You are on page 1of 18

Desalination 186 (2005) 111128

Modeling concentration polarization in reverse osmosis


processes
Suhan Kim, Eric M.V. Hoek*
Department of Civil and Environmental Engineering, 5732-G Boelter Hall, University of California,
Los Angeles, CA 90095-1593, USA
Tel. +1 (310) 206-3735; Fax +1 (310) 206-2222; email: hoek@seas.ucla.edu

Received 11 April 2005; accepted 23 May 2005

Abstract
Accurate prediction of concentration polarization (CP) phenomena is critical for properly designing reverse
osmosis (RO) processes because it enhances trans-membrane osmotic pressure and solute passage, as well as
surface fouling and scaling phenomena. The objective of this study was to compare available analytical CP models
to a more rigorous numerical CP model and experimental CP data. A numerical concentration polarization model
was developed to enable local description of permeate flux and solute rejection in crossflow reverse osmosis
separations. Predictions of channel averaged water flux and salt rejection by the developed numerical model, the
classical film theory model, and a recently proposed analytical model were compared to well-controlled laboratory
scale experimental data. At operating conditions relevant to practical RO applications, film theory and the numerical
model accurately predicted channel-averaged experimental permeate flux and salt rejection data, while the more
recent analytical model did not. Predictions of local concentration polarization, permeate flux, and solute rejection
by film theory and the numerical model also agreed well for realistic ranges of RO process operating conditions.
Keywords: Concentration polarization; Reverse osmosis; Film theory; Modeling; Membranes

1. Introduction
Concentration polarization (CP) is one of the
most important factors influencing the performance of membrane separation processes [1]. Prediction of solute concentration polarization is
crucial for designing reverse osmosis processes,
*Corresponding author.

predicting their performance, and especially for


understanding surface fouling phenomena [25].
For example, accurate description of local variations in permeate velocity, solute rejection, and
concentration polarization are required to predict
the onset of surface scale formation by sparing
soluble minerals [6]. However, some debate lingers among membrane scientists and engineers

0011-9164/06/$ See front matter 2005 Elsevier B.V. All rights reserved.
doi:10.1016/j.desal.2005.05.017

112

S. Kim, E.M.V. Hoek / Desalination 186 (2005) 111128

as to the most appropriate means of predicting


concen-tration polarization in reverse osmosis and
other semi-permeable membrane separations like
nano-filtration and ultrafiltration [7].
Concentration polarization is governed by
solute properties, membrane properties, and hydrodynamics [810]. It is influenced by both axial
and transverse flow fields, which are in turn influence by concentration polarization [1113]. The
coupling of momentum and mass transport makes
simultaneous solution of the NavierStokes, continuity, and convection-diffusion equations computationally intensive even for simple systems [13,
14]. Although concentration polarization in
reverse osmosis processes under laminar flow conditions can be rigorously modeled [15,16], these
accurate numerical solutions have little bearing
on real membrane separations in spiral wound elements with complex hydrodynamics due to feed
spacers [1719]. However, these relatively idealized
CP models are important for proper interpretation
of laboratory scale mechanistic studies, especially
those focused on elucidating fundamental mechanisms governing fouling and scaling phenomena.
Early CP models were limited to semi-empirical mass transfer analogies of heat transfer
correlations, which were developed to describe
stagnant film layer formation in conduits with
impermeable walls [20]. Stagnant film models are
fundamentally ill-suited to accurately describe all
mechanisms of concentration polarization in
crossflow membrane filtration processes because
these models assume uniform concentration along
the filtration channel wall and do not account for
the impact of permeate convection on the film
layer thickness. Despite these limitations, film
theory is simple, analytical, and (reasonably) accurate for most reverse osmosis separations. Further,
film theory can be extended to describe CP
phenomena in spacer-filled RO modules, which
is of tremendous value in practical process design
and evaluation.
Recently, a model was proposed [21], which
attempts to capture the fundamental interplay

between concentration polarization layer development and a locally varying permeate flux (in a
closed form analytical solution) by making use
of the retained solute concept [22]. This model
describes two dimensional convective-diffusive
solute transport without complex computational
effort, and hence, can be used to predict local concentration polarization and permeate flux in
spacer-free reverse osmosis flow channels. The
retained solute model appears an attractive alternative to film theory and is simpler than numerical
techniques, but it has not been subjected to
rigorous analysis or comparison to well-controlled
experimental data. A more rigorous numerical
convection-diffusion based concentration polarization model was developed for comparison with
the retained solute and film theory models. Predictions from all three models were compared to
experimentally-derived permeate flux and observed salt rejection data. The accuracy, limitations, and applicability of each model for RO
processes are discussed.
2. Theoretical
2.1. Analytical film theory (FT) model
A film theory approach to describe concentration polarization was developed by Michaels
and others [810]. Film theory simplifies a complex transport problem to a one-dimensional masstransfer problem by assuming axial solute convection near the membrane surface is negligible.
Integrating the one dimensional (transverse) convection-diffusion mass balance from the membrane surface out to a finite mass boundary (film)
layer thickness, , yields the relationship between
concentration polarization and permeate flux. The
result is
cw c p

v
= exp w
cb c p
D

(1)

where cw is concentration at the membrane surface,

S. Kim, E.M.V. Hoek / Desalination 186 (2005) 111128

or channel wall, for the rejected salt, cb and cp are


the bulk and permeate solute concentrations, respectively, vw is the permeate water velocity at the
channel wall, and D is the solute diffusion coefficient.
A procedure for estimating the convective
diffusion layer thickness in a channel with soluble
or rapidly reacting walls forms the basis of most
film theory models [23]. For fully developed
laminar flow in a thin rectangular channel, the
film layer thickness is described by

( x)

h
= 1.475
x
x

2/3

1/ 3

umax h

(2)

where x is the longitudinal coordinate, h is the


channel half-height, and umax is the maximum
crossflow velocity at channel center. Assuming
constant diffusivity, a local mass transfer coefficient is described by
1/ 3

