Desalination 157 (2003) 105–111

Predicting flux and rejection of multicomponent salts mixture in

nanofiltration membranes

Abdul Wahab Mohammad*, Mohd Sobri Takriff

Department of Chemical and Process Engineering, Universiti Kebangsaan Malaysia,
43600 UKM Bangi, Selangor, Malaysia
Tel. +60 (3) 89296410; Fax +60 (3) 89296148; email: wahabm@eng.ukm.my

Received 7 February 2003; accepted 12 February 2003

Abstract

This paper presents a modified Donnan-Steric-Pore model (DSPM) to predict the rejection of mixture of
NaCl:Na2SO4 in nanofiltration (NF) membrane based on the extended Nernst-Planck equation with the incorporation
of charge and steric effects for the transport of ions inside the membrane and incorporation of concentration polarization
effect for mixture of ions/solutes. With this approach, the permeate flux can be calculated based on the concentration
of ions at the membrane surface. The membrane performance was modelled using three parameters namely; effective
pore radius, rp, effective ratio of membrane thickness to porosity, ∆x/Ak, and the effective charge density, Xd. The
simulation results shows that the model can predict the tendencies and patterns of rejection and flux reduction
behaviour reasonably well for system containing NaCl:Na2SO4. The model can be used as a preliminary tool to
assess the rejection capability as well as the flux behaviour of NF membranes in mixtures of solutions. For future
works, the model can be improved further to account for non-ideal as well as real mixture solutions.

Keywords: Nanofiltration; Nernst-Planck equation; Multicomponent; Assessment tools

1. Introduction (RO) membranes. Their separation mechanisms

involve both steric (sieving) effects and electrical
Nanofiltration (NF) membranes have found
(Donnan) effects. This combination allows NF
various applications in many industries. NF mem-
membranes to be effective for a range of separa-
branes have properties in between those of ultra-
tions of mixtures of small organic solutes (either
filtration (UF) membranes, and reverse osmosis
neutral or charged) and salts. In desalination plant
for example, NF has been used as a pre-treatment
*Corresponding author.
Presented at the European Conference on Desalination and the Environment: Fresh Water for All, Malta, 4–8 May 2003.
European Desalination Society, International Water Association.

0011-9164/03/$– See front matter © 2003 Elsevier Science B.V. All rights reserved
106 A.W. Mohammad, M.S. Takriff / Desalination 157 (2003) 105–111

step for RO/thermal processes where divalent ions Na2SO4. In such system, the transport of all ions
are removed [1]. will be considered inside the membrane since all
The ability to model the NF membrane pro- ions will be partially rejected. The model will first
cesses will be very useful especially for process be described and subsequently some simulation
prediction, optimization and economic assess- results will be presented for different cases such
ment. Various models have been proposed in as (i) variation of NaCl:Na2SO4 ratios, (ii) varia-
modelling the rejection performance of salts and tion of pore radius and (iii) variation of the effec-
charged organics in NF membranes[2–5]. Most tive charge density.
of these models were based on the extended
Nernst-Planck equation to describe the transport
inside the membrane. One of the most commonly
2. Model development
cited model is the so-called Donnan-Steric-Pore
model (DSPM) [6]. Most reported works however Fig. 1 depicts the system within which the
have used the DSPM model to predict the rejec- transport takes place. The bulk components will
tion based on specified flux. In actual system, the consist of Na+, Cl– and SO42– ions. For ions/solutes
flux will depend on the osmotic effect caused by to permeate across the membrane, each ion/solute
the salt concentrations at the membrane wall. will move from the bulk solution through the film
Therefore, the ability to predict rejection and flux layer, membrane interface and inside the mem-
simultaneously will further improve the prediction brane. The proposed model takes into account the
capability. transport across all the mentioned mediums.
In previous work [7,8], a modified Donnan- Concentration polarisation close to the mem-
Steric-Pore model (DSPM) has been proposed to brane surface is assumed to occur within a boun-
predict the rejection of mixture of charged ions dary film layer of thickness, δ, which is dependent
(NaCl and dye) in NF membrane based on the on the mass transfer characteristics of the system.
extended Nernst-Planck equation with the incor- For a system containing charged ions, a mass
poration of charge and steric effects for the trans- balance for the film layer yields,
port of ions inside the membrane and incorpo-
ration of concentration polarization effect for dC i z i F dψ f
ji = − Di ,∞ − Ci Di ,∞ + Ci J v (1)
mixture of charged ions. With this approach, the dx RT dx
permeate flux can be calculated based on the con-
centration of ions solutes at the membrane surface. Di,∞ is the bulk diffusivity of ion i in the solution.
The membrane performance was modelled using The equation can be solved using the boundary
three parameters namely; effective pore radius, conditions at x = –δ, c = Cbi and at x = 0, c = Cwi.
rp, effective ratio of membrane thickness to poro-
sity, ∆x/Ak, and the effective charge density, Xd. film membrane
Comparison of the model calculation with own Cw1
Cb1
Na+
and published experimental data shows that the Cw3
model can predict the tendencies and patterns of Cb3
SO42-
Na+
rejection and flux reduction behaviour reasonably Cb2 SO42-
well for system containing NaCl-dye. In such Cl- Cl-
Cw2
NaCl-dye system, the dye rejection was almost
100% and thus only the transport of NaCl was
δ ∆x
considered inside the membrane.
In this paper, the applicability of the model Fig. 1. Schematic of the film layer and membrane for three-
will be extended to mixture of salts such as NaCl- component system.
 A.W. Mohammad, M.S. Takriff / Desalination 157 (2003) 105–111 107

