Bioseparation 7: 1–7, 1997. 1c 1997 Kluwer Academic Publishers. Printed in the Netherlands.

Fouling of ceramic membranes during microfiltration of model solutions

O. Surel1 & M.-H. Famelart2

1Ecole Superieure d’Agriculture de Purpan, 75, voie du TOEC 31076 Toulouse cedex, France; 2INRA,Laboratoire de Recherche de Technologie Laitiere, 65, rue de Saint Brieuc, 35042 Rennes cedex, France

Received 8 January 1997; accepted in revised form 12 September 1997

Key words: cross flow filtration, fatty acids, fouling, inorganic membranes, minerals

Abstract

A model study of the microfiltration of fatty acids (C2 to C8) dissolved in a synthetic ultrafiltrate was performed.Ceramic membranes of 0.1 and 0.2 �m were used. Results showed that minerals from the ultrafiltrate were involvedin membrane fouling, while fatty acids were poorly rejected.

On 0.1 �m membranes, modelisation of fouling exhibited two different steps of fouling, a first one due toadsorption of solutes, and a second one involving particles deposit onto the membrane. On 0.2 �m membranes,only fouling due to particles deposit was observed in modelisation. Results of fouling and rejection pointed out thegreat difference between 0.1 and 0.2 �m membranes.

Abbreviations: AA, AB, AC membrane – surface modified membrane type A, B, or C; C2, C3, C4, C6, C8 – fattyacids containing 2, 3, 4, 6 or 8 atoms of carbon; Cp – permate concentration (g l�1); Cr – retentate concentration(g l�1); FA – fatty acids; J – membrane flux (l h�1 m�2); St – standard membrane; SMUF – simulated milkultrafiltrate; Rf – total fouling hydraulic resistance (m�1), Rm – membrane hydraulic resistance (m�1); Rirr –irreversible fouling hydraulic resistance (m�1); TP – transmembrane pressure (bar); TR – retention rate (%).

Introduction

Short chain fatty acids occur in ruminant milks andrelated products, such as ripened cheeses, in which theyare responsible for their typical flavour. Furthermore,some of them (especially propionic acid) have antisep-tic activities (Playne, 1985), and are largely used in thefood industry as preservative. Thus, the obtention ofsuch acids could be useful to produce cheese aroma orpreservative for the food industry.

In the field of biotechnology, membrane biore-actors have been used in numerous processes. Insuch a device, a standard fermentor is coupled to anappropriate membrane recycling a broth composed ofsubstrate, microorganisms and macromolecules, andextracting soluble molecules. This process increasesbiomass concentration and delays inhibition to takeplace. Membrane bioreactors had been applied suc-cessfully to short chain fatty acids production (Boy-aval and Corre, 1995). This kind of process needs a

membrane carefully chosen to avoid fouling and toachieve specific separation. In the field of membraneseparation, new membranes made of metallic oxides(ceramic) are emerging as relevant materials. They areresistant to high pressure, steam sterilization and there-fore are often coupled to fermentors.

This study is the first step of investigations focusedon the production of short chain fatty acid concentrates,using byproducts from milk (serum or permeate) assubstrate. Therefore, we made a model fluid to inves-tigate rejection of short chain fatty acids in a syntheticUF permeate. New kinds of ceramic membranes weretested in the microfiltration process.

JS: PIPS No. 149849 BIO2KAP

*149849* bios366.tex; 15/04/1998; 15:21; v.7; p.1

2

Table 1. Hydraulic resistance of membranes, Rm. The membranehas just been conditioned by a cleaning step: Ultrasil 11, 0.1% (w/v)at 70 �C during 10 min

Membrane 0.1 �m 0.1 �m 0.2 �m 0.2 �m 0.2 �m 0.2 �mSt. AA St. AA AB AC

Rm (1011 m�1) 2.50 2.86 4.44 5.61 5.77 4.48

Material and methods

Membrane and filtration module

Membralox 7/10 monotubes (SCT, France) of 0.1 or0.2 �m pore diameter were fitted in a filtration loopdescribed elsewhere (Surel and Famelart, 1994). 0.1�m membranes were made of zirconia and alpha alu-mina (standard membrane, St) or zirconia and alphaalumina associated to other metallic oxides (modi-fied membrane, AA). 0.2 �m membrane were madeof alpha alumina (standard membrane, St) or of alphaalumina associated to other metallic oxides (modifiedmembranes, AA, AB, AC). Hydraulic resistances ofthe membranes are given in Table 1.

