EFFECTS OF THE COMPONENTS OF MILK ON FOULING AND
CLEANING OF POLYAMIDE REVERSE OSMOSIS MEMBRANES
Xiao Wei Tew & Ken R Morison *
Department of Chemical and Process Engineering, University of Canterbury, Private Bag
4800, Christchurch, New Zealand
ABSTRACT
Reverse osmosis is used for the concentration of milk to reduce transportation costs from
farms, and also within dairy factories to concentrate dairy liquids. A SEPA flat sheet reverse
osmosis apparatus with a polyamide membrane was used to measure the flux, rejection,
fouling and cleaning of components of milk, both as model solutions and also as whole or
skimmed milk at up to 24 bar transmembrane pressure (TMP).
Both whole milk and skimmed milk were found to have similar fluxes with a maximum at
about 14-16 bar TMP. At higher pressures the flux decreased below the maximum. The flux of
a solution of twice the concentration of SMUF caused significant flux reduction with a
maximum flux at 16 bar TMP. Neither the addition of lactose nor whey protein isolate to
normal SMUF caused additional flux reduction. Skimmed milk with enhanced and depleted
mineral and/or protein contents were also filtered. It was found that altering the mineral
content either up or down increased the fouling resistance for filtration, whereas both
changes in protein content reduced the amount of fouling. The fouling was found to be greater
at higher TMP values. It is thought that high mineral contents led to mineral precipitation but
low mineral contents cause casein micelle disintegration. The membranes were cleaned using
a cycle of 10 minutes each of 0.5% NaOH, 0.8% HNO3, 0.5% NaOH with water rinses before
each step. After 12 bar TMP filtration only one cleaning cycle was required but after 24 bar
filtration, two cycles were required.
INTRODUCTION
Reverse osmosis is used to concentrate raw milk to reduce transportation costs from farms,
and also within dairy factories to concentrate dairy liquids (Grandison and Lewis, 1996). Salts,
proteins, and fat in milk accumulate at the membrane causing increases in the effective
osmotic pressure and/or resistance to flux. Once the solubility limit of solutes is exceeded, a
deposited layer can form on the membrane reducing its permeability and at the same time
inhibiting the back diffusion of salts to the bulk solution (Hoek and Elimelech, 2003). Acids
such as nitric, phosphoric and citric, are generally used to clean inorganic precipitates from
membranes while high pH cleaners, primarily NaOH, are employed for the removal of
biofilms, proteins and other organic fouling (D'Souza and Mawson, 2005).
Protein, in particular casein, is the main component found in the deposit that forms on
membrane surfaces during reverse osmosis of milk solutions (Skudder et al., 1977). For
solutions without casein, calcium-phosphate precipitation appeared to cause strong fouling on
the membrane (Hiddink et al., 1980). The deposit formation for reverse osmosis with skim
milk is affected by the stability of the casein micelle structure (Kulozik, 1998). The deposited
layer behaves like a membrane whose selectivity depends on the solubility and mobility of the
dissolved substances inside it (Kulozik and Kessler, 1990). The fouling resistance appeared to
increase with the soluble calcium and phosphate fractions in skim milk and particularly with
the concentration of free calcium (Ca2+) (Bouzid et al., 2012). However, removal of calcium
from milk by ion exchange led to the disintegration of the micelle structure, causing a very
dense deposited protein layer consisting of casein submicelles and protein molecules. The
flow resistance of the deposited protein layer from diafiltered skim milk was less than ionexchanged skim milk as diafiltration was only able to remove soluble calcium from skim milk
and the casein micelle structure remained unchanged. A more loosely packed deposit structure
was formed during reverse osmosis of diafiltered skim milk as no soluble calcium was
available to form links between casein particles in the deposited layer (Kulozik, 1998).
