A Generalized Solution-Diffusion Model of The Pervaporation Part II PDF

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 17

ELSEVIER Journal of Membrane Science 89 ( 1994) 153- 169

A generalized solution-diffusion model of the pervaporation


process through composite membranes
Part II. Concentration polarization, coupled diffusion and the
influence of the porous support layer
A. Heintz*, W. Stephan
UniversitSit Heidelberg, Physikalisch-Chemisches Institut. Im Neuenheimer Feld 253, D-69120 Heidelberg, Germany
(Received May 20, 1993; accepted in revised form December 8, 1993)

Abstract

The central part of the model is based on the solution-diffusion process which describes fluxes of low molecular
weight components in the dense active layer. Using the Maxwell-Stefan theory, coupled equations for the diffusive
fluxes through the dense layer are derived. The concentrations on the phase boundaries inside the membrane are
determined by solubility equilibria with the adjacent phases of the feed and permeate phase, respectively. They are
calculated using the UNIQUAC model as described in Part I of this series. Fluxes through the porous sublayer are
calculated by a general kinetic gas flow model which includes Knudsen flow and laminar flow as special cases. The
solution-diffusion model and the pore flow model are combined resulting in a resistance-in-series model for fluxes
through the dense layer and the support layer.
The generalized model is applied to the pervaporation of six aqueous/organic mixtures through the standard
poly (vinyl alcohol) /poly (acrylonitrile ) (PVA/PAN) composite membrane. It turns out that the strong non-ideal
solubility behavior of the liquid mixture components in PVA and diffusive coupling effects play the dominant
roles in the pervaporation process. More simplified models which do not account for these physico-chemical ef-
fects fail to describe fluxes and separation diagrams in PVAIPAN membranes. Introducing the influence of the
porous support layer leads to almost complete agreement with experimental data even at elevated permeate
pressures.
The influence of concentration polarization has also been studied in some detail and is found to be of minor
importance compared to the other contributions.

Key words: Transport theory; pervaporation; Maxwell-Stefan theory; Pore flow model; Poly(viny1 alcohol) membrane

1. Introduction through dense membranes [ l-5 1. In the case of


composite membranes the porous sublayer sup-
It has been known for many years that the so- porting the dense active layer on the down-
called solution-diffusion model is suitable for stream side may exert an additional effect on the
describing transport processes in pervaporation fluxes and separation potential. This has been
pointed out by several authors [ 3,6,7 1. Further-
*Corresponding author. more, concentration polarization in front of the

0376-7388/94/$07.00 0 1994 Elsevier Science B.V. All rights reserved


SSDIO376-7388(93) E0223-7
154 A. Heintz, W. Stephan /Journal ofMembrane Science 89 (1994) 153-169

feed side of the active layer can influence the eous flow caused by the driving force of a pres-
fluxes and separation characteristics under spe- sure gradient P - Pp. Pp is the permeate pressure.
cial conditions [ 1,6,8-lo]. There is no phase change when the components
In this paper we develop a generalized model leave the support layer entering the permeate
which accounts for these contributions simulta- space with the weight fraction wip (mole fraction
neously. The importance of each contribution is xiP)-

discussed quantitatively. Whenever it is possible In the following sections the model is devel-
and justified by the results, simplifications and/ oped stepwise and tested for increasing levels of
or assumptions have been made in order to re- complexity by comparison with experimental
duce the complexity of the calculation procedure. data.
The composite membrane model is illustrated
in Fig. 1. Generally, a system consisting of three
layers is considered. On the left side in the liquid 2. Solution-diffusion model and the influence of
feed mixture the composition is given by the concentration polarization
weight fractions wjF (or mole fraction xii?) where
i indicates that component i permeating the We consider the flux Ji through the active layer
composite membrane. For each component it of the composite membrane. If diffusive cou-
holds that the molar flux Ji is constant and has pling is neglected, the fluxes J, are independent
the same value at any location in the composite from each other and it holds for J, according to
membrane under steady state conditions. The the solution-diffusion model:
layer in front of the active layer is the diffusive
J.=~.di”(,. .$, _S .w? )
(1)
laminar layer of the feed mixture which might
give rise to concentration polarization resulting
’ J,f* 6” lF IF lP IP
in weight fractions “t, of the liquid mixture at where p is the membrane density, Mi and DiM are
the boundary to the membrane which differ from the molar mass of component i and its averaged
the bulk weight fractions wLF. diffusion coefficient in the active layer,
In the framework of the solution-diffusion respectively.
model it is assumed that thermodynamic solu- DijiMis defined by
bility equilibrium exists between the liquid phase %F. SF
outside the membrane and the swollen mem- DiM( w:)dw:
F
brane phase adjacent to the phase boundary. As *J
a consequence there exists equality of the ther- DiM= ;;;f:sjF
- wTp .SiP) (2)
modynamic activity Ui at the phase boundaries
of the liquid/membrane (CZiF=a iF) . SiF=w&/$;F and Sjp are the solubility coeffl-
ei)iF (or wjF if no concentration polarization oc- cients at the feed side and the permeate side of
curs) and w:F, the weight fraction inside the ac- the active layer, respectively. SiF and Sip are
tive membrane layer at the feed side boundary, quantitative measures for the solubility equilib-
are connected by the phase equilibrium condi- rium of component i between the mixture phase
tion. Inside the active layer the components move and the membrane phase of the active layer ma-
by diffusion according to the gradients of their terial. At a given Weight fraction tt;F (or w:p ) in
weight fractions w:. the liquid feed (or permeate) mixture, w:F (or
At the boundary to the pore space of the po- wip ) can be calculated if SiF (or Sip) is known
rous support layer, again the existence of solubil- from experimental solubility data or from theo-
ity equilibrium is assumed between w& (inside retical model calculations. In this paper S, and
the active layer) and w$ (outside the active Sip are calculated using the UNIQUAC model for
layer). x7 is the mole fraction corresponding to all mixtures studied. Details of calculating S, and
the weight fraction wTP.The flux through the po- S,, at given $iF and wTp are described in part I
res of the support layer is considered to be a gas- [Ill.
A. Heintz, W. Stephan /Journal ofMembrane Science 89 (1994) 153-169 155

feed laminar active support permeate


layer layer layer

PP
aiF

aiP

WiF
wiP
(XiF)
@iPI

i I
=Q I I
Y
pressure

K
Pp 1
pore length

Fig. 1. Concentrations and pressures in the pervaporation process through composite membranes. w,&xiF): weight fractions
(mole fractions) in the feed mixture; titiF:weight fraction in the feed mixture adjacent to the active membrane layer; w&: weight
fraction inside the active membrane layer facing the feed mixture; w&: weight fraction inside the active membrane layer facing
the support layer; xt, P*: mole fraction and total pressure at the upstream side of the porous layer; ( w~~(x,~): weight fraction
(mole fraction) in the gaseous permeate identical to the values at the downstream side of the porous layer; Pp: total permeate
pressure.

