Journal of Membrane Science 271 (2006) 38–46

A numerical study on concentration polarization and system

performance of spiral wound RO membrane modules
Wenwen Zhou, Lianfa Song ∗ , Tay Kwee Guan
Division of Environmental Science & Engineering, National University of Singapore, 10 Kent Ridge Crescent, Singapore 119260, Singapore

Received 7 December 2004; received in revised form 1 July 2005; accepted 8 July 2005
Available online 15 August 2005

Abstract

The development of concentration polarization in a spiral wound reverse osmosis membrane channel and the depolarization effect of
spacers are important concerns for understanding the performance of membrane processes. However, an accurate quantification of these
effects derived from fundamental principles is impractical due to the complexity of the processes. In this study, a macroscopic method was
developed to estimate the effect of concentration polarization on the performance of the spiral wound membrane modules. Concentration
polarization in a channel filled with spacers was described as a combination of two extreme cases, namely the undisturbed concentration
polarization and complete depolarization (uniform distribution across the channel height). With the introduction of a polarization factor
for the degree of concentration polarization, a mathematical model was developed for the permeate flux in the spiral wound modules. The
proposed model was solved numerically to simulate the performance of a long membrane channel under various operation conditions. The
simulation results demonstrated that the model developed in this study was a feasible way to estimate concentration polarization in spiral
wound modules. Excellent fitness was found between the numerical simulations and experimental observations of the average permeate fluxes
in along membrane channel of spiral wound membrane modules.
© 2005 Elsevier B.V. All rights reserved.

Keywords: Concentration polarization; Spiral-wound module; Reverse osmosis; Mathematical modeling; Spacers

1. Introduction tion [3]. The behavior of a small piece of RO membrane can

Reverse osmosis (RO) processes have been widely used
for separation and concentration of solutes in many fields,
such as chemical and biomedical industry, food and beverage
processing, as well as water and wastewater treatment [1,2].
With the shrinkage of water sources and the more stringent
standards for drinking water quality, the applications of RO
membrane in water reclamation and seawater desalination
will continue to grow in the near future. Thus, a better model
for describing RO process performance is highly desirable
for designing and optimizing the system to further improve
its cost-effectiveness.
The performance of an RO process is generally governed
by membrane properties, operating conditions and feed solu-
∗ Corresponding author. Tel.: +65 874 8796; fax: +65 6779 1635.
E-mail address: eseslf@nus.edu.sg (L. Song).
be described with fundamental transport theories and models
[2,4–6]). However, it becomes questionable when these mod-
els are directly applied to describe the performance of long
membrane filtration channels as employed in full scale RO
filtration systems. For example, according to the membrane
transport theories, permeate flux is linearly related to driving
pressure, which has been well demonstrated with laboratory-
scale RO systems and small flat-sheet membrane processes.
However, it is observed in full-scale RO processes that the
average permeate flux does not always increase linearly with
driving pressure. Because the process parameters may vary
substantially along a long membrane channel, the average
permeate flux in the membrane channel cannot be ade-
quately described with the basic membrane transport theories
[7–9].
Many studies have been conducted to investigate the
mechanisms that control the performance of full-scale RO

0376-7388/$ – see front matter © 2005 Elsevier B.V. All rights reserved.
doi:10.1016/j.memsci.2005.07.007
 W. Zhou et al. / Journal of Membrane Science 271 (2006) 38–46 39

