Vous êtes sur la page 1sur 8

Journal of Membrane Science 379 (2011) 488495

Contents lists available at ScienceDirect


Journal of Membrane Science
j our nal home page: www. el sevi er . com/ l ocat e/ memsci
Computational uid dynamics simulations of ow and concentration
polarization in forward osmosis membrane systems
M.F. Gruber
a,b,
, C.J. Johnson
c
, C.Y. Tang
d,e
, M.H. Jensen
f
, L. Yde
c
, C. Hlix-Nielsen
a,g,
a
Aquaporin A/S, Ole Maales Vej 3, DK-2200 Copenhagen N, Denmark
b
Nano-Science Center, University of Copenhagen, Universitetsparken 5, DK-2100 Copenhagen , Denmark
c
DHI Water & Environment, Agern Alle 5, DK-2970 Hrsholm, Denmark
d
Singapore Membrane Technology Centre, Nanyang Technological University, Singapore 639798, Singapore
e
School of Civil and Environmental Engineering, Nanyang Technological University, Singapore 639798, Singapore
f
Center for Models of Life, Niels Bohr Institute, Blegdamsvej 17, DK-2100 Copenhagen , Denmark
g
The biomimetic membrane group DTU-Physics, Technical University of Denmark, DK-2800 Kongens Lyngby, Denmark
a r t i c l e i n f o
Article history:
Received 15 April 2011
Received in revised form10 June 2011
Accepted 14 June 2011
Available online 22 June 2011
Keywords:
Forward osmosis
Computational uid dynamics (CFD)
Internal concentration polarization
External concentration polarization
Desalination
a b s t r a c t
Forward osmosis is an osmotically driven membrane separation process that relies on the utilization of
a large osmotic pressure differential generated across a semi-permeable membrane. In recent years for-
ward osmosis has shown great promise in the areas of wastewater treatment, seawater/brackish water
desalination, and power generation. Previous analytical and experimental investigations have demon-
strated how characteristics of typical asymmetric membranes, especially aporous support layer, inuence
the water ux performance in osmotically driven systems. In order to advance the understanding of
membrane systems, models that can accurately encapsulate all signicant physical processes occurring
inthe systems are required. The present study demonstrates a computational uid dynamics (CFD) model
capable of simulating forward osmosis systems with asymmetric membranes. The model is inspired by
previously published CFD models for pressure-driven systems and the general analytical theory for ux
modeling in asymmetric membranes. Simulations reveal a non-negligible external concentration polar-
ization on the porous support, even when accounting for high cross-ow velocity and slip velocity at
the porous surface. Results conrm that the common assumption of insignicant external concentration
polarization on the porous surface of asymmetric membranes used in current semi-analytical approaches
may not be generally valid in realistic systems under certain conditions; specically in systems without
mass-transfer promoting spacers and low cross-ow velocities.
2011 Elsevier B.V. All rights reserved.
1. Introduction
During the last 40 years the separation of aqueous solutions
using pressure-driven membrane systems have been intensely
studied both experimentally and theoretically. Over this time,
many different semi-analytical models have been used to inves-
tigate various features found in the pressure-driven systems: e.g.
effects such as concentration polarization (CP) phenomena [1,2],
changes in solute rejection [3], the effect of wall slip velocity [4,5]
and gravitational effects [6] have been studied.
Developing a generic model to encapsulate all aspects of
membrane ltration in complex geometries can however not be
accomplished using an analytical model. During the last 15 years,

Corresponding author. Tel.: +45 31168385.

Corresponding author. Tel.: +45 27102076.


