A review of fouling and fouling control in ultrafiltration

A review of fouling and fouling control in ultrafiltration

Desalination, 62 (1987) 117-136 Elsevier Science Publishers B.V., Amsterdam - 117 Printed in The Netherlands A Review of Fouling and Fouling Control...

2MB Sizes 1 Downloads 170 Views

Desalination, 62 (1987) 117-136 Elsevier Science Publishers B.V., Amsterdam -

117 Printed in The Netherlands

A Review of Fouling and Fouling Control in Ultrafiltration* A.G. FANE and C.J.D. FELL School of Chemical Engineering and Industrial Chemistry University of New South Wales, Kensington, NSW 2033 (Australia) Tel. (02) 6974315, TX. AA 26054

SUMMARY

This paper discusses the properties of ultrafiltration (UF) membranes which make them susceptible to fouling. Various types of flux decline are described from early usage to long-term effects. For protein UF it is shown that flux decline occurs due to protein deposition,‘and that this depends on membrane and solute type, solution environment and operating conditions. Attempts to model UF fouling are reviewed and selected examples of fouling control are also described. Keywords: ultrafiltration, membranes, fouling

SYMBOLS

A, c d,d,d, fi Fi

J 5

Jf Ji JO &,K& 4

-

membrane area concentration diameter, equivalent diameter, pore diameter fraction of pores with diameter dpi fraction of solvent passing through pores dpi flux (velocity) membrane-averaged flux (observed flux) fouled flux local flux associated with pores of size dpi initial flux constants in deposition Eqs. (15) and (17) solute mass transfer coefficient

*Presented at the International Symposium on Synthetic Membrane Science and Technology, Dalian, China, April 13-18,1986.

OOll-9164/87/$03.50

0 1987 Elsevier Science Publishers B.V.

118 1

-

M AP

-

Qi

-

rd7re

-

R t

-

ub

V s% ; p Subscripts

-

b bl cl m

-

S

-

W

-

pore length mass deposited per unit area transmembrane pressure difference volumetric flow rate through single capillary c&i rate of deposition, rate of removal [ Eq. (13 ) ] hydraulic resistance time cross-flow velocity accumulated permeate volume specific resistance of deposit membrane thickness voidage contact angle viscosity bulk (feed) boundary layer deposit membrane solute at membrane surface

INTRODUCTION

In the process of ultrafiltration ( UF ) a pressurised solution or suspension is passed across the surface of a membrane. Ideally the solvent and low-molecular weight species pass through the membrane, the larger species are swept away from the surface, and the membrane remains as an inert film of constant hydraulic resistance, R,. Flux under UF conditions will be less than for pure solvent due to concentration polarisation, so

J=

AP p [Rm+R,l

(1)

where R, represents the resistance of the solute. Thus, in an ideal situation flux is constant, unless action is taken to alter R, (by change of concentration or hydrodynamics), p (by change of temperature) or AP. However, in most practical applications of UF, resistance increases and flux declines. The increase in resistance may be due to changes in R,, R, or both. If flux decline is not reversible by simply altering the operating conditions it is termed fouling. This is a major problem in UF since it reduces productivity, shortens membrane life (often due to aggressive cleaning agents) and impairs fractionation capability of the membrane. Unless the fouling effect is properly understood and controlled, the membrane process becomes nonviable. This

j

,

,'I"'

40

50 CONTACT

,

,

1

60 70 ANGLE191

80

Fig. 1. Contact angle versus UF flux loss. Fig. 2. Surface of Amicon XMlOO viewed by TEM.

paper describes the nature and causes of fouling of UF membranes methods to control the problem.

and some

THE NATURE OF MEMBRANES AND FOULING

Two characteristics of UF membranes appear to influence fouling, viz.: (a) the physico-chemical properties of the membrane, and (b) the porosity and morphology of the surface. Physico-chemical properties Many situations arise where physico-chemical interactions occur between solution species and membrane material. For example macrosolutes, such as proteins, can bind to polymer surfaces by a variety of mechanisms [ 1 ] including electrostatic interaction, hydrophobic effects, charge transfer (e.g. hydrogen bonding and rc-7c bonding) or through combinations of these. As an example, the hydrophobic nature of a membrane can be characterised by measuring its contact angle, 0. The higher the value of 0 the more hydrophobic the surface. Figure 1 shows how contact angle correlates with flux loss in protein UF (flux loss is the relative flux decline over a 2.5 h period= {Jo.,,-J3h}/J3h). Note that the most hydrophilic membrane, the Amicon YM30, achieved the least flux loss. A useful review of physico-chemi-

