Journal of Membrane Science 208 (2002) 157169

Application of fluidised particles as turbulence

promoters in ultrafiltration
Improvement of flux and rejection
T.R. Noordman a, , A. de Jonge a , J.A. Wesselingh a , W. Bel b ,
M. Dekker b , E. Ter Voorde b , S.D. Grijpma b
a

Department of Chemical Engineering, University of Groningen, Nijenborgh 4, 9747 AG Groningen, The Netherlands
b Unilever Research Laboratories, Olivier van Noortlaan 120, 3133 AT, Vlaardingen, The Netherlands
Received 27 October 1997; received in revised form 31 May 1999; accepted 9 May 2002

Abstract
To prevent fouling of ultrafiltration membranes during processing of protein solutions, a high degree of turbulence should
be introduced in the feed solution, keeping the energy consumption as low as possible. For this purpose, the application of
fluidised beds at the upstream side of the membrane can be very attractive. In this study, the effect of fluidised particles is
investigated under non-fouling and fouling conditions. In both systems, mass transfer is increased considerably by addition
of fluidised particles. The main mechanism of fouling reduction is by an increased diffusion of rejected solutes from the
membrane interface back to the bulk solution. Improvement of mass transfer is only a weak function of particle size and
density. Therefore, the particles can be chosen as light and small as possible. This greatly reduces the energy consumption in
the system and the risk of membrane damage, which may be a problem. Application of fluidised bed systems can be especially
attractive in highly viscous systems in which energy consumption can be a problem.
2002 Elsevier Science Ltd. All rights reserved.
Keywords: Membranes; Fouling; Fluidised beds; Mass transfer; Energy consumption

1. Introduction
A serious problem in ultrafiltration processes is fouling of the membranes. This can be due to adsorption
of solutes on outer membrane and inner pore surfaces,
blocking of pores by rejected solutes or formation of
a thick cake layer of precipitated solutes on top of
the membrane. The latter type of fouling is predomi Corresponding author. Present address: Heineken Technical
Services, P.O. Box 510, 2380 BB Zoeterwoude, The Netherlands.
Fax.: 31-71-5457800.
E-mail address: t.r. noordman@heineken.nl (T.R. Noordman).

nantly caused by concentration polarisation occurring

in the liquid boundary layer in front of the membrane.
Concentration polarisation, and thereby fouling, depends on the mixing of the liquid feed. This mixing
can be improved by various means. In cross-flow filtration, the classical way is by increasing the crossflow velocity. Also, static turbulence promoters have
been used [13]. However, these methods also introduce a large pressure drop across the membrane
modules, resulting in high energy consumption. An
alternative is adding fluidised particles. These can do
their job in two ways: mixing up the liquid boundary layer in front of the membrane and scouring of an

0376-7388/02/$ see front matter 2002 Elsevier Science Ltd. All rights reserved.
PII: S 0 3 7 6 - 7 3 8 8 ( 0 2 ) 0 0 2 2 9 - 6

158

T.R. Noordman et al. / Journal of Membrane Science 208 (2002) 157169

Nomenclature
C
CD
dp
D
Dt
E
f
g
H
k
J

(kg/m3 )

concentration
drag coefficient
particle diameter (m)
diffusivity (m2 /s)
tube diameter (m)
rate of energy consumption (J/s)
friction factor
gravity constant (m/s2 )
height of the fluidised bed
mass transfer coefficient (m/s)
volume flux through the
membrane (m/s)
m
mass (kg)
Mn
number average molecular weight
P
pressure (bar)
R
rejection, fraction of solute that
is retained
TMP transmembrane pressure drop (bar)
v
velocity (m/s)
z
co-ordinate in boundary layer (m)
Re
Reynolds number in empty tube
flow (vcross Dt /)
Reh Reynolds number used in relation
of Smith, Eq. (8) ((vsf /)dh /)
Reynolds number at terminal velocity
Ret
of a falling particle (vt dp /)
Sc
Schmidt number (/D)
Sh
Sherwood number (kDt /D)
Greek

letters
boundary layer thickness (m)
porosity of the fluidised bed
viscosity (Pa s)
density (kg/m3 )
kinematic viscosity (m2 /s)