Membrane cleaning was achieved by rinsing theplant with 5 l of distilled water and then by a cleaningsequence: Ultrasil 11 (Henkel, France) 1% (W/V) 70�C, 20 min, rinsing with distilled water until neutrali-ty, HNO3 (d=1.42) 1% (V/V), 65 �C, 20 min, rinsingwith distilled water. The membrane cleaning procedureallowed recovery of the initial hydraulic resistance val-ues. Before each experiment membranes were condi-tioned with Ultrasil 11 (Henkel, France) 0.1% (W/V),70 �C, 10 min.

Experimental conditions

Tangential flow velocity (V) was 2 m s�1, average TPwas 1 bar, temperature was 20 �C. Total experimenttime was 180 min. The permeate was recycled in thefeed tank. Experiments were done once, but experi-mental conditions and results were similar to those ofprevious work in our laboratory (Surel, 1993).

Model solutions

The model solution was a simulated milk ultrafiltrate(SMUF, Daufin et al., 1987). Its composition is givenin Table 2. If necessary, short chain fatty acids (aceticC2, propionic C3, butyric C4, caproic C6 and caprilicC8) purchased from Sigma (France) were added to the

Table 2. Composition of the simulated milk ultrafiltrate (SMUF)(Daufin et al., 1987). pH was 6.6. All the figures are given in mgl�1

Solute Concentration

Lactose 49500

Citrate 1601

PO43� total 401

PO43� soluble 358

Ca2+ total 400

Ca2+ soluble 362

Na+ 364

K+ 1601

Cl� 1180

SMUF to reach final concentration of 0.5 g l�1 for eachone. Fatty acids were added and pH was fixed at 6.6using NaOH 10 N.

Fatty acids quantification

To recover fatty acids, a sample of 10 ml of the solutionwas acidified with 3 drops of HCl 32% (W/V). Thisprocedure allowed transformation of salt into acids.This solution was mixed with 10 ml of ethylic etherduring 20 min with continuous stirring. This stage wasrepeated. An internal standard (valeric acid C5) wasused for GC quantification.

GC analysis were performed on a Varian GC 3400(France) apparatus. GC column was a FFAP Nukol(Supelco, France), length was 30 m, internal diameterwas 0.32 mm, and phase thickness was 0.25�m. Carri-er gas was hydrogen, and the flow velocity was 27 cms�1 The flame ionization detector was fixed on range10, its temperature was 260 �C. The injector temper-ature was 250 �C. The injector was kept splitless 30 sand then to a split ratio of 1/20. The temperature elu-tion profile was: isotherm 5 min 100 �C, 5 �C min�1 to160 �C, isotherm 5 min 160 �C. Integrated peak areaswere corrected according to response factor of eachfatty acid. This whole procedure allowed recovery ofC6 (106%), and C8 (103%). Extraction ratios of C2and C3 and C4 were 23%, 52% and 83%, respective-ly. These factors due to evaporation and solubility inwater were taken into account for each sample of thepermeate analysed.

bios366.tex; 15/04/1998; 15:21; v.7; p.2

3

Retention of fatty acids

Retention of fatty acids was calculated through theequation

TR = (1�Cp

Cr):100 (1)

TR is the retention rate in %, Cp and Cr are the con-centration of a fatty acid in the permeate and in theretentate, respectively.

Modelisation of fouling

The evolution of the flux with time was modelisedaccording to Aimar et al. (1986) for the adsorption pro-cess, followed by cake deposit by convection accord-ing to Coulson and Richardson (1967).The use of thesetwo models has been tested successfully by Taddei etal. (1989) on dairy fluids, and by ourselves in oth-er experiments performed with the same membranes(Surel and Famelart, 1994).

The equation used for fouling due to adsorptionwas

ln(1J��

1J(t)

) = � pCmxt+ ln(

1J��

1J0) (2)

where J(t) is the flux at time t, J0 is the initial flux attime t0, J* the flux at the end of the adsorption step(J , J0 and J* are in l h�1 m�2). p, Cm and x areparameters of the kinetic adsorption.

