Beruflich Dokumente
Kultur Dokumente
net/publication/223573542
CITATIONS READS
36 303
2 authors:
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Jaleh Mansouri on 10 July 2018.
Abstract
This paper describes the development of modi®ed hydrophobic membranes for osmotic distillation (OD) which are tolerant
to oily feeds. Three commercial membranes were chosen as substrates including the Celgard 2500, Millipore GVSP and the
UPVP. The focus has been on using PVA coatings which were found to have an insigni®cant effect on ¯ux.
For concentrated sucrose solutions, ¯ux was reduced signi®cantly by viscosity-related concentration polarisation effects.
The presence of the coating did not increase this effect, suggesting that the ¯ux limitation mostly occurred adjacent to the
feed-side of the laminate and not within the top-layer or inside the pores.
The laminate membrane is a necessity for oily feeds since the uncoated membranes were promptly wetted out even for low
concentration of oil (limonene) dispersion in water. The laminate membranes were all stable in oil emulsions when tested for
periods up to 24 h; i.e. the membranes did not wet out during ¯ux measurement and no visual damage nor coating detachment
was observed for laminates. These observations con®rm the ef®cacy of having coated membranes for OD of oily feeds.
# 1999 Published by Elsevier Science B.V. All rights reserved.
0376-7388/99/$ ± see front matter # 1999 Published by Elsevier Science B.V. All rights reserved.
PII: S0376-7388(98)00252-X
104 J. Mansouri, A.G. Fane / Journal of Membrane Science 153 (1999) 103±120
In this study the performance of the laminate mem- 2.2. Laminate membrane formation
branes was compared with the uncoated substrates.
2.2.1. PVA coating
Table 2
Substrate membrane specifications
was 53.650.21 and the EWC for the X-linked 2.2.2. PolyHEMA coating
laminate was 41.250.19.
2.2.2.1. Celgard membrane. PolyHEMA was exam-
2.2.1.4. One-step method. In the one-step approach a ined as an alternative material for top-layer coating
mixture of GA, catalyst (sulfuric acid, hydrochloric of membranes. The main problem with coating of
acid) and quencher (methanol) was added to the PVA Celgard with PHEMA was the penetration of an
solution and the hydrogel was cast onto the substrate ethanol solution of this polymer into the membrane
membrane. The amount of glutaraldehyde was varied pores. In an attempt to overcome this problem water
so that membranes could be prepared with initial (as non-solvent) was added to the ethanol polymer
(nominal) crosslinking ratio, Xr, from 0.0025 to solution. A 10 wt% polymer solution in ethanol/water
0.01 moles of glutaraldehyde per mole of PVA (90:10 wt%) was prepared. This solution still penetrated
repeating unit. These solutions were then cast onto theCelgardmembrane.Thisalsohappenedwhen20wt%
the Celgard membrane and kept at room temperature polymer solution was used for coating. In-creasing the
(258C) to dry (20 h). amount of water led to a more viscous solution which
These hydrogels showed the tendency for gel for- could not penetrate the Celgard, but formed a dispersion
mation after a few days storage at ambient conditions. that was not suitable for film forming.
In order to overcome the gel formation a small amount As an alternative a membrane (attached to a glass
of methanol as quencher and also acetic acid as buffer plate) was pre-wetted by soaking in EtOH and a thick
were added to the solution. The addition of methanol solution of PHEMA (20 wt%) was cast on this mem-
to the recipe postponed the gel forming to some extent brane. The coating was dried with a hot-air blower.
but it could not be eliminated. Thus for preparation of However, the coating did not look homogenous, prob-
the PVA/GA hydrogel the two-step approach was ably due to partial wetting of the substrate. Taking
considered more feasible. Figs. 2 and 3 summarise these dif®culties into consideration, the PHEMA coat-
optimum experimental protocols of laminate prepara- ing was excluded as an alternative coating for Celgard
tion for both cases. membranes.
Fig. 2. Schematic diagram for preparation of laminate membranes of Celgard and PVA/MA hydrogel.
J. Mansouri, A.G. Fane / Journal of Membrane Science 153 (1999) 103±120 107
Fig. 3. Schematic diagram for preparation of laminate membranes of Celgard and PVA/GA hydrogel.
