Professional Documents
Culture Documents
November 2006
© 2006 TECHNEAU
TECHNEAU is an Integrated Project Funded by the European Commission under the Sixth Framework
Programme, Sustainable Development, Global Change and Ecosystems Thematic Priority Area
(contractnumber 018320). All rights reserved. No part of this book may be reproduced, stored in a database
or retrieval system, or published, in any form or in any way, electronically, mechanically, by print,
photoprint, microfilm or any other means without prior written permission from the publisher
Colophon
Title
Ultrafiltration with pre-coagulation in drinking
water production – Literature review
Author(s)
Philipp Meier, Farhad Salehi, Christian Kazner,
Thomas Wintgens, Thomas Melin
Quality Assurance
By Ingo Machenbach
Deliverable number
D 5.3.4
Contents 1
1 Introduction 4
1.1 Requirements for water treatment 4
1.1.1 Surface water 4
1.1.2 Groundwater 5
1.2 Ultrafiltration in general 6
1.2.1 Process description 6
1.2.2 Operating mode and module design 7
1.2.3 Fouling and chemical cleaning 11
1.3 Use of ultrafiltration in drinking water production 11
1.4 Role of coagulation in combination with ultrafiltration in water treatment 13
2 Pre-treatment options 15
2.1 Coagulation 15
2.1.1 Aluminium-based coagulants coupled with UF 15
2.1.2 Coagulation with PACL coupled with UF 17
2.1.3 Comparison of coagulation with iron chloride and aluminium sulphate coupled with
UF 18
2.1.4 In-line coagulation 20
2.1.5 Comparison of in-line and standard coagulation coupled with UF considering the
effect on the removal of DOMs 21
2.1.6 Coagulation coupled with UF with an aerated submerged membrane 24
2.1.7 Coagulation coupled with filtration with metal membranes 26
2.1.8 Summary 27
2.2 Alternative pre-treatment options 27
2.2.1 Adsorption with IOP or PAC coupled with UF and pre-chlorination 27
2.2.2 Pre-chlorination coupled with UF 29
4 Fouling control 44
4.1 NOM fouling 44
4.2 Biofouling of UF membranes 53
4.3 Heterogeneous fouling in dead-end UF 55
6 Case studies 64
6.1 Roetgen (Germany 64
6.2 Hermeskeil (Germany) 67
6.3 Keldgate (UK) 68
6.4 Switzerland 68
6.5 Minneapolis (USA) 69
6.6 Choa Chu Kang (Singapore) 71
6.7 Iverness (UK) 72
7 Conclusions 75
8 References 77
One of the major problems of using surface water as source for drinking
water is the high content of natural organic matter (NOM). NOM can cause
odour, can influence the taste and can increase corrosion and biofilm growth
in the distribution network. Furthermore NOM could be a source for the
formation of disinfections by-products when water is disinfected. Hence,
NOM removal is one of the most important treatment requirements for the
production of drinking water [10]. Fig. 1 shows constituents of dissolved
organic mattes (DOC) in typical surface water. Within the DOC fraction the
humics play an essential role. Other DOC fractions are the higher molecular
polysaccharides and low molecular weight neutrals.
Humics
879 ppb
Low-Molecular-Weight-Neutrals
Polysaccharides 278 ppb
104 ppb
• Removal of particles
• Removal of dissolved matter
• Hardness increase and neutralising
• Disinfection
• Removal of micropollutants
The most common method to remove particulate matter and a portion of the
dissolved NOM from surface waters is by sedimentation and/or filtration
following the conditioning of the water by coagulation [2].
1.1.2 Groundwater
• Aeration
• Neutralising
• Iron removal
• Manganese removal
• Nitrification
Depending on the raw water quality disinfection and organic matter removal
might also be required [1]. In Germany groundwater can be inducted partly
into the drinking water distribution network without treatment [3]. In
catchment areas with high porous grounds, it could come temporary to
microbiological immersions into the groundwater. Additionally in such cases
disinfection can be also required. In the case of spring waters periodical
immissions of coliform bacteria into the water might occur. Furthermore
intensive rainfall and snowmelt can cause higher turbidity in the water [4].
200
100 Reverse
osmosis
[bar]
Nano-
10 filtration
Pressure Δp
Ultra-
filtration
1
Microfiltration
Filtration
0,1
0,0001 0,001 0,01 0,1 1 10 100
vP ρm
permeate flux
membrane vP 0
After certain time intervals, the cake layer has to be removed from the
membrane by module flushing. A decline in permeate flux during the
filtration interval usually is undesirable. Therefore, the reduced permeability
is normally compensated by an increase of the feed pressure (Fig. 5) [6].
F
feed pressure p
permeate flux vP
time t time t
fouling layer
detached
fouling layer
permeate
backflush
feedside
Cross-flow
air-bubble
flush
flush
membrane membrane
feed concentrate
permeate
backflush
fibre diameter
potting 0,5 - 1,5 mm
membrane
In general dead-end filtration is limited by the solid matter content in the raw
water. With increasing solid matter content filtration intervals decline and the
risk of module blocking rises. Thus, the module has to be flushed more often.
Since each module flushing consumes permeate and energy, the yield of the
process decreases considerably and energy costs increase. At higher solid
content levels, dead-end processes tend to be ineffective and cross-flow
filtration becomes more interesting [6].
In contrast to dead-end operations, dynamic filtration or cross-flow filtration
operates with a feed flow parallel to the membrane surface. Particles usually
deposit on the membrane forming a cake layer. This cake layer formation is
controlled by the cross flow (Fig. 8). The cross flow generates shear- and
buoyancy forces at the membrane surface. Depending on the feed
components deposited particles can be moved from the cake layer to the bulk
flow. If a partial removal of the cake layer occurs due to a higher cross-flow
velocity this is referred to as a reversible cake layer formation. After start-up
such systems reach a steady state condition where particle settlement and
removal are in equilibrium. A constant cake layer thickness builds up, which
results in a stable permeate flux [6].
