Hydrodynamics of a liquid-liquid-solid

fluidized-bed bioreactor


































Promotor:
Prof. dr. ir. J. Tramper
Hoogleraar Bioprocestechnologie


Co-promotoren:
Dr. ir. A. Rinzema
Universitair docent Sectie proceskunde

Dr. H.H. Beeftink
Universitair docent Sectie proceskunde


Promotiecommissie:
Prof. dr. ir. R.M. Boom (Wageningen Universiteit)
Dr. ir. A.J.B. van Boxtel (Wageningen Universiteit)
Dr. ir. S. Hartmans (DSM)
Prof. dr. ir. L.A.M. van der Wielen (Technische Universiteit Delft)





Erik van Zessen




Hydrodynamics of a liquid-liquid-solid


fluidized-bed bioreactor






Proefschrift

ter verkrijging van de graad van doctor
op gezag van de rector magnificus
van Wageningen Universiteit,
prof. dr. ir. L. Speelman,
in het openbaar te verdedigen
op woensdag 1 oktober 2003
des namiddags te half twee in de Aula.
































Erik van Zessen
Hydrodynamics of a liquid-liquid-solid fluidized-bed bioreactor
Thesis Wageningen University, The Netherlands, 2003  with Dutch Summary

ISBN 90-5808-875-8




Contents




Chapter 1 Introduction 7

Chapter 2 Fluidized-bed and packed-bed characteristics of gel beads 15

Chapter 3 Dispersed-phase hold-up in a liquid-liquid extraction spray
column 43

Chapter 4 Solid and droplet hold-ups in a liquid-liquid-solid 3-phase
fluidized-bed 71

Chapter 5 Mixing of the continuous phase in a liquid-liquid-solid
fluidized-bed bioreactor 99

Chapter 6 Design of liquid-liquid-solid fluidized-bed bioreactors 125

Chapter 7 General discussion 163

References 175

Summary 187

Samenvatting 192

Nawoord 197

Curriculum vitae 199



7

1
Introduction























Chapter 1
8
The potentials of biotechnology, the use of living organisms or parts thereof for the
benefit of mankind in an economic manner, are enormous. In ancient days, water of
doubtful quality was converted into beer and wine by micro-organisms to make it
potable (Coutouly, 2001). For ages, micro-organisms have also been used for bread
making (Coutouly, 2001). The use of micro-organisms to produce chemicals started
more or less with the production of penicillin in 1943 (Coutouly, 2001). Nowadays,
chemicals are produced by a number of different micro-organisms (Krab-Hüsken,
2002). Examples of commercial microbial production processes are the production of
adipyl-7-ADCA at DSM (Bruggink, 2001), production of 6-hydroxy-S-nicotine at
Lonza (Schmid et al., 2001), or production of 5-cyanopentazamide at DuPont
(Stieglitz et al., 1996).
These chemicals are mostly produced in stirred tank reactors, in which the
micro-organisms are suspended in an aqueous phase, in which the desired product is
also present. After the biotransformation, the product is recovered from the mixture
by subsequent downstream processing. Mostly, this operation strategy works
satisfactorily.
This overall production strategy, i.e. batch biotransformation and subsequent
downstream processing of the desired product, may result in a low overall
production in some cases. These are summarized in Table 1. In case the substrate
strongly inhibits the production rate, the substrate conversion within a practical
reaction time is limited and consequently, a small amount of product is formed. Only
after a prolonged period of time, most substrate will eventually be converted. Also
for product-inhibited biotransformations the amount of product is limited: only after
a long period of time, most substrate will be converted. Another type are those
biotransformations that suffer from an unfavorable thermodynamic equilibrium.
Substrate is only partially converted, and production halts due to the approach of
thermodynamic equilibrium. A last type is a biotransformation in which the applied
substrate has a low solubility; consequently, the amount of product is limited as well.
For the first three types the overall production is low, unless the unconverted
substrate is recycled after downstream.
So, if these biotransformations should be applied for industrial purposes, with
a standard operation strategy and within a practical biotransformation time, their
productivity should be enhanced. A possible solution to do so is direct removal of
the product from the reaction environment (in case of product-inhibited or
thermodynamically controlled biotransformations). In case of substrate-inhibited
transformations (or low substrate solubility), a controlled supply of substrate to the
Introduction
9
reaction phase can overcome the limited productivity. Supply can be arranged by
using an organic solvent with dissolved substrate, or controlled supply from a solid
carrier. A possible operation strategy for product-inhibited biotransformations is
depicted in Figure 1. In this strategy, the product is transferred to a second phase to
keep the product concentration in the reaction phase low. This phase is circulated
over the extraction device. So, for the biotransformations summarized in Table 1,
integration with either substrate supply or product removal may overcome the
limited overall productivity. In the remaining part of this introduction we will focus
on product-inhibited biotransformations. The same principles, however, can be
directly used for the other types of biotransformations.

Table I. Overview of different classes of biotransformations for which productivity is
limited using a standard production strategy (batch biotransformation and subsequent
downstream processing), after Vermuë et al. (1995).
Biotransformation
characteristics
Problem Possible solution
Low solubility of
substrate in water

Low product
concentrations
Introduce a second phase (organic
solvent) that acts as a large substrate
reservoir

Substrate inhibition/
toxicity

Low production rates Keep aqueous substrate concentration
low by its controlled supply
Product inhibition/
toxicity

High product
concentrations reduce the
production rate
Keep aqueous product concentration
low by its in situ removal into a second
phase

Unfavorable thermo-
dynamic equilibrium
Incomplete conversion
due to the equilibrium
Postpone attainment of equilibrium by
direct product removal into a second
phase

Chapter 1
10
Figure 1. Operation strategy to overcome limited productivity.


True integration of the biotransformation and product removal can be realized by
executing both processes in one vessel. Integration can also be achieved by physically
separating the bioreactor and product removal device; the reaction mixture is
continuously recirculated over the product removal device.
Various examples of this latter type of integration have been presented (e.g.
Qureshi and Maddox, 1995). The product removal device can be a membrane; the
reaction mixture is recirculated over the membrane, and at the other side of the
membrane an organic solvent is used to collect the product (e.g. Hüsken et al., 2002,
who describe the production of 3-methylcatechol using octanol as the organic
solvent). The product removal device can also be an adsorption column; the reaction
mixture is recirculated over the column and the product is adsorbed onto a solid
adsorption material. Boon et al. (2000), for example, described the production of
oligosaccharides and their selective adsorption onto activated carbon.
From an operational point of view this recirculation type of integration may
be advantageous, as both processes can be controlled and optimized independently.
A disadvantage is the extra amount of equipment, and the relatively high
recirculation rate between biotransformation vessel and separation device in order to
keep product or substrate concentrations low. Clogging of the membrane or
adsorption column by microbial cells can also be a serious problem. This latter
problem can be solved by immobilizing the cells.
True integration of the biotransformation and product removal may overcome
the disadvantages of dual vessel recirculation systems. Integration of both processes
in a single vessel will result in a reduction in the amount of process equipment used.
Biotransformation Product Removal
Organic Solvent
Introduction
11
Besides, transfer of product is not delayed, but takes place instantaneously. An
obvious disadvantage is the higher degree of complexity, and fewer degrees of
freedom for process control and optimization: the two processes have to be
controlled and optimized as a single system.
Two examples of true integration of biotransformation and product removal
could be:
1) The biocatalyst is immobilized in a solid carrier and another solid carrier is
used for the adsorption of the reaction product. Both types of solid are fluidized in
the same vessel, thus yielding a solid-solid-liquid 3-phase system. Such a design was
studied extensively by Van der Wielen (1997). As a model reaction, they studied the
deacylation of penicillin G.
2) A bioreactor and an extraction spray column can be integrated. Organic
solvent droplets rise through the reaction mixture that contains immobilized
biocatalyst particles. Such a solid-liquid-liquid 3-phase system was studied by
Vermuë et al. (1995) for the degradation of tetratlin, and by Mateus et al. (1996) for the
production of L-tryptophan from indole and L-serine. In both cases, the organic
solvent was used as a reservoir from which an inhibitory substrate was supplied to
the aqueous reaction phase. Both authors used a liquid-impelled loop reactor (Figure
2A) as a basis for their integrated system (Van Sonsbeek, 1992).
Figure 2. Liquid-impelled loop reactor (A) and 3-phase fluidized-bed bioreactor (B)


In a 3-phase reactor such as the liquid-impelled loop reactor (Figure 2A), only a
single phase (the organic solvent) is recirculated in a controlled way with an external
A
B
Organic solvent droplets
Gel beads
Chapter 1
12
pump. The circulation velocities of the other two phases (aqueous phase and solids
phase) are not controlled directly but are dependent on the density difference
between riser and downcomer. A different type of 3-phase liquid-liquid-solid
bioreactor is depicted in Figure 2B. It features a stationary fluidized bed of gel beads
that are not subject to (external) recirculation, with organic solvent droplets rising
through this bed. In this bioreactor, two external pumps are used for recirculating a
single liquid phase each: one pump recirculates the organic solvent, the other pump
recirculates the aqueous phase. So, the most prominent operational difference
between a 3-phase liquid-impelled loop reactor (Figure 2A) and a 3-phase fluidized
bed (Figure 2B) is the way in which the aqueous phase is recirculated: a 3-phase
fluidized bed features an extra external pump, thus offering an additional control
option.

Performance of a bioreactor is dependent on its hydrodynamics. Also for 3-phase
bioreactors mass transfer, phase hold-ups, and mixing characteristics are largely
determined by the velocities of both liquids. In a liquid impelled loop reactor, the
velocity of the organic phase influences directly the circulation velocity of the
aqueous phase (Van Sonsbeek, 1990). In a 3-phase fluidized bed, both liquid
velocities may be chosen independently of each other, as two external pumps are
applied. Hence, for a 3-phase fluidized bed more control and optimization
possibilities are available with respect to the hydrodynamics. Optimization of the
hydrodynamics in relation to the overall production of the desired product will
result in a maximal performance of the bioreactor.

A 3-phase fluidized-bed bioreactor was studied by Davison and Thompson (1993) for
the production of butanol from glucose, and they reported an increase of 50-90% in
the butanol production rate. Gas production and subsequent slugging necessitated
the use of a tapered column to prevent wash-out of the gel beads, and to obtain
stable operation of the fluidized bed. These authors focused on the
biotransformation. Hydrodynamics of this 3-phase system were not studied.


Objective
The objective of the study presented in this thesis is the physical behavior of a liquid-
liquid-solid 3-phase fluidized-bed bioreactor. It focused on the hydrodynamic
behavior of the 3-phase system, i.e. hold-up and mixing. As a basis, the
Introduction
13
hydrodynamics of the constituent 2-phase systems were studied: a liquid-solid
fluidized bed and a liquid-liquid droplet spray column. Another objective was to
identify the conditions for which this new type of bioreactor would perform superior
to a conventional bioreactor.


Outline
In Chapter 2, the fluidization behavior of different types of gel beads is presented for
a 2-phase liquid-fluidized bed. In Chapter 3, the droplet hold-up in a 2-phase liquid-
liquid extraction spray column is reported on for different sparger lay-outs, and for
co-current and counter-current operation. The results of hold-up experiments in a 3-
phase liquid-liquid-gel bead (solid) fluidized bed are presented in Chapter 4.
Continuous phase mixing in the 3-phase fluidized bed is discussed in Chapter 5. In
Chapter 6, the preceding results are combined into a design for a 3-phase liquid-
liquid-gel bead (solid) fluidized-bed bioreactor. Performance of this 3-phase
fluidized bed is compared to its 2-phase counterpart. The thesis concludes with a
general discussion on the current status of the newly developed 3-phase fluidized-
bed bioreactor.

14


15

2
Fluidized-Bed and Packed-Bed Characteristics of Gel Beads


Abstract
A liquid-fluidized bed or packed bed with gel beads is attractive as an immobilized-
cell bioreactor. The performance of such bioreactors is influenced by the physical
behavior of these beads. Three different but related aspects involving the drag force
between particle and flowing liquid were studied for 5 types of gel beads, differing in
diameter and density. These aspects were the terminal settling velocity of a single gel
bead, pressure drop over a packed bed and voidage in a liquid-fluidized bed. For gel
beads the same trends and characteristics of these aspects were observed as for
conventional solids. However, these aspects were not correctly predicted by
established models. Though, the model of Grbavcic et al. (1991) predicts the voidage
well.
It is concluded that the drag force between gel bead and flowing liquid is lower
than that for conventional solids. Two hypotheses are suggested to explain this lower
drag force. The first one attributes the drag reduction to small amounts of dissolved
polymer. The second one attributes the smaller drag force to the nature of the gel
beads; gel beads contain over 95% of water and can in that sense be regarded as
‘rigid’ water droplets. Hence, the gel bead surface might show waterlike properties.
We modeled the drag coefficient of a single gel bead in a packed bed or a
fluidized bed, using the equation of Foscolo et al. (1983). The drag coefficient of a
single gel bead, whether present in a packed bed or liquid-fluidized bed, was well
described.
This opens to a better design and understanding of a packed bed and fluidized
bed of gel beads.

Submitted for publication
Chapter 2
16
Introduction

A packed bed or liquid-fluidized bed has attractive characteristics for application as
an immobilized-cell bioreactor (Gòdia and Solà, 1995; Willaert et al., 1996). Cells can
be easily immobilized in solid particles, making a high biocatalyst concentration
possible in a bioreactor (Wijffels et al., 1996). An elegant way of immobilization is to
embed the catalyst in beads of natural gels, like κ-carrageenan, alginate or agar
(Hulst et al., 1985). These beads, applied in fluidized-bed bioreactors, can be used for
typical bio-processes like the production of ethanol (Gòdia et al., 1987), or lactic acid
(Wang et al., 1995).
As little is known in literature on the hydrodynamics of a liquid-fluidized bed
of gel beads as well as the hydrodynamics of a packed bed of gel beads, we studied
the behavior of these systems. Gel beads differing in diameter and density were
used.
In a packed bed upward liquid velocities up to the minimum fluidization
velocity can be applied. This velocity is determined by the fact that pressure drop
equals specific buoyant weight of the bed. In this paper pressure-drop experiments
with different kinds of gel beads are described. We will show that pressure drop over
a packed bed of gel beads is much lower than the pressure drop over a packed bed of
conventional solids, like lead shot, glass pearls, etc., with the same diameter and
voidage. Consequently, the pressure drop equations in literature, e.g. the Ergun
equation or the equation of Foscolo (Foscolo et al., 1983), cannot be applied to predict
the minimum fluidization velocity.
A liquid-fluidized bed can exists between the minimum fluidization velocity
and the terminal settling velocity of a single particle. The solids hold-up, ranging ca.
0.7 for the packed-bed to zero for wash out of particles, is determined by the
equilibrium between the buoyancy force and the drag force experienced by a single
particle in such a fluidized bed. The solids hold-up for different kinds of gel beads
was determined as a function of the superficial velocity. To predict this gel bead
hold-up several models are available in literature. It will be shown that the empirical
model of Wilhelm and Kwauk (1948) with parameters according to Richardson and
Zaki (1954) does not correctly predict the gel bead hold-up between minimum
fluidization velocity and terminal settling velocity. However, the data can be well
predicted with the model of Wilhelm and Kwauk (1948) with parameters obtained
by fitting. The more fundamental model of Grbavcic et al. (1991), using
independently determined parameters, predicts the data correctly.
2-phase liquid fluidized- bed of gel beads
17
The terminal settling velocity of different kinds of gel beads was also measured.
Many models predicting this terminal settling velocity for more conventional solids
are available in literature, but all these models underestimate the terminal settling
velocity for gel beads.
The common feature of pressure drop over packed bed, solids hold-up in a
fluidized bed and terminal settling velocity, is the fact that their magnitude is
reigned by the drag force experienced by a particle. Gel beads experience a smaller
drag force than conventional particles with the same diameter and density. Two
hypotheses are suggested to explain this feature. In the first one it is considered that
drag reduction occurs by small amounts of dissolved polymer. In the second one it is
hypothesized that the observed feature results from a different surface structure of
the gel beads: in a sense they can be regarded as ‘rigid’ water droplets.
Chapter 2
18
Materials and Methods

Different gel beads and salt solutions were used. Such salt solutions are needed to
prevent the elution of counter-ions from the gel beads. Using plain tap water without
any salt addition would result in the dissolution of the gel beads. Table I gives an
overview of the different gel beads used, and the applied molarity of the salt
solutions.
All κ-carrageenan gel beads were produced using a resonance nozzle as
described by Hunik et al. (1993). The specific aqueous κ-carrageenan (Genugel 0909
Copenhagen Pectin Factory) solution, kept at 35 °C, was pressed through a nozzle.
The drops were collected in a 80 mM KCl-solution for hardening of the beads. To
obtain spherical beads a butyl-acetate (Aldrich-Chemie) layer was brought upon the
hardening solution. After hardening for approximately two hours, the beads were
stored in a KCl-solution. The alginate beads filled with yeast were prepared with a
conventional dripping method (Hulst et al., 1985).
The temperature in each experiment was equal to the ambient temperature,
26±2 °C.

Table I. Different particles used for fluidization, pressure drop and terminal
settling velocity experiments.
Gel bead Salt solution
A 3% κ-carrageenan 10 mM KCl
B 3% κ-carrageenan 10 mM KCl
C 3.25 % κ-carrageenan filled with 5% Celite™ 30 mM KCl
D 3.25 % κ-carrageenan filled with 10% Celite™ 30 mM KCl
E 8 % alginate filled with yeast cells 25 v% 50 mM CaCl
2



Characterization of the beads

Diameter
The diameter was determined with image-analysis (CCD-camera with a Nikon
macro objective 50 mm). The image of the particles was digitized and analyzed with
the software package Genius (Applied Imaging). For a higher contrast the fluid
surrounding the beads (a 10 mM KCl-solution) was colored with blue dextran
(Pharmacia Biotech, M
w
=2,000,000 g mole
-1
). The coloring of the surrounding liquid
2-phase liquid fluidized- bed of gel beads
19
was used for gel beads A and B, κ-carrageenan beads without any additions, as they
are transparent.

Volume and Density
Gel beads, as a consequence of their nature, do have a certain amount of water
attached that cannot be removed by simply sieving. In all experiments that rely upon
the exact volume of the gel beads used, this amount of water attached, has to be
accounted for. The following procedure was used for determining this amount.
Simultaneously the bead density can be calculated. A calibrated flask was weighted
(=M
flask
) and partly filled with a blue dextran solution and weighted again (=M
1
).
Sieved gel beads were added, and the flask was again weighted (=M
2
). Next the flask
was filled with the blue dextran solution up to the calibrated volume of the flask;
again the flask was weighted (=M
3
). After shaking for over 2 hours, the adsorption of
the supernatant and the adsorption of the original blue dextran solution were
determined with a spectrophotometer (Ultrospec 2000, Pharmacia Biotech) at 280 nm.
The next equations show how from this procedure the exact volume of the gel beads
can be determined:

total water flask beads gel
t supernatan
solution
added water total water
water(T)
2 3 flask 1
added water
V V V
Abs
Abs
V V
M M M M
V
− =
⋅ =
ρ
− + −
=
1

The density of the gel beads is calculated with the next equations:

( )
beads gel
beads gel
beads gel
adhering water 1 2 beads gel
water(T) added water total water adhering water
V
M
M M M M
V V M
= ρ
− − =
ρ ⋅ − =
2


Diffusion of blue dextran into the gel beads was not observed after incubating the
solids in a blue dextran solution for 48 hours.


Chapter 2
20
Terminal settling velocity
The split-times at different heights of a particle falling in a glass column with a 6 cm
inner diameter, and a total height of 1 m filled with a salt solution were measured. A
vertical distance of 50 cm from the liquid interface was used to allow a constant
velocity to be attained. The terminal settling velocity of the beads followed from the
slope between time and height. This slope was determined with linear regression.
Wall effects might become important as the particle to column diameter ratio is not
extremely large. Therefore the terminal settling velocity of bead A was also
determined in a rectangular vessel with dimensions of 43.5 x 29.5 x 100.5 cm (l x b x
h).


Pressure drop of a packed bed of gel beads
A packed bed of gel beads was formed in a column with an inner diameter of 2.56
cm, and a total length of 120 cm. The first 5 cm of the column was filled with glass
beads to provide a good flow distribution. The pressure difference over a vertical
distance was measured with a Validyne DP103 pressure transducer together with a
CD23 digital transducer indicator (maximum pressure difference 140 Pa). In each set
of pressure measurements the pressure transducer was calibrated; a liquid
manometer filled with water and octanoic acid (ρ=903 kg m
-3
) was used for this
calibration. Flow rates were measured with a calibrated rotameter (Sho-rate, Brooks).
To check whether wall effects are important, pressure drop experiments were also
done for bead D in a 6 cm inner diameter column.


Fluidization of gel beads
Experiments were carried out in a glass column with an inner diameter of 6 cm and a
total height of 3 m. Figure 1 shows the experimental lay-out. Liquid was pumped
from a storage vessel to the glass column. The first 10 cm of this column were filled
with 0.8 cm glass pearls to provide a good flow distribution. The flow rate was
measured with one or more calibrated rotameters (Sho-rate, Brooks). As the fluidized
bed showed a quiescent behavior, the height of the bed could be determined visually.
The solids hold-up followed from the bed height and the initial volume of the gel
beads.



2-phase liquid fluidized- bed of gel beads
21

Figure 1. Experimental lay-out for the fluidization experiments. 1 pump, 2 rotameter(s), 3 bed of glass
pearls, 4 sieve plate, 5 gel bead fluidized bed, 6 storage vessel


The predicting quality of a model is expressed as the average deviation (AVD)
defined as:

N
i
measured
i
X
model
i
X
measured
i
X AVD








∑ − = 3
with X
i
a model quantity, and N the number of data points.

6
2
1
4
3
5
Chapter 2
22
Results

First physical parameters of the different gel beads are discussed: shape, diameter
and terminal settling velocity. This settling velocity is compared with literature
models. Next the data of the pressure drop experiments for a packed bed are shown
and compared with literature model predictions. Thereafter, we show the results of
the fluidization experiments. An empirical model (Wilhelm and Kwauk, 1948), and a
more fundamental model (Grbavcic et al., 1991) are tested.


Gel-bead characterization

Shape and diameter
The size and shape of the different gel beads were evaluated with image analysis.
Two important chord sizes for the shape characterization of the gel bead are the
object length (OL in insert of Figure 3), defined as the largest distance between two
points on the perimeter of the object, and the object breadth (OB in insert of Figure 2),
defined as the largest distance between two points on the perimeter of the object,
perpendicular to the object length. Figure 2 shows a chord size distribution for both
the object length and object breadth typical for all gel beads analyzed. This figure
shows that the object-length distribution is shifted a little to higher values compared
to the object-breadth distribution. This indicates that the beads were not completely
spherical. This deviation from a perfect sphere was visually verified as most of the gel
beads showed a droplet like shape.

Figure 2. Object length () and object-breadth () distribution for gel bead A.
0
4
8
12
16
20
0.7 1.2 1.7 2.2 2.7
Chord size (mm)
F
r
e
q
u
e
n
c
y

(
%
)

OL
OB
OL
OB
OL
OB
2-phase liquid fluidized- bed of gel beads
23
Two shape factors could be determined with the image analysis software: the
elongation and the object circularity, see equation 4.

perimeter
area object
y Circularit
adth object bre
length object
Elongation
=
=
4

These shape factors are given in Table II together with the standard deviation. It can
be concluded that the beads approach the spherical shape closely. Thus an equivalent
spherical diameter, based on the measured area of the gel bead, was calculated.
These diameters yielded the Sauter mean diameters in Table II.

Table II. Physical parameters of the different gel beads mentioned in Table I.
Size and Shape Density
Gel bead Mean (±σ)
mm
Median
mm
Sauter mean*
mm
Shape
factor 1*
Shape
factor 2*
kg m
-3

A 1.90 (0.28) 1.99 1.97 1.03 1.18 1007.4
B 2.99 (0.45) 3.03 3.12 0.98 1.10 1005.4
C 3.14 (0.17) 3.14 3.16 1.01 1.06 1029.8
D 2.90 (0.42) 2.83 2.76 1.02 1.15 1065.1
E 4.25 (0.24) 4.27 4.27 1.00 1.13 1039.9
* definition of the Sauter mean
∑ ∑
=
i
i
i
i
d d d
2 3
32
,
shape factor 1: circularity, shape factor 2 elongation.


Terminal Settling Velocity
The mean and median of the settling-velocity distribution for the different types of
gel beads are given in Table III. As the mean and median values are almost equal, it
is concluded that the distribution is symmetrical for each type of gel bead.
The terminal settling velocity is the largest liquid superficial velocity for which
a fluidized bed exists. So, it is a key parameter in modeling fluidization. As the
fluidization experiments were done in a 6 cm inner diameter column, the terminal-
settling-velocity experiments were also done in this column. However, the ratio
between column diameter and gel bead diameter is not large, and consequently wall
effects will be present. In order to compare our measurements with literature models,
these wall effects have to be accounted for. Therefore, the terminal settling velocity
Chapter 2
24
for gel bead A was also measured in a large rectangular vessel, in which wall effects
are negligible. The mean terminal velocity was equal to 1.518 (± 0.065) cm s
-1
, the
value between brackets is the standard deviation. The ratio of the terminal settling
velocity with and without wall effects is equal to 0.90. This value is close to the value
calculated with the correlation suggested by Richardson and Zaki (1954) to account
for wall effects:
0.92 10 10
effects wall without v
effects wall with v
60
2.02

D dp
= = =






This correlation is used to correct the measured terminal settling velocity of the other
gel beads for wall effects. This corrected settling velocity is also given in Table III.

Table III. Terminal Settling velocity of the different gel beads mentioned
in Table I in cm s
-1
.
Gel bead Mean (±σ) Median Corrected for
wall effects
Model of
Dallavalle
A 1.37 (±0.08) 1.38 1.52
1
0.99
B 1.96 (±0.25) 2.02 2.22 1.42
C 4.41 (±0.17) 4.44 4.98 3.49
D 5.12 (±0.24) 5.12 5.69 4.89
E 5.22 (±0.11) 5.22 6.15 5.62
1
Experimental value


The settling velocity of a single particle in a stagnant liquid can be calculated from a
force balance for this particle. This force balance accounts for the gravity force, the
buoyancy force and the drag force. This balance reads in dimensionless form:

2
p d 4
3
Re C Ga = 5

where Ga is the Galileo number, C
d
is the drag coefficient for a single particle and
Re
p
is the particle Reynolds number, which is based on the particle dimensions. To
calculate the terminal settling velocity with equation 5, a model for the drag
coefficient is needed. Models relating the drag coefficient to the particle Reynolds
number are abundantly available in literature (Dallavalle, 1948; Turton and
Levenspiel, 1986; Hidaka et al., 1994). Other models relate the Galileo number
directly to the particle Reynolds number (Zigrang and Silvester, 1981; Turton and
2-phase liquid fluidized- bed of gel beads
25
Clark, 1987; Khan and Richardson, 1989; Hartman et al, 1989). All these literature
models predict more or less the same terminal settling velocity. Table III shows the
terminal settling velocity for the different gel beads, calculated with the equation of
Dallavalle (1948), using the physical properties of the gel beads given in Table II and
the physical properties of water at 24 °C (Weast, 1979). Table III shows that the
calculated terminal settling velocity is considerably lower than the measured
terminal settling velocity of the gel beads corrected for wall effects.
Figure 3 shows the drag coefficient calculated from the experimental values for
the settling velocity (eq.5) and the drag coefficient calculated with literature models.
From this figure it can be concluded that published models overestimate the drag
coefficient for gel beads.
Figure 3. Drag coefficient for gel bead particles settling in stagnant water as a function of the particle
Reynolds number (physical properties given in Table I). Experimental values and model predictions:
data ; ── model of Dallavalle, 1948; ─┼─ Model of Turton and Clarke, 1987;
▬▬ model of Silvester and Zigrang, 1981.


Packed-Bed Pressure Drop

Figure 4 shows the measured pressure drop over a fixed bed length for the different
gel beads as a function of the Reynolds number (Re = ρUd
p
/η). This figure clearly
shows features commonly observed in packed-bed pressure-drop experiments:
pressure drop increases with an increasing Reynolds number; a smaller particle
diameter gives a higher pressure drop at the same Reynolds number, and the
maximum pressure drop is highest for the gel-bead type with the largest density.
A commonly used model to predict pressure drop over a packed bed is the
model of Ergun (Bird et al., 1960). Foscolo et al. (1983) revised the Ergun model by

0
1
2
3
4
5
6
7
0 50 100 150 200 250
Re
p
C
d
Chapter 2
26
introducing a voidage dependent tortuosity. This model predicts drag forces acting
on an individual particle, either present in a packed bed or settling in a stagnant
liquid. Both models overestimate the pressure drop over the packed bed of gel beads
(predictions not shown).
Figure 4. Pressure drop over a packed bed of gel beads as a function of the Reynolds number (see
Table I for physical properties of the different gel beads). Column diameter is 2.54 cm.
gel bead A; gel bead B; gel bead C; gel bead D;▲ gel bead E


The large discrepancy between measured and predicted pressure drop can be
elucidated by considering the drag force experienced by a single particle in the
packed bed. A good measure for this drag force is the drag coefficient. This drag
coefficient can be related to the pressure drop. Appendix A gives a detailed
derivation, which results in:

2
p
3
3
4
d
U
d
1 L
P
C
ρ








ε −
ε ∆
= 6

The drag coefficients derived from Ergun's and Foscolo's model are given by
equation 7, see for details Appendix A:

Foscolo
1.8
0.336
Re
17.3
3
4
d
C

ε






+ =
Ergun
( )






+
ε
= 1.75
- 1 Re
100
3
4
d
C

7

0
100
200
300
400
0 5 10 15 20
Re

P
/
L






(
P
a

m
-
1
)
2-phase liquid fluidized- bed of gel beads
27
The experimental drag coefficient for gel beads A and D, calculated with equation 6,
together with the model predictions calculated with equation 7, is shown in Figure 5
as a function of the hydraulic Reynolds number ( ( ) η ρ ε − =
p 3
2
h
Ud 1 Re ). This figure
clearly shows that the drag experienced by a single gel bead in a packed bed is
smaller than the drag predicted by Ergun's and Foscolo's equations.
As the particle diameter to column diameter ratio is less than 10 for gel beads B
and C, a wall effect might by expected. Therefore the drag coefficient for these gel
beads is not shown in Figure 5. For gel bead A this ratio is larger than 10. For gel
bead D this ratio is smaller but close to 10. Indeed, a small wall effect is observed for
gel beads D; the C
d
values for data corresponding with the 6 cm diameter column are
a little lower than those for data corresponding with the 2.54 cm diameter column.

Figure 5. Drag coefficient of a single particle in a packed bed of identical particles as function of the
hydraulic Reynolds number. Experimental data: gel bead A; ▲ gel bead D, column diameter 2.54
cm; gel bead D, column diameter 6 cm. Model predictions: ── Foscolo, voidage is 0.25; …...
Foscolo, voidage is 0.38; ▬▬ Ergun.
0.1 1.0 10.0 Re
h
0
100
200
300
C
d
Chapter 2
28
Fluidization of gel beads

The different gel beads mentioned in Table I, were fluidized by the appropriate salt
solutions. The solids hold-up as a function of the superficial velocity is shown in
Figure 6.

Figure 6. Gel bead hold-up in a fluidized bed as a function of the superficial velocity. Data and
description with the Wilhelm and Kwauk model (for fitted parameters see Table IV).
gel bead A; O gel bead B; ◊ gel bead C; ▲ gel bead D; □ gel bead E

Features commonly seen in liquid fluidization of solid particles are also observed for
the fluidization of gel beads: the solids hold-up decreases with an increasing
superficial liquid velocity (U); at the same U the solids hold-up is higher for gel
beads with the same diameter but larger density (type C versus type B); at the same
U gel beads with the highest terminal settling velocity show the highest solids hold-
up.