Subscripts
b
bulk
calc calculated
cross cross-flow
exp experimental
h
hydraulic
diss dissipated
fb
fluid bed
gel
concentration at gelation point

l
obs
p
true
s
sf
t
w

liquid
observed
permeate
true rejection, intrinsic property
of the membrane
solid
superficial
terminal
wall
in unbound liquid

already formed fouling layer by collisions. The potential of preventing fouling by using fluidised particles
was shown by Meijer [4,5]. He used them to prevent
scale formation in tubes of heat exchangers. Fluidised
particles were successfully applied to ultrafiltration
processes by van der Velden et al. [6,7] and Rios et al.
[8]. Van der Velden found a considerable increase in
mass transfer in the boundary layer during hyperfiltration of salts, when varying the cross-flow (and thereby
the bed porosity) with a constant transmembrane pressure drop. He fluidised glass beads with diameters
ranging from 0.4 to 2 mm. The increase in mass transfer could be predicted by an empirical mass transfer relation determined for fluid beds with non-porous walls
by Smith [9]. He also did a few ultrafiltration experiments using a high molecular weight polyethylene glycol (PEG4000000) to study improvement of flux under
gelation conditions. In his work, a major disadvantage
of applying fluidised particles came to light. SEM pictures of the polymeric ultrafiltration membranes used,
showed severe damage caused especially by the bigger
particles, those that removed the gel layer completely.
From these observations, it can be concluded that the
membrane should be protected from impacts of particles by at least a small amount of fouling. We also
encountered membrane damage under certain conditions. In our future work, we will go into this subject
in detail. Rios et al. [8] used inorganic membranes and
did not notice any damage of their membranes, despite the brittle nature of inorganic membranes. They
filtered gelatine solutions and assumed to be working
in a gel regime. They found considerable improvements on flux and modest changes in rejection of gelatine. In this study, the influence of fluidised particles
on mass transfer in the boundary layer is determined

T.R. Noordman et al. / Journal of Membrane Science 208 (2002) 157169

with aqueous solutions of both, a medium sized solute

(PEG10000) and a large solute (bovine serum albumin). This is done by measuring fluxes and rejections
at various transmembrane pressure drops.

2. Theory

where Cw , Cp , and Cb are the concentration of solute

at the membrane interface, in the permeate stream and
in the bulk, respectively, k the mass transfer coefficient
in the boundary layer. The degree of separation of a
solute from the solution is expressed in a rejection
coefficient, which is defined as:
Robs = 1

2.1. Ultrafiltration theory

Ultrafiltration is driven by a pressure gradient. This
induces convective transport of bulk liquid to the
membrane interface. At this interface, some of the
solutes are (partly) rejected. Their concentrations at
the interface build up until they get high enough to
give a diffusive flux back to the bulk solution. This
back diffusion is balanced by the convective transport
of solutes to the interface and the flux of the solutes
through the membrane. The increase in concentration
at the membrane interface is called concentration
polarisation and is visualised in Fig. 1.
A steady state mass balance over the liquid boundary layer results in the following equation:
JC = JCp D

C
z

(1)

Integration of Eq. (1) with boundary conditions, C =

Cb at z = 0 and C = Cw at z = , yields the
well-known concentration polarisation equation [10]:



Cw C p
J = kln
(2)
Cb C p

159

Cp
Cb

(3)

This Robs is influenced by concentration polarisation

effects. To correct for these, a true rejection, Rtrue is
defined according to:
Rtrue = 1

Cp
Cw

(4)

These two can be interrelated with the help of Eq. (2):





1 Robs
J
1 Rtrue
ln
= + ln
(5)
Robs
k
Rtrue
At low permeate fluxes, Rtrue increases with increasing flux (and thus with increasing pressure drop). At
sufficiently high fluxes, the Rtrue approaches a constant value, while the Robs decreases with increasing
flux due to concentration polarization effects. Under
these circumstances, mass transfer coefficients can be
obtained from a plot of ln(1 Robs /Robs ) versus J at
constant feed velocity (Appendix A).
The concentration at the membrane interface can
become quite high. Under certain circumstances, it
can even exceed the solubility of the solute. In that
case, a deposition layer will be formed upon the membrane surface, as depicted in Fig. 1b. This is a common

Fig. 1. (a) Concentration polarisation in ultrafiltration. A concentration gradient is built up over the liquid boundary layer, ; (b) formation
of a gel layer on the membrane. The concentration of solute at the membrane interface reaches a constant value, which is equal to the
solubility of the solute.

160

T.R. Noordman et al. / Journal of Membrane Science 208 (2002) 157169

phenomenon in the ultrafiltration of protein solutions.

