Sunteți pe pagina 1din 15

j o u r n a l of MEMBRANE SCIENCE

ELSEVIER
Journal of Membrane Science 136 (1997) 191-205

Modeling of transient permeate flux in cross-flow membrane filtration incorporating multiple particle transport mechanisms
S a n d e e p Sethl ' , M a r k R. W i e s n e r
Department of Environmental Science and Engineering, Rice University, 6100 Main Street, Houston, TX 77005, USA
Received 16 January 1997; received in revised form 17 June 1997; accepted 30 June 1997 -1"

Abstract
Dominant mechanisms of particle transport in cross-flow membrane filtration are unified to obtain a generalized model for time-dependent permeate flux. The unified model extends an earlier model based on shear-induced diffusion and a concentrated flowing layer to include Brownian diffusion and inertial lift. It is applicable over a broad range of contaminant sizes encompassing macromolecules, colloidal and fine particles, and large particles. The combined theory predicts an unfavorable particle size, of the order of 10 1 ~tm, where the net back-transport away from the membrane attains a minimum, leading to maximum cake growth. For the system simulated in this work, this implies minimum permeate fluxes in the size range of 0.01~0.1 ~tm, depending on the operating time. Inside-out hollow-fiber geometry is predicted to be favorable for feed suspensions with small particles and/or low concentrations which form thin resistive cakes. However, larger particles, which form thick cakes, may result in reduced surface area available for filtration due to curvature effects in inside-out membranes, making the slit or outside-in geometry more favorable for these particles. Fine particles (< 0.1 ~tm) are predicted to demonstrate mass-transport limited behavior. For larger particles, different combinations of fiber radius and cross-flow velocity, resulting in the same shear rate, demonstrate different permeate fluxes.

Keywords: Transient model; Ultrafiltration; Microfiltration; Theory; Permeate flux

1. Introduction
The semi-permeable synthetic membranes employed in ultrafiltration (UF) and microfiltration (MF) processes roughly correspond to effective pore sizes from ca. 0.001 to 0.1 ~tm for U F and 0.1 to 10 ~tm for M E Thus, U F and M F are capable of removing
*Corresponding author. Tel.: +1 404-881-8010; fax: +1 404-8723161; e-mail: sandeep sethi@air-water.com. 1Present address: Metcalf and Eddy, Inc., 1201 Peachtree Street, N.E., 400 Colony Square, Suite 1101, Atlanta, GA 30361, USA. 0376-7388/97/$17.00 1997 Elsevier Science B.V. All rights reserved. PII S0376-7388(97)00 168-3

species in a wide range of sizes, including macromolecules, colloids, fine particles, and cells, from suspensions. An important performance parameter in the design, operation and evaluation of membrane filtration processes is the permeation rate or the permeate flux. Quantitative estimation of permeate flux requires consideration of the different mass-transport mechanisms involved. Various theories of particle transport in low-pressure cross-flow membrane filtration have been put forward over the previous three decades to describe permeate flux [1-22]. These include, among others, the models based on Brownian

192

S. Sethi, M.R. Wiesner/Journal of Membrane Science 136 (1997) 191-205

diffusion [1-6], inertial lift [7-11], shear-induced diffusion [12-17], concentrated flowing layers [ 18,19], surface transport [20,21 ], and surface renewal [22]. Recently, a number of excellent reviews on these and other theories, applicable to low-pressure filtration, have appeared in the literature [23-26]. Some of these reviews also assess the predictive capabilities of various theories by comparison with published experimental data. Such comparisons have indicated that the predictive performance of a model is very much a function of the feed suspension in question. Among the theories requiring the least number of empirical parameters, the gel-polarization theory based on Brownian diffusion [1-5] and the integral theory based on shear-induced diffusion [13-16] appear to yield reasonable predictions for permeate flux in the cases of macromolecular suspensions [3,4] and suspensions containing colloids, fine particles and cells [17,27], respectively. The effect of inertial lift on the trajectories of larger particles has been tested in a number of studies [7-9,28]. The applicability of previous models tends to be restricted to a particular type of feed suspension due to the fact that these models evoke a single back-transport mechanism which is considered to be dominant for the feed suspension in question. However, in many cases several different transport mechanisms may be important for a particular suspension. Hence, predicting and comparing permeate fluxes for feed suspensions containing materials which may vary by several orders of magnitude in effective size calls for consideration of multiple transport mechanisms in masstransfer modeling. The applicability of previous models can be improved by simultaneously considering the effects of Brownian diffusion, shear-induced diffusion, inertial lift and concentrated flowing layers on particle accumulation near the membrane. In a recent work [29], we extended the steady-state model of Romero and Davis [14] to include Brownian diffusion as a particle transport mechanism. One key prediction was that species in the size range of 10 -~ tam should exhibit unfavorable accumulation on the membrane surface due to a minimum in net diffusivity. Consequently, species falling in this unfavorable size range are predicted to produce minimum permeate flux, while species either smaller or larger than this intermediate size are predicted to produce higher permeate fluxes. Until recently, there was only indirect experi-

mental evidence for this hypothesized minimum in back-transport as interpreted from permeate flux data [30-32]. However, the first direct experimental confirmation of the existence of a minimum in backtransport based on measurements of particle residence-time distributions was recently reported [33]. In this work, we undertake the synthesis of a unified transient model for cross-flow membrane filtration that considers several particle transport mechanisms acting in concert. The approach to formulating the involved mass balances extends that employed in the transient shear-induced diffusion model of Romero and Davis [15]. Theoretical comparisons are made between the flat-slit and the inside-out cylindrical membrane geometries by applying the model to a range of particle sizes and analyzing the cake growth and permeate flux behavior. Because the flat-slit geometry lies at the boundary between concave inside-out and convex outside-in depositional surfaces for hollow-fiber membranes, conclusions are drawn on the efficacy of these two hollow-fiber geometries for feed suspensions which differ in particle size and concentration. Results also include analysis of the effects of transmembrane pressure, shear rate, cross-flow velocity, and membrane diameter on the permeate flux decline for the inside-out cylindrical membrane. The model presented in this work may be used to compare the permeate flux behavior of various monodisperse suspensions, to investigate the relative impact of different particle size fractions in a single polydisperse suspension, or to approximate the permeate flux behavior of polydisperse suspensions characterized by an average/equivalent particle size. However, it should be noted that the model is ideally applicable to monodisperse suspensions and the effects of shear-induced diffusion for polydisperse suspensions may be quite different.