For the following discussion of concentration The theory of concentration polarization pro-
polarization we neglect the second term in Eq. vides a functional connection between iCZFand
( 1 ), i.e., S’ip*w$ =O. This simplification makes WiFin the feed mixture. For binary mixtures the
the discussion easier and it is also justified by the basic equation for concentration polarization in
fact that w& is usually close to zero in pervapor- non-electrolyte fluids (liquids or gases) is given
ation since the permeate pressure Pp (and as a by [ 1,10,12]:
consequence also P*) is kept as low as possible I’, -J1 + V, . J2 @,F(z=#)--7,
in real situations. Because the effect of concen- ew (3)
( k > @IF(Z=O) -.&
=
tration polarization decreases at higher per-
meate pressures, the discussion at Sip* w*, =O is with k=D12/d’, the feed side mass transfer coef-
sufficient in order to estimate the maximal effect. ficient, and
156 A. Heintz, W. Stephan /JournalofMembrane Science89 (1994) 153-169

constant values for DiM adjusted to the fluxes of


XlF * K
@IF= l-&= (4) the pure organic components through the PVA/
X1F’ VI +kF’ v, PAN membrane were used in the model calcula-
the volume fraction of 1 at z=8’ and z=O, re- tions rather than the experimental data taken
spectively (see Fig. 1) . D1 2 is the binary diffu- from the kinetic vapor sorption isotherms. 6”, p
sion coefficient in the feed mixture and V, and and DiM are taken from the data listed in Tables
V, are the molar volumes of component 1 and 2 1 (left column ) and 2.
in the feed mixture. VIJ1+ V2J2 is the total vol- For the example ethanol/water and the PVA/
ume flow which has the meaning of an averaged PAN composite membrane we calculated the
velocity. J1 is defined as fluxes Ji using Eq. (6) with GiF obtained from
Eq. (3) at a given feed composition w,F. The S,,
~~=v,.J,/(v,.J,+v*.J,)=l-~~ (5) values were calculated according to the proce-
dure described in Part I [ 111 and the Di, values
With ,!&* w*, = 0, Eq. ( 1) can be rewritten as for water in PVA were obtained from Eq. (2 ) us-
ing experimental data [ 13 1.
(6) The results for the partial mass fluxes jHzO =
J Hz0 %H20 and JEtOH-MElOHare shown in Fig. 2
By substituting Eq. ( 5 ) into (6 ) and Eq. (6 ) into for the cases k=m (no concentration polariza-
(3 ) we obtain an equation which allows to deter- tion, $LF= WiF) and k= lop3 (cm/s). This latter
mine @rF(z=#) at agiven @~F(z=O) provided value for k was chosen because it corresponds to
the DiM, SiF and p Values, as Well as 6” and k are the hydrodynamic situation in a plate and frame
known. module [ lo] used in the equipment for measur-
Furthermore, we have made the assumptions ingjHzO and&OH [ 1 4 1.

that @IF(z=S~)~CTJ~F and @~F(z=O)~W~F From Fig. 2 two conclusions can be drawn.
which means that the volume fractions have been First, the solution-diffusion model in the sim-
replaced by the corresponding weight fractions. plified form of Eq. ( 1) with uncoupled diffusion
Eq. (3) makes clear that $I,?% wlF is only ob- is not able to describe the experimental data of
tained if the exponent on the left side of Eq. ( 3 ) the fluxes adequately. Secondly, the effect of
is close to zero which is the case when DiM.SiF/ concentration polarization is small. This is dem-
da-pi is small compared to k. onstrated in more detail in Fig. 3. The flux ratio
S,= at a given & can be predicted by the UNI- ji( k) /ji( k= m) is larger than 90% for water and
QUAC model. The calculation procedure is de- 95% for ethanol in the k region of plate and frame
scribed in detail in Part I [ 111. DiM are averaged modules for concentrations which are of practi-
diffusion coefficients according to Eq. (2 ) . They cal interest for dewatering organic solvents.
have been calculated using the concentration-de- Fig. 3 reveals that the flux of ethanol, the com-
pendent coefftcients Di ( w: ) of the pure compo- ponent with the smaller permeability, is de-
nents in the membrane material of the active creased by concentration polarization as well as
layer. Data of water at 333 K are available from the water flux. Usually a higher flux of ethanol
time-dependent vapor sorption experiments with would be expected since &on,,> WE~OH,Fbut this
flat polymer sheets by measuring the mass up- is overcompensated by the fact that the concen-
take of pure solvent vapor as function of time and tration-dependent solubility coefficient S,,,,,,
analyzing the data on the basis of Fick’s second leads to lower solubilities of ethanol in the PVA
law of diffusion [ 13 1. The diffusion coefficient membrane when the weight fraction of ethanol
of water used in the integral of Eq. (2) for cal- in the liquid mixture increases in this range of
culating &o depends strongly on the concen- liquid mixture concentration. This is due to the
tration passing a maximum. The experimental strong synergistic solubility effect of ethanol in
data of the diffusion coefficients of the organic PVA described in detail in Part I of this series
components are not very accurate and therefore ill].
A. He&z. W. Stephan /Journal ofMembrane Science 89 (1994) 153-169 157

Table 1
Diffusion coefftcients and Maxwell-Stefan interaction parameters

6,x-10”” &r(~‘H20)-10”b L& (HZO/organic)-10”


(cm’/s) (cm’/s) (cm*/s)

Methanol 223 100+exp(-~I~25)~125 20


Ethanol 30 lO+ (W’,/0.36)‘-50 25
nPropano1 20 lO+ (m’, /0.36)‘-30 25
i-Propanol 20 lO+ (~;/0.36)‘-30 40
Dioxane 30 10+(~;/0.36)2~15 0.1
THF 25 lO+ (W’,/0.36)2*20 0.1

‘Constant values adjusted to the fluxes of the pure organic compounds (used in the calculations shown in Fig. 4).
bAdjusted to the experimental partial fluxes of the mixtures depending on the mean weight fraction of water in PVA:
*; = (whzo.P+ WL2O.P) /2 (used in the calculations shown in Fig. 5 and Figs. 7- 12 ) .