The deposit layer increases the hydraulic resistance
of the membrane and reduces the permeate flux. The
concentration at the interface attains a constant value,
which is called the gel concentration, Cgel . The steady
state flux will be independent on pressure drop; an increase of pressure will temporarily increase the flux,
but this will also result in transport of solute to the interface where it will deposit. A pressure increase will
merely increase the thickness of the deposition layer.
If the solute is completely rejected, we can derive the
following expression from Eq. (2), for the flux in this
regime:



Cgel
J = kln
(6)
Cb
2.2. Estimating mass transfer coefficients
The mass transfer coefficient in the liquid boundary
layer, k can be estimated from empirical mass transfer
relations presented in literature. For empty tubes operated in the turbulent regime, the relation of Dittus
Boelter is often used [11]:
Sh = 0.023Re0.8 Sc0.33

(7)

For fluidised beds, various empirical relations have

been given for heat or mass transfer. In this article, we
compare the experimental results to the mass transfer relations determined by Smith and King [9] and
Yasunishi et al. [12]. The relation of Smith for larger
particles (diameter ratio dp /Dt < 0.06) is:
Sh = 0.455Reh 0.56 Sc0.33

(8)

where Reh is equal to (vsf /)dh / and dh is defined

Dt
as: 1+1.5(1)(D
.
t /dp )
The relation of Yasunishi relates the Sh number
directly to the rate of energy consumption in the fluidised bed:
 1/3 4/3 3/4
Ediss Dt
Sh = 0.13
Sc1/3
(9)
1/3
ml
An expression for the amount of energy dissipated in
fluidised beds is given in Eq. (11) later. The relation of
Yasunishi holds for liquidsolid, gasliquid and three
phase systems. All relations presented earlier can be

corrected for viscosity gradients in the boundary layer

[11], resulting in the following general form of the
mass transfer relations:

0.11
b
(10)
Sh = AReb Scc
w
2.3. Energy dissipation
We want to compare the effect of fluidised particles
on mass transfer with empty tube cross-flow operation.
Therefore, it is best to relate mass transfer coefficients
in both systems to the energy input that is required. The
energy dissipation due to the presence of the fluidised
particles is equal to:

Ediss,fb = Dh2 vsf gH(1 )(s l )

(11)
4
Besides this dissipation, energy is also dissipated due
to friction with the wall, this amount is equal to:
Ediss,w =

2
2 H
Dh vsf Pw with P = 4f 21 vsf
4
Dh
(12)

The friction factor f can be calculated with the equation of Blasius, assuming that the tube has a smooth
surface. The friction loss at the wall is normally small
compared to the friction in the bed of particles. With
the Eq. (12), the energy dissipation in empty tube
cross-flow filtration can also be calculated.
2.4. Fluidisation velocity
The terminal falling velocity, i.e. the falling velocity of a single sphere in an unbounded liquid can be
obtained from the force balance:
3

2
dp (s l )g = CD dp2 21 l vt,
(13)
6
4
The friction factor CD is a function of the particle
Reynolds number at the terminal falling velocity,
Ret CD is equal to 24/Ret for Ret < 1 and equal to 0.43
for Ret > 1000. For intermediate Reynolds numbers,
Lali [13] gives the following relation:
24
CD =
(1 + 0.15Re0.687
(14)
t, )
Ret,
In a bounded tube, the falling velocity is reduced by the
presence of the tube wall. The correction of Richardson and Zaki can be applied [14]:
vt = vt, 10dp /Dt

(15)

T.R. Noordman et al. / Journal of Membrane Science 208 (2002) 157169

The dependency of bed porosity on superficial velocity

can be described by the relation:
vsf
(16)
n =
vt
For the exponent, n, the relation of Garside and Al
Dibouni [15] was used:
5.1 n
(17)
= 0.1Re0.9
t
n 2.7

161

3. Experimental
3.1. Materials
Polyethylene glycol (Sigma) was used with an average mass of 10,000 Da (PEG10000). Solutions of
2% were made in demineralised water. Diffusivity
and viscosity data were obtained from literature. The

Fig. 2. Ultrafiltration set-up.

