Journal of Membrane Science 279 (2006) 601607

Influence of solvent on the structure and performance

of cellulose acetate membranes
Zhansheng Li , Jizhong Ren, Anthony G. Fane, Dong Fei Li, Fook-Sin Wong
Institute of Environmental Science and Engineering, School of Civil and Environmental Engineering, Nanyang Technological University,
Innovation Centre, Block 2, Unit 237, 18 Nanyang Drive, Singapore 637723, Singapore
Received 29 April 2005; received in revised form 2 December 2005; accepted 24 December 2005
Available online 31 January 2006

Abstract
The liquidliquid demixing process of cellulose acetate (CA) in two nonvolatile solvents N-methyl-2-pyrrolidone (NMP) and -butyrolactone
(GBL) coagulating in water was investigated. Due to the relatively poor solvency of GBL for cellulose acetate and the higher interaction parameter
between GBL and water, the CA/GBL/H2 O system exhibits a delayed liquidliquid demixing, while the CA/NMP/H2 O has an instantaneous one.
The solvent quality and the interaction parameter between nonsolvent and solvent determine the different liquidliquid demixing behavior and the
distinct membrane structures. By adjusting the GBL content in the solvent mixture, the effect of solvent on the membrane structure and performance
was systematically studied. The membrane structure was tunable by changing the ratio of NMP/GBL. A critical ratio of NMP/GBL appears to
exist that denotes the transition of membrane structure from macrovoids to sponge-like.
2006 Elsevier B.V. All rights reserved.
Keywords: Cellulose acetate; Membrane; Immersion precipitation; Liquidliquid demixing process

1. Introduction
Immersion precipitation is the most common method for the
preparation of a range of membranes [13]. A homogenous polymer solution is cast as a thin film or spun as a hollow fiber and
then immersed into a nonsolvent coagulation bath. Due to the
exchange of solvent and nonsolvent between polymer solution
and coagulation bath, phase separation of polymer solution is
induced to form polymer-rich phase and polymer-lean phase.
Then a vitrification or gelation process takes place to fix the
membrane structure. The membrane structure is governed by
the thermodynamics and kinetic aspects of the system. Though
the membrane formation process is an interplay of different phenomena, it has been widely accepted that liquidliquid demixing
plays an important role in this process. Generally, a delayed
phase separation usually results in membrane with a dense thick
top layer and a sublayer with sponge-like structure, while an
instantaneous phase separation usually results in a membrane
with a thin porous top layer and a sublayer with macrovoids.

Corresponding author. Tel.: +86 25 8358 7775; fax: +86 25 8336 5813.
E-mail address: membranechina@hotmail.com (Z. Li).

0376-7388/$ see front matter 2006 Elsevier B.V. All rights reserved.
doi:10.1016/j.memsci.2005.12.054

One of the simplest and most efficient ways to investigate

the role of liquidliquid demixing during membrane formation
by immersion precipitation is to use a binary solvent mixture that can give rise to large morphological changes as a
function of the ratio between the two solvents [46]. Kools
[4] systemically studied the different morphologies obtained
from polysulfone/N-methyl-2-pyrrolidone (NMP)/tetrahydrofuran (THF) solutions with various NMP/THF ratios coagulating in water. The polysulfone/NMP/water systems give rise to
instantaneous demixing, and the obtained membranes contain
large macrovoids, while polysulfone/THF/water systems give
rise to delayed demixing, and the entire film is almost a dense
film with closed cell structure. Intermediate morphologies are
expected for solvent mixtures. By increasing THF content in
the solvent mixture, not only the number of the macrovoids
diminishes but macrovoids also broaden. Shieh and Chung [5]
investigated the effect of liquidliquid demixing on membrane
morphology and performance of cellulose acetate (CA) hollow fibers by adjusting the ratio of NMP and THF in a binary
solvent mixture. The CA/NMP/H2 O system exhibits an instantaneous liquidliquid demixing, while CA/THF/H2 O system a
delayed one. With an increase in THF content, the onset time of
liquidliquid phase separation increased, and the teardrop type