2. Background
2.1. Brownian diffusion
The Brownian diffusion coefficient for a spherical particle can be estimated by the Stokes-Einstein equation: DB -- - 67r/z0ap

kT

(1)

S. Sethi, M.R. Wiesner/Journal of Membrane Science 136 (1997) 191-205

193

where ap is the particle radius, k the Boltzmann's constant, #o the dynamic viscosity, and T the absolute temperature. Brownian diffusivity is inversely proportional to particle size and, hence, becomes increasingly important as the latter decreases to sub-micron dimensions, such as for macromolecules. Transport models based on back-transport solely due to Brownian diffusion, such as the traditional gel-polarization model or "film theory" [1-4], predict flux values within 15%-30% for macromolecular solutions [4], but grossly underpredict fluxes for colloidal suspensions due to the low diffusivity of colloids and small particles. This discrepancy was originally termed the "flux paradox" by Belfort et al. [7].

where bs(~b) is a dimensionless function of the local particle volume fraction ~b: bs(~b) = 0.33~b2(1 + 0.5e 88~) (4)

Although the data exhibited considerable scatter, this empirical correlation was reported as valid for particle volume fractions up to ~b = 0.5.

2.4. Shear-induced diffusion and concentrated flowing layers


In subsequent work, Romero and Davis considered the combined effects of shear-induced diffusion [13,34,35] and concentrated flowing layers [18,19] to describe the steady-state [14] and transient [15] permeate flux in cross-flow MF. The latter concept assumes that particles enter a boundary layer which flows along the membrane surface due to tangential shear. The theory predicts formation of such a concentrated boundary layer, with nonlinear concentration and velocity profiles, flowing toward the filter exit. As particles accumulate in this layer, the local concentration reaches a threshold where the existing tangential velocity is insufficient to maintain a completely flowing layer. At this point, formation of a compact stagnant cake is predicted underneath the concentrated flowing layer which constricts the channel and, effectively, increases the shear rate. This stagnant cake provides significant resistance to permeation, in addition to that of the membrane. Cake growth is described by formulating a mass balance on the control volume, represented by the cake and flowing layers. The critical time and distance required for the establishment of the stagnant cake layer are also predicted. The cake growth is related to flux decline through the standard filtration theory, incorporating the membrane and cake resistances. Good correspondence between model calculation and experimental data, obtained using rigid micron-sized particles, has been published [17].

2.2. Inertial lift


An early attempt to resolve this "flux paradox" evoked the concept of inertial lift [7-9]. The basic premise of the inertial lift theory lies in that the presence of walls induces a lift on particles, directed away from the membrane surface, which effectively reduces the net particle transport toward the membrane. For fast laminar flows (Re >> 1), as is typical in membrane filtration, the maximum value of the lift velocity in a clean channel is given by [9,11]:

Vlo = O.036pa3pa/2/ #o

(2)

where ~/0 is the shear rate in the absence of a cake layer, and p the fluid density. Due to its direct scaling with the cube of particle size and square of shear rate, inertial lift is expected to be important for large particles and/or high shear rates. It is noted that Eq. (2) is strictly limited to spherical particles in dilute suspensions where particle-particle interactions are negligible.

2.3. Shear-induced diffusion


Shear-induced diffusion arises due to particle-particle interactions in a shear field, resulting in lateral migrations of particles from the instantaneous trajectories. The diffusivity coefficient for suspensions of rigid spherical particles, assumed in this work, is given by the correlations, established by Leighton and Acrivos [34,35], of the form: Ds = "~a2pbs(~) (3)

3. Model synthesis
We present an extended model for predicting the transient permeate flux in low-pressure membrane filtration processes by combining Brownian diffusion and inertial lift with the shear-induced diffusion and

194

S. Sethi, M.R. Wiesner/Journal of Membrane Science 136 (1997) 191-205

flowing boundary layer model formulated by Romero and Davis [15]. 3.1. Cake filtration law The resistances produced by the membrane and the cake which accumulates on the membrane surface are considered to act in series, and the permeate flux is then described by Darcy's law:

3.2. Cake growth and conservation of particle flux It has previously [14,15] been shown that the axial convection term and the transient term in the convective diffusion equation are negligible as long as the bulk suspension volume fraction is much lower than the characteristic volume fraction in the flowing layer. Under this assumption of dilute suspensions, the particle mass balance in the polarized layer is obtained by integration of a reduced form of the convective diffusion equation, which includes back-transport due to both Brownian as well as shear-induced diffusion: [DB + Ds(~)] ~yy + Vwq~= 0

j -

Ap #0(Rm + Rc)

(5)

where A p is the transmembrane pressure and R m and Rc the membrane and cake resistances, respectively. The resistance due to the concentrated flowing layer is assumed to be negligible compared to the stagnant cake resistance and is not included in Eq. (5). The cake resistance is given by [23]: Rc = R~c rectangular slit (6a)

(9)

Re = R*~Ho ln[Ho/ (Ho - ~c)] inside-out cylindrical fiber (6b)

where R c is the specific resistance of the cake layer, Ho the fiber radius or channel half-height, and (~cthe cake layer thickness. Eq. (6b) accounts for the change in cake area with radial growth due to curvature. The relationship between the permeate flux, J, and the permeate velocity, Vw, is given, from mass balance considerations, as [23]: Vw = J rectangular slit (7a)

Vw = JHo/(Ho - 6c)

inside-out cylindrical fiber (7b)

The specific resistance of the cake layer can be estimated using the Carman-Kozeny correlation [36]:
R~* -

45+c ~
a2(1
,c) 3

(8)

where y is the coordinate normal to the membrane surface. The assumption of a perfectly rejecting membrane has been made in Eq. (9). It has also been implicitly assumed that Brownian diffusion and shear-induced diffusion can be considered to act simultaneously by adding the diffusivity coefficients associated with the two phenomena. Inertial lift, being a far field mechanism, is not considered in the mass balance within the boundary layer. Transient accumulation of particles in the stagnant cake is described by the particle flux conservation law. When the entire flowing particle layer is considered as an elemental volume, then the net transverse convection of particles normal to the membrane surface is balanced by the axial convection tangential to the membrane surface. In this formulation, the axial convection term is now included but the diffusion term is negligible across the boundaries of the elemental volume and is not considered. Inertial lift effectively reduces the net transport of particles into the boundary layer. Hence, the net transport of particles towards the membrane can be represented by the vector sum of the permeation and inertial lift velocities, thus:

where 4~ is the maximum packing density or solidosity of the cake. If the particles are monodisperse, spherical and nondeformable q~c may achieve a value of 0.58 [35]. Polydisperse suspensions or suspensions of deformable particles usually produce cakes with higher solidosities.

