Modeling and Simulation of the Phase-Inversion Process

During Membrane Preparation

G. R. FERNANDES, J. C. PINTO, R. NOBREGA

COPPE/Chemical Engineering Program, Federal University of Rio de Janeiro, P.O. Box 68502,
Rio de Janeiro 21945-970, Brazil

Received 15 November 2000; accepted 4 April 2001

Published online 3 October 2001; DOI 10.1002/app.2159

ABSTRACT: The objective of this work is to present a simple mathematical model to

describe the main features of phase inversion by immersion precipitation. The model
was developed for planar geometry and is used to simulate membrane synthesis. Two
systems commonly used for membrane preparation were used for simulations: cellulose
acetate/acetone/water and polyetherimide/N-methylpyrrolidone/water. The influence of
nonsolvent addition to initial polymer casting solutions, solvent addition to the coagu-
lation bath, and geometric variables, such as polymer film thickness, on the final
precipitation conditions were studied through simulation and compared to available
experimental data. The results are in good agreement with published results.1–3 It is
shown that polymer film composition profiles at the moment of precipitation may give
important information about the structure and substructure of formed membrane. It is
also shown, for both polymeric systems investigated in this work, that the dynamics of
the mass transfer process seems to be much more important than the influence of the
concentration on the diffusion and thermodynamic partition coefficients, as fair agree-
ment with available experimental data was obtained even when these coefficients were
assumed to be constant. © 2001 John Wiley & Sons, Inc. J Appl Polym Sci 82: 3036 –3051, 2001

Key words: membranes; modeling; phase separation; membrane preparation; phase

inversion; mathematical model; polymer solution

INTRODUCTION vent, that is compatible with the solvent but does

not dissolve the polymer species), which forces the
Polymer membranes are widely used for the sepa- polymer precipitation and the formation of a poly-
ration of chemical constituents and the purification mer film membrane.3 The final membrane proper-
of process streams in many important industrial ties are known to depend on the precipitation con-
fields. Usual applications include solid–liquid sepa- ditions and variables such as the initial polymer
rations, removal of ethanol from aqueous solutions, concentration in the polymer casting solution, ini-
and gas separations.4 These polymer membranes tial solvent concentration in the coagulation bath,
are usually prepared through phase inversion, relative amounts of polymer and nonsolvent in cast-
where a polymer solution (containing the polymer ing solutions, and so on, which may exert a major
species and a solvent) is immersed in a coagulation impact on the final performance of the membrane.5
bath (containing a chemical component, the nonsol- Therefore, the appropriate understanding of the
phase-inversion phenomena and the development
of mathematical models to describe the membrane
Correspondence to: R. Nobrega (nobrega@peq.coppe.ufrj.br).
formation may be of great value for those interested
Journal of Applied Polymer Science, Vol. 82, 3036 –3051 (2001)
© 2001 John Wiley & Sons, Inc. in membrane synthesis and its applications.
3036
 MODELING AND SIMULATION OF PHASE INVERSION 3037