162

T.R. Noordman et al. / Journal of Membrane Science 208 (2002) 157169

diffusion coefficient is 7.9 1011 m2 /s [16]. The viscosity was estimated from the following relation:
sp
water
(18)
= [] + 0.33[]2 C, sp =
C
water
The intrinsic viscosity was estimated from [17]:
[] = 6.04 105 Mn0.9

(19)

Albumin solutions were prepared by dissolving 10 g/l

BSA Fraction V (Boehringer Mannheim) in 0.15 M
NaCl (Merck) solution. The pH was adjusted to pH 7.0
with NaOH (Merck). The viscosity of BSA solutions
was determined with the following relation, given by
Kozinski and Lightfoot [18]:
103
=
1.11 0.00542C + 6.71 106 C 2

(20)

The diffusivity of BSA in 0.15 M NaCl solution at pH

7.4 was determined by Trettin and Doshi [19] to be
6.9 1011 m2 /s.
Four different kinds of particles were used for
fluidisation. Glass beads used were obtained from
P.M. Tamson B.V. (Zoetermeer, The Netherlands)
with mean particle diameters of 0.46 and 1.04 mm
(s = 2900 kg/m3 ). Stainless steel beads with diameters of 1.0 and 2.0 mm (s = 7800 kg/m3 ) were
purchased from INA Naaldlager Mij B.V. (Barneveld,
The Netherlands).
3.2. Equipment and measurements
Experiments were carried out in a configuration as
shown in Fig. 2. Tubular polysulfone membranes were
used of type Stork Wafilin WFBX0121, with an inner
diameter of 14.4 mm and 1.75 m length and a NMWC
of 10 kD. Solutions were fed to the membrane module using a Jabsco S2 gear Lobe pump and a Verder
2035 gear pump. In experiments with a fluidised bed,
the particles were added as soon as the permeate flow
was stabilised. The feed velocity was adjusted such
that the particles were visible just above and below
the membrane module to ensure that the membrane
tube is completely filled with a fluidised bed. Pressures on the feed and concentrate side were measured
using Econosto pressure gauges. The pressure at the
permeate side was assumed to be atmospheric. During experiments permeate fluxes were measured gravimetrically. The permeate solution was fed back to

the feed solution periodically to keep feed concentration constant. Concentrations of PEG10000 or BSA
in both concentrate and permeate streams were measured. Concentrations of PEG10000 were determined
by measuring the refractory index of the solution,
those of BSA by measuring UV adsorption at 280 nm.
The rejection of BSA turned out to be always larger
than 98% in all experiments, so 100% rejection was
assumed. All experiments were carried out at 25 C.

4. Results
4.1. Velocity voidage relationship
The relation between bed porosity and superficial
velocity was measured for the different particles used
in water at 25 C. The results are presented in Fig. 3.
It is seen that the experimental data can be welldescribed by Eq. (16), with n determined from Eq. (17)
and vt from Eqs. (13)(15).
4.2. Solutions of PEG10000
In Fig. 4ac results are presented of experiments
without fluidised particles. At two cross-flow velocities (0.29 and 1.0 m/s) the transmembrane pressure
drop was varied. In Fig. 4a, the fluxes are plotted
against the transmembrane pressure drop; in Fig. 4b,
the observed rejection against the filtrate flux. In
Fig. 4c, the results are plotted according to Eq. (5),
to obtain the true rejection of the membrane and the
mass transfer coefficient. The flux was divided by v 0.8

Fig. 3. Comparison between experimental data and theoretical

relation between bed porosity and superficial velocity.

T.R. Noordman et al. / Journal of Membrane Science 208 (2002) 157169

163

Fig. 4. Flux (a) and rejection; (b) of PEG10000 in empty tube experiments at varying transmembrane pressure drop and cross-flow
velocities; (c) ln(1 Robs )/Robs vs. the permeate flux, corrected for cross-flow velocity.

to compare experiments at two different cross-flow

velocities.
The graphs clearly show the effect of concentration polarisation; at increasing filtrate flux the rejection is decreased. At lower cross-flow velocity, this
effect is more severe. Also, the flux is influenced by
the cross-flow velocity, another effect of concentration
polarisation. From Fig. 4c, the value of Rtrue was de-

termined to be 0.84 by linear regression. The slope of

the curve was found to be 2.24105 , which is slightly
higher than predicted from the DittusBoelter Eq. (7),
1.93 105 .
Experiments with fluidised particles are presented
in Figs. 5 and 6. In Fig. 5, experiments with fluidised
beds of 5 and 11% steel particles of 2 mm diameter
are shown. The results are compared to an empty tube

Fig. 5. Flux (a) and rejection; (b) of PEG10000 in the presence of fluidised steel particles of 2 mm compared to the empty tube experiment
with a cross-flow velocity of 0.29 m/s.