Fouling
Feed - raw water p > p
F P
a Reversible
b Irreversible
m
Permeate flux vP ρ
Membrane
vP 0 b
Permeate - filtrate
Time t
filtrate
air bubbles
permeate
capillary
channel
membrane
air inlet
The occurrence of natural organic matter (NOM) in surface waters cause very
often fouling on the membrane. NOM includes molecular fragments of
plants, algae, microorganism and particularly humic acids [6].
Physicochemical interactions between retained substances with the
membrane are characteristic for all kinds of fouling. Fig. 10 shows potential
fouling mechanisms. Several substances adsorb on the membrane. Under
normal process conditions this kind of fouling can not be inhibited. The
efficiency of the process will be reduced and the permeability decreases [6].
Pore blocking
Particle diameter ~ Pore diameter
Biofouling
Microorganisms in film consisting of EPS
In the German drinking water production primarily bases, acids and H2O2 are
used for a CEB, because these chemicals form neither by-products nor raise
the germination. Chemicals like chlorine, citric acid or commercial cleaners
are used in the CIP/COP [7].
In recent years the use of ultrafiltration has rapidly increased in the field of
drinking water production. Stricter drinking water regulations and
requirements for the treatment are creating a rapid growth in use of
membranes to treat more challenging water sources [8]. The filtration with
porous membranes is a young, but worldwide quickly expanding technology
250.000 500
150.000 300
100.000 200
50.000 100
0 0
88
90
91
92
94
95
96
97
99
00
01
02
03
89
93
98
19
19
19
19
19
19
19
19
19
19
19
19
20
20
20
20
Worldwide MF/UF production capacity (kumulativ)
Number of plants > 8 m³/h
Fig. 11: Development of MF/UF plants worldwide since 1988; not comprehensive [9]
Fig. 12 shows the basic process chart of a conventional water treatment plant.
After pre-treatment by coagulation, precipitation, and sedimentation, the
water passes typically two filter stages. These filters are designed to sustain a
satisfactory hygienic drinking water quality under normal conditions. At
times with heavy rainfall often feed water conditions change rapidly and
dramatically and germ removal cannot be guaranteed with a sufficient
accuracy [6]. After the outbreak of cryptosporidiosis in Milwaukee, USA, in
1993 causing more than 400,000 illnesses and 50 deaths, it was determined
that Cryptosporidium oocysts had passed through the conventional water
treatment including rapid granular filters. Through this incident the interest
in membrane filtration with its ability to provide a superior level of filtration
increased rapidly and led to an unbroken boom of membrane application in
water treatment [2].
2 1
Surface water Drinking water
3
Filter backwash water
Fig. 12: Application concepts of membrane filtration for/at drinking water processing [6]
The reuse of the processed filter back wash water in the process of drinking
water processing today requires a safe removal of pathogens. Membrane
technology offers a feasible option while conventional processes have not
proofed successful. Due to relatively small capacities, filter backwash water
treatment plants often can be incorporated into existing buildings [6].
Colloidal and
precipitation
suspended matters
Coagulation
Sedimentation Flotation
Fig. 13: Typical plan of the treatment of surface water and groundwater [1]
2.1 Coagulation
Al3+ ←⎯ ⎯ ⎯ ⎯ ⎯
+
→ Al(OH)²+ ←⎯ ⎯ ⎯ ⎯ ⎯
+
→ Al(OH) +2 ←⎯ ⎯ ⎯ ⎯⎯
+
→
+ H 2O , − H + H 2O , − H + H 2O,− H
Al(OH)3 ←⎯ ⎯ ⎯ ⎯ ⎯ → Al(OH) −
+ H 2O , − H +
4
Fig. 14: The effect of the coagulant and the membrane type on the TOC removal efficiency [11]
Table 1 and Fig. 14 show that the best performance regarding TOC, colour,
and SUV254 removal was achieved with the PES membrane with Al2(SO4)3 and
PAC 10WA as coagulant. When sodium aluminate was used as coagulant, in
all cases the efficiency was lower. In the case of coagulation with sodium
aluminate prior to the filtration with the PES membrane the removal
efficiency was lower than the PES filtration without pre-coagulation. It also
leads to an increase of pH, which resulted in an increase of the dissociation
degree of the NOM particles. Concurrently the quantity of positively charged
products of the coagulant hydrolysis decreased. Hence, the elimination rate of
organic matter particles was reduced.
The dependence of the residual aluminium on the coagulant type was
investigated by measurements of permeate (Fig. 15).
The lowest aluminium residual was measured when the pre-hydrolyzed
coagulant PAC 10WA was used. Similar values were reported if 'alum'
(Al2(SO4)3) was used. A very high residual, thus the worst result, was noticed
when the coagulation was performed with the sodium aluminate. Both
membranes caused a decrease in the residual aluminium for all tested
coagulants.
In general, regarding the removal rates and the residual aluminium, PAC
10WA and Al2(SO4)3 as coagulants showed a good performance. Sodium
aluminate cannot be recommended as suitable agent for coagulation coupled
with UF [11].
The dosage and the type of coagulant used in a study provide different
results with regard to flux decline. One coagulant, which can be used, is
polyaluminium chloride (PACL). Xia et al. [12] used polyacrylonitrile (PAN)
membranes with a MWCO of 50kDa to investigate the best dosage. The raw
water was filtered inside-out in cross-flow mode. Every 30 to 80 minutes the
membranes were backwashed for 45 seconds. The raw water had a turbidity
of 23.0 NTU (Nephelometric Turbidity Unit).
