Journal of Membrane Science 405406 (2012) 284290

Contents lists available at SciVerse ScienceDirect

Journal of Membrane Science

journal homepage: www.elsevier.com/locate/memsci

Extraction of amaranth starch from an aqueous medium using microltration:

Membrane characterisation
Paul G. Middlewood a,b, , James K. Carson b
a
b

AgResearch Ltd., Private Bag 4749, Christchurch 8140, New Zealand

University of Waikato, Private Bag 3105, Hamilton 3240, New Zealand

a r t i c l e

i n f o

Article history:
Received 18 January 2012
Received in revised form 8 March 2012
Accepted 9 March 2012
Available online 18 March 2012
Keywords:
Amaranth starch
Microltration
Al-Hakkak process

a b s t r a c t
The research reported here investigated microltration as an alternative to density-based processes for
separating amaranth starch-milk produced by the pilot-scale Al-Hakkak process into a starch-rich concentrate and an aqueous stream containing the soluble proteins and carbohydrates. A Millipore ProFlux
M12 Tangential Filtration System, tted with a 1000 kDa regenerated cellulose membrane, was used as
the experimental apparatus. The selected membrane retained all the starch granules, but also retained
more protein than desired (protein retention was 67% and the starch-rich concentrate had a dry-basis
protein content of 12%). Analysis of the feed liquor, and dialtered concentrate, revealed the presence
of some non-starch insoluble material. This material, which may have been protein-based, was present
in the starch-milk produced using the pilot-scale method but not the laboratory-scale method, and its
presence determined the nal protein content of the dialtered concentrate. The optimal transmembrane pressure was approximately 100150 kPa, above which ux increased non-linearly with pressure.
However, the uxpressure relationship was weak, suggesting that higher operating pressures may be
sustainable.
2012 Elsevier B.V. All rights reserved.

1. Introduction
The size of a starch granule varies between plant species and
affects many functional and physicochemical properties of the
starch, and hence its potential uses [1]. The majority of commercially available starches have a medium (1025 m) or large
(>25 m) granule size (Table 1). Amaranth seed is one of the few
sources of small granule starch, typically 13 m in diameter, and
having regular granule size [2]. Amaranth has become a subject
of renewed interest due to the nutritional value of its seed, and
the potential for using the various seed components (especially
starch and proteins) as functional ingredients in both food, and nonfood (e.g. cosmetic) applications. Nutritionally, amaranth seed has
a higher protein content, higher digestibility, higher protein utilisation, and a higher protein efciency ratio than traditional cereals
such as corn and wheat [3]. The protein has an amino acid prole
that is well balanced, approximates the World Health Organisation
standard protein, and includes lysine (an essential amino acid that
most cereals lack, or have in small amounts) [4]. Amaranth does
not contain gluten, which makes it a suitable food for those with
celiac disease. Amaranth seeds are also a good source of dietary

Corresponding author.
E-mail addresses: paul.middlewood@agresearch.co.nz, octie@paradise.net.nz
(P.G. Middlewood).
0376-7388/$ see front matter 2012 Elsevier B.V. All rights reserved.
doi:10.1016/j.memsci.2012.03.027

bre [5,6]. Outside of nutrition, research has focused on investigating the properties, isolation methods, and uses, of amaranth starch,
proteins, and oil.
At the time of this research, no commercial amaranth starch
extraction methods exist. The Al-Hakkak process [7] has been
used to extract starch from amaranth seed on a laboratory-scale,
and work is currently underway to scale-up this process. The AlHakkak process produces an intermediate product stream known
as starch-milk, which contains the insoluble starch granules and
other seed components such as water extractable carbohydrates,
protein, and fat. On the laboratory-scale the starch is recovered
from the starch-milk using a high-speed centrifuge. However, at
pilot and commercial scales the use of density-based processes such
as centrifuges, settling tables, or hydrocyclones may not be practical as the small granule size either reduces the efciency of, or
completely excludes, density-based separations [8].
An alternative to density-based separations is tangential ow
ltration (TFF). To be technically viable in an industrial application
a TFF process must achieve the desired separation, perform the
separation using a realistic membrane area, and any fouling that
occurs must be removable. Unfortunately it cannot be assumed
that selecting a membrane with an apparently suitable pore size
will perform the desired separation since a membrane pore size, as
stated by the manufacturer, is a guide only. Small-scale trials are
recommended to conrm that the selected membrane achieves the
desired separation because of the many factors that can affect the