Experimental data for particulate bed expansion were correlated by Wilhelm and
Kwauk (1948) with the following empirical equation:

k U
n
ε = 8

where U is the superficial liquid velocity, ε is the voidage, and k and n are empirical
parameters. Table IV gives the parameters in equation 8 determined from fitting this
equation to the experimental data. Figure 6 shows that the experimental data are well
described.

0
0.2
0.4
0.6
0.8
0 1 2 3 4
U (cm s
-1
)
ε
s
2-phase liquid fluidized- bed of gel beads
29
Table IV. Fitted parameters in Wilhelm and Kwauk’s model for
liquid fluidization of different gel beads.
Fitted Parameters
1

Gel bead k n n from Rowe’s model
A 1.31 (±0.03) 2.29 (±0.05) 3.06
B 1.40 (±0.03) 3.33 (±0.06) 2.79
C 3.16 (±0.08) 2.29 (±0.04) 2.61
D 3.77 (±0.07) 2.35 (±0.06) 2.61
E 4.67 (±0.08) 2.10 (±0.04) 2.55
1
value between brackets gives the 95% confidence interval


Richardson and Zaki (1954) showed that the parameter k is equal to the terminal
settling velocity and that the power n is a function of the particle Reynolds number.
If we compare parameter k in Table IV with v

in Table III, it is apparent that the
parameter k is always smaller than the terminal settling velocity. As the voidage
approaches 1, the superficial velocity should approach the terminal settling velocity,
but extrapolation gives a lower value. This fact has also been reported by different
authors for liquid fluidization of glass pearls (Garside and Al-Dibouni, 1977; Riba
and Couderc, 1977; Grbavcic et al., 1991). This kind of expansion behavior is one of
the four different types of liquid-fluidized-bed expansion behavior mentioned by Di
Felice (1994). The observed expansion behavior is characterized by two different
regimes. For lower voidage, the voidage as a function of the ratio superficial velocity
to terminal settling velocity, U/v

, shows a straight line on a log-log scale, see Figure
7. Extrapolation of this straight line to ε=1 results in a ratio U/v

smaller than 1. For
higher voidages, the voidage as a function of U/v

shows also a straight line on a
log-log scale but the slope is smaller; the extrapolation of this line to ε=1 gives a ratio
equal to 1, see the dotted line in Figure 7. In our data this second regime is hardly
present. We did not do experiments at higher voidages, because the bed height could
not be clearly determined.







Chapter 2
30
Figure 7. Voidage for gel bead D as a function of the dimensionless velocity. The superficial velocity
(U) was divided by the terminal settling velocity of gel bead D (v

). O experimental data,
── model of Grbavcic, 1991 (U
mf
= 0.358 cm s
-1
, v

= 5.12 cm s
-1
, ε
mf
= 0.38),
- - - model of Wilhelm and Kwauk, 1948 with fitted constants.
- . - model of Grbavcic 1991 conventional solid
(U
mf
= 0.18 cm s
-1
, v

= 4.4 cm s
-1
, ε
mf
= 0.38, d
p
=2.76 mm, ρ
s
=1065 kg m
-3
).


Table IV shows the fitted parameter n in equation 8, and parameter n calculated with
a well-established model from literature (Rowe, 1987):

0.75
p
0.75
p
Re 0.175 1
Re 0.4112 4.7
n
⋅ +
⋅ +
= 9

For all gel beads, except type B, the fitted parameter n is smaller than the calculated
parameter. So, the parameter n in equation 8 cannot be estimated from literature
models in order to predict the gel bead hold-up.
The relationship between voidage and superficial velocity as described by
equation 8 does predict our data satisfactorily, provided the empirical constants are
known. However, in the preceding paragraphs it was demonstrated that these
constants cannot be determined from literature models.
An alternative to the Wilhelm and Kwauk approach (eq.8) is the more
fundamental model of Grbavcic et al. (1991). These authors describe a model
predicting the voidage in liquid fluidized beds without any adjustable constants. As
parameters this model uses only the minimum fluidization velocity (U
mf
), the
0.1 1.0
0.08 0.20 0.30 0.50 0.70
U / v

1.0
0.4
0.5
0.6
0.7
0.8
0.9
ε
2-phase liquid fluidized- bed of gel beads
31
voidage at the onset of fluidization (ε
mf
) and the terminal settling velocity;
parameters that are easy to measure. The equation is:
( )
5 . 0
3
3
d
C
mf ,
d
C
mf
1
mf
) 1 (
mf
U U










ε − ε
ε − ε
= 10

The drag coefficient ratio is given by:

5 . 0
2
1
c
mf
1
mf
1
1
2
c 1
mf d,
C
d
C
(
(
(
¸
(

¸

|

'

÷

|

'

ε −
ε − ε
λ − ⋅
λ
÷ − = 11

The constants, c
1
, c
2
, and λ, in equation 11 are functions of the minimum fluidization
velocity, the voidage at the onset of fluidization, and the terminal settling velocity,
see Nomenclature for the corresponding correlation.
An interesting aspect of this model, equations 10 and 11, is the fact that it uses
as one parameter a single particle characteristic, the terminal settling velocity. The
other two parameters are packed bed characteristics: packed bed voidage at the onset
of fluidization and the minimum fluidization velocity, which is equal to the
maximum velocity through a packed bed. The packed bed voidage has to be
determined experimentally, regardless of the kind of particles used. Literature
models are abundantly available for adequately predicting v

and U
mf
for
conventional solids. However, as shown in preceding paragraphs, these models fail
to predict v

and U
mf
for gel beads. These parameters, however, can be easily
determined.
The predicting quality of this model, equations 10 and 11, expressed as the
AVD (eq.3), is given in Table V for the different gel beads mentioned in Table I. This
table also gives the minimum fluidization velocity, the voidage at this velocity, the
terminal settling velocity of the different gel beads, and the corresponding constants
c
1
, c
2
, and λ. As a typical example, Figure 7 shows the measured voidage, and the
voidage predicted by equations 10 and 11 with the constants given in Table V, as a
function of the ratio of superficial velocity to v

for type D gel beads. Considering
that the AVD of all gel beads in Table V is less than 0.05, we can conclude that the
model of Grbavcic et al. (1991) predicts the experimental data well. This conclusion is
supported by Figure 7.
Chapter 2
32
Two models, equation 8 and Grbavcic’s model, were used to describe the gel bead
hold-up as a function of the superficial liquid velocity. Figure 7 shows that both
models describe the experimental data well, although Grbavcic’s model slightly
overestimates the voidage in the intermediate regime. The advantage of equation 8
over Grbavcic’s model is the less complicated mathematical functionality. However,
fluidization data are necessary to get the empirical constants. For application of
Grbavcic’s model estimates of terminal settling velocity, minimum fluidization
velocity and the voidage at this velocity suffice.

Table V. Constants for calculating the voidage in a liquid fluidized bed as
described by equation 10 and 11, together with its predicting quality expressed
as the average deviation (AVD).
Bead Characteristics Constants in equation 11 AVD
Gel bead ε
mf
U
mf
U
t
10C
1
-100C
2 -10λ
A 0.25 0.065 1.38 9.90 16.4 8.50 0.040
B 0.35 0.080 2.02 9.99 3.8 9.63 0.050
C 0.31 0.216 4.44 9.96 8.6 9.18 0.019
D 0.38 0.358 5.12 9.96 9.5 9.10 0.022
E 0.32 0.40 5.22 9.83 23.1 7.98 0.018
ε
mf
: voidage at minimal fluidization velocity
U
mf
: minimal fluidization velocity cm s
-1

U
t
: measured terminal settling velocity cm s
-1



General Discussion

In this section we show that the drag coefficient for a gel bead in a fluidized bed is
smaller than that of conventional solid with the same density and diameter. Next it is
concluded that a smaller drag coefficient is observed in three related aspects,
terminal settling velocity, pressure drop over a packed bed, and voidage of a liquid
fluidized bed. Two hypotheses are discussed to explain the difference between gel
beads and conventional solids. This sections ends with the derivation of a model to
predict the drag coefficient of an individual gel bead either present in a packed-bed
or a fluidized bed of identical gel beads.
The drag force experienced by a particle in a fluidized bed of identical particles
depends on the voidage in such a fluidized bed. Appendix A shows how the drag
coefficient is related to the voidage:
2-phase liquid fluidized- bed of gel beads
33

g d
u 3
4
C
p 2
3
s
drag
ε
ρ
ρ − ρ
= 12

So, to compare the drag coefficient for a gel bead and a more conventional solid
particle the voidage has to be known. For the gel beads we used the measured
voidage. For the conventional solid particle, the voidage was calculated with the
model of Grbavcic et al. (1991). This model is superior to all other literature models,
as discussed above. As the model of Grbavcic et al. (1991) uses the minimal
fluidization velocity, we used Foscolo’s pressure-drop equation for calculating this
velocity. The terminal settling velocity, the other parameter, was calculated with
Dallavalle’s model (1948). Figure 7 shows the voidage in the fluidized bed of
conventional solids as a function of a dimensionless superficial velocity (the
superficial velocity divided by the terminal settling velocity of gel bead D).
Obviously, the voidage is always smaller for the gel beads at an equal velocity ratio,
and consequently their drag coefficient is also smaller.
Above, it has been shown that the drag coefficient is smaller for a single gel
bead settling in a stagnant liquid, see Figure 3. The same conclusion has been drawn
for the drag coefficient in a packed bed, see Figure 5. Thus, for the three related
aspects it is concluded that the drag coefficient for a gel bead in an array of identical
gel beads is smaller than that of a conventional solid, e.g. a glass pearl. Two
hypotheses are discussed below to explain this observation.
The observed difference between gel beads and conventional solids might be
explained by the following hypothesis. It is known that extremely low amounts of
dissolved polymer can reduce the drag force between a solid surface and a flowing
liquid (Paireau and Bonn, 1999). In our experimental set-up small amounts of
polymer, κ-carrageenan, will be present in the liquid. Consequently, the drag force
between gel bead and flowing water is reduced.
A different hypothesis comes from considering the nature of these gel beads.
Gel beads are made by dripping a solution of a natural polymer like κ-carrageenan or
alginate into a salt solution; the liquid droplet transforms to a solid due to exchange
of ions. Consequently, gel beads consist over 95% of water and in a sense can be
considered as ‘rigid’ water droplets. Water streaming alongside the gel bead surface
will experience a water-like surface. Consequently, the drag force is less, compared
to the case in which water flows alongside a non-water-like surface.
In the next part of this section a model will be derived that describes the drag
Chapter 2
34
coefficient of a gel bead in an array of other identical gel beads, whether present in a
packed-bed configuration or in a fluidized bed. Most literature models do not predict
the drag coefficient for gel beads correctly, except the model of Grbavcic et al. (1991)
which predicts the drag coefficient in fluidized beds. Unfortunately, this model
cannot predict the drag coefficient for a gel bead in a packed bed of gel beads.
As the drag coefficient is related to the pressure drop over a bed of particles, see
appendix A, a pressure-drop relation is used to model the drag coefficient for gel
beads. Pressure drop is well correlated by the Ergun equation (Bird et al., 1960). In
the derivation of this equation it is not required that the particles are in contact with
each other. Therefore, this equation should also be applicable to a fluidized bed with
a voidage ranging from the packed bed condition to the fully expanded state.
Unfortunately the Ergun equation does not describe the pressure drop over a
particulate fluidized bed well (Foscolo et al., 1983). Foscolo et al. (1983) made a re-
examination of the basis for the Ergun equation, which resulted in an equation
describing pressure drop over a packed bed as well as a fluidized bed. Incorporating
this pressure drop equation in equation 6 resulted in equation 7 for the drag
coefficient. This equation does not predict the drag coefficient for gel beads correctly,
see Figure 8. We used the form of this equation and introduced two unknown
parameters:







ε + ε =


1.8
4
c
3
d
0.336
Re
c
3
4
C 13

Equation 13 consists of two parts: a part describing the drag coefficient at laminar-
flow conditions and a part describing this coefficient at turbulent-flow conditions.
Under turbulent flow conditions the dominant contribution to energy dissipation in a
bed of particles comes from the numerous expansions and contractions of the fluid
streams in their passage around the particles. As fluid streams do not follow the
surface of the particles strictly in the turbulent regime, we believe that in this regime
the surface structure does not influence the energy dissipation. In the laminar-flow
regime, however, fluid streams do follow the surface of the particles. So the structure
of the surface will influence the drag force between flowing liquid and a single
particle in a packed or fluidized bed. Consequently, the turbulent part of the model
of Foscolo, equation 13, was used unchanged but the constants appearing in the
laminar part of the model of Foscolo were fitted to the experimental data.

2-phase liquid fluidized- bed of gel beads
35
Figure 8. Parity plot for the drag coefficient of a single gel bead in a bed of identical gel beads. Model
of Foscolo et al. (1983) with fitted constants. ◊ gel bead A; gel bead B; ∆ gel bead C; O gel bead D.
• Model of Foscolo et al. (1983) with original constants.


To find the constants c
3
and c
4
, we minimized the average deviation (AVD), equation
3, with X
i
is the drag coefficient. Not only the drag coefficient calculated from the
pressure-drop experiments were used, but also the drag coefficients calculated from
the fluidization experiments; pressure drop over a fluidized bed is straightforwardly
correlated to voidage, see also appendix A. For gel bead A pressure-drop
experiments done in a small column could be used; a wall effect was not observed. In
this column with a small diameter a wall effect was observed for gel beads B, C and
D. So for gel beads B and C, we only used the fluidization experiments done in a 6-
cm inner diameter column to fit the constants in equation 13. For gel bead D
pressure-drop experiments carried out in a 6-cm inner diameter column were
extended above the minimal fluidization velocity. These experiments were used also
to fit the constants in equation 13. The fitted constants are given in Table VI.
Table VI shows that constant c
3
differs considerably for each gel-bead type, but
constant c
4
does not differ much for gel beads A – D, especially when the sensitivity
of the model for this constant is considered. For example, for gel bead B, a constant c
4
of 0.5 results in a AVD of 0.024. Therefore, a common voidage dependency was
assumed for gel beads A to D. This common voidage dependency was implemented
into fitting equation 13 to the experimental drag coefficient. The resulting constants
are given in Table VI. Figure 8 shows that the experimental drag coefficient is well
described by equation 13, and the constants from Table VI based on the common
voidage-dependency. Table VI shows that constant c
3
for gel bead A is almost equal

1
10
100
1 10 100
C
d
experimental
C
d

p
r
e
d
i
c
t
e
d
Chapter 2
36
to the value in the original model of Foscolo, equation 7, but for the other gel beads
this constant is increasingly larger. The fitted constants for gel bead E are quite
different from the fitted constants for gel bead A – D. We have no explanation for this
difference.

For a packed bed and fluidized bed the effect of the voidage on the pressure drop is
obtained by combining equations 6 and 7 for conventional solids and by combining
equations 6 and 13 for gel beads A - D:

( )
( )
43 . 2
.8 4
- 1 ~ P : D - A beads gel
- 1 ~ P : solid al convention


ε ⋅ ε ∆
ε ⋅ ε ∆
14

The influence of the voidage on the pressure drop is much smaller for gel beads than
for conventional solids.

Table VI. Constants for equation 13 describing the drag coefficient of a single particle in a
packed or fluidized bed of identical particles, as well as the average deviation (AVD) of
the model from the experimental data.
Each gel bead type individually fitted Common voidage dependency
Gel bead C
3
C
4
AVD C
3
C
4
AVD
A 16.77 0.55 0.085 16.95 0.571 0.104
B 25.45 -0.0235 0.004 33.05 0.571 0.075
C 66.19 0.85 0.041 54.46 0.571 0.051
D 92.00 0.549 0.073 95.47 0.571 0.073
E 166.96 2.84 0.079
For gel bead E the constants given in this table are fitted on the fluidization experiments as
the pressure drop experiments could not be satisfactorily fitted on equation 13.


Conclusions

Three different properties have been measured for five different types of gel beads:
terminal settling velocity of a single gel bead, pressure drop over a packed bed, and
voidage of a liquid fluidized bed. Although the experiments with gel beads show the
same trends and characteristics as previously reported for conventional solids,
pressure drop and terminal settling velocity are not correctly predicted by
established models. The terminal settling velocity is higher, the pressure drop is
lower. Voidage of a liquid fluidized bed is also not well predicted by the empirical
2-phase liquid fluidized- bed of gel beads
37
model of Wilhelm and Kwauk with parameters according to Richardson and Zaki.
The more fundamental model of Grbavcic et al., however, predicts the voidage well,
using a measured minimal fluidization velocity and a measured terminal settling
velocity.
It can be concluded that the drag coefficient for a single gel bead, whether
settling or present in a packed bed or fluidized bed of other identical gel beads, is
always smaller than the drag coefficient of a conventional solid having the same
diameter and density.
Two hypotheses are presented to explain the features above. The first considers
a lower drag coefficient by the presence of small amounts of dissolved polymer. The
other considers that the difference between gel bead and conventional solid results
from the difference in surface structure between gel beads and more conventional
particles; a gel bead consists over 95% of water and can be regarded as a ‘rigid’ water
droplet.
The drag force experienced by a gel bead in a liquid fluidized bed, and hence
the voidage in a liquid fluidized bed, is well predicted for the different gel beads by
the model of Grbavcic et al., derived for more conventional solids. Unfortunately, this
model is not applicable for a packed bed. So, a different model is presented for
predicting the drag coefficient of a gel bead present in a packed bed or fluidized bed
of other identical gel beads. This model is based on the model of Foscolo (1983) for
conventional solids. The turbulent part is equal for gel beads and conventional
solids. The laminar part, however, shows a different but common voidage
dependency for the gel beads.


Acknowledgements

We thank Arend Mulder for supplying the alginate beads.
This work was financially supported by the Ministry of Economic Affairs, the
Ministry of Education, Culture and Science, the Ministry of Agriculture, Nature
Management and Fishery in the framework of a research program of the Netherlands
Association of Biotechnology Centers (ABON).

Chapter 2
38
Appendix A. Relationship between the drag coefficient of a single particle in
packed bed or fluidized bed of other identical particles and the pressure drop

Water flows through a bed of particles. A steady-state macroscopic mechanical-energy
balance from point 1 to 2 reads (Bird et al., 1969):
( ) ( ) 0 E W v v h h g
dp
m v
2
1
2
2 2
1
1 2
2 p
1 p
m
= φ ⋅ + −






− + − +
ρ
⋅ φ

A.1
This energy balance simplifies to equation A.2, as the velocity of the fluid at point 1 equals
the velocity at point 2 (continuity), and no work (W) is done by the surroundings on the
fluid:
( ) 0 E h h g
dp
m v 1 2
2 p
1 p
m
= φ ⋅ +






− +
ρ
⋅ φ

A.2
with E
v
the amount of mechanical energy irreversibly converted to thermal energy due to
friction. Solving the integral in equation A.2 (the fluid density is supposed to be independent
of the pressure) results in:
( ) ( ) 0 E h h g P P
v 1 2 1 2
= + − + − ρ ρ A.3
The force exerted by the fluid on a single particle, is multiplied with the velocity of the fluid
and the total amount of particles present; this gives the total power dissipated. This
dissipated power is equal to E
v
φ
m
(Stammers et al. ,1986):
p
3
2
2
1
drag 2
3
v
3
p 6
1
p
v p
2
2
1
2
p 4
1
drag p drag
d
L 1
u C E
v u
d
A ) 1 ( L
N
uA E N v v d C N v F
ε
ε −
=
ε =
π
ε −
=
ρ = ⋅ ρ π = ⋅ ⋅
A.4
Combining equation A.3 and A.4 relates the drag coefficient for a single particle in a bed of
identical particles to the pressure drop:
h g
d
L 1
u C h g E P2 P1
p
3
2
drag 4
3
v
∆ ρ +
ε
ε −
ρ = ∆ ρ + ρ = −
ε
ε
ρ A.5
In the set-up (Figure A.1) the pressure difference between point 1 and 2 is measured, and
consequently the hydrostatic pressure, h g∆ ρ , is cancelled. So equation A.5 simplifies to:
p
3
2
drag 4
3
d
1 1
u C
L
2 P 1 P
ε
ε −
ρ =

A.6
Pressure drop over a packed bed due to friction between fluid and particles can be described
by the Ergun equation:
2-phase liquid fluidized- bed of gel beads
39
( )
3
p
2
3
2
2
p
1
d
u
75 . 1
1
d
u 150
L
P
ε
ε − ρ
+
ε
ε − η
=

A.7
Pressure drop can also be described with the equation of Foscolo et al. (1983). These authors
used the same basic elements but a porosity dependent tortuosity instead of a constant
tortuosity.
( )
4.8
p
2
1
d
U
0.336
Re
17.3
L
P

ε ε −
ρ






+ =

A.8
At a voidage of 0.4 both models give the same pressure drop.

Equation A.6 can be combined with equations A.7 or A.8 to give an equation for the drag
coefficient for a single particle in a packed bed of other identical particles:
Foscolo
1.8
d
0.336
Re
17.3
3
4
C

ε






+ =
Ergun
3 . 2
Re
133
C
h
d
+ =
with Re
h
the hydraulic Reynolds number defined as
( )
η
ρ ε −
p
3
2
Ud 1
.


Liquid Fluidization

The pressure drop in liquid fluidization is equal to the specific buoyant weight of the bed
(Foscolo et al. 1983):
( ) ( )g 1
L
P
s
ρ − ρ ⋅ ε − =

A.9
Combining equation A.9 and A.6 gives the drag coefficient of single particle in a fluidized
bed of identical particles:
g d
u
C
p 2
3
s
3
4
drag
ε
ρ
ρ − ρ
= A.10
Chapter 2
40
Nomenclature
c
1
: constant in equation 11 ( ) ( ) ( )
0.5
2
3
mf
2 2
mf
v / U 1


ε + -
c
2
: constant in equation 11 ( ) ( ) ( )
1
0.5
2
1
0.5
2
1
c c - 1 / c - 1 − -
c
3
: constant in equation 13 -
c
4
: constant in equation 13 -
C
d
: drag coefficient for a single particle -
d
p
: particle diameter m
g : gravity constant m
2
s
-1
Ga : Galileo number ( )
2
c c s
3
p
g d η ρ ρ − ρ -
k : constant in equation 8 -
L : length m
n : constant in equation 8 -
∆P : pressure drop N m
-2

Re
p
: particle Reynolds number η ρ d v
c ∞
-
Re : Reynolds number η ρ Ud
c
-
U : superficial liquid velocity m s
-1
v

: terminal settling velocity m s
-1

Greek
ε : voidage -
η : viscosity kg m
-1
s
-1
λ : constant in equation 11 ( )
1
0.5
2
1
c c 1 − − -
ρ : density kg m
-3

∆ρ : density difference kg m
-3



Subscript
mf : at minimal fluidization conditions
s : gel bead
c : liquid

41

42

43

3
Dispersed-phase hold-up in a liquid-liquid extraction
spray column


Abstract

Phase hold-ups were studied in a spray column with a water/dodecane 2-phase
system, in order to support the development of an extractive fermentation process.
Different sparger geometries were used. Water and dodecane were introduced in co-
current and counter-current flow. Although water and dodecane fluxes were of the
same order of magnitude, the water flux had a minor influence on the dodecane
hold-up. The nozzle geometry, however, influenced the hold-up to a large extent; the
sparger with the lowest number of nozzles showed the highest hold-up at the same
dodecane flux. A model that predicts the dispersed-phase hold-up in a spray column
regardless of the mode of operation is presented. This model uses the slip velocity
between the continuous phase and the dispersed phase to determine the dispersed-
phase hold-up. The proposed model incorporates the nozzle geometry, the physical
properties of the 2-phase system and the fluxes of the continuous and dispersed
phase. A Richardson and Zaki type equation is used for the slip velocity. We present
a new model for the power n in the Richardson and Zaki model, in which n depends
on the dimensionless drift flux. The model describes the hold-up data for the
water/dodecane system well. Validation of the model was obtained with a 60 mM
KCl solution/dodecane 2-phase system. This 2-phase system differs only in surface
tension.
Chapter 3
44
Introduction

Kinetic or thermodynamic product inhibition in bioconversion processes can be
reduced by integrating the conversion with an extraction process in one apparatus,
so-called in situ extraction. The (bio)catalyst can be freely suspended or immobilized
in a solid support. The protection against direct contact with extractant droplets is an
important advantage of immobilized biocatalyst systems (Vermuë and Tramper,
1995). With regard to the extraction, a choice has to be made between systems with
and without moving internals. Moving internals enhance the extraction process
(Godfrey and Slater, 1994), but are likely to cause inadmissible damage to the solid
biocatalysts. This holds especially for solid biocatalysts immobilized in a gel matrix
(Wijffels et al., 1995). Therefore, columns without moving internals are more
appropriate for immobilized biocatalysts.
We are currently developing a fluidized bed with extractant droplets rising
through the bed of biocatalytic solids. This system can be a suitable reactor set-up for
biotransformations using immobilized biocatalysts and applying in situ extraction.
This reactor set-up can offer high biocatalyst hold-up as well as high mass transfer
rates. Our first aim is to predict the hold-up of the two liquids and the solids phase in
this system, as these are important variables with respect to the conversion and
extraction rates. In order to understand the behavior of the 3-phase system we have
studied the behavior of the spray column with two liquids (the extractor) and a
fluidized bed with one liquid and suspended solids (the reactor) separately. This
paper describes the work on the spray column.
Although the dispersed-phase hold-up in spray columns has been studied
several times (Kumar and Hartland, 1991; Godfrey and Slater, 1994), the models
resulting from these studies cannot be extended straightforwardly to models
predicting the dispersed-phase hold-up in the above-mentioned 3-phase fluidized
bed. We believe that the prediction of droplet hold-up in such a 3-phase system can
be based on the prediction of droplet hold-up in a spray column. Such an extension
of a 2-phase model to a 3-phase model predicting the dispersed-phase hold-up, has
been demonstrated for the prediction of the gas hold-up in a liquid-solid-gas
fluidized bed (Chen and Fan, 1990). Richardson and Zaki's model (Richardson and
Zaki, 1954) for solid fluidization can be extended to predict the hold-up of the
different components in a multi-component fluidized bed (Van der Wielen et al.
1996).
This article presents a model that predicts the dispersed-phase hold-up in a
Droplet spray-column
45
spray column regardless of the mode of operation. This model can be easily extended
to predict the dispersed-phase hold-up in a 3-phase system. The slip velocity
between the continuous phase and the dispersed phase is used to determine the
dispersed-phase hold-up, analogous to models presented in literature. The slip
velocity is described with Richardson and Zaki's model for a fluidized system of
mono-sized particles. This model uses two parameters, viz. the terminal rise velocity
of a single droplet and a power n. The terminal velocity can be calculated with
models from literature, but models for determining the power n in liquid-liquid 2-
phase systems have not yet been published. In this paper we will propose a new
model for this power n.


Theory

The dispersed-phase hold-up can be predicted according to a model presented
below. A schematic overview of this model is shown in Figure 1. This figure shows
how the combination of a definition of the slip velocity and a model for this slip
velocity can be used to predict the dispersed-phase hold-up. This figure also shows
how the various operating variables (superficial velocity of both phases and lay-out
of the sparger) as well as the physical properties of the 2-phase system influence the
slip velocity, and hence the dispersed-phase hold-up.


Figure 1. Schematic representation of the model for the prediction of the dispersed-phase hold-up.
Droplet Hold-ups are less than 0.23. U
d
: superficial velocity dispersed phase; U
c
: superficial velocity
continuous phase; v

: terminal rise velocity of a single droplet; n: power n in equation 2; d: droplet
diameter.
Slip velocity
Definition (eq.1) ≡ model (eq.2)
hold-up
U
d
U
c
hold-up < 0.13 ⇒ n = constant
hold-up > 0.13 ⇒ n (eq.8)
U
c
U
d
v

(eq.3)
d (eq.4)
physical
properties
U
d
sparger
lay out
Chapter 3
46
Slip velocity
The slip velocity (v
12
) is defined as the difference between the velocities of both
phases taking into account the direction of flow:


α −

α
=
1
U U
v
sc sd
12
r r
1

with U
sd
is the superficial velocity of the dispersed phase, U
sc
is the superficial
velocity of the continuous phase and α is the dispersed-phase hold-up. Upward flow
is taken as the positive direction for both flows.
The relationship between slip velocity and hold-up is implicit: the dispersed-
phase hold-up is determined by the slip velocity and the slip velocity depends on the
dispersed-phase hold-up. Empirical models that predict directly the dispersed-phase
hold-up have been reviewed (Kumar and Hartland, 1995). However, these authors
demonstrated that those empirical models are less accurate than slip velocity models.
Different relations for slip velocity can be found in literature (Kumar and
Hartland, 1995; Godfrey and Slater, 1991). Godfrey and Slater state that the best
equation describing the slip velocity is the equation of Kumar and Hartland (Kumar
and Hartland, 1985). Godfrey and Slater (1991) also concluded that most of the
equations describing slip velocity can be summed up to one equation based on
Richardson and Zaki's equation for fluidized systems of mono-sized rigid particles:

( )
n
k 12
- 1 v v α = 2

with v
k
a characteristic velocity, in most cases equal to the terminal rise velocity of a
single droplet (v

). The power n indicates the influence of liquid flow around a
droplet. To calculate the slip velocity (eq.2), models for the terminal rise velocity and
power n are needed. As the terminal rise velocity depends on the droplet diameter,
also a model is needed to predict this diameter (see also Figure 1).

Terminal rise velocity
Godfroy and Slater (1994) give a selection chart, based on physical properties of both
phases, for choosing the best-suited terminal rise-velocity equation. The selection is
mainly based on the Morton number of the 2-phase system, with the Morton-number
Droplet spray-column
47
defined as
4 2
4

g
Mo
σ ρ
η ρ
c
c

= . We have studied a water/dodecane 2-phase system. The
surface tension between water and dodecane is high, which results in a low Morton
number. Therefore, based on the selection chart, Vignes' equation was used:
















η
ρ








ρ
ρ ∆
=

6
Eo
1
g
4.2
d
v
3 / 1
c
c
3 / 2
c
3

where d is the droplet diameter, g is the gravity constant, ∆ρ is the density difference
between water and dodecane, ρ
c
is the density of the continuous phase, η
c
is the
viscosity of the continuous phase. Eo is the Eötvös-number defined as
σ
ρ ∆
2
d g
,
representing the ratio of the buoyancy force over the resistive force defined by the
interfacial tension.

Droplet diameter
Liquid droplets can be made by pumping liquid through one or more nozzles. An
overview of different droplet formation mechanisms is given elsewhere (Dalingaros
et al., 1986). At a nozzle Weber number less than 2, single drops are formed at the tip
of a nozzle, with the nozzle Weber number defined as
σ
ρ
noz
2
noz
d u
. The Weber
number represents the ratio between inertial force and resistive interfacial force. At a
nozzle Weber number between 2 and a critical nozzle Weber number 8.64, drops are
formed by uniform break-up of the liquid jet existing in this range. Above a nozzle
Weber number of 8.64 the liquid jet starts to atomize and a broad droplet-size
distribution is formed. A single relation covering nozzle Weber numbers up to 8.64
has been derived by Kumar and Hartland (1996):

315 . 0
2
noz d 73 . 0
noz
33 . 0
noz
noz
gd
We 0393 . 0 Eo 55 . 0
d
d









σ
ρ
+
= 4
where d
noz
is the nozzle diameter, Eo
noz
is the Eötvös number based on the nozzle
dimensions, ρ
d
is the density of the dispersed phase, and σ is the interfacial tension
between water and dodecane. This relation was used here for calculating the droplet
diameter.
Chapter 3
48
Power n
A model predicting the power n used in the slip velocity model (eq.2), has not yet
been published for liquid-liquid 2-phase systems. It is concluded by Godfrey and
Slater (1991) that this power n has a value between 0 and 4. When the interfacial
tension is high, liquid droplets can be regarded as solid particles. In that case
relations for the power n derived for water and solid particles can be used. An
overview of different relations and their accuracy can be found elsewhere (Hartman
et al., 1994). However, we will show that these models for the power n do not
describe the slip velocity for a water-dodecane 2-phase system accurately. As slip
velocity and hold-up are closely related, see Figure 1, the dispersed-phase hold-up is
also not well predicted.
In this study we will propose a new model for the power n; for reasons of
clarity this model is integrated in the Results and Discussion section. Applying the
total model as given in Figure 1, i.e. the combination of equation 1 and 2, the new
model for the power n, and the presented equations for droplet diameter (eq.4) and
terminal rise velocity (eq.3), the experimental dispersed-phase hold-up is well
described. Data from a water/dodecane 2-phase system with a different surface
tension are well predicted by the model presented in Figure 1.


