See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/51990927

Characterization of nanofiltration membranes with uncharged solutes

Article · January 2008

CITATIONS READS

26 657

2 authors:

Zoltán Kovács Wolfgang Samhaber

Technische Hochschule Mittelhessen Johannes Kepler University Linz
80 PUBLICATIONS   406 CITATIONS    17 PUBLICATIONS   213 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

JEVA-Pomelo View project

Optimal Operation of Membrane Processes under Fouling View project

All content following this page was uploaded by Wolfgang Samhaber on 26 May 2014.

The user has requested enhancement of the downloaded file.

Characterization of nanofiltration membranes
with uncharged solutes

Z. Kovács ∗, W. Samhaber
Institut für Verfahrenstechnik, Johannes Kepler Universität Linz,
Welser Straße 42, A-4060 Leonding, Austria.

Abstract

The transportation of non-charged solutes through nanofiltration (NF) membranes

is usually characterized by the term of molecular weight cut-off (MWCO) that is
in the range of about 200-1500 Dalton. However, there are currently no standard
methods for characterizing and reporting MWCO. The meaning of this information
can vary between different membrane manufactures, and the applied experimental
technique has a significant impact on the measured MWCO value. In this study, the
separation performances of five commercial NF and tight ultrafiltration membranes
(DK, GE, GH, NP030, NP010) were investigated using experimental permeation
data of six noncharged solutes with various molecular weights. The membranes
were characterized with their hypothetical pore sizes and their MWCO values were
determined using two types of steric pore flow models, the steric-hindrance-pore
(SHP) and the Donnan-steric-pore (DSP) model. Thermodynamical analysis of the
experimental data was performed in order to obtain the reflection coefficients of
the solutes, and the obtained phenomenological parameters were then linked to
the membrane structural parameters. The overall best-fitting pore size radii in a
least-squares sense were computed for each membrane by fitting to the sets of ex-
perimental data of reflection coefficients and solute radii. The predictions of both
DSP and SHP models were found to be in a good agreement with the experimental
data. However, a significant difference was found in the estimated pore size values
depending on which measure of the solute size was employed and depending on the
applied models. The provided values of membrane pore size can be used to predict
the rejection of any noncharged solute for a given applied pressure.

Key words: nanofiltration, steric hidrance pore model, Spiegler-Kedem,

uncharged, molecular weight cut-off

∗ Corresponding author.
Email address: zoltan.kovacs@jku.at (Z. Kovács).

Preprint submitted to Elsevier 17 May 2008

1 Introduction

Nanofiltration (NF) membranes are a relatively recent development in the field

of pressure-driven membrane separations, and their properties lie between ul-
trafiltration (UF) and reverse osmosis (RO). Due to its potential advantages,
NF has gained a strong market position, brought about a number of patents,
industrial research projects and commercial installations [1]. Starting in the
late sixties, NF membrane processes have gradually found their way into in-
dustrial applications in various fields such as water softening, dye recovery,
treatment of metal contained waste waters, oil-water separation, demineral-
ization of whey, recycle of nutrients in fermentation processes, purification of
landfill leachate, removal of sulfates from sea-water, bioproduct separation [2].

NF systems are usually operated at medium pressures in the range of 10-50

bar, and have much higher water fluxes compared to RO membranes. NF can
be applied for separation between ions with different valences and for separa-
tion of low- and high-molecular weight components. Polymeric NF membranes
show diversity in separation behaviour but they are common in rejecting highly
charged ions (such as SO42− , CO32− , P O43− , Mg2+ ) in a higher degree, while in
comparison, rejection of single charge ions (Cl− , Na+ , K+ ) is much less. NF
also rejects uncharged, dissolved material and positively charged ions accord-
ing to the size and shape of the molecule in question [3].