P.G. Middlewood, J.K. Carson / Journal of Membrane Science 405406 (2012) 284290
Table 1
Granule size of different starches [5].
Source

Granule diameter

Amaranth
Barley
Corn
Oats
Rice
Sorghum
Wheat

13 m
Bimodal, 2025 m and 26 m
15 m
310 m
38 m
25 m
Bimodal, 2035 m and 210 m

actual separation [9]. A key parameter that determines the required

membrane area is the permeate ow rate, which is dependent on
a number of factors. At present no mathematical models exist that
can accurately predict, from rst principles, the TFF performance
of complex solutions. Therefore, data must be collected to determine the membrane selectivity, permeate ow rate, and optimal
operating conditions.
The objective of this research was to perform a preliminary
assessment of the suitability of using TFF for recovering amaranth
starch from the aqueous starch-milk produced by the pilot-scale
Al-Hakkak process. The key goals were to propose a suitable membrane, and determine whether it retained the starch granules while
passing the soluble carbohydrates and proteins and to determine
the permeate ow rate, and its relationship with operating conditions and liquor concentration. Membrane fouling and cleaning,
which can be a signicant issue in TFF processes [10], will be the
focus of a later publication.
2. Methods and materials
2.1. Feed liquor preparation
Organic amaranth whole our was purchased from Chantal
Organic Wholesalers (Napier, New Zealand). The amaranth seeds
were of the White Oscar variety, and the whole our contained
12.7% protein, 5.08% fat, 5.1% bre, and 59.4% carbohydrate (composition data provided by the manufacturer). The whole our was
sieved through a pilot-scale vibrating screen (tted with a 100 m
screen) to obtain the fraction less than 100 m. Healtheries Fine
Ground Gluten Flour was purchased from a local supermarket,
salt (sodium chloride) was obtained from SigmaAldrich. Feed
liquor (starch-milk) was produced using the pilot-scale Al-Hakkak
method [11]. This method was completed in two parts. Firstly, a
stiff dough was made by combining amaranth our (2485 g), vital
wheat gluten (620 g), 1% salt solution (185 g), and water (1710 g,
25 C) using a Varimixer AR40 planetary mixer. Once formed, the
dough was allowed to rest for 90 min. Secondly, two successive
washes were used to wash the starch from the dough. Water (20 L,
25 C) was added to the dough and mixed for 90 min. The solution
was passed through a pilot-scale vibrating screen (40 m) to separate the starch-rich liquor from the solid dough residue. The dough
residue was returned to the mixer and a second wash performed
(20 L of water, 40 min). As per the rst wash, a pilot-scale vibrating
screen was used to separate the starch-rich liquor from the solid
dough residue. The two lots of starch-milk were combined to make
a bulk batch of feed liquor. This bulk liquor was divided into 2 L
lots and frozen until required. The day before use, the starch-milk
was defrosted by standing at room temperature. Despite the starchmilk being screened before freezing, the defrosted starch-milk had
some small (25 mm) lumps. These lumps were brown in colour,
resembling the spent dough retained by the vibrating screen during the starch-milk preparation. Prior to use the starch-milk was
ltered through a 30 m screen, and heated to 25 C on an IKAmag
Ret-G hot plate.