o~

//
-4- ~

F6+6
(e
~)dy + (~c -

]
~)~

~+~

U(t~ -- (~b)dy : (~w -- V1)~o

(10)

S. Sethi, M.R. Wiesner/Journal of Membrane Science 136 (1997) 191-205

195

where 95b is the bulk suspension particle volume fraction. The first term in Eq. (10) describes particle accumulation in the flowing and cake layers, and the second term describes the "excess particle flux" [13] due to the cross flow. The local velocity, u, in Eq. (10) is evaluated from integration of the local velocity gradient in the polarized layer, given by [15]: d~ _ "Y= rw /ton(0) (1 1)

where 11 and/2 are integrals given by:

1'--L ,,(95)+6rr#oa~ ~b
/2

-waZ/)s(95)

kr

] (95 ~

95b) d95

(16)

+ E J6 aP'd')

where rw is the wall shear stress and ~/(95) the relative viscosity. This expression assumes that the suspension flow is fully-developed and steady, based on the fact that the time scale for a steady-state bulk profile is small compared to that required for the development of the concentrated layer at the membrane surface. Further, the polarized layer is assumed to be thin compared to the channel thickness, so that the y dependence of the shear stress can be neglected. The relative viscosity is empirically estimated as [34]: r/(95) = 1.595 ~2 1 + 1 --77qScJ (12)

x d951

(17)

11 and 12 can be expressed in terms of the cake thickness, using Eqs. (11) and (14), and component integrals which are a function of only the particle volume fraction, 95:
2 n 7-w0apH~ , kT 11 = #0(H0 - 5c)" 111 + 6--~0ap I12 _2 _ 4 u 2 n

and the shear stress is expressed as: rw = (n + 1)U#o~/(95b)/H (13)

(lS) (122 + 123)

12 =

~wUPn
k2T 2

where U is the average cross-flow velocity and H the effective fiber radius or channel half-height, calculated by accounting for the constriction of the fiber or channel due to cake formation. The shear rate, and consequently the shear stress, increases with axial distance due to constriction of the membrane channel as the cake builds up, hence: x/= 70 ~ H - - ~ _ ) ~

/ t 2 ( H 0 __ (5c)2n 21 - -

67r#2(~_--

kTv-woapH~
6"-c)n

-+ 367rZ#Zap2-I24

(19)

where Ii1, I12, 121, 22, 23, and 124 are component integrals given by the following expressions:
~w ^

.(

Ho "~"

(14)

/ D~(95)(95 -~ 95b)d95 1t, = v(95) ~b


q)w

(20a)

where n = 2 for channels with rectangular and n = 3 for channels with circular cross sections. This relationship is established from flow continuity and neglects the small fraction of fluid lost through the membrane. After substituting the expression for the local velocity, u, in Eq. (1 0), and then transforming the variable of integration from y to 95 (using Eq. (9)), the particle flux conservation can be expressed as: o--; ~ + (ec - 95u)~ + N \ / t o ~ ] (15)

112 = f (95 -- 95b)d95 95


~b

(20b)

b ' I21 = f f 951r/-7~--~)s(95) ,9 - 95b)~-~-~ ~p d~'('y /)s (95)d ~b 6 ~w~w f ~(95 ) d~'(d~ d~b

~w~w

(20c)

(20d)

196 ew ew
123 = jf j[

S. Sethi, M . R . W i e s n e r / J o u r n a l o f M e m b r a n e S c i e n c e 1 3 6 ( 1 9 9 7 ) 1 9 1 - 2 0 5

dO' ,v(,)

( -

bs()~

(20e)

the permeate flux and lift velocity, respectively, in the absence of the cake layer): (~111 + ~112)
w 2 kT

b
'w ~)w

de' de 124 = j f j f ~,~/(,) (0 - q~b)~ -

(20f)

tcr =

(Jm -- Vl0)q)b

(21)

The particle volume fraction at the wall, 0w, increases prior to the development of the stagnant cake layer, and achieves its maximum value, ~, at the onset of cake formation. The integrals 111 and I21 are identical to the integrals denoted as (5 and Q, respectively, in [15], whereas 112, 22, 23, and 124 arise due to the additional consideration of Brownian diffusivity. Numerical values of these integrals Eqs. (20a), (20b),(20c),(20d),(20e) and (20f), evaluated using the routine DQDAG [37], for a range of bulk suspension particle volume fractions, are presented in Fig. 1. As illustrated, these integrals decrease in value with increasing concentration. For dilute solutions, the evaluation of Eq. (15) is greatly simplified due to the fact that the integrals have approximately constant values. Thus, these integrals can be approximated as: Ill = 0.0282, 112 = 0.58, I21 = 9.84 l0 -5, 122 = 4.7 10 -3, 123 ---- 1.53 X 10 -3 and/24 = 0.207. The time required to establish the cake layer growth, or the critical time, t~, is given by integrating Eq.(10) from t = 0 to t = t c r , with 6 c = 0 , Vw = Jm and vl = vl0 (where Jm and vl0 represent

The distance required to establish the stagnant layer growth, or the critical length, xc~, is given by integrating Eq.(10) from x = 0 to X=Xcr, with ~5~=0, Vw = Jm, and Vl = Vlo: w 3 2 (~ #oJm
Xcr ---T2 .E2 a 4

121 + kTr~a(I22+I23)+~124) 67r#2 (Jm - Vl0)0b

k2T 2

(22) Using Eqs. (11)-(19) and Eqs. 20(a-f), Eq. (10) may be written as: O 7-wapH Ill-q-- 112 + (c -- (~b)rc Ot tZOVw(Ho--rc) n 67r#0apVw
2 2 kT

o[
+

( 'w a4"4 /2,


k2T2 )]

[ OV wIHo - 6cl n .

kTTwoapH~,
---- (Vw -- Vl)b

+ 67r#2(Ho - 6c)" (122 +/23) + 367r2#o2a~I24 (23)

0.6
0.5-

........