processes, among which concentration polarization is usu-

ally considered as a major factor for the deviation of the
average permeate flux from being linearly increasing with
driving pressure. For a better description of RO system, con-
siderable efforts have been made to estimate the mass transfer
coefficient in RO processes affected by concentration polar-
ization [10–12]. Recently, Song et al. [7] suggested that salt
accumulation along a long membrane channel might lead to
an equilibrium state under certain circumstances. At the equi-
librium state, the osmotic pressure of brine at some point in
the membrane channel becomes equal to the driving pressure
and permeate flux would therefore vanish from that point
onwards. It was further demonstrated that the increase of
salt concentration along the channel length could lead to a
strong non-linear relationship between the average perme-
ate flux and the driving pressure. Model simulations were Fig. 1. Schematic drawings of (a) spiral-wound module and (b) cross-section
compared and agreed with experimental observations in a of a feed channel.
full-scale RO plant [7]. However, concentration polarization
in the spiral wound modules was assumed negligible in these by:
studies without much in-depth delineation.
p(x) −
π(x)
The objective of this study was to develop a macroscopic v(x) = (1)
Rm
method to quantify the depolarization effect by spacers in
the spiral wound modules so that the effect of concentra- where Rm is the membrane resistance,
p(x) and
π(x) are
tion polarization on the performance of a long membrane the driving pressure and the osmotic pressure at x, respec-
channel can be better evaluated. The method treated concen- tively. The osmotic pressure can be determined with the
tration polarization as a combination of two extreme cases: modified van’t Hoff formula by adding an ionization number
undisturbed polarization and completely mixing in the direc- Nion to account for the effect of salt ionization in water,
tion of channel height. A mathematical model was then Nion Rg T
c
developed to investigate the effects of key operating param-
π = (2)
Mw
eters, such as driving pressure, feed salt concentration and
feed velocity, on the performance of the long membrane where Rg is the ideal gas constant (=8.31 × 103 Pa/(K mol/l)),
channel. Finally, model simulations were compared with T the absolute temperature (K),
c the salt concentration dif-
experiments of a long membrane channel of spiral wound ference across the membrane (mg/l) and Mw is the molecular
modules. weight of the salt (mg/mol).
By applying the mass conservation principle on water, the
cross-flow velocity along the membrane channel, u(x), can
2. Theoretical development be expressed as:


1 x
2.1. Governing equations u(x) = u0 − v(ξ) dξ (3)
H 0
Spiral-wound module is the typical RO element used in where u0 is the feed velocity at the entrance and ξ is the
water and wastewater treatment processes. It is a customary dummy integration variable.
practice to employ several membrane modules in series (up to At steady state, the total longitudinal flux of the retained
seven modules) held in a long pressure vessel for full-scale salt across the height of the channel at any point x is equal
RO processes [13]. Fig. 1 shows schematically (a) a spiral to the total amount of salt retained by the membrane from
wound RO module with (b) a feed channel sandwiched by two the inlet to the location x, which can be expressed by the
pieces of membranes. For convenience, a membrane channel following equation [14,15]:
in the following sections refers half of the channel shown in
H
x
Fig. 1b with a membrane on one side and an impermeable uC dy = rj c0 v(ξ) dξ (4)
wall on the other. The length and height of the membrane 0 0
channel are indicated by L and H, respectively. Permeate flux, where c0 is the feed concentration and rj is the membrane salt
cross-flow velocity and salt concentration in the channel are rejection. C is the concentration of retained salt (the actual
denoted by v(x), u(x) and c(x) with x being the distance from salt concentration c = C + c0 ).
the entrance. The distribution of salt concentration across the channel
According to the principle of membrane transport, perme- height is affected by both concentration polarization induced
ate flux at any point along the channel, v(x), is determined by permeate flow and the mixing facilitated by spacers in
40 W. Zhou et al. / Journal of Membrane Science 271 (2006) 38–46

the channel. In this study, the transverse distribution of the
retained salt is estimated by a combination of two cases. The
first case assumes that cross-flow in the channel is completely
mixed at any cross-section. Therefore, u and C in the left-
hand-side of Eq. (4) are constant with respect to y direction
and the equation can be rewritten as:

x
uCH = rj c0 v(ξ) dξ (5)
0 increasing permeate velocity because concentration polariza-
Rearranging Eq. (5) gives the average retained salt con-
centration at x:

completely mixed scenario. Although one of the purposes of
rj c0 x
Cave = v(ξ) dξ (6)
uH 0
where Cave is the average concentration of the retained salt
determined for the non-polarization case.
Another case considers a spacer-free channel where con-
centration polarization is undisturbed. In this case, the
retained salt concentration in a boundary layer satisfies the
following equation [14,15]:
dC
v(x)C + D =0 (7)
dy
where D is the salt diffusion or disperse coefficient (m2 /s). It
should be noted that Eq. (7) is the mass balance relationship
applied to the “retained salt” only. In most of the studies on
concentration polarization in cross-flow membrane filtration,
a similar relationship was applied to the real salt concen-
tration. Song and Elimelech [16] pointed out that only the
retained salt could be treated as stagnant with respect to the
distance from the membrane surface. Furthermore, Eq. (7)
does not require perfect membranes (100% rejection) because
it is applied only to the retained salt.
The solution of Eq. (7) takes the following form:
C = A e−[v(x)/D]y (8)
in between. Therefore, the actual salt concentration on the
membrane surface takes the following form:
cw = fp Cw + (1 − fp )Cave + c0 (12)
where cw is the salt concentration on the membrane surface,
and fp is the polarization factor that describes the degree
of concentration polarization in the RO channel. One basic
requirement for the choice of fp is that it increases with
tion is induced originally by permeate flow. In the extreme
cases, fp = 1 for the undisturbed polarization and fp = 0 for the
using spacers is to promote mixing and reduce concentration
polarization, there has not been any report in the literature
addressing the effect of spacers on concentration polarization
in the spiral wound RO modules. In this study, the following
formula is proposed for describing the polarization factor:
fp = 1 − e−v(x)H/D (13)
It can be seen from Eq. (13) that the polarization factor
increases monotonically with increasing permeate velocity
with 0 and 1 as two extreme values. The effect of the density
and shape of the spacers on the depolarization does not explic-
itly appear in Eq. (13). However, it is accountable implicitly
in the parameter D, which is termed as “hydraulic dispersion
coefficient” from now on. Owing to the existence of spac-
ers, the value of D can be several orders of magnitude higher
than the molecular diffusion coefficient, dm . The value of the
hydraulic dispersion coefficient can serve as an indicator of
the effectiveness of spacers for promoting mixing.
Combining Eqs. (6), (11)–(13), salt concentration on the
membrane surface under any degree of concentration polar-
ization can be determined as:



rj c 0 x rj c0 v(x) x
cw = c0 + e−v(x)H/D v(ξ) dξ + v(ξ) dξ
uH 0 uD 0
(14)

where A is an integration constant to be determined. The Eq. (14) shows that the depolarization effect of the spacers is
constant A can be determined by substituting Eq. (8) into Eq. more pronounced as D increases. The local permeate velocity
(4): can then be determined from Eqs. (1), (2) and (14):
x N R Tc  r x 
rj c0 v(x) 0 v(ξ) dξ

p − ionMwg 0 1 + e−v(x)H/D uHj 0 v(ξ) dξ
A= (9) v(x) = N R Tc r  x
(15)
uD (1 − e−v(x)H/D ) Rm + ionMwg 0 uDj 0 v(ξ) dξ
Then Eq. (8) becomes Eq. (15) can be used for calculating the permeate flux
x along a membrane channel for any operating conditions. The
rj c0 v(x) 0 v(ξ) dξ
C= e−v(x)y/D (10) case for complete mixing can be obtained by letting (pro-
uD (1 − e−v(x)H/D ) moted hydraulic dispersion coefficient) D → ∞.
Letting y = 0 in Eq. (10), the salt concentration on the mem- The driving pressure decreases along the membrane chan-
brane surface, Cw , can be determined as follows: nel due to the friction of the feed/concentrate flow with the
x channel walls and spacers in the channel. The reduction in
rj c0 v(x) 0 v(ξ) dξ driving pressure over an infinitesimal distance, dx, along a
Cw = (11)
uD (1 − e−v(x)H/D ) channel is calculated as follows [17]:
In a membrane channel with spacers, the salt concentration is 12uη dx

p =
p0 − k (16)
neither completely mixed nor fully polarized, but something H2
 W. Zhou et al. / Journal of Membrane Science 271 (2006) 38–46 41

where k is the friction coefficient due to the existence of
spacers and other irregularities, and η is the viscosity of the
solution.
2.2. Numerical solution
processes under various conditions are numerically investi-
gated. Unless other indicated, the friction coefficient k = 7.0,
the channel length L = 6 m, the channel height H = 0.6 mm,
the salt rejection rj = 1.0 and the number channel segments
n = 500 are used in the numerical simulations.

Because the involvement of v(x) in both sides of the 3.1. Non-linearity relationship in RO process
equation and the cross-flow velocity and driving pressure
are undetermined variables, analytic solution of Eq. (15) is Driving pressure is the driving force for separation in a
unavailable. Instead, a numerical solution procedure for the reverse osmosis process. It is one of the most important oper-
problem is proposed in this section. The membrane chan- ating parameters of a RO process and is usually manipulated
nel is first divided into n segments of equal intervals
x. to optimize the system performance. The effect of driving
The segments are numbered from the entrance to the end pressure on the average permeate flux under low feed salt
of the membrane channel from 1 to n. Correspondingly, the concentrations is plotted in Fig. 2. It can be seen from Fig. 2
concentration, cross-flow velocity, driving pressure and per- that the average permeate flux curves are clearly divided into
meate velocity in segment i are denoted as ci , ui ,
pi and vi , two regions. The average permeate flux increases linearly
respectively. In the first segment, one has: with the driving pressure in the lower pressure region. How-