E-mail addresses: nano.mathias@gmail.com(M.F. Gruber),
Claus.Helix.Nielsen@fysik.dtu.dk (C. Hlix-Nielsen).
computational uid dynamics (CFD) have become increasingly
popular for modeling complex owpatterns in membrane systems
because it provides a more robust approach capable of includ-
ing many parameters. Many different CFD models dealing with
pressure-driven membrane systems have been presented in liter-
ature. Some models focus on the effects of variations in membrane
properties like solute rejectionandsolutionproperties suchas den-
sity, viscosity, and diffusivity [79]. Others are centralized around
mass transfer optimization by changing the case geometry to mod-
ulate an optimal ow, e.g. by using eddy-promoting spacers [9,10].
In recent years, forward osmosis (FO) has emerged as a popu-
lar alternative to conventional pressure-driven membrane systems
suchas microltration(MF), ultraltration(UF), nanoltration(NF)
and reverse osmosis (RO). Unlike pressure-driven processes where
a hydraulic pressure is used to establish solvent ow through a
semi-permeable membrane, FO uses a concentrated drawsolution
and a dilute feed solution to generate solvent ow driven by an
osmotic pressure difference across a semi-permeable membrane.
The main advantages of FO compared to RO is the lack of a need for
0376-7388/$ see front matter 2011 Elsevier B.V. All rights reserved.
doi:10.1016/j.memsci.2011.06.022
M.F. Gruber et al. / Journal of Membrane Science 379 (2011) 488495 489
hydraulic pressure, whichmakes FOpotentiallymorecost-effective
[11,12]. It has also been shown that FO has a lower propensity to
membrane fouling, possibly due to the lack of hydraulic pressure
[1317]. Furthermore, the insertionof proteins withspecic water-
or ion-transporting properties into biomimetic membranes shows
great promise in FO applications [18].
So far all studies found in the literature have observed far
lower water uxes in FO experiments than expected based on
bulkosmotic pressuredifferences andmembranewater permeabil-
ities. The reason for this is that membrane development over the
past four decades has been focused on pressure-driven processes,
and as such current membranes typically utilize an asymmetric
designwitha dense rejectionlayer anda thick porous support layer
for mechanical stability [19]. The porous support has a negligible
inuence when it comes to CP effects in RO, since only CP on the
feed-side of the membrane (i.e. the dense layer) is signicant with
respect to mass transfer [20]. In FO, however, CP effects are impor-
tant on both sides of the membrane, and depending on how the
membrane is positioned either dilutive or concentrative CP will
occur within the porous layer [21]. CP within the porous support is
in a sense effectively decoupled fromthe external ow, and there-
fore poses a reduction in effective osmotic driving force that cannot
be diminished using optimal owconditions. CP within the porous
layer is generally termed internal concentration polarization (ICP)
as opposed to the external concentration polarization (ECP) which
occurs outside the membrane.
Several one-dimensional analytical and semi-analytical models
derived from the work by Lee et al. [21] and Loeb et al. [22] have
been proposed for determining the water ux through compos-
ite membranes in FO systems, many of which correlate well with
experiments [15,21,2325]. In CFD simulations of RO systems, the
membrane is often modeled as being symmetric: that is, as a sin-
gle active separation layer with no porous support [9,10,26]. This
is acceptable since the porous support has an insignicant effect in
this case. In the case of FO the effects of ICP must be included in
the model. To the best knowledge of the authors, no attempts have
been made so far to fully simulate a FOsystemwith an asymmetric
membrane using a CFD approach.
The objective of this paper is to extend the current semi-
analytical models found in the literature used to describe ICP and
water ux in FO systems into an efcient CFD model. Using this
CFD model, we evaluate the semi-analytical approaches as well as
investigate the effect of various major parameters such as cross-
ow velocity, bulk osmotic pressure difference, and slip velocity
on the CP proles and water uxes.
2. Model development
The basis for the model described in this paper is the open
source toolkit OpenFOAM

(Open Field Operation And Manipula-


tion). OpenFOAMis aregisteredtrademarkof OpenCFDLimited, the
producer of the OpenFOAMsoftware. CFDpackages have improved
signicantly during the last decade and OpenFOAM provides an
easily extendable framework with an extensive range of features
and robust solution algorithms. Custom implementation of the
model was performed in two steps: implementation of the solver
for the governing equations and implementation of proper bound-
ary conditions on the membrane surface. The model developed for
the governing equations was validated by using the solver in a RO
test case, for which it was possible to reproduce previously pub-
lished results in [9,26]. The boundary conditions were validated
by comparing obtained water uxes against published measure-
ments; using values for membrane characteristics obtained in [27]
for a chamber corresponding to that used in [27], the boundary
conditions (with a no-slip tangential velocity) resulted in water
uxes within 10% of what was obtained in [27]. This discrepancy
is easily explained based on discussions in Sections 3.2 and 3.3 as
well as various experimental factors, e.g. possibly peristaltic ow,
membrane oscillations, etc.
2.1. Governing equations and membrane model
The ow in the membrane chamber is governed by equations
for conservation of mass and momentum as well as a convection-
diffusion transport equation for the solute mass fraction. The
developed model is capable of solving the governing equations in
three dimensions, but can also be used without modications for
two-dimensional cases. The governing equations used inthe model
are:

t
+ (U) = 0 (1)
U
t
+ (U U) = [(U+U
T
)] p +g (2)
m
A
t
+ (Um
A
) (D
AB
m
A
) = 0 (3)
where symbol denitions can be found in the nomenclature. The
uid is assumed to be isothermal with the density being a function
of solute mass fraction only; i.e. the pressure dependence of the
density is neglected. This is generally knownas a weakly compress-
ible formulation of the governing equations, and has previously
been used successfully for studying pressure-driven membrane
systems [9,26]. The viscosity and diffusion coefcient are allowed
to be functions of the solute mass fraction. The ow is assumed to
be laminar, which is reasonable in most real membrane systems
[10].
The case considered in this paper is one in which the porous
support of a composite membrane faces the drawsolution and the
active separation layer of the membrane faces the feed solution: an
orientation commonly known as the AL-FS orientation. The com-
posite membrane is modeled as a two-dimensional plane which
means that the thickness of the membrane is not resolved, see Fig.
1. The fact that the membrane is modeled as a smooth plane means
that potential effects caused by surface roughness are neglected.
We believe that it is highly unlikely that roughness effects beyond
those represented by a slip velocity, especially on the micro-meter
scale, will signicantly affect the observed results. Assuming zero
hydraulic pressure applied over the membrane, the water ux J
w
can be written as [24]:
J
w
= A(
d,i

