Whey crossflow microfiltration using an M14Carbosep membrane: influence of initial hydraulic
resistance
G Gésan 1, G Daufin 1, U Merin2
1 Laboratoire de Recherche de Technologie Laitière,INRA, 65, rue de Saint-Brieuc, 35042 Rennes Cedex, France;
2 Dairy Science Laboratory, ARO The Volcani Center, PO Box 6, Bet Dagan 50250, Israel
(Received 31 December 1993; accepted 6 April 1994)
5ummary - Membrane hydraulic resistance (Rm) for the same nominal M14 Carbosep membrane(pore diameter 0.14 um) was found to vary significantly from 0.85 ± 0.05 to 1.22 ± 0.09 1012 m-1. Pre-treated whey microfiltration experiments carried out using these new membranes resulted in shorter oper-ating time and lower protein transmission for the low Rm membranes. When the same membranewas used new, irreversibly fouled or cleaned (Rm = 1.14, 2.48and 1.261012 m-1 respectively) foulingevolution was positively correlated with Rm increase. It was assumed that a low Rm of a new membraneindicated a larger population of large pores which under constant permeation flux experiments filteredlarger volumes and thus fouled faster. Residual fouling after c1eaning was found to have a negative effecton performance presumably due to alteration of the membrane's morphological characteristics and toelectrostatic and hydrophobie interactions of the feed stream with the residual fouling layer.
Résumé - Microfiltration tangentielle de lactosérum avec une membrane M14 Carbosep :influence de la résistance hydraulique initiale. L'objectif de ce travail est de montrer les consé-quences de la variation de la résistance hydraulique (Rm) de la membrane M14 Carbosep (diamètrede pores 0, 14 ~m) à l'état neuf et nettoyé (avec colmatage résiduel) sur les performances de la micro-filtration tangentielle de lactosérum doux pré traité. Deux lots de membranes (30 au total) montrent quela résistance Rm de cette membrane varie de 0,85 ± 0,05 à 1,22± 0,09 1012 trrt . En répétant une même
Abbreviations. A, membrane area (m2); Cp, concentration of a component in the permeate (gol-1); Cr,concentration of a component in the retentate (gol-l); J, permeation flux (Ioh-10m-2); 00, optical den-sity; Pr, mean pressure of the retentate compartment (105 Pa); Or, retentate extraction flow rate(mSos-1); T, temperature (OC); TP, transmembrane pressure (105 Pa); Tr, transmission (%); v, flowvelocity (mog-'); VCR, volume concentration ratio; Rf, ove rail fouling hydraulic resistance (rrr1); Rif, irre-versible fouling hydraulic resistance (m-1) Rm, initial membrane hydraulic resistance (m-1); RR, reten-tion (%).
268 G Gésan etai
microfiltration avec différentes membranes M14, il s'avère qu'une membrane neuve de forte perméa-bilité initiale (Rm petit) réduit le colmatage aux temps courts, mais diminue la durée opératoire de lamicrofiltration. Quand une même membrane est utilisée neuve ou irréversiblement colmatée ou net-toyée (Rm = 1,14; 2,48; 1,26 1012 ttr! respectivement), un grand Rm réduit les performances de la micro-filtration (durée opératoire plus faible, moins bonne récupération des protéines). De tels résultats peu-vent être expliqués par l'existence d'une plus grande population de grands pores dans les membranesneuves de petit Rm : lors de microfi/trations menées à flux de perméation constant, un écoulement pré-férentiel se crée dans les zones de faible résistance, c'est-à-dire dans les grands pores, qui filtrent pluset se colmatent plus vite. Le colmatage résiduel après nettoyage affecte les performances de l'opération,ce qui peut être la conséquence d'une réduction de l'aire filtrante efficace et d'interactions électrosta-tiques et hydrophobes entre le produit à filtrer et le colmatage résiduel. Ces résultats mettent en évi-dence la nécessité de rechercher des procédures adéquates de nettoyage de membranes de micro-filtration.
Characteristics of membranes are used asguidelines -for research, to acquire betterknowledge conceming the fouling and trans-mission properties of membranes, and forthe use of membranes in filtration processes.Several techniques can be used to assessmembrane pore size and distribution: scan-ning electron microscopy (Strathmann et al,1975; Kim et al, 1990, 1991); transmissionelectron microscopy (Merin and Cheryan,1981; Kim et al, 1991); tunnel and atomicforce microscopy (Chahboun et al, 1992;Dietz et al, 1992); liquid displacement meth-ods (Capannelli et al, 1983; McDonogh et al,1992); and size exclusion techniques (DeBalmann and Nobrega, 1989). However,those methods are almost impossible to beimplemented by the industrial user. A dis-advantage is the need to measure concen-trations of transmitted and/or rejectedsolutes using analytical instruments or tobreak or cut the membrane for direct obser-vation. Moreover, they are olten irrelevantwith respect to the membrane's actual per-meability. In this regard, a preferable methodis one which does not involve special instru-mentation and can rely on hydraulic rnea-surements which are always performed bythe user.