602

Z. Li et al. / Journal of Membrane Science 279 (2006) 601607

macrovoids became smaller. A combination of a volatile and

a nonvolatile solvent was used in these studies. However, the
highly volatile tetrahydrofuran would not be a favorable solvent
for membrane preparation due to the complicated handling of
the harmful or inflammable agents. Moreover, the superposition
of evaporation of the volatile solvent would make the investigation of liquidliquid demixing more complicated. Albrecht
et al. [6] investigated the mechanism of macrovoids formation
by changing the dope solvent composition. It was shown that
by a systematic variation of the solvent composition, the morphology of the membrane could be varied from macrovoids to
sponge-like structure under instantaneous demixing conditions.
Two types of macrovoids that started at different locations in the
cross-section of membranes were observed, which were induced
by the thermodynamic state of the dope and the kinetics of the
phase inversion, respectively. However, it is well known that
the process of hollow fiber spinning is very different from that
for flat-sheet membrane preparation [5]. In comparison with
the complicated hollow fiber formation process, the flat-sheet
membrane formation is to some extent more suitable for the
investigation of the mechanism of the liquidliquid demixing
process.
On the other hand, according to the work by Peinemann
et al. [7], -butyrolactone (GBL) is found to be an excellent membrane structure modifier that usually gives rise to
sponge-like membrane structures [7,8]. For cellulose acetate,
-butyrolactone is an interesting solvent since it combines nonvolatility, water miscibility and polymer dissolving properties.
However, cellulose acetate membrane formation using the GBL
system was seldom discussed in the literature. By spinning the
CA/NMP/GBL/triethylene glycol solution at 170 C into a water
bath consisting of NMP, GBL and triethylene glycol, Noriyuki
et al. [9] prepared cellulose acetate hollow fiber membranes for
hemodialysis. They also pointed out that the coagulation rate
was rapid in the case of NMP as a solvent, while the coagulation
rate was slow in the case of GBL as a solvent.
In this study, cellulose acetate was chosen as the polymer
due to its extensive use as membrane material covering from gas
separation to microfiltration. Furthermore, the inherent foulingresistant property due to its neutral surface and hydrophilicity make cellulose acetate a popular membrane material for
water and wastewater treatment [2]. The liquidliquid demixing behavior of the binary CA/NMP and CA/GBL solutions
coagulating in a water bath was investigated. Thereafter, by a

systematic variation of the ratio of the two nonvolatile solvents

NMP and GBL, the influence of solvent on the structure and
performance of cellulose acetate membranes was studied.
2. Experimental
2.1. Materials
Cellulose acetate used in this study was CA-398-30 with
averaged acetyl content 39.8 wt% supplied by Eastman Chemical Company. The polymer was dried in a vacuum oven at
110 C over night. N-Methyl-2-pyrrolidone (NMP, 99.0%) and
-butyrolactone (GBL, >99%) were purchased from Merck
and Aldrich, respectively. Both the solvents were used as
received.
2.2. Cloud point
The cloud point measurements on CA/NMP/H2 O and
CA/GBL/H2 O were carried out at 25 0.1 C as described elsewhere [10].
2.3. Intrinsic viscosity
The viscosities of dilute CA solutions and solvent were measured using an Ubbelohde viscometer (Schott AVS360, GmbH,
Germany) at 25 0.01 C. The concentration of the dilute CA
solution was about 0.5 g/dl. The intrinsic viscosity [] was determined by the one-point method [11],


2(sp ln r )
[] =
(1)
c
where c r and sp are the concentrations of polymer, the relative
viscosity and the specific viscosity, respectively.
2.4. Polymer solution preparation and viscosity
measurements
The homogenous CA/NMP/GBL solutions with various
NMP/GBL ratios were prepared (see Table 1). The viscosities
of the polymer solutions were measured using a RheoStress 300
rheometer (HAAKE Instruments Inc., Germany) with a 25 mm
cone-plate sensor at 25 0.1 C.