164

T.R. Noordman et al. / Journal of Membrane Science 208 (2002) 157169

Fig. 6. Effect of particle size on flux (a) and rejection; (b) of PEG10000. Bed porosity is 0.89.

experiment with a cross-flow velocity comparable to the superficial velocities in the fluidised bed
experiments.
It is clearly shown that both flux and rejections are
increased by the particles. Both effects are a result of
better mixing of the feed liquid. This leads to lower
concentrations of PEG at the membraneliquid interface. Hereby, the osmotic pressure loss is reduced,
as is seen by the higher fluxes and the rejections are
closer to the intrinsic values for the membrane. We see
a small effect of particle concentration; the rejections
with 11% particles are a little higher than with 5%, as
are the fluxes.
The effect of particle size is shown in Fig. 6. Neither fluxes nor rejections are influenced by particle
size. The same degree of mixing is obtained. Plotting
Eq. (5) yields empirical mass transfer coefficients for
the different fluidised beds. This is depicted in Fig. 7.
The plots can all be drawn to the same intercept (i.e.
Rtrue = 0.84), although the experiments with beds at

higher concentration (11%) begin to deviate at low

fluxes.
In Table 1, all the results obtained with PEG10000
are summarised. The experimentally determined mass
transfer coefficients are compared with those predicted
from relations (8) and (9). Also, the rate of energy dissipation in one membrane tube is calculated. Application of fluidised particles result in higher mass transfer
coefficients. If we compare mass transfer coefficients
at equal feed velocity, we obtain improvements up to
five times. They can also be compared at equal energy
consumption rate. The results obtained with a bed of
11% 1 mm steel particles can best be compared with
the empty tube experiment at a cross-flow of 1 m/s. At
the same energy input, the mass transfer coefficient is
increased by a factor 2. The mass transfer coefficients
measured are in good agreement with the empirical
relations of Smith and Yasunishi.
4.3. Experiments with protein solutions
In Fig. 8, the effect of fluid beds on fluxes is presented in concentration experiments. It can be seen
Table 1
Mass transfer coefficients in experiments with PEG solutions

Fig. 7. The ln(1 Robs )/Robs vs. the permeate flux, without and
with fluidised particles.

dp
(mm)

vcross
(m/s)

Ediss
(J/s)

kexp
(m/s)

kcalc
(m/s)

2
2
1

1
1
0.95
0.89
0.89

1
0.29
0.30
0.28
0.17

0.35
0.01
0.35
0.59
0.36

4.5
1.7
6.2
8.6
8.8

5.1
1.9
7.4a ; 6.5b
8.6a ; 7.8b
7.9a ; 6.9b

a
b

Calculated with the relation of Smith (8).

Calculated with the relation of Yasunishi (9).

T.R. Noordman et al. / Journal of Membrane Science 208 (2002) 157169

Fig. 8. Flux vs. logarithmic bulk concentration of protein solutions

in experiments with fluidised glass particles (1 mm) and without
particles.

that all curves tend to cut the x-axis at a protein concentration of 250 g/l. According to the gel layer model,
this is the concentration at the membrane interface at
which the protein starts to precipitate. This concentration is the same as was found by van Oers [20]. The
osmotic pressure of a protein solution of 250 g/l at pH
7.4 is 0.8 bar [21] and this reduction in effective transmembrane pressure is insufficient for the flux reduc-

165

tion obtained in all experiments with protein (clean

water flux of membranes is 300 l/m2 h at 3 bar TMP).
So, the flux appears to be limited by gel formation or
fouling of the membrane. We see that fluxes are increased by the presence of the fluid bed and that fluxes
are higher at higher particle concentrations. From the
slope of these plots, we can determine mass transfer
coefficients according to Eq. (6). Again, these are comparable with those predicted by Eq. (7) for the empty
tube experiment and by the relation of Smith, Eq. (8),
for the fluid bed experiments, after correction for the
viscosity at the membrane interface (Eq. (10)). Coefficients predicted by the relation of Yasunishi, Eq. (9),
are a little lower. Thus, the observed flux increases
can, for the larger part, be explained by an increase in
mass transfer. The main mechanism of fouling reduction is by augmentation of back diffusion.
In Fig. 9ad, the effect of the particles is shown
at constant protein concentration. The small periodic
flux decays that are shown in some plots are due to
a slight increase in concentration of the feed solution in time (from 1 to 1.5%). Fluxes were restored
after the permeate solution was fed back to the feed
solution, periodically. In Table 2, the mass transfer

Fig. 9. Effect of different kinds of particles on flux of BSA solutions.