Membrane flux, J, is defined as the mass flow rate of permeate passing through a unit area of
membrane (kg m-2 s-1). Membrane permeance is defined as the flux divided by transmembrane
pressure (TMP) (kg m-2 bar-1 s-1). The water flux or permeance can be normalized relative to
the water flux or permeance for a pre-treated membrane. Flux decline can be attributed to the
effects of osmotic pressure, irreversible fouling and other reversible phenomena such as
concentration polarization and gel formation (Bouzid et al., 2012). The governing equation
used in this study is:
𝐽=
∆𝑃 − ∆𝜋
𝜇(𝑅𝑚 + 𝑅𝑓𝑙𝑢𝑠ℎ𝑎𝑏𝑙𝑒 + 𝑅𝑛𝑜𝑛−𝑓𝑙𝑢𝑠ℎ𝑎𝑏𝑙𝑒 )
(1)
Here ∆𝑃 is the effective TMP, bar; ∆𝜋 is the osmotic pressure difference between feed and
permeate, bar; 𝜇 is the dynamic viscosity of permeating fluid, Pa.s; 𝑅 is the resistance,
bar m3 s2 kg-2. The subscript m refers to a clean pre-treated membrane, flushable refers to
resistance during milk filtration that can be removed by a water rinse and non-flushable refers
to resistance that remains after a water rinse.
The main objective of this study was to improve the understanding of the mechanisms by
which the components of processed liquids and chemical cleaning agents interact with thinfilm composite polyamide reverse osmosis membranes, so as to be able to reduce membrane
fouling and enhance cleaning effectiveness.
METHODS
Apparatus and operating conditions
A SEPA flat sheet membrane system (Sepa CF, Osmonics, USA), with a filtration area of 134
cm2 and with a feed spacer was used. The system was operated at up to 25 bar TMP using a
Hydra-Cell diaphragm pump (Wanner Engineering, USA) with a pulsation dampener that
reduced the pressure range to ± 10% of the operating pressure. The cross flow velocity was
0.2 m s-1. The feed solution was maintained at 30 °C by immersing the feed container in a
water bath. The dynamics of start-up caused temperature variations of ±4.0 °C but the flux
was corrected inversely proportionally to the viscosity of water at the feed temperature. The
feed temperature, mass of permeate and applied TMP were recorded by a data logging PC.
The DOW Filmtec FT30 polyamide thin-film composite membranes (DOW
HYPERSHELLTM RO-8038/48 element) were cut from a spiral module and kept moist in
sealed bags at about 4 °C. Permeate and retentate were recycled into the feed. Conductivity
and pH were used to determine concentrations for rejection. The rejection coefficient for
minerals was calculated using the electrical conductivity of the bulk feed and permeate.
Methods
All concentrations in this paper are given as percent by mass unless otherwise noted. The
water used in all experiments was supplied from a Millipore water purification system (Elix-5,
Millipore, USA). The membrane was thoroughly rinsed with water and left to soak in water
for 1 hour to ensure complete hydration. The immersed membrane was rinsed with water
again before placement in the rig. The membrane was pre-treated with 0.8% HNO3, water,
0.5% NaOH and water. The solutions were circulated for 10 minutes with no applied TMP.
After each solution the system was rinsed thoroughly with at least 3 L of water, each litre for
5 minutes, until the electrical conductivity dropped below 25 μS cm-1. The water fluxes
before and after pre-treatments were measured and are termed initial water flux and pretreated water flux respectively. The flux after pre-treatment was used as the reference flux.
For fixed TMP runs at 12 or 24 bar, the filtration step was for 3 or 5 hours. For increasing
TMP runs, the TMP was increased stepwise by 2 bar or 4 bar from 8 bar up to 24 bar. The
flux was measured for about 30 minutes or until a steady permeance was achieved depending
on the TMP. The water flux, corresponding to non-flushable fouling, was measured after
system rinsing. The feed solutions used in this study are listed in Table 1.
Table 1 Preparation method for feed solutions.
Feed solution
Skimmed milk
Whole milk
Simulated milk
ultrafiltrate (SMUF)
Concentrated SMUF
SMUF + lactose
SMUF + lactose + WPI
Mineral enhanced
skimmed milk
Mineral depleted
skimmed milk
Protein enhanced
skimmed milk
Protein depleted
skimmed milk
Description
Commercially available Trim milk from Meadow Fresh, Auckland, NZ, 4 g L-1 fat, 37
g L-1 protein.
