You are on page 1of 14

journal of MEMBRANE SCIENCE

ELSEVIER

Journal of Membrane Science 121 (1996) 229-242

Modeling of concentration polarization and depolarization with high-frequency backpulsing


Sanjeev Redkar

1, Vinod Kuberkar, Robert H. Davis

Department of Chemical Engineering, University of Colorado, Boulder, CO 80309-0424, USA.

Received 23 January 1996; revised 26 April 1996; accepted 19 June 1996

Abstract
Rapid backpulsing to reduce membrane fouling during crossflow microfiltration and ultrafiltration is studied by solving the convection-diffusion equation for concentration polarization and depolarization during cyclic operation with transmembrane pressure reversal. For a fixed duration of reverse filtration, there is a critical duration of forward filtration which must not be exceeded if the formation of a cake or gel layer on the membrane surface is to be avoided. The theory also predicts an optimum duration of forward filtration which maximizes the net flux, since backpulsing at too high of frequency does not allow for adequate permeate collection during forward filtration relative to that lost during reverse filtration, whereas backpulsing at too low of frequency results in significant flux decline due to cake or gel buildup during each period of forward filtration. In general, short backpulse durations, low feed concentrations, high shear rates, and high forward transmembrane pressures give the highest net fluxes, whereas the magnitude of the reverse transmembrane pressure has a relatively small effect. Rapid backpulsing experiments with yeast suspended in deionized water were performed with a flat-sheet crossflow microfiltration module and cellulose acetate membranes with 0.07 Ixm average pore diameter. The optimum forward filtration times were found to be 1.5, 3, and 5 s, respectively, for backpulse durations of 0.1, 0.2, and 0.3 s. Both theory and experiment gave net fluxes with backpulsing of about 85% of the clean membrane flux (0.022 c m / s = 790 1/m 2 h), whereas the long-term flux in the absence of backpulsing is an order-of-magnitude lower (0.0026 c m / s = 94 1/m 2 h).
Keywords: Biotechnology; Fouling; Microfiltration; Ultrafiltration; Theory

1. Introduction Crossflow ultrafiltration and microfiltration are often a c c o m p a n i e d by severe m e m b r a n e fouling,

* Corresponding author. Tel.: 303-492-7314; Fax: 303-4924341; E-mail: davisr@spot.colorado.edu. i Present address: Matrix Pharmaceutical, Inc., 34700 Campus Drive, Fremont, CA 94555

which leads to declines in permeate flux and m e m brane selectivity [1,2,7]. O n e technique for reducing m e m b r a n e fouling is backpulsing, in which the transm e m b r a n e pressure is reversed at high frequency (on the order of once per second) to provide in situ cleaning by forcing fluid in the reverse direction through the m e m b r a n e . T r a n s m e m b r a n e backpulsing has been shown to be effective in applications such as ultrafiltration of single proteins and protein mix-

0376-7388/96/$15.00 Copyright 1996 Elsevier Science B.V. All rights reserved. PH S0376-7388(96)00179-2

230

S. Redkar et al./Journal of Membrane Science 121 (1996) 229-242

tures, microfiltration of yeast cells, and protein recovery from cell lysates [5,6,8-12]. Backpulsing is similar to the more familiar operation of backflushing, except that the latter occurs at a much lower frequency, with backflush durations of a few seconds employed between intervals of several minutes or longer [3,7]. The key difference is that the shorter intervals between backpulses may be of the same order as the characteristic growth time of the fouling layer, and so the fouling layer may be removed or disrupted before significant flux decline occurs. It is anticipated that, for a given application, there is an optimum combination of backpulsing frequency and duration which maximizes the permeate flux. For example, a very short backpulse may not provide sufficient membrane cleaning or provide enough time for the cake layer to be lifted off the membrane by the reverse flow and then swept away by the tangential flow, whereas a very long backpulse leads to unnecessary loss of permeate. In addition, a very short interval between backpulses has too little collection of permeate relative to that lost by backpulsing, whereas a long interval of forward filtration leads to significant fouling and flux loss. Redkar and Davis [9] developed a simple model of the backpulsing operation, with the key assumptions being that the flux decline during the forward filtration portion of a cycle is described by dead end filtration theory, and that backpulsing causes instantaneous and complete cleaning of the membrane. Qualitative agreement between predicted and measured net permeate fluxes for yeast suspensions was obtained. For each backpulse time or duration, an optimum value of the forward filtration time interval which maximized the permeate flux was found. However, the simple model does not include a delay in cake formation at the start of each period of forward filtration, as was evident from the experimental data, and it is not able to predict a minimum backpulse duration required to achieve membrane cleaning. As a result, the measured net fluxes are higher than those predicted by the simple model [9]. The purpose of the present work is to develop a more complete model of the rapid backpulsing process, which includes the time required for the development of the concentration polarization layer during forward filtration, and for the depolarization of this

layer during reverse filtration. In addition, the possibility of sufficiently rapid backpulsing to avoid formation of a cake layer on the membrane surface altogether is explored by solving the convection-diffusion equation for cyclic operation. Finally, experiments with suspensions of washed yeast cells are presented and compared with model predictions.

2.

Theory

The backpulsing process for crossflow filtration is depicted in Fig. 1. Cyclic operation is employed in which a period of forward filtration of duration tf is followed by a period of backpulsing or reverse filtration of duration t b. During forward filtration, there is a delay in the formation of a cake layer on the membrane surface due to the small but finite time ( t crit for the clear fluid introduced through the memf) brane during backpulsing to be refiltered and for the subsequent development of the polarization layer until the concentration at the membrane surface reaches a maximum value at which a gel or cake layer begins to form and causes the flux to decline. During backpulsing, the reverse flow through the membrane removes the polarization and cake or gel layers from the membrane surface.