120 TABLE I SURFACE POROSITY OF TYPICAL UF MEMBRANES

“bw

Mean pore diameter (nm)

Surface porosity (o/o)

Reference

Notes

Amicon XM300

20 25

XMlOO PM30 Millipore PTGC PVDF DDSGRGlP Amicon YM30

18

0.3 4.7 0.8 2.0 7-12 9-15 - 10 -50”

3 4 3 4 5 6 7 8

TEM TEM TEM TEM

12 6 6-8 -

using unidirectional scattering. using rotating scattering. using unidirectional scattering. using rotating scattering.

Bubble point/permeability. High-resolution electron microscopy.

“Order of magnitude inferred by authors.

cal interactions between membranes has been given by Larsson [ 21.

and biological

species, such as proteins,

Membrane porosity Typical UF membranes have been found to have a relatively low surface porosity as illustrated in Fig. 2 for one membrane; collected data are shown in Table I. This low porosity means that solvent flowing towards the membrane does not meet a homogeneously permeable surface, but will have to follow streamlines to the openings of isolated pores. In this respect UF membranes differ from microporous membrane filters, which have a typical porosity of 7540%. At the other end of the spectrum reverse osmosis ( RO ) membranes can also be assumed to present a homogeneous surface to the solvent. A consequence of the sparse porosity of UF membranes is that local polarisation phenomena may be much greater than the average polarisation. UF membranes also have a distribution of pore sizes. Figure 3 shows the measured pore size distributions of Amicon XMlOOA and XM300 membranes obtained by electron microscopy [ 31. Similar distributions have been obtained for other UF membranes using combined bubble pressure and solvent permeability [ 91 and capillary condensation/permeability [ lo]. Figure 3 also includes the estimated flow through different pore sizes. A simplified model has been assumed with the membrane represented by a bundle of capillaries. The Hagen-Poiseuille equation for a single capillary of diameter o&igives a flow rate Qi = II $i AP/ [128/d]

(2)

121

100

, / , ~ / / . - ~ 100

100

80

8O

~80 x

/

~ 60 20 XM100A~,

/

I

10

>

15 20 25 30 PORE DIAMETERInto)

z,O

J

20 ~-

/z',-XM300

om

60

LO

/// //

~0 13_

J

U-

I

35

20 _

i

i

i

20

~*0

60

PORES

80

100

PLUGGED(%)

Fig. 3. Distribution of pores and solvent flow for typical U F membranes. ( - - ) pores, ( - - - ) flow. Fig. 4. Illustration of flux decline due to loss of pores by plugging with XM300 membrane.

If/~ is the fraction of pores with diameter dpi the fraction of solvent passing through those pores, Fi, is dpmax

Fi=[id4i/ E fid4i

(3)

dprain

Equation (3) has been used to calculate the flow distribution data in Fig. 3, which shows that solvent flow is strongly biased to the larger pores with 50% of the solvent flow through 20-25% of the pores. The effect on flux of losing pores is illustrated in Fig. 4 for the pore size and flow distribution of the XM300 membrane. Curve A assumes that the largest pores plug first, and curve B assumes that the largest 10% of pores are so large they are unpluggable and that the next largest pores plug first. If the flow is biased towards the larger pores they will experience a higher degree of concentration polarisation. According to the film model [ 11] the polarisation modulus {Cw/Cb) is related to the flux and the mass transfer coefficient ks by Cw/Cb=exp [J/ks]

(4)

Equation ( 4 ) can be examined in terms of the local flux J~, associated with pores of size dp~.If J is the membrane-average flux (i.e. the observed flux = total volumetric flow rate/total area), the local flux is = j - ( .fraction . . of. flow . through . . . size dpi ) Ji fraction of area associated with dpi

(5)

Ji : J Fi/

(6)

e

2

Using Eq. (3)

Ji = j d2i [Zf~dpJZf~dpi] 2 4

(7)

122

I

0’ 12

16

20

2L

28

PORE OIAMETERlnm)

INITIAL SOLVEN1 DECLINE

II FLUX

INITIAL UF FLUX OEC L INE

*‘I

LONGTERM FLUX

U F

OECLINE

I

3; T

I

M E

(sl

Fig. 5. Polarisation modulus and local flux versus pore diameter. Average flux = 20, k, = 8. Fig. 6. Typical flux history for a UF membrane.