p0 −
π(c0 ) ever, the flux only increases marginally in the higher pressure
v1 = (17) region.
Rm
The non-linear behavior of the average permeate flux as
u1 = u0 (18) shown in Fig. 2 cannot be explained with the membrane trans-
port theories, which dictates a linear relationship between
c1 = c0 (19)
the permeate flux and the driving pressure [18–20]. This

p1 =
p0 (20) non-linearity is likely a result of salt accumulation along the
membrane channel that exerts an increasing osmotic pressure
For segments 2–n, on membrane transport [21]. Song et al. [8,9] demonstrated

pi−1 −
π(ci−1 ) that in a sufficiently long membrane channel, the salt concen-
vi = (21) tration could reach an equilibrium value at which the osmotic
Rm
pressure is equal to the driving pressure. The recovery of a

x RO process under equilibrium condition (when complete salt
ui = ui−1 − vi−1 (22)
H rejection is assumed) can be calculated as:
vsum = vsum + vi−1 (23)
π0
  −vi H/D  R=1− (28)
rj e rj vi
p
ci = c0 1 + + v
x (24)
ui−1 H ui−1 D where
π0 is the osmotic pressure of the feed. More simu-
12kη lations of the recoveries as a function of driving pressure for

pi =
pi−1 − ui−1
x (25)
H2
With the numerical solutions, the recovery R and average per-
meate flux V, two commonly used parameters for the overall
performance of a RO process, can be easily determined with:
L
v(ξ) dξ un
R= 0 =1− (26)
u0 H u0
and
(u0 − un )H
V = (27)
L