Initial membrane hydraulic resistance(Rm) according to water flux is an easy wayto characterize water permeability of themembrane and consequently to roughlydescribe membrane c1eanliness on-line. Itis shown by Taddéi et al (1989) that the irre-versible fouling hydraulic resistance can becalculated using the method of measuringwater flux of the fouled membrane. Thesame technique is used for studying thec1eaning efficiency of milk and whey ultra-filtration (UF) membranes (Daufin et al,1991 b, 1992).
ln the analysis of the performance of anindustrial microfiltration (MF) plant Rm isfound to have a significant influence on oper-ating time and th us on the ove rail MF oper-ation (Gésan et al, 1993b). The role of Rmon membrane operating performancereceived some attention (Davis and Birdsell,1987; Nilsson, 1988; Persson and Nilsson,1991), but still many researchers neglectits effect on filtration modelling andschematic presentation of fouling which isused for comparison of filtration experi-ments.
The aim of this work was to study theinfluence of initial membrane hydraulic resis-tance on whey MF performance, which hasemerged as a necessary step in producinghigh protein concentrates. lt consists of
Membrane resistance and MF performance
showing the consequences of variation inRm values on fouling evolution in the courseof timeand on theschematic presentation offouling. The work was performed by usingmembranes of the same nominal pore sizein three different conditions (new c1ean; irre-versibly fouled; fouled and c1eaned) whichgave different Rm values for testing. In addi-tion, evaluation of the procedure of Rm rnea-surement for MF membranes is also pre-sented.
MATERIALS AND METHODS
Membranes
The membranes used were M14 Carbosep mem-branes (Tech-Sep, Miribel, France). The M14membrane is a composite membrane with a0.14 urn mean pore diameter and Zr02 and Ti02filtering layer on a carbon support (6 mm innerdiameter, 1.2 m long; 2.26 10-2 m2). The mem-branes used were from two production batches.One membrane was used in three consecutiveexperiments: first as a c1ean membrane for anMF test, secondly after rinsing of the reversiblefouling layer and finally after c1eaning at the endof an MF run.
Feed
Whey was obtained at a local dairy plant (Préval,Montauban de Bretagne, France) from Emmen-tal cheese production. The whey was prefilteredon a 20 urn industrial filter, defatted in the plant bycentrifugation (9 000 9 at SO°C) and cooled downat the laboratory (2-4°C in S min). MF was carriedout with the pretreated whey: this operation resultsin a retentate composed of aggregates of Iipidsand calcium phosphates and a microfiltrate thatcontains proteins suitable for producing high puritywhey protein concentrates. Aggregation of resid-ual phospholipids was performed by a procedureadapted from Fauquant et al, (198S): whey at2-4°C, plus CaCI2 up to a calcium concentrationof 1.2 g-kg-' and 10 N NaOH, pH adjusted to7.2, was heated to SO°C in 30 min with a hold-ing time of 1S min. During the pretreatment, whey
269
pH decreased down to 6:4. NaN3 at 0.2 g-I-' wasadded to the whey to prevent bacterial growth.The chemical composition of the pretreated wheyis presented in table 1.
Chemical analyses
The whey and permeate and retentate sampleswithdrawn during MF were analysed for œ-lac-talbumin and ~-Iactoglobulin by reverse phasehigh performance liquid chromatography accord-ing to Jaubert and Martin (1992). Calcium wasdetermined by atomic absorption (Varian AA 300,Les Ulis, France) as described by Brulé et al(1974), phosphorus by mineralization and col-orimetric determination following the AFNORstandard NF V 04-284 (AFNOR-ITSV, 1986) tur-bidity by optical density (00) at 600 nm (Beck-man DU 62, Gagny, France). Density was deter-mined using a densimeter (Haake DM 48,Karlsruhe, Germany), and the dynamic viscosityof f1uids using a microdensimeter (Haake 08,Karlsruhe, Germany).
Microfiltration experiments
The MF rig, described in detail by Daufin et al(1993), was equipped with a single membranetube. MF, cleaning and water flux measurementswere performed with permeate circulating co-eur-rent to the retentate in order to create a permeatepressure drop equal to the retentate pressure
Table 1. Average composition of pretreated whey.Composition moyenne du lactosérum pré traité.
Mean Standardvalue deviation
T of pretreatment (oC) SO.3 ±O.SFinal pH of pretreatment(at SO°C) 6.40 ± O.OSa-Iactalbumin (g-kg -1) 0.73 ± 0.04~-Iactoglobulin (g-kg-') 3.6S ± 0.24Optical density (600 nm) 1.7 ± 0.4Calcium (g-kg-') 1.16 ± 0.04Phosphorus (g-kg-') 0.38 ± 0.02
270 G Gésan etai
drop, so as to get no transmembrane pressure(TP) difference along the membrane's hydraulicpath(Sandblom, J 974; Plett,.1989).
Experimental procedure for condition-ing and cleaning
The membrane was conditioned or cleanedbefore each MF experiment following thesequence: 1) acid wash: 55% technical HN03 at3 mlol-1 (0.03 molol-1), pH z 1.2 at 50°C (45 min);2) water rinse using 0.2 urn filtered tap water after5 and 2 urn filters in series (10 min); 3) hypochlo-rite solution containing 1 gol-1 active chlorine andpH adjusted to 11 using 10 N NaOH (30 min);and 4) filtered water rinse (10 min).