Droplet spray-column
49
Materials and Methods

Wageningen tap water was used as continuous phase and n-dodecane (Aldrich,
99%+ pure) as dispersed phase. As the n-dodecane is lighter than tap water, it was
introduced in all experiments at the bottom of a 2 m high and 60 mm internal-
diameter column equipped with sample ports to make measurements at different
heights. Tap water was introduced at the bottom of the column in co-current
operation and at the top in counter-current operation. On top of this column a phase
separator was placed with a diameter of 180 mm. As water and n-dodecane are
highly immiscible, phase separation was easily achieved for the applied fluxes.


Co-Current Operation
Figure 2 shows schematically the experimental set-up. Both tap water and n-
dodecane were introduced through a sparger at the bottom of the column. The
separated phases at the top of the column were sent back to the storage vessels to
allow recirculation of both liquids. As the storage vessels were jacketed, temperature
could be maintained at 25±1°C by circulating water of appropriate temperature. No
temperature gradient was observed between top and bottom of the column.

Figure 2. Experimental set-up for co-current operation. 1 column; 2 phase separator; 3 sparger; 4
rotameters; 5 dodecane storage vessel; 6 tap water storage vessel; 7 inverse water manometer


4
5
1
2
4
6
7
3
Chapter 3
50
Liquid Distributor. The liquid distributor is schematically shown in Figure 3a. This
distributor consisted of an inner and an outer cone. The distance between both cones
was constant, with dodecane flowing through the inner cone, and the tap water
through the outer one. Both cones had a 10° angle and were filled with glass beads to
provide a good flow distribution. The maximum diameter of the outer cone was 60
mm, the inner cone had a 30 mm maximum diameter. On top of the inner cone a
sparger was placed, which consisted of a number of nozzles. An overview of the
different spargers used is given in Table I. The nozzles were made of hollow
stainless-steel tube with a length of 3 cm, which were hammered through a Teflon
plate. The distance between the nozzles was arranged in such a way, that the
growing droplets did not touch each other. The nozzles were evenly distributed over
the area of a circle with a 30 mm diameter.

Figure 3a. Liquid distributor in co-current
operation. 1 dodecane inlet; 2 water inlet;
3 sparger; 4 nozzles
Figure 3b. Alternative liquid distributor in
counter-current operation. 1 dodecane inlet; 2
water inlet; 3 sparger; 4 nozzles;
5 glass pearls.


Counter-Current Operation
The same experimental set-up was used as described in the co-current operation part,
except for the introduction and recirculation of the tap water. Figure 3c gives details
of the changes made in the co-current experimental set-up to make counter-current
water recirculation possible. The first adaptation was made in the glass column wall,
see Figure 3c1 and 3c2; the wall had four circular openings, diameter 10 mm, located
in one plane at 90° relative to each other, Figure 3c2. These openings were 1 cm
above the liquid distributor, Figure 3c1. The flows through these openings were
adjusted with throttle valves in such a way that they equaled each other. The second
adaptation was made on the tube used for the water introduction at the top of the
column, see also Figure 3c3. This tube was split in two parts, and both ends were
directed to the outer wall and bottom of the annulus made up between column and
enlarged column, see also Figure 3c3. In this way plug flow was assured through the

} 3
1
2 2
4
}
1
4
2
5
3
Droplet spray-column
51
column. This was visually verified by color-pulse experiments. With these color-
pulse experiments a flat front was still observed at the bottom of the column.
Although only the dodecane storage vessel was maintained at 25±1°C by circulating
water of appropriate temperature, a temperature gradient was not observed in the
column.
3c1 3c2 3c3
Figure 3c. Details on adaptations made on the co-current experimental set-up for counter-current
operation


Liquid Distributor. The same liquid distributor was used for dodecane dispersal as
described with the co-current operation. Also another sparger was used, see Figure
3b. This liquid distributor consisted of a sparger with 209 nozzles (5 cm length and 1
mm diameter), evenly distributed over the area of a circle with a 60 mm diameter. On
top of a cone a 10 cm high glass column, diameter 6 cm, was placed. The cone had an
angle of 10° and a maximum diameter of 60 mm. Both cone and column were filled
with glass beads to provide a good flow distribution over the nozzles that were
placed on top of the glass column.

Table I. Different spargers used in experiments.
# nozzles nozzle diameter (mm)
sparger 1 12 1.4
sparger 2 23 1.0
sparger 3 37 1.0
sparger 4 54 1.0
sparger 5
1
209 1.0
1
only in counter-current operation
1
2 2
1
Chapter 3
52
Measurements
The hold-up of n-dodecane at different flow rates of tap water and n-dodecane was
determined with a reversed U-tube water manometer. To increase the accuracy, the
water manometer was filled with hexanoic acid (ρ=921 kg m
-3
). These measurements
were used to determine the parameters in the model for the power n. Dodecane hold-
up measurements have also been done with a slightly different 2-phase system: a 60
mM KCl solution as continuous phase and the same dodecane as dispersed phase.
These measurements were used to validate the new model.

The fluxes of tap water and dodecane were measured with calibrated rotameters
(Sho-Rate). The flux of tap water ranged from 0 to 0.7 cm s
-1
and the dodecane flux
ranged from 0 to 1.25 cm s
-1
.

The static interfacial tension between continuous phase and dispersed phase was
determined with the Wilhelmy-plate method.

The average deviation (AVD) between measurement and model is defined as:
1 N AVD
meas
mod meas

α
α − α
=

5


Results and Discussion

The dispersed-phase hold-up was measured for different sections of the column and
different combinations of dodecane and water fluxes. From these data (not shown) it
can be concluded that an axial hold-up profile is not significant for any of the
combinations. Radial profiles may exist (Ueyama and Miyauchi, 1979). However, for
the sake of simplicity, a uniform distribution of the droplets across the column
diameter is assumed here.

Zero water flux
The effect of the sparger lay-out on the measured dodecane hold-up is shown in
Figure 4. In this figure experimental data are shown for spargers 3, 4 and 5 at zero
water flux. The same trends were observed for the co- and counter-current hold-up
data. Hold-up increases with an increasing superficial dodecane velocity. Figure 4
Droplet spray-column
53
clearly shows that the increase in hold-up is larger for a sparger with a lower number
of nozzles. This holds especially at dodecane fluxes larger than 3 mm s
-1
.
Figure 4. Dodecane hold-up versus dodecane superficial velocity for sparger 3 to 5 (Table I).
● sparger 3, 37 nozzles; ◊ sparger 4, 54 nozzles; ▲ sparger 5, 209 nozzles


The difference between these spargers can be explained by considering the one-
dimensional continuity equation for the dodecane phase:
d
d
d d
v
U
v U = α ⇒ α =
So the dodecane hold-up (α) is defined by the ratio between the superficial velocity
(U
d
), and the true velocity (v
d
) of the dodecane phase. If it is assumed that the
droplets do not influence each other, then the true velocity equals the rise velocity of
a single droplet. At an equal superficial dodecane velocity there is a difference in
droplet diameters generated by the different spargers, as the diameter depends on
the dodecane velocity through one nozzle (see also the section on the droplet
diameter). A smaller number of nozzles per sparger will result in a higher velocity
through one nozzle, and hence a smaller diameter. A smaller diameter results in a
lower terminal rise velocity (eq.3) and hence a higher hold-up.
Figure 4 shows data up to a nozzle Weber number of 8.64. Above this nozzle
Weber number atomization of the liquid jet starts; a broad droplet-size distribution
will occur, and the slip velocity is not described by equation 2. Moreover, phase
separation becomes less straightforward, as it can not be achieved by a single-stage
phase-separator that is based on gravity.
Figure 5 shows data for spargers 1 and 2. In this figure, contrary to Figure 4, the
superficial velocity goes beyond the velocity corresponding with the critical nozzle
0
0.04
0.08
0.12
0.16
0 0.003 0.006 0.009 0.012
U
d
(m s
-1
)
α
Chapter 3
54
Weber number. Figure 5 shows that for sparger two at each superficial dodecane
velocity, the hold-up is equal or larger compared with data for sparger 1. Below the
critical nozzle Weber number this is explained in exactly the same way as above: at a
dodecane superficial velocity of 3 mm s
-1
the velocity through one nozzle equals
0.459 m s
-1
for sparger 1 (12 nozzles, with a diameter of 1.4 mm), and 0.470 m s
-1
for
sparger 2 (23 nozzles, with a diameter of 1 mm). At this superficial velocity the
droplet diameters (eq.4) and terminal rise velocities (eq.3), are 3 mm and 0.123 cm s
-1

for sparger 1, and 2.4 mm and 0.103 cm s
-1
for sparger 2. Consequently, the hold-up is
expected to be higher for sparger 2. (The physical properties used for calculating
droplet diameter, and terminal rise velocity are given in Table II).

Table II. Physical Properties of the liquids applied at 25±1°C.


Density
kg m
-3

Viscosity
Pa s
Surface tension
1

mN/m
Surface tension
2

mN/m
Tap water 998 0.00093 68-63.5 40
Dodecane 742.7 0.0012 24
1
surface tension against water,
2
surface tension between tap water and dodecane


At a zero water flux the slip velocity reduces to the true velocity,
α
d
U
. In Figure 5 this
velocity is shown as a function of the superficial dodecane velocity. The true
dodecane velocity is almost constant above a certain value of the superficial
dodecane velocity.

For both spargers this value corresponds with the critical superficial dodecane
velocity, calculated from the critical nozzle Weber number. This Weber number is
indicated by the vertical lines in Figure 5. So above the critical nozzle Weber number
the true dodecane velocity is independent of the hold-up. This means that the hold-
up is linearly proportional to the superficial dodecane velocity, as indicated by the
dotted lines in Figure 5. Obviously the relationship between hold-up and superficial
dodecane velocity becomes different above the critical nozzle Weber number.

Droplet spray-column
55
Figure 5. Dodecane hold-up (closed symbols) and slip velocity (open symbols) versus dodecane
superficial velocity for sparger 1 and 2. ● sparger 1, 12 nozzles, i.e. 1.4 mm; ▲sparger 2, 23 nozzles,
i.e. 1 mm. Superficial water velocity is zero.

Co- and counter-current water flux
The above-described characteristics for dodecane hold-up at zero water flux are also
observed for co- and counter current operation, when the dodecane flux is changed
at a constant water flux (data not shown). At low superficial dodecane fluxes, i.e. low
hold-ups, no effect of the water flux can be observed. At higher superficial dodecane
fluxes, i.e. higher hold-ups, there is a small effect of the water flux. For co-current
operation an increase in water flux resulted in a slightly lower hold-up at the same
superficial dodecane flux, whereas for counter-current operation a slightly higher
hold-up was measured. Obviously, this is straightforwardly explained by the
reduction or increase of the true dodecane velocity at counter- or co-current
operation, respectively.
Under the conditions tested the influence of the superficial water velocity is
marginal. This can be illustrated by comparing the true water velocity with the true
dodecane velocity: at a superficial water velocity of 7.51 mm s
-1
, and a superficial
dodecane velocity of 7.14 mm s
-1
, a dodecane hold-up of 0.14 was measured in
counter-current operation using sparger 4 (54 nozzles). With these values the true
dodecane velocity becomes 5.8 times higher than the true water velocity:

8 . 5
1
86 . 0
51 . 7
14 . 0
14 . 7
1
1
c
U
d
U
=

⋅ =

α −

α













0.000 0.002 0.004 0.006 0.008
U
d
(m s
-1
)
0.100
0.200
0.05
0.07
0.09
0.12
0.15
v
12
(m s
-1
)
0.010
0.100
5
7
2
5
7
α
Chapter 3
56
Dodecane hold-up can be manipulated to a large extent with the sparger lay-out and
to a minor extent with the mode of operation. Less nozzles on a sparger give a higher
hold-up at the same superficial dodecane flux. However, for a sparger equipped with
few nozzles, the dodecane flux corresponding with a nozzle Weber number of 8.64 is
lower compared with a sparger equipped with more nozzles. Under our conditions
the water flux only slightly influenced the dodecane hold-up. A larger influence can
be expected with increasing water flux. In counter-current operation this would yield
a higher hold-up. However, at the higher applied water fluxes satellite-formation
was observed. As these satellites form emulsions, and phase separation of emulsions
is not straightforward, the water flux cannot be increased unlimitedly. Also, in
counter-current operation, flooding will occur at a certain water flux.


Modeling droplet hold-up

The droplet hold-up can be predicted from an accurate slip velocity model (Kumar
and Hartland, 1985):
( ) ( )








ρ
η
+ ⋅ α + ρ = α − ρ ∆
c 12
c 0.73 2
12 c 3
4
v d
24
0.53 4.56 1 v 1 g d 6
However, this model can not be extended easily to systems with more than two
phases. A slip velocity model (eq.2), analogous to the Richardson and Zaki model
(1954), can be easily extended. Figure 1 shows how from such a model the droplet
hold-up can be predicted. This model, however, depends on a model for the power n,
which has not been developed yet. Below a model will be proposed to predict this
power n. This model holds for dispersed-phase velocities up to velocities
corresponding with the critical nozzle Weber number of 8.64.
Obviously, the power n can be obtained from fitting equation 2 to the slip
velocity, with this velocity determined from the experimental hold-up. However,
especially at low hold-ups, the relative error in the hold-up is large, and
consequently there is large scatter in the slip velocity. This scatter largely disappears
when the slip velocity is transformed to a drift flux, U
dc
(adapted from Wallis, 1969):

( ) ( ) ( )
1 n
12
c d
dc
1 v 1 v 1
1
U U
U
+

α − ⋅ α = α − ⋅ α = α − ⋅ α






α −

α
= 7

Droplet spray-column
57
A drift flux represents the volumetric flow rate of the dispersed-phase relative to a
surface moving at the volumetric average velocity (U
d
+U
c
) (Wallis, 1969). Next, the
drift flux is made dimensionless by dividing this drift flux by the terminal rise
velocity (eq.3). For all the spargers applied, the dimensionless drift-flux, derived
from experiments for co- and counter-current operation, is shown in Figure 6 as a
function of the dodecane hold-up. This figure shows that the dimensionless drift flux
is virtually independent of the sparger lay-out.
Equation 7 was fitted to the dimensionless drift flux data up to a dispersed-
phase hold-up of 0.13. This gave a power n of 4.00 with a 95% confidence interval of
0.07.
Up to a hold-up of 0.13 the dimensionless drift flux is well described with a
constant power n. At higher hold-ups this model systematically underestimates the
dimensionless drift flux, as can be seen in Figure 6. An underestimation of the
dimensionless drift flux means that the power n in equation 7 is too large. It appears
that the power n, derived from the measured hold-up and equations 3 and 7,
decreases monotonically with the dimensionless drift-flux. So a new model is
proposed for the power n:
2
c
dc
1 0
v
U
c n n








− =

8
The dimensionless drift flux, defined as the drift flux divided by the terminal rise
velocity of a single droplet, is shown in Figure 6. This figure shows that the
dimensionless drift flux (U
dc
*
) becomes constant for higher hold-ups. The hold-up for
which U
dc
*
becomes constant, is called α
boundary
, and is also shown in Figure 6. So, the
new model for the power n, equation 8, is applicable for hold-ups up to α
boundary
.

The total model for predicting the dimensionless drift flux consists of two parts:
• for hold-ups below α
boundary
, the model is represented by Figure 1:
⋅ nozzle lay-out and physical properties of the 2-phase system determine
the terminal rise velocity of a droplet;
⋅ a definition and a model of the slip velocity transformed to a drift flux,
eq.7;
⋅ a model for the power n, eq.8.
• for holds-ups above α
boundary
, the dimensionless drift flux is constant, and equal
to 0.0975 with a 95% confidence interval of 0.0004.


Chapter 3
58
Figure 6. Dimensionless drift flux versus dodecane hold-up. O sparger 1; ● sparger 2; sparger 3; ▲
sparger 4; ◊ sparger 5.  total model with n = f(Udc), … total model with n=4


Both parts of the model hold at the α
boundary
, which gives an additional equation for
the coefficients n
0
, c
1
, and c
2
. Using equation 7, with the hold-up equal to α
boundary
,
and the dimensionless drift flux equal to its constant value (0.0975), a corresponding
power n can be calculated: this is n
boundary
. Substituting this n
boundary
in equation 8,
gives the following relationship between the coefficients in equation 8.

ln(0.0975)
1
c
n n
ln c
c
0.0975 c n n
1
boundary 0
2
2
1 boundary 0










= ⇒ ⋅ = − 9

If coefficient c
2
is determined with equation 9 at any given n
0
, c
1
and n
boundary
both
parts of the model for the dimensionless drift flux will hold at α
boundary
.
A clear value for α
boundary
cannot be determined unambiguously from Figure 6.
Therefore α
boundary
has been varied on the interval 0.19…0.24. At each α
boundary
a
corresponding n
boundary
is calculated, and the coefficients n
o
and c
1
are fitted to
equations 3, 7, 8 and 9 by minimizing the residual sum of squares of measured and
predicted droplet hold-up (equation 10), using a downhill simplex method (Nelder
and Mead, 1965):

( )

α − α =
i
2
predicted i, measured i,
SSres 10

In appendix A it is shown that the parameters n
o
and c
1
cannot have an arbitrary
value, but are subjected to some constraints. Applying these constraints, fitting was
0
0.025
0.05
0.075
0.1
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
α
U
dc
*
α
boundary
Droplet spray-column
59
successful with the next set of parameters yielding the lowest sum of squares of
residuals: α
boundary
=0.233, n
0
=6.55

and c
1
=33.2. Coefficient c
2
follows from equation 9
and is equal to 0.88. It is concluded from Fig 6, which shows the dimensionless drift
flux predicted with equations 3, 7, 8 and 9 and the fitted constants, as well as the
experimentally determined dimensionless drift flux, that the prediction of the model
is good. This is also supported by the parity plot in Figure 7.



Figure 7. Parity-plot between measured and predicted dodecane hold-up for different models.
A : presented model; B : slip velocity model of Kumar and Hartland 1985; C : explicit model of Kumar
and Hartland 1995; D : slip velocity model with n according Richardson and Zaki 1954


A complete analysis of the confidence intervals of the fitted parameters is beyond the
scope of this paper. However, to get some insight in the sensitivity of the model to its
parameters, the SS
res
was calculated at different combinations of parameters n
0
and c
1

for α
boundary
=0.23. The SS
res
for the fitted parameters is equal to 0.0244. Figure 8 shows
contour lines at 1, 5 and 10 % higher values than the minimal SS
res
. This figure also
shows the relation between the parameters n
0
and c
1
that resulted from the constraint

0
0.1
0.2
0.3
0 0.1 0.2 0.3
Measured α
Predict ed α
A
0
0.1
0.2
0.3
0 0.1 0.2 0.3
Measured α
Predict ed α
B
0
0.1
0.2
0.3
0 0.1 0.2 0.3
Measured α
Predict ed α
C
0
0.1
0.2
0.3
0 0.1 0.2 0.3
Measured α
Predict ed α
D
0
0.1
0.2
0.3
0 0.1 0.2 0.3
Measured α
Predict ed α
A
0
0.1
0.2
0.3
0 0.1 0.2 0.3
Measured α
Predict ed α
B
0
0.1
0.2
0.3
0 0.1 0.2 0.3
Measured α
Predict ed α
C
0
0.1
0.2
0.3
0 0.1 0.2 0.3
Measured α
Predict ed α
D
Chapter 3
60
used in the fitting procedure. At the left, the end of this constraint line is (25.5; 8.5);
for lower c
1
-values the constraint function does not exist. This Figure 8 ends at c
1
=60,
as for higher c
1
values the constraint is violated for the contour lines. For values of n
0

and c
1
in the area below the constraint line, the contours are continuous and smooth.
For values of n
0
and c
1
in the area above the constraint line, the contours are
discontinuous and behave not at all smooth. This behavior results from the constraint
violation, see also appendix A. The 5 and 10% contour lines cover a broad interval for
n
0
and c
1
. For n
0
the minimum and maximum values are 5.4 (0.82 × fitted value) and
11.1 (1.71 × fitted value), and for c
1
these values are 21.75 (0.66 × fitted value) and 60
(1.81 × fitted value), all referring to the 5% contour line. The large range might
suggest that the model seems insensitive to its parameters. However, at a given c
1
the
allowed range for n
0
is limited, as can be seen in Figure 8.

Up to a hold-up of 0.13 a constant n of 4 suffices to describe the data (see Fig 6). This
is in agreement with the results of Godfrey and Slater (1991), who concluded that this
power is between 0 and 4. This power n also compares well with values of n derived
from literature models for liquid fluidization of solid spheres at laminar flow
conditions (Richardson and Zaki (1954), n=3.65; Rowe (1987), n=3.7; Foscolo (1987),
n=3.8; Garside and Al-Dibouni (1977), n=4.09). Apparently, the dodecane droplets
behave as solid spheres at laminar flow conditions. At a high interfacial tension
droplets indeed tend to behave like solids. The interfacial tension between water and
dodecane is 40 mN m
-1
. Such an interfacial tension is regarded as high. So the solids-
like behavior of the droplets is to be expected.

Droplet spray-column
61
Figure 8. Sensitivity analysis for parameter n
0
and c
1
of the presented model. Contour lines of 1%
(black), 5% (dark grey) and 10% (grey) higher values than the SS
res
for n
0
=6.33 and c
1
=33.2 (●),
--- constraint line

The transition from a constant power n of 4 to a decreasing power n might indicate a
deviation from laminar flow to more turbulent flow. For this type of droplet flow,
this transition corresponds with a certain dimensionless drift flux (0.065) or
dispersed-phase hold-up (0.13), instead of a single Reynolds number. This is in
contrast to what is suggested in literature (Wallis, 1969). To illustrate that this
transition does not correspond to a single Reynolds number, but to an endless set of
Reynolds numbers, we calculated the superficial dodecane velocity at different
nozzle geometries, and two column diameters, that will result in a hold-up of 0.13.
Figure 1 shows how this can be achieved; the results are shown in Table III. This
table clearly shows that there is not a constant Reynolds number for the transition
from a constant power n to a decreasing power n.
Chapter 3
62

Table III. Droplet characteristics at the transition from laminar to more turbulent flow for
different sparger geometries (nozzle diameter is 1 mm), and two column diameters:
D
1
=6 cm, D
2
=9 cm.
U
d

(mm s
-1
)
v


(m s
-1
)

d
(mm)
Re
#
nozzles D
1
D
2
D
1
D
2
D
1
D
2
D
1
D
2

50
1
7.3 - 0.098 - 2.3 - 254 -
100 9.3 7.0 0.124 0.094 3.0 2.1 425 230
200 10.8 9.0 0.145 0.120 3.7 2.9 607 394
500 11.8 11.0 0.158 0.147 4.2 3.7 764 631
1
for this sparger, the liquid velocity through one nozzle to obtain a hold-up of 0.13 using a
column with diameter D
2
should be larger than the maximum velocity corresponding with a
nozzle Weber number of 8.64. So no calculations of terminal droplet rise velocity and droplet
diameter could be done.


Comparison with literature models

Different models from literature were evaluated with all the measured droplet hold-
ups (N=912). Their prediction is expressed as the average deviation (AVD), as
defined by equation 5. The results are given in Table IV, and parity plots for the
different models are shown in Figure 7 (the physical constants used are given in
Table II).
It can be concluded from Table IV and Figure 7 that, for the whole range of
measured hold-ups, only the slip-velocity model (Kumar and Hartland, 1985), eq.6,
predicts the measured hold-ups well. Up to a hold-up of 5% the prediction is good; at
larger hold-ups the model slightly underestimates the experimental data. This model
is based on a force balance around a single droplet and a hold-up dependent drag
coefficient to account for the presence of other droplets. This model was derived for a
large number of different 2-phase systems with a wide range of physical constants.
For hold-ups up to 15% the hold-up is well predicted with a slip velocity
according to equation 2 and a power n according to Richardson and Zaki (1954). At
higher hold-ups no solution for the hold-up is possible. The good prediction of the
Richardson and Zaki model is not surprising as the power n is between 4.2 and 4.3
(the value of the power n obtained from fitting the hold-up data up to 13% is equal to
4.0).

Droplet spray-column
63
Table IV. Comparison of measured data with models from literature. Physical constants
are given in Table II.
AVD Remarks
Droplets behave like solids
Richardson and Zaki (1954)
0.0728 Good prediction of the hold-up up to 15%, at
higher hold-ups no solution of the model for the
hold-up

A slip velocity model
Kumar and Hartland

(1985)
0.0797 Good prediction of the data, 68% of the data show
a deviation less than 10%, 95% of the data has a
deviation less than 20%

An explicit model
Kumar and Hartland

(1995)
0.2748 All data points are underestimated


Presented model

0.0750 Good prediction of the data, 72% of the data show
a deviation less than 10%, 97% of the data has a
deviation less than 20%


Validation of the presented model

As mentioned in the experimental part, a different 2-phase system was used to
validate the newly proposed model; the continuous phase is a 60 mM KCl-solution
instead of tap water and dodecane is again used as dispersed phase. The physical
properties are equal to those for a tap-water/dodecane 2-phase system, except for the
interfacial tension, which is equal to 29 mN m
-1
, in contrast to 40 mN m
-1
for the first
2-phase system. Figure 9A shows the measured dimensionless drift-flux, as well as
the predictions of our proposed model, and those from Kumar and Hartland

(1985),
eq.6, as a function of the dodecane hold-up. Figure 9B shows the parity plot for
predicted and measured droplet hold-up for both models. It can be concluded from
Figure 9A-B, that both models predict the droplet hold-up well; the AVD as defined
by equation 5, is 0.119 for the model of Kumar and Hartland

(1985), and 0.115 for the
presented model. At higher hold-ups, however, our model predicts the
dimensionless drift-flux and hold-up better. At these higher hold-ups the model of
Kumar and Hartland overestimates the dimensionless drift flux, and underestimates
the hold-up.




Chapter 3
64
Figure 9
A. Dimensionless drift flux versus dodecane hold-up. Continuous phase is a 60 mM KCl solution. ●
data; — presented model; --- model of Kumar and Hartland, 1985.
B. Parity plot between predicted and measured hold-up. O : model of Kumar and Hartland, 1985; ● :
presented model.


Conclusion

The sparger lay-out influences the dodecane hold-up. A higher hold-up is obtained
with a sparger with less nozzles and the same nozzle diameter, using the same
superficial dodecane velocity. The presented hold-up model predicts well the
dispersed-phase hold-up and related properties like slip velocity and drift-flux. This
model uses the definition of the slip velocity and a Richardson and Zaki type
equation for the slip velocity: an equation for the droplet diameter and an equation
for the terminal rise velocity of a single droplet. Up to a hold-up of 0.13, or a
dimensionless drift-flux of 0.065, a constant n of 4 in the slip-velocity model suffices
to describe the hold-up adequately. At hold-ups higher than 0.13 the parameter n
decreases with an increasing dimensionless drift- flux. A model for this parameter
based on the dimensionless drift-flux is used in the hold-up model, which predicts
the dispersed-phase hold-up from zero to 0.23 well. At hold-ups larger than 0.23 the
dimensionless drift flux becomes constant.
Up to 0.15 the slip velocity model of Richardson and Zaki (1954) for solids
predicts the data well, but for higher hold-ups this model gives no solution. The
model of Kumar and Hartland (1985) is applicable for the whole range of measured
hold-ups; at low hold-ups the prediction is good, at higher hold-ups the model
underestimates the hold-up. The average deviation between data and model (=AVD)
is for the model of Kumar and Hartland 0.080, whereas our model gives an AVD of
0.075.

0
0.025
0.05
0.075
0 0.03 0.06 0.09 0.12
α
Udc*
A
0
0.025
0.05
0.075
0 0.03 0.06 0.09 0.12
α
Udc*
A

0
0.03
0.06
0.09
0.12
0 0.03 0.06 0.09 0.12
α measured
α predicted
B
0
0.03
0.06
0.09
0.12
0 0.03 0.06 0.09 0.12
α measured
α predicted
B
Droplet spray-column
65
Our model was validated with measured data using a different
dodecane/water 2-phase system; all physical properties were the same, except for
the interfacial tension (29 mN m
-1
for the new data versus 40 mN m
-1
for the old
data). The new data were well predicted.


Acknowledgements

This work was financially supported by the Ministry of Economic Affairs, the
Ministry of Education, Culture and Science, the Ministry of Agriculture, Nature
Management and Fishery in the framework of a research program of the Netherlands
Association of Biotechnology Centres (ABON).


Appendix A. Constraints applied in the fitting procedure

The equations describing the dimensionless drift flux (U
dc
*) are:
( )
1 n *
dc
dc
1 U
v
U
+

α − ⋅ α = = A.1
with n given by:
( )
2
c
*
dc 1 0
U c n n − = A.2
The parameter c
2
is given by:
ln(0.0975)
1
c
n n
ln c
1
boundary 0
2










= A.3
with n
boundary
given by :
( ) 1 1 ln
0.0975
ln n
boundary
boundary
boundary









α −








α
= A.4

Parameter c
2
is constant at a given n
0
, c
1
and α
boundary
. Substituting equation A.2 into equation
A.1 yields an implicit equation for U
dc
*. A new function f is defined by rewriting equation
A.1:


( ) 0 1 U f
1 *) n(U *
dc
dc
= α − ⋅ α − =
+
A.5

The constants n
0
and c
1
were fitted with a downhill simplex method as suggested by Nelder
Chapter 3
66
and Mead (1969) without any constraint on the fitting procedure. It is only demanded that
U
dc
* is smaller than 0.0975. This gave a value for n
0
equal to 4.37, and for c
1
equal to 685.
However, as shown in Figure A.1, these fitted constants do not give a good fit at higher
hold-ups; U
dc
* decreases at higher hold-ups and is estimated far too low. At α
boundary
a
discontinuity can be observed.
Figure A.1. The dimensionless drift flux as a function of the hold-up for different combinations of the
parameters n
0
and c
1
. The add-in figure is an enlargement.  n
0
= 685, c
1
= 4.37; --- n
0
= 69.1, c
1
=
5.48; n
0
= 33.2, c
1
= 6.55.

Figure A.2 shows f as a function of U
dc
* (equation A.5). This figure shows that at a given α
two roots exist for U
dc
* that satisfy equation A.5. The smallest of both roots is taken as the
correct solution of equation A.5. When α approaches α
boundary
equation A.5 still gives two
roots for U
dc
*. None of these roots are near the constant dimensionless drift flux. However,
equation A.4 demands that the root of equation A.5 is equal to the constant dimensionless
drift flux (0.0975) at α
boundary
. This explains the discontinuity in Figure A.1.

Figure A.2. Function f versus the dimensionless drift flux for different values of the hold-up  hold-
up is 0.21 ;  -  hold-up is 0.20; … hold-up is 0.19; × hold-up is 0.21, U
dc
*=0.0975, f=0.

In order to avoid the discontinuity, we demand that the first root of the function f at α
boundary

is equal to the constant value of the dimensionless drift flux. This means that the derivative
of function f with respect to the dimensionless drift flux is less than or equal to zero.

0
0.025
0.05
0.075
0.1
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
α
U
dc
*
0.075
0.1
0.15 0.2 0.25 α
U
dc
*

-0.0015
-0.001
-0.0005
0
0.0005
0.001
0.0015
0.002
0.07 0.08 0.09 0.1 0.11
U
dc
*
function f
Droplet spray-column
67
From equation A.6 a maximum value for n
0
can be determined at a given c
1
, and α
boundary
.

