Sei sulla pagina 1di 8

j o u r n a l of MEMBRANE SCIENCE

ELSEVIER Journal of MembraneScience 119 (1996) 39-46

Analysis of mass transfer during ultrafiltration of PEG-6000 in a continuous stirred cell: effect of back transport
Chiranjib Bhattacharjee, Siddhartha Datta *
Department of Chemical Engineering, Jada~'pur Unit'ersity, Calcutta 700032, India

Received 27 November 1995; accepted 18 March 1996

Abstract
A mass transfer model based on an unsteady state mass balance over the concentration boundary layer, coupled with the back transport opposing the ultrafiltrate flux, has been developed in the present study. The model is capable of predicting permeate flux and rejection at different pressures, concentrations and stirrer speeds. This model first uses the experimental data of Bhattacharjee and Bhattacharya to integrate the governing partial differential equation and in the process, it calculates the back transport coefficient by a least squares fit. Once this coefficient along with other membrane and solute properties are known, permeate flux and rejection can be predicted at any desired operating conditions. Concentration profiles as a function of time for different experiments were also computed in order to analyze the effects of different operating parameters on the concentration boundary layer. The model predicts that the flux decline during ultrafiltration of PEG-6000 in a continuous stirred cell using a cellulose acetate membrane, occurs mainly due to the resistance offered by solute molecules during their transport back to the bulk. The effects of osmotic pressure and cake/gel formation are assumed to be negligible in this study.
Keywords: Ultrafiltration;Back transport; Flux; Rejection;Mathematical modelling

1. Introduction
Separation, purification and concentration are important steps in the chemical process industry accounting for over 50% of operating costs and a major portion of capital investments. Continuous efforts are being made world wide to improve existing processes or search for more efficient low energy processes. Over the last few decades, ultrafiltration (UF) has emerged as a successful competitor to conventional unit operations, particularly in the food,

* Correspondingauthor.

chemical and pharmaceutical industries. Glimenius [1] has presented some observations, viewpoints and trends in U F for various industries. Ultrafiltration is a rate governed separation process in which pressure is the main driving force. Usually its operational pressure varies in the range 10 to 140 psi. The feed solution containing macromolecular solutes is introduced into a membrane separator where solvent and certain solutes pass through a semipermeable membrane which are subsequently collected as ultrafiltrate. One of the problems associated with membrane processes is the phenomenon of concentration polarization. In UF, solute rejected by the membrane accumulates on the

0376-7388/96/$15.00 Copyright 1996 Elsevier Science B.V. All rights reserved. PII S0376-7388(96)00114-7

40

C. Bhattacharjee, S. Datta / Journal of Membrane Science 119 (1996)39-46

membrane surface. The concentration of solutes on the membrane surface thus becomes higher than the bulk solution. This is the so called concentration polarization phenomenon. This drawback is perhaps a major reason for the relatively slow acceptance of the ultrafiltration process by industry. Several models are available to describe the flux behavior during ultrafiltration. Wijmans et al. [2] suggested some phenomena to account for this flux reduction: (a) a decrease of the hydraulic driving force by osmotic pressure, (b) resistance of the concentration polarization boundary layer, (c) resistance of a gel layer, (d) an increase in the membrane resistance and (e) the resistance of an adsorption layer. The ultrafiltration permeate flux can be analyzed by the following models: The gel polarization model [3], the osmotic pressure model [4,5] and the resistance in series model [2]. According to Wijmans et al. [2] all the models mentioned above predict almost equivalent permeate flux under steady state conditions, especially at higher concentrations. For this reason, it is very difficult to conclude from a study of experimental flux data which mechanism is valid. It has been found in previous studies [6,7] that no gel layer is formed near the membrane for the case of PEG-6000 ultrafiltration with a cellulose acetate membrane of 5000 MWCO. The flux reduction during continuous stirred ultrafiltration has been attributed to the formation of a "gel type" layer whose concentration is far lower than the solubility limit prevailing for the true gel and the extra resistance of this layer to flow causes the flux to decline with time. Various workers have investigated this phenomenon. Most studies were based on the resistance in series model. Resistance of the additional " c a k e " or "gel type" layer was assumed to be proportional to its thickness and as the thickness increased, flux decreased due to the increase of resistance. This continuous accumulation of solutes near the membrane in the case of continuous stirred UF seems to be somewhat improbable, as the rejected solutes would increase the thickness as well as the resistance of the boundary layer continuously as the time goes on. Since, it has been experimentally found that the flux becomes constant after a particular time interval, this continuous increase in thickness of cake layer is