Fig. 16: Effect of coagulant dose and sand filtration on UF flux [12]
As shown in Fig. 16, dosing 1 mg/L and 2 mg/L of PACL had only little
impact on the flux compared to dosing no coagulant and direct UF. With a
2.1.3 Comparison of coagulation with iron chloride and aluminium sulphate coupled
with UF
Fig. 17 presents the results of the different tests. VP 1 and VP 2 showed that
intermitted aeration had not a large influence on the permeability. Therefore,
it could be considered as inefficient for ultrafiltration with submerged
membranes. If iron chloride was used as coagulant, the phenomenon of
permeability decline was much lower than for aluminium sulphate as
coagulant. The reason for this was presumably the different formation of
flocs, as iron flocs were larger than aluminium hydroxide flocs and had a
lower density. Therefore, they formed a more complex and permeable cake
layer on the membrane surface. The higher the dose of iron chloride (VP 3a to
VP 3c) the lower the decline of permeability was due to higher removal of
dissolved and colloidal matter. Permeate turbidity was constantly below 0.1
NTU. Bacteria and pathogenic germs were completely retained. Residual
coagulant was below the detection limit [13].
Müller et al. [13] showed that pre-treatment with iron chloride was more
efficient than coagulation with aluminium sulphate.
Schippers et al. [14] preferred iron instead of aluminium as coagulant because
of the difficulties with the aluminium sludge and the high cost of this
coagulant. In contrast with other researchers, they determined the best pH-
value was 8 with an iron dosage of 1 mg/L [14]. Schäfer et al. [15]
investigated the influence of ferric chloride pre-coagulation to the membrane
fouling behaviour. They showed that precipitated colloids can cause
membrane fouling by pore blocking or cake deposition. Fig. 18 illustrates the
flux decline over permeate volume for different ultrafiltration membranes
and different ferric chloride concentrations. The flux decreases with
increasing ferric chloride concentration and with higher molecular weight cut
offs (MWCO). An explanation for this phenomenon could be that more open
membranes with a higher initial flux J0 are more sensitive to cake formation
[15].
Fig. 18: Ferric chloride addition in ultrafiltration (100, 30 and 10 kDa membranes, 0,5 mM CaCl2, pH
7-8, 1 mM NaHCO3, 20 mM NaCl, 5 mg/L DOC IHSS FA). [15]
The particles formed under conditions #1 and #2 were so small that more
than 85% of turbidity passed through a 0.2 µm membrane filter. All other
conditions showed good removal efficiencies.
Transmembrane pressure (TMP) measurements during in-line coagulation
revealed a rapid increase for the unconventional conditions (#1 and #2) as
permeate flux increased. This was due to pore blocking or development of a
dense and impermeable cake on the surface of the membrane because under
these conditions the particles had the smallest size. The best TMP
performance was observed at condition #3 (under-dosing with acidic pH) and
#4 (charge neutralisation). The relationship between TMP and flux was linear
for these conditions.
Backwash efficiency was better for charged particles than for the destabilized
particles that were formed during conventional treatment. A larger absolute
value of ZP resulted in more effective hydraulic cleaning.
Summarizing the tests the under-dose condition #3 with acidic pH showed
the best performance since it was possible to obtain good removal rates of
contaminants and improved physical performance. Moreover, less waste
solids were produced when the low amount of coagulant was dosed. As
proved by the application on different waters the absolute values of
coagulant dose and pH depend on the nature of the raw water and on the
operational requirements [17].
2.1.5 Comparison of in-line and standard coagulation coupled with UF considering the
effect on the removal of DOMs
Dissolved organic matter (DOM) in raw water can cause serious fouling on
the surface of the membrane. Coagulation as a pre-treatment before UF is
known to remove DOMs and therefore reduce fouling. Park et al. [18]
examined the two different ways of coagulation, namely in-line and standard
coagulation, with respect to the removal of DOM and membrane
Fig. 20: Variation in TMP as a function of coagulation conditions in the cross flow mode [18]
In the dead-end mode (Fig. 21), like in the cross flow mode, ordinary
coagulation and in-line coagulation improved the membrane filterability. Best
improvement was reached with ordinary coagulation. In the dead-end mode,
the flocs remained intact and formed a cake layer. The cake characteristics
varied with the different coagulation conditions. At ordinary coagulation
Fig. 21: Variation in TMP as a function of coagulation conditions in the dead-end mode [18]
In the pre-coating experiments with PACL at 4.1 ppm, the filterability was
worse than for ordinary or in-line coagulation conditions. This was caused by
the high resistance of aluminium hydroxide particles. By reducing the dosage,
a better filterability could be reached. Therefore, pre-coating could reduce the
need of coagulant, but the removal of DOMs was not that efficient as in the
other conditions. When ferric chloride was used for pre-coating filterability
increased because the iron hydroxide particles formed a very permeable cake
layer (Fig. 22) [18].
Fig. 22: Comparison of TMP between pre-coating conditions with PACL and ferric chloride [18]
Fig. 23: Microscope analysis of 0.1 – 0.5 g/L clay particles without (left) and with coagulation of 0.01
g/L FeCl3 (right) [19]
Fig. 23 shows that the clay particle size ranged from 55 µm to 70 µm with
coagulation while the size ranged from 11 µm to 20 µm without coagulation.
The critical flux 1 was measured with regard to clay concentration and air
bubbling. Without air bubbling, the critical flux decreased with increasing
clay concentration because of a thick deposit set-up on the membrane surface.
With more air injected a lower critical flux could be obtained (Fig. 24a) [19].
Fig. 24b shows the influence of ferric chloride inside the tank. The clay
suspension was destabilized immediately and the critical flux could be
maintained without air bubbling. At very high concentrations (5 g/L), clay
deposit on the membrane surface was drastic and rapid; hence, air bubbling
was necessary to maintain a stable critical flux. At low concentrations, the
flux remained stable even without air bubbling [19].
1 “critical is a flux value below which there is no or negligible particle deposition on the membrane
surface and above which the deposition is significant during the filtration of suspensions” [51]
Fig. 25 shows that the large flocs from coagulation (50 µm to 250 µm) were
dissagregated in smaller flocs through air bubbling. Without air induction,
coagulation enhanced the filtration rate due to the larger porosity of deposit.
The flocs were very distant from each other and therefore increased small
particle retention by cake filtration and sorption phenomena. After air
bubbling, the smaller floc size had no consequence on filtration rate [19].