The transportation of non-charged solutes through an NF membrane is usu-

ally characterized by the term of molecular weight cut-off (MWCO), which is
a number expressed in Dalton indicating the molecular weight of a hypotheti-
cal non-charged solute that is in 90% rejected. The MWCO of NF membranes
is usually given by the manufacturers and typically in the range of about
200-1500 Dalton. However, there are currently no standard methods for char-
acterizing and reporting MWCO. The meaning of this information can vary
between different membrane manufactures, thus limiting its value. The applied
experimental technique has a significant impact on the measured MWCO.
The experimental conditions, such as the hydrodynamic flow conditions, the
temperature, the applied pressure or the range of the applied pressure-scan,
the type of solute(s), are often not reported. Besides, the different techniques
for MWCO make membranes from different manufacturers hardly comparable
without further experimental investigations. Moreover, the concept of MWCO
does not address the question of how great the permeation of solutes smaller
and larger than the indicated MWCO can be.

Membranes have been historically characterized by MWCO rather than by

membrane pore size [4]. It should be noted that this concept is based on prac-
tical aspects and has no true physical meaning. The molecular weight is not a
straightforward meausre of the size and it ignores the shape of the permeating

2
molecule, and thus, it gives only a rough estimation of the membrane’s ability
to remove dissolved uncharged components.

The term nanofiltration was introduced by the membrane manufacturer FilmTec

highlighting that such membranes are capable to retain uncharged molecules
with the size of about 1 nm, and which corresponded well with the early
predictions of Sourirajan and Matsuura about the membranes’ hypothetical
capillary pore size [5]. However, it is still the subject of scientific debates
whether NF membranes can be considered as porous or nonporous media.

Several direct characterization methods are known for NF membranes such as

permporometry, gas adsorption-desorption and microscopy techniques. How-
ever, the pore size determination of polymeric membranes seems to be still an
unsolved problem. Permporometry analysis requires pores larger than 2 nm
[6], and the nitrogen adsorption-desorption method only for inorganic mem-
branes can be effectively applied. As far as microscopy is concerned, Bowen
et al. [7] analysed the structural parameters of NF membranes with atomic
force microscopy (AFM) and claimed that AFM images provide direct con-
firmation of the presence of discrete pores of nanometre dimensions in such
membranes. However, the authors reported that AFM can only give informa-
tion of the surface pore dimensions as the tip cannot probe into the depth
of the pore. Singh et al. [8] also pointed out that an important requirement
for AFM pore size determination is a very low surface roughness, hence it is
difficult to distinguish between the pores and the depressions in the membrane
surface.

Models that adequately describe the separation behaviour of membranes are

important since these are needed in the design of NF processes. NF is a com-
plex process lying between UF and RO, thus models from both these areas
and their combinations can be used to describe NF performance. These models
rely on fundamentally different concepts and can be divided into the following
types: irreversible thermodynamics models, nonporous or homogenous models,
and pore models.

The properties of dry membranes are subject to changes in the presence of

water. Thus, during the filtration, a swollen polymeric network is obtained
which likely does not contain discrete pores. [5] In contrary to polymers, the
morphology of inorganic membranes does not change by contacting with wa-
ter. Well established characterization methods prove that inorganic NF mem-
branes contain discrete pores in the order of 0.5 to 2 nm. Despite this mor-
phological difference, both type of membranes show very similar separation
behaviours, and their performance can be predicted with the same models as-
suming hypothetical pores in the near- or sub-nanometre range. However, in
the pore models, the terms ’pore’ or ’capillary’ refer to any connecting polymer
material-free void space through which fluid transport can take place under a

3
driving force [9].

In this paper we investigate the separation behaviour of NF membranes using

experimental permeation data of six noncharged solutes. We use an indirect
characterization method employing two types of steric pore flow models which
are described in Sect.2.2, and we discuss possible concepts for solute size de-
termination in Sect.2.3. We characterize five NF and tight UF membranes
with their hypothetical pore sizes and determine their MWCO values with
the different models.

2 Theory

2.1 Spiegler–Kedem model

The Spiegler–Kedem model (SK) [10] is based on the principles of irreversible

thermodynamics and has found a wide use for the description of RO and NF
membrane separations. This phenomenological approach treats the membrane
as a black box, and no insight is given into the membrane structure and
morphology. The relation between the rejection R and the solvent volume
flux Jv is given by the expression:

σ−1
  
σ 1 − exp Jv
Ps
R=  , (1)
σ−1

1 − σ exp Jv
Ps

where the two membrane parameters are the reflection coefficient σ and the
solute permeability coefficient Ps .