D
1 3u D 2
k ( x) =
=

( x ) 1.475 2h x
1/ 3

 D 2
= 0.538 w
x

113

recovery is negligible. The channel average mass


transfer coefficient is expressed as
1/ 3

 D 2
k = 0.807 w
L

(4)

which is identical to the mass transfer correlation


for laminar flow in a thin channel commonly
reported in terms of a Sherwood number, e.g.,
1/ 3
Sh = kd H / D = 1.85 ( ReScd H / L ) [9].
In reverse osmosis processes, the driving force
for permeation is the difference between the
applied pressure, p, and the transmembrane
osmotic pressure, m. Thus, the permeate flux is
described by
vw = A ( p m )

(5)

where A is the apparent water permeability of the


membrane. The transmembrane osmotic pressure
is given as
m = f os ( cw c p ) = f os Ri cw

(3)

where u is the bulk average crossflow velocity


(=Q/2hW), W is the channel width, Q is the
volumetric feed flow rate) and  w ( = 3u / h) is the
wall shear rate [24,25]. A key limitation of applying film theory to membrane separations is the
assumption that the transverse component of convection (permeate flux) does not influence the
boundary layer thickness, .
The channel averaged mass transfer coefficient, k , derives from integrating the local flux,
vw(x), given by Eqs. (1) and (3) assuming the membrane, permeate, and bulk concentrations, as well
as the permeate and crossflow velocities remain
constant along the channel length, L. This assumption is unrealistic for large scale RO systems, but
reasonable for short membrane channels where

(6)

where Ri = (1 cp/cw) is the intrinsic salt rejection


of membrane and fos is a coefficient that converts
molar salt concentration to an osmotic pressure
via an appropriate expression [26]. vant Hoffs
equation gives fos = 2RT for NaCl (R is the universal gas constant and T the absolute temperature), which is accurate for dilute concentrations.
Provided with the system operating parameters
(p, u , T), solution and solute characteristics (cb,
D), and membrane properties (A, Ri), the wall concentration, cw, is the only unknown parameter
linking concentration polarization, permeate flux,
and observed rejection (Ro = 1 cp/cb). The wall
concentration can be solved for when Ri is known
from [5]
cw
v
= 1 Ri + Ri exp w
cb
k

which is called the CP modulus [1].

(7)

S. Kim, E.M.V. Hoek / Desalination 186 (2005) 111128

114

2.2. Analytical retained solute (RS) model


The retained solute model attempts to account
for the interplay between solvent permeation,
solute retention, and concentration polarization.
The coupling of water permeation and concentration polarization becomes more complex in
crossflow filtration when the local variation of CP
is considered along the filtration channel. An
analytical concentration polarization model was
previously developed to describe the combined
influences of colloidal particle diffusion, crossflow velocity, and permeate flux in crossflow
ultrafiltration using the concept of retained solute
[21].
At steady state, the concentration of retained
solute in the CP layer satisfies the following condition, assuming negligible axial transport

vw ( x ) C + D

C
=0
y

(8)

where x and y are the longitudinal and transverse


coordinates, respectively, vw(x) is the permeate
flux at x, and C indicates the concentration of
retained solute (the actual solute concentration in
the polarization layer is cb + C) [22]. At steady
state, the longitudinal solute flux at any point x
along the channel is equal to the total amount of
solute rejected by the membrane from the inlet to
the location x, that is

 w yCdy = Ro cb vw ( x ) dx
0

(9)

where Ro (= 1 cp/cb) is the observed salt rejection


and x is a dummy integration variable.
Using Eqs. (5), (8) and (9) a closed-form
analytical expression for the local permeate
velocity can be written as

vw ( x ) =

AF 0

(1 + 6F

X / Ns3 )

1/3

1/3

1/ 2

4
2

+
+

2
3
+

F
X
N
1
6
/

s )
(

(10a)

1/ 3

1/ 2

4
2

+

2
3
(1 + 6F X / Ns )

with F =

p 0
0

(10b)
1/ 3

and N s =

1 1 D 2  w

A 0 L

(10c)

Here, is the viscosity of water, 0 is the osmotic


pressure difference between feed and permeate
sides of the membrane (0 = fos (cb cp) = fosRocb),
X is the dimensionless longitudinal distance, x/L,
and F and Ns are the dimensionless driving force
and dimensionless membrane resistance, respectively [21]. The salt concentration on the membrane surface can be derived from vw(x) via
Eq. (5), that is
cw ( x ) =

p vw ( x ) / A + f os (1 Ro ) c0
f os

(11)

This analytical concentration polarization


model is attractive because it accounts for the
fundamental interplay between concentration
polarization and permeate flux in a closed form
analytical solution.
2.3. Numerical convectiondiffusion (CD) model
We have developed a new concentration polarization model accounting for the two-dimensional

S. Kim, E.M.V. Hoek / Desalination 186 (2005) 111128

mass transfer occurring in crossflow RO processes.


The velocity field is composed of axial (feed) and
transverse (permeate) components. In principle,
the velocity profile can be obtained by complete
solving the NavierStokes and the continuity
equations; however, a Poiseuille flow profile is
classically assumed if the permeation velocity is
small and does not significantly alter the axial flow
profile [15]. An additional complication arises
because both velocity components influence
concentration polarization of rejected salt ions.
Simultaneously, concentration polarization reduces
permeate flux, which in turn alters the polarization
layer characteristics. A complete description of
the process requires solving the fully coupled
NavierStokes, continuity, and convectiondiffusion equations.
In order to simplify the coupled momentum
and mass transport problem, the axial velocity
profile is assumed independent of concentration
polarization in our model. The axial (u) and
transverse (v) components of velocity are expressed as [15]:
r2
u ( x, y ) = umax ( x ) 1 2
h
r2
3
= u ( x ) 1 2
2
h

r2
r
v ( x, y ) = vw ( x ) 3 2 ;
h
2h
hvw ( x )