This model was suitable for experimental data if ln( 1J��

1J(t) ) versus time t was linear (J� is obtained

graphically from experimental data).Modelisation of cake deposit (i.e. build up of

deposited material on the membrane surface by con-vection, with the filtered material remaining on themembrane) was

1J(t)2

=1J 00

2+

2�aC0�S

TPt (3)

where J(t) and J00 (l h�1 m�2) are the permeate fluxesat time t, and at the beginning of the cake filtration step,respectively, � is the fraction of rejected particles, a isthe specific resistance of the cake, C0 is the concen-tration of particles in the solution, � is the dynamicviscosity of the permeate, S is the membrane area andTP is the transmembrane pressure. This model wasvalid if 1

J(t)2 versus time was linear.Total fouling was quantified using the equation

Rf

Rm

=Jc

J(t)� 1 (4)

where Rm is the hydraulic resistance of clean mem-brane measured with water. Rf is the total resistanceof the membrane and the deposit during experiment.(Rf and Rm are in m�1). J(t) is the flux at any timeof the experiment, Jc is the water flux of clean mem-branes at 20 �C and TP = 1 bar (J(t) and Jc are in lh�1 m�2).

The irreversible fouling was quantified by measur-ing water flux of fouled membranes before cleaningprocess. Results were expressed as ratios Rirr/Rm (Rirr

andRm are hydraulic resistances coming from Darcy’slaw, measured on fouled or clean membrane, respec-tively, in m�1).

Results

Table 3 summarizes the means of FA concentrationsin the permeate of the different membranes. Only 7of the 30 data showed a statistical difference with theamount of FA in the initial solutions. In those cases thecalculated FA retentions would raise from - 7% to 8%.Thus, the retention of FA appeared to be weak.

Despite of the small number of experimental datafor the adsorption model, two models were suitable for0.1 �m membranes. The cake filtration model alone,was suitable to modelise fouling of 0.2�m membranes.Typical curves are given in Figure 1. Model parametersare given in Tables 4 and 5. Correlation coefficientsranged from 0.82 to 0.96.

Fouling of 0.1 �m membranes was lowered with-out FA. Model parameters p Cm

x for adsorption and2�aC0S/TP for cake filtration were higher in the pres-ence of FA (0.06–0.09 against 0.03 without FA forthe former, and 1.7–1.8 against 1.1–1. 2 without fattyacids for the latter) (Tables 4 and 5). Similar resultswere obtained for irreversible fouling (2–1.6 against1.1–0.7 without FA) (Figure 2), and total fouling dueto adsorption Rae/Rm (2.3–2.6 against 1.2–2 withoutFA) (Table 4). On 0.1 �m membrane, adsorption wasthe main fouling phenomena. The difference betweenJ0 and J�, decrease due to adsorption, was 350–400 lh�1 m�2, whereas J� - J180, decrease due to deposit,was 36–83 l h�1 m�2 (Table 4). Cake filtration mod-el applied to fouling of 0.2 �m membranes showeda faster decrease of flux versus time (2 �aC0�S/TP)when SMUF free of FA was filtered (Table 4). Compa-rable results came from irreversible fouling measure-ments (averageRirr/Rm = 0.5 against 0.65 without FA)(Figure 2).

bios366.tex; 15/04/1998; 15:21; v.7; p.3

4

Table 3. Average concentrations of fatty acids (in g l�1) in the permeate of the different membranes

Membranes Fatty acid

C2 C3 C4 C6 C8

0.1 �m

St 0.50(0.035) 0.49(0.022) 0.47(0.001) 0.46*(0.005) 0.48*(0.001)

AA 0.51(0.013) 0.50(0.081) 0.47(0.001) 0.45*(0.015) 0.48*(0.014)

0.2 �m

St 0.56(0.034) 0.61(0.026) 0.57(0.019) 0.56*(0.001) 0.49 (0.016)

AA 0.54(0.043) 0.61(0.024) 0.57(0.015) 0.53 (0.050) 0.48 (0.014)

AB 0.55(0.033) 0.62(0.036) 0.58(0.021) 0.57*(0.013) 0.49 (0.032)

AC 0.55(0.029) 0.61(0.024) 0.57(0.017) 0.57*(0.006) 0.49 (0.019)

All the figures are the mean of 4 determinations at 0, 60, 120 and 180 min. Standard deviation is in brackets.*Shows figures significantly different from concentrations in the initial solution (P=0.01), using the student test. The initial concentrationswere C2 0.45(0.046), C3 0.48(0.025), C4 0.48(0.005), C6 0.49(0.005), C8 0.51(0.01) for 0.1 �m membranes and C2 0.54(0.056), C30.61(0.031), C4 0.57(0.005), C6 0.53(0.005), C8 0.46(0.01) for 0.2 �m membranes.