Foward filtration F e e d ~ t-,\\\\\\\\\\\\\\\\\, ,\\\\~ - i ~ ~

Reverse filtration

[N\N\\\\N\\\\\\N\\\\\\\\N 1

++++++++

[~\\\\\\\\\\\\\\\\\\\\\\~1

~,++(,(,+(,~,

tf
x ~

crit
....... Jo

Id_

......
lira

Js

TIME, t tb

Fig. 1. Schematic of rapid backpulsing and the associated permeate flux during repeated cycles of forward and reverse filtration.

s. Redkar et al. / Journal of Membrane Science 121 (1996) 229-242 2.1. Flux model

231

pression for the average or net permeate flux with backpulsing:

The net permeate flux over the entire cycle period is defined by

<J>

=Jo(t~rit-- OZtbJr- 21"((1


/ ( t f + tb), tf > t~rit

-~-(lf--t~rit)//T)

1/2-

1)

<j> =
tf + t b

(1)

(6)

where Jf and Jb are the magnitudes of the forward and reverse fluxes, respectively. When tf < t~TM, no cake or gel layer is allowed to form. It is assumed that no internal fouling of the membrane due to adsorption or pore plugging occurs, and that osmotic pressure effects are negligible. Then, the forward and reverse fluxes are simply the values for an unfouled membrane, which are related to the membrane resistance by Darcy's law: Jf
__

APf - = Jo, Jb ~oRm

APt,

- = o~Jo , tf < tf ~oRm

crit

(2)

where ~o is the permeate viscosity, R m is the membrane resistance, Jo is the clean membrane flux, A Pf is the magnitude of the transmembrane pressure during forward filtration, and a = A Pb/A pf is the ratio of the magnitudes of the transmembrane pressures during reverse and forward filtration. Then, the net permeate flux per cycle is

<J> = Jo(tf- o, t b ) / ( tf + tb),


crit

<_

erit

(3)

For simplicity, it is assumed here that the magnitude of the negative flux during backpulsing is given by Eq. (2), i.e. that the backpulse achieves instant and complete removal of the fouling deposit; this assumption will not be valid for highly adhesive deposits or very short backpulse times. Fig. 2 shows the predictions of Eqs. (2) and (6) when there is a delay in cake formation (t~rit :/: 0) during forward filtration, for a = 1 and t J r = 1. There is an optimum forward filtration time, tPt , ~f which maximizes the net flux. This optimum occurs because of the tradeoff between the reduced fraction of time spent in reverse filtration and the increased cake formation and flux decline during forward filtration, as tf is increased. A s /~rit is increased, the maximum flux increases and the optimum forward filtration time initially decreases and then increases. In general, /~pt > tfcrit , indicating that the net flux is maximized when the period of forward filtration is sufficiently large that a cake is allowed to begin forming before it is removed by backpulsing. However, it may be desirable to operate with t e </~rit

When tf > tf , a cake or gel layer of rejected material (i.e. particles for microfiltration, or macromolecules for ultrafiltration) forms on the membrane surface [13]. The associated flux decline has been shown to follow dead end filtration theory, even under crossflow conditions, for short times [3]:

1.0

0.8

Jf=Jo//(l+(l-t~rit)//T)

1/2 , t > t f

crit

(4)
A

where ~- is the time constant for cake growth and is given by [3]

- c )ZXPf
(5)

0.2

~-= 2/~c t.%CbJoZ

0.0

,~[

where c b and c c are the concentrations of the fouling species in the bulk suspension and cake layer, respectively, and /qc is the specific cake resistance (resistance per unit depth) of the fouling layer. Combining Eqs. (1)-(4) yields the following ex-

10 tf/T

15

....

20

Fig. 2. Average permeate flux versus the duration of forward filtration, as predicted by Eqs. (2) and (6) with a = 1, tb/~" = 1, and a delay in cake formation; the different curves are for t~rit/'r = 0, 0.5, 1.0, 2.0, 5.0, 10, and ~ (bottom to top).

232

s. Redkar et al./ Journal of Membrane Science 121 (1996) 229-242

when the cakes are expected to be adhesive, so that irreversible fouling is minimized. Moreover, the time delay in cake formation during forward filtration, t~rit, is not known a priori and is expected to vary with the backpulse duration, t b. This is explored by solving the convective-diffusion equations governing the concentration polarization and depolarization during the cycles of forward and reverse filtration.
2.2. Concentration polarization and depolarization

During forward filtration, the permeate flow due to the transmembrane pressure convects the suspension towards the membrane. A concentration gradient or polarization layer of rejected particles (microfiltration) or solute (ultrafiltration) forms, and the particles or solute molecules diffuse away from the membrane. As reviewed by Belfort et al. [1], the back-transport process is typically Brownian diffusion for solute molecules and particles less than about 0.1 Ixm in diameter, and shear-induced diffusion for particles with diameters between approximately 0.1 and 10 Ixm. For larger particles and high shear rates, inertial lift [4] may be important. As long as the particle or solute concentration at the membrane surface does not reach the maximum packing or gel concentration, the concentration polarization layer is mobile and does not offer a significant hydraulic resistance to the permeate flow. When the particle or solute concentration at the membrane reaches the maximum packing or gel concentration, however, a stagnant layer develops. Such a stagnant layer offers a high resistance to permeate flow, resulting in a reduction of the flux through the membrane. In rapid backpulsing, a primary goal is to prevent the formation of the stagnant cake or gel layer. Thus, the reverse filtration portion of a cycle should be started as soon as the concentration at the membrane reaches the maximum packing or gel concentration. Since convection and diffusion are both away from the membrane during reverse filtration, whereas they oppose each other during forward filtration, it seems plausible that reverse filtration portion of the cycle may be shorter than the forward filtration portion of the cycle. This is important, as the average flux in one complete cycle should be positive. In order to predict the concentration polarization