Fig. 25: Particle size (2.0 g/L, 0.2 g/L FeCl3) a) before and b) after 2 h of 50 L/h aeration [19]
Fig. 26: Typical TMP development for the respective 10 and 30 minutes operating modes [20]
2.1.8 Summary
Instead of dosing a coagulant before UF, the adsorption with iron oxide
particles (IOP) is a possible pre-treatment. It is known that adding IOP into
UF systems increases the NOM removal efficiency and decreases membrane
fouling. However, the interactions between IOP and the membrane have not
been clearly investigated. Ha et al. [21] investigated the characteristics of
adsorptive NOM removal by IOPs in an IOP/UF system with and without
pre-chlorination. To get a better understanding of the efficiency of this
process, they compared it with the removal efficiency of adsorption by
powdered activated carbon (PAC) coupled with UF, which is known to be
helpful in NOM removal. The amorphous IOP slurry (10g/L as Fe) was
prepared in a laboratory by neutralizing a ferric chloride solution using 5 N
NaOH. It had an average surface area of 240 m²/g. The PAC adsorbents had
an average surface of 1400 m²/g. The MWCO of the flat polyethersulfone
membrane was 100 kDa. When they investigate the dosage of both adsorbents
with an adsorption time of 30 minutes, different trends in NOM removal
were found, which was due to the different structures and sorption
mechanisms. Because of high molecular weights, the diffusion of NOM
molecules inside PAC was very low and it took a few days to reach the
adsorption equilibrium. On the other hand, most of the active sorption sites of
Fig. 27: Comparison of SUV254 removal efficiency between IOP and PAC adsorbents: adsorption time 30
min [21]
Fig. 28: Effect of pre-chlorination on flux in IOP/UF and UF alone treating raw water [21]
Groundwater and some water from the anoxic zones at the bottom of
reservoirs often contain Fe2+, and Mn2+, or their complexes with NOM. In
recent years, many water treatment technologies have been tested to remove
these compounds. Nevertheless, these technologies, like a biological trickling
filter, could not meet the WHO guidelines on the Fe and Mn removal, or they
had other disadvantages. Choo et al. [22] investigated a pre-treatment with
chlorination before UF in terms of the removal of iron and manganese from
feed water.
The UF hollow fibre membranes used were made of cellulose and had a
MWCO of 100 kDa. As shown in Fig. 30, there was a rapid increase in
removal efficiency of iron even with a low chlorine dosage of 0.5 mg/L.
Manganese removal efficiency was not as effective. To achieve an efficiency
higher than 40%, at least 2 mg/L as Cl2 had to be added. Hence, a large
amount of chlorine is required if both iron and manganese are present in the
raw water. Choo et al. detected that when a large amount of manganese was
removed by chlorination, serious membrane fouling will occur. At first, it was
assumed that this was caused by the formation of the cake layers on top of the
membrane surface. However, this could not be proved in a UF batch stirred
cell. Surprisingly, they determined that the backwashing was accountable for
the membrane fouling since the concentration of metal oxide particles on the
membrane surface improved the permeability rather than caused fouling,
because these particles could adsorb NOM, which is known as a foulant. The
oxidation of manganese was slower than iron oxidation [22].
Therefore, some of the manganese particles were still small enough to pass
through the membrane pores after chlorination. Afterwards, the growing of
these particles in the pores was assisted by the addition of chlorine during
backwashing, which was done to prevent microbial growth on the membrane
Fig. 30: Removal efficiency of: (a) iron and (b) manganese over time at different chlorine dosages [22]
Fig. 32: SEM pictures of the UF hollow fibre membranes tested: a) virgin membrane (lumen side) b)
Virgin membrane (shell side), c) used membrane (lumen side) and d) used membrane (shell side) [22]
The tests showed that membranes M3 and M4 were very similar in terms of
permeability in spite of their different MWCOs. This was probably due to the
different membrane materials. Acrylonitrile membranes performed better.
Another advantage of this material is that these membranes are more
resistant to free chlorine, which is important when membranes suffer from
biofouling and need a chemical cleaning.
All tested membranes removed the bacteria completely. Fig. 33 shows the
sample of an Enterococcus faecalis assay of feed and permeate for one of the
tested waters. The results were similar for the other permeates.
Thus, all the membranes are suitable for water disinfection.
UF membranes are also used to remove Giardia and Cryptosporidium from
drinking water. Measurements at Hermeskeil (Germany) proved that they
could also be completely removed. (See Chapter 6) [23].
NOM is one of the major reasons causing fouling on the UF membrane. The
major fraction of NOM is humic substances. These substances are negatively
charged when the environmental pH is in the range of natural waters.
Nevertheless, their structure is dependent on the solution chemical
characteristics such as pH and ionic strength. The solution characteristics also
affect the membrane performance, e.g. it can change the membrane surface
charge. However, the membrane pore size is important for the removal
Membrane permeability tests showed that the solutions with a high ionic
strength of 1 x 10-2 M NaCl had a slightly lower filterability than the solutions
with low ionic strength of 1 x 10-3 M NaCl. From these results, it can be
concluded that salt concentration in the feed water influenced the membrane
performance. Further tests revealed that the tighter membrane CA1 presented
a linear variation of the flux as a function of the operating pressure. The same
was discovered for CA2 and CA3 membrane. Fig. 35 also demonstrates the
decrease of flux with increasing ionic strength at both pH-levels. This can be
explained by the conditioning of the membrane due to adsorption of Cl- ions
[25].
Fig. 35: Permeate fluxes of humic acid model solutions. Membrane CA1; operating conditions:
circulation rate, 200L/h, temperature 25°C [25]
On the other hand, the membranes with the highest pore size displayed a
non-linear variation of the permeate flux as a function of the TMP (Fig. 36).
After cleaning with deionised water, the membranes did not recover the flux
Fig. 36: Permeate fluxes of humic acid model solutions. Membrane CA4 (a) and CA5 (b); operating
conditions: circulation rate, 200L/h, temperature 25°[25]
Experiments were carried out with deionised water, a humic acid solution
and well water from Senta, Serbia.