f,m
)n
d
(4)
where A is the pure water permeability coefcient,
d,i
is the
osmotic pressure between the porous support and the active layer
of the membrane,
f,m
is the osmotic pressure at the membrane
on the feed side, and n
d
is the unit normal vector on the porous
boundary, i.e. the normal facing the draw solution, see Fig. 1. A is
readily measured in a pressure driven systemwith clean water on
both sides of the membrane; in such a systemthe water ux will be
linearly proportional to the applied pressure with A as the propor-
tionality coefcient [19]. The model does not resolve the osmotic
pressure
d,i
withinthemembrane, soadditional informationabout
what happens in the porous support is required in order to calcu-
late the water ux. The most widely accepted analytical model for
describing the effect of the porous support layer onthe water ux is
that developed by Loeb et al., where a linear relationship between
solute concentration and osmotic pressure is assumed [22]:
J
w
=
1
K
ln
B +A
d,m
B +|J
w
| +A
f,m
n
d
(5)
490 M.F. Gruber et al. / Journal of Membrane Science 379 (2011) 488495
Fig. 1. Schematic representation of the geometry considered in all simulations. The close-up illustrates how the grid is graded such that a very ne resolution is obtained
close to the membrane. The membrane is represented using a dashed line, reecting the fact that the thickness of the membrane is not resolved, but rather described using
Eq. (5).
where B is the solute permeation coefcient and K is a parameter
termed by Lee et al. [21] that describes how easily solute can dif-
fuse into and out of the support layer. Eq. (5) provides the velocity
boundary condition in the normal direction of the membrane. It is
notedthat inthe cases where BA
d,m
andB |J
w
| Eq. (5) reduces
to the following:
J
w
= A(
d,m
e
|J
w
|K

f,m
)n
d
(6)
By comparing Eq. (4) with Eq. (6) an ICP modulus is obtained:

d,i

d,m
= e
|J
w
|K
(7)
The ICP modulus in Eq. (7) is only strictly valid for perfect mem-
branes (i.e. 100% solute rejection) and when assuming a linear
relationship between osmotic pressure and solute concentration.
Nevertheless, the ICP modulus nicely illustrates howthe coefcient
K inuences the ux equation to account for changes in concentra-
tion across the porous support layer.
Ano-slipvelocity boundary conditioninthe tangential direction
is often assumed at the membrane boundary in CFD simulations
[9,26], however this may not be accurate at potentially rough
porous supports [5]. On porous surfaces where the no-slip bound-
ary cannot be applied, the tangential slip velocity U
slip
can be
modeled as being proportional to the shear rate U/n
d
at the
boundary [4,5]:
U
slip
=

U
n
d
(8)
where is the slip coefcient, is the permeability (m
2
) and n
d
is the direction towards the draw which is normal to the mem-
brane. The slip coefcient often ranges from0.5 to 10 and describes
the owhydrodynamics adjacent to and directly inside the porous
support, such that it depends on surface characteristics such as
roughness, pore size and structure. It has for instance been shown
that has a tendency to be higher for a densely packed porous
material than for a less densely packed material [28]. The perme-
ability generally lies in the range of 10
11
m
2
to 10
15
m
2
for
most semi-pervious materials [29]. In this model it is assumed that
and do not depend on permeation velocity and in general that
they are simply membrane constants. The shear rate is obtained
directly from the velocity eld in the model and as such accounts
for potential oweffects caused by membrane permeation. Know-
ing the value for the parameter

/ it is possible to calculate the


slip velocity on the porous boundary explicitly in the CFD model.
The solute ux J
s
through the membrane can be written as
follows assuming that solute separation occurs over the active sep-
aration layer only [24]:
J
s
= B(C
d,i
C
f,m
)n
d
(9)
It is noted that the salt ux is negative to indicate that it is oppo-
site of the water ow. The constant Bis commonly determinedfrom
measurements of the salt separation coefcient R in a pressure-
driven reverse osmosis experiment [19,27]. More specically, B can
be related to R as follows [21,15]:
B =
1 R
R
|J
w
| (10)
R = 1
C
f,m
C
d,m
(11)
Given the denition of the salt separation coefcient in Eq. (11),
it is apparent that R describes how the solute is separated across
the membrane; e.g. a value of 0 means that the membrane is fully
permeable and that the concentration on the feed and draw side
of the membrane will be the same. A value of 1 indicates that the
membrane is completely impervious to a given solute. In a given
cross-ow experiment, one can estimate R using Eq. (11), and Eq.
(10) can then be used to calculate the membrane constant B.
Back to Eq. (9), if we assume a linear relationship between con-
centration and osmotic pressure, i.e. = C, we can combine Eqs.
(9) and (4) to obtain an expression for the solute ux:
J
s
=
B
A
J
w
(12)
Eq. (12) expresses the solute ux J
s
as a function of experimen-
tally determined parameters A, B and along with the water ux
J
w
which is calculated by solving Eq. (5). Knowing the solute ux
through the membrane, it is possible to write boundary conditions
for the solute mass fraction on both sides of the asymmetric mem-
brane because the convective anddiffusive uxes must be balanced
with the solute ux [21]:

m
D
AB
m
A
n
d
n
d
+
m
m
A,m
J
w
= J
s
(13)
This provides a boundary condition for the solute mass fraction
which must be satised at both sides of the membrane. In conclu-
sion, with Eq. (5) for the water ux through the membrane and Eq.
(13) for solute ux balance on the membrane, we have boundary
conditions available for the velocity eld and solute mass fraction
on both sides of the membrane.
2.2. Case geometry and parameters
Simulations were carried out in the simple cross-owchamber
shown in Fig. 1. The chamber was inspired by the one used in [27]
and the dimensions were 14cm6mm. The membrane segment
was 10cm and located in the middle of the chamber. The height
of both the feed and drawchannels were 3mm. The computational
mesh consisted of 150 cells perpendicular to the membrane in both
the feed and draw channel, graded such that the rst grid points
were located within 5mof the membrane in order to capture CP
effects [9]. Gridpoints were within 50m of the upper and lower
chamber walls, which was found to be sufcient. 280 cells were
used along the length of the chamber.
M.F. Gruber et al. / Journal of Membrane Science 379 (2011) 488495 491
At the drawand feed inlet uniforminitial solute mass fractions
were specied. Onbothinlets the component of the pressure gradi-
ent normal totheinlets was set tozero; theerror introducedbysuch
a boundary condition on the inlet pressure is believed to be mini-
mal, especially given the 2cminlet sections, which was conrmed
through a range of verication tests. Fully developed velocity pro-
les in the direction along the length of the channels were set at
both inlets as follows [9]:
U
x
= 6