Few models are available in the literature to face. The coefficients are also assumed to depend
describe the phase-inversion process on planar on the local states of both solutions. The model
systems.3,5–13 Because the phase-inversion pro- structure comprises a set of nonlinear, coupled
cess depends on the equilibrium thermodynamics partial differential equations with time-depen-
of the system, the surface properties of the inter- dent and moving boundary conditions. The cellu-
face that separates the polymer solution and the lose acetate (AC)/acetone/water system is gener-
coagulation bath, and the mass transfer of the ally used to validate the models because model
chemical species added to both solutions and be- parameters needed to simulate the phase-inver-
cause fast polymer precipitation leads to the de- sion process may be evaluated from available ex-
velopment of huge concentration gradients in perimental data in this case.
both polymer and coagulation solutions, these The goal of this work was to develop a much
models present a very detailed description of the simpler phase-inversion model that could be used
precipitation mechanism. For this reason, except for practical purposes for those involved with
in the case of molecular modeling proposed by membrane preparation and that would relax
Termonia,9 these models depend on parameters some of the usual assumptions and reduce the
that are difficult to obtain in the open literature number of parameters needed for simulations.
(e.g., ternary diffusion coefficients and the con- The main objective was to provide a simple math-
centration dependence of diffusion coefficients ematical structure that could be used to simulate
and thermodynamics parameters) and difficult to how concentration gradients change in each
estimate a priori on the basis of known constitu- phase as certain key preparation variables, such
tive equations. Besides, the final model structure as initial concentration, are changed. The model
may become very complex, in the sense that the must describe the phase-inversion conditions and
model is constituted by a coupled set of partial
the precipitation lag time as functions of the prep-
differential equations that depend on time-vary-
aration conditions. Polymer, solvent, and nonsol-
ing parameters that may be unbounded at certain
vent concentrations at the interface during poly-
preparation conditions, especially in the very be-
mer precipitation and the concentration gradients
ginning of the simulation.3 [eqs. 50 – 60] After dis-
inside individual phases provide information
cretization, due to the intrinsic nature of the
about the structure of the membrane.5,10,13,14 Pre-
available models and due to the preparation con-
cipitation time provides fundamental information
ditions, the Jacobian matrix is full at each dis-
for membrane manufacture and process design
cretization grid point, and the discretized system
and is also useful for parameter estimation, as
of ordinary differential equations may be very
stiff. This means that very small integration steps precipitation time may be easily obtained in the
and a very large computation time may be re- lab from standard kinetic experiments.2
quired for simulation of the phase-inversion It is shown here that a very simple mathemat-
model. As model solutions may be very complex ical model may be built to allow the simulation of
and may depend on many physical parameters phase-inversion phenomena. The model combines
that almost always are unknown, they can seldom simple mass balance equations for the different
be applied to practical purposes, such as the pre- chemical components in the different phases with
liminary investigation of the effects of phase-in- a thermodynamic equilibrium constraint. Besides
version operation variables on precipitation con- the thermodynamic equilibrium relationship, the
ditions, in the form presented. This is especially model depends only on a set of constant binary
true during the screening stages of membrane coefficients (diffusion coefficients) that may be ob-
development, when large sets of possible prepa- tained in the open literature15 or may be com-
ration conditions are taken into consideration and puted with the help of known constitutive equa-
most conditions are discarded before actual ex- tions.16 Additionally, model equations are coupled
periments are carried out. through simple linear algebraic constraints,
The models described previously are based on which means that standard integration proce-
the Maxwell–Stefan equations,4 where diffusion dures such as DASSL17 are able to integrate the
coefficients depend on the local states of the solu- model very efficiently. It is shown here that such
tion and cross-diffusion terms are not neglected. a model can be used very successfully to describe
The Flory–Huggins equation13 is used to describe the phase-inversion phenomena during mem-
the local activities of the chemical components brane preparation and can be used for practical
and the thermodynamic equilibrium at the inter- purposes and the preliminary investigation of
3038 FERNANDES, PINTO, AND NOBREGA

conditions, and the polymer constituent eventu-

ally precipitates. Then, the casting solution gives
birth to two distinct phases: a polymer-rich phase
and a diluted polymer phase. The polymer phase
leads to the membrane formation. If the precipi-
tation occurs with high polymer concentrations,
the membrane skin tends to be dense and very
selective; otherwise, the membrane skin tends to
be porous and to present low selectivity. If poly-
mer concentration gradients are very large dur-
ing precipitation, the skin and the sublayer tend
to present very different properties.
Figure 1 Schematic representation of the phase-in- Prediction of membrane morphology through
version process.
simulation, based on knowledge of the initial
states of both casting solution and coagulation
bath, is a highly desirable objective. Although this
membrane preparation conditions, despite the objective is far from realistic, modeling of mass
much less rigorous derivation of model equations. transfer phenomena during the quenching step
can provide important information about the
structure and substructure of the formed mem-
MATHEMATICAL MODEL brane, as discussed before. With this objective, a
mass transfer model is presented later. The model
The phase-inversion process is schematically rep- is based on very simple assumptions, which in-
resented in Figure 1. A thin film of a certain clude constant physical parameters, additivity of
polymer solution (Phase 1, which contains the volumes, thermodynamic equilibrium at the in-
polymer constituent, the solvent, and a certain terface (described through constant partition co-
amount of the nonsolvent) is initially casted onto efficients), absence of polymer in the coagulation
a glass support. This step is usually called the bath, and the existence of one-dimensional con-
casting step. This solution is then immersed into centration gradients. Cross-diffusion terms are
an appropriate coagulation bath (Phase 2, which neglected and the well-known Fick equations13
contains the nonsolvent and a certain amount of are used to describe mass transfer within each
the solvent). This step is normally called the phase. Despite the simplifying assumptions, the
quenching step. An intermediate step, called the model considers that the casting solution/coagu-
evaporation step, where the initial polymer solu- lation bath interface moves during the process.
tion is exposed to a controlled gas atmosphere to According to the model assumptions, the mass
allow the manipulation of the initial gradients in transfer of component i within the phase j can
the polymer solution, is sometimes performed be- then be written as
fore the immersion into the coagulation bath.
This step is not covered in this work, but the
model presented can be easily extended to include
⭸´ C 共i, j兲
⭸t
⫽ D 共i, j兲 䡠
⭸x 2 再
⭸ 2C 共i, j兲 i ⫽ 1, 2, for j ⫽ 1
, i ⫽ 1, for j ⫽ 2
this step. The result of the phase-inversion pro-
(1)
cess is the production of an asymmetric polymer
film with one thin and relative dense skin and a where C(i,j) is the molar concentration of compo-
porous sublayer. nent i in phase j (kg/m3), D(i,j) is the diffusion
The formed membrane must show high perme- coefficient of component i in phase j (m2/s), and x
ability and high selectivity. These properties may is the mass transfer direction (m). This is subject
be correlated to the membrane morphology so to the following boundary conditions:
that polymer concentrations during precipitation
may give clues in regard to the final properties of
the membrane.5 During the quenching step, a
⭸C 共i,1兲
⭸x
冏 x⫽0
⫽ 0, i ⫽ 1, 2, for j ⫽ 1 (2)
significant mass exchange of solvent and nonsol-
vent occurs through the casting solution/bath co-
agulation interface. The mass transfer between
the two phases leads to unstable thermodynamic
⭸C 共i,2兲
⭸x
冏 x⫽W
⫽ 0, i ⫽ 1, for j ⫽ 2 (3)
 MODELING AND SIMULATION OF PHASE INVERSION 3039