3. Simulations and discussions

In this section, the effect of the spacer on concentra-

tion polarization and system performance of spiral wound Fig. 2. Average permeate fluxes as functions of driving pressure for low feed
RO modules are first estimated with numerical simulations. salt concentrations: (1) 100 mg/l; (2) 500 mg/l; (3) 800 mg/l; (4) 1000 mg/l.
Then the average permeate flux and recovery of membrane L = 6 m, Rm = 1011 Pa s/m, u0 = 0.15 m/s, D = 9.6 × 10−9 m2 /s, T = 298 K.
42 W. Zhou et al. / Journal of Membrane Science 271 (2006) 38–46
Fig. 3. Permeate recoveries as functions of driving pressure for high
feed salt concentrations: (1) 1000 mg/l; (2) 5000 mg/l; (3) 10,000 mg/l;
(4) 20,000 mg/l; (5) 30,000 mg/l. L = 6 m, Rm = 1011 Pa s/m, u0 = 0.15 m/s,
D = 9.6 × 10−9 m2 /s, T = 298 K.
different feed salt concentrations are shown in Fig. 3. Strong
non-linearity between the recoveries and driving pressure is
again observed in Fig. 3. The recoveries from numerical sim-
ulations for different feed salt concentrations are compared
with the corresponding equilibrium values calculated with
Eq. (28) in Table 1. The results show that the recoveries of
all systems under study are equal to or near to corresponding
equilibrium values at pressure of 5.0 × 106 Pa. This finding
suggests that the recovery of the RO process under the sim-
ulation conditions is restricted by the equilibrium. Eq. (28)
shows that the recovery of an equilibrium system is non-
linearly related to the driving pressure.
3.2. Effects of concentration polarization
Concentration polarization was considered as one of the
major factors leading to the non-linearity between the average
permeate flux and driving pressure [22,23]. To investigate the
effect of concentration polarization on the behavior of per-
meate flux, the salt concentrations under different hydraulic
dispersion coefficients are simulated along a membrane chan-
nel and are shown in Fig. 4. It is assumed that the hydraulic
dispersion coefficient can be reasonably approximated by the
diffusion coefficient when there are no spacers present in
the channel. Fig. 4 shows that the salt concentration on the
membrane surface is lower for a larger hydraulic dispersion
coefficient. This observation indicates that the depolarization
Table 1
Comparison of system recoveries with equilibrium values
Feed concentration (mg/l)
1000
Osmotic pressure,
π0 (kPa) 84.7
System recovery from numerical simulations (%) 98.3
Equilibrium recovery by Eq. (28) (%) 98.3
Fig. 4. Salt concentration on membrane surface varying along the membrane
channel for different diffusion coefficients: (1) dm ; (2) 2dm ; (3) 4dm ; (4) 8dm .
Rm = 1011 Pa s/m, u0 = 0.15 m/s,
p0 = 1.5 × 106 Pa, T = 298 K.
effect of the spacers can be well represented by the magni-
tude of the hydraulic dispersion coefficient in this simulation
model. A larger value of hydraulic dispersion coefficient indi-
cates a stronger depolarization effect. It can also be seen from
Fig. 4 that further increase in hydraulic dispersion coefficient
beyond 8dm only results in a marginal change in concen-
tration profile. Therefore, the concentration profile under
D = 8dm can be taken as an approximation of non-polarization
case. Moreover, the development of concentration polariza-
tion along the membrane channel is strongly dependent on
the feed concentration. As shown in Fig. 4, concentration
polarization (the difference between lines 1 and 4) develops
earlier for higher feed salt concentration (5000 mg/l), while
it only becomes significant towards the end of the channel
for lower feed salt concentration (100 mg/l).
The average permeate fluxes of RO processes with var-
ious hydraulic dispersion coefficients are simulated over
a range of applied pressures and plotted in Fig. 5. As
shown in Fig. 5, the average permeate fluxes under differ-
ent hydraulic dispersion coefficients are almost the same for
feed salt concentration of 100 mg/l. It demonstrates that the
effect of concentration polarization on the performance of
RO processes is negligible when the feed concentration is
extremely low. On the contrary, the corresponding values
differ with each other for higher feed salt concentrations,
such as 5000 and 20,000 mg/l. This observation indicates
that the performance of RO process can be strongly affected
by concentration polarization at high feed concentrations,
5000 10000 20000 30000
423.5 847.0 1694.1 2541.1
91.5 83.0 65.7 48.1
91.5 83.0 66.1 49.2
 W. Zhou et al. / Journal of Membrane Science 271 (2006) 38–46