U
y
h

1
y
h

(14)
where x and y refer to the coordinate system seen in Fig. 1. Since
pressure is unimportant for the hydrodynamics of the weakly com-
pressible formulation, a gauge pressure of zero was specied at
both outlets. For the solute mass fraction and velocity at the outlets
we set m
A
/n=0 and U n=0. On all non-membrane walls the
no-slip velocity and zero gradient solute mass fraction boundary
conditions were applied; U=0 and m
A
/n=0, respectively. On the
membrane the boundary conditions described in Section 2.1 were
implemented for the velocity and solute mass fraction. The veloc-
ity on the draw side of the membrane was corrected for density
changes across the membrane as in [9].
The scope of this paper is limited to a theoretical and numerical
investigationof FOsystems andas suchall necessary model param-
eters were inspired by previous literature in the eld, rather than
being obtained from our own experiments. Empirical expressions
for the physical properties of a NaCl solution at 25

C were obtained
from[30]:
= 805.1 10
5
m
A
(15)
= 0.89 10
3
(1 +1.63m
A
) (16)
D
AB
= max(1.61 10
9
(1 14m
A
), 1.45 10
9
) (17)
= 997.1 +694m
A
(18)
where is the osmotic pressure (Pa), is the viscosity (Pa s), D
AB
is the diffusion coefcient of the solute (m
2
s
1
), and is the uid
density (kgm
3
). Using Eqs. (15) and (18) we estimate the propor-
tionality factor in Eq. (12) as:
= 805 10
2
Pa m
3
kg
1
(19)
The expressions (15)(19) are valid for NaCl mass fractions up to
0.09, which corresponds to a NaCl concentration of approximately
1.6M[30]. Inspired by FOexperiments performed by Yip et al. [27],
membrane characteristics were set as follows:
A = 1 10
12
m(s Pa)
1
(20a)
B = 1 10
7
ms
1
(20b)
K = 0.5s m
1
(20c)
For the slip-boundary condition on the porous support, the slip
coefcient andthe porosity must be set. Basedoninvestigations
in [31] the slip coefcient was set to 5. The porosity was var-
ied between the general values of 10
5
m
2
to 10
15
m
2
, i.e. values
ranging from highly pervious materials to more semi-permeable
materials [29]. It is noted that the value for and the values speci-
ed in Eqs. (20a)(20c) are merely inspired by the values found in
[31] and [27], i.e. it is not attempted to reproduce the experimental
results presented in [27].
2.3. Numerical procedure
The simulations were performed using the OpenFOAM library,
version 1.7. Anewsolver based on the weakly compressible formu-
lationdescribedinSection2.1 was implementedusing the supplied
solvers twoLiquidMixingFoam and rhoPisoFoam as templates. A
PISO algorithmwas used for treating the inter-equation pressure-
velocity coupling of the governing Eqs. (1) and (2) for mass and
momentumconservation[32]. Two iterations inthe PISOloopwere
found to be sufcient to achieve stable simulations for all cross-
ow velocities. In order to obtain temporal accuracy of the solute
mass fraction, Eq. (3) which describes the convection and diffusion
of solute, was solved within the PISO loop. Membrane boundary
conditions were explicitly implemented using eld values from
previous time steps. The water ux equation, Eq. (5), was solved
for J
w
at each point on the membrane using Ridders method for
root-nding [33].
In this work the Reynolds numbers for the majority of the
simulations were around 300. It has been argued that isotropic
turbulence models only are relevant to membrane chambers for
Reynolds numbers above 30,000 [10]. We therefore believe that
the laminar model developed here is capable of describing the gen-
eral owdynamics investigated in this work. The Reynolds number
is proportional to the cross-ow velocity and reaches a value of
3000 for a cross-owvelocity of 1ms
1
in the present system. One
simulation was carried out in this work with a cross-ow veloc-
ity of 1ms
1
. It is believed that the ow will enter a transitional
regime above a Reynolds number of about 2000, which may work
to decrease ECP at the membrane because of increased mixing [10].
Inclusionof turbulenceinthemodel is however beyondthescopeof
this work, and it is assumed that results obtained with the present
model at Reynolds numbers of 3000 are sufciently accurate.
3. Results and discussions
All simulations were run until steady-state solutions with con-
stant water and solute uxes were obtained. The exact time for
each simulation to reach steady-state was dependent on the ow
conditions; for the lowest cross-ow velocities it took about an
hour of simulation for the ow to reach steady-state and for the