Commercially available from Klondyke Fresh Limited, Christchurch, NZ, 34 g L-1 fat,
33 g L-1 protein.
Salts were dissolved in the amount and order specified in Table 1 of Jenness and
Koops (1962). The pH was adjusted to 6.6.
Solution was prepared with similar method as described for normal SMUF but with
twice the salt quantity.
A solution 55.36 g L–1 lactose monohydrate (52.59 g L–1 lactose) was used instead of
water to prepare SMUF.
A solution was prepared by dissolving 6.24 g of whey protein isolate (WPI)
(Balance, Vitaco Health Ltd., Auckland, NZ) for 0.6% or 34.32 g of WPI for 3.3% in
1 L of SMUF and lactose solution.
A solution was prepared by dissolving 25% of the amount of SMUF salts into 1 L of
commercial Trim milk.
A solution was prepared by removing 250 mL milk ultrafiltrate from 1 L Trim milk
by ultrafiltration with a PES 10 membrane (Synder Filtration, USA) at 30 °C, TMP of
4 bar and cross flow velocity of 0.2 m s-1. 250 mL of 52.59 g L-1 lactose solution was
added into the ultrafiltered milk to make it up to 1 L.
A solution was prepared by removing 250 mL milk ultrafiltrate from 1 L Trim milk
by ultrafiltration at the same conditions as the previous sample. Protein was enhanced
by 33.3%.
A solution was prepared by diluting the commercial Trim milk with 250 mL SMUF
and lactose solution. Protein was depleted by 20%.
After filtration the membrane was cleaned with a standard cleaning cycle consisting of 0.5%
NaOH, water, 0.8% HNO3, water, 0.5% NaOH and water. The concentrations were based on
previous work in our laboratory, but are not confirmed as being optimal. Each cleaning
chemical was circulated in the system for 10 minutes with no applied TMP to minimize the
production of permeate during cleaning and reduce the convective redeposition of foulant
back onto the membrane surface making it more adhesive (D'Souza and Mawson, 2005). The
cleaning cycle was repeated two to three times with the same concentration and order. The
water flux was measured after each rinse and the permeance ratio, a ratio between water
permeance after each chemical exposure and pre-treatment permeance, was used to indicate
cleaning effectiveness. The sequence of cleaning cycle, concentration of cleaning chemical or
cleaning duration was altered in some studies to investigate the interaction of cleaning
chemical with foulants as well as with thin-film composite polyamide reverse osmosis
membranes.
RESULTS AND DISCUSSION
Figure 1 shows the fluxes for reverse osmosis of milk solutions, both whole and skimmed
milk, as well as its model solutions simulating different components in milk at increasing
TMPs. As revealed by the figure, both whole milk and skimmed milk were found to have
similar behaviour with a maximum flux at 14 – 16 bar. The characteristic curve of milk is
consistent with the work of Rabiller-Baudry et al. (2009) and Bouzid et al. (2012) who
utilized the concept of limiting and critical fluxes to study the effect of feed pH. Similar
behaviour for both whole milk and skimmed milk suggested that fat does not play a
significant role in membrane fouling. The flux of simulated milk ultrafiltrate (SMUF) was
consistent with its osmotic pressure with a slight deviation from the straight line at high TMP,
while concentrated SMUF caused a significant flux reduction with maximum flux at 16 bar
TMP. Neither the addition of lactose or whey protein isolate to normal SMUF caused large
additional flux reduction.
Figure 1. Effect of feed composition on fluxes during runs with increasing TMP.
The permeance as a function of time for selected runs is shown in Figure 2. The raw data,
with oscillation of about ±10% or less, which was caused by the use of an electronic balance
for flow measurement, was smoothed by taking the average of each 10 successive permeance
values. The recorded data points for the first 10 minutes were retained to identify the initial
changes in permeance. After the initial drop, the permeance of whole milk at both TMPs
decreased slightly with time for the first 3 hours before it reached its steady state. The steady
state permeance for 12 bar and 24 bar after 5 hours filtration were 0.203 g s-1 bar-1 m-2 and
0.083 g s-1 bar-1 m-2 respectively. The reduction of permeance with time was higher for runs at
24 bar than at 12 bar.