Table 1
The composition and viscosity of polymer solutions
ID

CA (wt%)

GBL concentration in
solvent mixture (wt%)

NMP concentration in
solvent mixture (wt%)

Viscosity (mPa s)

CA-1
CA-2
CA-3
CA-4
CA-5
CA-6
CA-7

11
11
11
11
11
11
11

0
50
60
67
75
80
100

100
50
40
33
25
20
0

4335
5690
6028
6047
6285
6911
7875

Z. Li et al. / Journal of Membrane Science 279 (2006) 601607

603

2.5. Light transmission experiments

The traditional light transmission experiment [12] was carried out to monitor the phase separation processes of the
binary CA/NMP and CA/GBL solutions coagulating in water
at 25 1 C.
2.6. Membrane preparation and performance
characterization
Polymer solutions were cast on a glass plate with a thickness of 200 m, then immediately immersed into a water bath
without an evaporation step at room temperature. The resulting membranes were kept in the water bath for several days to
remove the residual solvents.
The experiments for pure water permeability (PWP) and
the retention of the membranes were carried out in a crossflow setup at 25 1 C. The transmembrane pressure was
1 105 Pa (1 bar), and the feed flow velocity was about
1 m/s. Milli-Q ultrapure water was used to determine the
pure water permeability. A 1500 ppm dextran mixture aqueous solution was used to characterize the molecular weight
cut-off (MWCO) of membranes with a gel permeation chromatography (GPC) as described in our previous paper [13].
The feed concentration was kept constant by recirculating
the permeate. The retention of membrane was calculated by
Eq. (2):


CP
R= 1
100%
(2)
CF
where CP and CF were the dextran concentrations in the permeate and feed determined by GPC, respectively.
For those membranes with very low permeability under the
pressure of 1 105 Pa (1 bar), the experiments were carried out
in a stirred cell at 25 1 C, a pressure of 1 106 Pa (10 bar)
and a feed concentration of 1000 ppm NaCl. The concentration
of NaCl in the feed and in the permeate were determined by
conductivity measurements using JENWAY, 4330 Conductivity
& pH Meter, UK. The salt retention coefficient was determined
by Eq. (2) using the conductivities of the permeate and feed as
measures of CP and CF , respectively.

Fig. 1. The phase diagram of CA/NMP/H2 O and CA/GBL/H2 O.

2.7. Scanning electron microscopy (SEM)

Membranes samples were freeze-fractured using liquid nitrogen and coated with a thin layer of gold using an SPI-Module
sputter. The cross-section of the flat-sheet membranes was examined using a Jeol JSM-5310LV scanning electron microscope.
3. Results and discussion
3.1. The binary polymer solution systems
The cloud points of the systems of CA/NMP/H2 O and
CA/GBL/H2 O are shown in Fig. 1. The CA/NMP/H2 O system
has a smaller phase separation gap than the CA/GBL/H2 O system. In other words, more water is needed for the CA/GBL/H2 O
system to initiate the liquidliquid demixing process, and different membrane structures could be obtained from the binary
CA/NMP and CA/GBL solutions coagulating in a water bath as
shown in Fig. 2. Using NMP as the solvent, the obtained membrane contains large macrovoids, which is the typical structure
of membranes from the instantaneous liquidliquid demixing.
Using GBL as the solvent, the typical sponge-like membrane
structure for the delayed liquidliquid demixing was obtained.
This interpretation is supported by the results of the light transmission experiments. As shown in Fig. 3, the CA/NMP/H2 O

Fig. 2. The cross-section of flat-sheet membranes obtained from the binary CA/solvent solutions coagulating in water.

604

Z. Li et al. / Journal of Membrane Science 279 (2006) 601607

Table 2
The physicochemical property of NMP and GBL at 25 C
Solvent

Molecular weight

Molar volume (cm3 /mol)

Viscosity (mPa s)

Boiling point ( C)

Vapor pressure (kPa)

Surface tension (mN/m)a

NMP
GBL

99.13
86.09

96.5
76.8

1.029
1.12

202
204

0.04
0.43

45.25
47.79

At 20 C.