115

the CP layer thickness is assumed negligible


compared to the channel height, which enables
the transverse velocity within the boundary layer
to be held constant and equal to the permeate
velocity at the wall, vw(x).
Conservation of mass is considered in two
regions of the crossflow filtration channel. One
region is above the boundary of the CP layer
and the other is within the CP layer. Fig. 1 shows
the domain for solving the mass balance problem.
At steady state, solute and solvent mass should
be conserved everywhere. Since the height of
concentration polarization layer is much smaller
than the channel height, solvent mass is considered
above concentration polarization layer, that is
Q ( x ) = Q ( x + x ) +

x +x

vw ( x )Wdx

(13)

where Q(x) and Q (x + x) are the axial flow rate


at x and x + x, respectively, and vw(x) is the permeate flux at x. The solute mass balance above
concentration polarization layer is described as

Q ( x ) cb ( x ) = Q ( x + x ) cb ( x + x )
x +x

(12a)

(12b)

= Re w  1

where, r is the distance from center line of a


rectangular channel (r = y h) and is kinematic
viscosity of water. As indicated, Eq. (12b) is valid
when the wall Reynolds number based on the
permeation velocity is much smaller than unity.
A further simplification used in our model is that

vw ( x )Wcb ( x ) dx

(14)

where cb(x) and cb(x + x) are the bulk concentration at x and x + x, respectively. By solving
Eqs. (13) and (14) simultaneously, cb(x) is proved
constant and denoted as cb.
Solute mass should be conserved within concentration polarization layer as well, that is,
vw ( x ) c ( x, y ) + D

c ( x, y )
y

= vw ( x ) c p ( x )

(15)

where c(x, y) is the solute concentration at (x, y)


within concentration polarization layer. Eq. (15)
is a form of convectiondiffusion equation and
its solution can be written as

116

S. Kim, E.M.V. Hoek / Desalination 186 (2005) 111128

Fig. 1. The problem domain used in developing the numerical convection diffusion model for a laboratory scale crossflow
membrane filter without mesh feed spacer. Feed water flows from left to right along the channel length and some portion
of water flows through the membrane.

v ( x)
c ( x, y ) = cw ( x ) c p ( x ) exp w
y
D

(16)
+ c p ( x ) ; y ( x )
At steady state, axial and transverse input and
output should be balanced in the problem domain
within concentration polarization layer as shown
in Fig. 1, that is

( x )

u ( x, y ) c ( x, y ) y
0

at x
( x )

u ( x, y ) c ( x, y ) y
0

at x +x
= cb c p ( x )

x +x

vw ( x )dx

(17)

The integrations in Eq. (17) are solved numerically using the trapezoidal rule. The concentration profile starting from an arbitrary (initial)
wall concentration is solved for by Eq. (16). As
the transverse coordinate (y) increases, the concentration decreases due to solute back transport by
convection and diffusion. The extent of the CP
boundary layer thickness is defined as the position
when the concentration, calculated from Eq. (16),
equals the bulk concentration. Concentration
polarization layer thickness is defined as the
distance from that position to the channel wall.
Above the CP layer, all the concentrations are
equal to the bulk concentration.
Since the initial wall concentration is an
arbitrary value, an iteration process is required.
This is performed by solving Eqs. (12), (16) and
(17) simultaneously to obtain the salt concentration locally in the channel. Once the local salt

S. Kim, E.M.V. Hoek / Desalination 186 (2005) 111128

concentration on the membrane surface (cw) is


determined, the local permeate flux can be calculated from Eq. (5) and the osmotic pressure from
Eq. (6). Provided with the system operating parameters (T, p, u ), solution and solute characteristics (cb, D) and membrane properties (A, Ri),
concentration polarization, permeate flux, and permeate concentration can be predicted. All mass
conservation equations above assume constant and
equal density and viscosity in feed, retentate, and
permeate solutions.
3. Experimental
3.1. Membrane and reagents
A commercial RO membrane designated as
XLE (Dow-FilmTec Corp., Edina, MN) was used
in this study. Upon receipt from the manufacturer,
the membrane was immersed in de-ionized water
(Milli-Q Synthesis, Millipore, Billerica, MA) and
stored at 5C. Deionized water was changed daily
for the first two weeks and subsequently changed
about every two weeks. Salt stock solutions were
prepared using ACS grade NaCl (Fisher Scientific;
Pittsburgh, PA) dissolved in de-ionized water.
3.2. Crossflow membrane filter
The crossflow membrane filter (CMF) used in
fouling experiments was a modified version of a
commercially available unit (Sepa CF, Osmonics,
Inc.; Minnetonka, MN). A schematic illustration
of the experimental apparatus and a complete
description of modifications are provided elsewhere [27]. The crossflow membrane filter was
rated for operating pressures up to 6895 kPa
(1000 psi), and had channel dimensions of 14.6 cm,
9.5 cm, and 1.73 mm for channel length, width,
and height, respectively. Membrane surface area
was 1.39102 m2 and cross-sectional flow area
was 1.64104 m2.
3.3. Pure water permeability
At the start of each fouling experiment, de-

117

ionized water was filtered through the membrane


overnight at constant temperature of 25C to allow
for membrane compaction and other unknown
causes of flux decline inherent to laboratory-scale
recirculation systems. After stable flux was
achieved, pure water permeability was determined
by measuring pure water flux over a range of
applied pressures. The relationship governing the
pure water flux is
vw = Ap

(18)

Pure water permeability (A) was determined


from a linear regression of the measured pure
water flux and applied pressure data.
3.4. Concentration polarization, salt rejection and
osmotic pressure
After membrane permeability was determined,
an appropriate volume of stock NaCl solution was
added to provide the desired feed ionic composition. Flux and crossflow were set at the desired
values for each fouling experiment and the system
was allowed to equilibrate up to 24 h to ensure
stable performance. Temperature was maintained
at 25C by a recirculating chiller. Experiments
were conducted with 27 (333) combinations
of cb (10, 20, and 50 mol/m3 NaCl), p (689, 1034,
and 1379 kPa), and u (0.017, 0.042, and 0.068 m/s;
290, 590, and 1170 as Reynolds number) at
unadjusted pH of 5.8 0.2.
Due to concentration polarization of rejected
ionic constituents, the driving force for permeation
is the difference between the applied pressure (p)
and the transmembrane osmotic pressure (m).
During electrolyte equilibration, observed salt
rejection (Ro) was determined from feed (f) and
permeate (p) conductivity measurements (Ro = 1
p/f).Conductivity of NaCl solutions was
determined linear over the range of ionic strengths
used.
Intrinsic salt rejection (Ri = 1 cp/cw) was determined from the calculated salt concentration at
the wall (cw), and the measured permeate concen-