not possible. Therefore it would be a better supposition to assume a turbulent back transport of the rejected solute into the bulk due to the effect of stirring. This turbulent back transport whose rate is assumed to be proportional to the bulk concentration and stirrer speed, takes place in addition to the diffusive back transport which is the consequence of concentration polarization. Accordingly in this work, a mathematical model based on turbulent back transport phenomena has been developed. In developing the model, the following assumptions have been made. (1) The formation of the gel type layer is neglected. (2) The concentration boundary layer is assumed to form initially and subsequently the turbulent back transport flux gets superimposed on convective flux and diffusive back flux. (3) The flux decline occurs predominantly from the resistance offered by the solute molecules moving in the opposite direction due to turbulent back transport. (4) This decline occurs until the system attains steady state. In the present study, the back transport coefficient has firstly been evaluated based on the experimental data of Bhattacharjee and Bhattacharya [8]. Then this value for the back transport coefficient was utilized in predicting the flux and rejection simultaneously. The predicted flux and rejection were found to agree with the experimental value [8] within a tolerance of _+ 10%. Unlike other studies, in this paper the flux decline occurs not due to the build up of "cake", " g e l " or "fouling", but mainly due to the resistance from the convective back current created by stirrer, which takes the solutes back from boundary layer into the bulk.

2.

Theory

Fig. 1 shows the concentration polarization phenomena for a membrane process. The present paper deals with the bulk transport of the solutes together with the back diffusive term in the differential mass balance over an element of thickness a x' lying at a distance x' from the membrane surface, along with the accumulation term. It is further assumed that the effect of stirring is insignificant within the boundary layer but it generates a backward convective current of accumulated solute from boundary layer into the

C. Bhattacharjee, S. Datta / Journal of Membrane Science 119 (1996) 39-46

41

Boundary layer ~. "xt~


i

if/, ~ -3--J(, i ":~:


dc "5///,
.....,

I----

V Membrane

Experimental values of flux, J, were correlated at time t' according to the equation [6] J = Jo(1 - 0 + Oyt') (6)

/2
Permeate

dy.
Bulk

;<<:

from which the total filtrate volume (V) collected up to time t' can be obtained as v/a=Ltjdt=Jo[1-O+O(y(-1)/ln y] (7)

! I

i *11/
....

"//2

i I///, - - C o
I...
I Ill, "

6 l" c'--c'b
x=

4
.

In Eqs. (6) and (7), t' has the dimension of minutes. For a stirred batch cell, the mass transfer coefficient can be found from the well known empirical correlation [9]: k = 0.0443( D / r ) ( v / D ) 3 3 ( o ~ r 2 / v ) 8 (8)

C Cm

x'-- 0

Fig. 1. Mass balance over the boundary layer.

Further, the diffusivity can be empirically related to the molecular weight of the polymer solution [10]: D = 2.74 l0 -9 M -1/3 (9)

bulk. With this assumption, the unsteady state mass balance gives the following differential equation:

OCt

OCt

02C'
(1)

Eqs. (4) and (5) were solved using a 6-point orthogonal collocation (OC) technique [11]. Eq. (4) can be written according to conventional nomenclature as: dc i J(t) n+2
n+2 j=l

~t' = J-~x' + D o x' 2

dt

The above equation assumes constant diffusivity and density and is to be solved under the following initial and boundary conditions: (i) At t' = 0, c' = c~ for all x' dc' (ii) At x ' = 0, d x ' (iii) At x ' = 6 , J D (C'lo - C'p) (2)

j=l

E Eijcj + E 8,jcj =f,

(10)

for i = 2,3 ...... n + 1, where n = 6, because a 6-point OC has been used. According to b.c. (iii), c,+ 2 1 at all time and c 1 can be calculated from b.c. (ii)
=

n+2

CI = ( J / k ) c p - - j~=2Eijcj](Ell-l-( J / k ) ) - I

c ' = c 'b for a l l t '

(ll)
Eij and Bij were calculated from the tabulated values of x i [11] for n = 6, according to the following formula [11]: Oij = x { - ' , G i j = ( j 1 ) x { -2,