166

T.R. Noordman et al. / Journal of Membrane Science 208 (2002) 157169

Table 2
Mass transfer coefficients in experiments with BSA solutions
s

dp
(mm)

vcross
(m/s)

Ediss
(J/s)

kexp
(m/s)

kcalc
(m/s)

2900
2900
2900
7800
7800

1
1
0.46
1
2

1
0.91
0.78
0.78
0.87
0.87

0.079
0.10
0.060
0.029
0.18
0.27

0.0003
0.052
0.073
0.035
0.45
0.71

0.54
4.3
6.4
5.9
7.6
8.1

0.60
4.5a ;
5.3a ;
4.7a ;
7.2a ;
7.6a ;

a
b

3.7b
4.3b
3.1b
5.7b
6.3b

Calculated with the relation of Smith (8).

Calculated with the relation of Yasunishi (9).

coefficients are given. These are calculated from the

experimental fluxes with Eq. (6), assuming a constant
gel concentration of 250 g/l. After a stationary flux
was reached in an empty tube, fluidised particles were
added and the flux increased within only a few minutes. This time-scale indicates that flux increase is
merely due to the effect of mixing up the boundary
layer, and scouring effects will not play an important
role. The absolute fluxes in the presence of fluidised
particles increase only slightly with increasing particle size and particle density. This is in agreement with
the results obtained with PEG10000 (Fig. 6). which
showed that mass transfer is hardly affected by particle size. These results have important consequences,
as the risk of membrane damage increases dramatically with increasing particle size or density. Besides
that, the energy consumption in the system is also
strongly increased. Therefore, the best choice of particles would be as small and as light as possible, as they
can provide quite acceptable fluxes with a low energy

consumption and low risk. With the use of the heaviest particles, these risks come clear to the eye. The
concentration polarisation layer of protein is unable to
protect the membrane against the impacts of the 2 mm
steel particles and permeate flux decreases in time.
In the results presented, it was shown that by adding
fluidised particles the fluxes are increased enormously
compared to empty tube experiments at the same feed
velocity, by up to a factor of 13. For practical considerations, it is interesting to compare the mass transfer
coefficients obtained in fluid bed and empty tube systems at equal energy consumption rates. This is done
in Fig. 10. The experimental data are compared to predictions with the equation of Smith (9) and Dittus and
Boelter (7). The effect of particle size is considered at
constant bed porosity. In the calculations, the particle
size was varied from 0.5 to 2 mm.
It is clear that at low energy consumption, a fluid
bed system is favoured over empty tube systems. Increases of mass transfer coefficients can be more than
a factor 2, at equal energy consumption rate. At higher
cross-flow velocities, this advantage is reduced. As
the application of fluidised beds also has two disadvantages, the risk of membrane damage and the lower
flexibility of the system, its use should be restricted to
the application in systems in which energy dissipation
is a problem. When highly viscous liquids are filtered,
energy consumption can be quite high and then application of fluidised particles is favoured over increasing
the cross-flow velocity.
We have carried out some calculations on the effect
of viscosity on mass transfer coefficients in both fluidised bed and cross-flow systems. These results are

Fig. 10. Comparison of mass transfer coefficients to energy consumption in empty tube and fluid bed systems, for BSA (a) and PEG10000;
(b) solutions. Symbols indicate experimental data, drawn lines are estimations based on the equations of Smith (8) and DittusBoelter (7).

T.R. Noordman et al. / Journal of Membrane Science 208 (2002) 157169

167

Appendix A. R true as function of the Flux

In this appendix, justification is given for the
assumption of constant value for Rtrue at the volume fluxes in the experimental work on PEG10000
presented here.
According to the hydrodynamic theory of membrane filtration [22,23], the actual rejection can be
related to both, the volume flux and the ratio of solute
diameter to pore diameter, :
Fig. 11. Calculation of the rate of energy consumption, per kilogram of liquid, needed to maintain k at 5 m/s in fluid bed and
empty tube systems at increasing relative viscosity. Calculations
were done for fluid beds of glass and steel particles. Porosities
varied from 0.95 down to 0.6 at increasing relative viscosity.