Table 2
1 .O
Membrane specific parameters for the PVA/PAN composite
membrane

(a) Active PVA layer: (b ) Porous PAN support 0.8


layer
F
p= 1.2 g/cm3 d,,,,=1.6-lo-sm II 0.6
6”=1.3*10-‘m 6P=1*10-4m 5
N=7 P
Y 0.4

k (cm/s)
Fig. 3. Ratio J(k) / J( k= co) calculated for the pervapora-
tion of a mixture water/ethanol at 333 K and different weight
fractions of water in the feed. w,,,,,,= 0.05: ( n ) water, ( A )
ethanol; w,,,.~_~= 0.01: (0 ) water, (A ) ethanol.

Since the effect of concentration polarization


is small we neglect this effect in the following
sections where increasingly more complicated
models are developed and tested which account
for diffusive coupling in the active layer and the
effect of the porous support layer.

3. Solution-diffusion model and the influence of


0.0 0.2 0.4 0.6 0.8 1.0 coupled diffusion according to the Maxwell-
wzFof ethanol in feed Stefan theory

Fig. 2. Partial fluxes for the system water/ethanol/PVA at In this section the solution-diffusion model in
333 K. Experiments [ 141: (0) water, (A ) ethanol, Pp=20 the active layer is extended by introducing dif-
mbar; calculation: (---) k=cq (-) k= lo-’ (cm/s). fusion coupling of the permeating components
158 A. Heintz, W. Stephan /Journal ofMembrane Science 89 (1994) 153-169

using a modified Maxwell-Stefan approach. This


approach is described in the Appendix.
The equations of coupled diffusion [Eqs.
(A 15 ) and (A 16 ) ] can be rearranged by divid- D,,\?I;
ing Eq. (A16) by (A15):
Aw; +1>,M
D12 +w’, D,, +\?I;&., >-
~/(PM,)-Aw; (9)

tw;.&M +42) -D,,-h


(7)
D,,\?I;
Aw; +&,
D,2 +jcl;D,, +‘%‘;D,, >’

p/(S”.M2)-Aw; (10)
This differential equation can be solved for given
permeate pressure, values of w:, and ratios of J2/
J, providing values of wiP and finally Wipe The
DiM in Eqs. (9) and ( 10) are averaged diffu-
integration has been performed numerically by sion coefficients for the pure components as de-
the following procedure. At a given composition fined by Eq. (2 ) with W:r = WjF’sir. The concen-
W$, W:F= &!?;r* WiFcan be calculated according to tration polarization is neglected ( WiF= Gir) and
the UNIQUAC model. At a given permeate pres- D12 has to be adjusted to the experimental par-
sure Pp the unknown weight fractions in the per- tial fluxes. Eqs. (9 ) and ( 10) represent non-lin-
meate wip have to be estimated in a first attempt. ear equations for w’,~ and wip at given values of
Using the UNIQUAC model for gaseous mixture W;F and wiF. Solving Eqs. (9) and (10) is much
solubilities, the corresponding equilibrium more convenient than Eq. ( 7 ). Approximate so-
weight fractions in the membrane w’,~ and wip lutions for w ‘,p and w ip deviate less than 5% from
are obtained. The partial fluxes are connected the values obtained by the exact solution of Eq.
with the permeate weight fraction wip by (7) in all cases. In order to obtain consistent re-
sults for the permeate composition ( wlp, wzp) at
J, -MI 1
(8) a given permeate pressure, the iteration process
W1P=J1441 +J2.M2=l+J2M2/J,M,
has to be repeated until the Wipvalues fulfill Eq.
wip defines the ratio J,M,/J,M, to be substi- ( 8 ) and are simultaneously in equilibrium with
tuted into Eq. (7 ). Numerical integration of Eq. the wip values according to the UNIQUAC cal-
( 7 ) provides values of w’,~ and wip which will culations. The effect of diffusion coupling is
differ from the first estimate and, consequently, demonstrated in Fig. 4.
from these values new values for wlp and J2M2/ The solid lines represent the model predic-
J&f, are obtained. Solving Eq. ( 7 ) again gives a tions with coupled diffusion and the broken lines
third set of solution for wip. The iterative proce- without coupling. Obviously the introduction of
dure is repeated until two following sets of solu- diffusion coupling according to the modified
tion for wIp differ by less than 0.0 1%. Maxwell-Stefan theory leads to a decrease of the
Eq. (7 ) has to be solved in each iteration step. water flux. The ‘fast moving’ component water
This calculation method is tedious and time con- is slowed down by frictional interaction with the
suming. Eqs. (A15) and (A16) can also be ‘slow moving’ organic compound. The flux of the
solved by an approximation which deals with organic compound is slightly increased because
difference equations rather than differential the slow moving organic compound is speed up
equations. Therefore, Eqs. (Al 5) and (Al 6) can by the fast moving water. Both fluxes are shifted
be rewritten as: into the direction required for an improved pre-
A. Heintz, W. Stephan /Journal ofMembrane Science 89 (1994) 153-169 159