The acid and the alkaline solutions were pre-pared with the filtered water. The washes wereperfarmed at 50°C with flow velocity (v) 5.9 rn-s-t,retentate compartment pressure (Pr) 2.1 105 Paand 0.4 105 Pa constant TP.
Water flux measurement
After the cleaning and before the experiment,water flux was measured at the experimental con-ditions, ie filtered tap water (5, 2, and 0.2 urn),T = 50°C, v = 4.5 mos-1, Pr = 2.1 105 Pa, asdescribed by Gésan (1993) after a comparativestudy using water of different quality. Fluxes weremeasured at four TP values between 0 and 0.5105 Pa. At each given TP value flux was left tostabilize for 5 min before data were acquired.
Microfiltration of whey
Micrafiltration was performed at constant per-meation flux (J). This mode is practised by theindustry in order to ensure a continuous feed tothe subsequent UF plant (Gésan et al, 1993b).Consequently TP increase meant that membranefouling occurred, therefore TP will be usedthroughout the text to describe fouling along withthe ove rail fouling hydraulic resistance (Rf).
When the rig was tumed to operate from waterto whey, flow velocity (v) was decreased to3.0 mos-1 and 2 1 (z x 2 the dead volume of theloop) of the pretreated whey were used to rinsethe filtration loop from water before operationstarted.
Operating conditions
Operating parameters such as temperature, vol-ume concentration ratio (VCR) and flow velocity(v) were chosen according to operating condi-tions practised by industrial plant (Gésan et al,1993b). J was chosen in arder to foui the mem-brane fast and this way to reduce the MF operat-ing time (Gésan, 1993). The retentate pressurelevel was set in order to get a final TP that willcause high enough fouling to allow to model theevolution vs time but not too severe to ensureRm recovery by cleaning.
Transient conditions ta stationary operatingconditions
When the feed tank was full with the whey, v wasraised to its set value (5.9 m-s ") at a rate of0.5 mos-10min-1 through the use of the automaticcontroller. In the meantime, permeate circulationwas set to its automatic mode (a detailed descrip-tion of the loop control was given by Daufin et al,1993).
When the set v was attained, Pr was set to itsautomatic mode to get to the target value (4.1105 Pa) in 5 min.
When the set Pr was reached, the permeationvalve was turned on to increase J to its set valueof 121 loh-10m-2 at the rate of 10 loh-10m-2 permin. The elapsed time fram start to steady Jwas z 30 min.
Stationary conditions
Filtration was performed in a concentration modeuntil VCR reached the set value (VCR=5.2) andVCR was maintained constant by extracting reten-tate at a flow of Or (m30s-1), working in a feedand bleed mode:
Or=JoA/(VCR-1)
where A is the membrane area (m2).
At the end of the run, retentate and permeatevalves were closed, Pr was relieved and flowvelocity of both retentate and permeate wasdecreased to the rinsing flow velocity manually.
Rinsing
After the completion of a run the membrane wasrinsed with filtered tap water at 45°C for30-35 min by increasing v to the test flow velee-
Membrane resistance and MF performance
ity (5.9 mos-1) in about three cycles, ie until nomore proteins were released from the materialsof theretentate'compartment(stainless steel,rubber seals, etc) and from the membrane sur-face (no visible foam in the feed tank). Then, Prwas set to 2.1 105 Pa and J to 100 loh-1om-2 for5 min. Alter the rinsing, water flux was measuredas indicated.
Calculations
Hydraulic resistances
Calculation of Rm, Rf and irreversible foulinghydraulic resistance (Rif) were done accordingto Darcy's law. The slope of a plot of J versus TPof the measured water flux of a c1ean membraneand alter the experiment gave the values of Rmand Rif (Taddéi et al, 1986). Rf was calculatedfrom TP measurements during the experimentsaccording to Daufin et al (1991 a).
Errors in measurements
Taking into account the errer of each sensor (fourpressure gauges (0.02 105 Pa each), tempera-ture (0.4%), viscosity (1%) and permeation flux(3%)) and the relative standard deviation thatcharacterizes the dispersion of results obtainedfrom a series of measurements of the same vari-able, the calculated errer amounted to ~ 27% ofthe resistance. Nevertheless, consecutive mea-surements made on the same type M14 Car-bosep membrane and on different days showed<5% errer in hydraulic resistance measurements,either for a clean or for an irreversibly fouledmembrane. lt was therefore concluded that thismeasured errer is a better estimate of the actualerrer in Rm and Rif and was therefore used inthe evaluation of cleaning, fouling and comparisonof results.
Transmission
Transmission (Tr) and retenti on (RR) were cal-culated as follows:
RR = 1 - Tr = 1 - Cp/Cr
with Cp the concentration of the component inthe permeate and Cr its concentration in the reten-
271
tate withdrawn at the same MF time. This calcu-lation was assessed assuming that the perme-ate compartment was a perfectly stirred reactor(Gésan, 1993). Accounting for the dilution of ana-Iysed components in the permeate compartment,relative errors of calcium and phosphorus reten-tions varied with operating time from 20 to 10%,and for a-Iactalbumin and p-Iactoglobulin trans-missions from 40 to 10%. As can be seen, theexperimental errors are much larger at short timeof operation, when the dilution is higher.