285

Additional feed solutions were prepared by splitting a sample of

the standard feed into its soluble and insoluble components. These
feed solutions were used to investigate how the soluble and insoluble components contributed to membrane performance. Four litres
of the standard feed liquor was divided into 200 mL lots and centrifuged (Beckman Avanti J-301 laboratory centrifuge, 2500 g,
15 min). The supernatant was decanted off and set aside as Feed
A1. The pellets were washed three times. For each wash, the pellet was re-suspended in 200 mL of distilled water, mixed for 5 min,
and re-centrifuged. The washed pellets were then combined, made
up to the original starting volume (4 L) with distilled water, and
set aside as Feed A2. Feed A1 was used the same day that it was
produced; Feed A2 was stored in a refrigerator and used two days
later.
2.2. Filtration equipment
All experiments were performed using a Millipore ProFlux M12
Tangential Filtration System. This unit has a 3 L feed tank with
heating/cooling coil, a variable speed feed pump, inlet and outlet pressure sensors, permeate and retentate owmeters, and a
back-pressure control valve. The unit can be used with a variety
of different membranes, and can be congured to operate in concentration mode, or recycle mode (Fig. 1). A Techne water bath
was used to supply hot, or cold, water to the heating/cooling coil,
a Mettler-PM34 balance was used to record process masses, and a
Center-305 thermometer tted with a K-type thermo-couple was
used to measure temperatures.
The membrane used for all trial work was a Pellicon 2 Ultracel PLCXK membrane (purchased from Millipore, through their
New Zealand agent Bio-Logic Solutions Ltd.). The PLCXK membrane
had a at plate (cassette) conguration with turbulence promotion
screens between the plates, a ltration area of 0.1 m2 , a nominal
molecular weight limit (NMWL) of 1000 kDa, and was constructed
of regenerated cellulose. Regenerated cellulose was chosen as the
membrane material because it is hydrophilic, which minimises
non-specic protein binding. This offers two advantages: (i) protein losses should be low, which is important as the protein-rich
permeate is a potential co-product; (ii) protein-based membrane
fouling should be minimised. One constraining issue is that regenerated cellulose membranes have a low tolerance to pH extremes,
high temperature, and some chemicals [12]. This reduces the range
and severity of steps that can be used to clean the membrane.
A NMWL of 1000 kDa was chosen as it was the closest available
size to the range needed to pass the soluble proteins while retaining
the starch granules. For retention applications, membrane suppliers recommend using a NMWL that is 1/3 the size of the material to
be retained. Amaranth starch granules may be as small as 0.5 m
[13]; therefore the recommended pore size is less than 0.2 m. For
passage applications, a factor of 5 is recommended. As the largest
protein present in the starch-milk is approximately 80 kDa [11], a
NMWL greater than 400 kDa is required to ensure all the protein
passes through the membrane pores. Millipore claim their NMWL
1000 kDa membrane will retain greater than 99% of molecules
larger than 0.03 m [14]. As such, this NMWL was expected to
retain the starch granules in the feed liquor and pass the soluble
proteins.
2.3. Procedures
Flux versus time data were collected while running in recycle mode, with constant cross-ow, trans-membrane pressure
(TMP), and temperature (typically 20 L h1 , 50100 kPa, and 25 C,
respectively). To begin a run, the desired mass of feed liquor was
transferred into the feed hopper, and the feed pump started and set
to give the required feed rate. The back-pressure valve was then

286

P.G. Middlewood, J.K. Carson / Journal of Membrane Science 405406 (2012) 284290

Permeate
(Concentration Mode)

Permeate
(Recycle Mode)

Permeate
Flowmeter

Outlet
Pressure
Gauge

Permeate
Retentate
Flowmeter

Temperature
Controlled
Water Bath

Retentate

Membrane
Back
Pressure
Valve

Feed
Vessel
with
Cooling
Coil

Inlet
Pressure
Gauge

Feed
Drain
Valve
Feed
Pump
Fig. 1. Schematic ow diagram of the microltration equipment.

adjusted to give the required TMP. Temperature was controlled

by altering the temperature of the water bath that fed the heating/cooling coil, and cross-ow rate was controlled by adjusting the
feed pump speed. Flux was measured at 5 min intervals for the rst
30 min, and at 15 min intervals for the remainder of the run (typically 34 h). As the ux was generally below the readable scale of
the permeate owmeter, permeate ux rates were determined by
measuring the mass of permeate collected over a 12 min period.
Flux versus TMP data were collected by bringing the system to
steady state by operating in recycle mode, at a low TMP (50 kPa),
for 23 h. Once at steady state the baseline ux was measured, and
the operating conditions (inlet and outlet pressure, cross-ow rate,
pump speed, and temperature) recorded. The TMP was increased
by approximately 50 kPa by closing the back-pressure valve. If
required the feed pump speed was increased to compensate for any
decrease in cross-ow resulting from the increased back-pressure.
The system was allowed 1520 min to stabilise, and then the ux
and operating conditions recorded. This was repeated for increments of 50 kPa, up to a maximum TMP of 200 kPa. The sequence
was then performed in reverse (incrementally decreasing TMP) to
check for hysteresis.
Flux versus concentration data were collected by bringing the
system to steady state by operating in recycle mode for 23 h. Concentration was then started by diverting the permeate away from
the feed tank and into a collection vessel. The operating conditions
were kept constant and the mass of permeate against time was
recorded, as were the key process variables. Retentate and permeate samples were taken each time the volumetric concentration
factor (VCF) doubled (i.e. VCF1, VCF2, VCF4, VCF8 or VCF nal).
A dialtration trial was performed by combining the retentate
solutions from two concentration runs, which had been stored in a
frozen state (approx. 20 C), to use as the feed. This combined feed
liquor had 11% total solids, 0.12% ash, 1.24% protein, 9.2% starch,
and 0.2% fat. The system was operated in recycle mode (cross-ow