( )
( )
( )
( )
( )
( )
( ) 0 U c c ) ln(1 1 1
dU
f d
0
U d
1
*
dc
U d
dU
f d
1 c
*
dc 2 1

2
dc 1 0
*
dc
boundary
*
dc

2
dc 1 0
*
dc
2
c
* U c n 1
c
* U c n 1
≤ − ⋅ α − ⋅ α − α − =
⇒ ≤






α − ⋅ α −
=













− +
− +
A.6

Equation A.1 and A.2 were fitted on the data, together with the constraint given by equation
A.6. This resulted in a n
0
equal to 5.48 and c
1
=69.1 at α
boundary
= 0.21. The dimensionless drift
flux as a function of the hold-up is given in Figure A.1.

This figure clearly shows that the fit is much better for the higher hold-ups; see the
enlargement in Figure A.1. However, around a hold-up of 0.21, the increase of the
dimensionless drift flux with the hold-up is really large, and shows an inflexion point. This
feature of the relationship between the dimensionless drift flux and the hold-up has no
physical meaning. Our data, however, suggest a monotonic increase in drift-flux with the
hold-up. To obtain a relationship between the dimensionless drift and the hold-up which
does not show the sharp increase, another more severe constraint has to applied to the fitting
procedure.

An obvious constraint would be demanding that the first derivative of U
dc
* with respect to α
is equal or larger than 0:
0
d
* U d
dc

α
A.7
Equation A.5 shows that U
dc
* is an implicit function of the hold-up. To obtain the first
derivative of U
dc
* with respect to α implicit differentiation has to be applied (equation A.8).
( )
( ) ( )
0
d
1 ) ( U d
0 1 U
1 n *
dc 1 n *
dc
=
α
α − ⋅ α − α
⇒ = α − ⋅ α −
+
+
A.8
If the numerical operation given in equation A.8 is applied, equation A.9 is obtained
(Almering, 1988) :
( ) ( ) ( )
( )
( )
( ) α − ⋅

⋅ ⋅
+
α − ⋅ α +
α − ⋅ α ⋅ + −
+
α −
=
α
1 ln
1 c2
* U c2 c1
1 n
1 1
n
1 1 n
1 n
1
d
* U d
dc
dc
A.9
This derivative should be larger than or equal to zero. So the nominator should be larger
than or equal to zero. This leads to:
2 n
1
+
≤ α
Chapter 3
68
This equation gives no extra information for the parameters c1 and n0.
However, a maximum α
boundary
can be derived. At the boundary the following two
equations hold.
( )
2 n
1
0.0975 1
boundary
boundary
1 n
boundary boundary
boundary
+
= α
= α − α
+

Solving these equations gives a maximum for α
boundary
: 0.2334. At a higher α
boundary
dU
dc
*/ dα
is negative.

The first derivative should also exist; this means that the denominator should not be equal to
zero. From the denominator of equation A.9 a maximum n
0
can be determined at a given c
1
,
α
boundary
, and U
dc
*=0.0975. Equation A.1 and A.2 were fitted on the data, together with the
constraint discussed above (the denominator in equation A.9 equals zero). This resulted in an
n
0
equal to 6.55 and c
1
=33.2 at α
boundary
= 0.23. The dimensionless drift flux as a function of the
hold-up, equation A.1 and A.2 with the fitted constants, is given in Figure A.1. This figure
shows that the obtained constants give a good fit.


Droplet spray-column
69
Nomenclature

c
1
, c
2
: parameter in equation 6 -
d : diameter droplet m
d
noz
: diameter nozzle m
Eo : Eötvös number,
σ
ρ ∆
2
d g
-
Eo
noz
: Eötvös nozzle number,
σ
ρ ∆
2
noz
d g
-
g : gravity constant (9.8) m
2
s
-1
N : number of data points -
n : power n -
n
0
: parameter in equation 6 -
U
dc
: drift flux m
3
m
-2
s
-1
U
sc
: superficial velocity of the continuous phase m
3
m
-2
s
-1
U
sd
: superficial velocity of the dispersed phase m
3
m
-2
s
-1

V
dj
: drift velocity m s
-1
v

: terminal rise velocity single droplet m s
-1
v
k
: characteristic velocity m s
-1
v
12 :
slip velocity m s
-1
We
noz
: nozzle Weber number,
σ
ρ
∞ noz
2
d
d v
-

Greek
α : dispersed-phase hold-up -
α
meas
: measured dispersed-phase hold-up -
α
mod
: predicted dispersed-phase hold-up -
∆ρ : density difference continuous and dispersed phase kg m
-3
ρ
c
: density continuous phase kg m
-3
ρ
d
: density disperse phase kg m
-3
η
c
: viscosity continuous phase Nm s
-1
σ : interfacial tension N m
-1


70


71

4
Solid and Droplet hold-ups in a liquid-liquid-solid 3-
phase fluidized-bed





Abstract

The hydrodynamics of a 3-phase fluidized bed reactor were studied. In view of
biocatalytic purposes, the reactor featured a bed of κ-carrageenan gel beads as solid
phases. The bed was fluidized by a continuous aqueous phase (30 mM KCl as
continuous aqueous phase), and a disperse organic phase (n-dodecane droplets) rose
through the bed of particles. At different aqueous and n-dodecane velocities, the type
of gel bead flow inside the 3-phase fluidized bed was characterized, and the various
phase hold-ups were measured and studied. Visual observations of the 3-phase
fluidized bed revealed five different flow regimes, characterized by the gel beads
flow: two homogeneous flow regimes, two heterogeneous flow regimes and a region
with wash-out of gel beads. Literature models for different types of 3-phase systems
could not be extended to the system under study and a new model for gel bead and
dodecane droplet hold-up in the fluidized bed was derived, based on stationary force
balances. A drag-force model for a single gel bead and a single droplet in the 3-phase
mixture was derived. All data for gel bead hold-up and for droplet hold-up up to
0.06 were well described. An empirical model was derived to describe the droplet
hold-ups for the entire range of measured hold-ups.



Chapter 4
72
Introduction

Gas-liquid-solid fluidized beds have been abundantly studied and are widely used in
industry. Different modes of operation were classified by Muroyama and Fan (1984).
Most frequently, the solids are fluidized by a rising liquid, while gas bubbles rise
through the bed. The liquid analogue of these systems, i.e. a liquid-liquid-solid
fluidized bed with liquid droplets instead of gas bubbles, has received little attention
in literature. An overview and classification of 3-phase systems is given in Table I, of
which the liquid-liquid-solid 3-phase systems are discussed in more detail below.
Application of a liquid-liquid-solid fluidized bed as extraction column was
firstly studied by Roszak and Gawronski (1979) and Dakshinamurty et al. (1979), who
studied a 3-phase system of water, organic solvent droplets, and a solid like glass
beads. Roszak and Gawronski (1979) focused on mass transfer and found it to be
faster than in a fluidized bed. Dakshinamurty et al. studied solids hold-up (1979), and
mass transfer (1984). In support of Roszak and Gawronski (1979), they found the
mass transfer rate to be higher than in a 2-phase droplet column. Physical aspects of
a similar 3-phase system, with water as continuous phase, and with kerosene and
glass pearls as disperse phases, have been reported by Kim et al. (1988, 1989, 1994,
1999), who also reviewed the disperse phase characteristics (Kim and Kang, 1997).
Here, we report on a different type of 3-phase liquid-fluidized bed, with water
as the continuous phase, and with n-dodecane droplets and polymeric gel beads as
disperse phases. This 3-phase system differs from the above described 3-phase
liquid-liquid-solid system in the type of solid used; gel beads versus glass pearls. The
physical characteristics (density and composition) between both types of solid differ
largely. The density of gel beads is much lower, and gel beads consist over 95%
percentage of water. Gel beads are an interesting type of solid, as they frequently are
used to immobilize a biocatalyst, e.g. enzymes or micro-organisms, thus allowing a
large amount of biocatalyst to be retained in the reactor (Wijffels et al., 1996). Such a
3-phase liquid-fluidized bed may have an advantage over more conventional
bioreactors, especially if the biological reaction is inhibited by toxicity of the reaction
substrate or reaction product (De Bont, 1998; Collins et al., 1995; Hüsken et al., 2001;
Marshall and Wooley, 1995; Tramper et al., 1992). In a liquid-liquid-solid fluidized
bed, the organic solvent droplets may serve as a substrate reservoir or a product sink.
As a result of such an exchange between the liquid phases, the concentration of
substrates and products are kept at sub-inhibitory levels, thus improving the reaction
Hold-up in 3-phase fluidized bed
73
rate significantly.
Application of such a liquid-liquid-solid fluidized bed as an extractive
bioreactor was studied by Davison and Thompson (1993). Clostridium acetobutylicum,
immobilized in gel beads, was used to ferment glucose to acetone and butanol, while
an organic solvent was used for in situ removal of inhibitory butanol. A substantial
increase in butanol production rate was reported. These authors, however, did not
study physical aspects of the 3-phase fluidized bed.
To our knowledge these are the only 3-phase liquid-liquid-solid fluidized bed
systems studied. In this paper we therefore focus on the physical aspects of such 3-
phase liquid fluidized beds. To this end, a 3-phase fluidized bed reactor has been
developed on a pilot scale.
This paper describes observed phenomena, like the bed stability and the type
of the solids flow inside the fluidized bed. Subsequently, droplet and solids hold-ups
at different velocities of water and dodecane are discussed. Hydrodynamics of our 3-
phase fluidized bed have not been studied before. Since available models for the
hold-ups in more or less similar 3-phase systems (gas-liquid-solid: Yu and Rittmann,
1997; liquid-solid-solid fluidized beds: Van der Wielen et al., 1996a) gave an
unsatisfactorily description of our data, a new model was developed. The solids
hold-up followed from the drag force of a single gel bead in the 3-phase mixture. For
this drag force, an existing 2-phase model was successfully adopted with minor
adaptations. For the droplet hold-up, an empirical model was derived from
experimental data. The combination of these models gave a satisfactory description
of the solid-liquid-liquid 3-phase system over a wide range of hold-up values.
Chapter 4
74

Table I. Overview of different 3-phase systems
Reference
Disperse
phase
Disperse
phase
Continuous
phase
Remarks
Muroyama and Fan,
1984

gas solid liquid hydrodynamics
Hunik et al., 1994

gas
(air bubbles)
solid
(reactive
gel bead)
liquid aerobic conversions
Roszak and
Gawronski 1979,
Dakshinamurty et al.
1979,
Kim et al. 1994



liquid
(organic
droplets)


solid
(glass beads)


liquid
(water)


hydrodynamics
Davison and
Thompson, 1993


liquid
(organic
droplets)
solid
(reactive
gel beads)
liquid
(water)

bioreaction
Van der Wielen et al.,
1996b

solid solid liquid reaction and
hydrodynamics
Michielsen et al. 2001



solid
(substrate
crystals)
solid
(product
crystals)
liquid
(water)
reaction not performed
in a packed or
fluidized bed
this study liquid
(organic
droplets)
solid
(gel beads)
liquid


Material and Methods

Gel beads
Celite™-enforced κ-carrageenan gel beads were used as disperse solid phase; they
were produced with a resonance nozzle (Hunik and Tramper, 1993). An aqueous 3.25
% κ-carrageenan solution (Genugel 0909 Copenhagen Pectin Factory), mixed with 10
w/w% Celite™, kept at 35 °C, was forced through a 0.8-mm nozzle at 202 Hz and 1.5
bar. Drops were collected in a 80-mM KCl solution for hardening. To obtain spherical
beads, butyl acetate (Aldrich) was layer upon the hardening solution. After
hardening for approximately two hours, the beads were stored in a 30-mM KCl
solution. The same KCl solution was used as continuous phase; it prevented the
Hold-up in 3-phase fluidized bed
75
elution of counter-ions from the gel beads. Density and diameter of the gel beads
were determined as described by Van Zessen et al. (2001). Physical properties of the
gel beads are given in Table II.

Table II. Phase properties in the 3-phase fluidized bed
gel beads n-dodecane 30 mM KCl
density (kg m
-3
) 1065 742.7 998
viscosity (mN s
-1
) - - 0.93
diameter ( mm ) 2.80 flux dependent -
settling velocity (cm s
-1
) 5.12 flux dependent -
surface tension 30 mM KCl/ n-dodecane (mN m
-1
) 40.0


Fluidized bed
The fluidized bed (Figure 1) consisted of a 2-m high and 6-cm diameter glass column.
On top of it, a widened section facilitated coalescence of the organic phase droplets
and separate discharge of the organic solvent (n-dodecane, Aldrich 99%+ pure) and
aqueous (30 mM KCl) phases. n-Dodecane was introduced at the bottom of the
column through a liquid distributor (Figure 2). It consisted of a 10 cm high glass
column filled with 8 mm glass pearls for a good distribution of the flow over the
flow-through column area, and with a sparger on top. The sparger consisted of 209
nozzles protruding through a Teflon plate with a diameter of 6 cm. The nozzles (5 cm
length and 1 mm internal diameter), were placed sufficiently far apart to prevent
contact between the growing droplets. As continuous phase, a 30-mM KCl solution
was introduced through four openings in the column wall, located below the nozzles
outlets (Figure 2). An equal flow rate through each of these continuous phase inlets
was obtained with throttle valves; this was verified with color-pulse experiments.
The storage vessels of both liquids (Figure 2) were maintained at 26 °C. A
temperature gradient over the column was not observed.

Solids hold-up
Solids hold-up (ε
s
) was calculated from the measured bed height (H
bed
, see also
Figure 1), the known total solids volume (V
s
), and the known column cross-sectional
area (A).
A H
V

bed
s
s
= ε 1

Chapter 4
76
Bed height was easily determined as there was a distinct transition between the 3-
phase and 2-phase regions in the upper part of the column (Figure 2). Total solids
volume was determined as described by Van Zessen (2001).

Figure 1. Experimental set-up. 1 column; 2 phase
separator; 3 sparger; 4 rotameters; 5 dodecane
storage; 6 30 mM KCl storage; 7 pressure
measurement device. • gel beads, • droplets; H
bed

height of the 3-phase fluidized bed

Figure 2. Sparger lay-out. 1
dodecane inlet, 2 glass pearls,
3 Teflon™ plate, 4 nozzles, 5
water inlet



Dodecane hold-up
At steady state conditions, the pressure drop (∆P) between two points a distance ∆z
along the column apart is equal to the total weight per unit of cross-sectional area of
the bed inbetween these points (Muroyama and Fan, 1984):

( ) z g P
c c d d s s
∆ ε ρ + ε ρ + ε ρ = ∆ 2

where ρ
s
, ρ
d
and ρ
c
are the densities of the solid gel beads, the dodecane droplets
and the continuous phase (30 mM KCl). Measurement of the pressure difference
between both points involved an external loop filled with continuous phase only
(Figure 1). The actually measured pressure difference (∆P
exp
) is therefore equal to:

( ) z g z g P P P
c c c d d s s 2 1 exp
∆ ρ − ∆ ε ρ + ε ρ + ε ρ = − = ∆ 3

As the sum of the various phase hold-ups is unity,
7
4
5
1
2
4
6
∆P
3
H
bed
P
1
P
2
2
1
3
4
5
3
4
5
Hold-up in 3-phase fluidized bed
77

1
c d s
= ε + ε + ε 4

the droplet hold-up follows from solving equations 3 and 4.

Pressure difference was measured over a vertical distance of 25 cm, at 24 cm and 49
cm above the sparger. A Validyne DP103 pressure transducer with a CD23 digital
transducer indicator (maximum pressure difference 140 Pa) was used. For each set of
measurements, the transducer was calibrated with a liquid manometer with water
and octanoic acid (ρ =903 kg m
-3
). As rising droplets lead to pressure fluctuations, the
analogue signal was monitored with a personal computer and an A/D converter
until a total of 1000 data points were collected that were normally distributed. The
mean of this distribution was used as a measure for ∆P
exp
.
Similar pressure measurements were also done in the 2-phase solids-free top
section of the column above the 3-phase fluidized beds (positions at 175 cm and 190
cm above the sparger).

Miscellaneous
Flow rates of the KCl solution (0.58 to 2.06 cm s
-1
) and n-dodecane (0 to 0.91 cm s
-1
)
were measured with calibrated rotameters (Sho-Rate).

Density of both liquids was measured with a pyknometer, total volume 25.00 ml. The
static interfacial tension between the continuous phase and n-dodecane phase was
determined by the Wilhelmy plate method (Hiemenz, 1986). Table II summarizes the
physical properties of both liquids.

Values for model parameters were obtained from a Gauss-Newton algorithm that
minimized the residual sum of squares.


Chapter 4
78
Results

First, observations on the nature of the 3-phase fluidized bed will be described and
discussed. Thereafter, the hold-ups of the different phases will be shown. It will be
demonstrated that literature models for comparable 3-phase systems do not predict
these hold-ups correctly. A new model describing the hold-ups correctly will be
presented.

Visual Observations
A large variety in water and dodecane fluxes was applied to study the behavior of
the fluidized bed. In the absence of dodecane, a minimal water flux (0.36 cm s
-1
) was
required to obtain fluidization; below this flux, the gel beads formed a packed bed.
At a water flux exceeding 5 cm s
-1
, the gel beads were washed out and a 2-phase
water/ dodecane system remained. At large dodecane fluxes (not applied in our
experiments), phase inversion will occur; the continuous phase will be dodecane, and
the dispersed phase will be water.
Inspired by visual observations on the behavior of the 3-phase fluidized bed,
five different regions were identified (Figure 3), but the transition from one region to
another region was not sharp. Below, only 3-phase fluidized bed regions are briefly
discussed.

Homogeneous flow pattern: mimic of a 2-phase fluidized bed (region 1)
In this region, the flow pattern of gel beads was not disturbed by the droplets. The
gel bead movement did not differ from its movement in a 2-phase fluidized bed. At
the top of the bed, little bead ‘eruptions’ were observed when dodecane droplets
escaped. The gel beads, however, fell back in the fluidized bed, and a so-called free
board was not formed.

Hold-up in 3-phase fluidized bed
79
Figure 3. Flow map for a 3-phase fluidized bed with gel beads, (ρ
s
=1067 kg m
-3
and d
s
=2.8 mm), and
dodecane droplets (ρ
d
=743 kg m
-3
) in an aqueous continuous phase. 1 Homogeneous flow pattern:
mimic of a 2-phase fluidized bed; 2 Homogeneous flow pattern: turbulent solids flow; 3 Heterogeneous
flow pattern: structured solids flow; 4 Heterogeneous flow pattern: unstructured chaotic solids flow; 5
Droplet coalescent region: slug formation


Homogeneous flow pattern: turbulent solids flow (region 2)
Characteristic for this region is bed expansion. The height of the 3-phase fluidized
bed always exceeded the 2-phase fluidized-bed. At none of the fluxes, a significant
freeboard of solids was formed. Gel beads moved randomly through the fluidized
bed in little groups, like aggregates. The velocity of the solid groups was much
higher in the 3-phase fluidized bed as compared to a 2-phase bed; due to a
continuous change in direction the group motion was much more chaotic (turbulent).
The size of these groups did not alter in time. However, at higher water fluxes these
aggregates became increasingly smaller, until at a water flux of ~ 2 cm s
-1
the gel
beads moved individually. The overall picture was not influenced by the dodecane
flux, although the solids group flow became more pronounced for the lower
dodecane fluxes.

Heterogeneous flow pattern: structured solids flow (region 3)
A clear characteristic of this region is bed contraction: the bed height of the 3-phase
fluidized bed was smaller as compared with the 2-phase fluidized bed. Another
characteristic for this region is the gel bead flow. The gel bead flow is characterized
by repetitive waves of solids concentrations traveling through the fluidized bed

U



c
m

s
-
1
d
0.2
0.4
0.6
0.8
0.5 0.75 1 1.25 1.5
U
c
cm s
-1
5
1
4
2
1
0.0
0
5.0
3
P
a
c
k
e
d
b
e
d
s
p
r
a
y
c
o
l
u
m
n
phase inversion
Chapter 4
80
(Figure 4). Wash-out of the gel beads did not occur, and a freeboard was not formed.
This gel bead flow behavior can be explained with the droplet flow. For the
dodecane fluxes in this region, dodecane droplets are released group wise from the
tip of the nozzles. Such a swarm of droplets rises then through the fluidized bed and
pushes the gel beads upwards (thus increasing the solids concentration). As the
droplet velocity is larger than the solids velocity, the swarm of droplets passes this
high-concentration area. No other swarm of droplets follows immediately, and the
gel beads fall back (decrease of the solids concentration). A subsequent swarm of
droplets arrives and the cycle is repeated.

Figure 4. Schematic gel bead flow pattern through a 3-phase fluidized bed in heterogeneous flow
region: structured solids flow. Dark areas represent high solids concentration, white areas are low
solids concentrations.

Heterogeneous flow pattern: unstructured chaotic solids flow (region 4)
Behavior in this region could not be characterized by a single key feature. An overall
characteristic, however, was the very heterogeneous axial and radial distribution of
the gel beads. The flow pattern of both gel beads and droplets was random and
chaotic. Locally, the solvent droplets pushed the gel beads towards one side of the
column, thus creating high-voidage pockets. Droplets entering into such pockets rose
very fast, and entailed some gel beads. Once gel beads approached each other
closely, they stopped moving and fell back, tumbling chaotically.
At dodecane fluxes above 0.8 cm s
-1
and water fluxes below 0.8 cm s
-1
, droplet
coalesce was observed to some extent. As opposed to full coalescence, full slug
formation was not observed, as smaller slugs were rapidly disintegrated.

Droplet coalescent region: slug formation (region 5)
At the bottom of the column, dodecane droplets rose very slowly through the
fluidized bed, and the droplets start to coalesce. Eventually at a certain height,
dodecane slugs spanning the column diameter were formed. These slugs pushed the
increase in time
Hold-up in 3-phase fluidized bed
81
full bed of gel beads upwards out of the column. In this region a fluidized bed could
not exist.


Particle and droplet hold-ups
Dodecane droplet hold-up as a function of the water flux at various dodecane fluxes
is shown in Figure 5a; gel bead hold-up as a function of the water flux at various
dodecane fluxes is shown in Figure 5b.

Droplet hold-up
The droplet hold-up was found to increase with the dodecane flux, and to decrease
with the water flux, (Figure 5a), which is in full agreement with the findings of Kim
et al. (1999). For the higher dodecane velocities (U
d
≥ 0.44 cm s
-1
), hold-ups were only
measured in region 2 (homogeneous turbulent solids flow). For the lower dodecane
velocities, hold-ups were measured in region 2 and region 3 (heterogeneous
structured solids flow). The transition from region 3 to region 2 had no effect on the
dodecane hold-up, which was found to increase smoothly with the decreasing water
flux.

At dodecane fluxes above 0.36 cm s
-1
and water fluxes below 1.25 cm s
-1
, the 3-phase
fluidized-bed is in the unstructured chaotic flow region 4 (Figure 3), and thus
features prominent axial and radial distributions of solids and droplets.
Consequently, hold-ups could not be measured unequivocally in this region.

Solids hold-up
The solids hold-up was found to decrease with increasing water fluxes (Figure 5b),
both in 2-phase and 3-phase fluidized beds. At water fluxes above 0.75 cm s
-1
and
dodecane fluxes above 0.42 cm s
-1
, the solids hold-up in the 3-phase system was
always smaller (i.e. the bed was higher) than in a 2-phase system (U
d
=0) at the same
water flux. Similar results have been reported for a water-kerosene-glass pearl
fluidized bed (Dakshinamurty et al., 1979; Kim et al. 1989).
At water fluxes below 0.75 cm s
-1
the bed height decreased with the dodecane
flux (bed contraction, data not shown). This phenomenon is not unique: Muroyama
and Fan (1984), who reviewed gas-liquid-solid fluidization, cite various authors who
observed it. Yu and Rittmann (1997), who studied a gas-liquid-solid fluidized bed
with and without biofilms, also reported on bed contraction. For liquid-liquid-solid
Chapter 4
82
fluidization the phenomenon was observed before by Dakshinamurty et al. (1979)
and Kim et al. (1989).
According to figure 4, at lower dodecane fluxes (U
d
≤ 0.27 cm s
-1
), the bed will
transform from an expanded state,
phase 2
s
phase 3
s
− −
ε ≥ ε (region 3), to a contracted state,
phase 2
s
phase 3
s
− −
ε ≤ ε (region 2), upon decreasing the water flux. This transition was
indeed observed (data not shown).

Figure 5a. Droplet hold-up as a function of
water and dodecane fluxes. Numbers refer to
the different flow regions.
Figure b. Gel bead hold-up as a function of
the water and dodecane flux.
● U
d
= 0.05 cm s
-1
, ◊ U
d
= 0.10 cm s
-1
, ● U
d
= 0 cm s
-1
, ◊ U
d
= 0.10 cm s
-1
,
■ U
d
= 0.18 cm s
-1
, O U
d
= 0.27 cm s
-1
, ■ U
d
= 0.29 cm s
-1
, ♦ U
d
= 0.42 cm s
-1
,
▲ U
d
= 0.44 cm s
-1
, □ U
d
= 0.54 cm s
-1
, O U
d
= 0.54 cm s
-1
, ▲ U
d
= 0.75 cm s
-1
,
♦ U
d
= 0.73 cm s
-1
, × U
d
= 0.91 cm s
-1
□ U
d
= 0.91 cm s
-1



Figure 6 shows the dodecane hold-up and the gel bead hold-up for various dodecane
fluxes. When the gel bead hold-up increased (i.e. when the water flux decreased) the
dodecane hold-up increased as well. When compared to a 2-phase droplet column of
dodecane in water (ε
s
=0), the mere presence of gel beads in a 3-phase fluidized bed
had a positive effect on droplet hold-up:
0
s
d
0
s
d

= >
>
ε ε
ε ε in all cases.


0
0.02
0.04
0.06
0.08
0.1
0.5 0.75 1 1.25 1.5 1.75 2
U
c
cm s
-1
ε
d
4
3
2
1

0.15
0.25
0.35
0.45
0.55
0.5 0.75 1 1.25 1.5 1.75 2
U
c
cm s
-1
ε
s
Hold-up in 3-phase fluidized bed
83

Figure 6. Droplet hold-up (ε
d
) as a function of solids hold-up (ε
s
) and dodecane fluxes. Points on the
ε
d
-axis are calculated from equation 14 (assuming a water flux of 2 cm s
-1
).
● U
d
= 0.05 cm s
-1
, ◊ U
d
= 0.10 cm s
-1

■ U
d
= 0.18 cm s
-1
, O U
d
= 0.27 cm s
-1

▲ U
d
= 0.44 cm s
-1
, □ U
d
= 0.54 cm s
-1

♦ U
d
= 0.73 cm s
-1
, × U
d
= 0.91 cm s
-1



Available models for solids and droplet hold-up
3-Phase hold-up models are available in literature, although not for liquid-liquid-
solid (gel bead) fluidized beds. For gas-liquid-solid fluidized beds, Muroyama and
Fan (1984) gave an extensive overview of models, which were either empirical or
based on the wake model. A more fundamental model for the gas hold-up in a
liquid-solid-gas fluidized bed was derived by Chen and Fan (1990). The solids hold-
up was assumed to be known. Yu and Rittmann (1997) presented a model for all
three hold-ups in a gas-solid-liquid fluidized bed, based on the wake model and with
an empirical model for the gas hold-up. For a solid-solid-liquid 3-phase system, Van
der Wielen et al. (1996a) presented a model for predicting all phase hold-ups.

Droplet hold-up
We tested the model of Chen and Fan (1990) and Yu and Rittmann (1997) for the
droplet hold-up. Although gas bubbles differ in physical characteristics from
dodecane droplets, the fundamental force balances are the same. In these calculations
we used the measured solid hold-ups. It appeared that both models failed in
predicting the droplet hold-up. The model of Chen and Fan largely underestimated
the dodecane hold-ups below 0.03, and overestimated dodecane hold-ups above 0.06.

0
0.02
0.04
0.06
0.08
0.1
0 0.2 0.4 0.6
ε
s
ε
d
Chapter 4
84
In the intermediate region (0.03 – 0.06), the prediction was fair. When the model
parameters were fitted to our data, the description did not improve. The model of Yu
and Rittmann highly overestimated all hold-ups. When the model parameters were
fitted to our data, the hold-ups were well predicted (81% of all data showed a
deviation less than 25%).
For a different 3-phase system, a solid-solid-liquid fluidized bed, Van der
Wielen et al. (1996a) presented a model for all hold-ups. This model uses parameters
of the individual 2-phase systems. Using these parameters, low droplet hold-ups
(<0.02) were well predicted, but the higher hold-ups were largely overestimated.

Solids hold-up
When the liquid-liquid-solid model of Van der Wielen was used to predict the solid
hold-up in our 3-phase fluidized bed, hold-ups up to 0.3 were well predicted, but the
higher hold-ups were systematically underestimated (measured parameters of the 2-
phase liquid-solid fluidized bed were used). The model of Yu and Rittmann largely
overestimated the hold-up. When model parameters were fitted on our data, the
description improved. However, the fitted parameters resulted in a negative wake
volume, which has no physical meaning.

Hold-up in 3-phase fluidized bed
85
Table III and IV summarize the above described results.

1
For ε
s
→ 0 the 3-phase fluidized bed simplifies to a droplet column. Thus, this
model describes the hold-up in a droplet column. Coefficient m2 is equal to 5 for
droplet hold-ups less than 0.15, as shown by Van Zessen et al. (2003)

2
For ε
s
→ 0 the 3-phase fluidized bed simplifies to a droplet column, and coefficient
n
d
=5, for droplet hold-ups less than 0.15 as shown by Van Zessen et al. (2003).

3
The average deviation (AVD) is defined as:
( ) ( )/N AVD
1..N i
i measured, i model, i measured, ∑
=
ε ε − ε =
10% is the percentage of data with an AVD less or equal to 10%; 25% be the
percentage of data with an AVD less or equal to 25%.

4
The parameters in this model are fitted together with the parameter in the gel bead
hold-up model based on the total gel bead drag coefficient model in Table IV.
T
a
b
l
e

I
I
I
.

M
o
d
e
l
s

f
o
r

t
h
e

d
r
o
p
l
e
t

h
o
l
d
-
u
p

i
n

a

3
-
p
h
a
s
e

f
l
u
i
d
i
z
e
d

b
e
d

D
r
o
p
l
e
t

H
o
l
d
-
u
p

R
e
f
e
r
e
n
c
e

E
q
u
a
t
i
o
n

R
e
s
u
l
t
s
3

R
e
m
a
r
k
s



A
V
D

1
0

%

2
5
%


C
h
e
n

a
n
d

F
a
n

1
9
9
0

m
2
d
c
c
m
1
s
c
c
d
,
s
c
d
d
d
v
1
U
U
U
   
   
ε
+
ε
ε
   
   
ε
+
ε
ε
+
ε

+
=
ε


w
i
t
h

m
1
=
1
.
0
4
5
,

m
2
=
5
1

2
4

3
2

6
4

L
o
w
e
r

h
o
l
d
-
u
p
s

a
r
e

u
n
d
e
r
e
s
t
i
m
a
t
e
d
,

w
h
e
r
e
a
s

h
i
g
h
e
r

h
o
l
d
-
u
p
s

a
r
e

l
a
r
g
e
l
y

o
v
e
r
e
s
t
i
m
a
t
e
d
.