The local fluxes for given pore sizes have been calculated from Eq. ( 7)) using the size distribution data for the XM300 membrane (Fig. 3) and J= 20 l/m2h. Figure 5 shows a &fold increase in Ji over the pore size range 14-30 nm. Polarisation modulus has been estimated from Eq. ( 4 ) for k, = 8 l/m2h ( approximately 2 x 10F6 m/s), which is typical of laminar flow with protein solutions [ 12,3]. Figure 5 shows that polarisation at the largest pores is over an order of magnitude higher than at the smallest pores. The implications of Figs. 3-5 are - Overall membrane flux will be very sensitive to the population of large pores, and their loss by plugging or obstruction. - The plugging and obstruction of larger pores will be encouraged by the considerably higher local fluxes (i.e. velocities) and concentration polarisation experienced. These considerations suggest that some fouling of UF membranes by pore blockage is almost inevitable. However this may be tolerable until the membrane resistance R, increases to the same magnitude as R, [ Eq. (1) 1. An alternative view of the membrane is that it is a porous body of voidage E. From the Carman-Kozeny relationship [ 131 the resistance will be R,=K

[~-E]~/c~

(8)

where K= 180 6,/e Equation (8) shows that R, will be very sensitive to changes in t. For examle, as Echanges from 0.3 to 0.2 the resistance R, increases 4-fold, and as it changes from 0.2 to 0.1 the increase is by a factor of 10. Thus, whether the membrane

123

I

\

( RZjnn

f=lL TERED)

10

0 1

20 I

t4

E

(min

1

Fig. 7. Effect of water purity on water flux decline with PM30 membranes.

suffers loss of pores or reduced voidage the result can be a significant flux.

drop in

FLUX DECLINE HISTORY

The phases solvent ditions

time dependence of UF membrane flux is illustrated in Fig. 6. Three can be considered which include (i) an initial loss of flux due to pure passage, (ii) an early flux decline during polarisation under UF conand (iii) a long-term flux decline over hours, days or weeks of operation.

Water (solvent)

flux

decline

Membrane suppliers usually specify water quality requirements for the washing, testing and cleaning of industrial UF systems. Although distilled water typically exceeds this water quality it can cause significant flux decline. Figure 7 shows the effect of water purity on water flux of an Amicon PM30 membrane [ 14 1. Passage of standard distilled water over an extended period caused water flux to drop to about 10% of the initial value (compare Fig. 4). This flux decline (fouling) was attributed to pore plugging by bacteria and trace colloids, since decline was arrested by use of ultrapure water prepared by prefiltration and addition of a bacteria-stat. The flux declines in Fig. 7 show how sensitive an unused membrane can be. As a general observation, the passage of excessive amounts of water through a membrane before usage is unwise. Initial UF flux decline In this phase the flux drops, in a period of seconds or minutes from the pure water flux to the UF flux due to the build-up of the concentration boundary

124 1 I M E lmlnl 0

2

1,

6

8

10

12

II

*-*w2'

BSA /SYTON .--_.-_

----*-__

1% 4%

---?_--7 0 0

I

I

I

,

10

20

30

u)

OEXTRAN$-50

Fig. 8. Initial flux dynamics for stirred UF. BSA 133 kPa, 600 rpm; Syton 31 kPa, 800 rpm; Dextran 133 kPa, 100 rpm. (-.-.-) experiment, (---I model. layer. These polarisation dynamics have been measured [ 151 and are illustrated in Fig. 8. This flux decline is mostly reversible and is therefore not fouling. However, fouling by pore blockage can occur during these initial stages. This effect is best illustrated by the results of constant pressure unstirred UF tests which can be interpreted in terms of filtration theory [ 151.

t/V=

[R,p/dPA,]

+ [C,apV,W~,,,

(9)

The intercept of a t/ Vversus V plot gives membrane resistant R,, and early changes in the plot give changes in R,. For protein [bovine serum albumin, BSA] UF Fig. 9 shows that in the first 20-30 s R, rose by an order of magnitude from about 2 x 1011 m-l to about 2 x 1012 m-l. The changes in R, depended on the ionic environment which is known to influence the shape and charge of the protein molecule. Long-term

flux decline

Long-term flux decline is caused by time-dependent hydraulic resistances, so Eq. (1) can be rewritten, J{t)=dP/p[R,i+dR,

{t)+R,

{t}+Rbl]