The rejection is defined as

R = 1 − cp /cr , (2)

where cp and cr represent, respectively, the solute concentrations in the per-

meate and retentate streams.

4
2.2 Models of hindered transport

The modelling of solute transport through the membrane is based on the ex-
tended Nernst–Planck equation accounting for diffusion, convection and elec-
tromigration. For uncharged species, when electrostatic effects are negligible,
the transport equation for the molar solute flux js is given as the sum of the
diffusive and the convective terms as follows

dc
js = Kc cV − Dp , (3)
dx

where V is the solvent velocity, Kc is the convective hindrance factor and Dp

is the hindered solute diffusivity in the membrane pores. This latter is given
as the product of the bulk diffusivity D∞ and the diffusive hindrance factor
Kd , and corrected with the factor η0 /ηp which accounts for the solvent pore
viscosity ηp versus the bulk viscosity η0 . There it holds:

η0
Dp = K d D∞ . (4)
ηp

The Donnan-steric-pore-model (DSP) was originally developed by Bowen et

al. [7]. They have introduced the following relationships between the ratio of
solute to pore radius λ and the hindrance factors Kc and Kd :

 
Kc = (2 − Φ) 1 + 0.054λ − 0.988λ2 + 0.441λ3 , (5)
2 3
Kd = 1 − 2.30λ + 1.154λ + 0.224λ , (6)

where the steric partitioning coefficient Φ (neglecting concentration polar-

isaton) is given as

Φ = (1 − λ)2 . (7)

The model assumes a fully developed solute velocity inside the pores and as-
sumes a Hagen-Poiseuille type parabolic profile. The solvent velocity in cylin-
drical pores with constant circular cross-sections can be obtained from the
Hagen-Poiseuille law:

rp2 (∆P − ∆π)

V = (8)
8ηp (∆x/Ak )

5
where x/Ak is the ratio of effective membrane thickness to membrane porosity,
ηp is the solvent viscosity in the pore, ∆P is the applied pressure, and ∆π is
the osmotic pressure difference across the pore.

The bulk solvent properties can significantly differ from the properties of thin
liquid layers present in narrow pores. In a recent study of Bowen et al. [11],
the ratio of bulk solvent viscosity η0 and pore viscosity ηp was expressed as a
function of pore size rp in the following form:

ηp
= 1 + 18 (d/rp ) − 9 (d/rp )2 , (9)
η0

where d is the thickness of oriented solvent layer.

Eq.(8) can be introduced into Eq(3), and since the solute flux can be rewritten
as js = cp V , the integration of the concentration gradient across the membrane
thickness yields:

Kc Φ
R=1− , (10)
1 − (1 − Kc Φ) exp (−P em )

where the modified dimensionless Peclet number P em is defined as

Kc V ∆x
P em = , (11)
Kd Dp Ak

and indicates the relative importance of convection over diffusion.

Thus, Eq. (10) allows us to determine the limiting rejection:

Rlim = 1 − Kc Φ. (12)

A similar approach to Bowen et al. [7] was used by Nakao et al. [12] to develop
the steric-hindrance pore (SHP) model, which was later applied by Wang et
al. [13] to predict the separation performance of NF membranes. In the SHP
model, the phenomenological coefficient σ obtained from the Spiegler–Kedem
analysis is linked to the membrane morphological parameter λ as follows:

16
  h i
σ = 1 − 1 + λ2 (1 − λ)2 2 − (1 − λ)2 (13)
9

Note that as far as neutral solutes are considered, Eq.(13) only differ from
Eq.(12) in the analytical expressions for Kc and Kd described in Eqs. (5).

6
More detailed description of the SHP model can be found in other studies
[12–14].

It should be pointed out that these models assume a spherical solute transport
in paralel cylindrical membrane pores with identical and constant circular
cross-sections. Pore size distribution is not considered in this paper.