0.0 0.2 , 0.4 0.6


w of ethanol in feed
0.8
I
1.0

iP

Fig. 4. Partial fluxes for the system water/ethanol/PVA at 0.0 0.2 0.4 0.6 0.8 1.0
333 K. Experiments [ 141: (0 ) water, (A ) ethanol, WF of ethonol in feed
Pr= mbar; calculation: (---) uncoupled diffusion, (-) cou-
Fig. 5. Partial fluxes for the system water/ethanol/PVA at
pled diffusion.
333 K. Experiments [ 141: (0) water, (A ) ethanol, P,=20
mbar; ( n ) water, ( A ) ethanol, Pp= 100 mbar; (0 ) water,
diction. This was achieved by introducing the (0 ) ethanol, Pp= 200 mbar. Calculation: coupled diffusion,
additional adjustable parameter D12 which is a diffusion coefficient of ethanol adjusted to experimental par-
measure for the mutual friction effect of water tial flux data (see text); (-) P,=20, (----) Pp= 100, (--
and the organic compound. Parameters used for - -) Pp= 200 mbar.
all systems studied are listed in Tables 1 and
At low permeate pressures (20-30 mbar) the
2(a).
In order to obtain better agreement with the model predictions using coupled diffusion and
experimental data, it is necessary to make the av- adjusted diffusion coefficients of the organic
eraged diffusion coefficients of the organic com- components in the dense layer are satisfactory.
pounds 6,M dependent on the average water con- However, at higher permeate pressures ( 100-200
centration in the PVA layer. In Table 1 6, is mbar) significant deviations of the calculated
given as function of WHzo, the average weight fluxes from the experimental data are observed
fraction of water in the membrane. The numeri- (Fig. 5 ). In all cases the predicted water flux was
cal parameters of these functions have been ad- to high at higher permeate pressures. It turned
justed to partial flux data. As an example for the out that a further improvement could only be
obtained improvement Fig. 5 shows the calcu- achieved by taking into account the influence of
lated partial fluxes for the system water-ethanol the membrane porous support layer which acts
at three different permeate pressures with diffu- essentially as a flow resistance.
sion coupling and the diffusion coefficient of
ethanol adjusted to the experimental partial flux
over the whole range of composition. The same 4. The influence of the porous support layer on
D,z value has been used as in Fig. 4. the fluxes and separation diagrams
These results show that diffusion in multi
component mixtures depends not only on the The porous support layer of a composite mem-
coupling parameter Q2 but also influences to brane is usually 50-1000 times thicker than the
some extent the diffusion coefficients of the in- dense, active layer. The porous.support acts es-
dividual components. sentially as a mechanical stabilization, but it can
160 A. Heintz, W. Stephan /Journal ofMembrane Science 89 (1994) 153-169

also have an effect on the separation character- starting with the Maxwell-Stefan equations as
istics of the membrane. After leaving the thin discussed in the Appendix. The following equa-
dense active membrane layer, the components tions are derived for the molar fluxes J, and J2:
enter the porous support layer as a gaseous mix-
ture with a certain mole fraction xl at pressure 4x1 ‘P)
dz
P* as illustrated in Fig. 1. The driving force for

1 (12)
the flux through the pores is the pressure differ-
+L-B ‘XI .P dP
----+x, -A-P%
ence between P and the permeate pressure Pp.
l+B.P dz
The gaseous composition Xip inside the porous
layer may also differ from XT. d((l-x,)-P)
The support layer is assumed to consist of a dz
bundle of pores with different diameters d. The

1
pore distribution can be described by the follow- +s,-B. (1 -xl).P dP
sz+ ( 1 -x, )-A-P$
ing function according to Rosin, Rammler and 1 +B*P
Sperling [ 15 ] : (13)
Eqs. ( 12 ) and ( 13 ) were first derived by Present
and de Bethune [ 17 ] on the basis of the kinetic
theory of gases. The parameters A, B, b, a, and
where Y(d) is the probability of finding a pore f, are defined in the Appendix. They include the
of diameter d statistically weighted by the area molecular parameters which are listed in Table
rr- (d/2)2. The parameters N and d,,, were ad- 3. The general condition of the steady state re-
justed to the experimental pore size distribution quires that
data [ 16 1. They are listed in Table 2b. The pore
size distribution function used in the model cal- -- Jl J1
-- XIP
(14)
culations is shown in Fig. 6. x’p-J, +J2 Or JZ-l-x,p
The flow of a gaseous mixture through a single
A solution of Eqs. ( 12) and ( 13) is obtained
pore of a given pore diameter d can be treated
when Eq. ( 12) is divided by Eq. ( 13 ). This leads
to the following differential equation:

4’oE+081 h(x,,).P.$+ [h(X,p) +c, .p+c, .P2] ‘2,

+(cuO--l)~x~p~(l-x,p)=O (15)
with
h(XIp)=l+(Q)o-l)*(l -x1p) (16)
X1 =x1 +x1p (17)
c1 =f;B+A (18)
c2 =A.B (19)
A solution of Eq. ( 15 ) gives x, as a function
,i - _,_
of P inside the pore at given values P=P, and
O.OE+OO -
0 10 20 30 40 x1 =xIp at the right-hand side of the pore (z= 6’,
pore diameter (nm) seeFig. 1) [16]:
~%-wx,P~w-X1P)_
Fig. 6. Pore size distribution function V(d) of the PAN sup- x, =x*p -
h(XlP).P
port layer.
A. Heintz, W. Stephan /Journal ofMembrane Science 89 (1994) 153-169 161

P Eq. (20) represents an approximate solution


accurate to within l-2% of the exact numerical
exp( -c”). exp(c’)dP (20)
I result. For x1 =xr and P=P, Eq. (20) is the so-
Pp lution of pressure and gaseous composition at the
with left-hand side of the pore (see Fig. 1).
The following steps have to be performed in
c, ~F+o.5~c*~F2 order to obtain a complete solution for a com-
c’ = (21)
h(xw) posite membrane consisting of an active layer and
C, ~P+o.5~C2~P2
a porous layer represented by a single pore hav-
C” = (22) ing a diameter d and length Sp.
h&P) (a) Since J, and J2 are functions of P only at
Table 3
a given feed composition according to the pro-
Parameters used for calculating gas mixture viscosities at 333 cedure described in the previous section, xIp is a
K 1151 function of P* at a given feed composition due
to Eq. (14).
elk P’ 0
(b) Ifthis functionx,p(P) is now substituted
(K) (nm) ~m’/mol)
into Eq. (20) with x1 =x; and P=P, x: is ob-
Water 775 1.0 0.252 18.4 tained as function of P* at a given permeate
Methanol 417 0.5 0.369 42.5 pressure Pp. xlp has been eliminated.
Ethanol 431 0.3 0.431 61 (c) A definite solution for x7 and P is ob-
n-Propanol 495 0.2 0.47 1 77.9 tained from calculating the total flux Jt,,t= J1 + J2.
i-Propanol 518 0.2 0.464 79.9
THF 362 0.08
Addition of Eqs. ( 12) and ( 13) and combina-
0.549 85.04
Dioxane 450 0.1 0.5 87.1 tion with Eq. ( 15 ) gives:

+f~.A(x’).P].g
I

(23)
Since x1 is a function of P at given values of xlp
and Pp [ Eq. (20) 1, Eq. (23) can be integrated:

*A(x,(P))-P
1
-dP (24)

Since Jtoton the left-hand side of Eq. (24) is a


definite function of P at a given feed composi-
tion as described in the previous section, Eq.
(24) provides a definite solution for P. As a
0.0 0.2 0.4 0.6 0.8 1 .o
W~F of methanol in feed consequence x7 has also been fixed now in a def-
inite way.
Fig. 7. Partial fluxes for the system water/methanol/PVA at
Eventually values of x:, PC and xlp are ob-
333 K. Experiments [ 141: (0 ) water, (A ) methanol, Pr= 30
mbar; ( n ) water, ( A ) methanol, Pp = 120 mbar. Calcula- tained at a given feed composition and permeate
tion: coupled diffusion including effect of support layer, (-) pressure Pp providing the complete mathemati-
P,=30, (-- ) Pp= 120 mbar. cal solution of the transport problem of binary
162 A. Heintz, W. Stephan /Journal of Membrane Science 89 (1994) 153-169

0.0 0.2 0.4 0.6 0.6


&
1.0
0.0 0.2
WIFof
0.4
i-propanol
0.6 0.6
in feed
4
1 .o
wiF of ethanol in feed
Fig. 9. Partial fluxes for the system water/i-propanol/PVA at
Fig. 8. Partial fluxes for the system water/ethanol/PVA at 333 K. Experiments [ 141: (0) water, (A) i-propanol,
333 K. Experiments [ 141: (0) water, (A ) ethanol, &.=20 Pr=30 mbar; (m) water, (A ) i-propanol, Pr= 130 mbar.
mbar; ( n ) water, (A ) ethanol, &.= 100 mbar; (0 ) water, Calculation: coupled diffusion including effect of support
(0 ) ethanol, Pr = 200 mbar. Calculation: coupled diffusion, layer; (-) P,=30, (-----) Pp= 130 mbar.
including effect of support layer, experimental partial flux data
(seetext); (-) P,=20, (--) P,=lOO, (----) P,=200
mbar.

mixtures permeating the active layer combined


with a pore flow in series at a given pore diame-
ter d. At this point it is necessary to account for
the fact that the real porous layer consists of a
bundle of pores rather than of a single pore. Us-
ing the pore distribution function of Eq. ( 11) for
calculating the molar fluxes Jr and .& averaged
over the whole surface of the composite mem-
brane one obtains
co
Ji(d)*Y(d).dd
I
.ig o3 (25)
P(d)-dd
s 0.0 y c
0
0.0 0.2 0.4 0.6 0.8 1.0
WIF of methanol in feed
JI and & can be calculated at a given feed com-
Fig. 10. Separation diagram for the system water/methanol/
position and a given total permeate pressure.
PVAat333K.Experiments[14]:(0)P,=3O,(~)P,=120
Figs. 7 to 9 show the results of the combined mbar. Calculations: coupled diffusion including effect of the
model for three selected examples of the six sys- support layer; (-) Pp= 30, (-----) Pr= 120 mbar. Upper
tems studied including coupled diffusion and the curve: (----) VLE, thermodynamic vapor liquid equilibrium.
A. He&z, W. Stephan /Journal ofMembrane Science 89 (1994) 153-169 163

0.8 0.8
0
s

i
a 0.6
.c

7
x
g 0.4
.-I
%
B
3 0.2

0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
WIF of ethanol in feed WF of i-proponol in feed

Fig. 11. Separation diagram for the system water/ethanol/ Fig. 12. Separation diagram for the system waterli-pro-
VAat333K.Experiments[14]:(Cl)Pr=20,(~)Pr=100, panol/PVA at 333 K. Experiments [ 141: (Cl) Pr=30, (m)
( 0 ) I+ 200 mbar. Calculations: coupled diffusion includ- Pr= 130 mbar. Calculations: coupled diffusion including ef-
ingeffect of the support layer; (-) Pr=20, (---) Pp= 100, fect of the support layer; (-) Pp = 30, (---) Pr.= 130 mbar.
(- - -) &.=200 mbar. Upper curve: (---) VLE, thermo- Upper curve: (----) VLE, thermodynamic vapor liquid
dynamic vapor liquid equilibrium. equilibrium.

influence of the porous support layer as de-


takes into account non-ideal multicomponent
scribed above. The diffusion coeffkients of the
solubility effects, non-ideal diffusivity behav-
organic components and coupling parameters
iour, diffusion coupling and the influence of the
have been used as presented in Table 1 (middle
porous support layer in order to describe semi-
and left column). The flow resistance of the po-
quantitatively the performance of a pervapora-
rous support layer causes a decrease in the pre-
tion membrane for a given separation problem.
dicted water flux especially at higher permeate
The effect of concentration polarization has also
pressures ( 100-200 mbar). It has to be pointed
be studied and found to be of minor importance.
out that no capillary condensation effects are ac-
Therefore it has been neglected. A few experi-
counted for in the theoretical treatment of the
mental data such as pure component vapour
flow through the porous layer. Separation dia-
sorption isotherms, pure component diffusion
grams of the three system examples are shown in
coefficients (both available from vapor sorption
Figs. 10 to 12 demonstrating the almost quanti-
experiments with the PVA membrane material)
tative agreement of the theoretical with the ex-
and vapor/liquid equilibrium data of the binary
perimental results. Similar quality of the results
mixture components are required for applying
is obtained for all other systems studied.
the model. The most simple calculation proce-
dure is based on the neglection of diffusion cou-
pling in the active layer and neglection of of the
5. Conclusions porous support layer. In this case the results are
unsatisfactory. Essential improvement of the
Pervaporation of mixtures through PVA/PAN model description can be achieved by introduc-
composite membranes is a complex process tion of diffusion coupling using the Maxwell-
which is influenced by several factors. Our model Stefan model with one additional adjustable pa-
164 A. Heintz, W. Stephan /Journal OfMembraneScience 89 (1994) 153-169