(10)

where R, is the initial membrane resistance and Rbl is the resistance due to solute in the boundary layer, the resistance AR,{ t} represents any increase in the membrane resistance, and strictly includes the initial plugging and obstruction illustrated in Fig. 9, subsequent pore blockage and any solventrelated deterioration of the membrane. The resistance, Rd{t} represents the

125

10

20

30 CO 50 TIME lsJ

60

70

80

Fig. 9. Change in membrane resistance due to pore blockage. 0.1% BSA, 100 kPa. (0 ) pH 10,0.2 M NaCI; ( l ) pH 3, no salt.

effect of the foulant, the resistance of which may increase because the foulant layer grows in thickness and/or because it consolidates and becomes less permeable. In practice it may be difficult to distinguish between the effect of AR, and Rd. Examples will now be given to illustrate the main characteristics of long-term decline. Example (l), the UF of biomass suspensions [16,17] Significant flux decline was observed during the UF of activated-sludge mixed liquor which contained both dissolved macrosolutes and bacterial cells, in a Dorr-Oliver thin channel system; see Fig. 10. In this case it was possible to restore substantially the flux by sponge-cleaning the membrane, and this allowed comparison of resistances before and after extended operation (typically 60 h to steady state). It was found that R, changed only marginally with service, but Rd changed considerably, particularly at low cross-flow velocity, ub-

The importance

0

10

of u), is apparent

20

30

4.0 T

IN

from Fig. 11 which shows that for systems

50

60 E

thl

Fig. 10. Flux decline for bacterial suspension (activated-sludge).

126

/

I

/ 0.81

_

A-S LIWOR

J
2 3 ub [ml)

3

I

I

0

1

I

7 1

I

I

3

L

5

I M E ihl

Fig. 11. Flux versus cross-flos velocity for bacterial suspension. Fig. 12. Flux declines for protein, 0.1% BSA. With ( 0 ) 0.2 M NaCl and ( 0 ) no salt.

exhibiting flux decline the flux/velocity relationship needs to be carefully defined. For example, initial flux, JO,was proportional to z&O,but the declined, was or long-term steady-state fluxes, Jf,, were proportional to ut4, where r.&b maintained at fixed values from t=O. However, when the fouled membrane was subjected to increase in cross-flow the flux change was negligible ) , because the dominant resistance, Rd, was unaffected by increases ( JLYZ.&~.~ in r.&This reduced dependency on cross-flow velocity is characteristic of fouled membranes. Example (Z), the UF of protein solutions There are many examples of flux decline with proteinaceous solutions, including process streams such as whey, starch factory effluent, tannery effluent, and numerous studies with albumin protein. (See references in [ 181. ) Typical data are given in Fig. 12 which shows flux decline for the UF of BSA in solutions of differing pH and ionic content [ 351. The decline in flux was matched by a gradual increase in irreversibly deposited protein, as shown in Fig. 13. This protein could not be removed by simple water washing but required chemical cleaning. Several factors influenced protein deposition [ 351. Figures 14 and 15 show that maximum deposition occurred around the isoelectric point of the BSA, and when the protein was charged (pH 2 or 10) it was less susceptible to deposition. Membrane type was also a significant factor (Fig. 15), as was the system hydrodynamics, with less deposition at higher cross-flow velocities (Fig. 15). These observations may be linked to the adsorptive tendency of the membranes which derive from hydrophobic effects, etc. as discussed in the section on physico-chemical properties. However, the results cannot be related simply to the membrane material, since the three polysulphone membranes [ PM30 (Amicon), GR61P (DDS)

127

,/

06

I L

I 2 TIME

I 6 Ihl

I a

_

02 0

PH

Fig. 13. Deposition kinetics for BSA. PM30,lOO kPa, 400 rpm. (0 ) 0.1% BSA, (0) 1.0% BSA, ( A ) 2.0% BSA. Fig. 14. Deposition versus pH. 3 h, 0.1% BSA, stirred. l=XMlOO, 2=PM30, 3=DDS GRGlP, 4=PTGC. 5=YM30.

and PTGC ( Millipore) ] gave noticeably different data. A possible explanation [ 351 is based on differences in the surface porosity and pore size distributions of the membranes. According to Table I the surface heterogeneity may be ranked XMlOO > PM30 > PTGC N GR61P > YM30. This ranking is similar to the observed tendencies shown in Fig. 14 to accumulate deposited protein. The initial deposition rates in Fig. 13 are found to be directly proportional to the initial filtration rate (JC,) [ 351 and if this effect applies on a local scale around individual pores it would give more rapid initial deposition for the heterogeneous membranes. (This effect is supported by the local flux and polarisation data in Fig. 5.) The course of the subsequent deposition would presumably depend on the extent and nature of this initial deposition. The amounts of deposited protein in Figs. 14 and 15 are substantial and equivalent to 100-200 monolayers or a thickness of - l-2 ,um [ 141. For comparison, Cheryan and Merin [ 191, using electron microscopy, observed protein

PH

Fig. 15. Deposition versus pH. 5 h, 0.1% BSA. H = 1.1 m/s, L = 0.3 m/s, S = 0.2 M NaCl, NS = no salt.