Fig. 37: Humic acid concentration in the permeate measured with UV absorbance, DOC and TOC [29]
Fig. 37 shows that the permeate concentrations were almost one order of
magnitude lower than the feed concentrations for all membranes. The
3.2.2 Use of additives for a higher specific flux and better humic acid removal
Fig. 39: SEM images of PES hollow fibre membrane H2b with an inner skin [30]
Membrane Used Air gap Specific pure water flux Retention of humic acid
additive (mm) (x 10-5 L m-2 h-1 Pa-1 ) (%)
PES - 0 20 97.4
H2a 1,2-PG 0 93 95.2
H3a LiNO3 0 65 85.0
Membrane Used Air gap Specific pure water flux Retention of humic acid
additive (mm) (x 10-5 L m-2 h-1 Pa-1 ) (%)
H2a 1,2-PG 0 93 95.2
H2b 1,2-PG 600 111 97.4
H3a LiNO3 0 65 85.0
H3b LiNO3 2 96 95.6
Iron is often used for coagulation pre-treatment before UF. However, when
coagulation conditions are not optimal, smaller colloidal and soluble iron
species may be present which can affect and clog the membrane [31].
Colloidal iron can also be found in the feed water itself. Soffer et al. [31] used
streaming potential and SEM-EDX (Scanning Electron Microscopy - Energy
Dispersive X-ray Analysis) in their study to investigate fouling interaction of
the colloidal iron with UF membranes and elucidate fouling mechanisms.
They carried out their measurements in a disc ultrafiltration module. Three
polysulfone membranes with MWCO of 10, 30 and 50 kDa and pure water
permeabilities of 22, 35 and 217 L/m²·bar·h and one type F membrane with a
MWCO of 20 kDa and 79 L/m²·bar·h were used. The pH was adjusted in the
range of 2 – 10 during streaming potential measurements using HCl-KOH.
The electrical potential ∆E was measured by varying the applied pressure ∆P
in the range of 10 – 50 kPa. Zeta potential was calculated using the Helmhotz-
Smoluchowski equation:
ΔE / ΔP = ε 0 ⋅ εrζ / η ⋅ k
where ε0 and εr are the vacuum pressure and dielectric constant of the
medium respectively. µ is viscosity of permeate and k is conductivity of
solution. As Fig. 40a) shows, the magnitude of zeta potential is higher for
regular type membranes with a higher MWCO. This is due to changes in the
dielectric constant of water in small pores changing the value of streaming
potential. The 20 kDa type-F membrane had a higher magnitude in zeta
potential at pH ≥7 and a lower one at pH ≤ 6.
Fig. 40: Effect of pH on zeta potential of: a) clean membranes and b) fouled membranes [31]
NOM character, molecular weight distribution, water pH, ionic strength, and
membrane cut-off are the most important factors influencing NOM removal
by UF membrane. Higher removal efficiencies can be obtained by coagulation
pre-treatment.
Kabsch-Korbutowicz [32] assessed the influence of the pH on coagulation
coupled with UF in a lab scale study.
For this purpose, the pH of the feed solution was adjusted by adding
0.1 M NaOH or HCl to the range between 5 to 10. Two membranes, made of
polyethersulfone (PES) and regenerated cellulose (C), both with a MWCO of
30 kDa were tested. Alum was used as coagulant in different doses. To
determine the best pH value during pre-treatment, coagulation was
performed by the jar test method. Thereby the samples were rapidly mixed
for 3 minutes, afterwards they were stirred slowly for 30 minutes and at the
end, they were allowed to settle for 1 hour. Most of the NOM in the feed
solution came from aquatic origin with high hydrophobicity and of high MW
[32].
As shown in Fig. 44 the removal of organic substances from the feed water
during coagulation was highly affected by the coagulant dose and the pH of
the solution. A higher pH caused a lower removal rate at given coagulant
dose. Kabsch-Korbutowicz concluded that the optimal pH for removal of
NOM is in the range of 5 to 6. The reason for this was the fact that the
ionisation of carboxyl groups of humic substances led to a condition in which
they had a negative charge. On the other hand, at pH conditions ≤ 6
aluminium formed substances with a positive charge, like Al3+ and Al(OH)2+.
These substances could either induce a complexation with humic substances
or they could neutralize the negatively charged sites of NOM (charge
neutralisation). At higher pH values, Al(OH)3 was predominant and the
NOM removal was dependent on the adsorption of humic substances on
Al(OH)3 crystals [32] (sweep coagulation).
Fig. 42: Comparison of TOC removal in different processes (coagulant dose: 3.59 g Al/m³) [32]
For all the processes, the maximum efficiency of NOM removal was observed
at pH 6 with PES membranes and an alum dose of 3.59 g Al/m³ [32] In a
range between pH 6 to 8 similar TOC removal rates were observed.
Fig 43 a) shows, that the NSP decreased immediately after each backwash
(BW) because of a temporary release of blocked pores, which soon were
blocked again. In the case of pre-treatment of the P+C+D fraction with
coagulation, an improved membrane performance could be detected. This is
probably due to a better permeable fouling layer or to prevented pore
blockage. Regardless heavy fouling occurred. Fig 43 b) shows comparable
results for the C+D fraction, but the permeability of the fouling layer seemed
to be lower than in the P+C+D case. Fig 34 c) shows high losses of
permeability during each filtration interval of the P+C fraction but good
recovery though backwashing. Permeability was almost stable with
coagulation pre-treatment. Fig 43 d) shows that NSP declined significantly
during the first filtration cycle of the C fraction and gradually stabilized after
backwashing (BW). Coagulation did not advance the NSP considerably.
Hence, the coagulation impact on fouling in the C case is limited. The
membrane performance for the D fraction was very poor and BW showed no
effect when performed without coagulation (Fig 43 e). With coagulation, the
membrane performance was much better. This might be due to ligand effects
caused by coagulation, which changed the dissolved NOM physicochemical
characteristics.