and depolarization profiles during cyclic backpulsing, we consider a Newtonian fluid which contains particles or macromolecular solutes. During forward filtration, the pressure above the membrane is higher than that below the membrane. The particles or solutes are retained within the channel, but the pure fluid (permeate) flows through the membrane with flux Jo. The particles or solutes accumulate near the membrane, resulting in the formation of a dynamic concentration polarization boundary layer (see Fig. 3). If the particle or solute size is small compared to the thickness of the boundary layer, then the boundary layer can be treated as a continuous medium. As justified below, this boundary layer is typically thin compared to the channel diameter or height, and so may be described by the Cartesian coordinates of Fig. 3 for both flat and tubular geometries. A microscopic mass balance in the boundary layer just above the membrane gives:
--+u--+v Ot Ox D

Oy

Oy

(7)

where c is the particle or solute concentration, t is time, u is the velocity in the x direction tangent to the membrane, v is the velocity in the y direction normal to the membrane, and D is the particle or solute diffusivity. The boundary conditions are c = c b as y ~ ~, Jo + DOc/Oy = 0 at y = 0 (complete rejection of the particles or solute has been assumed), and c = c b at x = 0. The initial condition for the first forward filtration period is c = c b at t = 0. In general, the fluid properties and the particle or solute diffusivity may vary with concentration. In order to keep the governing equation linear, we restrict our attention to suspensions which can be approximated as Newtonian fluids with constant den-

u(y) :::;~i~ccC(Y)
Membrane

Permeate

Fig. 3. Schematic of the concentration polarization boundary layer.

S. Redkar et al./Journal of Membrane Science 121 (1996) 229-242

233

sity and viscosity, and with constant diffusivity. Then, the axial velocity profile in the thin boundary layer near the porous membrane is linear: u = p y, where ~/ is the nominal shear rate at the membrane surface and is given by @= 3Uo/H o and @= 4Uo/H o, respectively, for fully-developed laminar flow in flat and tubular channels.[3] Here, Uo is the average velocity in the channel, and Ho is the channel half-height or tube radius. Under typical conditions, JoL/UoHo << 1, which implies that only a small fraction of the feed stream exits as permeate in a given pass, and so @ may be considered constant. The continuity equation for the bulk suspension, Ou/Ox + Ov/Oy = 0, then requires that v = constant =-Jo across the boundary layer. Moreover, the permeate flux, Jo = A Pf/iXo Rm, during forward filtration is independent of time and axial position, provided that the axial pressure drop, osmotic effects, and membrane fouling are negligible. Under these conditions, the transient convection-diffusion equation becomes
OC OC OC 02C

Ot

Jo -~y + @Y -~x = D --O 2 y

(8 /

tion reaches its maximum value, c c, is t c = 8 (c c Cb)/JoC b = D ( c c - Cb)/Jo2Cb. Except for the concentration dependence, this result is the same as that given by Rodgers and Sparks [10]. For long times, however, axial convection (the third term in Eq. (8)) becomes important, and this term is required for a steady-state solution to exist. If the shear rate exceeds a critical value, @> j,~, which scales as j3LCb/D2(c c -Cb), then the axial convection balances the transverse convection and prevents a cake or gel layer from forming on the membrane surface. The more typical case is @< @, for which a cake or gel fouling layer does form. The dynamic fouling layer reduces the permeate flux, which increases the boundary-layer thickness until both convection terms and the diffusion term are in balance. Then, the steady-state boundary-layer thickness scales as ( D L / j O 1/3, and the steady-state permeate flux scales as (D2j,/L) 1/3. However, in the current application of rapid backpulsing, we are interested in the shorttime transient development of the boundary layer, during which the flux remains equal to Jo and the appropriate boundary-layer thickness and time scales are 8 = D / J o and t c = D ( c c - - C b ) / / J 2 C b , r e s p e c tively.

At this point, a scaling analysis and order-of-magnitude estimates are in order. The convective flux toward the membrane and the diffusive flux away from the membrane (the second and fourth terms in Eq. (8), respectively) must be in balance, at least near the membrane surface. This requires that the boundary layer thickness scales as 8 = D / J o. The characteristic time for enough particles to be convected into the boundary layer so that the concentra-

Table 1 provides typical parameter values for microfiltration and ultrafiltration. For microfiltration, particles or cells with a diameter of 5 p~m are chosen, and shear-induced diffusion is dominant over Brownian diffusion. An approximate expression for the average shear-induced diffusion coefficient is [3] D = 0.03@a 2 (9)