........

........

........

.......

x~o'

" ' ' ~

"

It is convenient to non-dimensionalize this governing equation by introducing the dimensionless variables defined in Table 1. Hence, Eq. (23) can be expanded for the fiat and inside-out cylindrical geometries and, subsequently, expressed in dimensionless form as: Flat slit (n = 2): 0 [IH (1 +_ 13~) ll2(1 +/3(5) ] O~ LPe~(1 _ ,~),, + Peb ~- (c - Ob)$ 3

_~ 0 . 4 5
== 0 . 3 > 0.2

,,, x ~o'
l,

E
:~ 0.1 I,~ x 10~

0 [ ( 121 122+123 + ~ (1 + fld)2 Pe2s( _ $)3, + PebPes(1 - S) 2n 1

0.0

. . . . . . . .

. . . . . . . .

. . . . . . . .

10-5

1 0 .4

10-3

10-2 Particle Volume

1 0 -~ Fraction (-)

10 0

-t Pe2(1 -- 8)"

/24 )][1

+ ~6
(24a)

Vlo ]

Bulk Suspension

Fig. 1. Values of sub-integrals over a range of bulk suspension particle volume fractions (=0.58).

Inside-out cylindrical hollow fiber (n = 3):

S. Sethi, M.R. Wiesner/Journal of Membrane Science 136 (1997) 191-205

197

Table 1 Definitions of dimensionless variables Dimensionless variable Description Dimensionless cake thickness Dimensionless resistance Dimensionless permeate velocity for fiat slit Dimensionless permeate velocity for inside-out hollow fiber Brownian diffusion based Peclet number Shear-induced diffusion based Peclet number Dimensionless inertial lift velocity (in absence of cake layer) Dimensionless time Dimensionless distance

= ec/HO
/3 = HoRc/R m f~w = Vw/J~ = 1/(1 + j36) f~w ~ Vw/Jm = 1/{(1 - 6)[1 +/31n (1/(1 - 6))]} P eb = 67rlzoapJmHo / k T Pes = #oJmHo/%o@ f~to = Vlo/Jm = Jm(t - tcr)/Ho = Jmt/Ho - tcr

o
o~/pes(1 - S)" +
0

]
Pe~ + ( ~ - ~)~
(

When either xcr _> L~ or 0j0 > 1, then no cake formation is predicted and the solution is simply:
~c(X, t) = 0, Vw(X, t) = Jm.

+~

+flln(__~]

\a-SJJ

\ P e s ( 1 - S ) 3n-~

I21

+ PebPes(1 - S)2.-2 + pe2(1

~)n-2)J

(1 _ ~) [1 +/31n(i.~)]

(1 _ ~)2~j ~b

Z,o.]

(24b)

The initial and boundary conditions associated with Eq. (24a) or Eq. (24b) are given by S=0,
~=0,

Eqs. 24(a,b) are first-order nonlinear hyperbolic partial differential equations which can be solved using any of the popular explicit or implicit numerical finite-difference schemes. For hyperbolic problems, the explicit schemes are generally better suited than implicit schemes. Among the explicit schemes, the MacCormack method based on a predictor-corrector methodology is usually recommended [38]. Using the chain rule of differentiation, Eqs. 24(a,b) are first recast as:

g - ~ + h~x = f

o~

o~

(28)

~<0

for2_>0

(25a) (25b) where the functions g, h and f are given in the Appendix A. Eq. (28) is then solved using the MacCormack numerical method. An appropriate Courant stability condition is employed to calculate the time steps.

5~ <__;~cr for i > 0

where the dimensionless forms of the critical time and distance are given by (26)

4. Results and discussion


(~ kc r : ~_ (122+123) .q_ J?,.~'~
~ Pe~] \PeZs

(1 - % ) b

(27)

Although it is convenient to solve the problem in dimensionless form, presentation of the results in this

198

S. Sethi, M.R. Wiesner/Journal of Membrane Science 136 (1997) 191-205

Table 2 Parameter values used in performing simulations, representing typical ranges of operating and design conditions Parameter Transmembrane pressure, A p (kPa) Bulk suspension volume fraction, Fiber radius or slit half-height, H0 (mm) Average cross-flow velocity at inlet, U0 (cm/s) Membrane length (cm) Membrane resistance, em (1/cm) Low value 50
--

Baseline value 100


0.001

High value 200


--

0.25 30
---

0.5 60
105 1.46 101

1.0 120
---

form is hindered by the fact that the dimensionless variables are closely related and it is not possible to vary one dimensionless parameter while keeping all others fixed. Hence, the results in this section are presented assuming typical dimensional values of the parameters. The simulations are performed assuming a continuous load of inflnent particles in the feed stream characterized by a single diameter and volume fraction, since the model ideally assumes a monodisperse suspension.

"~" 0.8
0.7 -~ o -

........ - --

.......

........

.......

Hollow-fiber Slit

Steady-State

~- 0 . 6 ,~ 0.5-

0.4-

-~ o.3-

~0.1N 0.0E
........ j ........ I ........ I

4.1. Effects of particle size and comparison between flat-slit and inside-out cylindrical geometries
Simulations were performed to study the effects of particle size on the cake thickness and permeate flux, for a typical flat-slit and an inside-out hollow fiber. Particle size was varied in the 0.001-10 ~tm range, while other system parameters were held constant at their baseline values defined in Table 2.

10-3

10" 10-~

10-1
Particle Diameter (pro)

100

101

Fig. 2. Particle size effects on the length-averaged cake thickness at various times for typical flat slit and inside-out hollow fiber filters.