highest cross-ow velocities the simulation had reached steady-
state after less than a minute of simulated ow. It was conrmed
that the model ensured overall mass balance and that the solute
ux balance equation was satised at both sides of the membrane
boundary during the simulations. Simulations were run on a per-
sonal computer with a quad core processor (Intel Q9450) and it
took approximately 1h for each simulation to reach steady-state.
Using the total amount of solute at the draw side of the mem-
brane as the integral functioninananalysis of the GridConvergence
Index (GCI) [10,34], it was found that the GCI for the coarse grid
compared to that of a ner grid with twice the amount of cells in
both directions was below 0.1%. This conrms that grid indepen-
dence was achieved.
3.1. Mass fraction proles vs. cross-ow velocity
The ultimate goal of implementing CFD models that describe
membrane ows is to have models in which various owparame-
ters can readily be optimized to promote higher water ux across
the membrane. A signicant parameter to be investigated in this
regard is the average cross-ow velocity in the membrane cham-
ber. It is generally known that higher cross-owvelocities work to
reduce the severity of ECP. To showthis effect, solute mass fraction
proles are presented in Fig. 2(a) for mean inlet velocities of 0.001,
0.01, 0.1, and 1ms
1
. All mass fraction proles in this paper are
recorded along the y axis at the center of the membrane, refer to
Fig. 1. At each time-step in the simulation, the total water uxes
across the membrane were recorded and the steady-state results
are presented in Fig. 2(b). As expected it is evident that increased
cross-ow velocities reduce ECP at the membrane, which in turn
provides for higher water permeation rates.
492 M.F. Gruber et al. / Journal of Membrane Science 379 (2011) 488495
In Fig. 2(a), signicant ECP is observed at the draw side of the
membrane for cross-ow velocities <1ms
1
. This is interesting
considering that most current analytical approaches to modeling
FO ux assume that there is no ECP at this side of the membrane;
i.e. that
d,m
in Eq. (5) can be replaced with the bulk osmotic
pressure in the draw [21,2325,27]. The general reasoning behind
this assumption is that the support layer is completely perme-
able to the draw solute, and therefore the concentration at the
membrane is the same as in the bulk [21,23]. This is likely to be
true in real applications where mass-transfer promoting spacers
are used and cross-ow velocities typically are on the order of
0.1ms
1
[10,27,35]. The extension of the analytical ICP theory to
a CFD model combined with the concept of solute ux conserva-
tion however allowus to investigate cases where ECP on the draw
side may be very signicant, e.g. when a lowcross-owvelocity is
used or when the water ux is high.
In Fig. 2, both the feed and draw cross-ow velocities are
adjusted simultaneously. Its seen that the cross-ow velocity in
the feed does not result in any signicant ECP on the feed side for
the velocities investigated here. Naturally, if the feed cross-ow
velocity is reduced so much that the inow of water in the feed
compartment becomes comparable to the membrane water ux,
concentrative ECPwill start tobecome important onthe feedside of
the membrane, which was conrmed to happen in the model (data
not shown). Such lowcross-owvelocities are however unlikely to
be used in any real membrane systems, and it is concluded that for
realistic cross-ow velocities and a solute-free feed solution, ECP
on the feed side is insignicant.
We note that although we manage to decrease the osmotic
pressure across the membrane by around 40% by changing the
cross-ow, this only results in a water ux decrease of around 20%,
seeFig. 2(b). This occurs partlybecausetheseverityof ICPis reduced
with decreased draw concentration at the membrane, refer to Eq.
(7), and partly because of the lower water ux which also works to
reduce the severity of ICP.
3.2. The effect of a slip boundary condition
An important assumption made in the simulations presented
in Fig. 2 is the no-slip velocity boundary condition on the support,
which is not valid for a rough surface, see Section 2.1. Depending
on membrane characteristics and shear rate there will be a non-
zero slip velocity present on the porous support, which may work
to reduce the ECP found on this side of the membrane signicantly
[4]. Implementing Eq. (8) in the CFD model, solute mass fraction
proles away fromthe membrane on the drawside were obtained
for different