Table 1 Typical parameter values for microfiltration and ultrafiltration

Jo (cm/s) Microfiltration Ultrafiltration 0.01 0.003


O
(cm2/s) Microfiltration Ultrafiltration 6 X 10 -6 4 )< 10 -7

Uo (cm/s) 100 100


6
(Ixm) 6 1.4

Ho (cm) 0.1 0.1


Cb/C c
0.05 0.05

L (cm) 50 50
tc
(s) 1.1 0.9

~/ (s 1) 3000 3000
J3oL/O2
( s - 1) 1 X 106 7 X 106

a (Ixm) 2.5 0.005 1/3


(D2"~/L)
(cm/s) 1 X 10 -3 2 )< 10 -4

234

s. Redkar et al./ Journal of Membrane Science 121 (1996) 229-242 x/L, ~= tJo2/D, and ~ = c / c b. They were then solved using a finite element code written by Professor R.L. Sani at the University of Colorado. The bounds of the region over which the equations were solved are 0 < ~ < 1 and 0 < ~ < Y m a x . The upper bound Ymax was in the range 5-50, selected such that ? = 1 with 0 ~ / 0 ~ = 0 at ~ = Ymax" This region was divided into 20 equally spaced elements in the direction and 50 equally spaced elements in the ~9 direction. The code uses an adjustable time step which is set automatically to keep the error small. A typical cycle of forward and reverse filtration required 6 0 - 8 0 time steps. The forward filtration equation [nondimensional version of Eq. (8)] was solved until the concentration at the membrane surface (~ = 0) reached the maximum packing concentration (co = CJCb) for the particles or solutes, which first occurs at the filter exit (2 = 1). Then, the reverse filtration equation [nondimensional version of Eq. (11)] was solved for a specified backpulse duration. After this, the forward filtration equation was solved once again, using the concentration profile at the end of reverse filtration as the new initial condition. The cycle of solving the forward filtration equation until just before a cake layer would start forming, followed by solving the reserve filtration equation for a fixed dimensionless time, t'b, was repeated until the dimensionless forward filtration time, ~f, reached a steady value. When this occurred, the concentration profiles also became periodic. The forward filtration time identified by this procedure is the critical forward filtration time, which represents the maximum duration of forward filtration which may be allowed without a cake or gel layer forming. This procedure was repeated for a broad range of values of the reverse filtration time, so that the optimum backpulse duration and frequency could be identified. The other parameters that were varied are the dimensionless shear rate, ~ = "~D2/j3L, the ratio of reverse and forward transmembrane pressure, a = A P b / A pf, and the ratio of the cake or gel concentration and the bulk concentration, Oc = c J c b. In order to check the accuracy of the calculations, the element size in the ~ direction was doubled for several of the calculations; in no case did the critical forward filtration time change by more than 0.01%.

where a is the particle radius. For ultrafiltration, macromolecules with a Stokes-Einstein diameter of 10 nm are chosen, and Brownian diffusion is dominant:

kT
U - - 67r/x o a (10) where k = 1.38 10 -16 g cm2//s 2 K is the Boltzmann constant and T is the absolute temperature. A typical shear rate of ~ = 3000 s - ~ is chosen for each case, and the suspending fluid is water at 20C. The clean membrane flux is higher for the microporous membrane because of its higher permeability. In both cases, the condition of a thin boundary layer, 6 << H o, is met. For microfiltration, however, the boundarylayer thickness is only slightly greater than the particle size, and so the continuum approximation is in doubt. Since ~/<<J3oLCb/D2(Cc--Cb), a stagnant cake or gel layer will form if backpulsing is not employed, and the scale for the long-term flux, (DZ~//L) 1/3, is then an order-of-magnitude smaller than the initial flux. Of primary importance is that the time scale for development of the concentration polarization layer, t~, is approximately 1 s in each case, indicating that rapid backpulsing with a frequency on the order of 1 Hz is needed to prevent cake or gel formation. Indeed, backpulsing frequencies of approximately 1 Hz were employed previously for ultrafiltration [10-12]. For the reverse filtration portion of each backpulse cycle, Eq. (8) is replaced by

Oc Oc Oc 02c --O + a J ~y + Y Y -~x = D --oy 2 t

(11)

where it is assumed that the membrane resistance is independent of the direction of the flow through it, so that the negative flux during reverse filtration has magnitude Jb = APb/IZoRm = lJo" The initial concentration profile for reverse filtration is the final concentration profile for forward filtration. The boundary conditions during reverse filtration are c = c b as y-+o% - a j o c + D o c / a y = O at y = O , and C=Cb at x = 0 .

2.3. Solution procedure


The governing equations were first nondimensionalized using the scaling analysis: ~ = yJo/D, ~ =

S. Redkar et a l . / Journal of Membrane Science 121 (1996) 229-242

235
' . . . .
' . . . . J . . . .

3. Theoretical results and discussion


3.1. Local concentration polarization and depolarization profiles

100

. . . .

' . . . .

. . . .

80

o= 6o
Fig. 4 shows the dimensionless concentration plotted versus the dimensionless distance away from the membrane at 2 = 1.0 (filter exit) for the forward filtration portion of the first cycle, with Oc = 100, and ~ = 0.00005. Before the start of forward filtration, ? = 1.0 everywhere. As forward filtration proceeds, the particles or solutes accumulate near the membrane (concentration polarization). As the concentration of the particles or solutes builds up, diffusion away from the membrane becomes significant, and the net motion of the panicles or solutes towards the membrane slows down. Forward filtration is stopped at the time t~tit = 90.4 D / J 2, which is when c = co at y = 0, x = L . Note that this time is very close to the characteristic time scale, t~ = D ( c c Cb)//J#Cb , derived from scaling arguments. The reverse filtration equation was solved for a = 1 using the final concentration profile at the end of the first forward filtration cycle as the initial condition. As reverse filtration proceeds, the concentration profile relaxes (concentration depolarization), as shown in Fig. 5. Immediately after the reverse filtration is started, the diffusion of the particles or solutes away from the membrane due to the concentration gradient aids the convection due to the reverse permeate flow. As time proceeds, however, the
100 . . . . . . . . ' . . . . . . . . ' . . . . . . . . . '
. . . . . . . . . ' . . . . . . . .

roll 40 i'~
2G / ' \

10

15

20

25

30

~" = y J o / D
Fig. 5. Concentration depolarization at 2 = 1 during reverse filtration for the first cycle, with ~ = 0.005, a = 1, and ~c = 100; the different curves are for dimensionless backpulse durations of t'b = 0, 0.2, 1.0, 2.0, 5.0, 10, and 15 (as the m a x i m u m m o v e s f r o m left to right).