Table I Typical Values for the Diffusion

Coefficienta 冋 冉
C *共3,1兲 ⫽ ␳ 共3,1兲 䡠 1 ⫺
Pm 共1,1兲
␳ 共1,1兲
䡠 C 共1,1兲 冊
Component Phase
Diffusion Coefficient
(m2/s) ⫺ 冉 Pm 共2,1兲
␳ 共2,1兲
䡠 C 共2,1兲 冊册 (9)

Solvent Polymer solution 1 ⫻ 10⫺11

Nonsolvent Polymer solution 1 ⫻ 10⫺11 where C*(i, j) is the mass concentration of compo-
Solvent Coagulation bath 1 ⫻ 10⫺9 nent i in phase j (mol/m3) and ␳(i,j) is the density of
Nonsolvent Coagulation bath 1 ⫻ 10⫺9 component i in phase j, whereas the nonsolvent
concentration in the coagulation bath is
a
Reuvers and Smolders.2

C 共2,2兲 ⫽ 冉 冊冋 冉
␳ 共2,2兲
Pm 共2,2兲
䡠 1⫺
Pm 共1,2兲
␳ 共1,2兲
䡠 C 共1,2兲 冊册 (10)
C 共i,1兲兩 L ⫽ k 共i兲 䡠 C 共i,2兲兩 L, i ⫽ 1, 2 (4)
Finally, the interface displacement velocity can

冏 冏
⭸C 共i,1兲 ⭸C 共i,2兲 be derived from the global mass balance equation
D 共i,1兲 䡠 ⫽ D 共i,2兲 䡠 , i ⫽ 1, 2 of the polymer species. The polymer mass is as-
⭸x x⫽L
⭸x x⫽L sumed to be constant and completely contained by
(5) the casting solution. Thus

where L is the casting solution thickness at time

t (m), k(i) is the partition coefficient of component
i, and W is the external limit of the coagulation
M 共3,1兲 ⫽ A 䡠 冕
0
L
C 共3,1兲 䡠 dx (11)

bath.
Equations (2) and (3) mean that there is no where M(3,1) is the total polymer mass in polymer
mass exchange through the borders of the system. solution (kg) and A is the flow transversal area
Equation (4) describes the local thermodynamic (m2). If eq. (11) is derived in respect to time and
equilibrium at the interface, and Equation (5) made equal to zero, then
describes the mass transfer continuity at both

冕
sides of the interface. L
Assuming that the initial solutions are homo- dL 1 dC共3,1兲
⫽⫺ 䡠 䡠 dx (12)
geneous, it is possible to write dt C共3,1兲 兩L dt
0

C 共i, j兲兩 t⫽0 ⫽ C 0共i, j兲, i, j ⫽ 1, 2 (6) which is the convective interface displacement
velocity.
The global mass balance equation may be written Equations (1–12) constitute the mathematical
as model used here to simulate the phase-inversion
phenomena. For the model to be solved, it is nec-

冘␾ essary to know the molecular weights, densities,

NC

共i,j兲 ⫽1 (7) partition coefficients, and diffusion coefficients for

i⫽1 all the chemical species involved. The first two

where NC is the number of chemical species and

␾(i,j) is the volume fraction of component i in the Table II Physical Properties for AC/Acetone/
phase j, defined as Water Systema

C 共i, j兲 䡠 Pm 共i兲 Density Molecular Weight

␾ 共i, j兲 ⫽ (8) (kg/m3)
␳ 共i兲 Component (kg/kmol)

AC 1.428 27,000
where Pm(i) is the molecular weight of component
Acetone 757 58
i (kg/kmol) and ␳(i) is the density of the pure Water 997 18
component i, so that the polymer composition in
the casting solution is a
Reuvers and Smolders.2
3040 FERNANDES, PINTO, AND NOBREGA

