flux, JCfi,.The former can be obtained
from literature data. The average flux under
steady-state conditions, Jz, can be calculated
directly from Eq. (15) as a limit:
J, = lim3($) = (J~-J*)--&

IP

+J*

(17)

In support of this model, calculated values of flux


using Eqs. (15) and (17) agreed well with
experimental data. The two equations describe a
permeation cycle of duration, tp, as shown in
Fig. 1. This is excellent paper for those who are
operating large-scale continuous UF plants, and

141 (2001) 269-289

to a certain extent RO plants. The model


developed describes not only the dynamic behavior of a plant but it also allows for optimization
of operating conditions (i.e., permeation time,
cleaning time, cleaning strategy).
4.3. Flux decline due to colloids
Chudacek
and Fane [32] assessed
the
dynamics of polarization in unstirred and stirred
UF using three types of solute; a colloid (silica
sol), a protein (albumin) and a branched chain
polymer (dextran). The predicted profiles of
permeate flux for both stirred and unstirred UF
consistently over-predicted the flux values during
the early period. This was attributed to pore
blocking or obstruction which was not allowed
for in the model. In a similar study Lee and Clark
[33] developed a model to predict the flux
decline due to concentration polarization during
UF of colloidal suspensions. For verification of
the model, they conducted experimental studies
with polystyrene latex microspheres. They determined the influence of particle size, concentration and cross-flow velocity on the specific
cake resistance and permeate flux. The model
successfully explained the mechanisms involved
in flux decline during cross-flow UF of colloidal
suspensions.
4.4. Spacers in spiral-wound modules
A comprehensive finite-difference model was
presented by Madireddi et al. [l] to predict
concentration polarization in commercial spiralwound membranes.
Their equations,
which
described flow in spiral-wound membranes with
feed spacers, is useful for experimental studies
with various flow channel thicknesses.
Avlonitis et al. [34] presented an analytical
solution for the performance of spiral-wound
modules with seawater as feed. The analysis took
into account the effect of concentration polarization and pressure losses in both permeate and

281

S.S. Sablani et al. /Desalination 141 (2001) 269-289

brine channels. A computer program was written


to calculate the mass transfer coefficient and the
solute permeability
coefficient.
The results
showed that it was necessary to incorporate the
concentration and pressure of the feed into the
mass transfer correlation. In a similar study,
Boudinar et al. [35] developed a numerical model
to simulate the performance
of spiral-wound
modules. For calculation of mass transfer coefficients in channels equipped with a spacer, they
used the following relationship:

k= 0.X3(

&)122S4

2)

where Pe is Peclet number, K=0.5

(18)

and M=0.6

(cm).
Van der Meer et al. [36] studied the influence
of the number of spiral-wound membranes per
pressure vessel on concentration polarization and
ion rejection. They developed a simplified mathematical model which took into account the
hydraulics of the membrane modules. Their
studies showed that an increase in permeate
productivity could be achieved by lowering the
number of membrane modules from six per
vessel to two per vessel. This is a useful paper for
those designing desalination plants using spiralwound membranes.

5. Review of theoretical
tration polarization

studies on concen-

5.1. Variation in gel layer thickness and concentration polarization along flow channel
In earlier studies, the thickness of the gel layer
and the concentration of the solute were assumed
to be uniform over the membrane surface.
However, this assumption is only valid for
systems where the hydrodynamic conditions of
the solution flow near the membrane provide
equal accessibility
of solute to the entire

membrane surface [ 111. This is not true in the


case of cross-flow filtration. One can thus expect
that the gel layer thickness and/or the surface
concentration of the solute will vary with the
distance from the channel entrance. As a consequence, the local permeate flux will also vary
with longitudinal position. In an excellent article,
Denisov
[ 1 l] presented
a mathematically
rigorous theory of concentration polarization in
cross-flow UF, which takes account the nonuniformity of the local permeate membrane flux.
He derived the equations describing the pressure/
flux curve. In the case of the gel-layer model, the
theory led to a simple analytical formula for a
limiting or critical flux: the flux turned out to be
proportional to the cube root of the ratio of the
gel concentration to the feed solution concentration, rather than to the logarithm of this ratio,
as the simplified Michaels-Blatt theory predicted:
Jlim = (3/2)23KP,
c
=