3. Results and discussion water (Wf, unbound water in the polymer having
transition temperature and enthalpy similar to those
3.1. Coating layer properties of pure water), freezing bound water (Wfb , intermedi-
ate water having a phase transition temperature lower
3.1.1. Determination of EWC and water structure than 08C) and nonfreezing water or bound water (Wb,
The water absorbed by a hydrogel is quantitatively water with no phase transition). For PVA/GA001
expressed by the equilibrium water content (EWC). which has the lowest amount of crosslinker only free
This is the ratio of the weight of water in the hydrogel and nonfreezing waters were detected (Table 4).
to the weight of the hydrogel at equilibrium hydration, It can be seen that increasing the crosslinker ratio
expressed as a percentage. EWC is the most important reduces the water content, although the effect is not
property of a hydrogel as this can in¯uence other very signi®cant. Table 4 also shows the amount of free
properties such as permeability, selectivity and and nonfreezing water per gram of the hydrogel. As
mechanical properties. The water present in a hydro- can be seen free water content drops from 0.33 to 0.20
gel exists in a continuum of states between two with increasing the crosslink ratio from 0.0025 to 0.01.
extremes; water strongly bonded to the polymer net-
work through hydrogen bonding, bound water (Wb), 3.1.2. Determination of the laminate structure
and water with a much greater degree of mobility, free The cross-section and surface morphology of the
water (Wf). The water structure of the hydrogels was laminate membranes were observed with FESEM
estimated by differential scanning calorimetry (DSC). (®eld emission electron microscopy) (Figs. 5 and
EWC studies showed that PVA coatings are highly 6). Fig. 5 shows the SEM of GVSP membrane which
hydrated gels. Tables 3 and 4 show typical results for is deliberately half-coated with PVA. As can be seen
two different series of laminate membranes prepared the top-layer hydrogel is homogenous and free of
with maleic acid and glutaraldehyde as crosslinkers. artefacts. Fig. 6 shows the cross-section for the
DSC heating curves for water-swollen samples of GVSP/PVA laminate membrane. No penetration of
PVA/GA002 and PVA/GA003 revealed the presence coating layer into the pores of the membrane was
of three different states of water namely free freezing observed.
Table 3
EWC for PVA hydrogels crosslinked with maleic acid
Table 4
EWC and water structure for PVA hydrogel crosslinked with glutaraldehyde
Table 5
T-peel strength of GVSP/PVA composite
GVSP/PVA(1) 0.21350.0455
GVSP/PVA(2) 0.13020.0950
GVSP/PVA(3)a 0.04810.0042
a
Not pretreated by acid etch.
line solution. A piece of laminate membrane was stuck 10.2 cm2. The effective volume of each chamber was
on the bottom of a beaker and the beaker (covered with 74 ml.
a lid) was kept in a 508C water bath. A droplet of warm In order to ensure uniformity of temperatures and
(508C) caustic soda (0.5 wt% NaOH, pH: 14) was then concentrations in each chamber, the solutions were
put on an area of about 1 cm2 of the coated membrane stirred. The stirrer in each chamber was composed of a
for periods of 5±30 min. The exposed spots were Te¯on-coated magnetic bar and a blade which was
marked and the membrane was kept in an oven mounted on a polypropylene shaft and held in the
(508C) until it was dry. Then the exposed areas were chamber by means of a plastic bracket to the Perspex
brushed with oil. The absence of transparency of the wall. The stirring speed was adjusted to 475 rpm using
membrane at those spots were taken as an indication of a stroboscope. Experiments were carried out in an air-
the chemical stability of the coating. Various Celgard/ conditioned room (258C) and all solutions were stored
PVA laminate membranes were tested for periods of 5 in a water bath maintained at 258C prior to use. The
and 30 min. No wetting out was observed for these cell was also kept in a water bath maintained at
laminate membranes. 2518C.