Fig. 3. The light transmission experiment of 11 wt% CA/solvent immersed into

water bath.

system exhibits an instantaneous liquidliquid demixing, while

the CA/GBL/H2 O system shows a delayed one with a delay time
of about 30 s.
It is interesting that the two high boiling point, low vapor pressure solvents NMP and GBL have very similar physicochemical
properties [1416] (see Table 2), but exhibit dramatically different behavior in the membrane formation process. Yanagishita et
al. [17] and Ruaan et al. [18] concluded that the quality of solvent would influence the membrane morphology. The intrinsic
viscosity is a good parameter to characterize the solvent quality
because the intrinsic viscosity of a given polymer is different in
different solvents depending on the solvent quality. In a good
solvent, the polymer molecule is fairly loosely extended, and
the intrinsic viscosity is high. In a poor solvent, segments of
a polymer molecule attract each other in solution more strongly
than they attract surrounding solvent molecules. The polymer
molecule assumes a tighter conformation, and the solution has
a lower intrinsic viscosity [19]. As listed in Table 3, the intrinsic viscosity of CA in NMP (1.66 dl/g) is higher than that in

GBL (1.56 dl/g), which means NMP is a better solvent for CA

than GBL. Moreover, the specific viscosity could also be used
to determine the quality of solvent. In general, the specific viscosity of polymer solutions in a good solvent is higher than
that in a poor solvent in dilute solutions, whereas the reverse
is true in concentrated solutions [20,21]. The specific viscosities of CA/NMP are higher than those of CA/GBL in the dilute
solution for the experiment of intrinsic viscosity, while they are
lower than those of CA/GBL in the 11 wt% polymer solutions
(see Table 3). This further confirms that GBL has poorer solvency than NMP for CA.
It is widely accepted that the viscosity of the polymer solution
usually plays an important role in membrane formation process.
Increase in the viscosity of the polymer solution will decrease
the tendency for macrovoid formation [3]. As listed in Table 1,
the viscosity of binary CA/GBL solution was almost two times
of that of the binary CA/NMP solution, which could to some
extent determine the different membrane structures from the
CA/NMP and CA/GBL solutions coagulating in water bath. For
example, the sponge-like membrane structure was obtained from
the CA/GBL systems with higher viscosity.
From the viewpoint of FloryHuggins theory for polymer
solutions, a change in the solvent of the membrane-forming system is mainly reflected in a change in the nonsolvent/solvent
interaction parameter g12 and the position of the liquidliquid
demixing gap in the phase diagram. The H2 O/GBL interaction
parameter g12 was calculated from the Wilson parameters for
H2 O/GBL system as described elsewhere [2224]. As shown
in Fig. 4, the interaction parameter g12 of H2 O/GBL is higher
than that of H2 O/NMP [25], which means the mutual affinity

Table 3
The intrinsic viscosity and specific viscosity of the binary CA/solvent systems
System

[] (dl/g)

sp of dilute CA
solution

sp of 11 wt% CA
solution

CA/NMP
CA/GBL

1.66
1.56

1.10
1.05

2610
4551

Fig. 4. The H2 O/solvent interaction parameters g12 as a function of solvent

volume fraction.

Z. Li et al. / Journal of Membrane Science 279 (2006) 601607

between water and GBL are weaker than those between water
and NMP, which usually gives rise to sponge-like membrane
structure [3,17,18,26].
From the above analysis, the higher viscosity of CA/GBL
solution due to the relative poor solvency of GBL for cellulose
acetate and the higher nonsolvent/solvent interaction parameter
g12 of H2 O/GBL dominates the delayed liquidliquid demixing and sponge-like membrane structure of the CA/GBL/H2 O
system. The binary systems CA/NMP and CA/GBL appear to
be suitable model systems for the investigation of the mechanism of phase separation and gelation during the immersion
precipitation process. Further work should be carried out in this
field.
3.2. Effect of solvent mixture on membrane structure
Based on the different structures obtained from the binary
CA/NMP and CA/GBL systems coagulating in water bath and
some insight into the mechanism for liquidliquid demixing,
intermediate changes in the membrane structure are expected