Table III Physical Properties for rameters needed for simulation are shown in Ta-
PEI/NMP/Water Systema bles I–III.
A central finite difference discretization
Molecular
strategy was used to provide numerical solu-
Density Weight
Component (kg/m3) (kg/kmol)
tions of the mathematical model. This proce-
dure transforms the model equations into a set
PEI 1370 22,400 of algebraic differential equations, which may
NMP 1027 96.13 be solved with standard numerical codes, such
Water 997 18.0 as DASSL.17 To account for the moving bound-
a
ary, an adaptative discretization grid scheme
Furtado.18
was used. According to this scheme, the last
discretization point of each phase was placed at
the interface, and its axial coordinate was al-
sets of parameters are assumed to be known. The lowed to vary with the interface position. As
partition coefficients may be estimated from sim-
soon as the interface position crossed one of the
ple static phase-separation experiments.18 Al-
other static discretization points, the integra-
though we are aware that partition coefficients
tion was halted, the grid was updated, and the
may depend on the local states of the system,
numerical procedure was reinitialized. The ad-
changes of orders of magnitude are not expected,
aptative strategy is illustrated in Figure 2.
so rough estimates of these parameters can be
As the numerical integration proceeded, poly-
obtained very easily, either experimentally or
through simulation. Diffusion coefficients may be mer, solvent, and nonsolvent concentrations for
estimated from dynamic experiments (which is all discretization elements inside the casting and
time consuming and difficult to do in the lab) or coagulation solutions were computed. The dy-
may be guessed from published data with accu- namic trajectory of the grid points and especially
racy within an order of magnitude. of the interface could then be followed graphically
The model parameters used in these simula- or numerically in a thermodynamic equilibrium
tions were obtained from published material. Dif- chart. The equilibrium charts (the binodal curve
fusion coefficients were taken from Reuvers2 and of the system) were taken from Reuvers and
are shown in Table I. Solvent and nonsolvent Smolders2 for the AC/acetone/water system and
partition coefficients are typical values presented from Furtado18 for the polyetherimide (PEI)/N-
by Reuvers and Smolders2 and Furtado.18 The methylpyrrolidone (NMP)/water system as pre-
typical values for partition coefficients of solvent sented in Figure 3. As soon as the dynamic tra-
and nonsolvent are 1.0 and 0.5, respectively. Ad- jectory crossed the system binodal curve, it was
ditional physical properties were also presented assumed that precipitation occurred. Figure 4 il-
by Reuvers and Smolders2 and Furtado.18 All pa- lustrates the procedure.

Figure 2 Numerical representation of the boundary movement.

 MODELING AND SIMULATION OF PHASE INVERSION 3041

Figure 3 Phase diagram for the PEI/NMP/water system. Temperature ⫽ 25°C (from
Furtado,18 chapter 4, p 89, unpublished results).

In many polymeric systems used to prepare a general way, both situations (either binodal or
membranes by immersion precipitation, we can spinodal phase separation) could occur, depend-
suppose, mainly on the basis of the membrane ing on the polymeric systems and immersion pre-
morphology, that the spinodal phase separation cipitation conditions. In this work, the binodal
was present during the process. Nevertheless, in precipitation was used as the phase-separation

Figure 4 Strategy for the computation of the precipitation time.

3042 FERNANDES, PINTO, AND NOBREGA

Table IV Ts for the AC/Acetone/Water System as a Function of the

Precipitation Bath Composition for Two Polymer Concentrations in Casting
Solution

Experimental
Precipitation Bath Precipitation
Composition ␾ (%) T s (s) Timea (s)

␾(1,2) ␾(2,2) ␾(3,1) ⫽ 10% ␾(3,1) ⫽ 15% ␾(3,1) ⫽ 15%

0 100 26.7 17.3 19.8

16 84 128 25.1 23.0
Not
36 64 detected 46.0 43.0
a
Reuvers,1 chapter 6, p 136.