4.1.1. Cake growth


Effect of particle size on cake growth is depicted by plotting the instantaneous length-averaged cake thickness (normalized with the channel radius or halfheight) against particle diameter at various times (Fig. 2). The unified model suggests an intermediate range of particle size where net transport of particles away from the membrane approaches a minimum and results in thick cake buildup. This occurs .~0.4 ~tm for the system simulated in this work. However, at small times (1000 s curve in Fig. 2) the maximum cake thickness actually occurs at particle diameters larger than 0.4 ~tm. This is due to the fact that at smaller times the relative size of individual particles is significant. At larger times, when the number of particles accumulated becomes significant, the peak shifts to its

steady-state value of 0.4 ~tm. The unfavorable size window is dynamic and moves with the kinetics of cake growth, as well as the hydrodynamics of the system. Thus, the relative composition of cakes formed from polydisperse suspensions is expected to change over time. The cake buildup at steady-state is less for a insideout hollow fiber than for a slit, for all particle diameters. This is due to the fact that cake growth in the inside-out cylindrical geometry causes greater reduction in cross-sectional area for suspension flow, leading to less permeate flux (discussed subsequently), and consequently less potential for cake growth. For the curve at 1000 s, however, the thicker cakes in the case of a slit for larger particles are actually due to the lower shear rate in the slit geometry. This was confirmed by performing simulations after adjusting the shear rate in the slit to be the same as in the case of the hollow fiber. Another consequence of the different

S. Sethi, M.R. Wiesner/Journal of Membrane Science 136 (1997) 191-205


S 0.8_ 0.7-

199

1.2 ~-1.0

s,.__

Hollow-fiber ]

#-- o.a.
_ 0.8 0.5-

~
~
E z

0.40.4

<, 0.3J:: =m 0.20.1.

~
E Z 101 102 103 Time (s) 104 10s 106

0.2-

"
o.o
10_3
i

~
i

Steady-State
i

m 0.0.
100

10.2

10.1 Particle Diameter (Ixm)

100

101

Fig. 3. C a k e g r o w t h for various particle sizes for typical fiat-slit a n d inside-out h o l l o w fiber filters.

Fig. 4. Particle size effects on the l e n g t h - a v e r a g e d p e r m e a t e flux at v a r i o u s t i m e s for typical fiat-slit and inside-out h o l l o w fiber filters.

geometries is that steady-state is achieved relatively faster in the hollow fiber than in the slit. The transient variation in the length-averaged cake thickness is illustrated in Fig. 3 where the cake thickness for various representative particle sizes is plotted against time. The inset depicts the cake thickness on a logarithmic scale. Small particles (0.005 ~tm, 0.01 pm), which form cakes with high specific resistance, demonstrate a relatively fast approach to steady state. These simulations also imply that the time required to reach a steady state is the longest for particles favoring accumulation (0.1 pm, 0.5txm), as is expected. The 1.0 lam particle has less cake buildup than the 0.5 lain particle due to shear-induced diffusion and inertial lift effects.

1,2 1.0 2ap=1.0 ~m

~. 0.8 ~0.6-

0.4-

~
E z

0.20.0 i . . . . . . . . l . . . . . . . . j . . . . . . . . j . . . . . . . . j . . . . . . . . i . . . . . . . 10o 101 102 103 104 10s 106

Time (s)

Fig. 5. P e r m e a t e flux d e c l i n e for v a r i o u s particle sizes for typical flat-slit a n d inside-out h o l l o w fiber filters.

4.1.2. Permeate flux The influence of particle size on the permeate flux decline is illustrated in Figs. 4 and 5, by plotting the instantaneous length-averaged permeate flux (normalized with the membrane limited flux, Jm). The "unfavorable" particle size range (Fig. 4), where the largest decline in flux occurs, is determined not only by the cake thickness, but also by the cake permeability. Since the latter decreases with decreasing particle size, the unfavorable size range for permeate-flux decline shifts toward smaller particle sizes as compared to the unfavorable particle size range for cake growth. The transient variation of the permeate flux, for a set of representative particle sizes, is illustrated in Fig. 5. For large particles (Fig. 4) which significantly

constrict the channel due to cake buildup, the surface area for filtration reduces in the ease of an inside-out circular geometry. Hence, the permeate flux for the hollow fiber is less than that of the slit, even though the cake for the former is thinner (at longer times and steady-state, Fig. 2). For smaller particles, which form thin cakes, curvature effects are not important, and the hollow-fiber geometry (which results in a higher shear rate than the slit for the same cross-flow velocity and channel dimensions) results in higher permeate flux. Fig. 4 also illustrates that for the system simulated, a crossover from a favorable inside-out hollow-fiber geometry to a favorable slit geometry, in terms of a higher equilibrated permeate flux, occurs at a particle

200
10

S. Sethi, M.R. Wiesner/Journal of Membrane Science 136 (1997) 191-205

(a)
~" 0.014 -L . . . . . . ~'_ '_' '~'"" Outside-In v 0012-~ ' 0.010 Uneof"cm~ovet" i~ide~uthollow tofiatsl~ from fiber ~ /~/ \\ \\ \ 2ap 1 0 ........ ' \\ ....... ~ ';P;'200 kP';'i' AP=SOkPa J I /
.

,$ mE10 -I

_____

~. 0.008

Q_

Inside-Out
< ...............

7;i

10-2 10.5

=,
10 -4 10-3 10-2 Bulk Suspension Particle Volume Fraction (-) 10-1

o.oo2-1

] o ol.m . o.oool ........ ................ , 100 101 102 103


,

"~..~-"__~\, _ . . . . . .
,

i.:::.~, . _ ~ :-~ 104 105 10s (s)

Time

Fig. 6. Domains of favorable equilibrated permeate flux for different hollow-fiber configurations.

(b)
0.012 "~ 0.010 -

size of ca. 0.06 pro. When simulations are performed over a range of bulk suspension concentrations, the "crossover diameter" is seen to decrease with increasing concentration (Fig. 6). The flat-slit geometry represents a crossover in the radius of curvature from inside-out (concave) to outside-in (convex) cylindrical configuration. Thus, Fig. 6 can be interpreted as illustrating the respective domains of raw water quality where these two hollow-fiber configurations are expected to be favorable.

Y
=1.0 pm

j::v

+ oE0.008;=_~
eLL 0.006 i

~ g 0.004g_
0.002~ o z~ / 0.01 Izrn
- c

0.1 Izm
0.000 , , , , --

50

1O0 150 200 Transmembrane Pressure (kPa)

250

4.2. Effects of operating conditions on permeate flux for inside-out hollow fiber
Interactions between operating parameters and membrane fouling were investigated by performing simulations in which the operating variable being analyzed assumes three values (Table 2), while all other system parameters are held constant at their baseline values (Table 2). These simulations were conducted for feed suspensions containing one of three representative particle sizes (small particle: 0.01 ~tm; intermediate/unfavorable sized particle: 0.1 ~tm; and large particle: 1.0 ~tm).