~
II

6o
4o

20 , 0

- , '~

panicles or solutes move further away from the membrane and the diffusion slows down. Hence, the relaxation of the concentration profile is fast during the beginning pan of reverse filtration but slows down as time proceeds. For long reverse filtration times, the reverse permeate flow causes a depletion layer to form and expand further into the channel, so that axial convection becomes more important. For very long backpulse times, a steady state will be reached, with the depletion layer thickness scaling as ~d = ( t J o L / ~ / ) 1 / 2 =(c~/)1/26 from a balance of axial and transverse convection. For the typical case of (c~/~) 1/2 >> 1, this is larger than the boundary layer thickness scale of 8 = D / J o for forward filtration. Similarly, the characteristic time for the depletion layer to form during reverse filtration is of order t d = ( L / O l J o ~ / ) 1/2 = ( c e / ~ / ) l / Z D / J 2 o , which is large compared to the characteristic time for concentration polarization, t c = D ( C c -- C b ) / / J 2 C b , only if ( a / ~ / ) l / 2 C b / ( C c -- Cb) >> 1. In practice, however, the backpulse duration should be kept relatively short (in order to minimize permeate loss), and so a depletion layer is not established.
3.2. Periodic steady state

~" : yJo/U
Fig. 4. Concentration polarization at 2 = 1 during f o r w a r d filtration for the first cycle, with ~, = 0 . 0 0 0 0 5 and 8c = 100; the different curves are for 7 = 0, 10, 20, 30, 40, 60, 80, and 90.4 (bottom to top).

Repeated cycles of forward filtration followed by reverse filtration lead to a periodic steady state, in which the concentration polarization and depolarization profiles for one cycle are identical to those of

236
20

S. Redkar et al./Journal of Membrane Science 121 (1996) 229-242


20 , . . . .
i . . . . i . . . . i . . . . i

. . . .

u 15

u o o o o o o D o u o u u o u o u o o o u o o o o

tO

lO

10 o o o o

o o

II (tO
5

o o

o o o o o o o o o o o o o o o o o o o

II
<%_~

. . . .

. . . .

. . . . . . . . . . . . .

10

10

15

20

25

5O

F = YJo/D

cycle n u m b e r Fig. 7. Critical forward filtration time versus cycle number for backpulsing with #f = 0.005, a = 1, and ~c = 19.3; the four sets of results are f o r ~ b = 0.2 (A), 2.0 (*), 5.0 (o), and 10 (D). slopes of the forward filtration time versus reverse filtration time curves increase with increasing shear rate, because the axial flow is more effective in carrying the depolarized particles or solutes out of the filter channel during reverse filtration at higher shear rates. In fact, w h e n ~ > 0.063, it was found from the n u m e r i c a l results that the concentration of particles or solutes does not reach the critical concentration for cake or gel formation at the m e m b r a n e surface, even for very long forward filtration times. The corresponding d i m e n s i o n a l critical shear rate, 3'c = 0.063 J 3 o L / D 2 , is very close to the scaling estimate, J 3o C b / D 2 ( c c - Cb). L For long backpulse durations, the critical forward
70
. . . . . . . . . i . . . . . . . . . i . . . . . . . . . i . . . . . . . . . i . . . . . . . . . i . . . . . . . . . i...

Fig. 6. The concentration profiles at the end of forward filtration for the first (solid) and second (dashed) cycles and at the end of reverse filtration for the first (dotted) and second (dashed-dotted) cycles of backpulsing, with ~, = 0.005, a = 1, ~c = 19.3, and ~b= 2. the previous cycle. The critical periodic steady state (where cake formation is just prevented by backpulsing) was f o u n d by fixing the duration of reverse filtration while allowing forward filtration to continue until the concentration at the m e m b r a n e surface reached the m a x i m u m value for incipient cake or gel formation. Fig. 6 shows the concentration profiles at the end of forward filtration and at the end of reverse filtration at 2 = 1.0 for the first two cycles, with a = 1, ~ b ~--- 2, ~ = 19.3, and / = 0.005. The c o n c e n tration profiles reach a near periodic state in just two cycles. Fig. 7 illustrates the change in the forward filtration time with the cycle n u m b e r for different reverse filtration times o f ~ b = 0 . 2 , 2 . 0 , 5.0 and 10, with a = 1, ~c = 19.3, and / = 0.005. For each case, the forward filtration time reaches a nearly constant value after several cycles. This value is the critical forward filtration time, t~~it, as it represents the longest duration of forward filtration, for a particular reverse filtration time, without a cake or gel layer forming o n the m e m b r a n e surface. Fig. 8 is a plot of the dimensionless critical forward filtration time, ~rit, once a periodic steady state is reached, versus d i m e n s i o n l e s s reverse filtration time, t'b, for different values of the d i m e n s i o n less shear rate, ~,, with c~ = 1 and ~c = 19.3. The critical forward filtration time increases with increasing backpulse time, because longer backpulses give thicker depolarization or depletion layers which take longer to refilter during forward filtration. The initial

rm

50

40

% 3o II
2O

,o
0 ........ i .........

"

/
p

t .........

i .........

i .........

i .........

i~,,

10

20

30

40

50

60

~b = % tb/D
Fig. 8. Steady-state critical forward filtration time versus back-

pulse duration for backpulsing with a = 1 and ~c = 19.3; the different curves are for "~= 0.0005 (solid), 0.005 (dotted), 0.025 (dashed), and 0.05 (dashed-dotted).