criterion. It is possible to choose other criteria alyzed for the coagulation bath, including 0/100,
based on spinodal curve, for instance, but this 16/84, and 36/64 (vol %). Results are presented in
does not seem to be necessary, as shown in the Table IV and in Figures 5 and 6.
following sections. Table IV presents the precipitation time calcu-
lated through simulation for the AC/acetone/wa-
ter system and the experimental results obtained
RESULTS AND DISCUSSION by Reuvers.1 Results are very similar and indi-
cate that the time lag for precipitation increased
The phase-inversion model was used to simulate as solvent was added to the coagulation bath. The
two different systems that are commonly used for simulation results also indicate that the model
membrane manufacture: AC/acetone/water and was able to capture the most important dynamic
PEI/N-methylpyrrolidone (NMP)/water. The first features of the phase-inversion process.
system was used to compare the model predic- Figure 5 shows the effect of different solvent/
tions with published results.2,3 Simulations were nonsolvent ratios in the coagulation bath on the
carried out to investigate the effect of solvent polymer concentration profiles at precipitation
addition to the coagulation bath, the effect of non- conditions. The addition of solvent to the coagu-
solvent addition to the casting solution, and the lation bath led to lower interfacial polymer con-
influence of the initial casting solution thickness centration at precipitation conditions and lower
(Lo) on the polymer concentration profile and polymer concentration gradients at the casting
casting solution contraction rate. The parameters solution. Figure 6 shows how the interface posi-
needed for simulation are presented in Tables tion responded to changes in the coagulation bath
I–III, and the partition coefficients were kept composition. The addition of solvent to the coag-
equal to 1.0 for solvent and 0.5 for nonsolvent, as ulation bath delayed the phase-inversion phe-
discussed before. nomena and led to larger shrinkage of the casting
solution.
Qualitatively, the results presented in Figures
Influence of Solvent Concentration in the
5 and 6 agree fairly well with results presented by
Coagulation Bath
Tsay and McHugh3 for other process conditions.
To investigate the influence of the addition of According to Tsay and McHugh, the addition of
solvent in the coagulation bath on polymer con- solvent to the coagulation bath causes an increase
centration profiles and casting solution contrac- in thickness and a decrease in density of the
tion rates, simulations with different initial sol- membrane skin. These behaviors are confirmed in
vent/nonsolvent concentration ratios were per- Figure 5, through simulation. Figures 5 and 6
formed for the AC/acetone/water system. The show very clearly that depending on the initial
casting solution was assumed to contain 10% solvent/nonsolvent concentration ratios in the co-
polymer and 90% solvent (vol %) in all cases. agulation bath, membranes with different char-
Different solvent/nonsolvent feed ratios were an- acteristics can be formed.
 MODELING AND SIMULATION OF PHASE INVERSION 3043

Figure 5 Simulated polymer concentration profiles at precipitation for the AC/ace-

tone/water system for different solvent/nonsolvent coagulation bath compositions. Lo
⫽ 200 ␮m. (⫹) 0/100, simulated precipitation time (Ts) ⫽ 26.7 s; (‚) 16/84, Ts ⫽ 128 s;
and (夹) 36/64, no precipitation detected. Polymer casting solution: AC/acetone (10/90).

Additional simulation studies were carried out the polymer concentration at the interface de-
for the PEI/NMP/water system. To our knowl- creased. The simulation results (instantaneous
edge, no modeling results are available for this precipitation) suggest that very asymmetric
system, although experimental data with regard membranes were produced in this case, which
to the membrane formation process can be found could be confirmed experimentally.20 The qualita-
in the literature.2,19 The polymer concentration in tive behavior of the phase-inversion process for
the casting solution was assumed to be equal to the PEI/NMP/water system resembled very
10% (vol %), and the casting solution was as- closely the behavior of the AC/acetone/water sys-
sumed to contain only polymer and solvent in all tem, although the first presented much faster pre-
simulations. Table V shows the time lag for pre- cipitation dynamics. For this reason, all interest-
cipitation for different coagulation bath composi- ing dynamics were observed in the spatial region
tions as obtained through simulation. Figures 7 that was located very close to the interfacial
and 8 show the polymer concentration profiles at boundary.
the moment of precipitation and the contraction Figure 8 shows that the total contraction of the
rates of the casting solution as a function of time casting solution was very small for the PEI/NMP/
respectively. water system due to the early polymer precipita-
Figure 7 shows that the concentration gradi- tion, which contributed to the high asymmetry
ents were very steep in all cases analyzed and and sublayer porosity of these membranes. The
that the addition of solvent to the coagulation results obtained for the precipitation time, shown
bath tended to produce less dense membranes as in Table V, indicate that precipitation may be
3044 FERNANDES, PINTO, AND NOBREGA

Figure 6 Simulated interface position for the AC/acetone/water system for different
solvent/nonsolvent coagulation bath compositions. Lo ⫽ 200 ␮m. (⫹) 0/100, Ts ⫽ 26.7 s;
(‚) 16/84, Ts ⫽ 128 s; and (夹) 36/64, no precipitation detected. Polymer casting solution:
AC/acetone (10/90).

regarded as instantaneous in this case, confirm- When Figures 5 and 7 and Figures 6 and 8 are
ing experimental results.19,20 This occurred be- compared, one can see that the AC and PEI mem-
cause the ternary phase diagram of PEI/NMP/ branes produced in the conditions analyzed are
water presented a very narrow solubility range, expected to present different morphologies. AC
as shown in Figure 3, so that initial feed condi- membranes are expected to be denser and to
tions were always very close to the binodal curve. present much more heterogeneous sublayers,
which can be confirmed experimentally.19,20