The permeance values of skimmed milk for constant pressure runs at 24 bar TMP exhibited
behaviour similar to whole milk. Concentrated SMUF had a much higher permeance at the
start of the run as compared to whole milk and skimmed milk. Unlike milk, the permeance
dropped dramatically with time during the first hour of the run indicating the development of
mineral precipitate and concentration polarization.
Permeance (g s-1 bar-1 m-2)
0.8
Whole milk - 12 bar
Whole milk - 24 bar
Trim milk - 24 bar
Concentrated SMUF - 24 bar
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
Time (hours)
Figure 2. Permeance as a function of time for runs with whole milk, skimmed milk and
concentrated SMUF at 12 bar or 24 bar TMP. Data points shown after the first 10
minutes are the average of each 10 successive values.
Membrane and fouling resistance
To compare the amount of fouling caused by the various feed liquids, resistances of the
membrane, flushable fouling and non-flushable fouling were calculated using Equation 1 and
are presented in Table 2. Generally, flushable resistance increased with increasing nonflushable resistance. The resistance corresponding to flushable and non-flushable fouling for
runs with whole milk at 24 bar TMP was 5 – 6 times higher than at 12 bar. The flushable and
non-flushable resistance for SMUF and mixtures of SMUF, lactose and whey protein isolate
were relatively low compared with skimmed milk and whole milk. This indicates that casein,
probably with calcium phosphate, is the major contributor of resistance formation in milk.
Concentrated SMUF increased the flushable resistance by a factor of 10 and the non-flushable
resistance by a factor of 4 as compared to normal SMUF. This was probably because the
higher concentration increased the amount of precipitation, and any precipitation that did
occur increased concentration polarization and hence precipitation. Both enhancement and
reduction of protein concentration in skimmed milk reduced flushable resistance, but it is
notable that both mineral enhancement and depletion increase the flushable resistance.
Table 2 Membrane, flushable and non-flushable resistance.
Calculated osmotic
pressure difference
bar
SMUF
2.97
SMUF and lactose
6.71
SMUF, lactose + 3.3% WPI
6.86
Concentrated SMUF
5.97
Whole milk (12 bar)
7.10
Whole milk (24 bar)
7.10
Skimmed milk (24 bar)
7.10
Mineral enhanced skimmed milk
7.89
Mineral depleted skimmed milk
6.30
Protein enhanced skimmed milk
7.20
Protein depleted skimmed milk
7.04
Note: ± indicates the range of measurement over 6 runs.
Feed solutions
Membrane
resistance
1010 m2 kg-1
11.7
11.0
10.9
10.6
12.2
11.7
10.9 ± 0.4
10.9
10.3
11.7
10.6
Flushable
resistance
1010 m2 kg-1
3.5
5.8
8.2
35
10.8
81
68 ± 6
78
73
66
64
Non-flushable
resistance
1010 m2 kg-1
1.5
1.5
1.9
6.0
2.1
14
13 ± 1
16
12
11
11
Chemical cleaning
Before testing the cleaning of membranes fouled with milk, a virgin membrane was subjected
to pre-treatment, water fluxing for 1 hour and 4 cycles of cleaning. The permeance values of
all water flux tests are shown relative to the pre-treated permeance in Figure 3. Each cycle of
cleaning was found to have an additive effect on membrane permeability. It can be seen that
lower chemical concentrations produced less of an effect. The permeance ratios obtained from
these runs can be used as a control for cleaning efficiency after milk fouling. Similar results
were reported by Madaeni and Mansourpanah (2004) for the cleaning of reverse osmosis
membrane fouled by whey with 0.1% NaOH (pH 12.4) who suggested that alkaline damage
was the cause. Sohrabi et al. (2011) suggested that it was caused by changes in hydrophilicity,
while Nilsson et al. (2008) claimed reversible membrane swelling is the major reason for the
changes in water permeability. In the current work membrane damage was thought to be
unlikely as a run with 15 hours of soaking at each chemical step showed no deterioration in
either permeability or NaCl rejection.