I
n

t
h
e

i
n
t
e
r
m
e
d
i
a
t
e

r
e
g
i
o
n
,

~

0
.
0
3



0
.
0
6

,

t
h
e

p
r
e
d
i
c
t
i
o
n

i
s

f
a
i
r

w
i
t
h

c
o
n
s
t
a
n
t

m
1

f
i
t
t
e
d

o
n

o
u
r

d
a
t
a

m
1
=
0
.
5

2
6

2
1

5
4

F
i
t
t
i
n
g

d
o
e
s

n
o
t

i
m
p
r
o
v
e

t
h
e

d
e
s
c
r
i
p
t
i
o
n
:

l
o
w
e
r

h
o
l
d
-
u
p
s

a
r
e

u
n
d
e
r
e
s
t
i
m
a
t
e
d
;

h
i
g
h
e
r

h
o
l
d
-
u
p
s

a
r
e

l
e
s
s

o
v
e
r
e
s
t
i
m
a
t
e
d

Y
u

a
n
d

R
i
t
t
m
a
n
n

1
9
9
7

0
.
5
s
d
s
c
d
d
d
1
U
1
.
4
8
8
0
.
1
0
1
6
1
U
U
U
   
   
ε

+
+
ε

+
=
ε

1
5
0

0
.
4

1
.
2

a
l
l

h
o
l
d
-
u
p
s

a
r
e

h
i
g
h
l
y

o
v
e
r
e
s
t
i
m
a
t
e
d

w
i
t
h

c
o
n
s
t
a
n
t
s

f
i
t
t
e
d

o
n

o
u
r

d
a
t
a

0
.
5
s
d
s
c
d
d
d
1
U
4
5
.
3
2
3
.
4
1
U
U
U
   
   
ε

+
+
ε

+
=
ε

1
4

5
1

8
1

F
i
t
t
i
n
g

r
e
s
u
l
t
s

i
n

a

m
u
c
h

b
e
t
t
e
r

d
e
s
c
r
i
p
t
i
o
n
.

A
l
l

h
o
l
d
-
u
p
s

a
r
e

w
e
l
l

d
e
s
c
r
i
b
e
d
.

V
a
n

d
e
r

W
i
e
l
e
n

e
t

a
l
.

1
9
9
6

3
.
8
c
4
.
8
/
n
d

c
c
d
d
c
c
d
m
i
x
d
d
v
U
U


ε

      
      
   
   
ε

ε
ε
=
ρ

ρ
ρ

ρ

a
n
d

c
m
i
x
ρ
ε
ρ
ε
ρ
ε
ρ
c
s
s
d
d
+
+
=
,

n
d
=
5
2

7
1

6
.
3

1
6

u
p

t
o

0
.
0
2

t
h
e

h
o
l
d
-
u
p
s

a
r
e

w
e
l
l

p
r
e
d
i
c
t
e
d
.

T
h
e

h
i
g
h
e
r

h
o
l
d
-
u
p
s

a
r
e

l
a
r
g
e
l
y

o
v
e
r
e
s
t
i
m
a
t
e
d
.

T
h
i
s

p
a
p
e
r
:

D
r
a
g

c
o
e
f
f
i
c
i
e
n
t

b
a
s
e
d

0
6
.
0
d
,
c
d
d
d
c
c
s
s
v
v
1


   
   
=
ε

+
ρ

ρ
ρ

ρ
ε

1
6
.
5

3
4

8
1

T
h
e

n
u
m
b
e
r
s

i
n

r
e
s
u
l
t
s

c
o
l
u
m
n

r
e
f
e
r

t
o

a

h
o
l
d
-
u
p

u
p

t
o

0
.
0
6
.

A
t

h
i
g
h
e
r

h
o
l
d
-
u
p
s

t
h
e

d
a
t
a

a
r
e

u
n
d
e
r
e
s
t
i
m
a
t
e
d

E
m
p
i
r
i
c
a
l

0
.
7
6
-
d
0
.
8
8
d
d
,
c
d
U

3
7
.
5
v
v
ε

=


1
3

5
5

8
9

A
l
l

h
o
l
d
-
u
p
s

a
r
e

w
e
l
l

d
e
s
c
r
i
b
e
d

E
m
p
i
r
i
c
a
l


2
4

0
.
7
6
-
d
0
.
9
0
d
d
,
c
d

U
1
5
.
6
v
v
ε

=


1
2

6
3

8
9

A
l
l

h
o
l
d
-
u
p
s

a
r
e

w
e
l
l

d
e
s
c
r
i
b
e
d

86
T
a
b
l
e

I
V
.

M
o
d
e
l
s

f
o
r

t
h
e

g
e
l

b
e
a
d

h
o
l
d
-
u
p

i
n

a

3
-
p
h
a
s
e

f
l
u
i
d
i
z
e
d

b
e
d
.

G
e
l

b
e
a
d

H
o
l
d
-
u
p

R
e
f
e
r
e
n
c
e

E
q
u
a
t
i
o
n

R
e
s
u
l
t
s
1

R
e
m
a
r
k
s



A
V
D

1
0
%

2
5
%


Y
u

a
n
d

R
i
t
t
m
a
n
n

1
9
9
7

(
)
d
1
/
n
1
d
d
1
/
n
s
,
d
s
, c
c
k
k
1
v
U
k
v
U
ε
+
ε

ε


   
   

=
ε




)

5
.
0
8
e
x
p
(

3
.
5
k
d
3
c
ε

ε
=
,

n
=
2
.
3
5
,

v

s
=
3
.
7
7

c
m
/
s

3
3

1
3

4
2

t
h
e

g
e
l

b
e
a
d

h
o
l
d
-
u
p

i
s

h
i
g
h
l
y

o
v
e
r
e
s
t
i
m
a
t
e
d

w
i
t
h

c
o
n
s
t
a
n
t
s

f
i
t
t
e
d

o
n

o
u
r

d
a
t
a

)

1
7
.
3
4
e
x
p
(

0
1
.
5
k
d
3
c
ε

ε

=

1
6

3
5

8
2

F
i
t
t
i
n
g

r
e
s
u
l
t
s

i
n

a

b
e
t
t
e
r

d
e
s
c
r
i
p
t
i
o
n

o
f

t
h
e

d
a
t
a
,

b
u
t

t
h
e

d
a
t
a

a
r
e

s
t
i
l
l

o
v
e
r
e
s
t
i
m
a
t
e
d
.

M
o
r
e
o
v
e
r
,

w
i
t
h

t
h
e

f
i
t
t
e
d

c
o
n
s
t
a
n
t
s

a

n
e
g
a
t
i
v
e

v
o
l
u
m
e

r
a
t
i
o

i
s

p
r
e
d
i
c
t
e
d
.

A

n
e
g
a
t
i
v
e

v
o
l
u
m
e

r
a
t
i
o

h
a
s

n
o

p
h
y
s
i
c
a
l

m
e
a
n
i
n
g
.

V
a
n

d
e
r

W
i
e
l
e
n

e
t

a
l
.

1
9
9
6

(
)
3
.
8
c
4
.
8
/
n
s
,
c
c
s
m
i
x
s
s
v
U


ε

   
   

ρ

ρ
=
ρ

ρ

a
n
d
c
c
s
s
d
d
m
i
x
ρ
ε
+
ρ
ε
+
ρ
ε
=
ρ
,

n
s
=
2
.
3
5
2
,


v

s
=
3
.
7
7

c
m
/
s

7

7
5

1
0
0

H
o
l
d
-
u
p
s

u
p

t
o

0
.
3

a
r
e

w
e
l
l

p
r
e
d
i
c
t
e
d
.

T
h
e

h
i
g
h
e
r

h
o
l
d
-
u
p
s
,

a
r
e

s
y
s
t
e
m
a
t
i
c
a
l
l
y

u
n
d
e
r
e
s
t
i
m
a
t
e
d
.


T
h
i
s

p
a
p
e
r

d
r
a
g

c
o
e
f
f
i
c
i
e
n
t

b
a
s
e
d

(
)
5
1
.
6
d
-
1
.
2
6
s

,
c
s
2
c
s
s
,
s
c
s
m
i
x
1
v
v
v
v
2
4
.
1




ε

   
   
=
   
   
ρ

ρ
ρ

ρ

7
.
2

7
6
.
5

9
7
.
2

H
o
l
d
-
u
p
s

a
r
e

w
e
l
l

p
r
e
d
i
c
t
e
d

o
v
e
r

t
h
e

w
h
o
l
e

r
a
n
g
e


87
Chapter 4
88
None of the available models thus gave a fair description of the whole range of both
solvent and solids hold-up. Consequently, a new model was derived.


New model for gel bead and droplet hold-up

As a basis to a new model for all three phase hold-ups, drag coefficients for single
droplets or gel beads in a 3-phase mixture were calculated from the experimental
data on hold-ups and liquid fluxes. Subsequently, a model was developed to
describe these drag coefficient data. This model, finally, was used to describe the
droplet and gel bead hold-up. This model was based on a steady state force balance
for a single particle in a multi-component system, in which one species of particles
settled at constant velocity in a fluidized bed of a second species of particles (Van der
Wielen et al., 1996a). It was assumed that the time-averaged velocity of the gel beads
is zero.

drag buoyancy gravity
F F F + = 5

where F
drag
is the total drag force experienced by a single particle from its mixed 3-
phase surroundings. This drag force is the sum of all liquid-particle and particle-
particle drag forces, including the drag forces originating from like particles. As in
general the contributions of both fractions cannot be distinguished, a total drag force
for a single particle is used (Van der Wielen et al., 1996a). The total drag force for a
single particle in a 3-phase mixture is given by a general equation:

( )
d s d
2
ci c 2
1
2
4
1
drag
, C v d F ε ε ⋅ ρ ⋅ π = 6

where d is the particle diameter, v
ci
is the slip velocity between water and particle
and C
di

s

d
) is the total drag coefficient for a single particle in a 3-phase mixture.
Combining equations 5 and 6 and evaluating gravity and buoyancy force gives for a
gel bead:

( ) ( )
d s s d,
2
cs c 2
1
2
s 4
1
mix s
3
s 6
1
, C v d - g d ε ε ρ π = ρ ρ π 7a

Hold-up in 3-phase fluidized bed
89
and for a dodecane droplet:

( ) ( )
d s d d,
2
cd c 2
1
2
d 4
1
mix d
3
d 6
1
, C v d - g d ε ε ρ π = ρ ρ π 7b

The bulk density ρ
mix
is determined by all phases (Duijn and Rietema, 1982):

c c s s d d mix
ρ ε + ρ ε + ρ ε = ρ 8

The slip velocities between water/ particle or water/ droplet are defined as the
velocity differences between water and particle and between water and droplet.
Water and droplet flows were assumed to be strictly vertical, while the solids flow
was assumed zero. Based on continuity, the slip velocities thus should equal:

Slip velocity droplets/water :
c
c
d
d
cd
U U
v
ε

ε
= 9a
Slip velocity gel beads/water :
c
c
cs
U
v
ε
= 9b
The gel bead diameter (d
s
) is given in Table II. The droplet diameter (d
d
) is calculated
with the equation of Kumar and Hartland (1996):

1
315 . 0
2
noz d 0.73
noz
0.33
noz
noz
d
gd
0.0393We 0.55Eo
d
d


















σ
ρ
+ = 10

and the rise velocity can be calculated with Vignes' equation (Godfrey and Slater,
1994):
















η
ρ








ρ
ρ ∆
=

6
Eo
1
g
4.2
d
v
3 / 1
c
c
3 / 2
c
d ,
11

The validity of equations 10 and 11 for the present case might be argued, as they
were developed for a droplet column. However, the hold-up in the column section
above the 3-phase bed (a droplet column equivalent) could be well predicted with
equations 9a, 10, 11 and 14, assuming the droplet diameter to be the same throughout
the whole column. Based on the validity of these hold-up predictions, equations 10
Chapter 4
90
and 11 were adopted for the present 3-phase fluidized bed.

The total drag coefficient in a 3-phase mixture for a droplet (C
d,d
) and for a single gel
bead (C
d,s
) were calculated from the measured hold-ups and the water and dodecane
fluxes by equations 4 and 7-11.
Figure 7a shows the total droplet drag coefficient in a 3-phase mixture as a
function of the dimensionless slip velocity between droplets and water, which was
defined as the actual slip velocity divided by the rise velocity of a single droplet in
stagnant water (equation 11); its value ranges from zero to one. Drag coefficients
were classified according to the measured solids hold-up. Figure 7a reveals two
interesting features. The behavior of the total drag coefficient (C
d,d
) for the friction
between a droplet and its surroundings (gel beads, droplets and 30 mM KCl mixture)
was found to be fully accounted for by its dimensionless slip velocity v
cd
`. There is no
influence of the solids hold-up ε
s
other than v
cd
`= f (ε
s
). In, addition the total droplet
drag coefficient in a 2-phase water/dodecane droplet column follows directly the
behavior of the total droplet drag coefficient in a 3-phase fluidized-bed (Figure 7a,
ε
s
=0).
As opposed to the behavior of the total droplet drag coefficient C
d,d
, the
behavior of the drag coefficient C
d,s
for the friction between a single gel bead and its
surroundings (droplets, gel beads and 30 mM KCl) was not fully accounted for by
variations in v
cs
’ (Figure 7b). Apart from v
cs
’=f (ε
d
), there is an additional dependency
of C
d,s
on ε
d
: a larger droplet hold-up was found to yield a relatively high C
d,s
value.
Hold-up in 3-phase fluidized bed
91

Figure 7a. Single droplet drag coefficient in a
3-phase mixture of gel beads, dodecane
droplets and 30mM KCl as function of the
dimensionless slip-velocity.
Figure 7b. Single gel bead drag
coefficient in a 3-phase mixture of gel
beads, dodecane droplets and 30 mM KCl
as function of the dimensionless slip-
velocity.
O ε
s
= 0 ● ε
s
= 0.17 – 0.20 ♦ ε
d
= 0 ◊ ε
d
= 0.0125
◊ ε
s
= 0.20 – 0.25 ♦ ε
s
= 0.25 – 0.30 ● ε
d
= 0.0275 O ε
d
= 0.0475
ε
s
= 0.30 – 0.35 ▲ ε
s
= 0.35 – 0.40 ■ ε
d
= 0.0725 □ ε
d
= 0.09
□ ε
s
= 0.40 – 0.45 ■ ε
s
= 0.45 – 0.50
× ε
s
= 0.50 – 0.56
The drawn line is the prediction with equation 12.

A model was developed for the total drag coefficients of droplets and gel beads in a
3-phase mixture. This model used features demonstrated in Figure 7. Similar to Van
der Wielen et al. (1996a), the model was based on force balances for a single droplet
or gel bead in a mixture of droplets, gel beads and continuous liquid. Different
correlations for the various total drag coefficients were used, however, as the
correlations of Van der Wielen et al. (1996a) did not gave satisfactory predictions. The
newly derived models for the drag coefficients were evaluated for predicting the
various hold-ups.

Model for the droplet hold-up
In a 3-phase fluidized bed, the bed height is easily measured, and, from a known gel
bead volume, the gel bead hold-up may be calculated directly. Therefore, the gel
bead hold-up is assumed to be known in this paragraph. Figure 7a shows that the
total drag coefficient for a single droplet in the 3-phase mixture may be described by
a general equation:
( ) ( )
2
2
2
c
cd d d,
c
d ,
cd
1
c
cd 1 d d,
' v C
v
v
c ' v c C


=








= = 12

0.1
1
10
100
0 0.2 0.4 0.6 0.8 1
V
cd
'
C
d,d

2
3
4
5
6
0.25 0.35 0.45 0.55
V
cs
'
C
d,s
Chapter 4
92
For ε
s
→ 0 and ε
d
→ 0, the slip velocity is equal to the terminal rise velocity of a single
droplet (v
∞,d
). Consequently, parameter c
1
is equal to the drag coefficient for a single
droplet rising in stagnant water (C
d,∞d
).
The drag coefficient ratio C
d,d
/C
d,∞d
(cf. equation 12) may also be derived from
force balances. The force balance for a droplet in the 3-phase mixture (equation 7a) is
divided by the force balance for a single droplet rising in stagnant water (an
equivalent of equation 7a), which results in the drag coefficient ratio. Substituting
this relation for the drag coefficient ratio in equation 12, and rewriting the resulting
equation (with equation 8 for the mixture density), yields an implicit correlation for
the droplet hold-up:

2 c
d ,
cd
d
d c
c s
s
2
v
v
1
+









= ε − +
ρ − ρ
ρ − ρ
ε 13

In the next paragraph, it will be shown how parameter c
2
can be derived from a 2-
phase droplet hold-up model.

Figure 7a shows that for ε
s
= 0, i.e. a 2-phase droplet column, the total droplet drag
coefficient in the droplet/water 2-phase mixture, can also be described by equation
12. An adequate model for the slip velocity between droplet and water in a 2-phase
system has been presented by Van Zessen et al. (2003) for hold-ups below 0.13:

( )
4
1 v v
d d , cd
ε − =

14

Substituting equation 14 in 13 and solving for parameter c
2

s
=0), results in c
2
= -1.75.
Although the droplet hold-up and drag coefficient (equation 12) for the 2-phase
droplet water mixture are well described, the 3-phase experimental data (drag
coefficient, and droplet hold-up) are systematically underestimated (data not
shown). Therefore, parameter c
2
was fitted on the 3-phase droplet hold-up data,
which resulted in c
2
=-2.06. The total droplet drag coefficient for both a 2-phase and 3-
phase system is now well described (Figure 7a). Apparently, the prediction of droplet
drag coefficients in a 2-phase water/ droplet mixture is not very sensitive to
parameter c
2
.

Hold-up in 3-phase fluidized bed
93
Up to a hold-up of about 0.06, the description of the data (equation 13, c
2
=-2.06) is
fair (Figure 8). The average deviation between data and model for hold-ups less than
0.06 is 16.5%; 81% of the model predictions show a deviation less than 25%. For hold-
ups larger than 0.06 the prediction is not good (Figure 8).

Figure 8. Parity plot for the droplet hold-up in a 3-phase fluidized bed. o the model fitted on the data
is equation 12; ● the model fitted on the data is equation 15.


In view of this partial inadequacy of the total droplet drag coefficient-based model
(good description of the drag coefficient, but underestimation of the higher droplet
hold-ups), an empirical model was tested as an alternative. A power-law
dependency of the dimensionless slip-velocity on both the solvent phase velocity and
its hold-up were assumed according to:

c5
d
c4
d 3 cd
d ,
cd
U c ' v
v
v
ε ⋅ ⋅ = =

15

A fit of this equation to the data in figure 5a yielded c
3
=5.4, c4=0.88 and c5=-0.76. A
parity plot between predicted and measured droplet hold-ups (Figure 8) shows that
droplet hold-up is well described over the entire range of measured hold-ups. The
average deviation between data and model is 13%; 89% of the model predictions
show a deviation less than 25%. The prediction of the drag coefficient with this
empirical model is fair; the average deviation is 17%, and 76% of the model
predictions show a deviation less than 25%.

0
0.02
0.04
0.06
0.08
0.1
0 0.02 0.04 0.06 0.08 0.1
measured ε
d
p
r
e
d
i
c
t
e
d


ε
d
Chapter 4
94
Model for droplet and gel bead hold-up
In the experimental set-up, the droplet hold-up was determined from the gel bead
hold-up (equations 1-4). So, contrary to the preceding paragraph, in which a known
gel bead hold-up was assumed, now both hold-ups were assumed unknown and will
be predicted simultaneously. This overall model for hold-up prediction in a 3-phase
fluidized bed is outlined in Figure 9, showing how physical parameters, nozzle
geometry and liquid fluxes determine the three hold-ups.

Figure 9. Hold-up calculations in a 3-phase fluidized bed.


Figure 7b suggests that the total drag coefficient of a single gel bead in mixture of gel
beads, droplets and 30 mM KCl can be described by:

( ) ( ) ( ) ( ) ( )
c8
d
c7
s c, 9 s , d
c8
d
c7
s c, 6
c8
d
c7
s ,
cs
6 s d,
1 ' v c C 1 ' v c 1
v
v
c C ε − = ε − = ε −








=


16

For ε
d
→ 0 and ε
s
→ 0, the slip velocity (v
cs
) is equal to the terminal settling velocity
of a gel bead in stagnant water (v
∞,s
). So, parameter c
6
should be equal to the drag
coefficient for a single gel bead settling in stagnant water (C
d ∞,s
). The ratio of drag
coefficients (C
d,s
/C
d∞,s
) may be obtained by dividing the force balance for a single gel
bead in the 3-phase mixture (equation 7b) with the force balance for a single gel bead
settling in stagnant water (an equivalent of equation 7b). Using this drag coefficient
ratio, together with equations 8 and 16, the gel bead hold-up may be predicted by:

Drag ratio gel beads
(eq. 16)
Drag ratio droplets
(eq. 12) / Empirical
equation (eq.15)
ε
s
ε
w
ε
d
ρ
mixture
(eq. 8)
v12 gel beads
(eq. 9b)
v12 droplets
(eq. 9a)
v∞ droplets (eq.11)
d droplets (eq.10)
∑ε
i
= 1
(eq. 4)
Ud
Uc
Uc
ρ
s
ρ
d
ρ
ρ
s
ρ
ρ
d
ρ
v∞,s
Drag ratio gel beads
(eq. 16)
Drag ratio droplets
(eq. 12) / Empirical
equation (eq.15)
ε
s
ε
w
ε
d
ρ
mixture
(eq. 8)
v12 gel beads
(eq. 9b)
v12 droplets
(eq. 9a)
v∞ droplets (eq.11)
d droplets (eq.10)
∑ε
i
= 1
(eq. 4)
Ud
Uc
Uc
ρ
s
ρ
d
ρ ρ
s
ρ
d
ρ
ρ
s
ρ
ρ
d
ρ
v∞,s
Hold-up in 3-phase fluidized bed
95
( )
c8
d
c7
s ,
cs
9
2
cs
s ,
s c
s c c s s d d
1
v
v
c
v
v
ε −








=








ρ − ρ
ρ − ρ ε + ρ ε + ρ ε



17

In the next paragraph, it will be shown how parameters c7 and c
9
can be derived
from the 2-phase hold-up model of a liquid fluidized bed.

Van Zessen et al., 2003 concluded that for a 2-phase fluidized bed of gel beads the slip
velocity can be described by:
1.35
c
c
c
cs
k
U
v ε =
ε
= 18
where parameter k is not equal to the terminal settling velocity of a gel bead
(5.12 cm s
-1
) but equals 3.77 cm s
-1
. Substituting equation 18 in equation 17, and
solving for parameter c
7

d
=0, c
9
=1 and c
8
=0) resulted in c
7
= f(ε
c
), and since ε
c
= f(U
c
),
also c
7
=f(U
c
). However, Figure 7b suggests that parameter c
7
might have a single
value. Therefore, it was fitted together with parameter c
9
to the solids hold-up data
for a liquid-fluidized bed using equation 17 (data in figure 5b, U
d
=0 cm s
-1
). This
resulted in c
7
=-1.26 and c
9
=1.24. Figure 10b, a parity plot between measured and
predicted gel bead hold-up, shows that the data are well described. Parameter c
9
had
to be fitted on the data, as for c
9
=1, there was a systematic underestimation of the
higher solids hold-up, and the relative residuals were not randomly distributed, as
they should be.

Figure 10a. Parity plot for droplet hold-up in
a 3-phase fluidized bed.
Figure 10b. Parity plot for gel bead hold-up o
3-phase fluidized bed, ● 2-phase fluidized bed
Models fitted on the data are equations 15 and 17.
0
0.02
0.04
0.06
0.08
0.1
0 0.02 0.04 0.06 0.08 0.1
measured ε
d
p
r
e
d
i
c
t
e
d


ε
d
0.15
0.25
0.35
0.45
0.55
0.15 0.25 0.35 0.45 0.55
measured ε
s
p
r
e
d
i
c
t
e
d


ε
s
Chapter 4
96
As equation 15 was found adequate for the entire range of droplet hold-ups, this
model was adopted as a part of the simultaneous prediction of all hold-ups.
Equations 15 and 17 were now used to describe the droplet hold-up and gel bead
hold-up simultaneously (cf. Figure 9). We used for c
7
and c
9
the parameter values of
the 2-phase fluidized bed. Parameters c
3
, c
4
, c
5
and

c
8
were obtained from fitting the
model equations on the hold-up data (Figure 5); resulting in c
3
= 6.15 s m
-1
, c
4
= 0.90,
c
5
= -0.76 and

c
8
=-6.51. Our data are well described by this model (Figure 10). The
average deviation between model and droplet hold-up data is 12% (89% of all data
show a deviation less than 25%), for the gel bead hold-up data the average deviation
is 7.2% (97% of all data show a deviation less than 25%).


Conclusion

The liquid analogue of a gas-liquid-solid fluidized bed, a liquid-liquid-solid (gel
bead) fluidized bed was studied. Visual observations of the 3-phase fluidized bed
showed how the behavior of the fluidized bed depended on the water and dodecane
fluxes. Characterized by the gel beads flow, five flow patterns were discerned: two
homogeneous regions, two heterogeneous regions and one unstable region in which
droplets coalesced and gel beads were washed out from the fluidized bed. Gel bead
hold-up and dodecane hold-up were measured. At a constant water velocity, the
dodecane hold-up increased with the dodecane velocity. At a constant dodecane
velocity, however, the droplet hold-up decreased with the water velocity, while at
the same time gel bead hold-up decreased. For all fluxes, the dodecane hold-up was
always higher than the dodecane hold-up in a 2-phase droplet column.
Literature models for related 3-phase systems, a gas-liquid-solid 3-phase
fluidized bed and a solid-solid-liquid fluidized bed, could not predict correctly the
entire range of hold-ups in the present system. A new model, based on steady state
force balances for both a single gel bead and a single droplet in a 3-phase mixture
(gel beads, droplets and 30 mM KCl) was developed for predicting gel bead hold-up
and dodecane droplet hold-up. The gel bead hold-up was well described for all data,
but droplet hold-ups were described adequately for hold-ups up to 0.06 only. An
empirical model for droplet hold-up was developed as an alternative. Its predictions
were good for the entire range of measured droplet hold-ups.

Hold-up in 3-phase fluidized bed
97
Nomenclature
A : cross-sectional column area m
-2

c
1
,c
2
: parameters in equation 12 -
c
3
: parameter in equation 15 s m
-1
c
4
,c
5
: parameter in equation 15 -
c
6-9
: parameters in equation 16 -
C
di
: total drag coefficient for particle i -
d : diameter droplet m
d
i
: diameter particle i m
d
noz
: nozzle diameter m
Eo : Eötvös number, σ ρ ∆
2
d g -
Eo
noz
: Eötvös nozzle number, σ ρ ∆
2
noz
d g -
g : gravity acceleration (9.8) m
2
s
-1
H
bed
: bed height m
∆P : pressure drop N m
-2

∆P
exp
: measured pressure difference N m
-2

U : superficial velocity m s
-1
V
s
: solids volume m
3

v

: terminal rise/settling velocity m s
-1
v
ci
: slip velocity between water and component i m s
-1
We
noz
: nozzle Weber number, σ ρ
∞ noz
2
d , c
d v -
∆z : height between two points m

Greek
ε : hold-up -
η : viscosity Nm s
-1
ρ : density kg m
-3

∆ρ : density difference continuous and disperse phase kg m
-3
σ : interfacial tension N m
-1

subscript
c : continuous
d : dispersed (organic solvent droplet)
mix : mixture
s : gel bead

98

99

5
Mixing of the continuous phase in a liquid-liquid-solid
fluidized-bed bioreactor



Abstract
A liquid-liquid-gel bead (solid) fluidized bed is a new type of bioreactor to overcome
biokinetic or solubility limitations present in different kinds of biotransformations. In
this 3-phase fluidized bed both liquid phases (n-dodecane dispersed as droplets, and
30 mM KCl as the continuous phase) are separately introduced to the bioreactor. In
this work the mixing of the continuous phase was studied by measuring the
residence-time distribution for different superficial velocities of both liquid phases. A
KCl-tracer was used. Due to mass transfer to the gel beads, the limitations using this
tracer were studied as well. A pulse volume of 3 ml (4 M KCl) gave reliable results.
An axial dispersion coefficient for the continuous phase was determined from inlet
and outlet residence time distribution curves. The axial dispersion coefficient in the
continuous phase (D
ax
) increased with an increasing superficial n-dodecane velocity.
For a superficial velocity of the continuous phase (Uc) smaller than ~ 0.8 cm s-1 the
D
ax
increased with an increasing Uc, whereas for Uc larger than ~ 0.8 cm s-1 the D
ax

decreased with an increasing Uc. These results were easily explained by the isotropic
turbulence theory of Baird and Rice (1975). The mixing of the continuous phase of
this 3-phase fluidized bed was comparable with the mixing of either the continuous
phase of a bubble column or the continuous phase in the main tube of a liquid-
impelled loop reactor.




Chapter 5
100
Introduction

The potentials of using an organic solvent as an extraction aid to overcome biokinetic
or solubility limitations in different kinds of biotransformations have been shown by
many authors (e.g. Adlercreutz, 2000; Vermuë, 1995). For a biotransformation in
which the production rate is strongly reduced by the product itself, a new type of
bioreactor has been studied: a liquid-liquid-solid 3-phase fluidized bed bioreactor. In
this bioreactor there are three different phases: 1) the solid phase that consists of gel
beads with biocatalyst, 2) one liquid phase, an organic solvent, which is dispersed as
droplets, 3) another liquid phase which is continuous, and this phase is used for the
supply of nutrients to the immobilized biocatalyst. This continuous phase fluidizes
the gel beads, whereas the organic-solvent droplets, rising through the fluidized bed,
extract the formed product.
The hold-ups of the different phases in a liquid-liquid-solid fluidized bed
bioreactor have been studied before (Van Zessen et al., 2003). In this paper the
residence time distribution of the continuous phase is studied. A residence time
distribution can be related to mixing characteristics of the bioreactor. As such, it
serves as an important design parameter.
To measure a residence-time distribution, a tracer is injected and its
concentration is followed in time at a fixed point in the bioreactor downstream from
the point of injection. As mixing of the continuous phase is studied, this tracer must
not be transferred to either of the two dispersed phases, i.e. the gel beads and the
organic droplets. A suitable tracer can be a blue dextran solution (Gommers et al.
1986). In our experimental set-up, however, the concentration of the blue dextran
comes close to the detection limit of the analysis equipment. Consequently, as the
signal is full of noise, the interpretation of this signal can not be performed reliably.
So, we chose a different tracer, viz. a KCl solution. This salt, however, can be
transferred to the gel beads (Thompson and Worden, 1992) and this measuring
method had to be validated.
In the first part of the results section the validation of this measuring method is
discussed. In the second part of the results section this validated measuring method
was applied for measuring the residence-time distribution of the continuous phase of
the 3-phase fluidized bed. Different flow patterns regimes have been distinguished in
the 3-phase fluidized bed (Van Zessen et al. 2003). Measurements have been
performed in the structured solids flow and in the turbulent solids flow regimes.
A dispersion coefficient of the continuous phase is calculated from the
Mixing of the 3-phase fluidized bed
101
residence-time distribution curve, using a suitable flow model. Based on the
magnitude of this dispersion coefficient it can be concluded that mixing of the
continuous phase in a liquid-liquid-solid 3-phase fluidized bed bioreactor is more
comparable to an ideally mixed flow pattern than to a plug flow pattern. It will be
shown that the dispersion coefficient of the continuous phase is well described by the
model of Baird and Rice (1975), using the energy-dissipation rate per unit mass for a
3-phase fluidized bed.


Theoretical aspects

Residence-time distribution represents the time-distribution of volume elements in a
specific part of a device, in this case the continuous aqueous phase in the bioreactor.
The shape of this time-distribution can be translated to flow characteristics of the
bioreactor. To measure this distribution, a tracer is injected and its concentration is
followed in time. The translation from tracer profile, C(t), to a residence time
distribution curve, E(t), follows from (Levenspiel,1972):

C(t)
m
E(t)
Φ
= 1

with Φ is the flux of the continuous phase and m is the total amount of tracer
injected. This amount is equal to:

dt C(t) m
0


Φ = 2

The shape of the distribution curve at the outlet is influenced by the tracer profile of
the incoming fluid and the system itself, see Figure 1. Unless the shape of the inlet-
tracer profile is known, this inlet-tracer profile should be measured. So, time-profiles
of the tracer concentration at the outlet as well as time-profiles of the tracer
concentration at the inlet have been measured.





Chapter 5
102
Figure 1. Transformation of the residence-time distribution curve at the inlet of a (reactor)system to
the residence-time distribution curve at the outlet of a (reactor)system by the (reactor)system itself
between the inlet and the outlet.


Fahim and Wakao (1982) presented a review on the parameter estimation from
tracer-response measurements. Different methods were analyzed, e.g. methods based
on 1) the moments of the distribution curve, 2) the application of transfer functions
and 3) time-domain analysis. They concluded that the most reliable parameter values
are obtained from the time-domain analysis. Thus in this paper, time-domain
analysis is used for parameter estimation. To use time-domain analysis for parameter
estimation, a dynamic model describing flow behavior of the continuous phase is
needed.