Figure 1. Variations of fluxes with time during microfiltration on 0.1 �m and 0.2 �m standard membranes (SCT) of a solution of short chainfatty acids dissolved into simulated milk ultrafiltrate. Flow velocity: 2 m s�1, TP: 1 bar. Solid lines: deposit by an adsorption model (Aimar etal., 1986). Dashed lines: model of cake filtration (Coulson and Richardson, 1967).

The behaviour of modified membranes was not dif-ferent regarding FA rejection (Table 3). Fouling ofmodified membranes AA and AB was lower than foul-ing observed on standard membranes when SMUF freeof FA was filtered (Figure 2).

Discussion

The weak rejection of FA was expected since theyare soluble in the SMUF. Assuming that pK of theseacids range from 4.75 to 4.89, nearly 98% of themare negatively charged at pH 6.6. In the present case,FA remain soluble and do not precipitate into calci-um salts as medium chain soaps would (Schwartz and

Perry, 1955). Furthermore, diameter of such organicions ranges from 4.5 to 8 �A (Kielland, 1937). All thisgoes to show that weak rejection of FA was due totheir small size. Negative retentions have already beenreported (Nau et al., 1995), but in the present case thisresult could be due to chemical analysis since a littlevariation of FA concentration in the permeate (about0.05 g l�1) can lead to significative rejection (around10%).

Several authors (Cheryan and Merin, 1981; Taddeiet al., 1986) showed membrane fouling due to synthet-ic ultrafiltrate. Their experiments were performed inorder to understand mineral fouling occurring duringwhey ultrafiltration. Minerals, especially phosphate,interact with membrane material (Cheryan and Merin,

bios366.tex; 15/04/1998; 15:21; v.7; p.4

5

Table 4. Parameters for the model of deposit by adsorption (Aimar et al., 1986) Logn ( 1J�

- 1J(t) ) = Logn( 1

J�- 1J0

) - pCxmt, where J(t)

is the flux at time t, J� is the final flux at the end of the adsorption step, and J0 is the flux at initial time t0. p and Cxm are the model

parameters, and r, the correlation coefficient. Ra is the hydraulic resistance with time due to adsorption, and is Ra = Rae (1-exp(�pCxmt)),according to Aimar et al. Rae/Rm is the normalized resistance due to adsorption (Rae/Rm = Jc/J*-1, where Jc and J* are the flux of theclean membrane and the flux at the end of the adsorption step, respectively)

Membrane/solution J� J0 J0-J* J* - J180 pCxm r Rae/Rm

(l h�1 m�2) (l h�1 m�2)

0.1 �m St./SMUF 501 912 411 46 0.0324 0.96 1.97

0.1 �m AA/SMUF 563 936 373 83 0.0303 0.93 1.22

0.1 �m St./SMUF+FA 390 732 342 36 0.0939 0.96 2.57

0.1 �m AA/SMUF+FA 383 708 325 70 0.0644 0.93 2.32

Table 5. Parameters for the model of cake deposit by convection(Coulson and Richardson, 1967) 1

J(t)2 = 1J00

2 + 2�aC0�S

TPt,

where J(t) is the flux at time t, J0‘ is the flux at t0, i.e. at thebeginning of the cake filtration step, � is the fraction of particlesretained onto the membrane, a is the specific resistance of the cakedeposit, C0 is the particle concentration in solution, � is the dynamicviscosity of the solute phase, S, is the membrane surface and TP isthe transmembrane pressure. r is the correlation coefficient

Membrane/solution J0’ 2�aC0�S/TP r

(l h�1 m�2) (10�8)

0.1 �m St./SMUF 610 1.18 0.85

0.1 �m AA/SMUF 648 1.09 0.86

0.1 �m St./SMUF+FA 407 1.80 0.89

0.1 �m AA/SMUF+FA 453 1.71 0.93

0.2 �m St./SMUF 548 1.81 0.95

0.2 �m AA/SMUF 491 2.45 0.92

0.2 �m AB/SMUF 471 1.83 0.93

0.2 �m AC/SMUF 559 1.95 0.94

0.2 �m St./SMUF+FA 581 1.35 0.82

0.2 �m AA/SMUF+FA 514 1.66 0.91

0.2 �m AB/SMUF+FA 487 1.24 0.93

0.2 �m AC/SMUF+FA 597 1.47 0.94

1981; Daufin et al., 1991; Noordman et al., 1995) orprecipitate into the membrane pores as calcium phos-phate (Hickey et al., 1980; Taddei et al., 1986), ratherwhen the solution is free of proteins (Maubois, 1980).Modelisation of fouling confirmed both hypothesis,since fouling exhibited a first step of adsorption fol-lowed by cake deposit. Ceramic is supposed to bepositively charged at pH 6.6 since its zero point chargeis around 9 (Hoogland et al., 1989). 0.1 �m mem-branes have a filtration layer made of zirconia whichzero point charge is lower than alumina’s (6 to 7, Pater-son, 1989). Thus, at the operating pH, 0.2 �m standard