1.31

$
i

m II3 ~213

~113

(19)

113 /, 113

where

Pg =

(20)

where K is hydraulic permeability of membrane


to pure solvent (m3/Ns), C, is the gel concentration (kmol/m3), C, is the solute concentration
in feed solution (kmol/m3), m is the channel
parameter, D is the solute diffusion coefficient
(m/s), U, is the longitudinal component of fluid
velocity averaged over the channel cross section
(m/s), L is the channel length (m), and h is the
transversal dimension of the channel (m). In case
of the osmotic-pressure
model, the rigorous
theory allowed the conclusion
that at high
applied transmembrane pressure, the permeate

282

S.S. Sablani et al. /Desalination

I41 (2001) 269-289

Table 2
Summary oftheoretical and semi-empirical studies on concentration polarization in ultrafiltration (UF) and reverse osmosis
(RO) membrane systems. Specific papers are recommended (0) and highly recommended (0.)
Reference

System

Operating parameters

Significant results

Maskan et al.
[31

RO

Mathematical study, constrained


variables non-linear optimization

Optimal network designs are the one that


minimize concentration polarization and
thus produce the most permeate

Madireddi
et al.* [l]

Pilot scale RO, three


spiral-wound membranes in series,
length = 1.O m,
diameter = 0.1 m

NaCl water system concentration,


= 0.2-l .Okg/m3, temperature
= 20C, pressure = 7001250 KPa, feed flow =
1.28~10-~ to 1.9~10^~ m3/s

Comprehensive finite difference model


was developed which can predict
parameters such as flow loss due to
concentration polarization. Model
verified experimentally

Parvatiyar
[431

UF, tubular
membrane

Newtonian and non-Newtonian


fluids, pectin solutions,
concentration = 0.355-0.93 wt%,
Reynolds number = 2x IO3to
11x103, flow index = 0.8857-1.0,
consistency factor = 0.00240.0296

Theoretical models for describing mass


transfer in membrane tube were
developed. Viscous stresses are more
sensitive to rheological parameters of the
solution during mass transfer process
compared to mixing stresses present in
turbulent layer and buffer layer

Lee and Clark


[331

Pilot scale UF, channeltype, width = 01039 m,


length = 0.138 m

Polystyrene latex microspheres


(0.064-2.16 pm), concentration
= 50-l 50 mg/l, temperature
= 2 I C, transmembrane pressure
= 10-20 psi, cross flow velocity:
= O.l- 0.5 m/s

Step-wise pseudo-steady-state model


developed which could predict flux
decline due to concentration polarization
during cross flow ultrafiltration. Good
agreement with experimental results

Van der Meer


et al. [36]

RO, spiral wound,


length = 0.92 m

Monovalent ions = 250 g/m,


bivalent ions = I50 g/m3,
temperature = 12.5C,
pressure = 700-l 100 kPa

Model shows that increase in permeate


productivity of 20% can be achieved by
using only two membranes per pressure
vessel. Can be applied to hydraulic
optimization of spiral wound membrane
plants

Song 1461

UF, data from


Blatt et al. [IO]

Protein and dextran solutions;


data from Blatt et al. [lo]

Developed mechanistic model for


predicting limiting flux. Also showed that
formation of cake layer on membrane
surface is necessary condition for limiting
flux

Elimelech and
Bhattacharjee
[451

RO, UF, flat-sheet

Mathematical study

Model for concentration polarization was


developed using combined hydrodynamic
(filtration) and thermodynamic (osmotic
pressure) approaches. Model prediction of
permeate flux compared well with the
numerical solution of convection
diffusion equation coupled with osmotic
pressure model

S.S. Sablani et al. /Desalination

283

141 (2001) 269-289

Significant

results

Reference

System

Operating parameters

Agashichev**

UF

Tubular, mathematical

Murthy and
Gupta [ 121

RO, flat sheet cell,


diameter = 0.09 m

NaCI-water system concentration


= 1,OOO-30,00Oppm, temperature
= 25C pressure = 20-l 00 atm,
feed flow = 18-90 l/h

Combined Spiegler-Kedem/film
theory
model may be best method for
establishing mass transfer correlation.
Marked variation in calculated k values
may be attributed to presence of reflection coefficient in model. Developed
Dittus-Boelter type relationship between
Sherwood, Schmidt and Reynolds
numbers

Alvarez et al.