605

for solvent mixtures of NMP and GBL. A systematic study was

carried out to examine the effect of solvent mixture on membrane
structure and performance by changing the GBL content in the
solvent mixture from 0% to 100%.
A series of membrane structures obtained from the 11 wt%
ternary CA/NMP/GBL system with various ratios of NMP/GBL
are shown in Fig. 5 Figs. 2 and 5. As can be seen, as the GBL
content in the solvent mixture increased from 0% to 75%, all the
membranes obtained consisted of large macrovoids. There was
no obvious change except that the skin layer differed slightly.
However, by increasing the GBL content from 75% to 80%,
the macrovoids membrane structure changed dramatically to the
typical sponge-like membrane structure. There could exist a critical ratio that denotes the point of change of membrane structure
from macrovoids to sponge-like.
3.3. Effect of solvent mixture on membrane performance
In order to further investigate the effect of solvent mixture of NMP/GBL on membrane formation, the PWP and

Fig. 5. The cross-section of flat-sheet membranes obtained from the ternary CA/NMP/GBL solutions with various GBL content coagulating in water.

606

Z. Li et al. / Journal of Membrane Science 279 (2006) 601607

Acknowledgements
This work was supported by the Agency for Science, Technology and Research, Singapore, Temasek Professor Program
on Membrane Technology for Sustainable Water, Grant No. 022
104 0029.
References

Fig. 6. The influence of GBL concentration on the membrane separation properties.

retention of the membranes obtained were examined. As shown

in Fig. 6, with the increase of GBL content in the solvent mixture from 0% to 75%, the PWP decreased from 2803 105 to
178 105 l/m2 h Pa (2803 to 178 l/m2 h bar), while the retention of the membrane increased from 0% to 90% for 100 K
dextran, which implies that the membrane structure became
tighter. By further increasing the GBL content from 75% to 80%
and then to 100%, the PWP decreased dramatically. Filtration
experiments with 1000 ppm NaCl aqueous solution at 1 106 Pa
(10 bar) showed that these membranes behaved as nanofiltration
membranes.
In summary, this paper shows how membrane performance
can be finely tuned by use of selected (good/bad) solvent mixtures. It points to a convenient method for control of membrane
separation properties.
4. Conclusions
The liquidliquid demixing process of cellulose acetate in
two nonvolatile solvents, NMP and GBL, coagulated in water
was investigated. The different liquidliquid demixing behavior and the distinct membrane structures obtained were ascribed
to the difference in the solvent quality of NMP and GBL and
the difference in the nonsolvent/solvent interaction parameters between H2 O/NMP and H2 O/GBL. The poor solvency
of GBL for CA and the higher nonsolvent/solvent interaction
parameter g12 of H2 O/GBL dominate the delayed liquidliquid
demixing of the CA/GBL system coagulating in water bath and
gives rise to the sponge-like membrane structure.
By adjusting the GBL content in the solvent mixture, the
effect of solvent on the membrane structure and performance
was studied. The membrane structure was tunable by changing
the ratio of NMP/GBL. A critical ratio of NMP/GBL appears
to exist that denotes the point of change of membrane structure
from macrovoids to sponge-like.