20 L h1 , TMP 100 kPa, temperature 25 C) until a steady ux was

obtained. Batch-wise dialtration was then performed by adding
RO water to the feed (at a ratio of 1:1), and running in concentration
mode until the original feed volume was reached. This sequence
was repeated six times, i.e. six diavolumes were performed. Retentate samples were taken after the rst, second, forth, and sixth
diavolumes, and permeate samples were taken after each of the
six diavolumes.
2.4. Sample analysis
The starch, protein, moisture, fat, ash, total solids and suspended
solids contents of samples were determined using standard methods [15]. Samples were examined visually using a JEOL JSM 7000F
eld emission gun scanning electron microscope (SEM) (operated
at 5 kV and viewed at a working distance of 15 mm). Sample viscosities were measured using a Cannon-Ubbelohde 50 M649 tube
viscometer.
3. Results and discussion
3.1. Feed composition analysis
The composition of the feed liquor is shown in Table 2. The soluble fraction of the feed obtained by centrifuging (Feed A1) was
tested for total solids, ash, and protein, while the insoluble fraction
(Feed A2) was washed in distilled water, and then subjected to the
same tests. Results for the soluble fraction (Table 3) and insoluble
fraction (Table 4), show that most of the protein present in the feed
went into the soluble fraction, but a small amount also remained
with the insoluble fraction. Insoluble material will not generally
pass through a 1000 kDa regenerated cellulose membrane as it is
usually too large to enter the pores; if it could enter the pores it

P.G. Middlewood, J.K. Carson / Journal of Membrane Science 405406 (2012) 284290

287

Table 2
Starch-milk composition.
Component

Concentration
(%g g1 )

Starch (insoluble granules)

NSP (soluble carbohydrate)
Protein
Fat
Ash

1.10
0.54A
0.33
0.09
0.10

Total solids

2.16

(%DB)
51
25A
15
4
5
100

Calculated by difference, DB, dry-weight basis.

Table 3
Composition of the starch-milk soluble fraction (Feed A1).
Component

Concentration
(%g g1 )

NSP (soluble carbohydrate)

Protein
Ash

0.48A
0.27
0.10

Total solids

0.84

(%DB)
57A
32
11
100

Fig. 2. SEM image of the starch-milk insoluble fraction. The starch granules are
polyhedral the amorphous material is insoluble non-starch material.

Calculated by difference (if fat or bre is present it will be included here).

14
TS

Table 4
Composition of the starch-milk insoluble fraction (Feed A2).
Concentration
(%g g1 )
NSP (soluble carbohydrate)
Protein
Ash

1.07A
0.05
0.01

Total solids

1.13

(%DB)
95A
4
1
100

Concentraon (% g g-1)

Component

Ash

Protein

Starch

NSP+fat

12
10
8
6
4

Calculated by difference (if fat or bre is present it will be included here).

3.2. Membrane characterisation

The change in composition of the retentate and permeate
streams was measured, and plotted, across a concentration run. The
retentate and permeate data are shown in Figs. 3 and 4, respectively.

0
2

10

Volumetric Concentraon Factor (VCF)

Fig. 3. Plot of retention composition during concentration.

1.2
TS

Ash

Protein

NSP+Fat

1.0

Concentraon (% g g-1)

would be unlikely to pass all the way through due to the tortuous
pore path. Identifying the presence of protein in the insoluble fraction was a key nding as it indicated that some of the protein may
have been in a form, or associated with the starch in such away,
that microltration could not remove it from the starch-milk.
The dry-basis (DB) protein content of the starch-rich stream
(Table 4) was 4%, which is considerably higher than the 0.1%
achieved by Al-Hakkak and Al-Hakkak [7] during their laboratorybased study. A possible explanation is that the Al-Hakkak
laboratory process included a step to scrape a thin proteinaceous
layer off the starch pellet; this step was not replicated during the
preliminary pilot-scale process used to generate the starch-milk for
the present study. An alternative explanation is that in the laboratory process the dough was very gently hand-massaged to release
the starch, and the dough remained mostly intact, whereas in the
pilot-scale process the dough was vigorously mechanically agitated
to release the starch. During agitation the dough disintegrated into
small fragments, some of which may have been small enough to
pass through the ne mesh used to separate the starch-milk from
the spent dough, and hence contaminate the starch-milk. Indeed,
the insoluble fraction of the starch-milk was examined using SEM
which revealed that, in addition to the polygonal shaped starch
granules, some other insoluble material was present (Fig. 2), potentially containing protein.