Fig. 16. Underside (a) and top surface (b) of deposited protein. pH 5,0.2 M NaCl, 0.1% BSA.

layers 0.5-1.0 pm thick following whey UF. Kinetic data in Fig. 13 show that the deposited material can increase rapidly and approach an equilibrium value. This equilibrium value is relatively insensitive to bulk solute concentration, C,,, but it depends on system hydrodynamics (Fig. 15). The picture that emerges for protein fouling in UF is that the deposition process starts with the initial accumulation of polarised solute. Local convective forces around pores help to overcome any repulsive forces between the protein and the membrane surface. The polarised layer then provides a ‘reservoir’ of molecules which gradually become fixed due to slow aggregation and flocculation. Deposition of protein ceases when the yield stress of the topmost layers of aggregated solute is less than the local shear stress due to cross-flow or stirring. The strong influences of pH and ionic environment (Figs. 14 and 15) are expected since they affect protein charge, stability and tendency to aggregate. Electron microscopy has been used to examine the nature of foulant layers in protein UF. Glover and Brooker [ 201 studied deposits from whey which were found to have an asymmetric structure with densest layers close to the membrane. This effect has also been observed with BSA [ 141, and is illustrated in Fig. 16, which show the under side of the protein layer (adjacent to membrane) and the upper layer (adjacent to stirred feed solution) respectively. The observed asymmetry is consistent with a concentration gradient in the aggregating polarised layers, and compression of the lower aggregates and the effect of tangential shear on the upper aggregates.

129

-60

Oom PERMEATE

VOLUMElml)

TIME

(h)

Fig. 17. Flux declines for 0.1% lysozyme ( A ) and 0.1% BSA ( A ) and rejection of 0.1% lysozyme ( 0 ) with PM30 membrane. Fig. 18. Flux decline model for particulate systems (Green and Belfort [ 231) . Jo = 10m4 m/s. A = narrow tube, B = larger tube, C = more dilute feed.

Example (3), partially permeable membranes Membranes with an initially low rejection to a solute can also become fouled. In this case fouling modifies both flux and rejection. An example of this is given in Fig. 17 [ 211 which compares flux declines through an Amicon PM30 membrane for a permeating species, lysozyme (mol.wt. = 14 000 D) and a nonpermeating species, BSA (mol.wt. =69 000 D) . The point of interest is that the lower molecular weight lysozyme commenced with a higher flux but declined much more rapidly than the BSA to a lower final flux. Internal fouling of the membrane by the lysozyme can be inferred. Lysozyme rejection showed a profile which is characteristic of fouling permeant species. It should be noted that the effect of fouling is to convert the originally nonrejecting membrane into a rejecting dynamic membrane. This is sometimes a desirable effect and can be exploited. FOULING MODELS

Ink this section some fouling models for UF are summarised. Two general types of model have been proposed, the (semi) empirical and the fundamental. (Semi)empirical models The simplest form of model, which has been widely used for RO is Jf=J,f’,

nC0.0

(11)

where Jf is the fouled flux and J, is the initial flux. Thomas and Mixon [ 221 have shown that n varies with cross-flow, &,. A problem with Eq. (11) is that

130 TABLE II (SEMI) EMPIRICAL FOULING MODELS FOR UF Model equations

[iI

Jf=J‘,r,

[ii]

J=dP/p[R, Rd=ad Md

n-co.0

+R, +R,,,]

dMJdt=rd-r,

=flJcbl --r, dM,/dt = Kd[Md * -MJ Md = Md * [ l-exp( -Kdt)] Figure 13 data 0.1% BSA Md * = 65 ,ug/cm*, Kd =

or, i.e.