RESULTS
Rm variability of new membranes
The calculated Rm 01 30 new membranes 01two different production batches (n1=20; n2= 10) showed Rm to vary Irom0.85 ± 0.05 to 1.22 ± 0.091012 rrr ". Themean Rm value 01 those 30 membraneswas 1.041012 m-1 (standard deviation:0.10 1012 rrr "). The two batches signili-cantly differed (P<0.05) in their averageresistance (1.09 and 0.921012 m-1).
There were visu al differences betweenthe membranes with respect to the color ofthe filtering layer which ranged from white togreyish black showing the heterogeneityfrom one batch to another and Irom onemembrane or part of it to another (such acolor heterogeneity was also noticed whiledismantling an industrial 8252 Carbosepmodule).
Influence of Rm on microfiltrationperformance
New membranes
The present detailed reported results wereobtained under the operating conditionsdescribed in Materials and methods. Theinfluence of initial membrane hydraulic resis-tance, Rm on MF performance was con-
272 G Gésan etai
firmed using other operating conditions:static counter pressure mode (without recir-culation of permeate); different whey pre-treatment (temperature of whey -55°C, andpH maintained at 7.5 during the pretreat-ment (Gésan et al, 1994)).
Permeability
The measured TP wh en J reached its setvalue was positively correlated (r2 = 70 %)to Am (fig 1). The lower the Am, the lowerthe TP. Two runs with similar Am mem-branes resulted in almost identical foulingevolution (fig 2), underlining the quality ofrig operation control and the repraducibilityof the pretreatments. In comparison to mem-branes with this Am (1.17 -1.181012 m-1)
the low Am membrane (1.03 1012 rrr ")(from the same production batch) resulted inless fouling in time < 1 h. Thereafter, thelow Am membrane fouled fast and its oper-
Transmembrane pressure (10' Pa)when steody J is reached
0'
r2 = 0.70
0.5
o .•
0.30+-.,----~----___:c----~1.0 1.2
Initial membrane resistance, Rm (1012
nï1 )
Fig 1. Transmembrane pressure when the steadypermeation flux (J) is reached, vs the initial mem-brane hydraulic resistance, Rm. Operating con-ditions: Pretreated whey; M14 membrane; v =5.9 mos-1; Pr = 4.1 105 Pa; J = 121 loh-1om-2;VCR = 5.2; T = 50°C.Pression transmembranaire quand le flux de pet-méation stationnaire (J) est atteint, en fonctionde la résistance hydraulique initiale de la mem-brane, RM. Conditions opératoires: Lactosérumpré traité; membrane M14; v= 5,9 m-s:'; Pr= 4, 1105 Pa; J = 121 loh-1om-2; VCR (facteur deconcentration volumique) = 5.2; T = 5CJ'C.
ating time was 30 min shorter than that ofthe high Am membranes (fig 2).
Selectivity
Calcium and phosphorus retentions weresimilar for the whole range of Am valuesstudied and increased with time and VCA(fig 3). Calcium retention increased from65% to 80% du ring VCA increase fram 1.0to 5.2 and stabilized at 80% at VCA = 5.2.Phosphorus retenti on increased from 80% to90% for the same conditions. The 00 of thepermeate along the experiment was0.02-0.05, independent of Am, indicatingefficient clarification.
The transmission of the two major pro-teins of whey (a-Iactalbumin and ~-Iac-toglobulin) decreased with time as foulingincreased (fig 4). a-Lactalbumin transmis-sion was always higher than that of ~-Iac-toglobulin. At short time, until ~ 80 min oper-ation the evolution of both proteins'transmission had a similar behaviour regard-
Fig 2. Volume concentration ratio (VCR) and trans-membrane pressure vs filtration time for three newM14 membranes with different initial membranehydraulic resistance Rm (expressed in 1012m-1).Operating conditions: see legend to figure 1.Facteur de concentration volumique (VCR) etpression transmenbranaire en fonction du tempsde filtration pour 3membranes M14 neuves de dif-férentes résistances hydrauliques initiales Rm(exprimées en 1012nru Conditions opératoires:voir figure 1.
Membrane resistance and MF performance
less of membrane Rm, and decreasedthereafter faster with the low Rm membrane.
Membrane of different stateof cleanliness
Permeability
A second set of experiments was performedusing the same membrane after condition-ing (Rmuse 1 = 1.14 ± 0.05 1012 m-1, for theclean membrane), after water rinse (Rmuse2 .= 2.48 ± 0.16 1012 m-1, for the irreversiblyfouled membrane) and after cleaning(Rmuse 3 = 1.26 ± 0.26 1012 m-1). Rmuse 3
was about 10% higher than Rmuse1 indi-cating that the membrane was not properlycleaned. Figure Sa shows the evolution ofTP vs time during the experiment whichlasted 60 min. The higher Rm membrane
Fig 3. Calcium and phosphorus retentions vs fil-tration time for three new M14 membranes withdifferent initial membrane hydraulic resistanceRm: Rm = 1.03 ± 0.041012 m-1 ( ....... ); Rm = 1.17± 0.101012 m-1 (--); Rm = 1.18 ± 0.041012
rrr1 (---). VCR: volume concentration ratio. Oper-ating conditions: see figure 1.Rétentions du calcium et du phosphore en fonc-tion du temps de filtration pour 3membranes M14neuves de différentes résistances hydrauliquesinitiales Rm: Rm = 1,03± 0,04 10'2 nri (.......);Rm = 1,17± 0,1010'2 rtrt (-); Rm = 1,18±0,04 10'2 rrr' (----).VCR: facteur de concentra-tion volumique. Conditions opératoires: voirfigure 1.