2.3 Molecular size

Several measures of the molecular size are known such as the Stokes radius, the
diameter derived from the hydrodynamic volume or from molar volume, the
effective diameter, and the molecular width [15]. Among them, the effective
diameter was found by Van der Bruggen et al. [16] to be the most suitable for
describing filtration phenomena, as it reflects the projection of the molecule
on the membrane. An empirical correlation between the effective radius rE
and the molecular weight Mw was determined as follows [15]:

rE = 0.0325(Mw)0.438 . (14)

The Stokes radius rS for the uncharged molecules at a given molecular weight
Mw can be determined by using the following empirical relation given by
Bowan et. al. [17]:

log rS = −1.4854 + 0.461 log(Mw). (15)

Finally, the relation between the diffusivity and molecular size can be calcu-
lated with the Stokes–Einstein equation:

kT
D∞ = . (16)
6rS πη

3 Experimental

Five commercial polymeric NF and tight UF membranes were employed in this

study. The membrane specifications given by the manufacturers are shown in
Table 1.

Chemicals of analytical degree used in the experiments were purchased from

Sigma–Aldrich and Serva–Feinbiochemica. Their properties are shown in Table

7
Table 1
Membrane specifications given by the manufacturers
Name DK GE (G-5) GH (G-10) NP030 NP010
Manufacturer GE W&P Technologies Microdyn-Nadir
Internet http://www.geawater.com http://www.microdyn-nadir.com
Material of sepa- proprietary permanently hydrophilic PES
ration layer
Retention of un- 150-300 1 1000 2 2500 2 70–90 3 25-45 3
charged solutes
Pure water flux 5.5 ± 25% 1.2 ± 25% 3.2 ± 25% 1–1.8 5–10
[L/h/m2 /bar]
Classification NF UF UF NF NF
1 MWCO [Da] characterized on glucose and sucrose

2 MWCO [Da] characterized on polyethylene glycol

3 Lactose rejecton [%] (test conditions: 4% solution, 40 bar, stirred cell (700 rpm), 20 o C)

Table 2
Molecular weights, effective and Stokes radii, and diffusivities of neutral solutes in
aqueous solutions at 25 o C.
Solute Mw [g/mol] rE 1 [nm] rS 2 [nm] D∞ 3 [10−9 m2 /s]
n–Butanol 74.1 0.21 0.24 1.03
D–Ribose 150.1 0.29 0.33 0.74
D–Glucose 180.2 0.32 0.36 0.68
D–Lactose Monohydrate 360.3 0.43 0.49 0.50
Polyethylene glycol 570–630 0.54 0.62 0.39
Dextran ≈ 900 0.64 0.75 0.33
1 calculated with Eq.(14); 2 calculated with Eq.(15); 3 calculated with Eq.(16)

3.

Membrane filtration was performed using the experimental set-up shown in

Fig.1.

The membrane unit is equipped with a DDS LAB20 plate-and-frame mod-

ule. Membrane discs with the effective membrane surface area of 350 cm2
were placed in series and simultaneously tested. Single-component solutions
of 100 ppm concentration were used. Both permeate and retentate streams
were recycled into the feed tank. The cross-flow rate was set to 500 L/h which
ensures a negligible concentration polarisation effect [18]. During the filtration
experiments the applied trans-membrane pressure was changed between 0 and
40 bar. In all runs, stabilization time of 30 minutes was used before taking
samples. Permeate flux was manually measured by using a calibrated cylinder

8
 Fig. 1. Schematic diagram of the lab-scale membrane system
and a stopwatch. Concentrations of solutes were determined by total organic
carbon analyzer (TOC 500, Shimadzu). All experiments were carried out at
25 oC.

4 Results and discussion

The measured permeate fluxes versus applied pressure for the five membranes
are shown in Fig.2. The slopes of the curves are equal to the hydraulic perme-
ability of the respective membranes, which indicates that the osmotic pressure
effects of the solutes are negligible due to the low feed concentrations.