Figure 3 makes it clear that when cleaning results are obtained from a fouled membrane, they
must be compared to results from the same cleaning of a similar membrane but without
fouling. Restoration of the flux back to its original value does not confirm that a membrane is
clean.
Figure 3. Permeance ratios during cleaning of a clean membrane.
The permeance ratios for concentrated SMUF, modified and unmodified milk solutions are
listed in Table 3. Considering the second row for skimmed milk filtered at 24 bar, it can be
seem that the permeance while operating with milk was about 8% of the pre-treated water
permeance. After a water rinse this increased to 45% showing that much of the flux reduction
is due to the temporary build-up of a concentration polarization layer. After the first NaOH
clean the permeance ratio increased to 85% showing that NaOH is not sufficient to remove all
the foulant. After the subsequent HNO3 and NaOH steps of cleaning cycle 1, the permeance
ratio increased to about 97%. This needs to be compared to 103% for cleaning of a clean
membrane (first row) and it shows that cleaning is not complete. Even after another cleaning
cycle the permeance was not restored the levels obtained for other milk solutions. A single
standard cleaning cycle was sufficient to remove the deposit and recover the water permeance
for whole milk run at 12 bar, probably because the cake layer formed at low TMP is loose and
easier to remove. A similar observation was reported by Kulozik and Kessler (1990). For
whole milk, skimmed milk and most modified milk solutions that operated at 24 bar, two or
three cleaning cycles were needed to restore the water permeance back to the level expected
for a clean membrane. The exception was protein depleted skimmed milk, which required
only one cleaning cycle.
Table 3 Effect of feed composition on the permeance ratio and cleaning performance.
Feed solutions
Fouled on
product
0.968
0.081 ± 0.002
0.199
0.077
Permeance ratio = Permeance/pre-treated permeance
After first
After cleaning After cleaning
Rinsed
NaOH
cycle 1
cycle 2
0.99
1.03
1.05
0.45 ± 0.005
0.97 ± 0.01
0.99 ± 0.01
0.85± 0.01
0.86
1.04
0.45
0.97
1.05
Water
Skimmed milk
Whole milk (12 bar)
Whole milk (24 bar)
Mineral enhanced
0.070
0.40
skimmed milk
Mineral depleted skimmed
0.080
0.46
milk
Protein enhanced
0.093
0.52
skimmed milk
Protein depleted skimmed
0.088
0.50
milk
Concentrated SMUF
0.155
0.64
Note: ± indicates the range of measurement over 2 runs
0.83
0.96
1.02
0.91
0.96
1.00
0.90
0.97
1.01
0.90
1.00
1.05
0.83
0.99
1.02
OVERALL DISCUSSION
There are a number of different mechanisms contributing to the results and future work will
seek to differentiate them. It is clear that operation at 24 bar TMP, rather than 12 bar, not
only reduces the operating flux but considerably increases the difficulty of all stages of
cleaning. It is unclear why the flux at 24 bar dropped more than that at 12 bar. Normally
higher pressures give higher fluxes which increase the deposit formation rate and the pressure
gradients. However in this case the flux at 24 bar was lower than at 12 bar, so the deposit
build-up should have been slower. Several mechanisms could be involved:
1. The initial flux at higher pressure is higher than at lower pressure causing a short term
increase of minerals and casein in the cake layer.
2. At higher TMP more minerals are forced into the membrane thus increasing its
resistance.
3. At higher TMP the pressure fluctuations are slightly higher causing cake build-up.
4. At higher concentrations of casein and mineral in the cake, higher static pressures
cause greater cross-linking.
The role of minerals in fouling is shown by the rapid fouling caused by concentrated SMUF.
There is little doubt that the concentration of minerals at the membrane increases to such a
level that precipitation occurs. The precipitate is likely to form a cake which will reduce the
back diffusion of minerals into the cross flow and thus fouling will be accelerated. The same
mechanisms are feasible when minerals in milk are concentrated within a casein cake.