Similarly the transport of ions across the mem- ∆π = RT (∑ C wi − ∑ C pi ) (12)

brane will be governed by similar equation based
on the extended Nernst-Planck equation. It is acknowledged however that the Van’t Hoff
model is only valid at dilute concentrations and
dci zi ci Di , p dψ thus only provide a rough estimation of the osmo-
ji = − Di , p − F + K i ,c ciV (2a)
dx RT dx tic pressure effect. At higher concentrations, the
Van’t Hoff model will over estimate the osmotic
where ji is the flux of ion i and the terms on the
pressure and thus a more accurate model such as
right hand side represent transport due to diffu-
the Pitzer model should have been used [9]. This
sion, electric field gradient and convection respec-
will be considered in our future work.
tively (see List of symbols for definition of sym-
In this model, the transport of the ions will be
bols). The hindered nature of diffusion and con-
described in term of only three parameters charac-
vection of the ions inside the membrane are ac-
teristic of the membrane namely, rp, ∆x/Ak and
counted for by the terms Ki,d and Ki,c. Further
Xd. By providing these parameters and the opera-
details for solution of this equation for transport
ting conditions of the system (concentration, pres-
through NF membranes can be found elsewhere
sure, temperature, interface boundary film thick-
[7,8]. Table 1 shows other supporting equations
ness), the model will be able to calculate the rejec-
used in this model.
tion of each ion. The flux will also be calculated
The permeate flux will be calculated as follows:
based on the osmotic pressure at the membrane
J v = Pm (∆P − ∆π ) (11) wall [Eq. (11)]. Fig. 2 shows the flowchart for the
solution of the model. Once the input data are
where the osmotic pressure difference, ∆π, is provided, the osmotic pressure, ∆π, can be cal-
calculated based on the Van’t Hoff law [8]: culated based on the bulk concentrations as well
Table 1
Equations for the DSPM model

Main equations:

Concentration gradient: Potential gradient:

dc i J
= v K c −C
dx Di , p i
(
, c i i , p
)zc
− i i F
RT
dψ m
dx
(2b) ∑
n zi J v
Di , p
(K c −C
i, c i i, p
)
dψ m i =1
= (3)
dx
RT i =1
F n
( )
∑ z i ci
2

Donnan-steric partitioning:
where:
 γc   zi F 
 ii   
Hindrance factors:
 o  = Φ exp − RT ∆ψ D  (4) Ki,d=K-1(λ,0) Ki,c = (2 –Φ)G(λ,0) (5)
γ C   
 i i Di , p = K i , d Di , ∞ (6)

Hydrodynamic drag coefficients: Steric partitioning:

K-1(λ,0) = 1.0 – 2.30λ + 1.154λ2 + 0.224λ3 (7) Φ = (1 – λ)2 (9)
G (λ,0) = 1.0 + 0.054λ – 0.988λ2 + 0.441λ3 (8) Electroneutrality conditions:
n n
∑ zi Ci = 0 ∑ z i ci = − X d (10)
i =1 i =1
108 A.W. Mohammad, M.S. Takriff / Desalination 157 (2003) 105–111
Fig. 2. Flowchart for DSPM model for mixture system.
as a guessed permeate concentration. Then the
permeate flux, Jv, can be determined based on the
pressure driving force. Knowing Jv, Eq. (1) can
be solved to obtain the wall concentrations for each
solute, Cwi. This will be done iteratively until the
calculated ∆π is the same as the previously
guessed ∆π. Eqs. (2)–(10) will then be solved for
the transport inside the membrane to obtain the
permeate concentration, Cpi, for each solute. Again
iterative calculations are required until the calcu-
lated Cpi is the same as initially guessed Cpi. Once
Cpi is determined, the real rejection, R, can be
calculated. The transport of Na+, Cl– and SO42–
will take place in both the film layer as well as
inside the membrane.
3. Results and discussion
As an illustrative purpose, simulation results
will be presented for three different cases namely
(i) variation of NaCl:Na2SO4 ratios, (ii) variation
of rp, and (iii) variation of Xd. Unless mentioned
otherwise, the membrane characteristics used are
rp = 1 nm, Xd = 200 mM and ∆x/Ak = 1 µm. The
boundary layer thickness, δ is taken to be 8.24 µm
based on a typical mass transfer coefficient for
cross-flow membrane system [8].
 ∆
A.W. Mohammad, M.S. Takriff / Desalination 157 (2003) 105–111 109

600 80

500
I
I 60
II
III II
Jv x 10-6 (ms-1)

400 40

Rejection (%)
III
300 20
Cl-
2-
SO4
200
Pure Water 0 III
100 permeate 0 100 200 300 400 500
-20 II
I
0
-40
0 2 4 6 8 10 12
Jvx10-6(ms-1)
(a) ∆ P (bar) (b)

Fig. 3. Effects of different ratios. (a) Plots of permeate fluxes vs. applied pressure and (b) plots of rejection against fluxes.
I: 10:40 ratio, II: 25:25 ratio and III: 40:10 ratio.

3.1. Effect of NaCl:Na2SO4 ratios
Fig. 3 shows the simulation results for variation
of NaCl:Na2SO4 ratios. The total concentration
is 50 mol/m3 and the ratios were set at 40:10,
25:25 and 10:40. Fig. 3a shows the plots of
permeate flux as a function of applied pressure
while Fig. 3b shows the plots of rejection as a
function of permeate flux. It can be seen that the
permeate flux was slightly reduced compared to
the pure water flux for all the three ratios. The
rejection plots show that the Cl– rejection is
negative but become less negative as the permeate
flux increases. Similar pattern of rejection has
been confirmed experimentally by several re-
searchers [2,7,10]. As the Cl– content is decreased,
the rejection of Cl– tends towards more negative
rejection indicating that more Cl– ions are being
washed out due to the higher SO42– content in the
feed. Again, similar behaviour has been shown in
NaCl:dye system [2].
3.2. Effect of varying rp
Fig. 4 shows the results for variation of effec-
tive pore radius, rp. In this case rp was set at 0.5 nm,
0.75 nm and 1 nm which are representative of NF
membrane pore radius. The total concentration is
still 50 mol/m3 with NaCl:Na2SO4 ratio was set
at 25:25. By reducing rp, the permeability of the
membrane is reduced as well as predicted by the
Hagen-Poiseuille equation ∆ [8]:

600 100
I
Pure water 80
500 III
permeate II
60
III
Jv x 10-6 (ms-1)

400
Rejection (%)

40
300 II 20
200 0
I 0 100 200 300 400 500
100 -20 III
II
-40 I
0 SO42- Cl-
0 2 4 6 8 10 12 -60
(a) ∆ P (bar) (b) Jvx10-6(ms-1)

Fig. 4. Effects of rp. (a) Plots of permeate fluxes vs. applied pressure and (b) plots of rejection against fluxes. I: rp = 0.5 nm,
II: rp = 0.75 nm and III: rp = 1 nm.
110 A.W. Mohammad, M.S. Takriff / Desalination 157 (2003) 105–111