[1] R.W. Baker, Membrane Technology and Applications, second ed., John
Wiley & Sons, Ltd., UK, 2004.
[2] S.P. Nunes, K.-V. Peinemann, Membrane Technology in the Chemical
Industry, Wiley-VCH Verlag GmbH, Weinheim, 2001.
[3] P. van de Witte, P.J. Dijkstra, J.W.A. van den Berg, J. Feijen, Phase
separation processes in polymer solutions in relation to membrane formation, J. Membr. Sci. 117 (1996) 1.
[4] W.F.C. Kools, Membrane Formation by Phase Inversion in Multicomponent Polymer Systems: Mechanisms and Morphologies, Ph.D. Dissertation, University of Twente, Enschede, 1998.
[5] J.-J. Shieh, T.S. Chung, Effect of liquidliquid demixing on the membrane morphology, gas permeation, thermal and mechanical properties of cellulose acetate hollow fibers, J. Membr. Sci. 140 (1998)
67.
[6] W. Albrecht, Th. Weigel, M. Schossig-Tiedemann, K. Kneifel, K.V. Peinemann, D. Paul, Formation of hollow fiber membranes from
poly(ether imide) at wet phase inversion using binary mixtures of solvents for the preparation of the dope, J. Membr. Sci. 192 (2001)
217.
[7] K.-V. Peinemann, J.F. Maggioni, S.P. Nunes, Poly(ether imide) membranes obtained from solution in cosolvent mixtures, Polymer 39 (1998)
3411.
[8] K. Kneifel, K.V. Peinemann, Preparation of hollow fiber membranes
from polyetherimide for gas separation, J. Membr. Sci. 65 (1992)
295.
[9] T. Noriyuki, S. Hidehiko, O. Kazutake, Cellulose hollow fiber membrane
and its production, JP10,216,489, 1998.
[10] W.W.Y. Lau, M.D. Guiver, T. Matsuura, Phase separation in polysulfone/solvent/water and polyethersulfone/solvent/water systems, J.
Membr. Sci. 59 (1991) 219.
[11] O.F. Solomon, I.Z. Ciuta, Determination de la viscosite intrins`eque de
solutions de polym`eres par une simple determination de la viscosite, J.
Appl. Polym. Sci. 6 (1962) 683.
[12] J. Reuvers, C.A. Smolders, Formation of membranes by means of
immersion precipitation Part II. The mechanism of formation of membranes prepared from the system cellulose acetateacetonewater, J.
Membr. Sci. 34 (1987) 67.
[13] J. Ren, Z. Li, F.-S. Wong, Development of asymmetric BTDA-TDI/MDI
(P84) co-polyimide hollow fiber membranes for ultrafiltration: the influence of shear rate and approaching ratio on membrane morphology and
performance, J. Membr. Sci. 248 (2005) 177.
[14] B.E. Boling, J.M. Prausnitz, J.P. OConnell, The Properties of Gases
and Liquids, fifth ed., McGraw-Hill, NY, 2001.
[15] D.R. Lide, Handbook of Chemistry and Physics, 84th ed., CRC Press,
NY, 20032004.
[16] P. Ma, Handbook of Essential Data for Petrochemical Industry (Continuation), Chemical Industry Press, Beijing, 1993 (in Chinese).
[17] H. Yanagishita, T. Nakane, H. Yoshitome, Selection criteria for solvent
and gelation medium in the phase inversion process, J. Membr. Sci. 89
(1994) 215.
[18] R.-C. Ruaan, T. Chang, D.-M. Wang, Selection criteria for solvent and
coagulation medium in view of macrovoid formation in the wet phase
inversion process, J. Polym. Sci. Part B: Polym. Phys. 37 (1999) 1495.
[19] M.J. He, W.X. Chen, X.X. Dong, Polymer Physics, Fudan University
Press, Shanghai, 1990 (in Chinese).
[20] Y. Isono, M. Nagasawa, Solvent effects on rheological properties of
polymer solutions, Macromolecules 13 (1980) 862.

Z. Li et al. / Journal of Membrane Science 279 (2006) 601607

[21] Z. Li, C. Jiang, Investigation into the rheological properties of
PES/NMP/nonsolvent membrane-forming systems, J. Appl. Polym. Sci.
82 (2001) 283.
[22] G. Wilson, Vaporliquid equilibrium. XI. A new expression for the
excess free energy of mixing, J. Am. Chem. Soc. 86 (1964) 127.
[23] D.H.S. Ramkumar, S.V. Odak, A.P. Kudchadker, Mixture properties
of the water + -butyrolactone + tetrahydrofuran system. Part 3. Isobaric
vaporliquid equilibrium of water + -butyrolactone and tetrahydrofuran + -butyrolactone at 1.013 bar, J. Chem. Eng. Data 34 (1989) 466.

607

[24] F.W. Altena, C.A. Smolders, Calculation of liquidliquid phase separation in a ternary system of a polymer in a mixture of a solvent and a
nonsolvent, Macromolecules 15 (1982) 1491.
[25] Z. Li, C. Jiang, Determination of the nonsolventpolymer interaction
parameter 13 in the casting solutions, J. Membr. Sci. 174 (2000)
87.
[26] M.L. Yeow, Y.T. Liu, K. Li, Morphological study of poly(vinylidene
fluoride) asymmetric membranes: effects of the solvent, additive, and
dope temperature, J. Appl. Polym. Sci. 92 (2004) 1782.