PI

RO, tubular,
diameter = 0.0125 m,
length = 1.2 m

Apple juice, temp. = 25-4OC,


pressure = 175-550 kPa,
velocity = 0.1-2.95 m/s

Modeled and predicted permeate flux


using combined solution-diffusion/film
theory model. Transmembrane pressure
was found to be an important variable
controlling process

Parvatiyare

UF, tubular

Gelatin solution, Re =
3,000-1,000, particle diameter
= 0.001-0.009 m, void fraction
= 0.4-0.75

Presented theoretical analysis to study


interaction of dispersed phase (particles)
with concentration polarization. Addition
of inert solid particles on solute side
produced smaller eddies which in turn
reduced concentration polarization effect

UF, flat sheet test cell,


membrane area =
1.45x10~4m2

Pulp mill effluent, temperature


= 50C, pressure = 345 kPa

Used phenomenological
model and
quantified relative contribution of several
flux reducing mechanisms such as
adsorption, pore plugging and
concentration polarization. Adsorption
fouling underestimated and concentration polarization overestimated.
Alternative method proposed based on
flux decline due to particular mechanism
as fraction of overall flux decline

Pilot-scale RO,
tubular, length = 1.3 m,
diameter = 0.0127 m

NaCl water system, pressure


= 25-40 bar, feed flow
= 200-400 l/h

Used Kimura-Sourirajan
transfer kinetics
and material-balance equations under
steady-state conditions and quantified
transport parameters. Study focused on
brakish water and rejection of six
different ions

study

[441

[421

Dal-Cin et al.
1301

Voros et al.

[=I

Concentration polarization model was


developed by integrating balance
equations for viscous and diffusion
boundary layers over control volume.
Influence of mass flow, bulk concentration and velocity on degree of
polarization was quantified

284

S.S. Sablani et al. /Desalination

141 (2001) 269-289

Reference

System

Operating parameters

Significant results

Bader and
Veenstram
u41

UF

Dextran = 0.5 wt%,


temperature = 25C,
pressure = 0.5 MPa

Used experimental data of Gekas and


Olund [27] and tested three models of
concentration polarization : (1) modified
film theory, (2) original film theory and
(3) Sherwood correlation model.
Modified model, which was found to be
the best, relates convective and diffusion
mechanisms in terms of mass transfer
coefficients and eliminates need for
external solute rejection mechanism to
complement concentration polarization
model.

Polyakov and
Polyakov
1371

RO, spiral wound

NaGwater system,
mathematical study

Theoretical model for calculation of


permeate flux and salt rejection was
developed. Model accounted pressure loss
along channel and considered spacer
geometry. Allows for optimization of flow
conditions

Jonsson and
Jiinsson [ 181

UF, mathematical
study

Colloidal dispersion,
transmembrane pressure,
the fluid shear, concentration,
particle size, pH, ionic strength

Derived thermodynamic model of


concentration polarization. Model was
based on balance between thermodynamic force, due to osmotic pressure
gradient, and frictional force, due to fluid
flow around each particle. Concen-tration
profile at membrane surface can be
derived. Operating conditions under
which critical precipitation concen-tration
is reached can be calculated

Denisov**

1111

UF, tubular,
slit (with one porous
and one solid wall)

Mathematical study
Laminar flow, porous channels,
non-uniform wall suctions

Developed concentration polarization


model for steady-state cross-flow in
tubular and slit geometries using NavierStokes equations. For gel-layer model,
flux proportional to cube root of ratio gel
concentration to feed solution concentration rather than logarithm of this ratio. In
case of osmotic-pressure model, permeate
flux increases as cube root of pressure, so
that limiting flux is never reached

Koltuniewicz and
Noworyta**
[41

UF, hollow fiber, fiber


length = 0.6 m, fiber
diameter = 0.0011 m

Skim milk, concentration (of


solids) = 10-85.5 g/l,
temperature = 30C
pressure = O-l 60 kPa,
velocity = O-O.622 m/s