/ values, see Fig. 3.


It is clear that as

/ and thereby the slip velocity on the
boundary is increased, ECP is reduced. Even for materials with a

/ value as high as 10
3
m, which corresponds to an extremely
pervious material [29], ECP is still observed at the porous support.
For

/ values below 10
5
m the slip velocity had a negligible
effect on ECP. It is noted that the slip velocity for

/ = 10
3
m
was foundtobe 0.086ms
1
, whichis very large considering a mean
cross-ow velocity of 0.1ms
1
. Based on these observations, we
conclude that albeit the additional convectionat the surface caused
by a slip velocity may work to reduce ECP at the porous support, it
does not seemto be able to completely negate ECP effects in realis-
tic systems, for which it is expected that

/ will be signicantly
less than 10
3
m, e.g. for typical high-ux MF membranes a value
on the order of 10
7
mis expected [19].
3.3. ICP modulus
As was noted in Section 2.1, it is customary to assume the
osmotic pressure at the membrane boundary in Eq. (5) to be equal
to the bulk osmotic pressure in the draw. Based on Fig. 2 we note
that ECP cannot be ignored on the draw side of the membrane for
the ow conditions and geometry investigated here. The conse-
quence of using an analytical model in which it is assumed that
ECP is absent to describe a systemwhere that is not strictly true, is
that the internal membrane parameter Kas used in Eq. (5) becomes
dependent on ow conditions outside the membrane. To see this,
we re-dene the actual resistivity to diffusion within the porous
support layer to be F instead of K. For illustrative purposes, we
assume that the membrane has complete (or very high) rejection
of solute, in which case we can utilize the ICP modulus dened in
Eq. (7) with F instead of K:

d,i
=
d,m
e
|J
w
|F
(21)
Noting that ECP is signicant on the draw side means that
it cannot be assumed that
d,m
=
d,b
. Based on lm theory, the
effect of ECP can be included in Eq. (21) by an ECP modulus given
Fig. 2. Results for different cross-ow velocities. (a) Solute mass fraction proles recorded along the y-axis at the center of the membrane (see Fig. 1), so that y =0.003
corresponds to the lower wall of the feed compartment and y =0.003 to the upper wall of the draw compartment. The membrane is indicated with a dashed vertical line at
y =0. (b) Water uxes through the membrane for experiments shown in (a).
M.F. Gruber et al. / Journal of Membrane Science 379 (2011) 488495 493
Fig. 3. Solute mass fraction proles away frommembrane at the drawside for dif-
ferent