To compare the NSP of the different fractions better, two factors were
introduced. The permeability decline coefficient (PDC) is defined as the
Fig. 44: Permeability decline coefficient (PDC) and influence of coagulation on membrane performance
[33]
Fig. 46: LC-OCD analysis for feed and product water [35]
Fig. 47: TMP vs. Time for each NOM fraction (138.5 L/m²·h) [35]
Fig. 48 shows a detailed analysis of the filtered water and the three NOM
fractions. As can be seen, only the hydrophilic and the filtered water
contained polysaccharides. This was the reason for the higher fouling rates of
these fractions.
The researchers also investigated the reversibility of the fouling for each
fraction.
Lee et al. [37] also investigated NOM fouling. In their study they compared
four different membranes, two MF membranes with hydrophilic properties
(HPI) and two UF membranes, one with hydrophilic and one with
hydrophobic (HPO) properties. A contact angle of over 50° is considered as
hydrophobic. Tests were performed in a dead-end stirred cell filtration unit.
Taken the different water qualities and their fouling potential into
consideration, four different surface waters were selected. The waters were
pre-filtered with a 0.45 µm filter. After that, they were fractionated with XAD-
8/4 resins (Table 8).
Source water Hydrophobic DOC [%] Transphilic DOC [%] Hydrophilic DOC [%]
Marne River 50 22 28
Cazau Lake 39 25 35
Bultière Reservoir 44 26 31
Yffiniac River 48 28 24
Permeation tests showed a greater flux decline for the Cazau Lake and
Bultière Reservoir waters, which contain a higher hydrophilic fraction than
for the other two waters. As in the tests by Kennedy et al. (2005) [35], organic
colloids and polysaccharides passed through the XAD 8/4 resins and ended
up in the hydrophilic fraction. SEC-DOC and UV responses of source waters
based on the relative MW distribution determined by PEG standards showed
a high first peak for these two waters. The first peak is typically assumed to
be proteins and/or polysaccharides. Flux decline trends of the different
membranes based on delivered DOC showed the most significant decline for
the hydrophobic MF membrane. UF membranes showed less flux decline
than MF membranes. Nevertheless, distinctive differences could not be
observed, suggesting that other membrane properties might be more
influential.
AFM and SEM analyses showed that the clean UF membranes had a
smoother surface than the MF membranes. The PES (UF-HPO) membrane
were smoothest and had the least flux decline for all the feed waters. Fig. 49
to Fig. 51 show the 3-D views, section analysis, and SEM images of clean and
fouled membranes, respectively. The fouled membrane surfaces appeared
smoother than the corresponding clean membrane surfaces and had a surface
coverage with a higher topography. AFM images supported the notion of
surface coverage as a fouling mechanism of UF membranes (Fig. 49). On the
other hand, the AFM images of MF membranes indicated that the filling of
the membrane pore structure with foulant material was the reason for
fouling, because the AFM “valleys” largely disappeared for the fouled
membranes (Fig. 50). The roughness of each membrane was very different.
Lee et al. concluded that the roughness could be more important in
membrane fouling by controlling interaction between fouling components
and the membrane surface / pores than the hydrophobic / hydrophilic
character of the membrane. Fig. 51 shows a reduction of the pores of the
GSWP (MF-HPI) membrane when filtered with Marne River water and pore
blockage with NOM material when filter with Cazau Lake water. This could
be due to the adsorption of NOM around membrane pores by smaller
molecules and/or pore blockage by larger molecules and colloids. This
showed the contribution of source water characteristics on membrane fouling.
Fig. 50: Section analysis of MF membrane surface by AFM (membrane: GVHP (MF-HPO); from left:
clean Cazau Lake and Bultière Reservoir) [37]
Fig. 51: SEM Images of GSWP (MF-HPI) membrane (from the left: clean and filtered with Marne River
and Cazau Lake; magnification: 9500 x) [37]
Contact angles were measured before and after filtration. At first, the contact
angle of hydrophilic membranes was less than 20° and it increased after
filtration. The contact angle of hydrophobic membranes was primarily greater
than 50° and it decreased after filtration. This reflected the character of the
foulant material. Backwash tests revealed that the flux recovery of UF
membranes was higher than that of MF membranes. This fact supports the
assumption that fouling of UF membranes was caused mainly by cake/gel
layer formation while MF membranes were affected more by the blocking of
pores.
Overall, they concluded that natural waters with a high content of
hydrophilic fraction resulted in more significant flux decline. This might be
due to colloidal and macromolecular matter in this fraction [37]. Later Lee et
al. [38] reported the same in another study. Membrane roughness was more
Fig. 52: SEM of the UF membrane, (a) cross-section of the new, unfouled membrane, (b) cross-section of
the membrane with fouling layer after 33 d of operation, arrow indicates the surface of the active layer,
(c) top-view with deposited particles after 33 d of operation [39]
Fig. 53: FISH of the UF fouling layer after 5 days of operation. (a) DAPI (total cell count), (b) α-subclass
(ALF1b, Fluos), (c) β-subclass (BET42a, Cy3), (d) γ-subclass (GAM42a, Cy5) [39]
Fig. 54: Pictures are taken after a run of 18 h. The run was stopped after a hydraulic cleaning. The
blocking of the capillaries is both observed on the feed side (left) as well as on the concentrate side (right)
of the module [40]
The injection of air during a filtration or cleaning phase, in order to limit the
formation of a particle deposit or concentration polarisation, is known as air
sparging. It was first used in 1986 by Hitachi in electro dialysis systems [41].