Fig. 7. (a) Effect of transmembrane pressure on the transient permeate flux for an inside-out hollow-fiber filter. (b) Effect of transmembrane pressure on the steady-state permeate flux for an inside-out hollow-fiberfilter. Illustration of mass transport limited (or pressure-independent) behavior of permeate flux.

4.2.1. Transmembrane pressure


The effect of transmembrane pressure on the transient and steady-state permeate flux is illustrated in Fig. 7(a) and (b), respectively. The permeate flux is not presented in a normalized fashion in these plots since the membrane limited flux changes with trans-

membrane pressure. Fine particles (0.1 and 0.01 ~tm, respectively) are seen to depict mass transport limited behavior since the limiting or steady-state permeate flux does not change much with pressure. However, large particles (1.0 ~tm) show significant variation in the limiting permeate flux with transmembrane pressure. This is illustrated more clearly in Fig. 7(b), where the steady-state permeate flux is plotted against transmembrane pressure for the three representative particle sizes. Operating at higher pressures causes the steady-state to be achieved earlier and the permeate flux to increase over the entire operation (Fig. 7(a)). However, it should be noted that with increasing

S. Sethi, M.R. Wiesner/Journal of Membrane Science 136 (1997) 191-205

201

1.0

2ap=1.0prn

--1

.7. 0.8-LL Q~
a..

0.1p m

i!;. . . . . . . . . . .

~ ........

~0.6

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

S--

I
. . . . .

Ho=O,25m m Ho=0.50mm Ho=l.00mm


~ He=0.25, m 0.5 m

~ 0.4-

g
o

~
0.2[11,/o=19200s -~ [21 ~,o=9600s ~ [al ~,o=4800s ~ [417o=2400s -~
0.0

"

Ho=0,25, 1 mm 0.5, Ho=0.5,1mrn

100

[51 ~,o=1200s -~ ........ i . . . . .... I

101

...... I ........ I ........ I ........ I


10 3 10 4 10 5 10 6

10 2

Time (s)
permeate flux

F i g . 8. E f f e c t o f s h e a r r a t e o n t h e l e n g t h - a v e r a g e d f o r a n i n s i d e - o u t h o l l o w - f i b e r filter.

filtration area. For these particles the kinetics of cake growth, coupled with the curvature effects on available filtration area, determines the transient permeate flux profile. Hence, as seen in Fig. 8, for these particles the transient permeate flux can improve with an increase in fiber radius. This behavior is not observed in the case of small particles, since the cakes formed are very thin and, hence, the reduction in available filtration area is not significant. The time to reach equilibration decreases with a higher cross-flow velocity or a smaller radius, but the change is seen to be more significant in the latter case. Small particles demonstrate the same behavior in permeate flux at the same shear rate, regardless of the individual values of the cross-flow velocity and the membrane radius. However, the intermediate and large-sized particles show variation in the permeate flux even at the same shear rate, depending on the value of cross-flow velocity and membrane radius. For these particles, at the same shear rate, a narrower fiber with a lower cross-flow velocity results in a higher steady-state permeate flux, but a lower transient permeate flux. This behavior can be explained from the earlier discussion on the effects of membrane radius on the permeate flux.

pressure the steady-state flux actually decreases relative to the initial flux, for all particle sizes. 4.2.2. Shear rate Shear rate can be adjusted by varying either the cross-flow velocity or the channel diameter/height. A combination and permutation of the variable crossflow velocities and channel diameters/heights listed in Table 2 results in five different values of the shear rate, ranging from 1200 to 19200 s -1. Fig. 8 illustrates the effect on the permeate flux as the shear rate is varied over this range. When the cross-flow velocity is varied, the effect on the transient decline of the permeate flux is less significant than the effect on the steady-state flux, particularly for the unfavorable particles. Increasing the membrane radius decreases the shear rate and, hence, the equilibrated permeate flux decreases. However, in case of larger particles (0.1 and 1.0 gm, respectively), which form thick cakes resulting in significantly reduced area available for filtration, a larger hollow-fiber radius increases the

5. C o n c l u s i o n s

A previous time-dependent model for cross-flow microfiltration, based on shear-induced diffusion and convective particle layer transport, has been extended to include Brownian diffusion and inertial lift as additional particle transport mechanisms. This extension makes the theory applicable to a broad range of contaminant sizes ranging from macromolecules, colloidal and fine particles, to large particles. For the typical system simulated in this work, the combined theory predicts an unfavorable size, ca. 0.4 ~tm, where net back-transport due to Brownian diffusion, shear-induced diffusion and inertial lift is at a minimum, and results in highest particle accumulation on the membrane surface. This, coupled with the decrease in permeability of cakes with particle size and cake growth over time, implies lowest permeate flux in the size range of 0.01-0.1 ~tm, depending on the operating time. These results support the indirect experimental observations of a minimum

202

S. Sethi, M.R. Wiesner/Journal of Membrane Science 136 (1997) 191-205

in permeate flux [30,32], as well as recent direct experimental confirmation of the existence of a minimum in back-transport [33]. Cakes formed from small particles are predicted to attain their steady-state thickness earliest, followed by large particles, and finally the particles in the intermediate size range. Particle size, coupled with the effects of curved cakes in cylindrical geometries, determines whether the flat-slit or inside-out hollow-fiber yields higher permeate flux. For the same cross-flow velocity and diameter/height, the inside-out cylindrical geometry results in a higher shear rate. The inside-out geometry may, therefore, be expected to be favorable (in terms of higher permeate flux) for feed waters with small particles which form thin resistive cakes. However, larger particles (> 0.1 ~tm), which form thick cakes, may result in reduced surface area available for filtration due to curvature effects in inside-out membranes, making the slit or outside-in geometry more favorable for these particles. For the typical ultrafiltration system, simulated in this study, variations in transmembrane pressure produced significant changes in both the transient and equilibrated flux for the large particles, and in the transient flux for the intermediate sized particles. The equilibrated flux was not significantly affected in the case of the intermediate-sized and small particles, indicating that fine particles (< 0.1 ~tm) demonstrate mass transfer limited, or pressure-independent, behavior. For large particles, the kinetics of cake growth coupled with the curvature effects on available filtration area determine the permeate flux profile, and the transient flux can improve with increased fiber radius. This behavior is not observed in the case of small particles, since the cakes formed are very thin and, hence, the reduction in available filtration area is not significant. The time to equilibration decreases with a higher cross-flow velocity or a smaller radius, but the change is seen to be more significant in the latter case. Small particles demonstrate the same behavior in permeate flux at the same shear rate, regardless of the individual values of the cross-flow velocity and the membrane radius. However, the intermediate and large size particles show variation in the permeate flux even at the same shear rate, depending on the value of cross-flow velocity and membrane radius. For these particles, at the same shear rate, a narrower fiber with a lower cross-flow velocity results in a higher