1%

81

2% 87 Empirically observed (Suki et al. [ 351) dM,/dt = K,J,,c,, - K2

Number

Reference

(11)

Thomas and Mixon [ 221

WI) (12) (13) (14)

Kimura and Nakao [ 251 Gutman [ 26 ] > Bhattacharyya et al. [ 271

(15) (16) 0.3 h-’ 1.2 1.8

Suki et al. [ 351

iI71

Jf asymptotes to zero, whereas real systems tend to a low, but stable, non-zero value. A number of models use a version of Eq. (10) with the deposit (or fouling) resistance given by Rd = CY&~

(12)

where ad is a constant equal to the specific resistance of the deposit and Md is the load, or mass/area, of deposit. Table II lists several references to this approach. It should be noted that the model data in Fig. 12 were obtained from the empirical model Eqs. (lo), (12 ) and (16). Fundamental

models

The complexity of the fouling process has discouraged the development of mechanistic models. One appraoch, developed by Green and Belfort [ 23 1, considers the trajectories of colloidal particles flowing through a channel. The particles move axially with the bulk steam and laterally due to convection (flux) which is opposed by lateral migration [ 241. Particles reaching the membrane gradually build up a cake which causes flux to decline. Because the cake thickness is increasing the channel height, or radius, decreases causing an increase in cross-flow velocity, &,. Eventually lateral migration (a function of ub) and convection achieve a balance and flux then remains constant. Figure 18 (from [ 231) shows predicted flux declines. Curve A is for a narrow

131

Ol

0

60

30 T IM E

(min)

Fig. 19. Flux decline for BSA (aggregation model [ 281) .

tube, curve B is for a larger diameter tube and curve C shows the effect of a more dilute feed. The limitations of this model are that it applies only to particulate fouling, and neglects physico-chemical effects which can greatly influence the behaviour of colloids. A different approach for protein fouling is based on physico-chemical effects and considers solute-solute interaction in the polarised layer [ 281. In this model the accumulated polarised solute provides a reservoir of macromolecules which gradually flocculate to form aggregates, layer by layer. As flocculation proceeds layer permeability falls and flux declines. Aggregation is described by flocculation theory using available electrokinetic parameters for BSA. For each incremental layer x there will be an aggregate size distribution from which an average voidage, tAv,x can be calculated [ 281. The resistance of layer X, of thickness a,, is obtained from the Carman relationship [ Eq. (8) with E= eAv,x, 6,-+6, and d,=d, (the solute monopartitle size ) ] . For a boundary layer subdivided into k sections, the flux can be calculated from

(18) Figure 19 [ 281 shows computed flux declines for typical BSA UF conditions, experimental data from Suki et al. [ 351 are included for comparison. Qualitatively the model predicts increased flux decline for higher.feed concentrations and for a greater degree of concentration polarisation. It also predicts the densest deposit layer adjacent to the membrane, as observed experimentally (Fig. 16a). The limitations of this protein aggregation model are the lack of reliable electrokinetic data for BAS and other proteins, and the fact that the model

132

2

0.4

,

,

0

30

60

1

,

90 120 T IM E Imid

1

, ]

150

180

I

OO

2

6

6

8

10

RECYCLES

Fig. 20. Flux and solute resistance versus time. PM30,0.1% BSA, ( 0 ) treated 3 h with 100 mg/l NlOO, (0 ) untreated. Fig. 21. UF flux increase versus usage. ( 0 ) NlOO, (0 ) methylcellulose.

does not allow for differences in membrane properties, which have been shown to influence fouling. FOULING CONTROL

The strategies available for control of fouling are (1) tailor or pretreat the membrane, (2) modify or pretreat the feed, ( 3 ) adjustment of operating conditions. A comprehensive review of fouling control has been given by Matthiasson and Sivik [ 291, and some illustrative examples are given below. Pretreat the membrane The ideal UF membrane for most applications would be hydrophilic and homogeneously permeable. Attempts have been made to achieve these characteristics by pretreatment with hydrophilic surfactants and polymers [ 30,311. Figure 1, discussed earlier, shows that a polysulphone (Amicon PM30) membrane pretreated with a nonionic surfactant is more hydrophilic and suffers less flux decline than an untreated membrane. The effect is also illustrated in Fig. 20 which compares the flux histories of treated and untreated membranes over a 3 h period. In this case treatment was obtained by contacting the membrane for 3 h with a 100 mg/l solution of NlOO (a nonionic nonylphenol polyethoxylate obtained from ICI Australia). Similar effects were obtained with