Table V Ts for the PEI/NMP/Water System as a

Function of the Precipitation Bath Composition Influence of the Thickness of the Initial Casting
Solution
Precipitation Bath
Composition ␾ (%) If x is divided by L and t is multiplied by D/L2 in
Ts eqs. (1–14), the model becomes dimensionless. In
␾(1,2) ␾(2,2) (s) this case, model solutions obtained are shown to
be similar when the boundary conditions are sim-
0 100 0.7
ilar in the scaled variables x and t, as shown by
16 84 0.9
36 64 1.5
Barton et al.21 (It is assumed in this case that the
dimensions of the coagulation bath can be scaled
Polymer casting solution: PEI/NMP ⫽ 10/90. in a similar manner; otherwise, solutions will not
 MODELING AND SIMULATION OF PHASE INVERSION 3045

Figure 7 Simulated polymer concentration profiles at precipitation for the PEI/NMP/

water system for different solvent/nonsolvent coagulation bath compositions. Lo ⫽ 200
␮m. (⫹) 0/100, Ts ⫽ 0.7 s; (‚) 16/84, Ts ⫽ 0.9 s; and (夹) 36/64, Ts ⫽ 1.5 s.

be equivalent.) This does not mean that resulting time, and the excess of the casting solution is
membrane characteristics would be the same for discarded (and possibly recovered and recycled).
three main reasons. First, membrane perfor- For this reason, the comparison of profiles at sim-
mance will depend on the absolute values of the ilar times may be useful and appropriate. Third,
skin dimensions and properties. For instance, after the formation of the first film of precipitated
when similar scaled membranes of different polymer, the model becomes invalid, as the poly-
thickness are used at actual separation processes, meric phase splits into two different polymer
permeability and selectivity are not necessarily phases. For scaling to be preserved, model formu-
the same. Permeability is expected to decrease, lation for this three-phase system must be equiv-
and selectivity, probably, is expected to increase alent to the model formulation used in the case of
as the membrane thickness increases. Second, two-phase system, which must not be taken for
during actual membrane preparation, it is not granted given the complex nature of the addi-
possible to monitor and control the position x of tional polymer phase. This was also acknowl-
the polymer precipitation front so that similar edged by Barton et al.21 To the best of our knowl-
membranes may be obtained. Actually, mem- edge, attempts to model the three-phase system
brane preparation is usually carried out in ma- have not yet been made.
chines where the extrusion velocity (and so, the To investigate the influence of Lo on polymer
coagulation time) is kept between well-defined concentration profiles and casting solution con-
values. Therefore, time and spatial scaling are traction rates, simulations with different Lo’s
not preserved in actual preparation conditions. were performed for the AC/acetone/water sys-
The usual procedure controls the coagulation tem. Typical values used for membrane prepa-
3046 FERNANDES, PINTO, AND NOBREGA

Figure 8 Simulated interface position for the PEI/NMP/water system as a function of

the quenching time for different solvent/nonsolvent coagulation bath compositions. Lo
⫽ 200 ␮m. (⫹) 0/100, Ts ⫽ 0.7 s; (‚) 16/84, Ts ⫽ 0.9 s; and (夹) 36/64, Ts ⫽ 1.5 s.

rations were assumed as 200, 500, and 1000 from the interface, gradients were larger for the
␮m. The casting solution contained 10% poly- thinner casting solutions. This means that thin-
mer and 90% solvent (vol %), whereas the coag- ner skins and more uniform sublayers may be
ulation bath was formed by pure water. As dis- produced when the initial casting solution is thin-
cussed in the previous paragraph, to allow a ner. Figure 10 shows that the precipitation dy-
direct comparison of the results, concentration namics were also much slower when the casting
profiles are presented for similar contact times. solution was thicker due to the same reasons pre-
It was assumed here that the coagulation bath sented before.
was much thicker than the casting solution (at Figure 11 shows the dynamic evolution of the
least 50 times thicker); otherwise, the amounts polymer concentration at the interface in the con-
of solvent removed from the casting solution ditions analyzed. Due to the larger amounts of
would exert a great influence on the precipita- solvent available for mass exchange, the thicker
tion conditions, as shown before. Results are casting solutions led to lower polymer concentra-
presented in Figures 9 –11. tions at equilibrium conditions. This means that
Figure 9 shows polymer concentration profiles denser skins may be produced when the casting
after a contact time of 15 s. Polymer concentra- film is thin. Interfacial equilibrium was reached
tion gradients were much larger (in dimension- after about 0.3– 0.5 s in all cases. In all of the
less scale) for thicker casting solutions because a simulations studied, it took approximately 0.3–
considerable amount of the initial feed behaved as 1.0 s for equilibrium to be attained. Therefore,
a reservoir of solvent. However, when the abso- thermodynamic equilibrium between the bulk co-
lute scale was analyzed in terms of the distance agulation bath and the interface was not attained
 MODELING AND SIMULATION OF PHASE INVERSION 3047