Figure 2. Normalised hydraulic resistance for final fouled mem-branes (Rirr/Rm) after microfiltration of simulated milk ultrafiltrate(SMUF) alone, or of short fatty acids dissolved into the SMUF.

membrane is supposed to wear a quite high positivecharge, whereas 0.1 �m membrane could be near fromglobal neutrality. According to our results, adsorp-tion of solutes occurred on 0.1 �m but not on 0.2�m membranes. Also, modified membranes are sup-posed to have a lower zero point charge than standardmembranes, but its effect was not drastic regardingfouling of 0.2 �m membranes. Therefore, adsorptiondeposit due to electrostatic interactions are not suffi-cient to explain fouling. Thus, rejection mechanismmust involve either a drastic modification of electro-static interactions arising from the ionic environment(Laurent et al., 1984), or the pore size of the mem-branes. Convection deposit due to aggregates in the

bios366.tex; 15/04/1998; 15:21; v.7; p.5

6

bulk solution was consistent with calcium phosphateprecipitation (Jenness and Koops, 1962; Taddei et al.,1986). The latter authors clearly demonstrate that asolution made of lactose and calcium phosphate canform a precipitate after 4 h of filtration. If this phe-nomenon occurs, they observe a drastic decrease ofmembrane flux, and a rejection of lactose and calcium.The resulting fouling of the membraneRirr/Rm is 5.67which is quite high compared to our results (0.6 to 1.1,Figure 2). In fact, during our experiments, the smallerTP (1 bar instead of 2.8) did not lead to pore blocking.

Modelisation of fouling of 0.2 �m membranes didnot exhibit fouling due to adsorption whereas foul-ing due to adsorption was observed on 0.1 �m mem-brane. Therefore, adsorption ability of 0.2 �m mem-brane must be lower than 0.1 �m membranes. Thisresult has also be noticed in microfiltration experimentsperformed in very different conditions with the samemembranes (Surel and Famelart, 1994). During thestep of fouling due to cake deposit, the kinetic parame-ter 2�aC0�S/TP was higher for 0.2�m than for 0.1�mmembranes without fatty acids. This phenomenon wasprobably due to calcium phosphate aggregates enteringinto the pores of 0.2 �m membrane while they wererejected by 0.1 �m membranes. If FA were added inthe SMUF, they increased fouling and were rejected(5.6 to 7.7% retention of C6 and C8), and transmittedinto the permeate of 0.2 �m membranes, decreasingfouling. So, kinetic parameter 2�aC0�S/TP becamehigher on 0.1 �m membranes if FA were added.

The choice of a membrane must be based on mem-brane porosity but also on its material compositionsince it can modify rejection and fouling.Further inves-tigation is needed to understand how calcium phos-phate is precisely involved in membrane fouling oneach type of membranes. Particularly, further modeli-sation with other models (Bowen et al., 1995) or mod-els applied to cross flow filtration (Field and Arnot,1995) must be relevant to investigate fouling mecha-nisms.

Conclusion

Ceramic membranes appeared to transmit short FA.Therefore, they could be successfully used by cou-pling to fermentors if short chain FA would be pro-duced. Their sensitivity to fouling was different fromone membrane to an other. From this point of view,0.2 �m would be the best to use since the FA rejec-tion is negligible, and their fouling level quite low. In

all cases, mineral substances were the major cause offouling. Such membranes could be used in a membranebioreactor process, using dairy byproducts containingresidual milk fat that would be hydrolyzed by lypoliticmicroorganisms. The final product would be a FA solu-tion, used as blue cheese aroma or food preservative.

Acknowledgements

The authors thank the Societe des Ceramiques Tech-niques and the French Ministry of Research and Tech-nology for their financial support.

References

Aimar P, Baklouti S and Sanchez V (1986) Membrane solute inter-actions : influence of pure solvents transfer during ultafiltration.J. Membrane Sci. 29: 207–224.

Bowen WR, Calvo J and Hernandez A (1995) Steps of membraneblocking in flux decline during protein microfiltration. J. Mem-brane Sci. 101: 153–165.