273
resulted in higher TP du ring the experiment,indicating the higher pressure needed tomaintain the set J-(sl:Jch results were con-firmed by further experiments performed atother operating conditions and longer dura-tion). When the results were plotted as resis-tance (Rf) evolution vstime (fig Sb) showedlower Rf value for the higher Rm membrane,but only for about 30 min when Rf started toincrease at a faster rate compared to theother two membranes. lt should be notedthat Rf is the addition of fouling during theexperiment to the already existing resis-tance, that of the membrane and of theuncleaned residual fouling.
Selectivity
During the above experiments with the threemembranes at different fouling state, cal-
Fig 4. Protein transmission vs filtration time forthree new M14 membranes with different initialmembrane hydraulic resistance Rm: Rm = 1.03 ±0.041012 m-1 ( ....... ); Rm = 1.17 ± 0.101012 m-1(-); Rm = 1.18 ± 0.041012 m-1 (----). VCR:volume concentration ratio. Operating conditions:see legend to figure 1. Errer bars are similar for p-lactoglobulin.Transmission des protéines en fonction du tempsde filtration pour 3membranes M14 de différentesrésistances hydrauliques initiales Rm: Rm = 1,03± 0,04 10'2 rrr' (.......); Rm = 1,17± 0,1010'2ttrt (----);Rm = 1,18± 0,0410'2 rtrt (----). VCR:facteur de concentration volumique. Conditionsopératoires: voir figure 1. Les barres d'erreur sontsimilaires pour la {3-lactoglobuline.
Fig 5. Evolution of transmembrane pressure (TP) (a); overall fouling hydraulic resistance (Rf) (b) vs fil-tration time for the same membrane after conditioning (Rm = 1.14 ± 0.05 1012 m-1, clean membrane--) after water rinse (Rm = 2.48 ± 0.16 1012 m-1 fouled membrane .....) and after cleaninq (Rm = 1.26± 0,26 1012 m-1• not properly cleaned membrane-Jo Operating conditions: see legend to figure 1.Évolution de la pression transmembranaire (fP) (a) ; la résistance hydraulique de colmatage global (Rf)(b) en fonction du temps de filtration pour la même membrane après conditionnement (Rm = 1, 14 ± 0,0510'2 tïr", membrane propre --) après rinçage à l'eau (Rm = 2,48± 0,1610'2 trrt; membrane col-matée ....... ) et après nettoyage (Rm =1,26±0,26 10'2ttrt, membrane nettoyée, non propre -). Condi-tions opératoires: voir figure 1.:
cium and phosphorus retentions (68-69%and 81- 82% respectively) and proteintransmissions (Tra-Lactalbumin= 90-95% andTr~-Lactoglobulin= 73-80%) were not signifi-cantly different (this was the consequence ofthe too short time of experiment, and of thelarge error on calculations due to dilution ofcomponents in the permeate compartment).
DISCUSSION
The hydraulic resistance of new M14 mem-branes used in this work highlighted theinfluence of their wide range (=o 40% differ-ence) of permeability on MF performance.The same trend was also observed withorganic UF membrane pieces cut from aDDS GR 61 P membrane (Nilsson, 1988),or Romicon PM50 hollow fiber polysulfonemembranes with various Rm of 1.2-7.8 1012
m-1 (Devereux and Hoare, 1986).
The variation in Rm of different produc-tion batches and within the same batch
1
might be due to the heterogeneity of the fiI-tering layer (thickness, quality of the Zrand/or Ti oxides, production procedures,etc) and/or of the carbon support charac-teristics (po rosity, pore size distribution, etc)of relative equal resistance of 0.691012 m-1
(Nau, 1991). Variation in Rm could not beblamed on differences in hydrophobicityand charge since the filtering layer and thesupport of the two membranes were of thesame nature. Therefore it seems that Rmdepends on morphological characteristicsof the filtering area, and consequently couldmainly be the result of the membrane poresize distribution. Pore size distribution ofMF organic membranes was very largecompared to UF membranes (Dietz et al,1992; Persson et al, 1993).
For the following discussion on pores andtheir influence on performance, the mem-brane pores will be assurned to be cylindri-cal in shape, of the same size and equalthroughout their length (as was consideredby Nilsson and Hallstrôm (1991) for usingRm as a function of pore size distribution).
Membrane resistance and MF performance
The determination of Rm of M14 mem-brane, new, fouled and cleaned after MFexperiment, revealed the Rm variation dueto residual fouling present on. the surfaceand within the membrane after cleaning(Daufin et al, 1992). In the case of cleanedmembranes, the Rm increase is assumednot to be solely due to the membrane's mor-phological characteristics but also, in com-parison to new membrane's Rm, due to thedifferences in hydrophobicity and chargebecause of residual fouling. The cleaningprocedure used was probably inefficient,which is in agreement with the observationsmade by Gésan et al (1993b) of an industrialplant and those of Heinemann et al (1988)and Warren et al (1991) who have observedwater penneability reduction (33%) in hollowfiber MF of solutions of proteins and yeasts.