Introducing the dimensionless variables in the following form, c = c ' / c ' b, x = x ' / 6 and t = D t ' / 6 2 Eqs. (1) and (2) become: ac Ot with (3)

J(t)
k
b.c.:

H u = (j02c + -cOx cOx2 ac (4) and dc


-

1)(j-

2) x[ -3

(i) c = 1 at t = 0 for all x dc (ii) d x (iii) J ~(Clo-Cp)at for all


t

dx

= G Q - i c = Ec where E = G Q - t

d2c d x 2 = H Q - l c = Bc where B = H Q - l x=O


(5)

c = 1 at x = 1

The explicit Euler method with small time interval At has been used to eliminate mathematical corn-

42

C. Bhattacharjee, S. Datta / Journal o f Membrane Science 119 ( 1 9 9 6 ) 3 9 - 4 6 t m+

plexities. Concentration at time from

l can be obtained

clm+ l)

clm) _~_Atfi(m )

(12)

Eqs. (11) and (12) were solved to find c i, i = 1 to n + 2 at time tin+ I" While doing the overall mass balance across the system, it has been assumed in this study that the total solute rejected by the membrane results in partly raising the average concentration in the boundary layer over that of the bulk with the remainder being transported away to the bulk by the convective back current. It has been further assumed that the back transport rate is proportional to the stirrer speed and bulk concentration. Then the overall mass balance equation becomes:
V ( c'b - C'p) = A 6 ( U C'b) + kbC'bWt'

or in terms of dimensionless concentration: V(1-cp)=A6(?1) + k b W t ' (13)

where k b is the constant of proportionality and can be termed as back transport coefficient. Now ? is the average concentration prevailing over the boundary layer and can be calculated at time tm+ 1 by the following formula [11]:
n+2
~ ( m + 1) : E

WjC~m+ 1)

(14)

j=1 All the concentration values at different points, c i, i = 1,2....n + 2, were calculated from Eqs. (11) and (12) and substituted into Eq. (14) along with the tabulated values of w i [11] to calculate average concentration ? over the boundary layer at any desirable time. Using various values of V [obtained from Eq. (7)], ? at different times for a single experiment at any fixed operating conditions, the k b value can be obtained by least squares fit of all the data at different times. The back transport coefficient values calculated for different experiments under various operating conditions were found to be essentially constant. Therefore the average of k b for all experiments was taken as the back transport coefficient value for the given system. This coefficient remains constant for all operating conditions as long as the membrane solute system, cell geometry, and the temperature remain unaltered. In fact the maximum

deviation for coefficient values for different experiments under various operating conditions from the average value was found to be within + 10% Once the back transport coefficient for the particular system is known, the permeate flux and rejection can be predicted without performing any experiments. For this, one has to know membrane hydraulic resistance, Rm, solute permeability, Pm and reflection coefficient, o', along with the back transport coefficient for the particular system, which can be calculated by the method discussed above. Once these four parameters are known, flux and rejection can be predicted at any operating conditions. For this purpose, an iteration scheme has been proposed. Permeate flux was predicted in the initialization step using the osmotic pressure model. At time t = 0, c" is approximately equal to c~,. The osmotic pressure was calculated for a PEG-6000 solution by Flory's equation [12]. The overall iteration scheme is outlined below: (i) At was chosen (ii) t = 0 , Cg for i = l to n + 2 were set to 1.0. V ~) = 0, Cp was assumed (Cp = 0.1 is a good choice) (iii) J = ( A p - o ' A ' r r ) / ( R m/~s) where A~-= ~r(C'b)--lr(C'p). Here Cp 0.1c b. Also from irreversible thermodynamics [13], cp = ( ( 1 - o - ) / ( 1 c r F ) ) c m where F = e x p ( - ( 1 - ~ ) J / P m ) . The above two equations were solved iteratively to find the flux J and permeate concentration Cp at time t = 0. (iv) With the known values of c}m), j(m), v(m), c(m) at time t m (total n + 5 parameters), the followP ing steps have been followed: c} m+ 1) = clm) + Atfi(m) where fi has been defined by Eq. (10). c~2 m+O, .(m+ ~3 1). . . . . . ,,~(m+ 1) were calculated in this manner. n+l c(m+ 1) remains 1.0 at all times. n+ 2 (v) The average concentration ~(m+l) over the boundary layer was calculated from Eq. (14) using known values of c~ m + calculated from step (iv). [Note: since w 1 and wn+ 2 are zero, so c~ m+l) and c ( m + l ) are not required in the calculation of ~(m + 0.] ,,+ 2 (vi) Eq. (13) was used to calculate V ~m+ i) (vii) Permeate flux was calculated from j~m + J) = (l/A)(dV/dt) = ( 1 / A ) ( V ~ + 1) _ V ( r ~ ) ) / A t , . (viii) c~m+ 1) was calculated from Eq. (11). (ix) tm+ 1 = tm + At. (x) Steps (iv) to (ix) were repeated up to say, t = 100, during which time the boundary layer was fully developed.