Fig. 5. Average permeate fluxes for different diffusion coefficients: (1)
dm ; (2) 2dm ; (3) 4dm ; (4) 8dm . L = 6 m, Rm = 1011 Pa s/m, u0 = 0.15 m/s,
T = 298 K. The feed concentrations are indicated in the figure.
which is consistent with the findings reported by Song and
Yu [15].
Simulations also show that the adverse effect of concentra-
tion polarization on process performance can be substantially
alleviated with spacers. It is evident from Fig. 5 that the aver-
age flux of dm is significantly lower than that of 8dm for
concentrations of 5000 and 20,000 mg/l. The intensity of mix-
ing within the membrane channel increases with the value of
hydraulic dispersion coefficient D. Mathematically, a limit-
ing case exits where the hydraulic dispersion coefficient D
approaches infinity. Physically, this limiting case represents
a condition where the flow in the membrane channel can
be regarded as completely mixed and concentration polar-
ization is negligible. The enhanced mixing resulting from
hydraulic dispersion with the use of spacers in the chan-
nel can be magnitudes stronger than that attributed to pure
molecular diffusion. That the average flux of 8dm only shows
marginal increase over that of 4dm implies that a hydraulic
dispersion coefficient of 8dm is sufficiently close to the lim-
iting case, where the effect of concentration polarization on
process performance can be neglected.
It should be pointed out that the non-linear relationship
between the average permeate flux and the driving pressure
is fundamentally due to thermodynamic restriction [8,9]. As
demonstrated before, a hydraulic dispersion coefficient of
8dm would produce a mixing condition that is close to com-
plete mixing in which the effect of concentration polarization
on the average permeate flux could be negligible. Therefore,
any attempt to increase average permeate flux by further pro-
moting mixing (with innovative design of spacers) may not
achieve the desired result.
3.3. Effect of channel length
Owing to the effect of salt accumulation, the local per-
meate flux decreases along the RO channel. If the channel
length is sufficiently long, the osmotic pressure of the reten-
43
Fig. 6. Variations of (a) average permeate fluxes and (b) recoveries with
driving pressure in RO processes with different channel lengths: (1) 1 m; (2)
2 m; (3) 3 m; (4) 4 m; (5) 5 m; (6) 6 m. Rm = 1011 Pa s/m, c0 = 1.5 × 103 mg/l,
u0 = 0.15 m/s, D = 9.6 × 10−9 m2 /s, T = 298 K.
tate inside the RO channel can become equal to the applied
pressure. As a result, the local permeate flux will vanish.
Thus, for a sufficiently long RO channel, the average flux
will be controlled by equilibrium mechanism. The recover-
ies or average permeate fluxes in RO channels of different
lengths are simulated under different driving pressures. The
simulation results are plotted in Fig. 6a and b. It can be seen
from Fig. 6a that the 1 m channel is too short to reach equi-
librium state and the average permeate flux of this channel
increases linearly with driving pressure in the pressure range
simulated. For all channels longer than 1 m, the average per-
meate fluxes are shown to level off at different pressures. In
addition, under sufficiently high driving pressures, the aver-
age permeate flux is lower for a longer RO channel.
Fig. 6b shows that all channel lengths, except that of 1 m,
are able to attain the equilibrium recoveries. The recover-
ies increase almost linearly with the pressure before they
become close to the equilibrium value and the equilibrium
recovery increases marginally with pressure. It is also noted
that a longer channel achieves the equilibrium value at a lower
pressure. It is interesting to note from Fig. 6b that the chan-
nel length does not affect the recovery when RO processes
are controlled by equilibrium. This explains the lower aver-
age permeate flux in a longer membrane channel as shown in
Fig. 6a.
3.4. Membrane fouling monitoring
Membrane fouling is one of the most important consid-
erations in the design and operation of a RO system. It is
desirable that the development of membrane fouling can be
accurately characterized and assessed.
The consequence of membrane fouling in channels of dif-
ferent lengths can be rather different. As shown in Fig. 7a and
44 W. Zhou et al. / Journal of Membrane Science 271 (2006) 38–46
Fig. 7. Effects of membrane resistance on (a) average permeate fluxes and
(b) recoveries in RO processes with different channel lengths: (1) 1 m; (2)
2 m; (3) 3 m; (4) 4 m; (5) 5 m; (6) 6 m. c0 = 1.5 × 103 mg/l, u0 = 0.10 m/s,