Microfiltration performance
The influence of initial Rm of new mem-branes was evidenced in a range of experi-mental conditions which corresponded todifferent localized filtration conditions result-ing on different localized fouling layers' build-up. The major features were: fouling het-eroqeneity connected to counter-pressuremodes, static and dynamic as reported byGésan et al (1993a); aggregates size relatedto pretreatment and po rosit y of the foulingdeposit th us created (Gésan et al, 1994).
Evolution of fouling and transmissionversustime
Fouling increased in the course of timeaccording to a 'complete blocking' filtrationlaw accounting for a progressive decrease inthe filtering area (Gésan et al, 1993a).According to previous and recent studies(Gésan et al, 1993a, 1994) the increase offouling versus time could be the outcome ofan increasing deposit layer thickness on themembrane surface. This deposit, composed
275
mainly of calcium and phosphate aggregatescould entrap protein and consequently affecttheir transfer to the permeate.
Effect of Rm'on microfiltrationperformance
During the first hour of MF and at whateverthe state of the membrane (new, irreversiblyfouled, cleaned), higher Rm coincide withhi.gher TP and fouling (fig 1). MF perfor-mance appeared to be governed by Rmwhich characterized the membrane poresize distribution.
The ove rail quantity of mate rial brought tothe membrane by convection was indepen-dent of Rm since the experiments were per-formed at constant flux. For this reason themore pronounced decrease of protein trans-mission with low Rm membrane andincrease in TP vs time (as presented in figs4,5) could be explained by: i) an initial actualfiltering layer characterized by a small num-ber of pores (due to manufacturing) or aninitial residually fouled membrane (badc1eaning); and ii) a large population of smallpores, even i! the global actual filtering areais the same. Using bi-Iiquid permporometrymethod, Persson et al (1993) showed thatlarge pores governed the major part of per-meability. They observed that 10% of thelarger pores were responsible for 90% ofthe permeability in the case of Nylon-66 MFmembrane from Pail with a me an pore dia-meter of 0.2 urn, Similar observations werereported by Fane et al (1981) with AmiconUF membranes. Such a phenomenon wasobserved during filtration performed at con-stant TP. The phenomenon is the same dur-ing constant J experiments since perme-ation occurs preferentially in regions of lowpressure drop, ie area of low resistance.
Consequently, it appears that the betterhydraulic MF performance at short time isobserved with the membranes which wouldhave the larger population of large pores.
276 G Gésan etaI
After 1 h of MF using new membranes,fouling increased faster for low Rm mem-branes (fig 2). Such an evolution could beexplained by the aforementioned assump-tions about the presence of the large popu-lation of large pores in the membrane (com-pared to new membranes of higher Rm).Increase of MF membrane fouling withincreased pore size was attributed to particles(Merin et al, 1983; Gatenholm et al, 1988;Ben Amar and Jalfrin, 1989; Attia etal, 1991)or to proteins (Brink et al, 1993). The largerpores filter more solvent and consequentlya large amount of solutes and particles accu-'mulates at their entrance or passes throughthem. This accelerates pore blocking andleads to a decrease in pore number resultingin a decrease of filtering area. Blocked poresobserved by scanning electron microscopywere reported to be the cause of decreasedpermeability (Kelly et al, 1992).
Consequently, taking into account theoverall evolution of MF performance, a newM 14 membrane with low Rm can be con-sidered as a membrane with a large popu-lation of large pores, representing the majorpart of the actual filtering area.
ln case of an used membrane theincrease of Rm could correspond to a sizereduction and/or disappearance of pores.Due to the reduction of filtering area, thequantity of material which must be filteredper unit area of residual filtering surface ishigher (in an experiment performed at con-stant J) and therefore fouling increasesfaster. These conclusions emphasize theneed for better knowledge of MF membranec1eaning, since MF performance appear tobe related to the characteristics of the usedmembrane.
ln the above discussion, we assumedthat Rm depended mainly on the mem-brane's pore size distribution. Some experi-mental proofs of our assumptions could beobtained using specified methodology (mer-cury porosimetry, etc) to get quantitativeinformation.
Reduction of the heterogeneityof filtration conditions
At the membrane's scale
According to these results obtained with anevenly distributed transmembrane pressurealong the membrane, it appears that theheterogeneity of the pore size of the mem-brane (new or c1eaned) governs the distri-bution of localized flux and in that way, thequality and quantity of global (irreversibleand reversible) fouling. Therefore, in order toreduce the effect of large pores and toimprove whey MF performance, it would bebetter to use a membrane of a more homo-geneous pore size. This optimal mean poresize has to be determined taking intoaccount the size distribution of solute andthe system performance versusthe ratio ofsolute size to pore size (Le and Atkinson,1985; Matsumoto et al, 1988) and operatingparameters. Moreover,adequate proce-dures for assessment of proper cleaninghave to be established since inadequate'washes lead to residuel fouling unevenlydistributed throughout the membrane andto a poor performance as described in thiswork for three membranes.
At the industrial scale
Il should be taken into account that mixingdifferent Rm membranes in a module couldresult in uneven fouling layer distribution(as seen in our experiments) and thus leadto fast fouling and reduced protein trans-mission. The replacement of membranes ina module by new ones has to be done afterRm determination.