/ values. The mean cross-owvelocity for all simulations was 0.1ms


1
.
The values seen in the legend represent

/.
as [24]:

d,m

d,b
= e
|J
w
|/k
d
(22)
where k
d
is a mass transfer coefcient. Combining Eqs. (7), (21) and
(22) we obtain:

d,i
=
d,b
e
|J
w
|(F+1/k
d
)
=
d,b
e
|J
w
|K
(23)
Fromthis we see that what is usually dened as the solute resis-
tivity Kin the porous support is actually a combination of the actual
solute resistivity in the porous layer F and a mass transfer coef-
cient k
d
:
K = F +
1
k
d
(24)
In accordance with the model developed by Lee et al., the solute
resistivity can be written as [21]:
K =
S
D
AB
(25)
where S is a structural parameter for the membrane whichdepends
onits thickness, tortuosity andporosity only. The structural param-
eter S is usually calculated using K in place of F in Eq. (25), and K is
in turn estimated by tting a water ux equation to experimental
results for thewater ux. SinceKdepends onthemass transfer coef-
cient it inturnalsodepends ontheowconditions, sotoobtainthe
Fig. 4. Using a geometric relation for k
d
as described in [36], this gure depicts an
analytical expression for K as a function of mean cross-owvelocity using Eq. (24).
Assuming negligible ECP, i.e.
f,m
=
f,b
and
d,m
=
d,b
, the mass-transfer coefcients
K as determined fromthe numerical results presented in Fig. 2 are also depicted in
the gure. The dotted line represents the solute resistivity to diffusion within the
porous support layer F.
Fig. 5. Simulations performed at different bulk draw/feed solute concentrations. (a)
Zero feed solute concentration with increasing drawsolute concentration. (b) Con-
stant draw solute mass fraction of 0.09 with increasing feed solute concentration.
(c) Water permeation uxes corresponding to the simulations shown in (a) and (b).
actual structural parameter S which should only depend on mem-
brane characteristics, one must consider the ow conditions. It is
recommended that ow conditions which minimize ECP are uti-
lized in order to ensure a large k
d
and thereby a value of K which
approaches F, thus making the calculated S value independent of
owconditions.
Based on correlations for the geometry used here, the mass
transfer coefcient k
d
can be calculated as a function of ow rate
using lm-theory models [36]. Using the relation for k
d
found in
[36] we depict the value of K as a function of cross-ow veloc-
ity using Eq. (24) in Fig. 4. From the gure it is clear that for K to
attain a value close to F the cross-owvelocity must be higher than
494 M.F. Gruber et al. / Journal of Membrane Science 379 (2011) 488495
0.1ms
1
. It is notedthat turbulence-promotingspacers, morecom-
plex case geometries and slip velocity at the boundary may work
to enhance mass transfer and thereby reduce K at a given mean
cross-ow velocity [4,10,35]. Naturally, the coefcient F could be
estimated from a given experiment simply by using Eq. (24) and
expressions for k
d
obtained with lm-theory or by some other
means. Such simple expressions for the mass transfer coefcients
k
d
are however only available for the simplest of geometries, and
even for the geometry investigated here some discrepancy is seen
between the analytical expression and the results obtained in the
CFD model, see Fig. 4. The advantage of using a CFD model is that
the ECP proles are fully resolved directly in the model.
3.4. Osmotic pressure dependence
To study howthe osmotic pressure difference across the mem-
brane inuences the solute mass fraction proles and water uxes,
a series of 9 simulations were performedwithdrawinlet m
A
-values
ranging from 0.01 to 0.08, keeping the feed inlet m
A
=0. Resulting
mass fraction proles are presented in Fig. 5(a). Another series of 8
simulations were run with the drawinlet m
A
kept at 0.09 and with
feed inlet m
A
-values ranging from 0.01 to 0.09. Solute mass frac-
tion proles are shown in Fig. 5(b). The water uxes recorded for all
simulations are presented in Fig. 5(c). A no-slip velocity boundary
conditionwas usedinall thesimulations, seediscussionSection3.2.
As the bulk draw concentration is increased in Fig. 5(c), a non-
linear increase in water ux was observed as expected from Eq.
(5). FromFig. 5(a), it is seen that increasing the drawconcentration
results inincreased ECP, whichis to be expected fromthe ECP mod-
ulus in Eq. (22). Upon addition of solute to the feed, the water ux is
decreased by the combined effect of increased ECP on the feed side
anddecreasedbulkdrivingforceacross themembrane. InFig. 5(b) it
is seen howwhen the feed concentration is increased initially, ECP
on the feed side becomes more signicant, however as the feed
concentration approaches the draw concentration, the decreased
water ux results in lower ECP on both sides of the membrane. The
lower water ux additionally decreases the severity of ICP and all
the effects taken together are what cause the hysteresis observed
for the water ux in Fig. 5(c), which corresponds to what has pre-
viously been observed experimentally for FO membranes [24].
4. Concluding remarks
We have developed a generalized computational model for
membrane systems and used it to investigate water ux and con-
centration polarization in FO systems with composite membranes.
The model was developed within an open source CFD framework
with a custom built weakly compressible transient solver and
explicit boundary conditions for the velocity and solute mass frac-
tion on the membrane. The code was validated by comparing the
results to previously published water and solute uxes, and veri-
ed by the existence of overall mass balance in the chamber and
ux balance at the membrane boundary.
The model was used to reveal the existence of external dilutive
concentration polarization on the drawside of an FOexperiment in
whichanasymmetric membrane was facedwithits porous support
against the draw side. The external concentration polarization on
the porous support is not normally considered in semi-analytical
approaches, since for realistic applications it can be assumed that
the presence of eddy-promoting spacers and highcross-owveloc-
ities will largely negate the external polarization. In cases where
such optimized conditions cannot be assumed, our results show
how CFD simulations can be used to reveal and quantify external
concentration polarization.
It was demonstrated that a tangential slip-velocity could be
included in the CFD model in order to parameterize and investi-
gate the effects of complex ow dynamics near the membrane.
The results show that a tangential ow along the membrane
result indecreased external polarizationand therefore enhance the
water ux, however not to such a degree that external polariza-
tion becomes negligible. Investigations of the inuence of osmotic
pressure difference across the membrane revealed additional
information about how different feed and draw concentrations
contribute differently to the effective osmotic pressure across the
active layer of the membrane.
The main advantage of the presented CFD model over tradi-
tional semi-analytical approaches is its capabilityof simultaneously
resolving the effects of parameters such as slip velocity, external
and internal concentration polarization, physical uid properties
and cross-owvelocity in a membrane chamber. The model is eas-
ily extended to three-dimensional geometries and canfurthermore
readily be usedtostudy the effect of mass-transfer promoting spac-
ers, chamber geometry and chamber size, albeit at a signicant
extra computational cost.
Acknowledgements
C. Hlix-Nielsen, L. Yde and C.Y. Tang were supported by the
Environment & Water Industry Programme Ofce of Singapore
(EWI) through a collaboration project #MEWR 651/06/169. L. Yde
andC.J. Johnsonwere also supportedby the DanishScience of Tech-
nology and Innovation.
Nomenclature Nomenclature
Symbol
A pure water permeability (m(s Pa)
1
)
slip coefcient
B solute permeation coefcient (ms
1
)
C solute concentration (kgm
3
)
D
AB
solute diffusion coefcient (m
2
s
1
)
g gravitational acceleration (ms
2
)
h height of chamber (m)
J
s
solute ux (kg(m
2
s)
1
)
J
w
water permeation ux (ms
1
)
K diffusion resistivity (s m
1
)
k
d
mass transfer coefcient (ms
1
)
permeability (m
2
)
m
A
solute mass fraction (kgkg
1
)
n surface normal vector
n surface normal direction (m)
viscosity of uid (Pa s)
p pressure (Pa)
osmotic pressure (Pa)
R solute separation coefcient
Re Reynolds number
uid denisty (kgm
3
)
S structural parameter (m)
U uid velocity vector (ms
1
)