P
rameter. Introducing a flow resistance of the po-
rous support layer leads to further improvement +RTlnaj(T,Po)+JVidp (AZ)
particularly at higher permeate pressures. This PO
pore flow model requires experimental data
about the structure of the porous support layer. where py( T,P”) is the standard chemical po-
This extended model leads to a satisfying de- tential of pure component i in the ideal gas state,
scription of all systems investigated over wide A,,( T,P”) is the fugacity of the real pure fluid at
ranges of feed concentrations and permeate a standard pressure P” and temperature T. V, is
pressures. the partial molar volume of component i and P
With respect to practical applications it might the considered value of pressure. ai is the ther-
be sufficient to use a simple empirical expression modynamic activity defined by
for the flow resistance of the support layer rather
than the quite complicated pore flow model in (A3)
order to reduce the amount of computational
work. However, UNIQUAC model calculations whereJ( T,P”) is the fugacity of component i in
and accounting for diffusion coupling based on the mixture and yi is the corresponding activity
the Maxwell-Stefan theory is recommended coefficient.
whenever the solution-diffusion model is ap- For the special case of ideal gas mixtures, Eq.
plied to diffusive resistance in pervaporation (A2 ) can be written
membranes.
~i=~~(T,P”)+RTltio+RTl~~+RTlnP
P0
6. Appendix =pp(T,P“)+RTln(p.xi) (A4)
A generalized Maxwell-Stefan model for and for the case of condensed liquid mixtures one
transport processes of mixtures through dense obtains if liquid compressibility is neglected:
and porous membranes ~i=~~“(Typ”)+RTln(xi*yi)+~(p-po)
(A5)
Starting from the concept of irreversible ther-
modynamics, the following expression is ob- with
tained for the relation between driving forces and T,P”) +RTln&( T,P”)
~2” =pp( (Ah)
frictional resistances in a multicomponent
mixture: Usually the last term in Eq. (A5 ) can be ne-
glected at ambient pressure (pmp’). pgq is the
(Al 1 chemical potential of the pure liquid at p” and T.
In the cases where polymer substances are in-
Eq. (Al ) is known as the Maxwell-Stefan volved, Eq. (A 1) is modified using the following
equation [ 18 1, where ,uiis the chemical potential expression:
of component i and dpJdz is its local gradient.
Xi are the mole fractions of the components j= 1, (A7)
2 .. . n and Vjare their local velocities. The recip-
rocal of 0; has the meaning of a friction coeffi- where @j is the volume fraction of component j,
cient accounting for the frictional effect exerted
by component j on component i. (A81
The general expression for the chemical poten-
tial pi is given by
V, and b are the partial molar volumes of the
corresponding components. Replacing Xj by @j
A. Heintz, W. Stephan /Journal of Membrane Science 89 (1994) 153-169 165

exhibits a generalization of the Maxwell-Stefan -J, =-c,v,


equations. The physical argument is that @j
rather than Xj is a realistic measure of the prob-
ability that the friction influence of component j
on component i contributes to the total friction
force applied to component i. Application of Eq.
(Al ) or (A7) requires the definition of a frame
of reference concerning the velocities Vi and Vj (Al2)
which is given by suitable linear combinations of
-J2=-c2v2
the Ujvalues.
We first consider liquid mixtures by combin-
ing Eq. (A5 ) with Eq. (A7 ). We account for three
components with the condition k= 0 for the
frame of reference (index M denotes the mem-
brane polymer ) :
(A13)
din@, ~ v2-v, ~ v,
7= 277 3_1M Obviously, the fluxes depend on the volume
D
fractions and concentration gradients of both
din@, Q, v, -v2 ~ v2 components. If 6r2+oo, the second term in Eqs.
7= 1x- 37 (A9) (A 12 ) and (A 13 ) disappears and one obtains
D 2M

with (A14)

with DiM =BiM/@M.


(AlO) This is the simple first Fick’s law of diffusion
in the Hittorf frame of reference where DIM is the
diffusion coefficient in the Hittorf system. The
where DUare effective diffusion coefficients gen-
fluxes become independent. If Fick’s frame of
erally dependent on the concentrations of all the reference would be chosen v1@1+v2@2
components present in the mixture. The frame +%&=O, &, instead of DIM appears in Eq.
of reference (v, = 0) chosen in Eq. (A9 ) is called (A14). & is the diffusion coefficient in Fick’s
the Hittorf frame of reference [ 19 1. The low mo- frame of reference. b,, has the meaning of a cou-
lecular weight components 1 and 2, are moving pling parameter with the dimension of a diffu-
against the background of the fixed membrane sion coefficient. 6,, and B2M are the diffusion
polymer component. From this point of view the coefficients of component 1 and 2, respectively,
membrane remains at rest and no swelling of the in the polymer. For practical purposes we use the
membrane would be observed when diffusion Hittorf system. If weight fractions w: instead of
occurs. volume fractions @i are introduced, the follow-
Eq. (A9 ) has now to be rearranged in order to ing equations are obtained from Eqs. (A 12 ) and
obtain the diffisive molar fluxes Ji = c,* Vi with the (A13):
molar concentration ci given by
DZMW; +D,2 dw;
P/Ml x
Dlz+w;D2~+~;4~ >
@i
ci=E (All)
DIM w’, dw;
mfl~
D,,+w;DzM+w;D,M >
Solving Eq. (A9) for J, and J2 is a straightfor-
ward procedure and leads to the result (A151
166 A. Heintz, W. Stephan /Journal ofMembrane Science 89 (1994) 153-169