Models for describing the flow behavior of the continuous phase can range from
simple to highly complicated. Two of the most common and simple models are the
tanks-in-series model and the plug-flow-with-dispersion model (axial dispersion
model). The tanks-in-series model reads:
Inlet
System
Outlet
0
0.03
0.06
0.09
0 15 30 45 60
time (s)
E(t)
0
0.03
0.06
0.09
0 15 30 45 60
time (s)
E(t)
0
0.03
0.06
0.09
0 15 30 45 60
time (s)
E(t)
0
0.03
0.06
0.09
0 15 30 45 60
time (s)
E(t)
0
0.03
0.06
0.09
0 15 30 45 60
time (s)
E(t)
0
0.03
0.06
0.09
0 15 30 45 60
time (s)
E(t)
1
i
N
11
ii
NN
Mixing of the 3-phase fluidized bed
103
( )
( )
( )
out 1 - N
out
i 1 - i
i
1 in
1
C C
1
dt
C d
1 - N 2 i C C
1
dt
C d
C ) t ( C
1
dt
C d

τ
=
= ∧ −
τ
=

τ
=
K 3

in which C
in
is the inlet tracer profile, C
out
is the outlet tracer profile, C
i
is the tracer
profile in tank i, t represents time and τ is the mean residence time given by:

V
N

c
Φ ε
= τ 4

in which ε
c
is the continuous phase hold-up, and V is the total volume of the
bioreactor section considered in this analysis.

The axial dispersion model reads:

2
2
ax
c
c
x
C
D
x
C U
t
C


+


ε
− =


5

in which U
c
is the superficial velocity of the continuous phase and the D
ax
is the axial
dispersion coefficient. The initial conditions are:

) t ( C 0) x 0, C(t ; 0 x) 0, C(t
in
= = > = ∀ =

When U
c
is given and ε
c
is known, both models contain only one parameter; the
number of tanks in the tanks-in-series model and the axial-dispersion coefficient in
the latter model. For a flow pattern comparable to plug flow, i.e. a large number of
tanks-in-series, or a small axial-dispersion coefficient, these parameters are related to
each other, as pointed out by Levenspiel (1972):

c
c
ax
N 2
LU
D
ε
= 6

with L is the height of the bioreactor section.
Chapter 5
104
For a flow pattern comparable to an ideally mixed tank, the basic idea of the axial
dispersion model is violated. However, Gommers et al. (1986) showed the validity of
applying the axial dispersion model to describe axial mixing for gas-liquid-solid
fluidized beds. The axial dispersion model was also successfully applied to describe
axial mixing in a liquid-liquid-solid fluidized bed (Kim et al., 1989). Applying the
axial dispersion model instead of the tanks-in-series models is tempting, as the axial
dispersion coefficient can take continuous values, whereas the number of tanks can
only be a discrete value. Especially for a small number of tanks, the limitation of
using only discrete values might result in a less accurate description of the flow
pattern. In this paper both models were used to describe the experimental data.


Materials and Methods

Two sets of different experiments were performed. The first set of experiments aimed
at developing and validating the measuring method. In the second set of experiments
this validated method was used to determine the mixing characteristics of the
continuous phase of the 3-phase fluidized bed.

Materials
The gel beads used in all experiments were κ-carrageenan gel beads filled with 10%
Celite
®
. The preparation of these gel beads can be found elsewhere (Van Zessen et al.
2003). The density of the gel beads was 1065.1 kg m
-3
and the diameter was 2.76 mm.
The density and diameter were determined as described by Van Zessen et al. (2001).
A 30mM KCl solution was used as the continuous phase (density is 998.0 kg m
-3
). The
organic solvent used in the 3-phase fluidized bed was n-dodecane (Aldrich, 99%+
pure), density is 742.7 kg m
-3
. As tracer solution a mixture of a blue-dextran solution
and a KCl-solution was used.

Development and validation of the measuring method
The experimental set-up for developing and validating the measuring method is
shown in Figure 2. A 30 mM KCl solution was pumped through the column with a
flow rate high enough to establish a fluidized bed. A tracer pulse was introduced to
the column by switching a three-way valve. Most of the liquid was purged, but a
small part was pumped through a home-made flow-through conductivity cell
connected to a conductivity meter (WTW). Next, the liquid flew through a flow-
Mixing of the 3-phase fluidized bed
105
through cuvette, in which the absorption at 280 nm was measured with a
spectrophotometer (Ultrospec 2000, Pharmacia Biotech). The spectrophotometer and
conductivity meter were connected to an A/D converter (Hewlett Packet) and a
personal computer for data acquisition. Two different columns were used. One
column was 20 cm high; the other was 130 cm high. Both columns had an internal
diameter of 2 cm. At different flow rates tracer concentrations ranging from 40 mM
up to 1 M and pulse volumes from 1 ml up to 7 ml were used to study whether mass
transfer took place. To verify whether mass transfer took place in the 3-phase
fluidized bed, the set-up for measuring absorption and conductivity, as described
above, in combination with the experimental set-up of the 3-phase fluidized bed, see
below, was also used.


Figure 2. Experimental set-up for developing and validating the measuring method. 1 gear pump, 2
glass pearls to provide a good flow distribution, 3 sieve-plate, 4 fluidized bed of gel beads, 5 three-way
valve, 6 storage vessel for 30 mM KCl, 7 flow-through conductivity cell, 8 WTW Conductivity Meter, 9
flow-through cuvette, 10 Ultrospec 2000 spectrophotometer, 11 A/D converter and personal computer,
12 tracer solution vessel.


Residence time measurements
Measurements were performed in an experimental set-up as shown in Figure 3. A
detailed description of this set-up is given elsewhere (Van Zessen et al., 2001). Water
and n-dodecane flew co-currently upwards, and as both liquids were highly
immiscible, they were easily separated with a one-stage gravity settler. The column
was 2 m high with an internal diameter of 6 cm.
6
7
1
3
2
4
9
8
10
11
5
5
12
Chapter 5
106
Figure 4 shows the set-up used for introducing a tracer pulse to the column.
While a 30 mM KCl-solution was circulated over the column, a tube between two
three-way valves was filled with the tracer solution. By adjusting the length of this
tube, the tracer volume could be changed. In the residence-time experiments the tube
length was 6 cm that resulted in a pulse volume of 3 ml (internal diameter of the tube
is 0.8 cm). By switching both three-way valves simultaneously, the pulse was
introduced to the column. The conductivity of the solution was measured at two
different column heights: 25cm and 75cm. Residence-time distribution measurements
were performed for the two (liquid-solid) and three phase (liquid-liquid-solid)
fluidized bed.

Figure 3. Experimental set-up of the 3-phase fluidized bed for measuring the residence time
distribution. 1 3-phase fluidized bed, the gel beads, dodecane droplets, 2 separator for 30 mM KCl
and dodecane, 3 conductivity measurement (a conductivity cells, b data acquisition device), 4
rotameter, 5 storage vessel for dodecane, 6 storage vessel for 30 mM KCl-solution, 7 absorbance
measurement (a flow-through conductivity cell, a1 WTW conductivity meter, b flow-though cuvette, b1
Ultrospec 2000 spectrophotometer, c A/D converter and personal computer), 8 device for introducing
the tracer solution.
3
4
5
1
2
4
6
a
a
b
a
b
a1
b1
c 7
8
Mixing of the 3-phase fluidized bed
107
Conductivity cell
The conductivity cell was home made. It consisted of two parallel placed platinum
wires, 1 mm thick and 1 cm long. The distance between both wires was equal to the
column diameter, i.e. 6 cm, see also Figure 5. The cell was connected to a
conductivity meter (WTW). Using an A/D converter (Hewlett Packard) and a
personal computer both conductivity signals were stored in time.


Figure 4. Set-up for the introduction of the tracer
pulse. A) flow from the 30mM KCl-storage vessel, B)
flow to the fluidized bed. a) By-passing the tracer
pulse section, b) flow through the tracer pulse
section. 1 three-way valve, 2 tube for tracer pulse (6
ml), 3 storage vessel tracer solution, 4 valve
Figure 5. Home-made conductivity
cell. 1 platinum wire, 2 (glass)-
holder for platinum wire, 3 copper
wire for connecting to a conductivity
meter, 4 column wall.


Flow rates were measured with calibrated rotameters (Sho-Rate). The different flow
rates applied in the experiments performed with either the 3-phase fluidized bed or
the liquid fluidized bed are summarized below. U
c
represents the superficial velocity
of the continuous phase, and U
d
represents the superficial velocity of the dodecane
phase.

U
c
(cm s
-1
) U
d
(cm s
-1
)
3-phase fluidized bed
Structured solids flow regime 0.55 – 0.8 0.1 – 0.4
Turbulent solids flow regime 1 – 1.8 0.05 – 0.7

Liquid fluidized bed 0.6 – 1.8 -


1
2
3
4
1
2
3
4
1 1
a
b
2
3
4
4
b
A B
Chapter 5
108
Fitting procedure
To obtain the parameters used by both presented flow models, the tracer
concentration-profile has to be known. Although conductivity and UV-absorption
are measured, these signals are linearly proportional with the KCl concentration and
blue dextran concentration, respectively:

S(t) k C(t)
i
= 7

in which S is the measured signal and k
i
is a proportionality constant for component
i.

Implementing equation 7 into equation 1 and 2 gives the residence-time distribution
based on a measured signal:


=
dt S(t)
S(t)
E(t) 8

The measured outlet signal is transformed to a residence-time distribution using
equation 8. Both models proposed for describing the mixing of the continuous phase,
equation 3 and 5, are transformed in order to calculate a residence-time distribution.
This was done by multiplying the respective equations with the water flux and by
dividing the equations by the total mass of the injected tracer. For the tanks-in-series
model the resulting set of equations is:

( )
( )
out 1 - N
out
i 1 - i
i
1
in
in 1
E E
1
dt
E d
1 - N 2 i E E
1
dt
E d
E
dt (t) S
(t) S 1
dt
E d

τ
=
= ∧ −
τ
=









τ
=

K 9

For the axial dispersion model the resulting equation is:

2
2
ax
c
c
x
E
D
x
E U
t
E


+


ε
− =


10
Mixing of the 3-phase fluidized bed
109
The initial conditions are transformed as well


= = > = ∀ =
0
in
in
(t)dt S
) t ( S
0) x 0, E(t ; 0 x) 0, E(t
The hold-up of the continuous phase followed from a previous presented model
predicting the different hold-ups in this type of fluidized bed (Van Zessen et al.,
2003), equations 14 and 15a-c. Equation 10 was then solved by discretization of the
axial coordinate. To determine the number of ideally stirred tanks-in-series in
equation 3 or the axial dispersion coefficient in equation 5, the calculated outlet
profile (E
out,m
) was compared with the measured outlet profile (E
out,exp
). Both
parameters were determined by minimizing the residual sum of squares as defined
by equation 11:


− =
t
2
exp out, m out, res
) (t) E (t) E ( SS 11


Chapter 5
110
Results and Discussion

First the results of the validation experiments are shown. These results are applied
for performing reliable residence time distribution measurements in the 3-phase
fluidized bed. Next, the results of these measurements are shown.

Development and validation of the measuring method
For a column without gel beads, the residence time distribution, determined with a
KCl and blue dextran solution, is shown in Figure 6. As the molecular weight of blue
dextran is ~ 2,000,000 g mol
-1
, diffusion of blue dextran into the gel beads can be
neglected. In different experiments diffusion was indeed not observed. Figure 6
demonstrates that both tracers yield identical response curves. So, in order to
evaluate whether mass transfer takes place, both response curves can be compared to
each other. When mass transfer of KCl takes place, the salt-E(t) curve will differ from
the blue dextran E(t) curve remarkably at two points: 1) the top of the salt-curve is
lower and 2) the tail is elongated over a longer period of time. Consequently, the
KCl-response curve crosses the blue dextran curve, see Figure 7.
Figure 6. Residence time distribution curve
of a 20 cm high column filled with water.
···· blue dextran tracer; — KCl tracer.
Figure 7. Residence time distribution curve
for a 130 cm high column filled with gel
beads. Pulse volume 2.9 ml (2 g/l blue
dextran, 1 M KCl), . ▬ blue-dextran tracer;
── KCl tracer.

From the experiments performed in the small column (20 cm high) it was observed
that mass transfer was only important for a large pulse volume (6 ml) (data not
shown). A better understanding whether mass transfer took place, was obtained
from experiments in the larger column (130 cm). For a low flow rate (1 cm/s) mass
transfer was absent for any salt concentration applied and each pulse volume used
(data not shown). At a high flow rate (1.9 cm/s) mass transfer was observed for the
0
0.0025
0.005
0.0075
0.01
0 100 200 300 400 500 600 700
Time (s)
E(t)
0
0.01
0.02
0.03
0.04
0 40 80 120 160
Time (s)
E(t)
Mixing of the 3-phase fluidized bed
111
larger pulse volumes used, regardless of the salt concentrations applied (data not
shown). By reducing the pulse volume, mass transfer almost disappeared (data not
shown). So, a KCl-solution as tracer for E(t)-measurements is well possible, provided
the pulse volume is not too large.

For determining the E(t) in a 3-phase fluidized bed, a pulse volume of 3 ml was used
and a tracer concentration of 4 M KCl. This high concentration was used as
experiments showed that a high salt concentration did not result in mass transfer,
provided the pulse volume is small. Whether mass transfer took place, was verified
by using the blue dextran/ KCl-salt tracer pulse for a dodecane flux of 0.1 cm s
-1
and
a water flux of 1.2 cm s
-1
. Although the UV-absorption of the blue dextran was close
to the detection limit, and consequently the signal is full of noise, it could be
concluded that mass transfer was not significant (results not shown). Applying the
same tracer pulse (3 ml, 4M KCl) in a 2-phase liquid-fluidized bed, it was concluded
that even for high water fluxes mass transfer is not detectable (see Figure 8).
Figure 8. Residence time distribution curve for a liquid-solid (gel beads) fluidized bed. Puls volume 3
ml, ▬ 4M KCl tracer, ── 2 g l
-1
blue dextran tracer.



0
0.005
0.01
0.015
0.02
0 50 100 150 200 250 300
Time (s)
E(t)
Chapter 5
112
Residence time distribution experiments
First, a typical residence-time distribution curve for a 2-phase and a 3-phase fluidized
bed will be shown. Thereafter, the dispersion coefficient derived from the residence
time distribution curve, will be given as a function of the water and dodecane flux.

Figure 9 shows a typical measured E(t)-curve at the inlet and the outlet of a liquid-
solid fluidized bed (figure 9a), as well as of a liquid-liquid-solid fluidized bed (figure
9b). It appears that the inlet E(t)-curve for a liquid-liquid-solid fluidized bed is much
broader and the top is much lower. Regarding the outlet E(t)-curves for both
fluidized beds it can be said that the E(t)-curve for the 3-phase fluidized bed is much
broader. At equal superficial water velocity the differences between the E(t)-curves of
both fluidized beds shown in figure 9, were always observed.
A B
Figure 9. Residence time distribution for a liquid-solid fluidized bed (A) and a liquid-liquid-solid
fluidized bed (B). 30 mM KCl-solution superficial velocity is 1.2 cm s
-1
, dodecane flux is 0.4 cm s
-1
. 
inlet, ▬ outlet.

Two different, but comparable models for describing the mixing of the continuous
phase were used: the tanks-in-series model (eq. 9) and the axial dispersion model (eq.
10). Using the experimental data for a superficial continuous phase velocity of
1.4 cm s
-1
and a superficial dodecane velocity ranging from 0.05 to 0.7 cm s
-1
, the
number of tanks-in-series and the axial dispersion coefficient were fitted. The
number of tanks-in-series can be transformed to a dispersion coefficient using
equation 6. Figure 10 shows the fitted dispersion coefficient, using the axial
dispersion model, and the dispersion coefficient based on the tanks-in-series model.
From this figure it can be concluded that both models give the same relationship
between the superficial dodecane velocity and the dispersion coefficient.

From duplicate experiments (the same superficial dodecane velocity) it can be
0
0.25
0.5
0.75
1
0 30 60 90 120
time (s)
E(t)
0
0.25
0.5
0.75
1
0 30 60 90 120
time (s)
E(t)
Mixing of the 3-phase fluidized bed
113
concluded that the experimental method gives reproducible results (see Figure 10).
Using the tanks-in-series model, the number of tanks-in-series can only take discrete
values. As a result, for a liquid-liquid-solid fluidized bed, in which the number of
tanks-in-series is between 2 and 9, the accuracy of the transformed dispersion
coefficient is small: for N=2, D
ax
is between 21 and 63 cm
2
s
-1
; for N=9, D
ax
is between
5.5 and 6.9 cm
2
s
-1
. As the dispersion coefficient in the axial dispersion model can take
any value, the axial dispersion model was used for fitting the dispersion coefficient
on the other experimental data.

Figure 10. Dispersion coefficient as a function of the dodecane flux for a liquid-liquid-solid fluidized
bed, 30 mM KCl-solution superficial velocity is 1.4 cm s
-1
.
● dispersion coefficient fitted with the axial dispersion model, o dispersion coefficient based on the
number of tanks-in-series fitted with the tanks-in-series model.

U
d
(cm s
-1
) 0.05 0.1 0.1 0.26 0.26 0.4 0.4 0.55 0.58 0.71 0.71
N 9 4 3 3 2 3 3 4 2 2 3

In this type of liquid-liquid-solid fluidized bed different distinct flow regimes can
exist (Van Zessen et al. 2003) depending on the superficial velocities of both phases.
Measurements have been performed in the structured solids flow and turbulent
solids flow regime. The transition from structured solids flow regime to turbulent
solids flow regime occurred at a superficial velocity of the continuous phase of about
1 cm s
-1
. For these two different flow regimes the dispersion coefficients of the
continuous phase are discussed separately.
A clear characteristic of the structured solids flow regime is bed contraction: the
bed height of the 3-phase fluidized bed is smaller than the bed height of the 2-phase
fluidized bed at the same superficial water velocity. Another distinctive characteristic
is the movement of the gel beads: the gel beads flow a short distance in upward
0
5
10
15
20
25
30
35
0 0.2 0.4 0.6 0.8
U
d
(cm s
-1
)
D
a
x




(
c
m
2

s
-
1
)
Chapter 5
114
direction, and then they stop moving. Next they flow a short distance in downward
direction – falling back - and they stop moving again. This cycle is continuously
repeated in time. In flowing upwards the gel beads approach each other, so the solids
are a little concentrated. In flowing downwards the gel beads are a little further
apart, and the solids are a little diluted (Van Zessen et al. 2003).
A clear characteristic of the turbulent solids flow regime is bed expansion: the
bed height of the 3-phase fluidized bed is always larger compared with the 2-phase
fluidized-bed at the same water flux. The movement of the gel beads depends on the
superficial water velocity; at a high velocity the gel beads move individually and
randomly through the fluidized bed, at a lower velocity the gel beads move in small
groups, and these small groups move randomly through the fluidized bed (Van
Zessen et al. 2003).

Structured solids flow regime
In figure 11 the dispersion coefficient of the water phase fitted on the measured E(t)-
curve at the outlet, using the measured E(t)-curve at the inlet and equation 10 and 11
is shown as a function of the superficial water velocity (U
c
) for two superficial
dodecane velocities (U
d
) (0.1 cm s
-1
and 0.2 cm s
-1
). This figure clearly shows that up
to an U
c
of 0.7 cm s
-1
the dispersion coefficient increases with an increase in both
superficial velocities. At a higher U
c
the dispersion coefficient decreases.
The structured solids flow regime is characterized by the flow of the gel beads
(Van Zessen et al., 2003). Large groups of gel beads move up and down the fluidized
bed, but in such a large group the individual gel beads stick together. At higher U
d
,
the up-and-down movement of the large groups of gel beads intensifies, and hence
the mixing of the continuous phase is better; the dispersion coefficient increases. At
higher U
c
, the movement of the individual gel beads inside the large groups of gel
beads intensifies, and hence mixing of the continuous phase increases; the dispersion
coefficient increases as well.
The decrease of the dispersion coefficient at an U
c
higher than 0.7 cm s
-1
follows
the trends observed for the turbulent solids flow regime. At an U
c
of about 1 cm s
-1
,
the flow pattern of the 3-phase fluidized bed is in turbulent solids flow. This
transition from structured solids flow to turbulent solids flow regime is not sharp.
Increasing U
c
, the large groups of gel beads fall apart gradually into smaller groups
and the mixing of the continuous phase becomes more comparable to mixing
characteristics of the turbulent solids flow regime.

Mixing of the 3-phase fluidized bed
115
Figure 11. Axial dispersion coefficient as a function of the 30 mM KCl superficial velocity. Structured
solids flow regime. ● U
d
= 0.1 cm s
-1
; ∆ U
d
= 0.2 cm s
-1
.


Turbulent solids flow regime
Figure 12 shows the dispersion coefficient of the water phase fitted on the measured
E(t)-curve at the outlet, using the measured E(t)-curve at the inlet and equation 10
and 11 as a function of U
c
at different values of U
d
. From figure 12 it follows that the
dispersion coefficient of the continuous phase in a liquid-liquid-solid fluidized bed,
regardless of the superficial water velocity applied, is always much larger than the
dispersion coefficient of the continuous phase in a liquid-solid fluidized bed.
For a liquid-solid fluidized bed the dispersion coefficient increases with U
c
(the
dispersion coefficient at U
c
is 1.8 cm s
-1
is 84% higher than the dispersion coefficient
at U
c
is 1 cm s
-1
). However, for a liquid-liquid-solid fluidized the dispersion
coefficient is virtually constant or decreases with an increasing U
c
. This holds for all
superficial dodecane velocities applied.

0
10
20
30
40
0.4 0.6 0.8 1
U
c
cm s
-1
D
a
x

c
m
2

s
-
1
Chapter 5
116
Figure 12. Axial dispersion coefficient as a function of the 30 mM KCl superficial velocity at different
superficial dodecane velocity. ● Ud= 0 cm s
-1
(liquid-solid fluidized bed), o Ud=0.05 cm s
-1
, ■ Ud=0.1
cm s
-1
, ∆ Ud = 0.3 cm s
-1
, ▲Ud= 0.6 cm s
-1
.


In figure 13 the dispersion coefficient of the continuous phase for the 3-phase
fluidized bed is shown as a function of U
d
at different values of U
c
. This figure 13
shows that at up to an U
d
of 0.4 cm s
-1
the dispersion coefficient increases with an
increasing U
d
. At U
d
higher than 0.4 cm s
-1
the dispersion coefficient is hardly
increasing with an increasing U
d
. These observations hold for all the different
superficial water velocities applied.
Figure 13. Axial dispersion coefficient as a function of the superficial dodecane velocity at different 30
mM KCl-solution superficial velocities. ■ Uc = 1.0 cm s
-1
, o Uc = 1.4 cm s
-1
, ● Uc = 1.8 cm s
-1
.


The experimental results (Figure 12 and 13) can be explained as follows. When
dodecane droplets are introduced into a fluidized bed of gel beads, this will result in

0
10
20
30
40
0.9 1.1 1.3 1.5 1.7
U
c
cm s
-1
D
a
x





c
m
2

s
-
1

0
10
20
30
40
0 0.2 0.4 0.6 0.8
U
d
cm s
-1
D
a
x




c
m
2

s
-
1
Mixing of the 3-phase fluidized bed
117
a much more intense movement of the solids (visually, this change in behavior of the
fluidized bed could be easily observed). The density of the gel beads is relatively low,
and consequently they can be moved easily by the rising solvent droplets. Such an
increase of the movement of the gel beads in comparison with a 2-phase (liquid- gel
beads) fluidized bed results in a better mixing of the continuous phase of a 3-phase
fluidized bed. Consequently, the dispersion coefficient of the continuous phase
increases. A higher superficial dodecane velocity results in more dodecane droplets
rising through the fluidized bed. The movement of the gel beads increases, and the
mixing of the continuous phase increases as well. So, the dispersion coefficient
increases with an increasing dodecane superficial velocity. An increase in U
d
always
results in an increase in the movement of the gel beads. Obviously, at a certain U
d
the
continuous phase is very well mixed, and a further increase in U
d
will hardly result
in better mixing of the continuous phase. As a result, the dispersion coefficient is
constant, or it will increase only slightly.

A decrease in dispersion coefficient with an increasing superficial continuous phase
velocity, at a constant U
d
, can be explained by considering the energy dissipation rate
per unit mass of the continuous phase, as will be shown in the next paragraph. This
energy-dissipation rate decreases with an increasing superficial water velocity, and
consequently the dispersion coefficient is lower.

The effect of the superficial dodecane velocity on the axial dispersion coefficient,
found in this study is in complete agreement with the results of Kim et al. (1989),
which is to our knowledge the only study on the dispersion coefficient for liquid-
liquid-solid 3-phase fluidized bed. As 3-phase system these authors used water,
kerosene and glass beads. For the relationship between U
c
and the dispersion
coefficient Kim et al. (1989) reported that the dispersion coefficient increases with an
increasing U
c
, which is in complete contradiction with our results. However, the
density of the glass beads (2500 kg m
-3
) differs very much from the density of the gel
beads. This large difference in density can explain the different dependence of the
dispersion coefficient on U
c
. Kim et al. (1989) could well explain their data with the
energy dissipation rate per unit mass (P
m
). A higher U
c
resulted for their system in a
higher P
m
. Our results could be explained in the same way, as will be shown below.
Apparently, a general relationship between U
c
and the dispersion coefficient does not
exist; it depends on the physical constants of the 3-phase system.

Chapter 5
118
Modeling the dispersion coefficient of the continuous phase

The isotropic turbulence theory (Baird and Rice, 1975) has been successfully applied
to predicting the axial-dispersion coefficient for two- and 3-phase fluidized beds
(Kim and Kim, 1983) as well as for liquid-impelled loop reactors (Van Sonsbeek et al.,
1992). The dispersion coefficient is given by:

3
1
3
4
m 1 ax
P L c D ⋅ ⋅ = 12

in which c
1
is a dimensionless constant, L is the length scale of the largest eddy (L
equals the column diameter), and P
m
is the energy-dissipation rate per unit mass. In
analogy to Van Sonsbeek et al. (1992) the P
m
for a liquid-liquid-solid 3-phase
fluidized bed is given by, see also appendix A:

d
c c
d mix
c
c c
c mix
m
U g U g P ⋅
ρ ε
ρ − ρ
+ ⋅
ρ ε
ρ − ρ
= 13

in which g is the gravitational constant, ρ
c
and ρ
d
are the densities of the continuous
and dodecane phase, and ε
c
is the hold-up of the continuous phase. The mixture’s
density, ρ
mix
, is given by:

s s d d c c mix
ε ρ + ε ρ + ε ρ = ρ 14

in which ρ
s
is the density of the gel beads, ε
s
and ε
d
are the hold-ups of the solid
phase and dodecane phase. The hold-ups of the three different phases in a liquid-
liquid-solid 3-phase fluidized bed are given by Van Zessen et al. (2003):

( )
51 . 6
d
-1.26
s ,
c c
2
c c
s ,
s c
s mix
1
v
U
U
v
24 . 1




ε −








ε
=








ε ρ − ρ
ρ − ρ
15a

d ,
0.76 -
d
0.9
d
c
c
d
d
v U 15 . 6
U U

⋅ ε ⋅ ⋅ =
ε

ε
15b

1
s d c
= ε + ε + ε 15c
Mixing of the 3-phase fluidized bed
119
in which v
∞,s
is the terminal settling velocity of a single gel bead (5.12 cm/s), and v
∞,d

is the terminal rise velocity of a single dodecane droplet. The rise velocity of a single
droplet depends on the droplet diameter, and the droplet diameter depends on the
superficial dodecane velocity and geometry of the sparger. For calculating the single
rise velocity we used the Vignes' equation (Godfroy and Slater, 1994), and for the
droplet diameter the equation of Kumar and Hartland (1996), see also Van Zessen et
al. (2003).
Figure 14. Axial dispersion coefficient as a function of the energy-dissipation rate per unit mass of the
continuous phase. ● data,  isotropic turbulence model with c
1
=0.33, ---- 25% higher and lower
predictions with the isotropic turbulence model.


Figure 14 shows the axial-dispersion coefficient as a function of the energy-
dissipation rate per unit mass, calculated with equations 13, 14 and 15, for the
different applied superficial velocities. An increase in P
m
obviously results in a
higher dispersion coefficient. It can be calculated that an increase in U
c
, at a constant
U
d
, results in a decrease of P
m
; the dispersion coefficient decreases with an increasing
U
c
. Figure 14 also shows equation 12 with the constant c
1
fitted on the experimental
value of the axial dispersion coefficient: c
1
is equal to 0.330 ± 0.017 (R=0.78). The
measured axial dispersion coefficients are well described by the model of Baird and
Rice (1975), equations 12-15: the average deviation is 27.9%. A deviation less than
25% holds for 67.9 % of the data, see also Figure 15. It is shown in Figure 15 that most
of the measured axial dispersion coefficients between 12.5 cm
2
s
-1
and 30 cm
2
s
-1
have
a deviation less than 25%. A measured axial dispersion coefficient less than 12.5 cm
2
s
-1
is found for U
d
=0.05 cm s
-1
and U
c
between 1 and 1.6 cm s
-1
, and for U
d
=0.1 cm s
-1

and U
c
between 1.6 and 1.8 cm s
-1
. At these flow conditions it could be visually
verified that the fluidized bed of gel beads is only slightly disturbed by the rising

0
10
20
30
40
0 0.01 0.02 0.03 0.04
P
m
J (kg s)
-1
D
a
x




c
m
2
s
-
1
Chapter 5
120
dodecane droplets. Apparently, at these conditions the relationship between energy
dissipation rate and axial dispersion coefficient does not hold. For a measured axial
dispersion coefficient larger than 30 cm
2
s
-1
, the deviation is always larger than 25%,
see Figure 15. However, there is always a duplicate experiment (using the same
superficial velocities of both phases) that gives an axial dispersion coefficient with a
deviation less than 25%. So, the isotropic turbulence theory (Baird and Rice, 1975)
describes well the relationship between axial dispersion coefficient and energy
dissipation rate per unit mass for a liquid-liquid-solid fluidized bed, if those
combinations of superficial velocities are excluded for which the solids flow in the 3-
phase fluidized bed is not disturbed by the rising droplets.

For a bubble column (length/diameter ratio > 5) a c
1
value of ~ 0.35 was found in the
absence of slugging, and for the main tube of a liquid-impelled loop reactor, at lab-
scale (diameter 0.06m and length 0.44m), a c
1
value of 0.35 was also found (Van
Sonsbeek, 1992). So, based on the theory of Baird and Rice (1974), it can be said that
the mixing behavior of the continuous phase of a liquid-liquid-solid 3-phase
fluidized bed is comparable with both the mixing behavior of the continuous phase
of a bubble column and the mixing behavior of the continuous phase in the main
tube of a liquid-impelled loop reactor.
Figure 15. Absolute deviation between the measured axial dispersion coefficient and the predicted
axial dispersion coefficient as a function of the measured axial dispersion coefficient.