Figure 9 Simulated polymer concentration profiles for the AC/acetone/water system

for different Lo’s at t ⫽ 15 s: (⫹) Lo ⫽ 200 ␮m, (‚) Lo ⫽ 500 ␮m, and (夹) Lo ⫽ 1000 ␮m.
(a) Dimensionless position and (b) distance from interface.

instantaneously, as assumed by Reuvers et al.,13 ning of the phase-inversion process. As the time
and the mass transfer resistance may have lag to attain more stable mass transfer conditions
played an important role during the very begin- may be of the order of magnitude of the precipi-
3048 FERNANDES, PINTO, AND NOBREGA

Figure 10 Simulated interface position for the AC/acetone/water system as a function

of the quenching time for different Lo’s at t ⫽ 15 s: (⫹) Lo ⫽ 200 ␮m, (‚) Lo ⫽ 500 ␮m,
and (夹) Lo ⫽ 1000 ␮m.

tation time, precipitation may occur very far from ing water concentration, the initial polymer cast-
equilibrium for more incompatible systems. ing solution composition came close to the binodal
curve. Table VI also presents the experimental
precipitation time obtained by Reuvers and Smol-
Addition of Nonsolvent to the Casting Solutions
ders2 for the AC/acetone/water system. The re-
To investigate the influence of the addition of sults are similar and indicate that the time lag for
nonsolvent to the initial casting solution on the precipitation decreases as water is added to the
polymer concentration profiles and casting solu- casting polymer solution. Once more, the simula-
tion contraction rates, simulations with different tion results also indicate that the model was able
initial casting solution concentrations were per- to capture the most important dynamic features
formed for the AC/acetone/water system. The ini- of the phase-inversion process.
tial polymer concentration was set to 10% (vol %), Figure 12 shows that the increase of the non-
and different polymer/solvent/nonsolvent feed ra- solvent concentration of the casting solution led
tios were analyzed. The coagulation bath was as- to an increase in the polymer concentration at
sumed to contain pure water. Results are pre- the interface and to an increase in the concen-
sented in Table VI and Figure 12. tration gradients inside the casting solution
Table VI shows that the precipitation time de- during precipitation. Both conditions favor the
creased with the increase of water concentration formation of denser and thinner membranes.
in the polymer casting solution. This expected Besides, the precipitation time was reduced
behavior was related to the fact that with increas- very significantly when small amounts of non-
 MODELING AND SIMULATION OF PHASE INVERSION 3049

Figure 11 Simulated interface polymer concentration as a function of the quenching

time for the AC/acetone/water system for different Lo’s: (⫹) Lo ⫽ 200 ␮m, (‚) Lo ⫽ 500
␮m, and (夹) Lo ⫽ 1000 ␮m.

solvent were added to the casting solution. All CONCLUSIONS

these results agree very well with the experi-
mental data presented by Tsay and McHugh3 A simple mass transfer model was developed
with regard to the phase inversion of AC/ace- and implemented to simulate the phase-inver-
tone/water systems. sion process, which takes place during mem-

Table VI Ts for the AC/Acetone/Water System as a Function of the Water Content

Polymer Casting Solution Composition ␾ (%)

Experimental
␾(3,1)/␾(1,1)/␾(2,1) Free Polymer Base Ts Precipitation Timea
(AC/Acetone/Water) (Acetone/Water) (s) (s)

10/90/0 100/00 26.7 26

10/81/9 90/10 12.1 18
10/80/10 88.75/11.25 7.1 —
10/78.75/11.25 87.5/12.5 4.1 0.0
10/76.5/13.5 85/15 0.9 0.0

Precipitation bath: water.

a
Reuvers and Smolders.2
3050 FERNANDES, PINTO, AND NOBREGA

Figure 12 Simulated polymer concentration profiles at precipitation for the AC/

acetone/water system for different solvent/nonsolvent casting solution compositions. Lo
⫽ 200 ␮m. (⫹) 10/90/00, Ts ⫽ 26.7 s; (‚) 10/80/10, Ts ⫽ 7.1 s; and (夹) 10/76.5/13.5, Ts
⫽ 0.9 s. Precipitation bath: water.