C. Bhattacharjee, S. Datta / Journal of Membrane Science 119 (1996) 39-46

43

(xi) Then using Cp = ((1 - t r ) / ( 1 o'F))Cl, a new Cp was calculated using the J value calculated in step (vii). (xii) Calculations were repeated from step (iv) with the same initialization at time t = 0 except for the Cp value, which has been calculated in step (xi). The whole procedure was repeated say, up to t = 100 and the Cp value was re-evaluated. If this value coincides with the previous one within the allowable range of tolerance, then the integration procedure may be continued up to any desired time. In the above iteration scheme, the permeate concentration was determined from the irreversible thermodynamic equation which is based on a membrane surface concentration corresponding to a fully developed concentration profile. It has been found in this study and also in previous papers [6,7] that the boundary layer develops within a very short span of time and in most cases, this time is less than one second. Here it has been found that after attainment of a fully developed boundary layer, the concentration at any point in the boundary layer decreases very gradually with time, until the steady state permeate flux is reached. This gradual decline in the concentration profile occurs due to the proposed back transport mechanism. The membrane surface concentration required for calculating the permeate concentration was taken to be that prevailing for a fully developed boundary layer.
-

40
~ 0.33s

35
E
e.-

\\\
-~ O-023 S

30

25

t"
u e-.
o U

20

15

10

0.0

0.2

0.4.

0.6

0.8

1.0

1.2

Dimensionless d i s t a n c e
Fig. 2. Initial d e v e l o p m e n t o f the b o u n d a r y layer as a function o f time u n d e r fixed operating conditions ( A p = 120 psi, c b = 16 k g / m 3, (.9 = 55.502 r a d / s ) .

3. Results and discussion

In this study, the choice of time interval (At) is very important. The set of ordinary differential equations, obtained from the partial differential equation. by orthogonal collocation, are quite stiff and for this reason, At has to be made very small (say, 0.1) to achieve convergence. It has been found that the stiffness of the set of differential equations in fact increases as the number of internal collocation points is raised. On the other hand, an increase in the number of collocation points generally gives better accuracy. Here the four point collocation technique has been used. Of course one can use any "A-stable" technique, but this increases mathematical complexities in the iterative loop. In this paper the osmotic pressure relationship,

solute permeability, etc., have been used only in the initialization step during prediction of flux and rejection. This initialization of flux and various concentration terms before starting the iterative loop are very significant because they can affect the convergence and its rate very adversely. The back transport coefficient was calculated for each experiment by a least squares fit of the experimental data at various times. The variation of this coefficient for different experiments calculated under various operating conditions was found to be very small. In fact, the average value for all experiments was taken to be the desired value of the back transport coefficient for this cell configuration operating with the specified membrane solute combination and at the specified temperature. The value of the coefficient thus found to be 1.25935 10-11 m 3. Fig. 2 shows the initial development of the concentration profile for a particular run under a specified set of operating conditions. Initially a flat concentration profile exists throughout the region. After starting the ultrafiltration process, the profile develops very quickly. The concentration near the membrane increases more rapidly compared to the region

44

6o \
5OI
~E \
O

C. Bhattacharjee, S. Datta / Journal of Membrane Science 119 (1996) 39-46

55.502 rod/s .... ..... 47.124 r o d / s 34.034 r0d/s

'kk
"\

t)

o {_)

20

10

0-0

0.2

0.4

0.6

O.B

1.0

1.2

Dimensionless

distance

Fig. 3. Fully developed concentration profile as a function of stirrer speed at constant bulk concentration (c b = 18 k g / m 3 ) and pressure differential ( A p = 120 psi).