133

only 5 min contact with a 1000 mg/l solution. The improvements shown have been maintained for UF runs longer than 24 h. Recent results in our laboratory show that similar or better improvements can be achieved by forming a monolayer (so-called Langmuir-Blodgett layer) on the membrane surface. Two aspects of Fig. 20 are noteworthy. Firstly the initial flux is higher for the treated membrane. This is believed to be due to the effect the precoat has on improving the homogeneity of the membrane surface, which should increase flux [ 3}. Secondly, Fig. 20 shows that the treatment reduces the flux decline, and this is found to coincide with a reduced rate of protein deposition. The precoat presumably lowers the high levels of local polarisation and reduces the availability of hydrophobic sites for adhesion. When the membrane goes through a number of cyles of treatment and usage the treated membrane shows a significant improvement. Figure 21 gives the results of pretreatment by surfactant ( NlOO) and by soluble polymer methylcellulose (high substitution, BDH Chemicals, UK). Flux improvements range from 40% for the NlOO to almost 100% for the methylcellulose over a few cycles. Adjustment of operating conditions A most useful operating condition for fouling control is the cross-flow velocity us. Increasing ub has the effect of decreasing the degree of polarisation by increasing mass transfer and other back-transport mechanisms. Figures 10 and 11 illustrate the advantages of increasing ub, although there are obvious practical and economic limits to its magnitude. Some advantages have been found for use of pulsating flow [ 32 ] and backwashing [ 33 1. A novel approach to the latter has been developed by Memtec Ltd. in Australia who produce microfiltration/ultrafiltration systems for particulate and colloid removal; typical applications being fruit-juice, beverage and water clarification and effluent treatment. The Memtec system uses hollow fibre membranes with a relatively low bubble-point. Feed suspension is pumped across the outside of the fibres and Eltrate passes out through the lumen (Fig. 22 ) . Flux decline occurs as the colloid deposits on and within the membrane. This effect is reversed by pulsing the lumen with gas (air, nitrogen, etc.) and backwashing with a gas/permeate mixture. The gas pulse expands the fibres and opens the pores allowing fouling material to be flushed out. Figure 23 [ 34 ] shows that the gas backwash is more effective than a liquid backwash in controlling fouling with particulate systems. This unique method of fouling control is only feasible for hollow fibres with low bubble-point.

134

ooeratina

wremane moulding loIds fibres in ~loce

mode

concentrated waste moteriol is rejected from shell

backwash

011 pumped

man feed

a “bundle” of hollow fibres with microporous walls

feed stream 1s pumped into shell and separated into two COmpOnents

clecm filtrate exits from end of shell

mode

concentrated waste buildup is forced out

feed theam

Fig. 22. Operation and backwash with Memtec cartridge. CONCLUSIONS

Flux decline due to fouling is an important problem in UF. The nature of UF membranes, particularly their relatively low surface porosity and pore size distributions, makes them sensitive to fouling by pore blockage. Surface chemistry, such as hydrophilicity is also important. UF membranes may suffer flux declines at various stages in their history. Significant loss of flux can occur due to foulants in apparently clean water.

135

0.25% 0

0

20

10

TIME

wilds 30

Imtnl

Fig. 23. Flux declines with air backwash and permeate backwash.

Under UF conditions flux drops, as expected, due to boundary layer formation and may also drop due to rapid fouling as pores are plugged. Long-term flux decline can occur due to solute deposition and possible compression of cake solids. Factors influencing solute deposition and long-term decline, include membrane properties, solute type, solution environment and those factors affecting polarisation such as cross-flow velocity. Models of the fouling process range from the (semi ) empirical to fundamental attempts to allow for the hydrodynamics of particle motion, and the coagulation of solute due to interactive forces. The strategies for control of fouling include pretreating the membranes with surfactants or polymers to increase the homogeneity and hydrophilicity of the surface. Operational factors such as backwashing also help to control fouling, and particular advantage is gained with air back-flushing for certain types of membrane. ACKNOWLEDGEMENTS

Much of the work described in this review was performed in the Membrane Research Group at the University of New South Wales. The authors wish to acknowledge the contribution of the following, Michael Chudacek, Kyu-Jin Kim, Tusirin Nor, Anhar Suki and Alan Waters. Financial support for some of this work was provided by the Australian Research Grant Scheme. The authors are also grateful for the information provided by Dr. Doug Ford of Memtec Ltd. Mr. Xiang-Zhou Jiang (Dalian Institute of Chemical Physics) is acknowledged for his help in preparation of the figures.