steady-state permeate flux, but a lower transient permeate flux. The combined model formulated in this work converges to the previous models based on a single mechanism of back-transport for very small or very large particle sizes under certain operating conditions. However, as expected, the combined model predicts different permeate fluxes at intermediate particle sizes where multiple mechanisms of particle transport become important. The previous models based on a single mechanism of particle back-transport may be sufficient under certain conditions of particle size and shear rates, when the particle transport mechanism on which the model is based dominates, and other backtransport mechanisms are negligible. It should be noted that dominance based on particle size is not the sole governing factor in mass transport. The operating conditions also play a part in determining the importance of a transport mechanism. For example, shear-induced diffusion may become important even for small particles when the flow is sufficiently fast, and/or the membrane channel is quite narrow, and/or the suspension is relatively concentrated. Similarly, inertial lift may become important for smaller particles under fast flows and/or narrow channel geometry. Hence, previous models may be sufficient for particular ranges of particle sizes only under certain operating conditions. In general, models based on Brownian diffusion may be sufficient for very small particles and low shear rates. However, it should be noted that previous models based on Brownian diffusion predict only the steady-state permeate flux. At very large particle sizes and shear rates, the models based on shear-induced diffusion and inertial lift may be sufficient. Under all other conditions of particle sizes and operating conditions, the combined model may be considered necessary for obtaining better estimates of the permeate flux.

6. List of symbols
ap

DB Ds

bs
f

particle radius (cm) Brownian diffusion coefficient (cm2/s) shear-induced diffusion coefficient (cm2/s) dimensionless shear-induced diffusion coefficient dimensionless function defined by Eqs. (29c) and (30c)

S. Sethi, M.R. Wiesner/Journal of Membrane Science 136 (1997) 191-205

203

g h H

/40
I1
Ill 112

12
121

122 124
121

J
Jm

k
Le F/ ee b

Pes
Rc

Rm
t

tcr T
U

U
V1 V1 o Vw X Xcr

dimensionless function defined by Eqs. (29a) and (30a) dimensionless function defined by Eqs. (29b) and (30b) effective slit half-height or hollow-fiber radius (H = H0 - 6c) (cm) slit half-height or hollow-fiber radius (cm) integral defined by Eq. (16) integral defined by Eq. (20a) integral defined by Eq. (20b) integral defined by Eq. (17) integral defined by Eq. (20c) integral defined by Eq. (20d) integral defined by Eq. (20e) integral defined by Eq. (20f) permeate flux (cm/s) permeate flux limited only by membrane in absence of cake layer (cm/s) Boltzmann's constant (1.38 x 10 -16 g cm2/s 2 K) length of membrane element (cm) constant (n -- 2 for flat geometry and n -- 3 for cylindrical geometry) Peclet number based on Brownian diffusion Peclet number based on shear-induced diffusion cake resistance (1/cm) specific cake resistance per unit height (1/cm 2) membrane resistance (1/cm) time (s) critical time (s) absolute temperature (K) axial velocity (cm/s) average cross-flow velocity of bulk suspension (cm/s) lift velocity (cm/s) lift velocity in absence of cake (cm/s) permeate velocity (cm/s) axial coordinate (cm) critical length (cm) transverse coordinate (cm)

6c Ap ~b

(~b
~c ~bw x/ //o ~7 # #o p pp ~-w TwO

cake thickness (cm) transmembrane pressure (g/cm S2) particle volume fraction bulk suspension particle volume fraction maximum packing or solidosity of cake particle volume fraction at wall shear rate (l/s) shear rate in absence of cake layer (l/s) relative viscosity (7/= #/#0) dynamic viscosity (g/cm s) dynamic viscosity of particle-free fluid (g/cm s) fluid density (g/cm 3) particle density (g/cm 3) wall shear stress (g/cm s2) wall shear stress in absence of cake (g/cm s2)

6.2. Superscripts

dimensionless value

Acknowledgements
This work has been supported in part by the United States Environmental Protection Agency, Hazardous Substances Research Center S&SW. Thanks are due to S. Veerapaneni and S. Krishnamoorthy for useful discussions.

Appendix A
Flat slit (n = 2):
Ill

/3

n(1 +/36)l

/3112

g=Pe~

(1-6)" (29a)

h = 2/3(1 +/36) 124


Pe 2 (1 -- 6)"

121

(122 + 23)
Pebee (1 -

Pe (1 -

6.1. Greek symbols

+ (1 +/3S)2

3ni21 pe2( 1 _ ~)3n+l


n/24

3
6

dimensionless resistance boundary layer thickness (cm) dimensionless cake thickness

2n(122 + 23)

PebPes(1 -- tS)2n+l ~- pe2(1 _ S)n+l ]

(29b)

204

S. Sethi, M.R. Wiesner/Journal of Membrane Science 136 (1997) 191-205

Ou

i 1 V'o ] 1 + rid) (1 - ~)2.J (n = 3): [~(1 6)~ ~ (n-1)[l+rln(~)]


(1 - d)n

(29c)

Inside-out cylindrical hollow fiber


Ill g = ~es

(30a) h = 2rll + tin (i~_l~) 1 pe2s(1 /21 ~)3n-1 _ +


(/22 + / 2 3 )
P e b P e s ( 1 - (~)2n 1

124

Pe~(1

d)n-1

(3n - 2)/21
PeZs (1 - t~)3"-1

+ (2n
f = ~b

-- 2)(/22 + 123)
-

PebPes(1

~)2n-1

(n - 2)/24
P e ~ ( 1 - ~)n-1

(30b)

. 1 Vlo (1- 6)[1 + rln(l1_-~)] (1 _--~)2~ (30c)

References
[1] P.L.T. Brian, Concentration polarization in reverse osmosis desalination with variable flux and incomplete salt rejection, Ind. Eng. Chem. Fundam., 4 (1965) 439--445. [2] A.S. Michaels, New separation technique for the CPI, Chem. Eng. Prog., 21 (1968) 31-43. [3] W.E Blatt, A. Dravid, A.S. Michaels and L. 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, pp. 47-97. [4] M.C. Porter, Concentration polarization with membrane ultrafiltration, Ind. Eng. Chem. Prod. Res. Develop., 11 (1972) 234-248. [5] J.S.S. Shen and R.E Probstein, On the prediction of limiting flux in laminar ultrafiltration of macromolecular solutions, Ind. Eng. Chem. Fundam., 16 (1977) 459-465. [6] D.R. Trettin and M.R. Doshi, Limiting flux in ultrafiltration of macromolecular solutions, Chem. Eng. Comm., 4 (1980) 507-522.