Effect of Rm on fouling schematicpresentation
The initial membrane fouling resistance influ-ences fouling kinetics. Consequently the
·- '."
Membrane resistance and MF performance
schematic presentation of fouling with TPor Rf vs time does not take into account theinfluence of Rm on MF performance. Dif-ferent ways to compare filtration experi-ments' performance were proposed. Dejmekand Nilsson (1989) showed for example thatwhen proteins were adsorbed to UF mem-branes Rf/(Rm+Rf) was a better parame-ter compared to Rf alone for the schematicpresentation of results since it reduced theeffect contributed by Rm. Persson and Nils-son (1991) used the normalized resistanceRf/Rm cons ide ring Rf to be directly propor-tional to Rm. Such an assumption was notverified in our experiments except for a shorttime of MF wh en a low Rm induced a lowRf. Rf/Rm did not take into account thedirect effect of Rm on operating durationand selectivity. In fact there is no singleparameter that can schematically describefouling evolution and compare performanceof membranes of different Rm values. More-over, if a membrane is not thoroughlyc1eaned, Rm corresponds to the Rm of theinitial new and c1ean membrane plus theresistance of a residual fouling; the latter isnot taken into account in Rf, which describesthe fouling build-up during the MF. Cense-quently, Rf is an improper parameter toassess performance as could be wronglydeduced from its low value as seen in figure5b.
CONCLUSION
MF performance (permeability and selec-tivity) during microtiltration of pretreatedsweet whey using an M 14 Carbosep mem-brane depended on the initial membranehydraulic resistance (Rm). A new M14 mem-brane of low Rm resulted in a shorter oper-ating time and lower protein transmissionat the end of the operation. An improperlycleaned M14 membrane, with a high Rmresulted in a reduction of MF performance.Considering that Rm depends mainly on
277
morphological characteristics of the tilter-ing area, this reduction of performance couldbe attributed to the preferential flow throughlow resistance area of large pores (for thenew membrane of low Rm) and to the reduc-tion of filtering area by residual fouling (forthe used membrane). Therefore, in order toimprove MF performance a membrane witha sharp pore size distribution and adequatec1eaning procedures should be developed.
Rm is a major characteristic of a mem-brane as it is used at industrial plants inorder to assess its initial state of c1eanli-ness. One of the aims of this work is to drawthe membrane user's attention to the poten-tial consequence of initial Rm values onwhey MF performance even though its con-tribution may be affected according to oper-ating conditions.
ACKNOWLEDGMENTS
This work was partially supported bya grant fromthe Etablissement Public Régional de Bretagnë'(Contract N° 3051 B). -
REFERENCES
AFNOR· ITSV (1986) Contr61ede la qualité des pro-duits laitiers. Recueil des normes françaises. 5.(Amariglio S, ed). AFNOR - ITSV, Paris, 3rd Edn
Allia H, Bennasar M, Tarodo de La Fuente B (1991)Study 01 the fouling of inorganic membranes by acid·ilied milks uslng scanning electron microscopy andelectrophoresis. II. Membrane with pore diameter0.8 urn. J Dairy Res 58, 51·65
Ben Amar R, Jaffrin MY (1989) Microfiltration tangen-tielle d'un vin rouge sur membranes minérales: étudedu colmatage au cours du temps. Entropie 149, 30·40
Brink LES, Elbers SJG, Robbertsen T, Both P (1993)The anti·louling action 01 polymers preadsorbed onultrafiltration and microliltration membranes. J Mem- .brane Sci76, 281·291
Brulé G, Maubois JL, Fauquant J (1974) Etude de lateneur en éléments minéraux des produits obtenuslors de l'ultrafiltration de lait sur membrane. Lait 54,600·615
Chahboun A, Coratger R, Ajustron F, Beauvillain J,Aimar P, sanchez V (1992) First investigations on theuse of scanning tunnelling microscopy (ST) for thecharacterisation of porous membranes. J MembraneSci67,295-3oo
Daufin G, Labbé JP, Quémerais A, Michel F (1991a)Fouling of an inorganic membrane during ultrafiltra-tion of defatted whey protein concentrates. Neth MilkDairy J 45, 259-272
Daufin G, Merin U, Labbé JP, Quémerais A, Kerhervé Fl(1991 b) Cleaning of inorganic membranes alter wheyand milk ultrafiltration. Biotechnol Bioeng 38, 82-89
Daufin G, Merin U, Kernervé FL, Labbé JP, QuémeraisA, Bousser C (1992) Efficiency of cleaning agents foran inorganic membrane after milk ultrafiltration. JDairy Res 59, 29-38
Daufin G, Radenac JF, Gésan G, Kerhervé Fl, Le Berre0, Michel F, Merin U (1993) A novel rig design forultra- and microfiltration experiments. Sep Sci Tech-no128,1635-1642
Davis RH, Birdsell SA (1987) Hydrodynamic model andexperiments for crossflow microfiltration. Chem EngCommun 49, 217-234