rangement of Eqs. (Al 7) and (Al 8) gives for


DIM& +Dl2 dw;
-Jz=DZM mf2x the molar fluxes J;’ and Jz :
Dl,+w;D,,+w’,D2~ >

-J: =-X,V,~~

DIM& /i/M2% X1 *d23 +Bl2


+DZM a3 dpl
=-* M
D~,+w;~M+w’,DzM >
RT 012 +X1 -023 +~z.D13 *-iii-
(Ale)
xl dp2
+%._ _‘D23 _
where p is the mean density of the swollen mem- RT 012 +x, -023 +x2.013 YG-
brane. DZM and DIM are the diffusion coeffi- (A21)
cients of 2 and 1 in M if the Hittorf frame of ref-
erence is used and D,2=8,2/@M.
- J2’ = -x~v~;~
The flow of a gaseous mixture through narrow
pores can be considered in a similar way as dif-
o”23 X2.D13 +B,, dp2
fusive fluxes in dense membrane polymers lead- =-* I

ing to solutions which involve laminar flow and RT 012 +~2’D,3 +X1 *D23’dZ
Knudsen flow as limiting situations.
We start with Eq. (Al ) using pi from Eq. (A4). +a._ RT 012 +x2*013
_
x2.0”13

+x,
I

sD23.Z
dpl

We consider a binary gaseous mixture. The ‘third


component’ is the wall of the pore. This auto- (A=)
matically requires that v3= 0. The Maxwell-Ste- J;’ and Jz are molar fluxes in the pore due to
fan equations obtained are diffusive processes only. However, there also ex-
ists a laminar (viscous) flow of the whole mix-
1 dp, v2-Vl Vl
(A17) ture which is superimposed to the individual dif-
p, dz ‘r-d,,
fusive fluxes. A velocity u has to be added to each
1 dP2_x VI -v2 ~2 of the individual velocities Viresulting in the to-
---
p2 dz lb,,-8,,
(Al8) tal velocity Wi = Vi + 24.
According to the theory of viscous flow in cy-
with p1 = Pax, and p2 =x2. d, 3 and o”23 are recip- lindrical pores [ 2 1 ] u is given by
rocal friction coefficients for the interactions of
component 1 and 2 with the wall, respectively. 1 d2 dp
With x2=0 or x1 =O, Eqs. (A17) and (A18) re- (~23)
u= -32’G-z
duce to the Knudsen flow of pure gas 1 and pure
gas 2 through the pore. From the theory of where qmix is the viscosity of the gaseous mix-
Knudsen flow of pure gases it follows that [ 201 ture. Adding u to v, and q in Eq. (A2 1) and
(A22), and substituting Eqs. (A19) and (A20)
ij 13=
J8 RT
_._.
9 nM,
d
(A19)
into Eqs. (A2 1) and (A22 ) one obtains for the
molar fluxes Ji:

J
D23=
8 RT d
-._.
J, =x1 (v, +u)FT
9 RM,
(‘420)
4x1 -f’)
where Ml and M2 are the molar masses of gas 1 (~24)
dz
and 2, respectively, and d is the diameter of the
pore. For x1 > 0 and x2 > 0 the term containing
D12 does not disappear. d,, is the binary diffu-
sion coeffkient in the gaseous mixture. Rear-
A. Heintz, W. Stephan /Journal ofMembrane Science 89 (1994) 153-169 167

r7,=26.69% [/IPI (A35)


,
= --.
b [ -. 1 d( ( 1 -x1 1 ‘PI where o is the molecular diameter and Qs the so-
NL l+B*P dz called collision integral based on the Stockmeyer

1
potential. The viscosity of a binary mixture gas
fo’B.( l-x,)-P dP
+ mixture according to Chapman and Enskog is

(25)
1 +B.P .dz+(l-x,)./I.Pg [22]:

rtmix = xI tJ1 + x2R WI (A36)


with XI +x2h2 x2 +x,tfJ2,

The parameters 912 for gases are approximated


b=$
Js-4 2
(26) by the method of Brokaw [ 221:

(~27)
@*j=
J $Sjj.A$
J

with the parameters S, and A, containing the di-


(A37)

mensionless Stockmeyer parameter SF and the


(A2g)
reduced temperature TT = kT/Ci as molecular
properties of the pure components. Ei represents
w_ MI ‘M2 the minimum of the intermolecular potential
(~29)
-M, +M2 curve.

(A30) s12 =s21

(A31)

(~32)

A. =m__M70.5.
Y V IJ
(A33)

1
-
JCz=x,*JM,+(l-x,)*,/M, (A34) M.. _&f+“5
1+
Eqs. (A24) and (A25) are identical with cor- 2(l+M~,)~(l+~~45)m,o-’
II
responding equations derived first by Present and l+m,
de Bethune [ 171 using the kinetic gas transport
formalism. The equations are rederived here on (A39)
the basis of the Maxwell-Stefan theory. g,, ap- with
pearing in Eq. (A27 ) can be calculated following

1
0.25
the procedure given by the kinetic theory of gases. 4
The viscosities of the pure substances vi are es- (A40)
(l+M,‘)(l+Mii)
timated according to the method of Chapman
and Enskog [ 22 ] and
168 A. Heintz, W. Stephan /Journal ofMembrane Science 89 (1994) 153-169