S. Redkar et al./Journal of Membrane Science 121 (1996) 229-242


,35

237

50 ~

,'-"..~.". / .-

S~S

:::::::::::::::::::::::::::::::::::::::::

~_~ ~_)o

20 15 10

/ / / /" // /"/ /

/// // /

,'i';/ ,"

0 ,

'

20

40

60

80

1O0

Fig. 9. Steady-state critical forward filtration time versus backpulse duration for backpulsing with ~ = 0.005 and 8c = 19.3; the different curves are for & = 0, 0.025, 0.5, 0.75, and 1.0 (bottom to top). filtration times in Fig. 8 become independent of the backpulse duration. This is because the depletionlayer profile reaches a steady state for backpulse durations longer than approximately t b = t d, or ?b = ( a / ~ ) 1/2. T h e depletion layer for long backpulse durations is thinner for higher shear rates, scaling as 6 d = ( a / ~ / ) l / 2 D / J o . Thus, the critical forward filtration time to refilter this layer is lower for higher shear rates. Fig. 9 shows ~rit versus ?b for different values of oz = A P b / A P f , with 8c = 19.3 and ~ = 0.005. W h e n a = 0, there is no permeate lost and no convection of particles or solutes away from the membrane during reverse filtration. The concentration profile then re120
1 O0 C'~ / / /~ . . . . . . . . . . . . . . . . . . . . . .

laxes only by the back-diffusion of the particles or solutes. Since this relaxation is slow, relatively little depolarization occurs during the duration t b, and a smaller value o f t~tit is needed to reestablish the polarization layer during forward filtration. As a is increased, the extent of depolarization or cleaning achieved during reverse filtration for the same t b is increased. Hence, the concentration polarization layer requires a longer time to redevelop during the next forward filtration period, and s o t~tit for a particular t b increases with increasing a . Fig. 10 is a plot of ~rit versus ~b for different values of 8 C = c J c b, with a = 1.0 and ~ = 0.005. As the bulk or feed concentration is decreased relative to the gel or cake concentration of the solutes or particulates, the dimensionless time taken to first form a gel or cake layer increases in agreement with the scaling estimate, ~'~ = (c c - C b ) / C b . These findings indicate that backpulsing is more efficient for dilute suspensions, which has been observed in experiments [8,9].
3.3. A v e r a g e f l u x

E
60 /

<~+-~ 20

/// "

10

20 A t"b ---- J~ f b / D

30

40

Fig. lO. Steady-state critical f o r w a r d filtration time versus back-

pulse duration for backpulsing with oz = 1 and "~= 0.005; the different curves are for 8c = 10, 19.3, 45, and 100 (bottom to top).

W e showed in the previous section that the critical forward filtration time before a cake or gel layer begins to form increases with increasing backpulse duration, increasing reverse transmembrane pressure during backpulsing, increasing (decreasing) shear rate for short (long) backpulse durations, and decreasing bulk concentration of particles or solutions. However, a longer critical forward filtration time does not necessarily give higher average or net flux for the cycle, since the time and permeate loss associated with the reverse filtration portion of the cycle must be accounted for. Figs. 11-13 give the net flux over one backpulsing cycle, once a periodic steady state is reached, as a function of the backpulse duration for different shear rates, reverse transmembrane pressures, and bulk concentrations, respectively. The curves were computed from Eq. (3) with tf = I fc r i ! . In general, short backpulse durations, low bulk concentrations, and high shear rates give the highest average flux. The ratio of transmembrane pressures has little effect for short backpulse durations. For long backpulse durations, the significant fluid loss during reverse filtration gives low or even negative net flux, espe-

238
1.0

S. Redkar et al./ Journal of Membrane Science 121 (1996) 229-242


1.0

0.8

0.8

0.6

~)o

0.6

.\\ A ~ 0.4 A V 0.4

0.2

0.2

0.0

...............
5 10 15

?:-.i: ......
20 25

~ , W . , ~. . . . . .
50 55

0.0
20

-,

40

60

~'b

J~ofb/D
Fig. 13. Net flux per cycle once a periodic steady state is reached versus backpulse duration, for ~f = ~-~,it,~, = 0.005, and ~ = 1; the different curves are for ~ = 10, 19.3, 45, and 100 (bottom to top).

Fig. 11. Net flux per cycle once a periodic steady state is reached versus backpulse duration, for t'f = ~it, c~ = 1, and ~c = 19.3; the different curves are for ~, = 0.0005 (solid), 0.005 (dotted), 0.025 (dashed), and 0.05 (dashed-dotted). cially at high reverse t r a n s m e m b r a n e pressures. For short backpulse durations, concentration polarization during forward filtration occurs more slowly than does concentration depolarization during reverse filtration. This is because diffusion opposes the convection toward the m e m b r a n e during forward filtration, whereas it aids the c o n v e c t i o n away from the m e m brane during reverse filtration for short times. As a result, a positive net flux which is a significant fraction of the clean m e m b r a n e flux is achieved with rapid backpulsing. O f course, if the shear rate is too low, the depolarized particles are not effectively swept out of the channel during the brief periods of
0.50

reverse filtration, and so a high flux is not achieved at low shear rates. For dead end filtration ( ~ = 0), no periodic steady state with a positive net flux is possible. The n o n m o n o t o n i c behavior of the curves for short backpulse times in Figs. 1 1 - 1 3 is real, and not the result of n u m e r i c a l errors; the global maxim u m net flux is at t'b ~ 0 for some conditions, whereas it is at a small but finite value of t'b for other conditions.