Cabassud et al. [41] investigated the effect of air sparging in their lab-scale
study in an inside-out UF of hollow fibres to prevent particulate fouling. The
mean pore diameter was 0.01 µm and the membrane inner diameter was 0.93
mm. Air was added to the feed water at the bottom of the vertical module
through a porous membrane. The up-flow velocity of the raw water and the
air was constant. TMP was constant as well. ULS (liquid slug velocity) was
between 0.5 and 0.9 m·s-1 and UGS (gas slug velocity) was between 0 and 1 m·s-
1. The first test was conducted with river water and a clay suspension. It
revealed that air sparging was more efficient with water containing a high
fouling ability in terms of external particle deposit or of concentration
polarisation. The reduction of the hydraulic resistance by air increased with
the fouling ability of water and therefore with decreasing critical flux. Fig. 56
shows the results of a cake filtration modelling for the two-phase flow.
Fig. 56: Influence of gas velocity on cake porosity(a), specific resistance (b) and thickness(c) –
(clay- calim = 1 g l-1, ULS = 0.9 ms-1, TMP = 1 bar, T= 20°C) [41]
The two-phase flow expanded the cake until a gas velocity of 0.35 m·s-1 was
reached. From there, the cake porosity was too high to have a physical
significance and no more cake deposit was observed because the cake
thickness reached the range of the inner radius of the hollow fibre. Even near
the membrane surface, the particles were in suspension and could not form a
cake layer. Without a cake layer there was no more additional resistance to
mass transfer, thus flux was enhanced. Further analysis demonstrated that
the flux enhancement was mainly related to mixing and turbulence created by
Fig. 57: Conceptual diagram of permeate flux behaviour before and after bleeding [42]
The curve in Fig. 58 can be subdivided into four categories. Within the first
three categories the temperature showed no significant impact and the PIC
stayed almost constant at certain level. In the first category, there was almost
no change in PIC when cleaning was performed with H2O2 or distilled water.
The acids in category 2 reached a PIC of 1.5 whereas the oxidants, bases and
the combination of acid and alkaline cleaning in category 3 had a PIC of
approximately 2.5. The remarkable difference between neutral and alkaline
H2O2 stands out. This might be due to a more open structure of the fouling
layer at basic conditions that enabled a better cleaning with the agent. The
best results were achieved by a successive alkaline and acid cleaning at 40°C
and by H2O2/NaOH at 40°C. The order of basic and acid cleaning is
important because HCl followed by NaOH resulted in a lower PIC as the
other way round.
After that, they focused on the cleaning efficiency of the agents on different
compounds. Inorganic compounds were poorly removed by uncharged
oxidants and H2O2 showed no effect at all. On the other hand, citric acid was
an effective agent for the elution of iron and formed complexes with iron.
Charged H2O2 followed by HCl was also successfully in iron and manganese
elution. The order of basic / acid cleaning steps showed an impact neither on
iron nor on aluminium dilution. However, NaOH followed by HCl showed
better results for manganese dilution as the other way round.
After that, the CDOC fraction was subdivided into polysaccharides, humic
substances, building blocks, neutral and amphiphile substances and acids
(Fig. 61).
Here, acidic cleaning showed almost no effect. NaOCl achieved the best
results among all the cleaning agents at a temperature of 40°C. Base and
combinations of base and acid showed good removal efficiencies. As
observed previously, the removal efficiency increased with increasing
Table 10: Overview of selected installed or planed UF plants in drinking water production
Fig. 62: The new and advanced drinking water line of WW Roetgen [45]
Fig. 65: Picture of WW Roetgen and draft design of the new drinking water line [46]
The operational experience of the last 6 years from the UF drinking water
treatment plant at Hermeskeil (Germany) was published in 2005 by Hagen et
al. [47]. This system began to operate in 1999 and has a maximum drinking
water production capacity of 140 m³/h. The average capacity produced was
60 m³/h. The raw water had a pH of approximately 5.0 to 6.5. The iron
content in the raw water changed from season to season between 0.01 and
0.03 mg/L. Manganese content was between 0.02 and 0.08 mg/L. The
Hermeskeil treatment plant consists of two stages. In the first stage, the water
is pre-treated with a limestone filter for alkalisation and to adjust the
hardness as well as to remove manganese and aluminium. After that,
membrane filtration is conducted with hollow fibre membranes with an inner
diameter of 0.8 mm in the dead-end mode with an inside-out configuration.
During operation, different chemical cleaning methods were tested. Best
results with respect to TMP and permeability were achieved with a
combination of HCl and NaOH. This confirms the results by Strugholtz et al.
(2005) [43]. Turbidity removal was successful. Table 11 shows that bacteria
and protozoa are completely removed by the treatment plant.
Overall, the water treatment plant works stable, fully automated and safe.
The quality of the filtrate complies with the German drinking water
guideline.
6.4 Switzerland
Another five UF treatment plants in the Switzerland have been used for the
production of drinking water for more than three years [49]. Their capacity
varies from 36 to 450 m³/h. The hollow fibre membrane modules are run in a
combined dead-end/cross-flow mode. Backwash is conducted every
30 minutes, disinfection one to three times a day, and chemical cleaning was
necessary between one to six times per year.
The water quality of permeates was good for all five plants. No bacteria were
detected and the turbidity was always below 0.01 FNU.
During the three years of operation, only four fibres were broken in more
than 200 modules. Raw water yields were between 93 and 98%. NaOCl was
used for disinfection and NaOH for chemical cleaning. Table 13 shows the
operational costs. Klahre et al. [49] conclude that UF with membranes is an
interesting alternative.
The following cleaning agents are used for chemical cleaning [8]:
Fig. 68 shows the process flow diagram of the water treatment plant in Choa
Chu Kang. The existing treatment includes aeration, pre-chlorination for
disinfection and algae control, and coagulation with alum and a coagulant aid
polymer. The submerged UF membranes, which are manufactured by
ZENON, will be integrated into an existing sand filter bed.
After the membrane step the water will be ozonated for disinfection, before it
passes to a granular activated carbon (PAC) contactor, which will be used for
DOC removal to improve the water stability. The last treatment steps are
chlorination followed by CO2 dosing, lime addition for corrosion control, and
ammonia addition to convert chlorine to chloramines [8].