the increased concentration gradient. In fact, at very low stirrer speed, back transport may become totally diffusion controlled, making the turbulent transport resulting from the stirrer quite insignificant. The variation of the concentration profile with pressure is shown in Fig. 4. This figure shows that with an increase in pressure, the concentration at any point within the boundary layer increases which is more pronounced near the membrane. As the pressure increases, the flux also increases due to an increase in the driving force. This higher flux results in increased rate of bulk transport of solutes towards the membrane surface, where most of it get rejected by the membrane, thus raising the concentration in the boundary layer. This occurs due to the fact that the back transport rate remains effectively the same, whereas the forward transport rate due to bulk movement increases and this difference results in greater accumulation and a higher concentration build up near the membrane. Fig. 5 shows that the concentration at any point within the boundary layer increases with bulk concentration. This is expected because higher bulk concentration means more solute build up near the

further away from the membrane. As shown in Fig. 2 the concentration gradient is very high near the membrane. This is because of the sieving action of the membrane, i.e. the solute gets rejected by the membrane which results in an increase of solute concentration very rapidly at all points adjacent to the membrane. This figure also shows that the concentration profile develops very rapidly over a short initial period after which changes become sluggish and in fact, it has been observed that there is hardly any change after one second. This very rapid development of the boundary layer is also supported by film theory. Fig. 3 shows the fully developed concentration profile as a function of stirrer speed, keeping all other parameters constant. As the stirrer speed decreases, the back transport rate also decreases, because the back transport rate is proportional to stirrer speed. Due to this decrease in the back transport rate, more solute accumulates near the membrane and hence the concentration also increases in the boundary layer. It is further seen that the concentration profile becomes steeper with a decrease of stirrer speed which increases the back diffusion rate due to

40
_~
35 E (J "-" 30 01
_h{

- ~

t 2 o psi

.... t o o psi . . . . . . . 8 o psi

\
\'\.'\,

=o

25

"\5. \
"~.~.~

.u

20

15

10 0.0

I 0.2

~ 0.4

~ 0.6

I O.B

i 1.0

1.2

Di men sion l e s s d i s t a n c e
Fig. 4. Fully developed concentration profile as a function of pressure at constant bulk concentration (c b = 18 k g / m 3) and stirrer speed ( w = 47.124 r a d / s ) .

C. Bhattacharjee, S. Datta/ Journal of Membrane Science 119 (1996) 39-46

45

120 \ "\.,, 100


~n ,-~ 8 0 0 -

20 Kg/m 3 50 Kg/m 3 \'~..,~


.....

70

Kg/m

" ~.
~ "-,~.

" ' - - '~..~.


'~.,~,,

ID U

~o

40

2OC I 1 i

~
i i

membrane as has been explained above. Though the back transport rate also increases to cope partly with the effect of increased concentration, the boundary layer concentration must be greater than the bulk due to solute accumulation near the membrane. Variation of membrane surface concentration with time, starting from the initial development of the boundary layer, is shown in Fig. 6 for a particular run under specific operating conditions. At first, concentration develops very quickly from its bulk value and then decreases very gradually. This decrease in membrane surface concentration occurs due to back transport created by the stirrer. The solute gets transported back into bulk due to stirring and this results in a gradual decrease of concentration at any point within the boundary layer.

0-0 0.2 0.4 0.6 0.1~ 1.0 Dimensionless distclnce

1.2 4. List of symbols


A
B

Fig. 5. Fully developed concentration profile as a function of bulk concentration at fixed pressure (AP = 120 psi) and stirrer speed (~o = 55.502 tad/s).

4-0

35
D E EU

30
tO

E 2,5

20'

L
F G H

transport area (m 2) collocation matrix; Bij a n element in that matrix concentration ( k g / m 3 ) ; c~,, bulk concentration; Cp, permeate concentration; c ' , ' membrane surface concentration dimensionless concentration = c /' c 'b, c i, at ith point; c b, Cp and c m means bulk, permeate and membrane surface concentration, respectively, in dimensionless form diffusivity ( m 2 / s ) collocation matrix element in ith row and jth column of E matrix function defined by Eq. (10) a dimensionless constant = e x p [ - (1 o')Jm/P]

I0

0"0

50

I0"0 Time,

15"0
rain.