brane formation. The model depends on a small the precipitation lag may be of the order of magni-
number of physical parameters, which are gen- tude of the time needed for the system to attain a
erally available or may be estimated from the dynamic equilibrium condition, so the assumption
literature. The model is able to reproduce both of instantaneous equilibrium between the interface
quantitatively and qualitatively experimental and the bulk solution may lead to large model de-
results available in published material regard- viations in these cases. Finally, it was shown for
ing the morphological aspects of the membranes both polymeric systems investigated in this work
prepared and the precipitation conditions of the that the dynamics of the mass transfer process
polymer material for the AC/acetone/water and seem to be much more important than the influence
PEI/NMP/water systems. Simulations indicate of the concentration on the diffusion and thermody-
that (1) the addition of solvent to the coagulation namic partition coefficients, as fair agreement with
bath leads to less dense and thicker skin layers in
available experimental data was obtained even
the polymeric solution before precipitation and
when these coefficients were assumed to be con-
longer time lags for precipitation, (2) the addition of
stant.
nonsolvent to the casting solution causes the oppo-
site effects, and (3) the reduction of the thickness of
the initial casting solutions may cause densification The authors thank Conselho Nacional de Desenvolvi-
of the skin and the formation of denser membranes. mento Cientı́fico e Tecnológico for scholarships and
For incompatible systems such as PEI/NMP/water, support of this research.
 MODELING AND SIMULATION OF PHASE INVERSION 3051

NOMENCLATURE REFERENCES

Abbreviations
1. Reuvers, A. J. Ph.D. Thesis, Twente University of
Technology, Twente, The Netherlands, 1987.
AC cellulose acetate
2. Reuvers, A. J.; Smolders, C. A. J Membr Sci 1987,
NC number of chemical species
34, 67.
NMP N-methylpyrrolidone 3. Tsay, C. S.; McHugh, A. J. J Polym Sci Part B:
PEI polyetherimide Polym Phys 1990, 28, 1327.
4. Nobel, R. D.; Stern, S. A. Membrane Separations
Symbols Technology—Principles and Applications; Elsevier
Science: Amsterdam, 1995.
A mass transfer area (m2) 5. McHugh, A. J.; Tsay, C. S. J Membr Sci 1991, 64,
C(i,j) molar concentration of component i in 81.
phase j (kg/m3) 6. Castellari, C.; Ottani, S. J Membr Sci 1981, 9, 29.
C*(i, j) mass concentration of component i in phase 7. Cheng, L. P.; Soh, Y. S.; Dwan, A. H.; Gryte, C. C.
j (mol/m3) J Polym Sci Part B: Polym Phys 1994, 32, 1413.
D(i,j) diffusion coefficient of component i in 8. Cheng, L. P.; Lin, D. J.; Shih, C. H.; Dwan, A. H.;
phase j (m2/s) Gryte, C. C. J Polym Sci Part B: Polym Phys 1999,
k(i) partition coefficient of component i (dimen- 37, 2079.
sionless) 9. Termonia, Y. J Membr Sci 1995, 104, 173.
L casting solution thickness at time t (m) 10. Tsay, C. S.; McHugh, A. J. J Polym Sci Part B:
Lo initial casting solution thickness (m) Polym Phys 1991, 29, 1261.
M(3,1) total polymer mass in polymer solution (kg) 11. McHugh, A. J.; Tsay, C. S.; Barton, B. F.; Reeve,
J. L. J Polym Sci Part B: Polym Phys 1995, 33,
Pm(i) molecular weight of component i (kg/kmol)
2179.
Ts simulated precipitation time (s)
12. Radovanovic, P.; Thiel, S. W.; Hwang, S. T. J
x mass transfer direction (m)
Membr Sci 1992, 65, 213.
x(0) initial interface position (m)—Figure 1 13. Reuvers, A. J.; van den Berg, J. W. A.; Smolders,
x(t) interface position at time t (m)—Figure 1 C. A. J Membr Sci 1987, 34, 45.
14. Shojaie, S. S.; Krantz, W. B.; Greenberg, A. R. J
Greek Membr Sci 1994, 94, 255.
15. Brandup, J.; Immergut, E. H. Polymer Handbook,
␾(i,j) volume fraction of component i in phase j 3rd ed.; Wiley: New York, 1989.
(dimensionless) 16. Reid, R. C.; Prausnitz, J. M.; Poling, B. E. The
␳(i,j) density of component i in phase j (kg/m3) Properties of Gases and Liquids, 4th ed.; McGraw-
␳(i) density of pure component i (Kg/m3) Hill: New York, 1987.
17. Petzold, L. R. DASSL Code, version 1989; Comput-
Index (i,j) ing and Mathematics Research Division, Lawrence
Livermore National Laboratory: Livermore, CA,
First Index (i) 1989.
18. Furtado, C. J. A. M.S. Thesis (in Portuguese), PEQ/
1. Solvent COPPE/UFRJ, Rio de Janeiro, Brazil, 1993.
2. Nonsolvent 19. Wang, D. L.; Li, K.; Teo, W. K. J Appl Polym Sci
3. Polymer 1999, 71, 1789.
20. Roesink, E. Ph.D. Dissertation, University of
Second Index (j) Twente, Twente, Enschede, The Netherlands,
1989.
1. Polymer solution phase 21. Barton, B. F.; Reeve, J. L.; McHugh, A. J. J Polym
2. Precipitation bath phase Sci Part B: Polym Phys 1997, 35, 569.