118

S. Kim, E.M.V. Hoek / Desalination 186 (2005) 111128

tration (cp). In order to calculate cw, osmotic


pressure was first calculated using Eq. (5). Then,
cw was calculated using the experimental permeate
concentration (cp) from the conductivity measurement as in Sutzkover et al. [28].
4. Results and discussion
4.1. Comparison of CP modulus predicted by
models
Prediction of observed salt rejection and permeate water flux at any applied pressure and
crossflow rate are dependent on the concentration
polarization modulus (=cw/cb), intrinsic salt rejection, and intrinsic water permeability. Once the
CP modulus, rejection, and permeability are known,
the channel averaged permeate flux may be predicted from Eqs. (5) and (6). Observed salt rejection is then predicted from the intrinsic salt rejection and known feed salt concentration. Hence,
the variable CP modulus combined with intrinsic
water permeability and salt rejection determines
RO process performance. Table 1 lists experimentally determined intrinsic salt rejections and
CP moduli, plus CP moduli predicted by the FT,
RS, and CD models.
According to results in Table 1, CP modulus
increases with increasing applied pressure, decreasing Reynolds number, and decreasing bulk
concentration. All model and experimental results
are in qualitative agreement. Higher applied pressure and lower crossflow rate make the CP modulus larger by increasing permeate convection and
decreasing shear rate (mass transfer), respectively.
At the same applied pressure and crossflow rate,
the permeate flux decreases with increasing bulk
concentration because the feed solution osmotic
pressure is larger. Salt rejection increased with
applied pressure and crossflow, but decreased with
increasing ionic concentration.
All three models underestimated the CP modulus at the lowest operating pressure. At the intermediate pressure, the FT and CD models were
quite accurate, but the RS model dramatically

underestimated CP. At the highest pressure, the


FT and CD models overestimated the CP modulus,
whereas the RS model underestimated CP again.
All three models were increasingly accurate as
ionic concentration increased. At a given applied
pressure, the accuracy of all models improved at
higher crossflow.
The overall average prediction errors by the
FT, RS, and CD models were 4.0 10.4, 37.8
7.9, and 2.1 10.0%, respectively, where the
value represents the standard deviation of all percent differences. A negative percentage error
means that the average predicted value is smaller
than the experimental one. The RS model consistently underestimated the CP modulus by as
much as 51%, while the FT and CD models never
deviated by more than 29 and 26%, respectively.
The numerical CD model was most accurate, but
the predicted CP moduli by the FT model were
also quite reasonable.
4.2. Accuracy of models for predicting flux
The predictive accuracy of the FT, RS, and CD
models were tested by comparison to experimental
flux data. Fig. 2 shows channel averaged permeate
flux produced by model predictions and experimental measurements. For the FT model, the
channel averaged permeate fluxes were predicted
directly. For the RS and CD models, the channel
average fluxes were determined by integrating the
locally predicted permeate flux over the entire
channel. All predicted and measured flux data
increased as applied pressure increased, as Reynolds number increased , and as feed concentration
decreased. Higher applied pressures induce a
larger driving force for permeation. Higher Reynolds number flows enhance mass transfer, which
causes a smaller CP modulus and osmotic pressure
drop resulting in a larger relative permeate flux.
Increasing bulk salt concentration causes a higher
osmotic pressure drop, which results in a smaller
permeate flux.
As shown in Fig. 2, fluxes predicted by the FT
and CD model are quite close to the experimental

50 (1170) 94.1% 1.42 1.37 -3.5% 1.09 -23.2% 1.40 -1.4% 96.1% 1.70 1.72 1.2% 1.18 -30.6% 1.76 3.5% 97.1% 1.99 2.13 7.0% 1.35 -32.2% 2.17 9.0%

20 (1170) 95.9% 1.89 1.60 -15.3% 1.18 -37.6% 1.66 -12.2% 97.0% 2.17 2.12 -2.3% 1.33 -38.7% 2.19 0.9% 97.5% 2.54 2.73 7.5% 1.60 -37.0% 2.80 10.2%

10 (1170) 96.6% 2.43 1.73 -28.8% 1.22 -49.8% 1.80 -25.9% 96.9% 2.38 2.29 -3.8% 1.38 -42.0% 2.39 0.4% 97.3% 2.76 3.04 10.1% 1.71 -38.0% 3.13 13.4%

93.7% 1.48 1.41 -4.7% 1.10 -25.7% 1.44 -2.7% 96.0% 1.83 1.83 0.0% 1.23 -32.8% 1.87 2.2% 97.0% 2.23 2.30 3.1% 1.47 -34.1% 2.33 4.5%

CD % err

50 (590)

p = 1380 kPa
EX FT % err RS % err

95.9% 2.01 1.71 -14.9% 1.20 -40.3% 1.76 -12.4% 96.9% 2.39 2.32 -2.9% 1.42 -40.6% 2.39 0.0% 97.5% 2.87 3.07 7.0% 1.82 -36.6% 3.12 8.7%

Ri

20 (590)

CD % err

96.5% 2.57 1.86 -27.6% 1.25 -51.4% 1.96 -23.7% 97.0% 2.69 2.59 -3.7% 1.51 -43.9% 2.68 -0.4% 97.4% 3.21 3.53 10.0% 2.00 -37.7% 3.59 11.8%

p = 1035 kPa
EX FT % err RS % err

10 (590)

Ri

93.0% 1.62 1.54 -4.9% 1.15 -29.0% 1.57 -3.1% 95.4% 2.09 2.04 -2.4% 1.39 -33.5% 2.07 -1.0% 96.6% 2.72 2.67 -1.8% 1.82 -33.1% 2.67 -1.8%