The feed chamber was ®lled with water, sucrose
3.2. Performance of membranes solutions and oil-containing feeds and the brine cham-
ber with calcium chloride solution. The brine chamber
3.2.1. Permeation rate of laminate membranes was connected to a pipette and the feed chamber to a
feed reservoir. Both pipette and reservoir were posi-
3.2.1.1. Experimental setup. The water permeability tioned at the same level in order to ensure the absence
of membranes was measured using a two of hydrostatic pressure difference across the mem-
compartment cell. The experimental setup was brane. The level of the brine in the pipette and the mass
similar to that used by Godino et al. [21] and is of feed in the reservoir were recorded every 5 or
illustrated in Fig. 7. The central part of the 10 min and ¯ux was calculated from the slope of
experimental device is a cell, which consists of two the weight decrease (or brine volume increase) over
symmetric cylindrical semicells of 7.6 cm length and the steady region versus time. Flux was normalised to
3.6 cm in diameter separated by the membrane. The water vapour pressure difference between feed and
membrane surface area exposed to the solution was brine.
3.2.1.2. Feed solutions. Pure water, different is the highest and the Celgard is the lowest (Table 6).
concentrations of sucrose solutions (40, 55, 60, 65, These results which identify the obtainable ¯ux
and 68 wt%), concentrated feed (Gordo grape juice, for a given driving force for different membranes
688 Brix i.e. ca. 68 wt% sucrose) and different oil are predictable according to the structure of these
(limonene) dispersions (0.2, 0.5 and 1 wt% in water membranes. Celgard has the lowest porosity and
and in sucrose solution) were used as feed. Sucrose pore diameter and the UPVP membrane has the
solutions were chosen for experimentation because same porosity and pore diameter as GVSP with less
they behave similarly to fruit juices. The physical thickness (see Table 2). For laminate membranes
properties of sucrose (vapour pressure, viscosity and when ¯ux is normalised to vapour pressure differ-
diffusivity) vary significantly with concentration. ence, the Celgard has a comparable ¯ux to the
Variation of viscosity at 258C is shown in Fig. 10 GVSP laminates (within experimental error the
[22]. laminates have similar ¯uxes to the substrates) but
Vapour pressure is expressed by the following signi®cantly less than the UPVP laminate membrane
equation [23]: (Table 6).
log
Pv =Pvsat log
Xw ÿ 2:6X 2 ; (1)
3.2.3. Discrepancy between feed and brine flux
where X is the mole fraction of sucrose in the feed, Xw There is a trend that the ¯ux from the feed side is
the mole fraction of water, Pv the vapour pressure and higher than that from the brine side. This difference is
Pvsat is the saturated vapour pressure. generally higher for concentrated feeds (Fig. 8). Esti-
mates of the volume reduction on mixing could not
3.2.1.3. Strip solutions. The main criteria for choice explain the difference. Errors due to entrapped bub-
of brine are that it does not wet the membrane, it is bles were suspected but no gas bubbles were observed
non-volatile, the vapour pressure is below that of the and the use of deaerated water had no effect. Other
feed and it is relatively cheap. Considering these attempts were made to reduce or eliminate this dif-
requirements calcium chloride (24.5 and 36 wt% ference by more precise control of the temperature of
CaCl2) was chosen as the strip solution in this solutions prior to and during the experiments. The fact
work. The physical properties (such as viscosity) of that the difference is more in the case of concentrated
this salt are much less sensitive to concentration than feeds suggests that this discrepancy could be related to
those of sucrose solutions. non-uniformity in temperature or concentration due to
inadequate mixing which is unavoidable in the present
3.2.2. Normalised flux for pure water cell. Comparisons between membranes in this paper
Results for pure water and 24.5 wt% brine are based on brine-side ¯ux data; the feed and brine
(Pv649 Pa) for the uncoated membranes showed ¯ux discrepancies do not effect the overall conclu-
that the normalised ¯ux for the UPVP membrane sions.
Table 6
Normalised flux for different substrates
Membrane Feed (wt%)/ Flux (kg mÿ2 hÿ1) measured from Normalised flux (kg mÿ2 hÿ1 kPaÿ1)
brine (wt%) measured from
Fig. 8. Ratio of feed flux to brine flux for different feeds (water, sucrose solutions and oil emulsions (ones designated by %)) using Celgard
and Celgard laminate membranes.