An excellent paper which uses


Danckwerts surface renewal theory to
predict permeate flux decline due to
development of concentration
polarization layer. Model enables
membrane cleaning strategies for largescale plants to be determined

S.S. Sablani et al. /Desalination

141 (2001) 269-289

285

Reference

System

Operating parameters

Significant results

Bouchard et al.
1461

UF, slit geometry

Mathematical study

Studied concentration polarization


phenomena by solving Navier-Stokes and
continuity equations using different finitedifference schemes. Good agreement
between numerical and analytical solution
was shown. Separation profile along
membrane length greatly affected by true
separation through membrane and by
permeation velocity

Avlonitis et al.
1341

RO, spiral wound,


length = 0.95 m,
diameter = 0.064 m

NaCI-water system,
concentration = 25,000-40,000
ppm, temperature = 20-35C
pressure = 50-80 bar,
feed flow = 7 1-2 15 cm/s

Used solution diffusion and film theory


model to predict permeate flux and
concentration. Developed Sherwood
correlation which incorporated feed
concentration and pressure. Mass transfer
coefficient, calculated by simple equation,
dependent on water flux

Boudinar et al:
1351

RO, spiral wound,


length = 0.88 m

NaCl-water system,
concentration = 187W0,ooo
ppm, temperature = 20-35C,
pressure = 28-80 bar,
feed flow = 90-528 cm/s

Used numerical finite-difference method


to solve equations for transport and
calculated concentration, pressure and
flow rates in brine and permeate channel.
Operating conditions for spiral-wound
modules optimized. Need more accurate
correlation for mass transfer coefficient

Gupta**
[3gi

RO, plate and frame,


tubular and spiral
wound

Mathematical study

An excellent paper which uses Kedem


and Katchalsky three-parameter model
and concentration polarization model of
film theory to develop design equations
that predicted length of module and solute
permeate concentration for given
recovery. Significance of solute reflection
coefficient (i.e., coupling of solute and
solvent fluxes) is brought out

Clifton et al.
[91

UF, hollow fiber

Dextran, poly-vinylpyrrolidone,
concentration = 1.0-2.0 g/l,
pressure = 24-l 70 kPa,
velocity = 0.1-2.65 m/s

Developed concentration polarization


model which accounted flux reduction
due to osmotic pressure and considered
concentration dependent viscosity.
Obtained local permeation data under
different experimental conditions that
compared well with data predicted using
mathematical model. Transmembrane flux
is limited by effect of osmotic pressure.
Polarization layer is unstable under
certain circumstances

286

S.S. Sablani et al. /Desalination

flux increased as a cube root of the pressure, so


that the limiting flux was never reached:

113 m 113 ~2

L II3

jyl3

(21)

h II3

where

PO =

(22)

where J is the average flux over the channel


(m/s), P is the transmembrane pressure (N/m*), R
is the gas constant (J/kmolK), and T is the
temperature (K). However, one minor weakness
of the study is that the analysis ignored the
concentration dependence of the viscosity and
partial transmission of the solute through the
membrane.
Polyakov and Polyakov [37] in their study
included the pressure loss along the flow channel
for a spiral-wound membrane system. Using the
material balance approach, they derived simple
explicit equations which could be employed in
designing and optimizing flow conditions in
reverse osmosis plants.
5.2. Flux prediction with a three-parameter
model
Various elaborate theoretical models have
been presented with different levels of simplification in order to predict concentration polarization effects. Most ofthe earlier models used for
predicting transport behavior in RO of one solute
are based on two or three parameters. Although
a two-parameter
model accounts for several
mechanisms
of solvent and solute transport
through membranes, it generally fails to predict