% err

50 (290)

CD

95.7% 2.31 1.98 -14.3% 1.31 -43.3% 2.04 -11.7% 96.8% 2.97 2.85 -4.0% 1.76 -40.7% 2.89 -2.7% 97.5% 3.89 3.92 0.8% 2.51 -35.5% 3.88 -0.3%

p = 690 kPa
% err RS % err

20 (290)

EX FT

96.4% 2.91 2.27 -22.0% 1.39 -52.2% 2.34 -19.6% 96.9% 3.49 3.35 -4.0% 1.95 -44.1% 3.40 -2.6% 97.6% 4.68 4.89 4.5% 2.96 -36.8% 4.79 2.4%

Ri

10 (290)

c b (Re)

Table 1
Predicted and experimental CP moduli

S. Kim, E.M.V. Hoek / Desalination 186 (2005) 111128


119

S. Kim, E.M.V. Hoek / Desalination 186 (2005) 111128

120

28
24

Permeate flux (m/s)

RS model

Experiment
FT model
RS model
CD model

20
16

<

10 mol/m

<

20 mol/m

<

50 mol/m

Experiment,
FT model,
CD model

12

10 mol/m >
3

20 mol/m >

8
3

50 mol/m >

4
400

600

800

1000

1200

1400

1600

Applied pressure (kPa)


(a) Crossflow Reynolds number = 290

28

Experiment
FT model
RS model
CD model

Permeate flux (m/s)

24

<

10 mol/m

<

20 mol/m

<

50 mol/m

20
16

Experiment,
FT model,
CD model

12

10 mol/m >

20 mol/m >

50 mol/m >

4
400

600

800

1000

1200

Applied pressure (kPa)


(b) Crossflow Reynolds number = 590

1400

1600

S. Kim, E.M.V. Hoek / Desalination 186 (2005) 111128

28

Experiment
FT model
RS model
CD model

Permeate flux (m/s)

24

121
3

<

10 mol/m

<

20 mol/m

<

50 mol/m

1400

1600

20
16

Experiment,
FT model,
CD model

12

10 mol/m >

3
3

20 mol/m >

50 mol/m >

4
400

600

800

1000

1200

Applied pressure (kPa)


(c) Crossflow Reynolds number = 1170
Fig. 2. Channel averaged permeate fluxes determined experimentally (circles) and predicted (lines) by film theory (FT),
retained solute (RS), and convection diffusion (CD) models plotted as a function of applied pressure and salt concentration for crossflow Re of (a) 290 (b) 590 (c) 1170. Constant experimental and simulation conditions employed were
temperature and pure water permeability of 298 K and 2.151011 mPa1s1. Solution pH was 5.80.2.

flux data, while fluxes predicted by the RS model


appear significantly off scale. Although the slope
of pressure-flux predictions by the FT and CD
models is not identical to the experimental data,
the slope of RS model prediction gives a less
accurate slope. The average prediction errors of
the FT, RS, and CD models were 2.0 3.6,
21.9 12.1 and 0.5 3.6%, respectively. These
errors were less than those of the CP modulus
predictions. The CP modulus prediction errors
result in the deviation of predicted osmotic pressure from the real value. The osmotic pressure is
used to predict the channel average permeate flux
with the pure water permeability and transmembrane pressure using Eq. (5). When channel
average values are compared, the prediction errors
were significantly reduced. The FT and CD
models appear to be most accurate for the highest

ionic strength and lowest pressure, which are the


most realistic set of conditions for practical RO
membrane desalination processes. It appears that
the FT and CD models reasonably predicted permeate flux of the experiments performed.
4.3. Accuracy of models for predicting rejection
Observed salt rejection cannot be predicted by
the RS model because it is an input parameter for
the model. The observed salt rejection may be determined experimentally or by a non-equilibrium
thermodynamic model [29]. Since the bulk concentration (cb) is an input for all models, observed
salt rejection (Ro = 1 cp/cb) is determined by predicting permeate concentration (cp). The permeate
concentration is determined from the product of
salt passage and the salt concentration on the
membrane surface, cp = (1 Ri) cw.

122

S. Kim, E.M.V. Hoek / Desalination 186 (2005) 111128

Intrinsic salt rejection is an input for the FT


and CD models and was determined experimentally for every set of operation conditions simulated. Since the intrinsic rejection is known, the
observed salt rejection predictions are solely dependent on the accuracy of predictions of CP
modulus. In Figs. 3a and 3b the predictions of
observed salt rejection by the FT and CD models
are plotted against experimental data. The average
prediction errors of the FT and CD models were
0.2 0.1 and 0.3 0.9%, respectively. Hence,
1.0

(a)

FT model

0.9

0.8

0.7

0.6
0.6

0.7

0.8

0.9

1.0

0.9

1.0

Experiment
1.0

(b)

CD model

0.9

0.8

0.7

0.6
0.6

0.7

0.8

Experiment

Fig. 3. The comparison between experimental and observed salt rejection (Ro) by (a) film theory (FT) model
and (b) numerical convection-diffusion (CD) model. The
experimental and simulation conditions are the same as
those for Fig. 2.

it can be concluded that the FT and CD models


give reasonably accurate prediction of salt rejection, as well as CP modulus and permeate flux.
4.4. Sources of errors in modeling CP, rejection
and flux
Potential sources for deviations between model
predictions and experimental results include entrance, exit, and side wall effects as well as density
differences for high salinity feed and low salinity
permeate solutions. For example, the entrance
length, over which a full developed parabolic flow
profile forms, is predicted by Lu = 0.08Reh,
where Re = u ( 2h ) / [30]. For Reynolds numbers
of 38, 191, 380 and 1910, Lu/L is 0.02, 0.09, 0.18
and 0.91 for the crossflow channel geometry used
in this study. Hence, the assumption of a fully
developed parabolic profile at the inlet is not satisfied for any experimentally employed conditions,
and is a source of common error for predictions
by all three models.
The analytical retained solute model is based
on a concept introduced by Song and Elimelech
[22] to describe particle concentration polarization
in membrane filtration processes. Use of the retained solute concept enabled analytical solution
of a two-dimensional convection-diffusion mass
balance applicable for crossflow membrane filtration. This approach worked well to predict the
limiting flux during ultrafiltration of nanoparticle
suspensions [31]. However, the RS model did not
accurately predict concentration polarization
phenomena for the reverse osmosis experiments
reported in this study. A possible explanation for
the inaccuracy of the retained solute model in
predicting concentration polarization is provided
below.
At the membrane surface, net solute mass
transfer by convection and diffusion is equal to
the solute flux through membrane, that is,
c
vw cw + D = vw c p
y y = 0

(19)

S. Kim, E.M.V. Hoek / Desalination 186 (2005) 111128

Here, y = 0 means at the membrane surface.