U mean cross-owvelocity (ms


1
)
U
slip
surface slip velocity (ms
1
)
x membrane tangential axis (m)
y membrane normal axis (m)
Index
d draw-side of membrane
f feed-side of membrane
i interface between active layer and support
m at the membrane surface
M.F. Gruber et al. / Journal of Membrane Science 379 (2011) 488495 495
References
[1] A.L. Zydney, Stagnant lmmodel for concentration polarization in membrane
systems, J. Membr. Sci. 130 (12) (1997) 275281.
[2] S. Kim, E. Hoek, Modeling concentration polarization in reverse osmosis pro-
cesses, Desalination 186 (13) (2005) 111128.
[3] P. Brian, Concentration polarization in reverse osmosis desalination with vari-
able ux and incomplete salt rejection, Ind. Eng. Chem. Fundam. 4 (4) (1965),
439&.
[4] H. Yeh, T. Cheng, Analysis of the slip effect on the permeate ux in membrane
ultraltration, J. Membr. Sci. 154 (1) (1999) 4151.
[5] S. Chellam, M.R. Wiesner, C. Dawson, Slip at a uniformly porous boundary:
effect on uid owand mass transfer, J. Eng. Math. 26 (4) (1992) 481492.
[6] K. Youm, A. Fane, D. Wiley, Effects of natural convection instability on mem-
brane performance in dead-end and cross-ow ultraltration, J. Membr. Sci.
116 (2) (1996) 229241.
[7] D.E. Wiley, D.F. Fletcher, Computational uid dynamics modelling of ow and
permeation for pressure-driven membrane processes, Desalination 145 (13)
(2002) 183186.
[8] D.E. Wiley, D.F. Fletcher, Techniques for computational uid dynamics mod-
elling of owin membrane channels, J. Membr. Sci. 211 (1) (2003) 127137.
[9] S. Wardeh, H. Morvan, CFD simulations of owand concentration polarization
in spacer-lled channels for application to water desalination, Chem. Eng. Res.
Des. 86 (10A) (2008) 11071116.
[10] G. Fimbres-Weihs, D. Wiley, Review of 3D CFD modeling of ow and mass
transfer in narrow spacer-lled channels in membrane modules, Chem. Eng.
Process 49 (7) (2010) 759781.
[11] R.L. McGinnis, M. Elimelech, Energy requirements of ammonia-carbon dioxide
forward osmosis desalination, Desalination 207 (13) (2007) 370382.
[12] T.Y. Cath, A.E. Childress, M. Elimelech, Forward osmosis: principles, applica-
tions, and recent developments, J. Membr. Sci. 281 (12) (2006) 7087.
[13] A. Achilli, T.Y. Cath, E.A. Marchand, A.E. Childress, The forward osmosis mem-
brane bioreactor: a lowfouling alternative to MBR processes, Desalination 239
(13) (2009) 1021.
[14] R.W. Holloway, A.E. Childress, K.E. Dennett, T.Y. Cath, Forward osmosis for con-
centrationof anaerobic digester centrate, Water Res. 41(17) (2007) 40054014.
[15] C.Y. Tang, Q. She, W.C.L. Lay, R. Wang, A.G. Fane, Coupled effects of internal
concentration polarization and fouling on ux behavior of forward osmo-
sis membranes during humic acid ltration, J. Membr. Sci. 354 (12) (2010)
123133.
[16] B. Mi, M. Elimelech, Organic fouling of forward osmosis membranes: fouling
reversibility and cleaning without chemical reagents, J. Membr. Sci. 348 (12)
(2010) 337345.
[17] S. Lee, C. Boo, M. Elimelech, S. Hong, Comparison of fouling behavior in for-
ward osmosis (FO) and reverse osmosis (RO), J. Membr. Sci. 365 (12) (2010)
3439.
[18] C.H. Nielsen, Biomimetic membranes for sensor and separation applications,
Anal. Bioanal. Chem. 395 (3) (2009) 697718.
[19] R.W. Baker, Membrane Technology and Applications, 2nd ed., John Wiley and
Sons, 2004.
[20] M. Goosen, S. Sablani, H. Ai-Hinai, S. Ai-Obeidani, R. Al-Belushi, D. Jackson,
Fouling of reverse osmosis and ultraltration membranes: a critical review,
Sep. Sci. Technol. 39 (10) (2004) 22612297.
[21] K. Lee, R. Baker, H. Lonsdale, Membranes for power generation by pressure-
retarded osmosis, J. Membr. Sci. 8 (2) (1981) 141171.
[22] S. Loeb, L. Titelman, E. Korngold, J. Freiman, Effect of porous support fabric on
osmosis through a Loeb-Sourirajan type asymmetric membrane, J. Membr. Sci.
129 (2) (1997) 243249.
[23] J.R. McCutcheon, M. Elimelech, Modeling water ux in forward osmosis: impli-
cations for improved membrane design, Am. Inst. Chem. Eng. 53 (7) (2007)
17361744.
[24] J.R. McCutcheon, M. Elimelech, Inuence of concentrative and dilutive internal
concentration polarization on ux behavior in forward osmosis, J. Membr. Sci.
284 (12) (2006) 237247.
[25] D. Xiao, C.Y. Tang, J. Zhang, W.C.L. Lay, R. Wang, A.G. Fane, Modeling salt accu-
mulation in osmotic membrane bioreactors: implications for FO membrane
selection and systemoperation, J. Membr. Sci. 366 (12) (2011) 314324.
[26] D. Fletcher, D. Wiley, A computational uids dynamics study of buoyancy
effects in reverse osmosis, J. Membr. Sci. 245 (12) (2004) 175181.
[27] N.Y. Yip, A. Tiraferri, W.A. Phillip, J.D. Schiffman, M. Elimelech, High perfor-
mance thin-lmcomposite forward osmosis membrane, Environ. Sci. Technol.
44 (10) (2010) 38123818.
[28] E.V. Mosina, Numerical study of ow at a liquid-porous medium interface,
Theor. Found. Chem. Eng. 44 (5) (2010) 679685.
[29] J. Bear, Dynamics of Fluids in Porous Media, Dover publications, 1972.
[30] V. Geraldes, V. Semiao, M.N. de Pinho, Flow and mass transfer modelling of
nanoltration, J. Membr. Sci. 191 (12) (2001) 109128.
[31] H.X. Bai, P. Yu, S.H. Winoto, H.T. Low, Boundary conditions at the interface
betweenuidlayer andbrous medium, Int. J. Numer. Meth. Fluid60(7) (2009)
809825.
[32] R. Issa, Solution of the implicitly discretized uid-owequations by operator-
splitting, J. Comput. Phys. 62 (1) (1986) 4065.
[33] W.H. Press, S. Teukolsky, W. Vetterling, B. Flannery, Numerical Recipes, the Art
of Scientic Computing, 3rd ed., Cambridge University Press, 2007.
[34] P. Roache, Quanticationof uncertaintyincomputational uiddynamics, Annu.
Rev. Fluid Mech. 29 (1997) 123160.
[35] Y. Wang, F. Wicaksana, C.Y. Tang, A.G. Fane, Direct microscopic observation
of forward osmosis membrane fouling, Environ. Sci. Technol. 44 (18) (2010)
71027109.
[36] E. Hoek, A. Kim, M. Elimelech, Inuence of crossow membrane lter geom-
etry and shear rate on colloidal fouling in reverse osmosis and nanoltration
separations, Environ. Eng. Sci. 19 (6) (2002) 357372.

Vous aimerez peut-être aussi