0
25
50
75
100
125
150
0 10 20 30 40
measured axial dispersion coefficient cm
2
s
-1
a
b
s

(

(
d
a
t
a

-

m
o
d
e
l
)

/

d
a
t
a
)

x

1
0
0
%
Mixing of the 3-phase fluidized bed
121
Conclusion

A method to determine the residence time distribution of the continuous phase of a
liquid-liquid-solid (gel beads) 3-phase fluidized bed bioreactor, using a KCl-tracer,
was evaluated and validated. It is concluded that the use of a KCl-tracer gives a
reliable residence-time distribution curve provided the pulse volume is not too large.
For this study a pulse volume of 3 ml 4M KCl resulted in reliable results.
Using the residence-time distribution curves measured for different superficial
velocities of the continuous phase (30 mM KCl) and dispersed phase (n-dodecane),
two different but comparable models, the axial dispersion model and the tanks-in-
series model, were applied to calculate the dispersion coefficient of the continuous
aqueous phase. It is concluded that the dispersion coefficients determined with both
models, show the same relationship between dispersion coefficient and superficial
velocities of both phases. However, the accuracy of the dispersion coefficient
calculated with the tanks-in-series model is much lower, because the number of tanks
can only have discrete values.
The dispersion coefficient of the continuous phase increases with the superficial
velocities of both phases, for flow conditions that belong in the structured solids flow
regime. For flow conditions that belong in turbulent solids flow regime, the
dispersion coefficient slightly decreases with an increasing continuous phase
superficial velocity and increases with an increasing dodecane superficial velocity.
These results are well explained by the energy dissipation rate per unit mass of the
continuous phase (P
m
); an increase in P
m
results in a higher dispersion coefficient and
a decrease in P
m
results in a lower dispersion coefficient.
The relationship between dispersion coefficient and P
m
as suggested by Baird
and Rice (1975) describes the experimental dispersion coefficients of the continuous
phase well, provided the dodecane superficial velocity is larger than 0.05 cm s
-1
, and
for a dodecane superficial velocity of 0.1 cm s
-1
the continuous phase superficial
velocity is smaller than 1.6 cm s
-1
. With these restrictions the total deviation between
experiment and model is 17.5%.
It is concluded that, regardless of the flow pattern, the mixing of the continuous
phase in liquid-liquid-solid (gel beads) fluidized bed is comparable with the mixing
of the continuous phase of either a bubble column or the main tube of a liquid-liquid
impelled loop reactor.


Chapter 5
122
Appendix A. Energy dissipation rater per unit mass for a 3-phase
fluidized-bed bioreactor.

In analogy to Van Sonsbeek et al. (1992) we assume that the power input is totally dissipated
in liquid motion of the continuous phase. The power input is equal to the difference between
the gain in potential energy and loss in pressure energy:

potential pressure disp
E E E ∆ − ∆ = A.1

The hydrostatic pressure gradient is given by
mix c hydrostati
H g P ρ = ∆ A.2

with the mixture’s density given by:
s s d d c c mix
ε ρ + ε ρ + ε ρ = ρ 14

For a liquid-liquid-solid fluidized bed, as described in the experimental part, both liquid
phases loose energy when they flow through this pressure gradient:

( )
d c mix pressure
U U H g A E + ρ = ∆ A.3
Both liquid phases gain potential energy

( )
c c d d potential
U U H g A E ρ + ρ = ∆ A.4

The energy dissipation rate per unit mass follows from implementing A.3 and A.4 in A.1 and
dividing by the mass of the continuous phase (ε
c
ρ
c
AH). After rewriting the resulting
equation for P
m
is:
d
c c
d mix
c
c c
c mix
m
U g U g P ⋅
ρ ε
ρ − ρ
+ ⋅
ρ ε
ρ − ρ
= 13
Mixing of the 3-phase fluidized bed
123
Nomenclature

A : area of the column m
2
C : concentration mol m
-3
c
1
: constant -
D
ax
: axial dispersion coefficient of the continuous phase m
2
s
-1
E : residence time distribution s
-1

E
disp
: dissipated energy J s
-1
∆E : change in energy J s
-1

g : gravitational constant m
2
s
-1
H : height of the fluidized bed m
k
i
: proportionality constant mol mA
-1
L : length scale m
m : mass mol
N : total number of tanks-in-series -
∆P
hydrostatic
: hydrostatic pressure gradient N m
-2
P
m
: energy dissipation per unit mass J kg
-1
s
-1
S : signal mA
t : time s
U
i
: superficial velocity of phase i m s
-1
V : volume of a section of the bioreactor m
3

v
∞,i
: terminal rise velocity of phase i m s
-1
x : axial coordinate m

greek
Φ : volume flux of the continuous phase m
3
s
-1
ε
i
: hold-up of phase i -
ρ
i
: density of phase i kg m
-3
τ : residence time s

subscript
c : continuous phase
d : dodecane phase
in, out : inlet, outlet
mix : mixture of the different phases
s : solid (gel beads) phase

124


125

6
Design of liquid-liquid-solid fluidized-bed bioreactors

Abstract

A liquid-liquid-solid 3-phase system is automatically formed, when in one apparatus
an immobilized biocatalyst is used for a biotransformation and an organic solvent for
extraction. The aim of this paper is to show that the application of such a 3-phase
reactor system for biotransformations strongly inhibited by product, will result in a
higher degree of conversion, compared to a conventional reactor system. For that,
different reactor configurations are discussed. These include a simple 3-phase
fluidized bed, but also more fancy options, like the different possibilities with a loop
reactor. It is concluded that building and operating a 3-phase system is nothing more
than a minor extension of conventional bioreactors. In the second part of this paper a
simple reactor model was developed for a 3-phase liquid-liquid-solid fluidized-bed
bioreactor. The influence of different parameters are discussed, e.g. the distribution
coefficient of product over medium/organic solvent, the toxic product concentration,
and the flux of the organic solvent. In the last part of this paper conditions have been
established for which this 3-phase reactor performs better than a conventional 2-
phase fluidized bed. At a given maximum substrate conversion rate, the distribution
coefficient is determined for which the 3-phase fluidized bed performs equally well
as the 2-phase fluidized bed. It appears that already a low distribution coefficient
(larger than 1 but less than 2) suffices for a better operation. Provided in situ
extraction is needed for such a biotransformation, a 3-phase fluidized bed, or less
simple 3-phase bioreactors are a good choice.

published as: Erik van Zessen, Arjen Rinzema and Johannes Tramper. 2001. In Multi-phase
bioreactor design. Chapter 12 Design of liquid-liquid-solid fluidized-bed bioreactors. Ed.
Cabral JMS, Mota M and Tramper J, Taylor and Francis (London, New York). p 225-247.
Chapter 6
126
Introduction

Advantages of using a liquid-liquid 2-phase system in biotransformations have been
discussed (Van Sonsbeek et al. 1993; Vermuë and Tramper 1995a; Tramper et al. 1992;
Adlercreutz 2000). Depending on the nature of the biotransformations, the reason for
applying a liquid-liquid 2-phase system is different. Examples of different
biotransformations carried out in a 2-phase system and the advantages of using such
a system are given by Vermuë and Tramper (1995a). In general one can profitably
choose for a liquid-liquid 2-phase system when:
− the substrate dissolves poorly in the medium; the organic solvent acts as a
reservoir;
− there is product inhibition; by using an organic solvent the product concentration
in the medium is lowered;
− the reaction equilibrium is unfavorable; the equilibrium products dissolve in the
organic solvent and the degree of conversion can be enhanced.
Another reason for using in situ extraction, the combination of bioconversion with
the first unit operation in the downstream processing, might be a facilitated product
recovery.
Obviously, using an organic solvent in bioconversions also leads to
disadvantages. For example, the toxicity of the solvent might reduce the biocatalytic
activity and stability (Vermuë et al. 1995b). Furthermore, in any bioconversion with
micro-organisms involved, surface active agents are present, whether excreted by the
micro-organism, or present due to cell lysis. This can result in non-desirable, stable
emulsions that are difficult to separate (Vermuë et al. 1995b).
Inhibition due to direct contact between micro-organisms and the organic
solvent can be reduced by immobilizing the micro-organism. Immobilization may
also reduce the amount of surface-active agents, and consequently prevent a stable
emulsion (Vermuë et al. 1995b). Immobilization of micro-organism has been studied
extensively (Wijffels et al. 1996). An elegant way of immobilizing micro-organisms is
a method that uses a natural gel solution (e.g. κ-carrageenan, alginate). Broth is
mixed with the gel solution and gel beads are made as described by for instance
Hunik and Tramper (1993). Obviously, immobilization will only work when there is
hardly any outgrowth of the micro-organisms, and little excretion of proteins.
In operating an extractive bioconversion one should decide whether to use a
bioreactor with bioconversion and extraction integrated, or use a plant set-up with
Design of a 3-phase fluidized bed
127
both processes separated. In the latter case the medium has to be circulated between
both apparatus at a high velocity, see also the section on loop reactors for more
detail. Using in one apparatus immobilized cells for a bioconversion and an organic
solvent for extraction, a 3-phase system is automatically formed. As apparatus,
stirred tank reactors, although favorable for good mixing and a high interfacial area
between medium and organic solvent, are not very useful; the harsh conditions in the
tank will destroy the gel beads. A stronger immobilization matrix would be required
than the natural gels mentioned above. Another option is using column reactors. For
this purpose a liquid analogue of the air-lift loop reactor, the liquid-impelled loop
reactor, was designed (Tramper et al. 1987) and hydrodynamically characterized by
Van Sonsbeek (1992a). Different biotransformations have been executed in this type
of reactor (Vermuë et al. 1995; Buitelaar et al. 1991; Van den Tweel et al. 1987; Mateus
et al. 1996). The application of this type of reactor is discussed in the section on loop
reactors. Another simple bioreactor is the liquid-liquid-solid 3-phase fluidized bed.
The reactor is almost identical to a liquid-impelled loop reactor, but the water flow is
controlled with a pump and not induced by a density difference. The hydrodynamics
of this reactor have been studied: the hold-up of the different phases as a function of
the fluxes of both phases (Van Zessen et al. 2003), as well as the mixing of the
medium phase (Van Zessen et al. 2003). This type of reactor has good mixing
characteristics, and seems promising for application in an in situ extraction
bioprocess.
The aim of this chapter is to show that the application of a 3-phase reactor
system for bioconversions strongly inhibited by product, will result in a higher
degree of conversion, compared to a conventional reactor system. Not all conditions
favor a 3-phase system, but it will be shown that with the right combination of
operating conditions (fluxes of both liquids), and physical parameters (distribution
coefficient and inhibition constant), a 3-phase system is always favorable.
Figure 1 summarizes different aspects looked upon in this paper. First,
possible process lay-outs and different reactor configurations for a liquid-liquid-solid
3-phase system are discussed. Next, the focus is on one of the reactor configurations:
a 3-phase fluidized-bed bioreactor. Then, a reactor model is developed, and
characteristics of this model are discussed. Thereafter, the conditions for which a 3-
phase fluidized-bed gives a better performance than a liquid-solid fluidized bed are
demonstrated.

Chapter 6
128
• Reactor Configuration
- 3-phase liquid-liquid-solid fluidized-beds, both liquids flow co-currently
- 3-phase liquid-liquid-solid fluidized-beds, water flow counter-currently
- 3 Liquid-impelled loop reactors

• Model derivation for reactor configuration 1
Simulations for a batch operation
Profiles inside a gel bead
Different distribution coefficients
Different toxic product concentrations
Different fluxes of organic solvent

• Comparison between a conventional reactor and reactor configuration 1
Evaluation of the distribution coefficient for which reactor configuration 1
performs equally
Batch operation is studied extensively
Continue operation is briefly discussed
Figure 1. Short overview of different aspects looked upon


Process lay-out and operation

A schematic presentation of a possible plant is shown in Figure 2. The core of the
plant is the bioreactor, i.e. a 3-phase fluidized bed. This column is fed by a
continuous medium flow, introduced at the bottom of this column. This up-flowing
stream is also used for fluidizing the biocatalytic particles. An immiscible solvent is
pumped through a sparger, and the formed droplets rise through the fluidized bed.
If coalescence of the droplets is easily achieved, which might be a prerequisite for
good operation, both liquids are separated at the top of the column and sent back to
their respective storage vessels. If separation between both liquids is not achieved by
a simple one-stage gravity settler, the 2-phase mixture can be separated outside the
column.
If the biocatalyst requires oxygen, than aeration of one or both storage vessels
will be an adequate method. Temperature control of the bioreactor can be done by
keeping the temperature of both storage vessels constant. pH is controlled by
keeping the pH of the substrate storage vessel constant.
Operating the bioreactor in the above-described manner is a batch operation;
accumulation of the product in the organic phase and depletion of the substrate will
occur. A continuous operation with respect to the substrate phase can be achieved by
adding a high substrate concentration to the storage vessel, keeping the substrate
Design of a 3-phase fluidized bed
129
concentration constant. To prevent overflow of the storage vessel part of the
circulating medium flow has to be purged, see also Figure 2.
Operating the bioreactor with a continuous supply of substrate will still result
in the accumulation of the product in the organic phase. Operating the organic phase
continuously is easily done by sending the total out-flowing organic phase to the
recovery section, and pumping purified organic phase into the column. This strategy
will be disadvantageous when the product concentration in the organic phase is low;
a large volume of organic solvent has to be clarified. A better strategy will be to allow
accumulation of the product to a certain concentration, below the equilibrium
concentration, and to send part of the out-flowing organic phase to the recovery section,
see Figure 2.
Figure 2. Process lay-out for a 3-phase fluidized-bed bioreactor

Above a plant lay-out and a process operation is discussed for a 3-phase fluidized-
bed bioreactor. It is assumed that both liquid phases flow upward co-currently. This
is true when the organic phase has a smaller density than the medium phase. When
the organic-phase density is larger, this liquid has to be introduced at the top of the
column, and both liquids have to be separated at the bottom of the column.
Another assumption is that the 3-phase fluidized-bed bioreactor is stable,
which means that the fluidized bed is in a hydrodynamic equilibrium; particles
remain in the column and the droplets do not coalesce in the fluidized bed.
Experiments with different kinds of gel beads have shown that a stable fluidized bed
exists, at least, for gel beads with a density of 1060 kg m
-3
and a diameter of 2.3 mm.
Applying gel beads with a density of 1010 kg m
-3
and a diameter of 2 mm in a 3-
phase fluidized bed resulted in the wash out of the beads at any organic flow rate
larger than 10
-4
m
3
(m
2
s)
-1
. It can be concluded that a stable 3-phase fluidized bed is
possible for particles with a settling velocity larger than 5.0 cm s
-1
. Coalescence of
droplets can be prevented by carefully adjusting both flows.
air
Buffer
vessel
2
to purification section
Purified organic solvent
air
Buffer
vessel
1
pH
DO
Purge
Substrate
T
Chapter 6
130
Alternatives
Below two different alternatives are given for operating a 3-phase liquid-liquid-solid
system, see also Figure 1.

Counter-current water flow
A 3-phase fluidized bed with both liquids flowing co-currently is not possible for
each type of particle. For particles with a terminal settling velocity larger than 5.0 cm
s
-1
a stable bed is possible, but for particles with a velocity smaller than 2 cm s
-1
a
stable bed could not be obtained (Van Zessen, unpublished). We do not know
whether a stable fluidized bed is possible for particles with a terminal settling
velocity between 2 and 5.0 cm s
-1
.
A stable 3-phase fluidized bed is possible for those particles with a settling
velocity smaller than 2 cm s
-1
, if a different strategy is followed. In operating this
strategy, droplets rise from bottom to top and medium flows counter-currently, i.e.
from top to bottom, see Figure 3. A limit situation of this strategy is the case in which
there is no medium flow. Firstly, this situation is shortly discussed. Thereafter, the
counter-current operation is discussed.
For bubble columns, it is known that gel beads can be fluidized - kept in
suspension - by the rising gas bubbles. As a liquid-liquid-solid 3-phase fluidized bed
is the liquid analogue of a gas-liquid-solid 3-phase fluidized bed, it should be
possible to create a droplet column with solids kept in suspension. Indeed, it was
possible to keep two different kinds of κ-carrageenan gel beads in suspension in a
droplet column (Van Zessen, unpublished). These gel beads had a density of 1007.4
kg m
-3
and 1005.4 kg m
-3
and a diameter of 1.97 mm and 3.12 mm. However, to
maintain the gel beads in suspension a minimal solvent flux was required.
Experimental data on hydrodynamic parameters, such as hold-up, were not
determined; only visual observations were made on this operating strategy. Provided
the solvent flux was higher than a minimal value, this 3-phase system was stable for
at least 24 hours.
Design of a 3-phase fluidized bed
131
Figure 3. Alternative process lay-out: water flows counter-current with respect to the droplets flow

As the droplet column with suspended gel beads worked well without any water
flow, we also tried to operate the set-up with water flowing counter-currently, see
Figure 3. The water flux puts a downward directed force on the gel beads, whereas
the force put on the gel beads by the organic solvent flux is directed upwards.
Obviously, when the water flux becomes too high, the gel beads will settle. Again,
we only made visual observations for this set-up. These observations can be
summarized as follows:
- at a fixed water flux, the organic solvent flux has to have a minimal value
in order to make this system work. Increasing the organic solvent flux
resulted in a larger bed height, i.e. a lower gel-bead hold-up;
- at a fixed organic solvent flux, higher than the minimum flux required for
stable operation, the water flux can be increased to a maximum value. At
higher values the system will not work, i.e. gel beads settle, and droplets
coalesce.
This operating strategy gives a couple of advantages over a co-current strategy. In
extraction processes counter-current operation is always better than co-current
operation, as there is always a higher driving force for mass transfer. In co-current
operation the gel-bead hold-up decreases with an increasing medium flux, whereas
in counter-current operation the gel-bead hold-up increases with an increasing
medium flux. It is also observed that at a higher organic solvent flux a higher water
flux can be applied; this means that the throughput of substrate can be higher in
counter-current operation.

Buffer
vessel
2
Buffer
vessel
1
Chapter 6
132
Loop reactors
Another type of bioreactors are the so-called loop reactors: these reactors consist of
two tubes connected to each other; internal and external designs exist (Van ‘t Riet
and Tramper, 1991). In order to function, there has to be a difference in density of the
mixture in both tubes. A density difference results in a pressure difference and water
will flow between both tubes: the direction of this water flow goes from a high to a
low density.
This principle has been applied in air-lift reactors (Chisti, 1989). The liquid
analogue, droplets instead of gas bubbles, the so-called liquid-impelled loop reactor,
has been described by Van Sonsbeek (1992a). This type of reactor with particles, κ-
carrageenan gel beads, has been used by Mateus et al. (1996) and Vermuë et al. (1995)
for biotransformations.
Figure 4 shows four possible reactor configurations for the application of a 3-
phase system in a loop reactor. First, it is assumed that the particles remain in one
column, and are not circulating. Particles circulating between both tubes are
discussed below. In parts A and B of this figure, a bed of biocatalytic particles and a
spray extraction column are separated. In part C and D, conversion and extraction
are fully integrated.
Figure 4. Loop reactors (see text for explanation)


In part A droplets of an organic solvent rise from the bottom to the top of the column
as their density is lower than the water density. Consequently, pressure is lower in
the spray column, and water flows from top to bottom in the column with particles.
B
C
D
A
Design of a 3-phase fluidized bed
133
If the density of the particles is higher than the water density, a packed bed will be
formed. If the density of the particles is lower than the water density, and the water
flux is larger than the rise velocity of the swarm of particles, a fluidized bed will be
formed. This so-called inverse fluidization has been studied by Fan et al. (1982).
In part B, droplets of an organic solvent settle from the top to the column, as
the density is higher than the medium density. Obviously, water flows from bottom
to top in the column with particles. This set-up will work, if the particle density is
higher than the water density. A fluidized bed will be observed, if the water flow is
higher than the minimum fluidization velocity.
Part C of Figure 4 shows a 3-phase mixture in one column and medium in the
other column of the loop reactor. The density of the organic solvent is less than the
medium density and solvent droplets will rise. A 3-phase fluidized bed will only
exists when the density of this mixture is less than the density of the medium in the
other column. In that case water will flow from bottom to top in the 3-phase column.
For example, in our laboratory experiments have been done on the hold-ups in a 3-
phase system, i.e. water, dodecane and gel beads (density equals 1060 kg m
-3
). At a
water flux of 1.8×10
-2
m s
-1
and a dodecane flux of 0.91×10
-2
m s
-1
, a gel bead hold-up
of 0.20 and a dodecane hold-up of 0.08 was measured. This gives a mixture density of
992 kg m
-3
. The density difference between both columns is in this case large enough
to obtain this water flux of 1.8×10
-2
m s
-1
(Van Sonsbeek et al. 1990).
A different situation exists when the density in the 3-phase system is larger
than the density of the medium in the other column, due to a higher solids hold-up
or a lower organic solvent hold-up. In this case water will flow from top to bottom.
Provided the density of the solids is close to water, this configuration can also work;
see the preceding paragraph ‘Counter-current water flow’.
The column with only medium can be aerated. However, aeration results in
the decrease of the mixture density in this column, and the circulation velocity of the
medium between both columns is changed. In order a 3-phase fluidized bed exists,
the aeration must be carefully carried out. Aeration results in an air hold-up, and
hence a smaller mixture density in the aerated column. Consequently, the density
difference between both columns becomes less, and the circulation velocity
decreases. At this smaller medium velocity a 3-phase fluidized bed must still exist.
Part D of Figure 4 is the same as part C, but in this case the density of the
organic solvent is larger than the density of water, hence droplets settle. Water will
always flow in the direction indicated in this figure; from top to bottom in the 3-
Chapter 6
134
phase column. Only if the biocatalytic particles have a density smaller than water, i.e.
they rise in the column, this configuration will work.
So far it is assumed that the particles remain in a fluidized state. However, the
water circulation velocity can be so high, that it exceeds the terminal settling velocity
of a single particle. In that case particles flow together with the water, and are
present in both columns. This principle is widely used in air-lift bioreactors (Heijnen
et al. (1997).


Performance of a conventional bioreactor and a 3-phase fluidized bed
bioreactor

In the preceding section possible reactor configurations have been discussed.
Whether such a 3-phase reactor will work better than a conventional reactor, is the
topic of this section. We decided to compare a 2-phase fluidized-bed bioreactor, the
conventional reactor, with a 3-phase liquid-liquid-solid fluidized-bed bioreactor.
Comparing two different reactors, one should consider a fair comparison. This
comparison can be based on cost, but in that case one should consider total cost, i.e.
cost of a total production plant. Making a total process design and optimizing this
design for both reactors, goes beyond the scope of this paper. One can also compare
both reactors on their performance, i.e. degree of conversion of the substrate, or
synthesis rate of product.
Before we go into a detailed comparison of both bioreactors, the model used
for the calculations is described.


Model Derivation

Below, the model we used to calculate the performance of both bioreactors is briefly
discussed. A general total model for the design of a reactor is schematically shown in
Figure 5. This model consists of three major parts:
• hydrodynamics of the reactor
• kinetics of the bioconversion
• mass transfer of compounds between the different phases

Design of a 3-phase fluidized bed
135
Together with mass balances over each phase for the different components, the total
model is complete.

Figure 5. Model outline

Investigated system
In our laboratory hydrodynamic studies have been done on a 3-phase system
consisting of water, dodecane and gel beads. This system is used as the model
system. Obviously, highly complicated bioconversions can be involved in producing
industrially interesting products, in which more than one substrate and product
influence the kinetics. We have chosen a very simple one-to-one conversion:
substrate S gives product P. It is further assumed that substrate is dissolved in the
medium and insoluble in the organic solvent, transported to the gel bead and there
converted to product. The product itself is transported out of the gel bead to the
medium phase and subsequently to the organic solvent phase.

Hydrodynamics
This part of the model considers the hold-up and mixing of the different phases
present. To model the medium flow, the fluidized bed is divided in a number of
ideally mixed tanks. For a fluidized bed of 1 m high, and for medium fluxes between
0.5×10
-2
and 2 ×10
-2
m s
-1
, the number of tanks has been experimentally determined. It
appears that for medium fluxes the number of ideally mixed tanks is equal to 13 for
the 2-phase fluidized bed and 3 for the 3-phase fluidized bed.
In order to make the reactor model less complicated, it is assumed that gel
beads remain in one tank and do not circulate between the different tanks.
A model has been developed for predicting the hold-ups in a 3-phase
fluidized bed (Van Zessen, 2003), but we chose to use here the experimentally
Hydrodynamics
U
d
U
c
Hold-ups Mixing
phases
Biokatalysator
Kinetics
Substrate
consumption
rate
Product
formation
rate
C
s
C
p
Mass transfer
between phases
Reactor
Performance
C
s
C
p
Chapter 6
136
determined hold-ups as they are more accurate. The dodecane hold-up and gel bead
hold-up as a function of medium flux and dodecane flux is shown in Figure 6.
Figure 6a. Droplet hold-up as a function
of the water and dodecane flux.
● U
d
= 0.05 cm s
-1
◊ U
d
= 0.10 cm s
-1

■ U
d
= 0.18 cm s
-1
O U
d
= 0.27 cm s
-1

▲U
d
= 0.44 cm s
-1
□ U
d
= 0.54 cm s
-1

♦ U
d
= 0.73 cm s
-1
× U
d
= 0.91 cm s
-1

Figure b. Gel bead hold-up as a function of
the water and dodecane flux.
● U
c
= 0.00 cm s
-1
◊ U
c
= 0.10 cm s
-1

■ U
c
= 0.29 cm s
-1
♦ U
c
= 0.42 cm s
-1

O U
c
= 0.54 cm s
-1

▲ U
c
= 0.75 cm s
-1
□ U
c
= 0.91 cm s
-1


Table I gives an overview of the assumptions made and the models used for
determining the hold-up and mixing characteristics of both reactors.

Table I. Hydrodynamic characteristics for a liquid-fluidized bed and a liquid-liquid-solid
fluidized bed
2-phase fluidized bed 3-phase fluidized bed
mixing
water phase
gel bead phase
droplet phase
number of tanks-in-series (15)
ideally mixed per tank
-
number of tanks-in-series (3)
ideally mixed per tank
plug flow

hold-up Richardson and Zaki model with
experimentally determined
parameters

n
w c
v U ε

=
Experimentally determined,
see Figure 6


Mass transfer
Mass transfer between the different phases in a 2- or 3-phase system can
schematically be presented as shown in Figure 7. To describe mass transfer inside the
gel bead we used Fickian diffusion of substrate and product. For describing the mass

0
0.02
0.04
0.06
0.08
0.1
0.5 0.75 1 1.25 1.5 1.75 2
U
c
cm s
-1
ε
d

0.15
0.25
0.35
0.45
0.55
0.5 0.75 1 1.25 1.5 1.75 2
U
c
cm s
-1
ε
s
Design of a 3-phase fluidized bed
137
transfer from gel bead interface to the bulk of the medium, we used the film model
for substrate as well as product. The film model was also used for the mass transfer
of the product from the bulk to the interface between medium and organic-solvent
droplet. Mass transfer from this interface to the bulk of the organic-solvent droplet
was also described by the film model. Further it is assumed that the distribution
coefficient between gel bead and surrounding liquid is equal to one. Next, it is
assumed that there is thermodynamic equilibrium at the interface between droplet
and medium. The equilibrium distribution at the interface can be described with a
constant distribution coefficient (m):

int w,
p
int o,
p
mC C =
The mass transfer rate between the different phases can be described with the
equations given in Table II; the rates are a combination of mass transfer per gel bead
or droplet, and the number of gel beads or droplets present, respectively. The
resulting rates are given per volume reactor.
Figure 7. Profiles in a multi-phase system: — product profile, ··· substrate profile


The mass transfer coefficient for the transfer of product from medium to organic
phase (K) is a combination of the transfer coefficient from bulk to interface at both
sides of the interface:
1
o
p drop,
w
p drop,
mk
1
k
1
K









+ =

The mass transfer coefficients k
bead,s
, k
bead,p
, and k
drop,p
w
are calculated with the
equation of Ranz and Marshall (1952); the mass transfer coefficient k
drop,p
o
is
calculated with the equation of Newman (1931); see appendix A for equations. As

drop water gel bead
component profile
C
i
C
bulk
C
bulk
C
interface
Chapter 6
138
contact time in the latter equation, we used the residence time of a droplet ( t = H
bed

ε
o
/U
d
).

Table II. Substrate and product transfer rates between the different phases
medium / gel bead medium / droplet
substrate
( )
s bead
int w,
s
bulk w,
s s bead,
d
6
C C k
ε
⋅ −
no transfer to the droplet

product
( )
s bead
int w,
p
bulk w,
p p bead,
d
6
C C k -
ε
⋅ −
o drop
bulk d,
p bulk w,
p
d
6
m
C
C K
ε












Kinetics
Let us consider bioconversions carried out by either non-growing cells or enzymes
immobilized in κ-carrageenan gel beads. It is further assumed that there is only one
limiting substrate, which dissolves well in medium and not in the organic solvent.
Medium is water containing all other essential nutrients. Straightforward Michaelis-
Menten kinetics with first order product inhibition are considered:









+
= −
tox
p
p
s m
s max
s
C
C
1
C K
C v X
r
When the product concentration reaches the toxic concentration (C
p
tox
) the substrate
consumption rate is zero. More complicated kinetics can be used, but this is the
simplest model to demonstrate the benefits for in situ extraction.