136 REFERENCES B.H.J. Hofstee, in: A.R. Cooper (Ed.), Polymeric Separation Media, Plenum, New York, 1982, pp. 87-92. K. Larsson, Desalination, 35 (1980) 105-l 14. A.G. Fane, C.J.D. Fell and A.G. Waters, J. Membr. Sci., 9 (1981) 245-262. A.B. Suki, Ph.D. Thesis, University of New South Wales, 1984. U. Merin and M. Cheryan, J. Appl. Polym. Sci., 25 (1980) 2139-2142. G. Capannelli, F. Vigo and S. Munari, J. Membr. Sci., 15(3) (1983) 289-313. S. Kimura, Presentation at the symposium on Membrane Technology, Tylosand, Sweden, May 28-30,1985. 8 H. Reihanian, C.R. Robertson and A.S. Michaels, J. Membr. Sci., 16 (1983) 237-258. 9 S. Munari, A. Bottino, G. Capannelli and P. Moretti, Desalination, 53 (1985) 11-24. 10 A. Mey-Marom and M.G. Katz, Proc. Europe-Japan Congress on Membranes, Stresa, June 1984, in press. 11 T.K. Sherwood, P.L.T. Brian and R.E. Fisher, MIT Desalination Research Laboratory, Rept. 295-1, Aug. 8,1963. 12 M.C. Porter, in: P.A. Schweitzer (Ed.), Handbook of Separation Techniques for Chemical Engineers, McGraw-Hill, New York, 1979, pp. 2-3 to 2-104. 13 P.C. Carman, Trans. Inst. Chem. Eng., 165 (1938) 168. 14 A.G. Fane, C.J.D. Fell, R.M. McDonogh, A. Suki and A.G. Waters, Proceedings Chemeca 83, Inst. Chem. Engineers, Brisbane, Sept. 1983,729-737. 15 M. Chudacek and A.G. Fane, J. Membr. Sci., 21 (1984) 145-160. 16 A.G. Fane, C.J.D. Fell and M.T. Nor, in: A.R. Cooper (Ed.), Ultrafiltration Membranes and Applications, Plenum, New York, 1980, pp. 631-658. 17 A.G. Fane, C.J.D. Fell and M.T. Nor, Inst. Chem. Eng. Sym. Ser., (73) (1982) Cl. 18 A.G. Fane, in: R.J. Wakeman (Ed.), Progress in Filtration and Separation, Vol. 4, Elsevier, Amsterdam, 1986, pp. 101-179. 19 M. Cheryan and U. Merin, in: A.R. Cooper (Ed.), Ultrafiltration Membranes and Applications, Plenum, New York, 1980, pp. 619-630. 20 F.A. Glover and B.E. Brooker, J. Dairy Res., 41 (1974) 89-93. 21 A.G. Fane, C.J.D. Fel1,andA.G. Waters, J. Membr. Sci., 16 (198313) 211-224. 22 D.G. Thomas and W.R. Mixon, Ind. Eng. Chem. Process Des. Dev., ll(3) (1972) 339. 23 G. Green and G. Belfort, Desalination, 35 (1980) 129-147. 24 M.C. Porter, Ind. Eng. Chem. Prod. Res. Deve., ll(3) (1972) 234. 25 S. Kimura and S.I. Nakao, Desalination, 17 (1975) 267. 26 R.G. Gutman, Chem. Eng., London, July 1977,510-523. 27 D. Bhattacharyya, A.B. Jumawan and R.B. Grieves, Sep. Sci. Technol., 14(6Z) (1979) 529-549. 28 A.B. Suki, A.G. Fane and C.J.D. Fell, J. Membr. Sci., 27 (1986) 269-283. 29 E. Matthiasson and B. Sivik, Desalination, 35 (1980) 59-103. 30 A.S. Michaels, C.R. Robertson and H. Reihanian, Proc. (Abstracts), International Membrane Technology Conference, Sydney, Nov. 1983,59-63. 31 A.G. Fane, C.J.D. FeIl and K.-J. Kim, Desalination, 53 (1985) 37-56. 32 H. Bauser, H. Chmiel, N. Stroh and E. Walitxa, J. Membr. Sci., 11 (1982) 321-332. 33 B.R. Breslau, A.J. Testa, B.A. Milnes and G. Medjanis, in: A.R. Cooper (Ed.), Ultrafiltration Membranes and Applications, Plenum, New York, 1980, pp. 109-128. 34 D. Ford, Memtec Ltd., Private Communication, 1986. 35 A.B. Suki, A.G. Fane and C.J.D. Fell, J. Membr. Sci., 21 (1984) 269-283. 1