Eq. (19) can be transformed by substituting c = cb
+ C, where C is the excess solute concentration at
the membrane surface or the retained solute concentration, into Eq. (19) yielding

( cb + C )
vw ( cb + Cw ) + D
= vwc p
y

y =0

(20)

Here Cw is the excess concentration at the membrane surface. Since the bulk concentration (cb)
is constant, the derivative is zero.
Rearranging the left hand side of Eq. (20) and
applying the definition of retained solute, i.e.,
Eq. (8), yields

C
vw cb + vwC + D
= vw c p
y y = 0

(21a)

or

C
vw C + D
= vw c p vwcb
y y = 0

(21b)

The term in parentheses is the retained solute


mass balance for the concentration polarization
domain. According to Eq. (21b), this term does
not equal zero at the membrane surface (y = 0), as
specified in the development of Eq. (8), because
the bulk and permeate concentrations are significantly different and non-negligible in RO
separations. This means that Eq. (8) cannot be used
as a boundary condition at the membrane surface.
However, the assumption that Eq. (8) is true
just at the membrane-solution interface is the key
step in deriving an analytical solution to the
integral in Eq. (9) and, ultimately, arriving at the
closed form analytical solution. The assumption
is acceptable if the bulk and permeate concentrations are approximately equal and if both are
negligible compared to the concentration at the
membrane surface (wall) as is the case for

123

ultrafiltration of dilute particle suspensions.


However, in RO separations the solute wall concentration is only slightly elevated over the bulk;
hence, the bulk concentration is not negligible.
The lack of mass conservation is likely the primary
source of error in predictions by the RS model;
however, the retained solute model does more
realistically simulate the fundamental interplay
between permeation, crossflow, rejection, and
concentration polarization than does film theory.
Thus, the retained solute model offers significant
conceptual insight into concentration polarization
phenomena in all crossflow membrane processes.
4.5. Applicability of film theory for RO processes
Film theory models are widely preferred by
process engineers for their simplicity and reasonable performance predictions. The data of Figs. 2
and 3 demonstrates that the film theory model
predicts permeate water flux and salt rejection
nearly as well as a more rigorous numerical convectiondiffusion model. For the experimental
conditions tested, the FT model was less accurate
than the CD model at low fluxes, but slightly more
accurate at the higher fluxes examined for all
crossflows and salt concentrations. Although this
is counter-intuitive, the film theory neglects the
effect of solute permeation through the membrane,
which may suppress the boundary layer thickness
at high fluxes.
Since film theory has been shown to give
reasonable predictions of CP, flux, and rejection,
it is of interest to further compare the FT and CD
models across an array of operating conditions,
membrane properties, and solution chemistries
likely to be encountered in real applications. In a
new set of simulations, constant conditions assumed were temperature, water permeability (A), and
salt rejection (Ri) of 298 K, 1.121011 mPa1s1,
and 97%. Table 2 summarizes the simulation conditions that were varied and Fig. 4 compares channel averaged permeate flux and observed salt
rejection data predicted by both models.

124

S. Kim, E.M.V. Hoek / Desalination 186 (2005) 111128

Table 2
Simulation conditions for the comparison of FT and CD models

#
Re
p, kPa
cb, mol/m3

30

10

11

38
912
50

191
912
50

380
912
50

1910
912
50

191
576
50

191
1580
50

191
2930
50

191
912
10

191
912
100

191
912
500

191
912
1000

1.0

(a)

(b)

25

FT model

FT model (m/s)

0.9

20
15
10

0.8

0.7

5
0

10

15

20

25

30

0.6
0.6

0.7

0.8

0.9

1.0

CD model

CD model (m/s)

Fig. 4. The comparison between the film theory (FT) and the numerical convection-diffusion (CD) model predictions for
(a) channel averaged permeate flux and (b) observed salt rejection. Constant simulation conditions employed were temperature, water permeability, and salt rejection of 298 K, 1.121011 mPa1s1, and 97%. Table 2 summarizes the simulation conditions that were varied. Crossflow hydrodynamics were varied within the laminar range (Re < 2100).

Crossflow hydrodynamics were varied within


the laminar range (Re < 2100). Permeate fluxes
employed were in the range frequently used in
real operations and higher. Salt concentrations
ranged from fresh water conditions to about twice
the molarity of seawater. Permeate flux and observed salt rejection data predicted by both models
are well-matched according to the data in Fig. 4.
Overall there was a difference of 3.5 2.8%, where
the FT model consistently over predicted the permeate flux and salt rejection because it under predicted the CP modulus.
4.6. Using film theory to predict local concentration polarization
Accurate prediction of local concentration

polarization phenomena is very important for


reverse osmosis processes because, for example,
it enables prediction of the onset of scaling by
sparing soluble minerals. The FT model may be
used to predict the local concentration polarization
modulus, cw/cb, via

cw ( x )

v ( x )
= 1 Ri + Ri exp w

cb ( x )
k ( x )

(22)

where vw(x) is the local permeate flux predicted


by the CD model, k(x) is the local mass transfer
coefficient predicted by Eq. (3), and cw(x) and cp(x)
denote local wall and permeate concentration,
respectively.