presented in Fig. 11, in which the energy consumption

is plotted against the viscosity of the solution, relative to water. The energy consumption is calculated to
maintain a mass transfer coefficient of 5 m/s (D =
6.91011 m2 /s), based upon the relation of Smith (8)
and DittusBoelter (9). Calculations were carried out
for the four different particles that were used before in
this work. Bed porosities were varied from 0.95 to 0.6.
It is clearly shown that fluidised beds are especially
attractive for liquids with high viscosity. At a relative
viscosity of 10, the energy consumption in a fluid bed
system is predicted to be 35 times lower than in an
empty tube system. So, in processing viscous liquids
application of fluidised particles as turbulence promoters is attractive, especially because the risk of membrane damage is reduced with increasing viscosity.
5. Conclusions
Both rejection and fluxes can be improved considerably by the use of fluidised particles. The fluxes and
rejections obtained were at the most a weak function
of particle properties such as size and density. The best
choice would be to use particles as light and small as
possible, as this minimises the energy consumption in
the process and also the risk of membrane damage.
The application of fluidised beds is especially attractive for viscous liquids. The main mechanism of flux
improvement is the better mixing of the liquid in front
of the membrane.

Rtrue = 1

(1 )2 Kc exp (Pem )
(1 )2 Kc 1 + exp (Pem )

The membrane Peclet number, Pem is the ratio

between diffusive and convective transport through
the membrane and is given by:




Kc
(jv /)
Pem =
Kd
D
In this equation, is the porosity of the membrane.
The constants Kc and Kd are the convective and diffusive hindrance factors, respectively and are related
to the ratio of solute to pore diameter, . This relation
is based upon simulations done by Bungay and Brenner [24] and translated into these hindrance factors by
Deen [22]:
Kc =

(2 (1 )2 )Ks
2Kt

Kd =

6
Kt

The coefficients, Ks and Kt are themselves a function

of :


Ks
9
= 2 2 (1 )5/2
4
Kt

 

2

an
1+
(1 )n
bn
n=1


4

an+3
+
n
bn+3
n=0

The coefficients an and bn are given in Table A.1.

The true rejection at infinite Pe, i.e. independent
volume flux is given by:
Rtrue, = 1 (1 )2 Kc

168

T.R. Noordman et al. / Journal of Membrane Science 208 (2002) 157169

Table A.1
Values of the coefficients needed in the approach of Bungay and Brenner
n

an
bn

1.2167
0.1167

1.5336
0.00442

22.5083
4.0180

5.6117
3.9788

0.3363
1.9215

1.216
4.392

1.647
5.006

Now the relation of Rtrue to the volume flux can be

determined for the present work with PEG10000. The
solute diameter of PEG10000, as calculated with the
StokesEinstein relation for diffusion coefficients, is
equal to 6.2 nm. From the experimental data, it can
be estimated that the true rejection of the membrane
at high fluxes should be around 0.80.9, which means
that is about 0.60.75 (the corresponding to a Rtrue
value 0.84 is equal to 0.66). From these values, a pore
diameter can be calculated. In addition, the value of
the ratio / m needs to be determined to calculate the
Pe as function of the volume flux, Jv . This is done
from data on the pure water flux of the membrane.
The latter was determined to be 60 l m2 h1 bar1 .
If Poiseuille flow is assumed, the pure water flux is
given by:
Jc =

dp2 P
32

From this equation, the ratio / m can be calculated, if

dp and Jc are known. Next, the Pe number as a function
of the volume flux and thus of the actual rejection as
a function of the volume flux. This plot is shown for
three different values of in Fig. 12. It can be seen that
above a flux of 3 l m2 h1 , the Rtrue is independent of

Fig. 12. True rejection as a function of flux for different ratios

of solute to pore diameter, . Above roughly 3 l m2 h1 , the
rejection attains a constant value.

the volume flux. In the present work on PEG10000, the

volume flux varies from 5 to 80 l m2 h1 . A constant
Rtrue can therefore safely be assumed.