0.8
0.6
0.4
0.2
0.0
0

Volumetric Concentraon Factor (VCF)

Fig. 4. Plot of permeate composition during concentration.

The retention data, as dened by Eqs. (1) and (2), are summarised
in Table 5.
R=1

Cp
CR

(1)

288

P.G. Middlewood, J.K. Carson / Journal of Membrane Science 405406 (2012) 284290

Table 5
Retention and average retention of the various feed components.
Component

Initial retentionA

Average retentionB

Starch
Protein
NSP
Ash
Fat

1.00
0.67
nm
0.05
nm

0.98
0.63
0.14
0.12
0.37

nm, not measured.

A
Calculated using Eq. (1).
B
Calculated using Eq. (2).

Rav =

ln(Cf /C0 )
ln(V0 /Vf )

(2)

where R is the initial retention, CP is the concentrate of the species

in the permeate and CR is the concentration of the species in the
retentate, Rav is the average retention, Cf is the nal concentration
of the species, C0 is the initial concentration of the species, V0 is the
initial volume of the feed and Vf is the nal volume of the feed.
Selected permeate samples were analysed for the presence of
starch using the iodine test, and using SEM. In all cases the result
was negative, indicating 100% retention of the starch granules. High
retention does not always result in a high process yield since product can be lost by membrane fouling and system hold up (i.e.
product remaining in the membrane and associated equipment
after draining). The overall starch yield was measured across a
concentration run (nine-fold volume reduction) and a dialtration
run (six diavolumes). The concentration run had a starch yield of
94%, and the dialtration run a starch yield of 98%. A combined
concentrationdialtration run is therefore expected to have an
approximate starch yield of 92% (i.e. the product of 98% and 92%).
Future research will focus on the application of methods to increase
yield by altering how product is removed from the membrane, e.g.
using an air purge to completely drain the system, or using a water
ush and adding the ush water to the next batch of feed.
The protein retention was 0.67, indicating that while some protein passed through the membrane, some protein was also retained
by the membrane. This is evident in Fig. 3, in which the protein
content increases with volumetric concentration. If all the protein
was passing through the membrane (i.e. a retention of 1) the plot
would be a horizontal line equal to the feed protein level. Although
the protein concentration increased with volumetric concentration, because some protein was passing through the membrane
but the starch was fully retained, the protein content on a dry
basis decreased from 16 to 12%. The measured protein retention
was higher than expected, given that the membrane NMWL was
ten times larger than the molecular weight of the largest protein.
A number of factors can work in isolation or combination to promote high retention. The three-dimensional shape of some of the
proteins may prevent them from passing through the membrane
pores. The proteins may interact with themselves or other feed
components (lipids, non-starch polysaccharides (NSPs)) to form
complexes too large to permeate the membrane. A gel layer may
form on the membrane surface, or a cake layer may develop, both
of which can act as a secondary membrane.
To test what inuence the cake layer was having on protein
retention the cake-forming fraction of the feed was removed by
centrifuging (see Section 2.1) and the resulting claried liquor (Feed
A1) membrane ltrated. The average protein retention was found
to be 0.64. Thus we believe the overall protein retention to result
from the interactions between the membrane and soluble proteins
(rather than the cake and soluble proteins), and the presence of
some insoluble protein in the feed stream.
The high protein retention has two negative consequences:
rstly, assuming all the protein can permeate the membrane, it