The results point to the need for independent measurements to test the feasibility of different
mechanisms. In the next stage, surface analysis techniques will be used to determine the
composition and position of deposits at various stages of fouling and cleaning. As examples,
it is not clear why the first NaOH solution is equally effective in removing the resistance from
concentrated SMUF and the resistance from mineral enhanced skimmed milk (Table 3).
CONCLUSIONS
Flux reduction of reverse osmosis is caused by several different mechanisms. Casein micelles
form a cake appears to increase the hydraulic resistance. This cake also contributes to
concentration polarization which increases osmotic pressure and reduces flux. The
polarization, with or without casein micelles, can cause minerals, such as calcium phosphate,
to precipitate further enhancing the cake and adding hydraulic resistance by coating the
membrane. Higher TMP caused significantly greater fouling.
The effect of cleaning chemicals on a clean membrane needs to be measured before cleaning
of fouled membranes can be evaluated. Both acid and caustic cause changes in permeability
of clean membranes. At least two cycles of acid/caustic/acid cleaning were required to
achieve a steady flux. The foulant produced at 24 bar TMP was much more difficult to clean
than that at 12 bar.
REFERENCES
Bouzid, H., Rabiller-Baudry, M., Derriche, Z. and Bettahar, N. (2012) Fluxes in reverse
osmosis of model acidic and alkaline transient effluents issued from skim milk filtration.
Desalination and Water Treatment 43(1-3), 52-62.
D'Souza, N.M. and Mawson, A.J. (2005) Membrane cleaning in the dairy industry: A review.
Critical Reviews in Food Science and Nutrition 45(2), 125-134.
Grandison, A.S. and Lewis, M.J. (1996) Separation Processes in the Food and Biotechnology
Industries: Principles and Applications. Woodhead Publishing Ltd, England
Hiddink, J., de Boer, R. and Nooy, P.F.C. (1980) Reverse osmosis of dairy liquids. Journal of
Dairy Science 63(2), 204-214.
Hoek, E.M.V. and Elimelech, M. (2003) Cake-enhanced concentration polarization: A new
fouling mechanism for salt-rejecting membranes. Environmental Science & Technology
37(24), 5581-5588.
Jenness, R. and Koops, J. (1962) Preparation and properties of a salt solution which simulates
milk ultrafiltrate. The Netherlands Milk and Dairy Journal 16(3), 153-164.
Kulozik, U. (1998) Variation of the calcium content in skim milk by diafiltration and ion
exchange – Effects on permeation rate and structure of deposited layers in the RO.
Journal of Membrane Science 145(1), 91-97.
Kulozik, U. and Kessler, H.G. (1990) Effects of salt ions and deposit formation on the
permeation of organic molecules in complex media in reverse osmosis. Journal of
Membrane Science 54(3), 339-354.
Madaeni, S.S. and Mansourpanah, Y. (2004) Chemical cleaning of reverse osmosis
membranes fouled by whey. Desalination 161(1), 13-24.
Nilsson, M., Trägårdh, G. and Östergren, K. (2008) Influence of temperature and cleaning on
aromatic and semi-aromatic polyamide thin-film composite NF and RO membranes.
Separation and Purification Technology 62(3), 717-726.
Rabiller-Baudry, M., Bouzid, H., Chaufer, B., Paugam, L., Delaunay, D., Mekmene, O.,
Ahmad, S. and Gaucheron, F. (2009) On the origin of flux dependece in pH-modified
skim milk filtration. Dairy Science Technology 89, 363-385.
Skudder, P.J., Glover, F.A. and Green, M.L. (1977) An examination of the factors affecting
the reverse osmosis of milk with special reference to deposit formation. J. Dairy Res.
44(02), 293-307.
Sohrabi, M.R., Madaeni, S.S., Khosravi, M. and Ghaedi, A.M. (2011) Chemical cleaning of
reverse osmosis and nanofiltration membranes fouled by licorice aqueous solutions.
Desalination 267(1), 93-100.