141 (2001) 269-289

solute flux accurately because of significant pore


flow [3 81, whereas a three-parameter model, such
as that produced by Kedem and Katchalsky [39],
based on irreversible thermodynamics, is characterized by hydrodynamic permeability, osmotic
permeability and the reflection coefficient. The
reflection coefficient in such a model takes into
account the coupling of solute and solvent fluxes
which are important in more open (i.e., pore
flow) membranes. In an excellent paper, Gupta
[38] for the first time showed that by using the
simplifying assumption of replacing the local
solute permeate concentrations
by an average
solute permeate concentration over the module,
the design equations based on an analytical
solution of the three-parameter
model could
predict the length of the module and the solute
permeate concentration
for a given solvent
recovery. The model, however, can be applied
only for dilute solutions where the fluxes can be
assumed to be linearly dependent on the driving
forces.
5.3. Evaluation offinite difference schemes
More accurate solutions of mass transport
differential equations require the use of numerical solution schemes such as the finite difference
method. Bouchard et al. [40], for example,
reviewed different approaches for the finite
difference discretization of the mass transport
equation. Their study on two of the finite difference schemes revealed that convergence of the
solution depends on the choice of the finitedifference
scheme
(conservative
or nonconservative approaches) and the form of the
mass transport equation. The influence of key
variables on the separation for low and high
separation range showed good agreement with
analytical solution for the high separation range,
but strong disagreement was observed in the case
of low separation. This indicated the limits of
applicability of the analytical solution.

S.S. Sablani et al. /Desalination

5.4. Cell model for colloidal dispersion


A cell model used extensively to describe
thermodynamic equilibrium properties, such as
the osmotic pressure and other chemical potentials in colloidal systems, and diffusion through
concentrated suspensions, was employed by
Jonsson and Jiinsson [41] to model concentration
polarization. The model was based on the balance
between a thermodynamic force, due to the
osmotic pressure gradient, and a frictional force,
due to the fluid flow around each particle. The
model was able to predict the performance of
both non-interacting solutions of low molecular
weight as well as solutions of interacting solutes.
The model required concentration dependence of
the osmotic pressure and the diffusion coefficients. The model, however, was limited to
colloidal dispersions with particles smaller than
0.1 urn.

5.5. Interaction of dispersed phase and turbulence modeling


Parvatiyar [42] advanced the understanding of
mass transport theories describing the separation
mechanism by analyzing the turbulent structure
of fluid flow due to the presence of a dispersed
phase. He derived a mathematical procedure to
predict the effect of eddy size, the energy
dissipation rate per unit mass of fluid, the
turbulence intensity, the presence of a void
fraction, the effect of particle diameter, and the
stress produced in fluid, on the behavior of the
mass transfer coefficient and concentration
polarization. He showed that the smallest eddies
are cumulatively responsible for an effective
increase in the mass transfer coefficient. It was
observed that eddy size decreased with an
increase in Reynolds number. A high void
fraction and a large particle diameter induced the
growth of larger eddies. The foremost important
parameter in controlling the mass transfer coeffi-

141 (2001) 269-289

287

cient and concentration polarization was found to


be the ratio of membrane tube diameter to the
particle diameter and not the absolute value of
the particle diameter. Results also suggested that
concentration polarization in a membrane tube
can be controlled by placing inert solid particles
as a suspension or dispersed phase in the tubular
device. Based on the turbulent structure of the
fluid flow, he also developed an equation for a
mass transfer coefficient to be used with Newtonian as well as non-Newtonian fluids [43].
Agashichev [44], in a very thorough paper,
developed a correlation for longitudinal mass
flow of a dissolved component which was based
on balance equations for viscous and diffusion
boundary layers against the degree of concentration polarization. The correlation was obtained
by integrating these equations over the control
volume of the cylindrical configuration. While
the application of this model is limited to
isothermal, laminar, incompressible fluids under
steady-state conditions, it does permit the
analysis of the combined influence of mass flow,
bulk concentration and velocity on the degree of
concentration polarization. This is a highly
recommended paper.
5.6. Unifying filtration
pressure model

theory and osmotic

Elimelech and Bhattacharjee [45] pointed out


the conflict between the osmotic pressure model
and filtration (i.e., film or gel layer) theory
normally used for the prediction of permeate flux
during cross-flow filtration. The former does not
concern itself with transport phenomena in the
polarized layer, while the latter is based on the
hydrodynamic resistance of the layer. From the
basic governing equations of thermodynamics
and hydrodynamics, they derived the equivalence
ofthe osmotic pressure and filtration approaches.
The model was capable of predicting the local
variation of permeate flux in a filtration channel,