S. Kim, E.M.V. Hoek / Desalination 186 (2005) 111128

Note the local permeate flux predicted by the


CD model was used as an input for CP modulus
predictions by the FT model because film theory
can not predict the local flux and CP modulus independently. It is more common to assume a constant permeation velocity when applying film
theory to RO processes; however, this has no effect
on the local film layer thickness, but does subtly
influence the local solute rejection. Also, simulations were performed for the geometry of the lab
scale filter used in the experiments of this study.
Hence, the assumption of a constant bulk concentration introduces little error because the recovery
in this system ranges from about 0.11.0%.
Clearly, a separate mass balance for the axially
varying retentate concentration must be performed
for systems with significant recovery [26].
Figs. 5 and 6 describe local CP modulus (cw(x)/
cb) predicted by FT and CD model for various
inlet permeate fluxes and crossflow Reynolds
numbers. The difference between each predicted
factor increases as the initial flux at the membrane

12
11

CP modulus (cw/cb)

10
9

125

inlet becomes larger (Fig. 5) and Reynolds number


decreases (Fig. 6). Film theory works well when
CP factor is small (e.g., higher crossflow Reynolds
number and lower operating pressure). At permeate fluxes below 7.5410 6 m/s (~16 gfd,
gallons per square foot per day), the FT model
accurately predicts local concentration polarization. The key source of error at higher fluxes
relates to independence of the boundary layer
thickness from permeation. Considering most RO
applications employ fluxes smaller than 7.54
106 m/s, it appears reasonable to apply film theory
based concentration polarization models.
5. Conclusions
A two dimensional, numerical convectiondiffusion mass balance model was developed to
enable more fundamental description of concentration polarization phenomena in reverse osmosis
processes. Predictions by two analytical concentration polarization models and the numerical

p
FT model CD model
576 kPa
912 kPa
1580 kPa
2930 kPa

8
7
6
5
4
3
2
1
0.0

0.2

0.4

0.6

0.8

1.0

x/L
Fig. 5. Prediction of local CP moduli depending on different applied pressures using the film theory (FT) and the numerical convection-diffusion (CD) models. Constant simulation conditions employed were temperature, water permeability,
and salt rejection of 298 K, 1.121011 mPa1s1, and 97%. Simulation conditions employed were combinations #2, 5, 6,
and 7 in Table 2. Crossflow hydrodynamics were varied within the laminar range (Re < 2100).

126

S. Kim, E.M.V. Hoek / Desalination 186 (2005) 111128

CP modulus (cw/cb)

Re FT model CD model
38
190
380
1900

1
0.0

0.2

0.4

0.6

0.8

1.0

x/L
Fig. 6. Predicted local CP moduli depending on different crossflow Reynolds numbers using the film theory (FT) and the
numerical convectiondiffusion (CD) models. Constant simulation conditions employed were temperature, water permeability, and salt rejection of 298 K, 1.121011 mPa1s1, and 97%. Simulation conditions employed were conditions 14
in Table 2.

model were compared to experimental RO separation performance data. The three models allowed
the prediction of local solute concentration polarization, permeate water flux, and permeate salt concentration given inputs of operating parameters,
filter geometry, and membrane properties. Predictions by film theory and the numerical convectiondiffusion model compared favorably to experimental data and to each other, providing accurate and consistent predictions of observed water
flux and salt rejection. An important result of this
study is the validation of film theory as a simple
and reliable indicator of concentration polarization
phenomena in reverse osmosis separations, particularly for mechanistic studies of fouling and
scaling conducted in laboratory scale crossflow
membrane filters. Further, the use of film theory
based concentration polarization models in real
RO processes appears justified given the specific
operating conditions of high crossflow and low
flux used in most practical applications.

Acknowlegements
The research described above was performed
while Dr. Suhan Kim was a postdoctoral fellow
at University of California, Riverside. Partial
financial support for this work was obtained
through a post-doctoral fellowship from the Korea
Science and Engineering Foundation (KOSEF) for
Dr. Kim. Additional support was obtained from
the California Department of Water Resources
through the Desalination Research Innovation
Partnership (DRIP), which is managed by the
Metropolitan Water District of Southern California.
Symbols
A
C
cb
cp

Pure water permeability, mPa1s1


Concentration of retained solute, molm3
Solute concentration in the bulk, molm3
Solute concentration in the permeate,
molm3

S. Kim, E.M.V. Hoek / Desalination 186 (2005) 111128

cw

Wall concentration for the rejecting salt,


molm3
D Solute diffusion coefficient, m2s1
dH Channel hydrodynamic parameter, m
F Dimensionless driving force
fos Osmotic coefficient, Pamol1
h
Channel half-height, m
k
Mass transfer coefficient, ms1
k Channel averaged mass transfer coefficient, ms1
L
Channel length, m
Ns Dimensionless membrane resistance
p
Pressure, Pa
Q Axial flow rate, m3s1
R Universal gas constant,
= 8.314 Jmol1K1
Ri Intrinsic salt rejection of membrane
(e.g., Ri = 1 cp/cb)
Ro Observed salt rejection (e.g., Ro = 1
cp/cb)
r
Distance from center line of a rectangular channel, m
Re Reynolds number (e.g., Re = ud H / )
Sc Schmidt number (e.g., Sc = v/D)
Sh Sherwood number [e.g.,Sh = kd H / D =
1.85 (ReScdH/L)1/3]
T
Absolute temperature, K
u
Axial components of velocity filed, ms1
u Bulk average crossflow velocity, ms1
umax Maximum crossflow velocity at channel
center, ms1
v
Transverse components of velocity filed,
ms1
vw Permeate water flux, ms1
X Dimensionless longitudinal distance,
xL1
x
Axial coordinate, m
x Dummy integration variable
y
Transverse coordinates, m
Greek

Mass boundary (film) layer thickness,


m
Feed conductivity, Sm1

p
 w

127

Permeate conductivity, Sm1


Wall shear rate, s1 (e.g.,  w = 3u / h )
Viscosity of water, Nsm2
Kinematic viscosity of water, m2s1
Osmotic pressure difference between
feed and permeate sides of the membrane,
Pa (e.g., 0 = fos (cb cp) = fosRocb)
m Transmembrane osmotic pressure, Pa