increases the amount of washing needed to ensure the nal product (a dry starch powder) has an acceptably low protein level.
Secondly, the protein content of the permeate will be reduced. As
the permeate is a potential co-product, it would be advantageous if
it had as high a protein level as possible. Assuming this co-product
will take the form of a protein concentrate or powder, a high permeate protein content should mean a greater mass of product, and a
lower cost of downstream processing (e.g. concentration and drying). Additionally, the proteins retained by the membrane could
have signicant functionality, which would result in a higher value
co-product if they were included.
The fat content of the permeate was not specically measured
(a prohibitive 500 mL permeate sample would have been needed
to provide enough dry-matter to perform the standard test), which
prevented the fat retention in the permeate from being calculated.
Instead, the average fat retention was calculated, and found to be
0.37. It is possible that instead of permeating the membrane, the fat
formed part of the fouling layer, and as such was removed from the
retentate. This was observed by Sayed-Razavi [16] when concentrating a similar feed stream (soy our extract). Since the fat content
of the permeate was not specically measured, the NSP retention
could not be calculated by difference. Instead, the average retention
was calculated, and found to be 0.14. This indicates that most, but
not all, NSPs were able to permeate the membrane pores. Potential
reasons why some NSPs were retained are the same as those listed
for the protein retention.
The ash retention was very low (0.05). This was expected as
the ash component is comprised of metal salts, all of which are
many orders of magnitude smaller than the membrane pore size.
The retained starch and protein contain small amounts of minerals,
which explain why a small fraction of the ash was retained.

3.3. Dialtration
The concentrated retentate (dialtration feed) from the membrane characterisation trials had a protein content of 12% dry basis
(DB), which was much higher than typical commercial starches
(0.5% DB). The protein in the concentrated retentate was calculated to be approximately 45% soluble and 55% insoluble. By
comparison, starch that was recovered from the starch-milk feed
stream by centrifuging, and then washed in distilled water (Feed
A2), had a protein content of 4% DB. This value represented the
amount of insoluble protein present, was considered the lowest
level obtainable from the pilot-scale Al-Hakkak process (at this
stage in its development), and was the target of the dialtration
trial; even though it was higher than the 0.1% DB achieved by
Al-Hakkak and Al-Hakkak [7] during their laboratory-based studies. Possible reasons for this difference in protein concentrations
were discussed in Section 3.1. Based on the protein retention coefcient of 0.67, ve diavolumes were expected to lower the protein
content from 12 to 4% DB. As a safety-factor an extra diavolume was added. The protein content of the retentate decreased
with increasing diavolumes (Fig. 5). This decrease attened out
with successive diavolumes, indicating that protein retention was
increasing (because the amount of insoluble protein relative to
soluble protein was increasing). The dry-basis composition of the
starch-milk before and after the concentration and dialtration
steps, and the composition of the permeate generated during the
concentration step, are summarised in Table 6.
Dialtration lowered the protein, fat, and ash content of the
retentate, however the protein and fat levels were not reduced
enough to reach commercially acceptable levels. The high protein
retention, and inability of dialtration to reach an acceptably low
protein level, was in part due to the presence of some insoluble
protein in the starch-milk.

P.G. Middlewood, J.K. Carson / Journal of Membrane Science 405406 (2012) 284290

289

Table 6
Dry-basis composition data.
Sample

Protein (%DB)

Ash (%DB)

Fat (%DB)

NSPA (%DB)

Starch (%DB)

Feed
Concentrate
Permeate
Dialtration feed
Dialtered liquor

15.8
11.7
22
12.3
3.3

4.9
1.0
11.2
1.1
0.3

4.4
1.8
nm
1.8
1.4

24.2
6.2
66.8
1.2
3.0

50.6
79.2
0.0
83.4
98.0

nm, permeate fat level not tested, any fat present will be included with the NSP.
A
Calculated by difference.

3.4. Flux characteristics

Plotting ux against time (while running in recycle mode) gives
an indication of the fouling potential of the membrane. The data
from ve trials are presented in Fig. 6. The ux, J(t) was modelled
using Eq. (3)
J(t) =

J0

(3)

(1 + t)n

with values of 60 L m2 h1 for the initial ux (J0 ), 240 for , and

0.27 for n. The initial ux decrease is common during tangential
ow ltration processes. It is generally caused by concentration
polarisation in ultraltration processes, and cake formation in
microltration processes. The extent and time dependence of the
ux decrease are largely inuenced by the interactions between
the feed stream and membrane; as such they vary from process to
process. For the process studied here, the quasi-steady state ux
occurred after 23 h, and was signicantly lower than the start-up
14

Protein content (% DB)

12
10
8
6
4
2
0
0

Diavolumes

Fig. 5. Plot of protein content against number of diavolumes. TMP 70 kPa, feed
20 L h1 , and temperature 25 C.