Fig. 68: Process flow diagram for the water treatment plant [8]
Table 16: Summary of the Choa Chu Kang water treatment plant [8]
In Inverness (UK) ultrafiltration is used for surface water treatment. The raw
water is abstracted from two lakes. Most of the catchment area is forest or
pasture. In the pasture some sheep and deer are grazing. So the risk of
Cryptosporidium contamination is elevated. Table 18 shows the raw water
quality of the Inverness water treatment plant [8].
The ultrafiltration, which is installed, has a capacity of 34 439 m3/d. The start-
up of the plant was in 2002. The process flow diagram is shown in Fig. 69. The
raw water is pumped from the lakes to 80 µm rotating drum strainers. For
pre-coagulation PACL is used to remove colour from the water. The pH-level
for the coagulation process is 6.7. The installed membranes are ultrafiltration
membranes from X-Flow.
Fig. 69: Process flow diagram for the water treatment plant [8]
Test on the effect of ionic strength on humic substances removal show that
solutions with a low ionic strength have a higher permeability than solutions
with a high ionic strength. An increase in ionic strength neutralizes the
negative surface charge of the membrane and reduces the charge repulsion of
functional groups in the pores.
An improved removal of humic acids can be achieved if fibres with an
additive are used for filtration. Fibers spun under an air gap had both higher
flux and higher retention than wet spun fibres.
Instead of dosing a coagulant before UF, the adsorption with iron oxide
particles is also an option for pre-treatment. Pre-chlorination is not efficient.
Ozonation with a minimum concentration of 0.05 mg/L of ozone
permanently in permeate can also prevent fouling efficiently and yields of
almost 100% can be obtained. Bromide in the raw water limits this process
because toxic bromates can be generated as oxidation by-products.
The WW Roetgen (Germany) and other drinking water production plants use
UF membranes successfully. Permeate quality complies with the drinking
water guidelines and membrane durability is good.
[10] Leiknes, T., Ødegaard, H., Myklebust, H., Removal of natural organic
matter (NOM) in drinking water treatment by coagulation–microfiltration
using metal membranes. Journal of Membrane Science 242 (2004).
[12] Xia, S., Nan. J., Study of drinking water treatment by ultrafiltration of
surface water and its application to China. Desalination 170. (2004).
[13] Müller, S., Pilotversuche zur Aufbereitung von Elbuferfiltrat mit einer
getauchten Ultrafiltrationsmembran. TU Dresden. 6. Aachener Tagung
Siedlungswasserwirtschaft und Verfahrenstechnik (2005).
[15] Schäfer, A. I., Fane. A. G.. Waite. T. D.. Cost Factors and chemical
pretreatment effects in the membrane filtration of waters containing natural
organic matter. Water Research. Vol. 35 (2000).
[20] Leiknes, T., Myklebust, H., Metal Membranes for Drinking Water
Treatment. NTNU Trondheim.
[22] Choo. K.-H.. Lee. H. . Iron and manganese removal and membrane fouling
during UF in conjunction with prechlorination for drinking water treatment.
Journal of Membrane Science 267 (2005).
[23] Arnal, J.M., Sancho, M., Selection of the most suitable ultrafiltration
membrane for water disinfection in developing countries. Desalination 168
(2004).
[25] Costa, A. R., Pinho. M. N., Effect of membrane pore size and solution
chemistry on the ultrafiltration of humic substances solutions. Journal of
Membrane Science 255 (2005).
[27] Cornet, P., Diffusion of humic acid in dilute aqueous solutions. J. Colloid
Interface Sci. 110 (1986).
[28] Saravia, P., Der Einfluss von Ladung und Ionenstärke auf den
Membranrückhalt. Engler-Bunte Institut. Universität Karlsruhe. 6.
Aachener Tagung Siedlungswasserwirtschaft und Verfahrenstechnik
(2005).
[29] Domany, Z., Galambos, I., Humic substances removal from drinking water
by membrane filtration. Desalination 145 (2002).
[30] Qin, J.-J., Oo, M. H., Hollow fibre ultrafiltration membranes with enhanced
flux for humic acid removal. Journal of Membrane Science 247 (2005).
[31] Soffer, Y., Gilron. J., Streaming potential and SEM-EDX study of UF
membranes fouled by colloidal iron. Desalination 146 (2002).
[35] Kennedy, M. D., Chun, H. K., Natural organic matter (NOM) fouling of
ultrafiltration membranes: fractionation of NOM in surface water and
characterisation by LC-OCD. Desalination 178 (2005).
[38] Lee, N., Amy, G., Understanding natural organic matter fouling in low-
pressure membrane filtration. Desalination 178 (2005).
[39] Hörsch, P., Gorenflo, A., Biofouling of ultra- and nanofiltration membranes
for drinking water treatment characterized by fluorescence in situ
hybridization (FISH). Desalination 172 (2005).
[41] Cabassud, C., Laborie S., Air sparging in ultrafiltration hollow fibres:
relationship between flux enhancement. cake characteristics and
hydrodynamic parameters. Journal of Membrane Science 181 (2001).
[42] Choi, H., Kim, H.-S., Pilot plant study of an ultrafiltration membrane
system for drinking water treatment operated in the feed-and-bleed mode.
Desalination (2005).
[44] Dautzenberg, W., Holy, A., Planung und Bau der größten deutschen UF-
Membrananlage zur Trinkwasseraufbereitung und Spülwasserbehandlung in
Roetgen. 6. Aachener Tagung Siedlungswasserwirtschaft und
Verfahrenstechnik (2005).
[46] Panglisch, S. Holy, A., Transferring pilot experiments into the planning of
Germany’s largest two-stage ultrafiltration plant. Desalination 179 (2005).
[49] Klahre, J., Robert. M., Drinking Water from Ultrafiltration – first Results
with Full-Scale Membrane Filtration Plants in Mountain Areas.
[50] http://www.doc-labor.de.
[51] Zhang, Y.P., Fane, A.G., Law, A.W.K., Critical flux and particle
deposition of bidisperse suspensions during crossflow microfiltration. Journal
of Membrane Science, article in press (2006).