3.2.4. Concentrated feed 2. The presence of the coating did not enhance this
Different concentrations of sucrose solution (55± polarisation effect, suggesting that the polarisation
68 wt%) and one concentrated feed (Gordo grape has occurred adjacent to the feed-side of the lami-
juice, 688 Brix, i.e. ca. 68 wt% sucrose) were used nate and not within the top-layer or inside the
as feed. In these experiments the brine concentration pores.
was 36 wt%. 3. Celgard data show a wider spread of normalised
1. It was found that by increasing the sucrose content fluxes compared to the UPVP membranes, parti-
of the feed, normalised ¯ux has been reduced signi- cularly at 0 wt% sucrose (Fig. 9). Similar to results
®cantly (Fig. 9). For UPVP substrate this reduc- obtained for UPVP membranes, it was found that
tion is about 20% for 55 wt% sucrose solution. by increasing the sucrose content of the feed, flux
When the feed is 65 wt% the sucrose solution ¯ux has been reduced significantly for Celgard mem-
reduced by up to 90% for uncoated substrate and branes. This reduction is about 20% for 55 wt%
by up to 60% for the laminate membrane. These sucrose solution and up to 90% for 68 wt% sucrose
observations indicate that there is solute polarisa- solution. The reduction in flux for 60 and 65 wt%
tion at these sucrose concentrations, which may be sucrose solutions is marginally less than those for
linked to viscosity effects (see below). UPVP. Celgard has a smoother surface morphology
and probably the risk of local polarisation at these
concentrations is less for this membrane compared
to UPVP.
4. As can be seen at concentrations below 40 wt% the
normalised flux varies almost proportionally to the
membrane permeability, indicating that the process
is controlled by the membrane. The flux for the
more permeable membrane (UPVP) is therefore
higher. At higher concentration (55±68 wt%) the
characteristics of the feed-side dominate, and there
is no advantage in using a more permeable mem-
brane without improving the hydrodynamic con-
ditions.
The feed concentration of around 40 wt% is
representative of the beginning of the transition
Fig. 9. Normalised flux (brine side) for different Celgard laminate from a membrane limited process to a feed-limited
membranes and UPVP membranes. process. This onset concentration is expected to be
J. Mansouri, A.G. Fane / Journal of Membrane Science 153 (1999) 103±120 113
Fig. 10. Normalised flux (brine side) and bulk viscosity of sucrose (at 258C) versus sugar concentration in the bulk feed.
improved by modification of experimental condi- and viscosity of sucrose [22] against the bulk sucrose
tions (e.g. temperature, hydrodynamic conditions concentration in the feed.
of flow, etc.). Fig. 11 demonstrates that there is a semi-logarith-
5. Celgard data over the range of concentrations mic correlation between normalised ¯ux of Celgard
suggest that the coating has only a small adverse membranes and bulk feed concentration for feed
effect, if any, on the flux. For high concentrations concentrations above 40 wt%. This is similar to rela-
of sucrose two of the laminates (Celgard/PVA/ tionships based on the ®lm model for ultra®ltration.