De Balmann H, Nobrega R (1989) The deformation ofdex1ran molecules. Causes and consequences inultrafiltration. J Membrane Sci40, 311-327
Devereux N, Hoare M (1986) Membrane separation ofprotein precipitates: studies with crossflow in hollowfibers. Biotechnol Bioeng 28,422-431
Dietz P, Hansma PK, Inacker 0, Lehmann HO, Her-mann KH (1992) Surface pore structures of microand ultrafiltration membranes imaged with the atomicforce microscope. J Membrane Sci65, 101-111
Fane AG, Feil CJD, Waters AG (1981) The relationshipbetween membrane surface pore characteristics andflux for ultrafiltration membranes. J Membrane Sci9,245-262
Fauquant J, Vieco E, Brulé G, Maubois JL (1985) Clarifi-cation des lactosérums doux par agrégation thermo-calcique de la matière grasse résiduelie. Lait 65, 1-20
Gatenholm P, Feil CJD, Fane AG (1988) Influence ofthe membrane structure on the composition of thedeposit layer du ring processing of microbial sus-pensions. Desalination 70, 363-378
Gésan G (1993) Microtiltration tangentielle de lactosérumdoux prétraité. Mécanismes et conduite opératoire.Thèse Univ Rennes l, France
Gésan G, Daufin G, Merin U, Labbé JP, Quémerais A(1993a) Fouling during constant crossflow microfil·tration of pretreated whey. Influence of transrnern-brane pressure gradient. J Membrane Sci80, 131-145
Gésan G, Merin U, Daufin G, Maugas JJ (1993b) Per-formance of an industrial cross-flow microfiltrationplant for clarifying rennet whey. Neth Milk Dairy J47,121-135
Gésan G, Daufin G, Merin U (1994) Microfiltration per-formance : physico - chemical aspects of whey pre-treatement. J Dairy Res, submitted
Heinemann P, Howell JA, Bryan RA (1988) Microfiltrationof protein solutions: effect of fouling on rejection.Desalination 68, 243-250
Jaubert A, Martin P (1992) Reverse phase HPLC anal-ysis of goat caseins: Identification of as1 and as2genetic variants. Lait 72, 235-247
Kelly ST, Opong S, Zydney AL (1992) The influence ofprotein aggregates on the fouling of ultrafiltrationand microfiltration membranes. Proc Eng MembraneProcesses, Garmisch Partenkirchen, 13-15 May,Germany
Kim KJ, Fane AG, Feil CJD, Suzuki T, Dickson MR(1990) Quantitative microscopic study of surfacecharacterization of ultrafiltration membranes. J Mem-brane Sei 54, 89-102
Kim KJ, Dickson MR, Fane AG, Feil CJD (1991) Electronmicroscopy in synthetic polymer membrane research.J Microsc 162, 403-413
Le MS, Atkinson T (1985) Crossflow microfiltration forrecovery of intracellular products. Process Biochem20,26-31
Matsumoto Y, Nakao S, Kimura S (1988) Crossflow fil-tration of solutions of polymers using ceramic micro-filtration. Int Chem Eng 28, 677-683
McDonogh RM, Strathmann H, Buck R, Deppich R, GôhiH (1992) Estimation of membrane sieving and rejec-tion from pore size distribution measured by the Iiq-uid displacement method. In: Membrane Prepara-tion; Fouling - Emerging Processes (Aimar P, AptelP, eds) Vol 6, Lavoisier Technique et Documentation,Paris, France, 65-70
Merin U, Cheryan M (1981) Ultrastructure of the sur-face of a polysulfone ultrafiltration membrane. J ApplPolym Sci 25, 2139-2142
Merin U, Gordin S, Tanny GB (1983) Microfiltration ofsweet cheese whey. N Z J Dairy Sci Techno/18,154-160
Nau F (1991) Production et séparation par ultrafiltrationde peptides trypsiques issus de la caséine f3 bovine.Etude de la cinétique d'hydrolyse et des intéractionspeptides-membrane. Thèse, ENSAR, Rennes, France
Nilsson JL (1988) Fouling of an ultrafiltration membranebya dissolved whey protein concentrate and somewhey proteins. J Membrane Sci36, 147~160
Nilsson J, Hallstrôm B (1991) Deviations in the foulingresistance of UF membranes due to c1ean membranepermeability variations. J Membrane Sci 67, 177-189
Persson KM, Nilsson JL (1991) Fouling resistance rnod-els in MF and UF. Desalination80, 123-138
Persson KM, Capannelli G, Bottino A, Trâgardh G (1993)Porosity and protein adsorption of four polymerie micro-filtration membranes. J Membrane SCi76, 61-71
Membrane resistance and MF performance 279
Plelt E (1989) The constant pressure difference methodfor microfiltration. 3rrJ Int Conf on Foufing and Clean-ing in FoodProcessing(Kessler.HG, Lund DB, Eds.)prien, FRG, 283-293
Strathmann H, Kock K, Amar P (1975) The formationmechanism of asymmetric membranes. Desalina-tion 16, 179-203
Taddéi C, Aimar P, Daufin G, Sanchez V (1986) Etudedu transfert de matière lors de l'ultrafiltration de lac-tosérum doux sur membrane minérale. Lait 66, 371-390
Taddéi C, Daufin G, Aimar P, Sanchez V (1989) Role ofsome whey components on mass transfer in ultra-filtration. Blotecnnol Bioeng 34, 171-179
Warren RK, Mc Donald DG, Hill GA (1991) Crossflowmicrofiltration of Saccharomyces cerevisae. ProcessBiochem 26, 337-342