p0 = 2 × 106 Pa, D = 9.6 × 10−9 m2 /s, T = 298 K.
b, the average permeate flux and recovery of the 1 m chan-
nel decrease considerably with any increase in membrane
resistance. On the contrary, the average permeate fluxes and
recoveries for longer membrane channels remain constant
when the membrane resistance is smaller than some thresh-
old values. When the membrane resistance increases beyond
these threshold values, the average permeate flux and recov-
ery start to decrease with increasing membrane resistance.
The simulations show that the average flux and recovery in
a sufficiently long channel (>3 m) of the highly permeable
membranes may not be affected by the initial changes in
membrane resistance because the systems are controlled by
equilibrium.
The simulations show that membrane fouling cannot be
detected by measuring average permeate flux or recovery
when membrane resistance is below the threshold values. A
new technique for monitoring membrane fouling is needed
so that fouling development in a RO process under any con-
ditions can be accurately assessed.
3.5. Effect of feed flow rate
Effects of feed flow velocity on the average permeate flux
and recovery under different driving pressures are simulated
and presented in Fig. 8a and b, respectively. As shown in
Fig. 8a, the average permeate flux increases linearly with feed
flow velocity until a certain threshold feed velocity is reached.
Below the threshold, the average permeate flux is limited by
the availability of feed water. As a result, an increase in feed
velocity will increase the permeate flux. Beyond the thresh-
old, there is sufficient water in the channel but the average
permeate flux is limited by water transfer capability across
the membrane (approximately proportional to driving pres-
sure). It is obvious from Fig. 8a that a lower pressure results
Fig. 8. Effect of feed velocity on (a) average permeate flux and (b)
recovery under different driving pressures: (1) 5 × 105 Pa; (2) 1 × 106 Pa;
(3) 1.5 × 106 Pa; (4) 2 × 106 Pa; (5) 2.5 × 106 Pa; (6) 3 × 106 Pa. L = 6 m,
Rm = 1011 Pa s/m, c0 = 1.5 × 103 mg/l, D = 9.6 × 10−9 m2 /s, T = 298 K.
in a lower threshold feed velocity for average permeate flux
to level off. The average permeate fluxes decrease slightly
but noticeably beyond the threshold feed velocities for low
driving pressures. These marginal average flux reductions are
attributable to the pressure drop caused by friction between
cross-flow and channel walls, which is relatively significant
under small driving pressures.
Fig. 8b shows the relationship between the recovery and
the feed flow velocity. When the average permeate flux
increases linearly with the feed velocity, the recovery remains
constant according to Eq. (28) until the average permeate
flux reaches its ceiling value. A further increase in feed flow
velocity will result in a reduction in recovery. Fig. 8a and
b jointly suggests that the turning points in both figures are
the best operating points. The process would have a lower
permeate flux when the cross-flow velocity is below the turn-
ing points, while the recovery would decline when operating
with a cross-flow velocity that is higher than the value corre-
sponding to the turning point.
4. Experimental verifications
Pilot-scale filtration tests were conducted with a 4 m
long membrane channel. Four spiral wound RO modules
(TFC 2540SW, Koch Membrane Systems, Wilmington, MA,
USA), which are 1.016 m long (40 in.) and 0.0635 m (2.5 in.)
in diameter, were arranged in two pressure vessels (Advanced
Structures Inc., Escondido, California, USA) connected in
series. The thin-film composite polyamide membrane that
has a high salt rejection and a wide range of working pres-
sure is used in the modules. The average permeate fluxes in
the pilot RO system were measured over a pressure range up
 W. Zhou et al. / Journal of Membrane Science 271 (2006) 38–46 45