MA(3) and Celgard/PVAG(1)) show slightly An empirical relationship between ¯ux and bulk
higher flux and lower flux decline as can be seen viscosity is evident from Fig. 12, which can be
in Figs. 9 and 10. expressed in the form
6. The fact that the coatings have very little (if any)
Flux a ln
fb b: (2)
effect on flux shows that ``highly hydrated'' hydro-
gel coatings offer very little resistance to water The in¯uence of viscosity will be greater than that
permeation relative to that offered by the micro- implied in Eq. (2) due to concentration and viscosity
porous substrate in this OD process. gradients in the boundary layer. Fig. 13 depicts the
7. No sugar crystals were observed at the surface of boundary layer pro®les anticipated in OD. It is pos-
the feed side of the membranes confirming that sible to estimate Cwf and hence wf from our data. In
there was no excessive local polarisation and the OD process the strip-side mass transfer resistance
saturation at the membrane surface. is very low compared to the feed and membrane
Tables 7 and 8 list the normalised ¯ux data for (usually less than 1%) and is not the limiting resis-
Celgard and Celgard laminate membranes. tance. In order to make an estimation of the concen-
tration at the wall the simple membrane distillation
3.2.5. Viscous polarisation equation for the case of deaerated feed and assuming
Observation of the normalised ¯uxes for uncoated Knudsen ¯ow was used [24]. Since the pores in the
and coated membranes has shown that in general Celgard membrane (0.0190.05 mm) are smaller
¯uxes decline as concentration of sucrose increases. than the free path of the water vapour (0.18 mm)
This dramatic decline in ¯uxes occurs at about the it is reasonable to assume that the dominant resistance
same concentration (higher than 40 wt%) where the is the polymer network and thus the dominant ¯ow is
viscosity of sucrose increases rapidly, as can be seen in Knudsen. The following is the semi-empirical equa-
Fig. 10 which shows the variation of ¯ux for Celgard tion, based on the above assumptions:
114 J. Mansouri, A.G. Fane / Journal of Membrane Science 153 (1999) 103±120
Table 7
Normalised water flux for Celgard membranes with different feed concentrations (maleic acid as crosslinker)
Sample Membrane Feed (wt%)/ P (Pa) J Flux (kg mÿ2 hÿ1) Jn Normalised flux (kg mÿ2
brine (wt%) measured from hÿ1 kPaÿ1) measured from
Table 8
Normalised water flux for Celgard/PVA/GA membranes with different feed concentrations
Sample Membrane Feed (wt%)/ P (Pa) J Flux (kg mÿ2 hÿ1) Jn Normalised flux (kg mÿ2
brine (wt%) measured from h kPaÿ1) measured from
Fig. 11. The normalised flux for sucrose solution as a function of bulk feed concentration for Celgard membranes.
Fig. 12. Normalised flux as a function of bulk viscosity of sucrose at 258C. Line shown based on Eq. (2) with aÿ0.272, b1.548.
Table 9
Concentration and viscous polarisation for Celgard and Celgard/
PVA/MA(3) laminate
Table 10
Normalised flux for Celgard and Celgard laminate membranes for oil containing feeds
Sample Membrane Feed (wt%)/ P (Pa) J Flux (kg mÿ2 hÿ1) Jn, Normalised flux (kg mÿ2
brine (wt%) measured from hÿ1 kPa) measured from
polarisation has been reduced. However, this reduc- An alternative interpretation of the effect of high
tion did not result in a more rapid mass transfer as the concentration on normalised ¯ux is that the raised
normalised ¯ux is lower for 68 wt% feed. viscosity leads to ``viscous ®ngering'' across the
A similar effect has been observed for ultra®ltration membrane surface [27]. This phenomena, typically
of high viscosity solutions [26]. In this case ¯ux found in two-phase ¯ow through porous media, can
showed a minimum with concentration for laminar lead to regions of stagnation near the membrane sur-
¯ow conditions. The effect was ascribed to the face. Further evidence to support this interpretation is
increase in the shear stress in the boundary layer that OD of concentrated feed in a conventional hollow
leading to improved mass transfer. This phenomena ®bre module experienced a dramatic drop in normal-
may explain our observed ¯ux minima with viscous ised ¯ux (75% drop as feed increased from 60 to
polarisation. 70 wt%), which implies serious maldistribution due to
118 J. Mansouri, A.G. Fane / Journal of Membrane Science 153 (1999) 103±120
``viscous ®ngering''. Use of a helically-wound module 5. For oil dispersions of 0.2, 0.5 and 1 wt% the
with improved ¯ow distribution overcomes this pro- experiments were continued for 24 h while the
blem [28]. Similarly the Hoechst Celanese Liqui feed in the cell and in the container was stirred
CelTM module, which has transverse ¯ow across hol- in order to maintain the dispersion concentration of
low ®bres, is also much less sensitive to feed con- oil as uniform as possible. The laminate mem-
centrate when used for OD [13]. branes were visually examined for the presence
of any damage or transparency at specific intervals
3.2.6. Flux for oil containing feeds of time. These long-term experiments were carried
One of the aims of this work is to improve the out for three of the membranes prepared at opti-
operation of OD membranes with oily feeds. Limo- mum conditions. Visual examination of mem-
nene was chosen for the model solutions since this is branes after the experiments revealed that the
one of the major oil components of orange juice. laminates remained intact with no visual damage
Table 10 compares the ¯uxes for oily feeds to pure or any detachment from the substrate. The water
water and sucrose solutions for various coated and flux which was measured after this period of time
uncoated membranes. 0.2, 0.5 and 1 wt% limonene for membranes remained unchanged, confirming
emulsions were used as candidates for oily feed; the the performance stability of laminate membranes.