ux. Past the 1 h mark the ux continued to decline, but much more
gradually. Davis [17] has attributed this gradual decrease, for feedstocks containing particulates, to cake consolidation, compaction,
or fouling. The combination of a large rst-stage ux decrease, plus
a continual second-stage ux decrease, indicates that fouling is
signicant. The steady-state ux (10 L m2 h1 ) is low when compared to the values (4070 L m2 h1 ) reported by other authors
[16,1820], albeit they used different membranes and different
starch solutions.
The effect of trans-membrane pressure (TMP) on ux, for a
typical run, is shown by Fig. 7. Data were collected for the unconcentrated (VCF1) starch-milk, and after a two-fold concentration
(VCF2). For the unconcentrated starch-milk, ux increased nonlinearly with TMP. This indicates the resistance of the gel-layer,
or caked material, increased with increasing pressure, i.e. either
its thickness increased (due to a decrease in back-transport), or
the gel, or caked material, was compressible. Further evidence
that compression was occurring is the hysteresis between the a
(increasing pressure) and b (decreasing pressure) measurements,
particularly on the rst set of trials (VCF1). When the TMP was lowered to a previous value (e.g. from 200 to 150 kPa) the ux did not
return to the value recorded previously at that TMP. This suggested
the foulant had compressed and did not relax when the pressure
was removed. It is also possible that increasing the TMP accelerated fouling by forcing material into the membrane pores. The
ux versus TMP relationship for liquor that had been concentrated
two-fold (VCF2) was closer to linearity, and had less hysteresis.
However, the irreversible cake compression, or accelerated fouling, during the VCF1 evaluation could have inuenced the shape of
the subsequent VCF2 curves.
The effect of feed rate on ux was evaluated over a range of
TMPs between 50 and 200 kPa. Due to equipment pressure limitations it was not possible to test the ux at each TMP for all the
feed rates. The results indicated that ux was almost independent
of feed rate. If anything it decreased slightly with increasing feed
rate, which contradicts typical ux-feed rate trends. The fact that
ux did not increase with feed rate could indicate that the increase
in velocity was not enough to reduce the thickness of the caked

70
RunA

60

RunB

RunC

RunD

RunE

14

Modelled data

VCF1-a (unconcentrated feed, TMP

incrementally increased)

50

Flux (L m- h-1)

Flux (L m- h-1)

12
40
30
20

10
VCF1-b (unconcentrated feed, TMP
incrementally decreased)

8
6

VCF2-a (feed concentrated 2-fold,

TMP incrementally increased)

10

VCF2-b (feed concentrated 2-fold,

TMP incrementally decreased)

2
0
0

Time (h)
Fig. 6. Relationship between ux and time for ve identical runs. Data collected
while running in recycle mode; TMP 100 kPa, feed 20 L h1 , and temperature 25 C.

0
0

50

100

150

200

TMP (kPa)
Fig. 7. Inuence of TMP on ux. Feed 20 L h1 and temperature 25 C.

290

P.G. Middlewood, J.K. Carson / Journal of Membrane Science 405406 (2012) 284290

material. During the centrifuging step performed when preparing

Feed A1, the starch granules were observed to exhibit nonNewtonian behaviour, i.e. the settled starch formed a pseudo-solid.
A velocity higher than what was possible using the current equipment may be required to re-suspend any settled starch granules.
Alternatively it could indicate that the cake thickness was reduced,
but it did not result in an increased ux as the gel-layer (if assumed
to be closer to the membrane surface) was undisturbed by the
increased velocity. Where the gel layer formed will depend on how
restrictive the cake layer was. If the gel-forming material could pass
through the cake voids it would form on the membrane surface, if
not it would form on the cake surface. Flux versus time data from
Feed A1 and Feed A2 showed that the soluble feed components (gelforming) were the main contributor to the initial rapid ux decline
(Fig. 6, 01 h), while the insoluble feed components (cake-forming)
were the main contributor to the more linear ux decline (Fig. 6,
13 h).
4. Conclusions
The Millipore membrane successfully retained the starch granules and the retention of the non-starch polysaccharides and ash
components of the feed was acceptably low. However, the retention
of protein and fat was higher than desired. Dialtration lowered the
protein and fat content of the retentate, but not far enough to reach
commercially acceptable levels. The high protein retention, and
inability of dialtration to reach an acceptably low protein level,
was in part due to the presence of some insoluble protein in the
starch-milk. While operating in recycle mode ux declined signicantly with time; indicating severe membrane fouling. This fouling
appeared mainly to be caused by the soluble feed components. Flux
increased with increasing TMP, but the increase was non-linear and
indicated that the foulant (or caked material) was compressible.
The optimum TMP was 100150 kPa. Flux was almost independent
of feed rate across the range trialled; however, equipment pressure
limitations prevented a thorough investigation into using higher
feed rates.
Acknowledgement
This work was funded by the Biopolymer Network Ltd. (New
Zealand).