20'0

25"0
k kb

Fig. 6. Variation of m e m b r a n e surface concentration with time for a run under fixed operating conditions (AP = 120 psi, c b = 20 k g / m 3 and o~ = 55.502 r a d / s ) .

M MWCO

collocation matrix; Gij an element in G matrix collocation matrix; Hij an element in H matrix permeate flux ( m 3 / m 2 s), Jo permeate flux at time zero mass transfer coefficient = D / 6 l ( m / s ) back transport coefficient (m 3) defined by Eq. (13) molecular weight ( k g / k m o l ) molecular weight cut off (abbreviation)

46

C. Bhattacharjee, S. Datta / Journal of Membrane Science 119 (1996) 39-46

em
PEG Q
r

Rm t'
t UF V
Wi xp x

pressure (Pa) solute permeability ( m / s ) p o l y e t h y l e n e glycol (abbreviation) c o l l o c a t i o n matrix, Qij an e l e m e n t in Q matrix radius o f the m e m b r a n e (m) m e m b r a n e hydraulic resistance (m ~) time (min in Eqs. (6) and (7), s in all other cases) dimensionless time = Dt'/~ 2 ultrafiltration (abbreviation) c u m u l a t i v e v o l u m e (m 3) w e i g h t factor for ith collocation point distance in the film layer (m) d i m e n s i o n l e s s distance = x'/(3, x i distance o f the ith point f r o m the m e m b r a n e

References [1] R. Glimenius, Membrane process for water, pulp and paper, Desalination, 35 (1980) 259, [2] J.G. Wijmans, S. Nakao, J.W.A. van den Berg, F.R. Troelstra and C.A. Smolders, Hydrodynamic resistance of concentration polarization layer in ultrafiltration, J. Membrane Sci., 22 (1985) 117. [3] W.F. Blatt, A. Dravid, A.S. Michaels and L.M. Nelsen, Solute polarization and cake formation in membrane ultrafiltration: causes, consequences and control techniques, in J.E. Flinn (Ed.), Membrane Science and Technology, Plenum Press, New York, 1970, p. 47. [4] A.A. Kozinski and E.N. Lightfoot, Protein ultrafiltration: A general example of boundary layer filtration, AIChE J., 18 (1972) 1030. [5] R.L. Goldsmith, Macromolecular ultrafiltration with microporous membranes, Ind. Eng. Chem. Fundam., 10 (1971) 113. [6] C. Bhattacharjee and P.K. Bhattacharya, Prediction of limiting flux in ultrafiltration of Kraft black liquor, J. Membrane Sci., 72 (1992) 137. [7] C. Bhattacharjee and P.K. Bhattacharya, Flux decline analysis in ultrafiltration of Kraft black liquor, J. Membrane Sci., 82 (1993) 1. [8] C. Bhattacharjee, Prediction of limiting flux in ultrafiltration of Kraft black liquor, M.Tech. Thesis, IIT, Kanpur, 1991. [19] Q.T. Nguyen, P. Aptel and J. Neel, Characterization of UF membranes, Part II. Mass transport measurements for low and high molecular weight synthetic polymer in water solution, J. Membrane Sci., 7 (1980) 141. [10] T.K. Sherwood, R.L. Pigford and C.R. Wilke, Mass Transfer, McGraw-Hill, New York, 1975. [11] B.A. Finlayson, Nonlinear Analysis in Chemical Engineering, McGraw-Hill, New York, 1980. [12] P.J. Flory, Principles of Polymer Chemistry, Cornell University Press, Ithaca, NY, 1953. [13] S. Nakao and S. Kimura, Analysis of solute rejection in ultrafiltration, J. Chem. Eng. Jpn., 14 (1981) 32.

4.1. Greek letters


Y g 0 /x
fO 77"

P
/: (7"

a p a r a m e t e r in Eq. (6) f i l m thickness (m) a p a r a m e t e r in Eq. (6) viscosity ( k g / m s) angular v e l o c i t y ( r a d / s ) o s m o t i c pressure (Pa) density ( k g / m ) k i n e m a t i c viscosity = t z / p ( m 2 / s ) reflection coefficient difference operator

4.2. Superscript

p a r a m e t e r value at kth time point

Potrebbero piacerti anche