Boyaval P and Corre C (1995) Production of propionic acid. Lait75: 453–461.

Cheryan M and Merin U (1981) The effect of protein - mineral -membrane interactions on the fouling of ultrafiltration mem-branes. In: B Hallstrom, DB Lund and G Tragardh (eds) Foul-ing and Cleaning in Food Processing (pp. 204–214). Lund Univ.Press, Sweden.

Coulson JM and Richardson JF (1967) Chemical Engineering. Perg-amon Press, Oxford, 7 ed. UK.

Daufin G, Labbe JP, Quemerais A, Brule G, Michel F, Roignant Mand Priol M (1987) Fouling of a heat exchanged surface by whey,milk and model fluids. An Analytical study. Lait 67: 339–364.

Daufin G, Labbe JP, Quemerais A and Michel F (1991) Fouling ofan inorganic membrane during ultrafiltration of defatted wheyprotein concentrates. Neth. Milk Dairy J. 45: 259–272.

Field RW and Arnot TC (1995) Fouling mechanism and modellingwith due allowance for cross flow and back diffusion : testing ofrecent theoretical advances with data on the membrane filtrationof yeast cells. In: WR Bowen, RW Field and JA Howell, (eds)Euromembrane 18–20 Sept. 1995 (I: pp. 17–22), University ofBath, UK.

Hickey MW, Hill RD and Smith BR (1980) Investigations into theultrafiltration and reverse osmosis of wheys. 1–The effects ofcertain pretreatments. New Zeal. J. Dairy Sci. Technol. 15: 109–121.

Hoogland MR, Fane AG and Fell CJD (1989) The effects of pHof crossflow filtration of mineral slurries using ceramic mem-branes. In: AJ Burgraaff, J Charpin, L Cot, (eds) 2nd Int. Conf.of Inorganic Membranes, l-4 Jul, Montpellier (pp. 153–162).Trans. Tech. Publ., Switzerland.

Jenness R and Koops J (1962) Preparation and properties of a saltsolution which simulates milk ultrafiltrate. Neth. Milk Dairy J.16: 153–164.

Kielland J (1937) Individual activity coefficients of ions in aqueoussolutions. J. Am. Chem. Soc. 59: 1675–1678.

bios366.tex; 15/04/1998; 15:21; v.7; p.6

7

Laurent CJCM, Billiet HAH, Degalan L, Buyttenhuys FA, Van derMaiden FBP (1984) High performance liquid chromatographyof proteins on alumina J. Chromatogr 287: 45–54.

Maubois J-L (1980) Ultrafltration of whey. J. Soc. Dairy Technol.33: 55–58.

Nau F, Kerherve F-L, Leonil J and Daufin G (1995) Selective sep-aration of tryptic beta - casein peptides through ultrafiltrationmembranes: influence of ionic interactions Biotechnol. Bioeng.46: 246–253.

Noordman TR, Vonk P and Wesselingh JA (I995) Phosphate rejec-tions by a ZrO2 membrane. In: WR Bowen, RW Field and JAHowell (eds) Euromembrane 18–20 Sept. 1995 (I: pp. 190–192),University of Bath, UK.

Paterson R (1989) Surface charge polarity and ion binding char-acteristic of ceramic oxydes. In: AJ Burgraaff, J Charpin andL Cot (eds) 2nd Int. Conf. of Inorganic Membranes, 1–4 July,Montpellier, (pp. 127–134). Trans. Tech. Publ., Switzerland.

Playne MJ (1985) Propionic and butyric acid. In: Moo Young M(ed) Comprehensive Biotechnology, Vol. 3 Chap. 37, PergamonPress, Oxford, UK.

Schwartz AM and Perry PW(1955) Chimie et technologie des agentstensio - actifs, Dunod, Paris, France.

Surel O (1993) Lipides et microfiltration : contribution a la com-prehension des phenomenes de transfert PhD Thesis, Rennes,France.

Surel O and Famelart M-H (1994) Microfiltration of long chaincarboxylates: relationship between rejection and micellization.Desalination 95: 11–25.

Taddei C, Daufin G, Aimar P and Sanchez V (1989) Role of somewhey components on mass transfer during sweet whey ultrafil-tration. Biotechnol. Bioeng. 34: 171–179.

Taddei C, Aimar P, Daufin G and Sanchez V (1986) Etude du trans-fert de matiere lors de l’ultrafiltration de lactoserum doux surmembrane minerale. Lait 66: 371–390.

bios366.tex; 15/04/1998; 15:21; v.7; p.7