288

S.S. Sablani et al. /Desalination

as well as providing a simple expression for the


channel-averaged flux. The predictions of
permeate flux using the model compared well
with a detailed numerical solution of the
convective diffusion equation coupled with the
osmotic pressure model.
5.7. Critical or limiting flux
A key phase in membrane separation processes is the transition from concentration
polarization to fouling. This occurs at a critical
flux. Song [46] indicated that in most theories
developed the limiting or critical flux is
a priori knowledge rather than being predicted
from fundamental principles. To overcome this
shortcoming, he developed a mechanistic model,
based on first principles, for predicting the
limiting flux. Similar to the critical flux results of
Chen et al. [7] and the limiting flux of Koltuniewicz and Noworyta [4], Song showed that there
is a critical pressure for a given suspension.
When the applied pressure is below the critical
pressure, only a concentration polarization layer
exists over the membrane surface. A cake layer,
however, will form between the polarization
layer and the membrane surface when the applied
pressure exceeds the critical pressure. The
limiting or critical flux values predicted by the
mechanistic model compared well with the
integral model for a low concentration feed.
However, it deviated at high solute feed concentrations.

6. Conclusions
Membrane separation processes are very
complex and are influenced by numerous operating parameters such as the transmembrane
pressure, nature and concentration of the solute,
and feed velocity. Concentration polarization,
along with solute adsorption and gel layer
formation, cause flux reduction. It is thus

141 (2001) 269-289

difficult to predict membrane performance for


different applications. In order to better understand this problem, experimental and modeling
studies have been employed in an attempt to
quantify concentration polarization under different operating conditions. The newer experimental
approaches of nuclear magnetic resonance and
laser-based refractometric techniques are recommended for quantifying concentration polarization effects. More detailed models, such as the
surface renewal theory, can be used in place of
the conventional gel-polarization and osmotic
pressure models. These newer models have
advanced our knowledge about concentration
polarization. A more unified approach towards
treatment of membrane separation processes has
also emerged in recent years. This has led to the
concept of a limiting or critical flux.
By having a more fundamental understanding
of a specific membrane separation process, both
experimentally as well as theoretically, it will be
possible to run an UF/RO system just below the
critical flux. This will allow for maximum overall
productivity through fewer membrane cleaning
cycles, longer operating runs, longer membrane
lifetimes and higher fluxes. Concentration polarization, however, is only one part of a complex
separation process which includes feed pretreatment, membrane fouling, and membrane
selection/modification. All of these problems
cannot be solved at the same time. By operating
just below the critical flux, we can still maintain
good overall productivity while at the same time
avoiding membrane fouling. As our understanding of membrane separation increases, it remains
to be seen if we can develop a system which can
operate indefinitely, without cleaning. It is a
worthy cause that all of us can work towards.
Acknowledgments

Funding from the Middle East Desalination


Research Center (MEDRC), through contract
number 97-A-004, is gratefully acknowledged.