For each sets of measured data of flux Jv and the rejection R, Eq.(1) was used
to compute the values of the membrane parameters σ and Ps with nonlin-
ear two-parameter fitting in a least-squares sense. The σ corresponds to the
maximum rejection at infinite volume flux. The results of the SK analysis are
illustrated in Fig.3 using the permeation data of lactose as an example.

Prior to the steric pore model analysis, two main steps were performed. First,
the reflection coefficients of the 6 solutes were determined for the 5 membranes
using the SK model. Second, the Stokes radii rS and the effective radii rE
of all solutes were calculated from Eq.(15) and Eq.(14), respectively. Thus,
using the solute radius and the respective reflection coefficient allow us to
determine the hypothetical pore size for each membrane. The DSP model
using Eq.(10) or alternatively, the SHP model using Eq.(13) can be fitted to the

9
 −4
x 10
1
DK
GE
0.8 GH
NP030

permeate flux [m/s]

NP010
0.6

0.4

0.2

0
0 0.5 1 1.5 2 2.5 3
trans−membrane pressure [Pa] 6
x 10

Fig. 2. Measured fluxes of lactose solution (symbols) and predicted fluxes based on
pure water permeability (solid lines) for the 5 membranes.

1
DK
GE
0.8 GH
NP030
NP010
rejection [−]

0.6

0.4

0.2

0
0 2 4 6 8
permeate flux [m/s] −5
x 10

Fig. 3. Experimental data (symbols) of lactose rejection as function of flux for the
5 membranes. Solid lines were fitted using the SK model.

reflection coefficient of each uncharged solute individually in order to estimate

the pore size. This would result in slightly different estimated pore size values
for each solute and an average value could be calculated to characterize each
membrane. Nevertheless, in this study, the reflection coefficients of all solutes
were considered to compute the pore size of each membrane, which gives the
overall best-fitting solution in a least-squares sense.

The predictions of the DSP and the SHP models compared with the experi-
mental data are shown in Fig.4 and Fig.5, respectively. The quality of the fit
can be described with the coefficient of determination R2 defined as

R2 = 1 − (σi − σi∗ )2 / (σi − σ̄)2 ,

X X
(17)
i i

10
 1

0.8

0.6
σ [−]

DK
0.4 GE
GH
NP030
0.2 NP010

0
0 0.2 0.4 0.6 0.8 1
Stokes radius [nm]

Fig. 4. Relationship between reflection coefficient (e.g. limiting rejection) and Stokes
radii of solutes. The optimal estimates for the pore radii of the five membranes were
provided from the model proposed by Bowen et al. using Eq(12). Solid lines illustrate
model predictions with the best fitting rp values.

0.8

0.6
σ [−]

DK
0.4
GE
GH
0.2 NP030
NP010

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
effective radius [nm]

Fig. 5. Relationship between reflection coefficient (e.g. limiting rejection) and effec-
tive radii of solutes. The optimal estimates for the pore radii of the five membranes
were provided from the SHP model using Eq(13). Solid lines illustrate model pre-
dictions with the best fitting rp values.

where σ, σ ∗ and σ̄ are the experimental, the modelled, and the mean of the
experimental values, respectively. Using the determined pore sizes in the mod-
els allows us to compute the size of the molecule that has a limiting rejection
of 90% and thus, the empirical relations described in Sect.2.3 can be used to

11
Table 3
Estimated hypothetical pore sizes of the five membranes obtained from SHP and
DSP models using experimental data of neutral solutes.
SHP model DSP model
using effective radius using Stokes radius using Stokes radius
membrane rp[nm] MWCO R2 rp[nm] MWCO R2 rp[nm] MWCO R2
GE 0.46 290 0.954 0.52 280 0.951 0.70 380 0.912
DK 0.39 200 0.940 0.44 200 0.925 0.55 230 0.807
GH 0.73 840 0.980 0.85 820 0.976 1.14 1100 0.923
NP030 0.59 520 0.915 0.68 500 0.917 0.93 700 0.934
NP010 0.80 1040 0.973 0.93 1010 0.976 1.29 1400 0.949

calculate the MWCO from the solute radius. The obtained data of rp , MWCO
and R2 are listed in Table 4 for three different approaches. Overall, very good
approximates of the measured data are given by both models. However, the
pore sizes estimated by the DSP model are considerably bigger than the SHP
model predictions. There is also a significant difference in the estimated pore
size values depending on which measure of the solute size was employed in the
SHP model.