[7] G. Green and G. Belfort, Fouling of ultrafiltration membranes: Lateral migration and the particle trajectory model, Desalination, 35 (1980) 129-147. [8] EW. Altena and G. Belfort, Lateral migration of spherical particles in porous flow channels: Application to membrane filtration, Chem. Eng. Sci., 39 (1984) 343-355. [9] R.J. Weigand, EW. Altena and G. Belfort, Lateral migration of spherical particles in laminar porous tube flows: Application to membrane filtration, Physicochem. Hydrodyn., 6 (1985) 393-413. [10] J.A. Schonberg and E.J. Hinch, Inertial migration of a sphere in Poiseuille flow, J. Fluid Mech., 203 (1988) 517-524. [11] 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. [12] A.L. Zydney and C.K. Colton, A concentration polarization model for the filtrate flux in cross-flow microfiltration of particulate suspensions, Chem. Eng. Comm., 47 (1986) 1-21. [13] R.H. Davis and D.T. Leighton, Shear-induced transport of a particle layer along a porous wall, Chem. Eng. Sci., 42 (1987) 275-281. [14] C.A. Romero and R.H. Davis, Global model of cross-flow microfiltration based on hydrodynamic particle diffusion, J. Membrane Sci., 39 (1988) 157-185. [15] C.A. Romero and R.H. Davis, Transient model of cross-flow microfiltration, Chem. Eng. Sci., 45 (1990) 13-25. [16] R.H. Davis and J.D. Sherwood, A similarity solution for steady-state cross-flow microfiltration, Chem. Eng. Sci., 45 (1990) 3203-3209. [17] C.A. Romero and R.H. Davis, Experimental verification of the shear-induced hydrodynamic diffusion model of crossflow microfiltration, J. Membrane Sci., 62 (1991) 249-273. [18] R.H. Davis and S.A. Birdsell, Hydrodynamic model and experiments for cross-flow microfiltration, Chem. Eng. Comm., 49 (1987) 217-234. [19] E.E Leonard and C.S. Vassilieff, The deposition of rejected matter in membrane separation processes, Chem. Eng. Comm., 30 (1984) 209-217. [20] W.-M. Lu and S.-C. Ju, Selective particle deposition in crossflow filtration, Sep. Sci. Technol., 24 (1989) 517-540. [21] K. Stamatakis and C. Tien, A simple model of cross-flow filtration based on particle adhesion, AIChE J., 39 (1993) 1292-1302. [22] A. Koltuniewicz, Predicting permeate flux in ultrafiltration on the basis of surface renewal concept, J. Membrane Sci., 68 (1992) 107-118. [23] R.H. Davis, Modeling of fouling of cross-flow microfiltration membranes, Sep. Purif. Methods, 21 (1992) 75-126. [24] M.H. Lojkine, R.W. Field and J.A. Howell, Cross-flow microfiltration of cell suspensions: A review of models with emphasis on particle size effects, Trans. Inst. Chem. Eng., 70 (1992) 149-164. [25] G. Belfort, R.H. Davis and A.L. Zydney, The behavior of suspensions and macromolecular solutions in cross-flow microfiltration, J. Membrane Sci., 96 (1994) 1-58. [26] W.R. Bowen and E Jenner, Theoretical descriptions of membrane filtration of colloids and fine particles: An

S. Sethi, M.R. Wiesner/Journal of Membrane Science 136 (1997) 191-205 assessment and review, Adv. Colloid Interface Sci., 56 (1995) 141-200. S.G. Redkar and R.H. Davis, Cross-flow microfiltration of yeast suspensions in tubular filters, Biotechnol. Prog., 9 (1993) 625-634. S. Chellam and M~R. Wiesner, Particle transport in clean membrane filters in laminar flow, Envir. Sci. and Technol., 26 (1992) 1611-1621. S. Sethi and M.R. Wiesner, Performance and cost modeling of ultrafiltration, J. Envir. Engrg., ASCE, 121 (1995) 874883. A.G. Fane, Ultrafiltration of suspensions, J. Membrane Sci., 20 (1984) 249-259. M.R. Wiesner, M.M. Clark and J. Mallevialle, Membrane filtration of coagulated suspensions, J. Envir. Engrg., ASCE, 115 (1989) 20-40. V. Lahoussine-Turcaud, M.R. Wiesner and J.Y. Bottero, Fouling of tangential-flow ultrafiltration: The effect of colloid

205

[27]

[33]

[28]

[34]

[29]

[35]

[30] [31]

[36] [37] [38]

[32]

size and coagulation pretreatment, J. Membrane Sci., 52 (1990) 173-190. S. Chellam and M.R. Wiesner, Particle back-transport and permeate flux behavior in cross-flow membrane filters, Environ. Sci. Tech., in press. D. Leighton and A. Acrivos, Measurement of shear-induced self-diffusion in concentrated suspensions of spheres, J. Fluid Mech., 177 (1987) 109-131. D. Leighton and A. Acrivos, The shear-induced migration of particles in concentrated suspensions, J. Fluid Mech., 181 (1987) 415-430. P.C. Carman, Fundamental principles of industrial filtration, Trans. Inst. Chem. Eng., 16 (1938) 168-188. IMSL, User's manual: FORTRAN subroutines for mathematical applications, Volume 2.0, Houston, TX, 1991. K.A. Hoffman and S.T. Chiang, Computational fluid dynamics for engineers, Vol. I, Engineering Education System, Wichita, KS, 1993.

S-ar putea să vă placă și