S.S. Sablani et al. /Desalination

References
[l] K. Madireddi, R.B. Babcock, B. Levine, J.H. Kim
and M.K. Stenstrom, J. Membr. Sci., 157 (1999) 13.
[2] V. Alvarez, S. Alvarez, F.A. Riera and R. Alvarez, J.
Membr. Sci., 127 (1997) 25.
[3] F. Maskan, D.E. Wiley and L.P.M. Johnston, AIChE
J., 46 (2000) 946.
[4] A. Koltuniewicz and A. Noworyta, Industrial
Engineering Chem. Res., 33 (1994) 1771.
[5] KC. Khulbe, T. Matsuura, S. Singh, G. Lamarche
and S.H. Noh, J. Membr. Sci., 167 (2000) 263.
[6] J. Schwinge, D.E. Wiley, A.G. Fane and R. Guentha,
J. Membr. Sci., 172 (2000) 19.
[7] V. Chen, A.G. Fane, S. Madaeni and I.G. Wenten, J.
Membr. Sci., 125 (1997) 109.
[S] M.C. Porter, Industrial Engineering Chem. Production Res. Develop., ll(3) (1972) 234.
[9] M.J. Clifton, N. Abidine, P. Aptel and V. Sanchez, J.
Membr. Sci., 21 (1984) 233.
[lo] W. Blatt, A. Dravid, A.S. Michaels and L. Nelsen, in:
J.E. Flinn, ed., Membrane Science and Technology,
Plenum, New York, 1970, pp. 47-91.
[ll] G.A. Denisov, J. Membr. Sci., 91 (1994) 173.
[12] Z.V.P. Murthy and S.K. Gupta, Desalination, 109
(1997) 39.
[13] V. Gekas and B. Hallstorm, J. Membr. Sci., 30 (1987)
153.
[ 141 M.S.H. Bader and J.N. Veenstra, J. Membr. Sci., 114
(1996) 139.
[15] J.D. Nikolova and M.A. Islam, J. Membr. Sci., 146
(1998) 105.
[16] K.H. Ahn, H.Y. Cha, I.T. Yeom and K.G. Song,
Desalination, 119 (1998) 169.
[17] A.-S. JBnsson, Sep. Sci. Technol., 30(2) (1995) 301.
[18] A.-S. JBnsson and B. Jansson, Sep. Sci. Technol., 3 1
(1996) 2611.
[19] A.N. Cherkasov, S.V. Tsareva and A.E. Polotsky,
J. Membr. Sci., 104 (1995) 157.
[20] P.J. Soltys,N.J. Ofsthun and A.L. Zydney, J. Membr.
Sci., 118 (1996) 199.
[21] L.M. Gowman and C.R. Ethier, J. Membr. Sci., 131
(1997) 95.
[22] L.M. Gowman and C.R. Ethier, J. Membr. Sci., 131
(1997) 107.

141 (2001) 269-289

289

[23] J.M. Pope, S. Yao and A.G. Fane, J. Membr. Sci.,


118 (1996) 247.
[24] H. Mallubhotla and G. Belfort, J. Membr. Sci., 125
(1988) 75.
[25] R.M. McDonough, A.G. Fane and C.J.D. Fell, J.
Membr. Sci., 43 (1989) 69.
[26] G. Belfort, J. Membr. Sci., 35 (1988) 245.
[27] V. Gekas and K. Olund, J. Membr. Sci., 37 (1988)
145.
[28] N.G. Voros, Z.B. Maroulis and D. Marinas-Kouris,
Desalination, 104 (1996) 141.
[29] K. Konieczny and M. Bodzek, Desalination, 104
(1996) 75.
[30] M.M. Dal-Cin, F. McLellan, C.N. Striez, CM. Tam,
T.A. Tweddle and A. Kumar, J. Membr. Sci., 120
(1996) 273.
[31] P.V. Dan&we@ Industrial Engineering Chem., 43
(1951) 460.
[32] M.W. Chudacek and A.G. Fane, J. Membr. Sci., 21
(1984) 145.
[33] Y. Lee and M.M. Clark, J. Membr. Sci., 149 (1998)
181.
[34] S. Avlonitis, W.T. Hanbury and M.B. Boudinar,
Desalination, 89 (1993) 227.
[35] M. Ben Boudinar, W.T. Hanbury and S. Avlonitis,
Desalination, 86 (1992) 273.
[36] W.G.J. Van der Meer, M. Riemersma and J.C.
van Dijk, Desalination, 119 (1998) 57.
[37] V.S. Polyakov and S.V. Polyakov, Desalination, 104
(1996) 215.
[38] S.V. Gupta, Desalination, 85 (1992) 283.
[39] 0. Kedem and A. Katchalsky, Biochem. Biophys.
Acta, 27 (1958) 229.
[40] C.R. Bouchard, C.R. Carreau, P.J. Matsuura and
S. Sourimjan, J. Membr. Sci., 97 (1994) 215.
[41] A.-S. Jansson and B. Jbnsson, J. Coil. Interface Sci.,
180 (1996) 504.
[42] M.G. Parvatiyar, J. Membr. Sci., 115 (1996) 121.
[43] M.G. Parvatiyar, J. Membr. Sci., 148 (1998) 45.
[44] S.P. Agashichev, Desalination, 119 (1998) 159.
[45] M. Elimelech and S. Bhattacharjee, J. Membr. Sci.,
145 (1998) 223.
[46] L. Song, J. Membr. Sci., 144 (1998) 173.

You might also like