The overall fittings of the studied two models are shown in Fig. 6 and Fig. 7.
The estimated pore radii of the membranes listed in Table 4 and the Stokes
radii as the measure of solute size were used to compute σ for each solute
and for each membrane. Then, σ was plotted as a function of λ. As shown in
Fig. 6 and Fig. 7, the experimental values of σ are in good agreement with the
estimated σ values irrespective of which solute and membrane was applied.

R2=0.9613
0.8

0.6
σ [−]

DK
GE
0.4
GH
NP030
0.2 NP010
SHP

0
0 0.2 0.4 0.6 0.8 1 1.2
λ [−]

Fig. 6. Reflection coefficients of all solutes as a function of ratio of Stokes radii of

the solutes to the pore radii of the five membranes. The curve was calculated from
the SHP model with Eq.(13) using the pore radii rp reported in Table 4.

It is important to note that combining Eq. 8, Eq. 10 and Eq. 11 provides a

12
 1

2
R =0.9419
0.8

0.6

σ [−]
DK
GE
0.4
GH
NP030
0.2 NP010
after Bowen et al.

0
0 0.2 0.4 0.6 0.8 1
λ [−]

Fig. 7. Reflection coefficients of all solutes as a function of ratio of Stokes radii to

the pore radii of the five membranes. The curve was calculated from the DSP model
with Eq.(12) using the pore radii rp reported in Table 4.

direct estimation of the pore size without the necessity of a prior SK analysis.
The pore size can be obtained by fitting to the experimentally determined
set of applied pressure and rejection data, or vice versa, the rejection of a
noncharged solute for a given applied pressure can be calculated if the pore
size of the membrane is known. This latter concept was applied to estimate
the R of the different solutes as a function of ∆P using the prior determined
pore radius listed in Table 4. The model prediction is plotted together with
the experimental data in Fig. 8 for the membrane GE as a representative
membrane.

lactose glucose ribose butanol

1

0.8
rejection [−]

0.6

0.4

0.2

0
0 1 2 3 4
∆P [Pa] 6
x 10

Fig. 8. Rejection of solutes as a function of permeate flux for the membrane GE. The
curves were computed using Eq.(10) and assuming membrane pore radius rp = 0.70
nm.

13
5 Conclusions

The hypothetical pore radii of five commercial NF membranes (DK, GE, GH,
NP030, NP010) were evaluated from permeation experiments using different
noncharged solutes. Thermodynamical analysis of experimental data was per-
formed in order to obtain the reflection coefficients of each solute. This phe-
nomenological parameter was linked to the membrane structural parameter
using two steric pore flow models. The overall best-fitting solution in a least-
squares sense was computed for each membrane by fitting the pore size radii
to the sets of experimental data of reflection coefficients and solute radii. Pre-
dictions of both DSP and SHP models were found to be in a good agreement
with the experimental data. However, there is a significant difference in the es-
timated pore size values depending on which model is used and also on which
measure of the solute size is employed. The provided values of membrane pore
size can be applied to predict the rejection of any noncharged solute for a
given applied pressure.

References

[1] D. Bessarabov, Z. Twardowski, Industrial application of nanofiltration new

perspectives, Mem. Technol. 9 (2002) 6–9.

[2] J. Timmer, Properties of nanofiltration membranes: model development

and industrial application, Ph.D. thesis, Technische Universiteit Eindhoven,
Eindhoven, Netherlands (2001).

[3] J. Wagner, Membrane Filtration Handbook - Practical Tips and Hints,

Osmonics, Inc., Minnetonka USA, 2001.

[4] S. Allgeier, Membrane filtration guidance manual, Tech. Rep. EPA 815-R-06-
009, United States Environmental Protection Agency, Office of Water (2005).