concentrations chosen are typical of industrial feeds. 6. Flux was measured for sucrose solution and oil in
The ¯ux for one concentration of oil in sucrose solu- sucrose solutions (similar to fruit juice) for some
tion (0.5 wt% in 55 wt% sucrose solution) was also laminates. There was only a slight reduction in flux
measured. after the addition of oil to the sucrose solution. For
The results can be summarised as follows: Samples 2 and 3 this reduction is 9 and 14 wt%,
1. Oily feed could only be used with laminate respectively.
membranes as the uncoated membrane was
promptly wetted out even for 0.2 wt% oil disper-
sion in water. This con®rms the importance of 4. Conclusions
having coated membranes for OD of oily feeds.
2. Laminate membranes were all stable against oil In this work the validity of PVA hydrogel as a top-
emulsions; i.e. the membranes did not wet out layer coating for hydrophobic membranes used in OD
during flux measurement (no transparent spots to improve their resistance against oily feed has been
on the surface of membranes were observed). No proved.
visual damage nor coating detachment was The tolerance of laminates was evaluated against oil
observed for laminates. These observations con- containing feeds in short and long term experiments.
firm the validity of hydrogel coating as an approach For concentrations up to 1 wt% oil laminate mem-
to eliminate wetting out of the Celgard membrane branes remained unchanged when the experiment
for oily feeds without any significant effect on continued for up to 24 h. In contrast, membranes
permeability. without the hydrogel coating were wetted out very
3. Fluxes for laminates when oil emulsions in water rapidly by the oily feed.
were used as feed were not significantly different Due to their high water content and the small
(within experimental error) from ones when feed thickness of the hydrogel, the water permeabilities
was pure water. This suggests that the coatings are for laminate membranes in most cases were not sig-
basically free of defects. ni®cantly lower than the ones for uncoated substrate.
4. For selected laminates the water flux measurement This means that laminates with comparable perme-
was repeated after running the tests for oil emul- ability and superior resistance to organic solution have
sions. These fluxes were not significantly lower been fabricated.
than the initial flux (Samples 3 and 4; Table 10). We have also shown that in OD concentration and
This suggests that membrane exposure to oil dis- viscous polarisation is signi®cant for feed concentra-
persion had no adverse effect on the coating layer tions in excess of 40 wt%. This is accompanied by a
properties. dramatic decrease in ¯ux and normalised ¯ux. The
J. Mansouri, A.G. Fane / Journal of Membrane Science 153 (1999) 103±120 119
distillation through a porous hydrophobic membrane, Sep. [25] R.W. Schofield, A.G. Fane, C.J.D. Fell, Gas and vapour
Sci. Technol. 30(6) (1995) 993. transport through microporous membranes, J. Membr. Sci. 53
[22] R.C. Weast, Handbook of Chemistry and Physics, 65th ed., (1990) 159.
CRC press, Cleveland, OH, 1984±1985. [26] M. Pritchard, R. Field, J.A. Howell, The ultrafiltration of
[23] P.X. Hoynak, G.N. Bollenback, This is Liquid Sugar. Refined viscous fluids, J. Membr. Sci. 102 (1995) 223.
Syrups and Sugars, 2nd ed., 1966, New York. [27] A.S. Michaels, Personal communication.
[24] V. Calabro, B.L. Jiao, E. Drioli, Theoretical and experimental [28] P. Hogan, Thermal and isothermal membrane distillation,
study on membrane distillation in the concentration of orange Ph.D. Thesis, University of New South Wales, 1996.
juice, Ind. Chem. Res. 33 (1994) 1803.