rp2 3.4. Future work and improvement

Pm =
8µ(∆x / Ak )
As shown in Fig. 4a, the pure water flux
decreased as rp was decreased. The reduction in
permeate flux is greater for rp = 1 since in this
case the higher convective flux caused a signi-
ficant concentration polarization effect at the
membrane surface and thus higher osmotic pres-
sure. The rejection for rp = 0.5 is more negative
as shown in Fig. 4b. This is expected since the
hindrance due to the steric effect will be larger.
3.3. Effect of varying Xd
Fig. 5 shows the result for variation of effective
charge density, Xd. In this case Xd was set at 100,
200 and 400 mol/m3 while rp is set at 1 nm. The
total concentration is still 50 mol/m 3 with
NaCl:Na2SO4 ratio was set at 25:25. Since rp is
the same, the pure water flux for all the three cases
is the same. As shown in Fig. 5a, the reduction in
permeate flux increased as Xd increased. At higher
Xd, more rejection of the ions occurred as shown
in Fig. 5b and as a result more ions were accu-
mulated at the membrane walls. This leads to
higher osmotic effect and subsequently lower
permeate flux.
(13)
Based on the presented results, the model is
able to predict the pattern and tendencies of rejec-
tion for mixture of ions quite well. Similar
rejection and flux reduction behaviours have been
shown in several experimental studies previously.
The applicability of the model should be extended
to other mixture of salts including divalent cations
such as Ca2+ and Mg2+ which are present in many
brackish water systems. The effect of concen-
tration and non-ideality should also be incorpo-
rated in the model in order to make it more
relevant to real system. If this is successful, the
model can be used as a preliminary tool to assess
the rejection capability as well as the flux beha-
viour of NF membranes in mixtures of solution.
4. Conclusion
The DSPM model based on the extended
Nernst-Planck equation used in this work is very
useful in predicting and explaining the pattern of
rejection in mixture of salts in NF membranes.
The model was described in terms of only three
membrane parameters and a mass transfer film
layer thickness. Furthermore by incorporating the
concentration polarization
∆ effect at the membrane

600 100

500 Pure water 80

I III
permeate II 60
Jv x 10-6 (ms-1)

400
Rejection (%)

III 40 II
300
20 I
200 0
I
100 -20 0 II 100 200 300 400 500

-40 III SO42- Cl-

0
-60
0 2 4 6 8 10 12
Jvx10-6(ms-1)
(a) ∆ P (bar)
(b)

Fig. 5. Effects of Xd. (a) Plots of permeate fluxes vs applied pressure and (b) plots of rejection against fluxes. I: Xd =
100 mol/m3, II: Xd = 200 mol/m3 and III: Xd = 400 mol/m3.
 A.W. Mohammad, M.S. Takriff / Desalination 157 (2003) 105–111 111

surface, the permeate flux was also calculated. Ki,d — Hindrance factor for diffusion
Based on the simulation results, the model can Pm — Pure water permeability, m3m–2s–1bar–1
predict the tendencies and patterns of rejection rp — Effective pore radius, m
and flux reduction behaviour reasonably well for R — Gas constant, J mol–1 K–1
system containing NaCl:Na2SO4. The model can T — Absolute temperature, K
be used as a preliminary tool to assess the rejection x — Distance normal to membrane, m
capability as well as the flux behaviour of NF ∆x — Effective membrane thickness, m
membranes in mixtures of solutions. For future Xd — Effective membrane charge, mol m–3
works, the model can be improved further to zi — Valence of ion
account for non-ideal as well as real mixture δ — Thickness of film layer, m
solutions. ∆ψ — Potential difference, V

Acknowledgement References
[1] M.A. Al-Sofi, Desalination, 135 (2001) 121–139.
The authors would like to thank the Ministry
[2] R. Levenstein, D. Hasson and R. Semiat, J. Membr.
of Science and Technology, Malaysia for financial Sci., 116 (1996) 77–92.
support through IRPA Grant 09-02-02-0070- [3] M. Perry and C. Linder, Desalination, 71 (1989) 233–
EA197. 245.
[4] T. Tsuru, S. Nakao and S. Kimura., J. Chem. Eng.
Japan, 24 (1991) 511–517.
Symbols [5] J. Schaep, C. Vandecasteele, A.W. Mohammad and
W.R. Bowen, Separ. Sci. Technol., 34(15) (1999)
ci — Concentration in membrane, mol m–3 3009–3030.
Ci,b — Concentration in the bulk solution, [6] W.R. Bowen, A.W. Mohammad and N. Hilal, J.
mol m–3 Membr. Sci., 126 (1997) 91–105.
Ci,p — Concentration in permeate, mol m–3 [7] W.R. Bowen and A.W. Mohammad., AIChE J., 44
Di,p — Hindered diffusivity, m2 s–1 (1998) 1799–1812.
[8] A.W. Mohammad, Separ. Sci. Technol., 37(5) (2002)
Di,∞ — Bulk diffusivity, m2 s–1
1009–1030.
F — Faraday constant, C mol–1 [9] D. Van Gauwbergen, J. Baeyens and C. Creemers,
ji — Ion flux (based on membrane area), Desalination, 109 (1997) 57–65.
mol m–2 s–1 [10] W.R. Bowen and H. Mukhtar, J. Membr. Sci., 112
Ki,c — Hindrance factor for convection (1996) 263–274.