K partial molar volume of component i


(A41) ( cm3/mol)
WiF weight fraction of component i in bulk
If all molecular parameters are known Eqs. feed mixture
(A24) and (A25 ) can be used for calculating $iF weight fraction of component i in feed
fluxes of gaseous mixture components through mixture adjacent to membrane
narrow pores. Table 3 gives parameters taken wiP weight fraction of component i in bulk
from the literature [ 22 1. permeate mixture
ti weight fraction of component i in active
layer at arbitrary location
tiF weight fraction of component i in active
7. List of symbols layer on feed side of membrane
tiP weight fraction of component i in active
ai thermodynamic activity of component i layer on permeate side
ci molar concentration of component i xi mole fraction of component i
(mol/l)
Z axis coordinate perpendicular to mem-
d pore diameter ( m ) brane surface
d max maximum of pore size distribution func-
tion (m) 7.1. Greek letters
Di diffusion coefficient of component i
(cm2/s) volume fraction of component i in
Maxwell-Stefan interaction parameter mixture
(cm2/s) thickness of dense, active layer (m )
effective diffusion coefficient depen- thickness of porous support layer (m)
dent on concentration (cm2/s) Stockmeyer parameter
Dij averaged diffusion coefficient (cm’/s) difference operator
fi‘ fugacity of component i ( mbar ) characteristic energy of component i (J )
Ji molar flux of component i (mol/h m2) activity coefficient of component i
k mass transfer coefficient chemical potential of component i (J/
Mi molar mass of component i (g/m01 ) mol )
number of components viscosity of pure component i (kg/m s)
i parameter for pore size distribution viscosity of gas mixture (kg/m s)
function density of component i ( kg/dm3)
K Avogadro number ( 1/mol) mean density of swollen membrane (g/
P total pressure (mbar) cm3 )
Pi partial pressure of component i in gas- molecular diameter (nm)
eous mixture (mbar ) binary interaction parameter
PP saturation pressure of component i at ( UNIQUAC )
given temperature (mbar ) collision integral based on the Stock-
P” standard pressure (mbar ) meyer potential
R gas constant (J/mol K) pore size distribution function
siF solubility coefficient on feed side of ac-
tive membrane layer
si, solubility coefficient on permeate side of 8. Acknowledgement
active membrane layer
T absolute temperature (K) Financial support of the Bundesministerium
u velocity due to viscous flow (m/s) fur Forschung und Technologie is greatly
vi mean velocity of molecule i (m/s) acknowledged.
A. Heintz, W. Stephan /Journal ofMembrane Science 89 (1994) 153-169 169

9. References [ 11 ]A. Heintz and W. Stephan, A generalized solution-dif-


fusion model of pervaporation through composite
ill R. Rautenbach and R. Albrecht, Membrantrennverfah- membranes. Part I. Prediction of mixture solubilities in
ren, Otto Salle Verlag, Frankfurt, 198 1. the dense active layer using the UNIQUAC model, J.
[21 H. Strathmann, Membrane separation processes, J. MembraneSci., 89 (1994) 143-151.
Membrane&i.,9 (1981) 121-189. [ 121 W. Stephan, Entwicklung eines Modells zum Stofftrans-
131 R. Rautenbach and R. Albrecht, The separation poten- port durch porenfreie Polymermembranen und Durch-
tial of pervaporation. Part I. Discussion of transport fuhrung von Pervaporationsexperimenten, PhD-Thesis,
equations and comparison with reverse osmosis, J. University of Heidelberg, 1993.
Membrane Sci., 25 (1985) l-23. [ 13 ] J. Hauser, A. Heintz, B. Schmittecker and R.N. Lichten-
[41 R.Y.M. Huang and J.W. Rhim, Separation characteris- thaler, Sorption equilibria and diffusion in polymeric
tics of pervaporation membrane separation processes, membranes, Fluid Phase Equil. 5 1 ( 1989) 369-38 1.
in R.Y.M. Huang (Ed.), Pervaporation Membrane Pro- [ 14]M. Wesslein, A. Heintz, and R.N. Lichtenthaler, Perva-
cesses, Elsevier, Amsterdam, 199 1, Chap. 2. poration of liquid mixtures through PVA membranes. I.
151 M.H.V. Mulder, Thermodynamic principles of perva- Study of water containing binary systems with complete
poration, in R.Y.M.Huang (Ed.), Pervaporation Mem- and partial miscibility. J. Membrane Sci., 51 ( 1990)
brane Processes, Elsevier, Amsterdam, 199 1, Chap. 4. 169-179.
[61 R. Rautenbach, C. Herion and U. Meyer-Blumenroth, [ 15 ] P. Rosin and E. Rammler, Die Komzusammensetzung
Engineering aspects of pervaporation: calculation of des Mahlgutes im Lichte der Wahrscheinlichtkeitslehre.
transport resistances, module optimization and plant Kolloid Z., 67 (1934) 16-26.
design, in R.Y.M. Huang (Ed.), Pervaporation Mem- [ 16]H. Steinhauser, H. Scholz, A. Htlbner and G. Elling-
brane Processes, Elsevier, Amsterdam, 199 1, Chap. 3. horst, private communication.
[71 A. Heintz and G.A. Reinhardt, Model calculations of [ 17lR.D. Present and A.J. de Bethune, Separation of a gas
the pervaporation through hydrophilic membranes in- mixture through a long tube at low pressures, Phys. Rev.,
cluding the effect of the support layer, in R. Bakish (Ed. ) , 75(1949)1050-1057.
Proc. 5th Int. Conf. Pervaporation Processes in the
[ 18lJ.A. Wesselingh and R. Krishna, Mass Transfer, Ellis
Chem. Ind., Heidelberg, Bakish Mat. Corp., Englewood,
Horwood, New York, 1990.
NJ, 1991, p. 174.
[ 19]A. Heintz, H. Funke and R.N. Lichtenthaler, Sorption
[81 P. C&t and C. Lipski, Mass transfer limitations in per- and diffusion in pervaporation membranes, in R.Y.M.
vaporation for water and waste water treatment, in R.
Huang (Ed. ) , Pervaporation Membrane Separation
Bakish (Ed. ), Proc. 3rd Int. Conf. Pervaporation Pro-
Processes, Elsevier, Amsterdam, 199 1, Chap. 5.
cesses in the Chem. Ind., Nancy, Bakish Mat. Corp., En-
glewood, NJ, 1988, pp. 449-462. [20]E.L. Cussler, Diffusion: Mass Transfer in Fluid Sys-
tems, Cambridge University Press, Cambridge, 1985.
[91 B. Raghunath and S.-T. Hwang, Effect of boundary layer
mass transfer resistance in the pervaporation of dilute [ 211 D.P. Shoemaker, C.W. Garland and J.W. Nibler, Exper-
organics, J. Membrane Sci., 65 ( 1992) 147-l 6 1. iments in Physical Chemistry, 5th ed., McGraw-Hill,
[ 10 ] R. Rautenbach and F.P. Helmus, Some considerations New York, 1989.
on mass-transfer resistances in solution-diffusion-type [22]R.C. Reid, J.M. Prausnitz and T.K. Sherwood, The
membrane processes, J. Membrane Sci., 87 ( 1994) 17 l- Properties of Gases and Liquids, McGraw-Hill, New
180. York, 1977, Chap. 9.

You might also like