References
[1] N. Lindeboom, P. Chang, R. Tyler, Analytical, biochemical and physicochemical aspects of starch granule size, with emphasis on small granule starches: a
review, Starch/Strke 56 (2004) 8999.
[2] R. Hoseney, Principles of Cereal Science and Technology, 2nd edition, American
Association of Cereal Chemists Inc, Minnesota, USA, 1994.
[3] B. Salcedo-Chavez, J. Osuna-Castro, F. Guevara-Lara, J. Domnguez-Domnguez,
O. Paredes-Lpez, Optimization of the isoelectric precipitation method to obtain
protein isolates from amaranth (Amaranthus cruentus) seeds, J. Agric. Food
Chem. 50 (2002) 65156520.
[4] R. Bressani, L. Garcia-Vela, Protein fractions in amaranth grain and their chemical characterisation, J. Agric. Food Chem. 38 (1990) 12051209.
[5] E. Tosi, E. Re, H. Lucero, R. Masciarelli, Dietary bre obtained from amaranth
(Amaranthus cruentus) grain by differential milling, Food Chem. 73 (2001)
441443.
[6] R. Repo-Carrasco-Valencia, J. Pena, H. Kallio, S. Salminen, Dietary ber and
other functional components in two varieties of crude and extruded kiwicha
(Amaranthus caudatus), J. Cereal Sci. 49 (2009) 219244.
[7] J. Al-Hakkak, F. Al-Hakkak, New non-destructive method using gluten to isolate starch from plant materials other than wheat, Starch/Strke 59 (2007)
117124.
[8] R. Coulson, J. Richardson, Membrane separation processes, in: Coulson &
Richardsons Chemical Engineering, vol. 2 (Particle Technology and Separation
Processes), Pergamon, Oxford, 1993, pp. 868901.
[9] L. Zeman, A. Zydney, Microltration and Ultraltration Principles and Applications, Marcel Decker Inc, New York, NY, 1996.
[10] C. Liu, S. Caothien, J. Hayes, T. Caothuy, T. Otoyo, T. Ogawa, Membrane Chemical
Cleaning; From Art to Science, Pall Corporation, Port Washington, New York,
2006, http://www.pall.com/pdf/mtcpaper.pdf (accessed 14.07.09).
[11] R. MacManus, R. Macdonald, Luqum Process: Amaranth 2 Quasi-Continuous
Pilot-Scale Trials, AgResearch Ltd., Client Report SC110/2009/65, 2009.
[12] J. Wagner, Membrane Filtration Handbook Practical Tips and Hints, Osmonics
Inc., 2001.
[13] X. Kong, J. Bao, H. Corke, Physical properties of amaranthus starch, Food Chem.
113 (2009) 371376.
[14] Millipore BioProcess Division, Maintenance Procedures: PelliconTM and
PelliconTM 2 Cassette Filters. www.millipore.com (accessed 09.12.09).
[15] P.G. Middlewood, Extraction of amaranth starch from an aqueous medium
using microltration, M.Sc. Tech Thesis, University of Waikato, New Zealand,
2011.
[16] S. Sayed-Razavi, J. Harris, F. Sherkat, Fouling and cleaning of membranes in the
ultraltration of the extract of soy our, J. Membr. Sci. 114 (1996) 93104.
[17] R. Davis, Modeling of fouling of crossow microltration membranes, Sep. Purif.
Methods 21 (1992) 75126.
[18] A. Hinkova, Z. Bubnik, V. Pour, S. Henke, P. Kadlec, Application of cross-ow
ultraltration on inorganic membranes in purication of food materials, Czech.
J. Food Sci. 23 (2005) 103110.
[19] R. Shukla, R. Tandon, M. Nguyen, M. Cheryan, Microltration of starch suspensions using a tubular stainless steel membrane, Membr. Technol. 2000 (2000)
58.
[20] N. Singh, Z. Chen, N. Tomer, R. Wickramasinghe, N. Soice, S. Husson, Modication of regenerated cellulose ultraltration membranes by surface-initiated
atom transfer radical polymerization, J. Membr. Sci. 311 (2008) 225234.