[5] A. Scaefer, A. Fane, T. Waite, Nanofiltration – Principles and Applications,

Elsevier Advanced Technology, Oxford UK, 2005.

[6] F. P. Cuperus, D. Bargeman, C. A. Smolders, Permporometry: the

determination of the size distribution of active pores in uf membranes, J. Mem.
Sci. 71 (1992) 57–67.

[7] W. Bowen, A. Mohammad, N. Hilal, Characterization of nanofiltration

membranes for predictive purposes - use of salts, uncharged solutes and atomic
force microscopy, J. Membr. Sci. 126 (1997) 91–105.

[8] S. Singh, K. C. Khulbe, T. Matsuura, P. Ramamurthy, Membrane

characterization by solute transport and atomic force microscopy, J. Mem. Sci.
142 (1998) 111–127.

14
[9] K. Kosutic, L. Katelan-Kunst, B. Kunst, Porosity of some commercial reverse
osmosis and nanofiltration polyamide thin-film composite membranes, J. Mem.
Sci. 76 (168) (2000) 101–108.

[10] K. Spiegler, O. Kedem, Thermodynamics of hyperfiltration (reverse osmosis):

criteria for efficient membranes, Desalination 1 (1966) 311–326.

[11] W. Bowen, J. Welfoot, Modelling the performance of membrane nanofiltration-

critical assessment and model development, Chem. Eng. Sci. 57 (2002) 1121–
1137.

[12] S. Nakao, S. Kimura, Models of membrane transport ohenomena and their

applications for ultrafiltration data, Chem. Etg. Japan 15 (2) (1982) 200.

[13] X. Wang, T. Tsuru, M. Togoh, S. Nakao, S. Kimura, Evaluation of pore

structure pressure and electrical properties of nanofiltrationmembranes, J.
Chem. Eng. Jpn. 28 (2) (1995) 186–192.

[14] X. Wang, T. Tsuru, S. Nakao, S. Kimura, The electrostatic and steric-hindrance

model for the transport of charged solutes through nanofiltration membranes,
J. Membr. Sci. 135 (1997) 19–32.

[15] B. V. der Bruggen, C. Vandecasteele, Modelling of the retention of uncharged

molecules with nanoiltration, Water Research 36 (2002) 1360–1368.

[16] B. V. der Bruggen, J. Shaep, D. Wilms, C. Vandecasteele, Influence of molecular

size, polarity and charge on the retention of organic molecules by nanofiltration,
J. Mem. Sci. 156 (1999) 29–41.

[17] W. Bowen, A. Mohammad, Characterization and prediction of nanofiltration

membrane performance - a general assessement, Trans. IChemE 76 (Part A)
(1998) 885.

[18] M. Teixeira, M. Rosa, M. Nystroem, The role of membrane charge on

nanofiltration performance, J. Membr. Sci. 265 (2005) 160–166.

15
 Table 4
Nomenclature
Symbol Name Unit
cp Permeate concentration mol/m3
cr Retentate concentration mol/m3
d Thikness of oriented solvent layer m
D∞ Diffusion coefficient at infinite dilution m2 /s
Dp Hindered diffusion coefficient within the pore m2 /s
js Molar solute flux mol/m2 /s
Jv Volumetric permeate flux m/s
k Boltzmann constant (1.38 · 10−23 ) J/K
Kc Convective hindrance factor –
Kd Diffusive hindrance factor –
Mw Molecular weight g/mol
P em Modified Peclet number –
Ps Solute permeability m/s
R Rejection –
Rlim Limiting rejection –
rE Effective radius of solute m
rp Pore radius m
rS Stokes radius of solute m
T Temperature K
V Solvent velocity m/s
x Axial position within the pore m
Greek symbols
∆P Applied pressure Pa
∆π Osmotic pressure difference across the pore Pa
∆x Membrane thickness m
Φ Steric partitioning coefficient –
η0 Bulk dynamic viscosity Pas
ηp Dynamic viscosity within pores Pas
λ Ratio of solute to pore radius –
σ Reflection coefficient –

16

View publication stats