References
[1] G.N. Vatai, M.N. Tekic, Convection promotion and gel formation in an ultrafiltration process, Chem. Eng. Commun. 132
(1995) 141149.
[2] A.R. Da Costa, A.G. Fane, D.E. Wiley, Ultrafiltration of whey
protein solutions in spacer-filled flat channels, J. Membr. Sci.
76 (1993) 245254.
[3] A.L. Copas, S. Middleman, Use of convection promotion in
the ultrafiltration of a gel-forming solute, Ind. Eng. Chem.
Process Des. Dev. 13 (1974) 143145.
[4] J.A.M. Meijer, Inhibition of calcium sulfate scale by
a fluidized bed, Ph.D. Thesis, University of Delft, The
Netherlands, 1984.
[5] J.A.M. Meijer, J.A. Wesselingh, A. Clobus, M.L.A. Goossens,
Impacts against the wall of a scaled-up fluidized bed, Desalination 58 (1986) 118.
[6] P.M. Van der Velden, New membranes for hyperfiltration processes, synthesis, properties and applications, Ph.D. Thesis,
University of Twente, Enschede, The Netherlands, 1976.
[7] M.J. Van der Waal, P.M. van der Velden, J. Koning, C.A.
Smolders, W.P.M. van Swaay, Use of fluidized beds as
turbulence promoters in tubular membrane systems, Desalination 22 (1977) 465483.
[8] G.M. Rios, H. Rakatotoarisoa, B. Tarodo de la Fuente,
Basic transport mechanics of ultrafiltration in the presence of
fluidized particles, J. Membr. Sci. 34 (1987) 331343.
[9] J.W. Smith, D.H. King, Electrochemical wall mass transfer
in liquid particulate systems, Can. J. Chem. Eng. 53 (1975)
4147.
[10] G.B. Van den Berg, C.A. Smolders, Flux decline in membrane
processes, Filtration Sep. 25 (1988) 115121.
[11] V. Gekas, B. Hallstrm, Mass transfer in the membrane
concentration polarization layer under turbulent cross flow.
Part I. Critical literature review and adaptation of existing
Sherwood correlations to membrane operations, J. Membr.
Sci. 30 (1987) 153170.
[12] A. Yasunishi, M. Fukuma, K. Muroyama, Wall-to-liquid mass
transfer in packed and fluidized beds with gasliquid cocurrent
upflow, J. Chem. Eng. Jpn. 21 (1988) 522528.
[13] A.M. Lali, A.S. Khare, J.B. Joshi, Behaviour of solid particles
in viscous non-Newtonian solutions: settling velocity, wall

T.R. Noordman et al. / Journal of Membrane Science 208 (2002) 157169

[14]
[15]

[16]

[17]

[18]

effects and bed expansion in solidliquid fluidized beds,

Powder Technol. 57 (1989) 3950.
J.F. Richardson, W.N. Zaki, Sedimentation and fluidization.
Part I, Trans. Inst. Chem. Eng. 32 (1954) 3553.
J. Garside, M.R. Al-Dibouni, Velocity voidage relationships
for fluidization and sedimentation, Ind. Eng. Chem. Process
Des. Dev. 2 (1977) 206.
P. Pradanos, J.I. Arribas, A. Hernandez, Hydraulic permeability, mass transfer and retention of PEGs in cross-flow ultrafiltration through a symmetric microporous membrane, Sep.
Sci. Technol. 27 (1992) 21212142.
Th.C. Amu, The unperturbed molecular dimensions of
poly(ethylene oxide) in aqueous solutions from intrinsic
viscosity measurements and the evaluation of the theta temperature, Polymer 23 (1983) 17751779.
A.A. Kozinski, E.N. Lightfoot, Protein ultrafiltration: a general example of boundary layer ultrafiltration, AIChE J. 18
(1972) 10301040.

169

[19] D.R. Trettin, M.R. Doshi, Ultrafiltration in an unstirred batch

cell, Ind. Eng. Chem. Fundam. 19 (1980) 189194.
[20] C.J. Van Oers, Solute rejection in multicomponent systems
during ultrafiltration, Ph.D. Thesis, University of Eindhoven,
The Netherlands, 1994.
[21] V.L. Vilker, C.K. Colton, K.A. Smith, The osmotic pressure
of concentrated protein solutions: effect of concentration and
pH in saline solutions of bovine serum albumin, J. Coll.
Interf. Sci. 79 (1981) 548565.
[22] W.M. Deen, Hindered transport of large molecules in liquidfilled pores, AIChE J. 33 (1987) 14091425.
[23] W.S. Opong, A.L. Zydney, Diffusive and convective protein
transport through asymmetric membranes, AIChe J. 37 (1991)
14971509.
[24] P.M. Bungay, H. Brenner, The motion of a closely-fitting
sphere in a fluid-filled tube, Int. J. Multiphase Flow 1 (1973)
2556.