4. Experiments
Experiments were performed on suspensions of washed S. cerevisiae yeast cells in deionized water. The cells have a m e d i a n radius of a = 2 . 1 txm (measured with a Coulter Multisizer) and a specific cake resistance of /~c = 0.8 10 l cm - 2 , as determ i n e d from the rate of flux decline for forward filtration only [9]. A dilute concentration of c b = 0.0026 g / m l on a dry weight basis was employed, which yields 0.78% cells by v o l u m e [9]. Since yeast cakes are approximately 78% cells by v o l u m e [9], this corresponds to ~c = 100. The experimental setup is a modified version of that described by Redkar and Davis [9] (see Fig. 14). The feed reservoir was pressurized to 5 psia (34 kPa), which gave a clean m e m b r a n e flux of Jo = 0.022 c m / s (790 1 / m 2 h). The corresponding time constant for cake growth during dead end filtration is ~-= 4.1 s. A M i n i t a n - S flat-plate m e m b r a n e m o d u l e

0.40

0.10

".

- -

0,00 20 40

I 60

f'b

J'o "i'b/D

Fig. 12. Net flux per cycle once a periodic steady state is reached versus backpulse duration, for ~'f= ~rit, ~ = 0.005, and ~c = 19.3; the different curves are for a = 0 (solid), 0.25 (dashed), 0.5 (dashed-dotted), 0.75 (dotted), and 1.0 (dashed-triple dotted).

S. Redkar et al./Journal of Membrane Science 121 (1996) 229-242

239

regeneration reservoir regulator valve

regeneration " ( ~ r pump feed reservoir

I~-ol
I valve (A) 3ressure I

backpulse reservoir

I
I

~ - tFanstlucet"

pre eyli

re ~r electronic balance

membrane I module [

solenoid valve (B) personal computer

'1I
H
I

pre: cyli

er I

I i

permeate
reservoir electronic balance

i--1 l-jJ F--q

Fig. 14. Schematic of the experimental apparatus. made by Millipore Systems was employed; this module has nine parallel channels, each 0.4 mm high X 7 mm wide 50 m m long. A cellulose-acetate membrane manufactured by Satorious, with an average pore size of 0.07 Ixm, was used in this module. The feed rate was fixed at Q = 6 m l / s , which gives a shear rate at the membrane surface of ~/= 3600 s -1 . The backpulse reservoir was pressurized to 10 psia (5 psi above the feed reservoir), so that c~ = 1. Pressures are maintained using high-pressure nitrogen cylinders and regulator valves. Solenoid valves controlled by a microcomputer are used to switch between forward and reverse filtration. Both the backpulse and permeate reservoirs sit on electronic microbalances interfaced to the computer. A regenerator pump controlled by the computer keeps the yeast concentration in the feed reservoir constant by continually replacing the net permeate lost with deionized water. Backpulse times of t b = 0.1, 0.2, and 0.3 s were employed, which yield predicted critical forward filtration times before a cake begins to form of 0.29, 0.67, and 1.0 s, respectively. The forward filtration time was varied for each set of experiments, in order to determine the optional forward filtration time and maximum net flux. The duration of each experiment was 1000 s. The experimental results are given in Figs. 15-17 for t b = 0.1 s, 0.2 s, and 0.3 s, respectively. In all cases, the symbols are the global average net flux over the durations of three repeated experiments, and the error bars represent plus and minus one standard deviation. The net flux values were computed by taking the fluid gained in the permeate reservoir, subtracting the fluid lost from the backpulse reservoir, and dividing by the membrane area and the duration of the experiment. Small declines in the net flux of about 10% were observed over the duration (1000 s) of a typical backpulse experiment, apparently due to membrane compaction and incomplete removal of the fouling deposit. The mass measure-

240

S. Redkar et al. / Journal of Membrane Science 121 (1996) 229-242

o o2oi
CD tO 0.015

0.020

1 ........

' .........

I .........

i ........

E
,~OA 0"010 1 "~ 0.005 '~ '~

o.oofI
E T: V 0.005
0.000
0

l
, ......... I , ......... 2 , ......... 3 , ......... 4 , ......... 5

0.000 0

....... , ......... , ......... , ......... ~......... , ......... , ..... ,,, 1 2 .5 4 5 6

.i .......

,..,
6

....
7

t, (s~c)
Fig. 15. Global average net flux versus forward filtrationtime for a fixed backpulse durationof tb = 0.1 s, with AP b = ~ P f = 5 psi (34 kPa), j, = 3600 s- l, and cb = 2.6 g yeast/1 on a dry-weight basis.

f, (sec)
Fig. 16. Global average net flux versus forward filtrationtime for a fixed backpulse duration of tb = 0.2 s, with AP b = APf = 5 psi (34 kPa), ~, = 3600 s -] , and cb = 2.6 g yeast/l on a dry-weight basis.

ments from the two electronic balances showed that some fluid bypassed directly from the backpulse reservoir to the permeate reservoir; this could likely be minimized by incorporating a small delay in opening the valve (B) to the permeate reservoir after closing the valve (A) from the backpulse reservoir at the end of each backpulse [6]. The dotted lines in Figs. 15-17 are the predicted net fluxes from Eq. (3) for tf < t~tit, and the solid lines are the predicted net fluxes from Eq. (8) for tf > t~tit. In all cases, good agreement between theory and experiment is observed, except at long forward filtration times. For each backpulse time, there is an optimum forward filtration time which maximizes the flux. Shorter durations of forward filtration do not allow for sufficient collection of permeate relative to that lost during reverse filtration, whereas longer durations of forward filtration lead to flux decline due to cake buildup. For the longest periods

of forward filtration, the experimental data fall below the predictions of the theory; this is likely the result of the yeast cakes not being completely removed during backpulsing. In addition, the actual backpulse time for the lowest setting may be slightly longer than t b = 0.1 s, because the solenoid valves have a response time of about 10 ms and the computer clock has a response time of about 50 ms [8]. This may account for the slightly lower experimental values than the predicted values of the net flux in Fig. 15. Table 2 gives the predicted and experimental values of the maximum net fluxes and the corresponding optimal forward filtration times, The optimal forward filtration time increases with increasing backpulse duration, whereas the maximum net flux is relatively insensitive to the backpulse duration. Most importantly, the net fluxes obtained with backpulsing