an equilibrium established inside the membrane channel.

The conditions for the equilibrium to establish include a
long membrane channel, high membrane permeability and
sufficiently high driving pressure. When the equilibrium is
reached, the permeate recovery is solely determined by the
thermodynamic properties of the membrane system (i.e.,
feed salt concentration, temperature and driving pressure).
Membrane fouling in an equilibrium membrane process is
undetectable by monitoring the average permeate flux.
The depolarization effect of the spacers in the feed channel
is strongly supported by the good fitness between the exper-
imental measurements and numerical simulations with the
model developed in this study. This further strengthens the
claim that the non-linear behavior of the average permeate
flux in a long membrane channel is caused by a thermody-
namic equilibrium in the membrane system.
Fig. 9. Comparison of experimental average permeate fluxes from a
4 m pilot membrane channel with numerical simulations of the sys-
tem. L = 4 m, H = 0.6 mm, Rm = 8.41 × 1010 Pa s/m, rj = 0.99, u0 = 0.075 m/s, Nomenclature
D = 1.8 × 10−8 m2 /s, T = 298 K.
A an integration constant introduced by Eq. (7)
to 2.0 MPa (∼280 psi) for three feed salt concentrations. The and determined in Eq. (9)
experimental results were compared with numerical simula- c salt concentration (mg/l)
tions with the model developed in the previous sections of c0 feed salt concentration (mg/l)
this paper. cw salt concentration on the membrane surface
The experimental average permeate fluxes for salt con- (mg/l)
centrations of 500, 1000 and 3000 mg/l of sodium chloride C retained salt concentration (mg/l)
over a pressure range are plotted in Fig. 9 as symbols, as well Cave average concentration of retained salt across
as the theoretical average fluxes from model simulations as the channel height (mg/l)
lines. It is found that the theoretical fluxes correlate very well Cw concentration of retained salt on the membrane
with the experimental fluxes. The same parameters used in surface (mg/l)
the experiments are employed in the model simulations. The dm salt diffusion coefficient (m2 /s)
hydraulic dispersion coefficient D of 1.8 × 10−8 m2 /s, which D hydraulic dispersion coefficient (m2 /s)
is about 10 times of the molecular diffusion coefficient of fp polarization factor
sodium chloride, dm , is used in the model simulations. The H channel height (m)
good fitness of the simulation results to the experimental data k friction coefficient due to the existence of spac-
indicates that concentration polarization is not a factor on the ers and other irregularities
average permeate flux in the spiral wound modules. L channel length (m)
n number of segments of a channel used in
numerical simulations
5. Conclusions
p driving pressure (Pa)
rj salt rejection
A macroscopic method to investigate concentration polar- R recovery
ization in a spiral-wound membrane channel with spacers is Rg gas constant (=8.31 Pa s or
developed for describing the performance of a long mem- 8.31 × 103 Pa/(K mol/l))
brane channel. The polarization factor is introduced to relate Rm membrane resistance (Pa s/m)
the weights of undisturbed polarization and completely mix- T absolute temperature (K)
ing to salt concentration profile. The effectiveness of spacers u cross-flow velocity (m/s)
in promoting mixing is quantified by the magnitude of the u0 feed flow velocity (m/s)
hydraulic dispersion coefficient. It is demonstrated through v permeate flux (m/s)
numerical simulations that the effect of concentration polar- V average permeate flux (m/s)
ization is negligible in the spiral wound membrane modules
if the hydraulic dispersion coefficient is greater than eight Greek letters
times of the salt diffusion coefficient. η solution viscosity (Pa s)
Numerical simulations on the performance of membrane ξ dummy integration variable
systems under various concentrations show that that the non-
π osmotic pressure (Pa)
linearity of the average permeate flux is mainly caused by
46 W. Zhou et al. / Journal of Membrane Science 271 (2006) 38–46
References
[1] M. Hajeeh, D. Chaudhuri, Reliability and availability assessment of
reverse osmosis, Desalination 130 (2000) 185–192.
[2] L.F. Song, Thermodynamic modeling of solute transport through
reverse osmosis membrane, Chem. Eng. Commun. 180 (2000)
145–167.
[3] V. Todtheide, G. Laufenberg, B. Kunz, Waste water treatment using
reverse osmosis: real osmotic pressure and chemical functionality
as influencing parameters on the retention of carboxylic acids in
multi-component system, Desalination 110 (1997) 213–222.
[4] K.S. Spiegler, O. Kedem, Thermodynamics of hyperfiltration (reverse
osmosis), criteria for efficient membranes, Desalination 1 (1966)
311–326.
[5] M. Soltanieh, W.N. Gill, Review of reverse osmosis membrane and
transport models, Chem. Eng. Commun. 12 (1981) 279–363.
[6] E.A. Mason, H.K. Lonsdale, Statistical–mechanical theory of mem-
brane transport, J. Membr. Sci. 51 (1990) 1–81.
[7] L.F. Song, S.K. Hong, J.Y. Hu, S.L. Ong, W.J. Ng, Simulations of
full-scale reverse osmosis membrane process, J. Environ. Eng.-ASCE
128 (10) (2002) 960–966.
[8] L.F. Song, S.L. Ong, W.J. Ng, M. Elimelech, M. Wilf, Performance
limitation of the full-scale reverse osmosis process, J. Membr. Sci.
214 (2) (2003) 239–244.
[9] L.F. Song, J.Y. Hu, S.L. Ong, W.J. Ng, M. Elimelech, M. Wilf,
Emergence of thermodynamic restriction and its implications for full-
scale reverse osmosis processes, Desalination 155 (2003) 213–228.
[10] Z.V.P. Murthy, S.K. Gupta, Estimation of mass transfer coefficient
using a combined nonlinear membrane transport and film theory
model, Desalination 109 (1997) 39–49.
[11] N.M. Al-Bastaki, A. Abbas, Predicting the performance of RO mem-
branes, Desalination 132 (2000) 181–187.
[12] M. Taniguchi, M. Kurihara, S. Kimura, Behavior of a reverse osmosis
plant adopting a brine conversion two-stage process and its computer
simulation, J. Membr. Sci. 183 (2001) 249–257.
[13] M. Wilf, K. Klinko, Performance of commercial seawater mem-
branes, Desalination 96 (1994) 465–478.
[14] L.F. Song, A new model for the calculation of the limiting flux in
ultrafiltration, J. Membr. Sci. 144 (1998) 173–185.
[15] L.F. Song, S. Yu, Concentration polarization in cross-flow reverse
osmosis, AIChE J. 45 (5) (1999) 921–928.
[16] L.F. Song, M. Elimelech, Theory of concentration polarization in
cross-flow filtration, J. Chem. Soc., Faraday Trans. 91 (19) (1995)
3389–3398.
[17] C.R. Bouchard, P.J. Carreau, T. Matsuura, S. Sourirajan, Modeling
of ultrafiltration: predictions of concentration polarization effects, J.
Membr. Sci. 97 (1994) 215–229.
[18] C.S. Slater, C.A. Brooks, Development of a simulation model pre-
dicting performance of reverse osmosis batch system, Sep. Sci.
Technol. 27 (1992) 1361–1388.
[19] R.I. Urama, B.J. Marinas, Mechanistic interpretation of solute perme-
ation through a fully aromatic polyamide reverse osmosis membrane,
J. Membr. Sci. 123 (1997) 267–280.
[20] A. Kargol, Modified Kedam-Katchalsky equations and their applica-
tions, J. Membr. Sci. 174 (2000) 43–53.
[21] S.K. Karode, Unsteady state flux response: a method to determine
the nature of the solute and gel layer in membrane filtration, J.
Membr. Sci. 188 (2001) 9–20.
[22] W.F. Blatt, A. Dravid, A.S. Michaels, L. Nelsen, Solute polarization
and cake formation in membrane ultrafiltration: causes, consequences
and control techniques, in: J.E. Flinn (Ed.), Membrane Science and
Technology, Plenum Press, New York, 1970.
[23] H.M. Yeh, T.W. Cheng, Analysis of the slip effect on permeate flux
in membrane ultrafiltration, J. Membr. Sci. 154 (1999) 41–51.