References
[1] M. Mulder, Basic Principles of Membrane Technology. Kluwer Academic Publishers, Dordrecht,
NL, 1991.
[2] B.J. Marinas and R.I. Urama, Modeling concentration polarization in reverse osmosis spiral-wound
elements, J. Environ. Eng.-ASCE, 122 (1996) 292.
[3] K.K. Sirkar and G.H. Rao, Approximate design
equations and alternate design methodologies for
tubular reverse-osmosis desalination, Ind. Eng.
Chem. Prod. Res. Develop., 20 (1981) 116.
[4] K.K. Sirkar, P.T. Dang, and G.H. Rao, Approximate
design equations for reverse-osmosis desalination
by spiral-wound modules, Ind. Eng. Chem. Prod.
Res. Develop., 21 (1982) 517.
[5] E.M.V. Hoek and M. Elimelech, Cake-enhanced
concentration polarization: A new fouling mechanism for salt-rejecting membranes, Environ. Sci.
Technol., 37 (2003) 5581.
[6] S. Bhattacharjee and G.M. Johnston, A model of
membrane fouling by salt precipitation from multicomponent ionic mixtures in crossflow nanofiltration, Environ. Eng. Sci., 19 (2002) 399.
[7] A.L. Zydney, Stagnant film model for concentration
polarization in membrane systems, J. Membr. Sci.,
130 (1997) 275.
[8] A.S. Michaels, New separation technique for the
CPI, Chem. Eng. Prog., 64 (1968) 31.
[9] W.F. Blatt, A. Dravid, A.S. Michaels and L. Nelson,
Solute polarization and cake formation in membrane
ultrafiltration: Causes, consequences, and control
techniques, in J.E. Flinn, ed., Membrane Science
and Technology: Industrial, Biological, and Waste
Treatment Processes, Plenum Press, Columbus,
Ohio, 1970, p. 47.
[10] M.C. Porter, Concentration polarization with membrane ultrafiltration, Ind. Eng. Chem. Prod. Res.
Develop., 11 (1972) 234.

128

S. Kim, E.M.V. Hoek / Desalination 186 (2005) 111128

[11] S. Bhattacharjee, A.S. Kim and M. Elimelech, Concentration polarization of interacting solute particles
in cross-flow membrane filtration, J. Colloid Interf.
Sci., 212 (1999) 81.
[12] S. Bhattacharjee, J.C. Chen and M. Elimelech,
Coupled model of concentration polarization and
pore transport in crossflow nanofiltration, AIChE
J., 47 (2001) 2733.
[13] D. Bhattacharyya, S.L. Back, R.I. Kermode and
M.C. Roco, Prediction of concentration polarization
and flux behavior in reverse osmosis by numerical
analysis, J. Membr. Sci., 48 (1990) 231.
[14] K. Madireddi, R.B. Babcock, B. Levine, J.H. Kim
and M.K. Stenstrom, An unsteady-state model to predict concentration polarization in commercial spiral
wound membranes, J. Membr. Sci., 157 (1999) 13.
[15] A. Berman, Laminar flow in channels with porous
walls, J. Appl. Phys., 24 (1953) 1232.
[16] S.K. Karode, Laminar flow in channels with porous
walls, revisited, J. Membr. Sci., 191 (2001) 237.
[17] S.W. Ma, L.F. Song, S.L. Ong and W.J. Ng, A 2-d
streamline upwind petrov/galerkin finite element
model for concentration polarization in spiral wound
reverse osmosis modules, J. Membr. Sci., 244 (2004)
129.
[18] D.E. Wiley and D.F. Fletcher, Techniques for computational fluid dynamics modelling of flow in membrane channels, J. Membr. Sci., 211 (2003) 127.
[19] S.K. Karode and A. Kumar, Flow visualization
through spacer filled channels by computational
fluid dynamics i. Pressure drop and shear rate calculations for flat sheet geometry, J. Membr. Sci., 193
(2001) 69.
[20] C.O. Bennett and J.E. Myers, Momentum, Heat, and
Mass Transfer, 2nd ed., McGraw Hill, Inc., New
York, NY, 1974.

[21] L.F. Song and S.C. Yu, Concentration polarization


in cross-flow reverse osmosis, AIChE J., 45 (1999)
921.
[22] L.F. Song and M. Elimelech, Theory of concentration polarization in cross-flow filtration, J. Chem.
Soc. Faraday Trans., 91 (1995) 3389.
[23] R.F. Probstein, Physicochemical Hydrodynamics:
An Introduction. 2nd ed., John Wiley & Sons, Inc.,
New York, 1994.
[24] A.L. Zydney and C.K. Colton, A concentration
polarization model for the filtrate flux in cross-flow
microfiltration of particulate suspensions, Chem.
Eng. Commun., 47 (1986) 1.
[25] R.H. Davis, Modeling of fouling of cross-flow
microfiltration membranes, Sep. Purif. Methods, 21
(1992) 75.
[26] L. Song, S. Hong, J.Y. Hu, S.L. Ong and W.J. Ng,
Simulations of full-scale reverse osmosis membrane
process, J. Environ. Eng.-ASCE, 128 (2002) 960.
[27] E.M.V. Hoek, A.S. Kim and M. Elimelech, Influence
of crossflow membrane filter geometry and shear
rate on colloidal fouling in reverse osmosis and
nanofiltration separations, Environ. Eng. Sci., 19
(2002) 357.
[28] I. Sutzkover, D. Hasson and R. Semiat, Simple
technique for measuring the concentration polarization level in a reverse osmosis system, Desalination,
131 (2000) 117.
[29] L.F. Song, Thermodynamic modeling of solute
transport through reverse osmosis membrane, Chem.
Eng. Commun., 180 (2000) 145.
[30] H. Schlichting, Boundary-Layer Theory. 7th ed.,
McGraw-Hill, Inc., New York, NY, 1979.
[31] L.F. Song, A new model for the calculation of the
limiting flux in ultrafiltration, J. Membr. Sci., 144
(1998) 173.

You might also like