Table 2 Theoretical experimentalmaximumfluxes the correspondingoptimumforward filtrationtimes Theoretical <j>rnax (cm/s) tb = 0.1 s t b = 0.2 s t b = 0.3 s 0.019 0.018 0.017
t~ pt (S)

Experimental <j>max (cm/s) 0.018 0.019 0.019

t?pt (S)

Jo (cm/s) 0.022 0.022 0.022

J~ (cm/s) 0.0026 0.0026 0.0026

2.1 3.1 3.8

1.5 3.0 5.0

S. Redkar et al. / Journal of Membrane Science 121 (1996) 229-242


' ' ' i , ,

241

0.020

II~ 0 . 0 1 5 O3

E
,~,0 0.010

/k
0 . 0 0 5

0.000 0

I 2

i 4

I 6

t 8

theory predicts that the forward filtration time which maximizes the net flux is, in general, slightly longer than the critical value which leads to cake or gel formation. However, the present theory includes the assumption that the fouling layer is instantly and completely removed from the membrane surface by reverse filtration. This assumption may not be good for adhesive foulants; thus, an important area for additional theory and experiments is rapid backpulsing with foulants which adhere to the exterior or interior surfaces of the membrane.

t, (sec)
Fig. 17. Global average net f l u x versus f o r w a r d filtration time for a fixed hackpulse duration o f t b = 0.3 s, with A P b = A p f = 5 psi

Acknowledgements
This research was supported by the National Science Foundation (grant CTS-9107703), the Water Treatment Technology Program of the Bureau of Reclamation (contract 1425-5-CR-81-20250), and the Center for Separations Using Thin Films at the University of Colorado.

(34 kPa), 4/= 3600 s - 1, and c b = 2.6 g yeast/1 on a dry-weight basis.

are close to the clean membrane flux (Jo), whereas the long-term flux in the absence of backpulsing (J~) is nearly one order-of-magnitude lower.

5. Concluding remarks
Numerical solution of the convection-diffusion equation for repeated cycles of concentration polarization during forward filtration and concentration depolarization during reverse fltration with transmembrane pressure reversal has shown that rapid backpulsing is predicted to prevent the formation of a cake or gel layer on the membrane surface, provided that the duration of forward filtration does not exceed a critical value. This critical value is quite short, on the order of one second for typical microfiltration and ultrafiltration operations, and so short backpulses (on the order of one-tenth of a second in duration) with a frequency of approximately 1 Hz are needed to prevent membrane fouling. The critical forward filtration time scales as D(c c -eb)/JoZcb, and so the required frequency of backpulsing is reduced for dilute suspensions and low clean membrane fluxes. If the forward filtration time is greater than the critical value, then a cake or gel layer begins to form on the surface of the membrane. Although this fouling layer reduces the forward filtration flux, the

References
[1] G. Belfort, R.H. Davis and A.L. Zydney, The behavior of suspensions and macromolecular solutions in crossflow microfiltration, J. Membrane Sci., 96 (1994) 1-58. [2] W.F. Blatt, A. Dravid, A.S. Michaels and L. Nelson, 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, pp. 47-97. [3] R.H. Davis, Modeling of fouling of crossflow microfiltration membranes, Sep. Purif. Methods, 21 (1992) 75-126. [4] D.A. Drew, J.A. Schonberg and G. Belfort, Lateral inertial migration of a small sphere in fast laminar flow through a membrane duct, Chem. Eng. Sci., 46 (1991) 3219-3224. [5] G. Johnson and I.G. Wenten, Control of concentration polarization, fouling, and protein transmission of microfiltration processes within the agro based industry, Workshop ASEAN-EU on Membrane Technology in Agro-based Industry, Kuala Lumpar, July 1994, pp. 157-166. [6] J.A. Levesley and M. Hoare, The effect of high frequency backflushing on the microfiltration of yeast homogenate suspensions for the recovery of soluble proteins, Biotechnol. Bioeng., under review. [7] A.S. Michaels, Fifteen years of uhrafiltration: problems and future promises of an adolescent technology, in A.L. Cooper (Ed.), Polymer Science and Technology, Plenum Press, New York. [8] C.S. Parnham and R.H. Davis, Protein recovery from bacte-

242

S. Redkar et al./Journal of Membrane Science 121 (1996) 229-242 pressure pulsing on concentration polarization, J. Membrane Sci., 68 (1992) 149-169. [12] V.G.J. Rodgers and R.E. Sparks, Effects of solution properties on polarization redevelopment and flux in pressure pulsed ultrafiltration, J. Membrane Sci., 78 (1993) 163-180. [13] C.A. Romero and R.H. Davis, Global model of crossflow microfiltration based on hydrodynamic diffusion, J. Membrane Sci., 39 (1988) 157-185.

rial cell debris using crossflow microfiltration with backpulsing, J. Membrane Sci., 118 (1996) 259-268. [9] S.G. Redkar and R.H. Davis, Enhancement of crossflow microfiltration performance using high frequency reverse filtration, AIChE J., 41 (1995) 501-508. [10] V.G.J. Rodgers and R.E. Sparks, Reduction of membrane fouling in the ultrafiltration of binary protein mixtures, AIChE J., 37 (1991) 1517-1528. [11] V.G.J. Rodgers and R.E. Sparks, Effects of transmembrane

You might also like