Mass balances
Figure 8 shows how the liquid-liquid-solid 3-phase fluidized bed is divided in
ideally stirred tanks to represent flow characteristics, see also Table 1.
The total 3-phase fluidized bed is divided in a number of tanks. In each tank
the medium is ideally mixed. The gel beads remain in each tank. The organic-solvent
phase is in each tank divided in 15 ideally mixed tanks-in-series to mimic plug flow
behavior.
Design of a 3-phase fluidized bed
139
Figure 8. Schematic representation of the 3-phase fluidized bed


Mass balances are derived for single tanks. The total derivation of these balances is
given in appendix A. The resulting dimensionless mass balances with the initial and
boundary conditions are:

Within this model two different kinds of parameters can be discerned:
1. parameters that can be easy manipulated by the designer,
i.e. operating variables like the fluxes of the medium and organic solvent, and
hence the hold-up of the different phases, initial concentration of substrate and
product in the different phases, and reactor geometry.
2. parameters that are difficult to manipulate,
e.g. the physical parameters like density and viscosity of the pure liquids, the
mass transfer coefficients, and the diffusion coefficients of the compounds.
Ud
Uc
1
2
3 i=15
i=15
i=15
Chapter 6
140
Particle mass balances

substrate
( )
0 , w
s s
2
p
n , p
s
n , p
s
n , p
s
C D
r
s
r -
z
X
z
2
2
z
X
2
Fo
X



+


=




( )
0 X 0, t z,
0
dz
dX
0, z t,
dz
dX
r
D
X X k 1, z t,
0 z
s
1 z
s
p
s n int, w,
s
n w,
s s bead,
= = ∀
= = ∀
= − = ∀
=
=




product
0 , w
s s
s
n p,
p
s
p
2
n p,
p
2
s
p
n p,
p
C D
2
p
r
) r - (
z
X
z
2
D
D
z
X
D
D
Fo
X
+


+


=




( )
0 X 0, t z,
0
dz
dX
0, z t,
dz
dX
r
D
X X k 1, z t,
p
0 z
p
1 z
p
p
p n int, w,
p
n w,
p p bead,
= = ∀
= = ∀
= − = ∀
=
=



Design of a 3-phase fluidized bed
141
Medium-phase mass balances

substrate

( ) ( )
n int, w,
s
n w,
s
s
2
p
s bead,
w
s
bead
n w,
s
1 n w,
s
s
2
p
t w
w
n w,
s
X X
D
r
k
d
6
X X
D
r
H
U
Fo
X

ε
ε
− −
ε
=





batch operation t=0, X
s
w
=1 for each tank n
continuous operation t=0, X
s
w
=0 for each tank n



product

( ) ( )









ε
ε

+ −
ε
ε
+ −
ε
=



m
X
X
D
r
d
6
K
X X
D
r
k
d
6
X X
D
r
H
U
Fo
X
i
n w,
p
n w,
p
s
2
p
w
o
o
n w,
p
n int, w,
p
s
2
p
p bead,
w
s
bead
n w,
p
1 n w,
p
s
2
p
t w
w
n w,
p


batch or continuous operation t=0, X
p
w
=0 for each tank n





Organic-solvent-phase mass balances

( )








− + −
ε
=



m
X
X
D
r
d
6
K X X
D
r
H
U
Fo
X
i
n w,
p
n w,
p
s
2
p
o
i o,
p
1 i o,
p
s
2
p
i t, d
d
i o,
p

batch or continuous operation t=0, X
p
o,i
=0 for each tank i



Chapter 6
142
Calculations

Before we compare a 3-phase fluidized-bed bioreactor with its 2-phase analogue, we
will first show some typical results of the model. For these simulations we took a
batch operation strategy for both liquids, and both liquids flew co-currently, see also
Figure 2. Figure 1 gives a short overview of the different simulations done. First we
show a typical time course of substrate and product concentration inside the gel
bead. Next the influence of the distribution coefficient and toxic product
concentration is shown. These parameters are system dependent; a toxic product
concentration depends on the micro-organism used, and can hardly be influenced;
the distribution coefficient depends on the two liquids used, but can be influenced by
adding specific compounds to the organic solvent for enlarging the distribution
coefficient. The influence of the dodecane flux is discussed at the end of this section.
The parameters that were used throughout the simulations are summarized in Table
III.
Design of a 3-phase fluidized bed
143

Table III Overview of parameters used in simulations
Reactor Constants
# nozzles 209
diameter nozzle 0.001 m
reactor height 1 m
reactor diameter 0.06 cm

Physical parameters
density medium 998 kg m
-3
density dodecane 742.7 kg m
-3
density gel bead 1065.1 kg m
-3

viscosity medium 0.0009325 Nm s
-1

viscosity dodecane 0.00135 Nm s
-1

surface tension between water and
dodecane
0.040 N m
-1
diameter gel bead 2.76 mm
diffusion coefficient substrate 10
-9
m
2
s
-1

diffusion coefficient product 10
-9
m
2
s
-1

distribution coefficient simulation depended

Hydrodynamics
hold-ups 3-phase fluidized bed simulation depended
Number of tanks per m reactor
3-phase fluidized bed
2-phase fluidized bed

3
13

Richardson and Zaki constants 2-phase
fluidization
n
v



2.35
0.0377


m s
-1


Kinetics
Michaelis-Menten constant 1 mol m
-3

toxic product concentration simulation depended mol m
-3

maximum substrate conversion rate simulation depended mol (m
3
s)
-1

Operation parameters
medium flux simulation depended m s
-1

dodecane flux simulation depended m s
-1

initial substrate concentration 1000 mol m
-3

initial product concentration 0 mol m
-3

Chapter 6
144
In Figure 9 a typical time course of substrate and product concentration inside the gel
bead is shown. The maximum substrate conversion rate was 0.01 mol (m
3
s)
-1
, the
toxic product concentration was 10 mol m
-3
, and the distribution coefficient was 1000.
We took a medium flux of 1.29×10
-2
m s
-1
and a dodecane flux of 0.91×10
-2
m s
-1
,
which resulted in gel bead hold-up of 0.32 and a dodecane hold-up of 0.093.
Figure 9a:
Substrate diffuses into the gel bead and gradually penetrates towards the
centre of the gel bead. As time progresses the concentration increases, the
profile flattens, and a maximum concentration is reached. At larger times the
concentration decreases as result of consumption, until all substrate is
converted into product. With the set of parameters used for calculating the
profiles in Figure 9a, total conversion of substrate is reached. However, a
combination of parameters is easily found, for which the substrate is not
completely converted, hence substrate concentration inside the gel bead is not
equal to zero. Substrate will be converted to product, as long as the product
concentration at any place inside the gel bead is less than the toxic product
concentration.
Figure 9b:
As substrate is converted, product is formed. Early in the process little
product is present. Gradually product appears throughout the whole gel bead.
As product is initially formed near the surface of the gel bead, product not
only diffuses out of the gel bead but also towards the centre; consequently a
maximum concentration is found inside the gel bead. This maximum
concentration progresses towards the centre of the gel bead in time. Hence,
after some time product concentration is highest in the centre. After some time
also a maximum concentration profile is reached. If at this point of time, there
is still mass transfer between the organic and medium phase, the product
concentration inside the gel bead will decrease. This transfer of product will
continue until equilibrium is reached between medium and organic phase.
Design of a 3-phase fluidized bed
145


Figure 9. Profile of substrate (A) and product (B) concentration in a gel bead.
U
c
=1.29 cm s
-1
U
d
=0.91 cm s
-1
, ε
s
=0.32, ε
o
=0.0925;
Xv
max
=0.01 mol (m
3
s)
-1
, C
p
tox
=10 mol m
-3
, m=1000.
♦ t = 0:0:3 O t = 0:0:33, ▲t = 0:1:53, ◊ t = 0:6:40, ● t = 0:19:0,
□ t = 19:6:0 ■ t = 66:0:0, × t = 98:0:0, + t = 111:0:0.

Before discussing the influence of the distribution coefficient and toxic product
concentration on the time course of the substrate concentration in the medium phase,
and the time course of product concentration in the medium and organic phase, first
some general remarks:
• The time course profile of the substrate concentration in the water phase shows
two processes with a different time constant; the first process is completed after
roughly 100 seconds, and is more or less independent of the distribution
coefficient and the toxic product concentration. It can be seen in Figure 9b that
around that point in time the gel bead is completely filled with substrate. The
second process is the gradual conversion of substrate to product, and is
dependent on the distribution coefficient and the toxic product concentration.
• Complete conversion of substrate can be reached with some combinations of
distribution coefficient and toxic product concentration.
• If substrate is not completely converted, than the end-product concentration
inside the gel bead and the medium phase is equal to the toxic product
concentration.
• At the point in time where substrate is not longer converted, there is equilibrium
for the product concentration between the medium phase and the organic phase.

0
0.2
0.4
0.6
0 0.2 0.4 0.6 0.8 1
dimensionless place
d
i
m
e
n
s
i
o
n
l
e
s
s

s
u
b
s
t
r
a
t
e

c
o
n
c
e
n
t
r
a
t
i
o
n
A

0
0.002
0.004
0.006
0.008
0 0.2 0.4 0.6 0.8 1
dimensionless place
d
i
m
e
n
s
i
o
n
l
e
s
s

p
r
o
d
u
c
t

c
o
n
c
e
n
t
r
a
t
i
o
n
B
Chapter 6
146
The influence of the distribution coefficient (m) is shown in Figure 10a, 10b and 10c.

Figure 10a and 10b shows the concentrations in the medium phase, whereas Figure
10c shows the product concentration in the dodecane phase. The maximum substrate
conversion rate was 0.01 mol (m
3
s)
-1
, and the toxic product concentration was 10
mol m
-3
. We took a medium flux of 1.29×10
-2
m s
-1
and a dodecane flux of 0.91×10
-2
m
s
-1
, which resulted in a gel bead hold-up of 0.32 and a dodecane hold-up of 0.093.
Figure 10a clearly shows that a larger m results in the end in a lower substrate
concentration in the medium phase, hence substrate is converted to a higher degree.
Naturally, when substrate is not converted totally, the product concentration in the
medium phase is equal to the toxic product concentration, Figure 10b. As there is
equilibrium between the medium and organic phase, the product concentration in
the organic phase is higher for a higher m, Figure 10c. If the m is high enough, than
at the end of the batch operation the product concentration in the medium phase will
be lower than the toxic product concentration. Consequently, substrate is totally
converted. This is observed for a distribution coefficient of 1000. The end of the batch
operation is reached at an earlier time for a lower m.
The influence of the toxic product concentration (C
p
tox
) is shown in Figure 11.
Figures 11a and 11b show the concentrations in the medium phase, whereas Figure
11c shows the product concentration in the dodecane phase. Obviously, a higher
C
p
tox
results in a lower substrate concentration, hence a more complete conversion
and the product concentration goes to C
p
tox
. This concentration is reached earlier for
a lower C
p
tox
. At C
p
tox
equal to 100, substrate is completely converted, and C
p
tox
is not
reached in the medium phase.
Design of a 3-phase fluidized bed
147


F
i
g
u
r
e

1
1
.

P
r
o
f
i
l
e

o
f

s
u
b
s
t
r
a
t
e

(
A
)

a
n
d

p
r
o
d
u
c
t

(
B
)

c
o
n
c
e
n
t
r
a
t
i
o
n

i
n

t
h
e

m
e
d
i
u
m

p
h
a
s
e

a
t

d
i
f
f
e
r
e
n
t

t
o
x
i
c

p
r
o
d
u
c
t

c
o
n
c
e
n
t
r
a
t
i
o
n
s
.

P
r
o
f
i
l
e

o
f

p
r
o
d
u
c
t

c
o
n
c
e
n
t
r
a
t
i
o
n

i
n

t
h
e

d
o
d
e
c
a
n
e

p
h
a
s
e

(
C
)

a
t

d
i
f
f
e
r
e
n
t

t
o
x
i
c

p
r
o
d
u
c
t

c
o
n
c
e
n
t
r
a
t
i
o
n
s
.

U
c
=
1
.
2
9

c
m
/
s
,
U
d
=
0
.
9
1

c
m
/
s
,

ε
s
=
0
.
3
2
,

ε
o
=
0
.
0
9
2
5
;

X
v
m
a
x
=
0
.
0
1

m
o
l
/
m
3
s
,

m
=
1
0
0
,








C
p
t
o
x
=
1

m
o
l

m
-
3

,

-

-

-
,

C
p
t
o
x
=
1
0

m
o
l

m
-
3
,



C
p
t
o
x
=
5
0

m
o
l

m
-
3
,

O

C
p
t
o
x
=
1
0
0

m
o
l

m
-
3

F
i
g
u
r
e

1
0
.

P
r
o
f
i
l
e

o
f

s
u
b
s
t
r
a
t
e

(
A
)

a
n
d

p
r
o
d
u
c
t

(
B
)

c
o
n
c
e
n
t
r
a
t
i
o
n

i
n

t
h
e

m
e
d
i
u
m

p
h
a
s
e

a
t

d
i
f
f
e
r
e
n
t

d
i
s
t
r
i
b
u
t
i
o
n

c
o
e
f
f
i
c
i
e
n
t
s
.

P
r
o
f
i
l
e

o
f

p
r
o
d
u
c
t

c
o
n
c
e
n
t
r
a
t
i
o
n

i
n

t
h
e

d
o
d
e
c
a
n
e

p
h
a
s
e

(
C
)

a
t

d
i
f
f
e
r
e
n
t

d
i
s
t
r
i
b
u
t
i
o
n

c
o
e
f
f
i
c
i
e
n
t
s
.


U
c
=
1
.
2
9

c
m

s
-
1
,

U
d
=
0
.
9
1

c
m

s
-
1
,

ε
s
=
0
.
3
2
,

ε
o
=
0
.
0
9
2
5
;

X
v
m
a
x
=
0
.
0
1

m
o
l

m
-
3

s
-
1
,

C
p
t
o
x
=
1
0

m
o
l

m
-
3
,




m
=
1
,

-

-

-
,

m
=
1
0
,




m
=
1
0
0
,


O


m
=
1
0
0
0


0
0
.
2
0
.
4
0
.
6
0
.
8 1
1
1
0
1
0
0
1
0
0
0
1
0
0
0
0
1
0
0
0
0
0
1
0
0
0
0
0
0
t
i
m
e

(
s
)
d i m e n s i o n l e s s s u b s t r a t e
c o n c e n t r a t i o n
A
0
0
.
0
0
2
0
.
0
0
4
0
.
0
0
6
0
.
0
0
8
0
.
0
1
1
1
0
1
0
0
1
0
0
0
1
0
0
0
0
1
0
0
0
0
0
1
0
0
0
0
0
0
t
i
m
e

(
s
)
d i m e n s i o n l e s s p r o d u c t
c o n c e n t r a t i o n
B
0 1 2 3 4 5 6 7
1
0
0
1
0
0
0
1
0
0
0
0
1
0
0
0
0
0
1
0
0
0
0
0
0
t
i
m
e

(
s
)
d i m e n s i o n l e s s p r o d u c t
c o n c e n t r a t i o n i n o r g a n i c
p h a s e
C

0
0
.
2
0
.
4
0
.
6
0
.
8 1
1
1
0
1
0
0
1
0
0
0
1
0
0
0
0
1
0
0
0
0
0
1
0
0
0
0
0
0
t
i
m
e

(
s
)
d i m e n s i o n l e s s s u b s t r a t e
c o n c e n t r a t i o n
A
0
0
.
0
2
0
.
0
4
0
.
0
6
1
0
0
1
0
0
0
1
0
0
0
0
1
0
0
0
0
0
1
0
0
0
0
0
0
t
i
m
e

(
s
)
d i m e n s i o n l e s s p r o d u c t
c o n c e n t r a t i o n
B
0 2 4 6
1
0
0
1
0
0
0
1
0
0
0
0
1
0
0
0
0
0
1
0
0
0
0
0
0
t
i
m
e

(
s
)
d i m e n s i o n l e s s p r o d u c t
c o n c e n t r a t i o n o r g a n i c p h a s e
C
Chapter 6
148
The influence of the organic solvent flux is shown in Figure 12. Figures 12a and 12b
show the concentrations in the medium phase, whereas Figure 12c shows the
product concentration in the dodecane phase. We kept the medium flux constant and
changed the dodecane flux. The dodecane flux influences the hold-up of the different
phases. At a constant medium flux, a higher dodecane flux gives a higher dodecane
hold-up, but a lower gel bead hold-up. Figure 12a shows that complete conversion of
substrate is reached for the highest U
d
, although the gel-bead hold-up is the lowest
(U
d
=0.75×10
-2
m s
-1
, ε
s
=0.26; U
d
=0.14×10
-2
m s
-1
, ε
s
=0.30). Obviously, at the highest
dodecane flux, the product concentration in the medium phase is always below the
toxic product concentration, Figure 12b. For the other dodecane fluxes applied in the
simulations, the product concentration in the medium phase is equal to the toxic
product concentration, Figure 12b. This concentration is reached at an earlier point of
time for the lowest U
d
. The time course of product concentration in the organic phase
follows the time course for the product concentration in the medium phase.

Figure 12. Profile of substrate (A) and product (B) concentration in the medium phase at different
fluxes of the organic phase. Profile of product concentration in the organic phase (C) at different fluxes
of the organic phase.
U
c
=1.49 cm s
-1
, Xv
max
=0.01 mol m
-3
s
-1
, m=100, C
p
tox
=100 mol m
-1
. ▬ U
d
=0.14 cm s
-1
, ε
s
=0.30,
ε
o
=0.012; - - - U
d
=0.36 cm s
-1
, ε
s
=0.27, ε
o
=0.036; ● U
d
=0.75 cm s
-1
, ε
s
=0.26, ε
o
=0.069.

0
0.2
0.4
0.6
0.8
1
1 10 100 1000 10000 100000 1000000
time (s)
d
i
m
e
n
s
i
o
n
l
e
s
s

s
u
b
s
t
r
a
t
e

c
o
n
c
e
n
t
r
a
t
i
o
n
A
0
0.02
0.04
0.06
0.08
0.1
100 1000 10000 100000 1000000
time (s)
d
i
m
e
n
s
i
o
n
l
e
s
s

p
r
o
d
u
c
t

c
o
n
c
e
n
t
r
a
t
i
o
n
B
0
2
4
6
8
10
100 1000 10000 100000 1000000
time (s)
d
i
m
e
n
s
i
o
n
l
e
s
s

p
r
o
d
u
c
t

c
o
n
c
e
n
t
r
a
t
i
o
n

o
r
g
a
n
i
c

p
h
a
s
e
C
Design of a 3-phase fluidized bed
149
We kept the medium flux constant and changed the dodecane flux. The dodecane
flux influences the hold-up of the different phases. At a constant medium flux, a
higher dodecane flux gives a higher dodecane hold-up, but a lower gel bead hold-up.
Figure 12a shows that complete conversion of substrate is reached for the highest U
d
,
although the gel-bead hold-up is the lowest (U
d
=0.75×10
-2
m s
-1
, ε
s
=0.26; U
d
=0.14×10
-2

m s
-1
, ε
s
=0.30). Obviously, at the highest dodecane flux, the product concentration in
the medium phase is always below the toxic product concentration, Figure 12b. For
the other dodecane fluxes applied in the simulations, the product concentration in
the medium phase is equal to the toxic product concentration, Figure 12b. This
concentration is reached at an earlier point of time for the lowest U
d
. The time course
of product concentration in the organic phase follows the time course for the product
concentration in the medium phase.


Comparison between a 2-phase and a 3-phase fluidized-bed bioreactor

In our comparison we did the simulations for batch operations:
substrate is circulated until a certain amount of substrate is converted. In a 2-
phase fluidized bed substrate is not totally converted due to product
inhibition. When the product concentration reaches the toxic product
concentration in the medium phase the conversion stops. At this moment, the
product concentrations inside the gel bead and in the medium phase are the
same and equal to the toxic product concentration. So, the total amount of
moles of product produced is equal to the toxic product concentration times
the total reactor volume. In order to compare a 3-phase fluidized bed with a 2-
phase fluidized bed we calculated the time necessary to reach 99% of the total
amount that can be converted in a 2-phase fluidized bed. Obviously, there is a
lot of freedom in totally or partly recirculating the organic solvent in the 3-
phase fluidized-bed bioreactor, see Figure 2. We used in this strategy a total
reflux as a fair comparison. An economic evaluation must show whether a
total reflux yields the lowest cost, but this is not the topic of this paper.
At the end some remarks are made about operating the bioreactor continuously.
Chapter 6
150
Batch operation

The comparison between both bioreactors was made for different combinations of
medium and dodecane fluxes. Two different toxic product concentrations were used,
i.e. 1 mol m
-3
and 100 mol m
-3
, and an initial substrate concentration of 1 kmol m
-3
.
Introducing an organic solvent in a 3-phase fluidized bed, will in most cases
result in a lower gel-bead hold-up. This means loss of biocatalytic activity per unit
volume of bioreactor. This loss of activity has to be compensated for by lowering the
product concentration in the gel beads, resulting in a higher substrate conversion
rate. This is achieved by extracting the product into the organic solvent. The
extraction rate is influenced by the distribution coefficient. As can be deduced from
the equation in Table II, a higher distribution coefficient gives a higher transfer rate.
So, at a given maximum substrate conversion rate, the time to reach a certain
degree of conversion can be manipulated by the distribution coefficient (m): a higher
m results in a shorter time, see also Figure 10. Thus, to conclude whether a 3-phase
fluidized bed performs better than a 2-phase fluidized bed, at a given maximum
substrate conversion rate, a distribution coefficient is determined, that yields the
same time to reach 99% of the total amount of moles converted in a 2-phase fluidized
bed. A higher distribution coefficient results in a better performance, whereas a
lower distribution coefficient gives a worse performance.
We chose to evaluate the performance of the bioreactor with two parameters,
i.e. the maximum substrate conversion rate (X⋅v
max
) and the distribution coefficient
(m), and to keep the other parameters constant. The maximum substrate conversion
rate can be influenced by using more or less biocatalyst, and the distribution
coefficient might be influenced by adding specific compounds to the organic solvent
for enlarging the distribution.
The distribution coefficient at a given maximum substrate conversion rate is
thus calculated for which a 3-phase fluidized bed performs the same as 2-phase
fluidized bed. This was done for a number of organic solvent fluxes and medium
fluxes, and for different toxic product concentrations. The results are shown in
Figures 13a-c.
Design of a 3-phase fluidized bed
151

F
i
g
u
r
e

1
3
.

D
i
s
t
r
i
b
u
t
i
o
n

c
o
e
f
f
i
c
i
e
n
t

a
s

a

f
u
n
c
t
i
o
n

o
f

t
h
e

m
a
x
i
m
u
m

s
u
b
s
t
r
a
t
e

c
o
n
v
e
r
s
i
o
n

r
a
t
e

f
o
r

w
h
i
c
h

a

3
-
p
h
a
s
e

f
l
u
i
d
i
z
e
d

b
e
d

p
e
r
f
o
r
m
s

e
q
u
a
l
l
y

w
e
l
l

a
s

a

l
i
q
u
i
d

f
l
u
i
d
i
s
e
d

b
e
d
.

O
p
e
n

s
y
m
b
o
l
s

C
p
t
o
x

=

1

m
o
l

m
-
3
,

c
l
o
s
e
d

s
y
m
b
o
l
s

C
p
t
o
x

=

1
0
0

m
o
l

m
-
3

A


U
c

=

0
.
7
5

c
m

s
-
1





U
d

=

0
.
4
0

c
m

s
-
1
;



U
d

=

0
.
1
4

c
m

s
-
1


B

U
c

=

1
.
2
9

c
m

s
-
1





U
d

=

0
.
1
4

c
m

s
-
1
,



U
d

=

0
.
5
5

c
m

s
-
1



U
d

=

0
.
9
1

c
m

s
-
1


C

U
c

=

2

c
m

s
-
1




U
d

=
0
.
1
4

c
m
/
s
,



U
d

=

0
.
2
8

c
m

s
-
1





U
d

=

0
.
5
4

c
m

s
-
1


t
b
a
t
c
h

(
9
9
%
)


(
m
i
n
)


X

v
m
a
x


m
o
l

m
-
3

s
-
1

C
p
t
o
x

=


1

m
o
l

m
-
3

C
p
t
o
x

=


1
0
0

m
o
l

m
-
3


C
p
t
o
x

=

1
0
0

m
o
l

m
-
3




C
p
t
o
x

=


1

m
o
l

m
-
3

C
p
t
o
x

=

1
0
0

m
o
l

m
-
3


0
.
0
0
1

1
6
6
.
3

1
5
5
4
5


2
0
9
5
0




3
5
5
.
8

3
2
5
0
4


0
.
0
1

2
2
.
1
3

1
5
6
0


2
1
0
8




5
5
.
3
1

3
2
7
8


0
.
0
5

1
5
.
0
4

3
1
9
.
4


4
3
5
.
5




2
3
.
8
1

6
9
9


0
.
1

1
5
.
4
7

1
6
7
.
2


2
2
7
.
9




1
9
.
6
3

3
6
1
.
4


1

1
4
.
9
5

2
1
.
4


3
2
.
3




1
7
.
0
2

5
5
.
9
4



0
.
9
1
.
1
1
.
3
1
.
5
1
.
7
1
.
9
2
.
1 0
.
0
0
1
0
.
0
1
0
.
1
1
X

v
m
a
x








(

m
o
l

m
-
3

s
-
1

)
m A
0
.
9
1
.
1
1
.
3
1
.
5
1
.
7
1
.
9
2
.
1 0
.
0
0
1
0
.
0
1
0
.
1
1
X

v
m
a
x











(

m
o
l

m
-
3

s
-
1

)
m B
0
.
9
1
.
1
1
.
3
1
.
5
1
.
7
1
.
9
2
.
1 0
.
0
0
1
0
.
0
1
0
.
1
1
X

v
m
a
x




(

m
o
l

m
-
3

s
-
1
)
m C
Chapter 6
152
A 3-phase fluidized bed performs better for combinations of the maximum substrate
conversion rate (X⋅v
max
) and distribution coefficient in the area above the lines in
these figures.

The distribution coefficient at each X⋅v
max
in favor of the 3-phase fluidized bed is
relatively low, see Figures 13a-c. Especially, when we consider the examples in
Table IV, in which distribution coefficients for different solutes for different liquid-
liquid 2-phase systems are given. Based on this table one might suggest that at any
pre-set medium flux, a 3-phase fluidized bed performs better than a conventional 2-
phase fluidized bed.

Table IV. Distribution coefficients for solute for different liquid-liquid 2-phase systems
Solute 2-phase system distribution
coefficient
reference

oxygen FC40/water 12 Sonsbeek et al. ,1992a
octene/water 11.4 Meer, 1993

propenoxide toluene/water 3.5 Brink and Tramper, 1985
butylacetate/water 4.6 Brink and Tramper, 1985

butanal toluene/water 6.9 Kawakami, et al., 1992
hexadecane/water 1.52 Kawakami, et al., 1992

tetraline FC40/water 120 Vermuë, et al., 1994
dodecene/water > 5000 Vermuë, et al., 1994

Steroids
Androstenolone-
actate
Hexane/water 1500 Boeren, et al., 1992
4-Androstene-
3,17–dione
Hexane/water 8.5 Boeren, et al., 1992
dehydro-epi
androsterone
Hexane/water 24 Boeren, et al., 1992

phenylalanine
propyl ester
isobutyl methyl
keton / water
123 Flaschel, et al., 1992


Looking in more detail to Figure 13a to c, the results in Figure 13a and Figure 13b – a
lower water flux applied - show that a lower dodecane flux requires a higher
distribution coefficient, whereas the results in Figure 13c – the highest water flux
Design of a 3-phase fluidized bed
153
applied – show that a higher dodecane flux requires also a higher distribution
coefficient. Although the latter observation might seem unexpected, both
observations can be easily explained. In most cases the dodecane flux decreases the
gel-bead hold-up and, as explained above, this decrease in gel-bead hold-up has to
be compensated for with a high enough transfer rate to the dodecane phase. The
transfer rate itself is influenced by the dodecane flux, as the dodecane flux
determines largely the dodecane hold-up and the mass transfer coefficient, see also
the equation in Table II. A higher dodecane flux results in a higher transfer rate. The
transfer rate is also influenced by the distribution coefficient, a higher distribution
coefficient gives also a higher transfer rate.
At the lower water fluxes, Figure 13a and Figure 13b, the decrease in gel-bead
hold-up is not that large, less than 15% for all dodecane fluxes applied, and there is
hardly any difference in gel-bead hold-up between the highest and lowest dodecane
flux. Thus, the transfer rate in the 3-phase fluidized beds, necessary to perform
equally well as the 2-phase fluidized bed, are more or less the same. So, at these
water fluxes, a higher distribution coefficient is needed for the lower dodecane flux,
to establish a high enough transfer rate.
For the highest water flux applied, Figure 13c, there is a significant decrease in
gel-bead hold-up with an increasing dodecane flux (compared to the 2-phase
fluidized-bed analogue, 39% for the highest dodecane flux and 7% for the lowest).
Consequently, the transfer rate must be higher for the higher dodecane fluxes. When
a lower distribution coefficient is used for the higher dodecane fluxes, as above for
the lower water fluxes, the resulting transfer rate is not enough for establishing the
required transfer rate, although dodecane hold-up is higher for a higher dodecane
flux. Hence, the distribution coefficient has to be higher for the higher dodecane
fluxes.
The open symbols in Figure 13a and c show the distribution coefficient as a
function of the X⋅v
max
for which the 3-phase fluidized bed performs equally well for a
low toxic product concentration (1 mol m
-3
, i.e. 0.1 % of the initial substrate
concentration). At the lower water flux, Figure 13a, the difference between both
dodecane fluxes is almost insignificant. At the higher water flux, Figure 13b, this
difference is a little more pronounced. A general remark about the difference
between a high and low toxic product concentration can hardly be made: at X⋅v
max

the distribution necessarily for giving an equal performance is sometimes lower
other times higher.

Chapter 6
154
Continuous Operation

Executing a biotransformation continuously in a 2-phase fluidized bed without a
recycle is highly inefficient, as the medium flux is high compared to the maximum
substrate conversion rate. Consequently the residence time is short and the
conversion rate is low. Only for fast conversions or high reactors, a high degree of
substrate conversion is possible, and hence a fluidized bed might be attractive. As an
example, a substrate conversion of 1 mol s
-1
, a medium velocity of 0.02 m s
-1
, and a
reactor height of 1 m results in conversion of 50 mol, assuming zero-order kinetics.
This is 50% of the attainable conversion at a toxic product concentration of
100 mol m
-3
.
Otherwise one should use a set-up with a partial recycle and fresh substrate
supply. The incoming substrate concentration is kept constant. The product
concentration in the medium phase will accumulate until steady state is reached. As
the product concentration increases with an increasing recycle, substrate will be less
converted. Application of an organic solvent, i.e. using a 3-phase fluidized bed, will
result in a higher conversion, as we calculated for the next case (the organic solvent
was not recycled): medium flux 0.75×10
-2
m s
-1
, dodecane flux 0.40×10
-2
m s
-1
, hence
ε
s
=0.48, ε
o
=0.045; X⋅v
max
= 0.01 mol s
-1
, C
p
tox
= 100 mol m
-3
, and m=100. The
dimensionless concentrations for a 2-phase fluidized bed and 3-phase fluidized bed
are calculated and given in Table V for different recycles.

Table V. Dimensionless substrate and product concentrations in a continuously
operated fluidized bed
2-phase fluidized bed 3-phase fluidized bed
Recycle substrate product substrate product product organic solvent
0.1 0.9993 6.50e-4 0.9994 6.07e-4 0.72e-4
0.9 0.9994 0.0061 0.9994 0.0019 2.98e-4
0.99 0.9996 0.0396 0.9994 0.005 8.5e-4


Obviously, little substrate is converted, and hence a small amount of product is
made. The introduction of the organic solvent results in a smaller product
concentration in the medium phase. In optimizing a continuous operation, there is lot
of freedom, e.g. the initial substrate concentration, recycle of medium and/or organic
solvent. So, whether a continuous operation is feasible depends on a total cost
calculation of the plant.
Design of a 3-phase fluidized bed
155
Conclusion

Different aspects of liquid-liquid-solid 3-phase systems have been discussed. The
benefits of such a system for typical biotransformations have been discussed in
literature. However, the ideal reactor configuration can still be argued about.
Different reactor configurations are discussed here, including a simple 3-phase
fluidized bed, but also more fancy options, like the different possibilities with a loop
reactor. It can be concluded that building and operating a 3-phase system is nothing
more than a minor extension of conventional bioreactors.
For a liquid-liquid-solid 3-phase fluidized-bed bioreactor conditions have
been established for which this reactor performs better than a conventional 2-phase
fluidized bed. Keeping all parameters constant, changing the operation variables,
medium flux and organic solvent flux, the distribution coefficient is determined, at
given maximum substrate conversion rate, for which the 3-phase fluidized bed
performs equally well. It appears that already a low distribution coefficient (larger
than 1 but less than 2) suffices for a better performance.
So, it can be concluded that a 3-phase fluidized bed, or less simple 3-phase
bioreactors, are good options for operating specific biotransformations, provided in
situ extraction is needed for such a specific biotransformation.


Chapter 6
156
Nomenclature
A
r
: area reactor m
2

C
i
j
: concentration component i in phase j mol m
-3

C
s
w,0
: substrate concentration in medium phase at t=0 mol m
-3

D
i
: diffusion coefficient component i m
2
s
-1
d
bead
: gel bead diameter m
d
drop
: droplet diameter m
d
noz
: nozzle diameter m
Fo : dimensionless time -
g : gravity constant m
2
s
-1
H
bed
: height fluidized bed m
H
t
: height of one medium tank m
H
t.i
: height of disperse phase tank m
K : overall mass transfer coefficient water/droplet m
3
(m
2
int
s)
-1
K
m
: michaelis-menten constant mol m
-3

k
bead,i
: mass transfer coefficient component i water/gel bead m
3
(m
2
int
s)
-1
k
drop
o
: mass transfer coefficient interface droplet/bulk droplet m
3
(m
2
int
s)
-1
k
drop
w
: mass transfer coefficient bulk water/interface droplet m
3
(m
2
int
s)
-1
N_dis : number of tanks in disperse phase per one medium tank -
m : distribution coefficient organic/water -
R
p
: gel bead radius m
r : space-coordinate in gel bead m
t : time s
U
c
: medium phase flux m
3
(m
2
s)
-1
U
d
: organic solvent flux m
3
(m
2
s)
-1
V
r
: reactor volume m
3

v

: terminal rise velocity droplet m s
-1
v
max
: maximum substrate conversion rate mol (g
cat
s)
-1
v
wp
: slip velocity water particle m s
-1
v
wo
: slip velocity water droplet m s
-1
X : concentration biocatalyst g
cat
m
-3

X
i
j
: dimensionless concentration component i in phase j -
z : dimensionless space coordinate in gel bead -


Design of a 3-phase fluidized bed
157
subscripts
p : product
s : substrate

superscripts
bulk : bulk phase
int : at the interface between two phases
n : tank number in medium phase
o : organic solvent phase
p : gel bead phase
tox : toxic
w : medium phase

greek
η
w
: viscosity medium phase Nm s
-1
∆ρ : densitydifference kg m
--3

ε
o
: organic solvent phase hold-up -
ε
s
: gel bead phase hold-up -
ε
w
: medium phase hold-up -
ρ : density kg m
-3

σ : interfacial tension N m
-1
ϕ
pw
p
: mass transfer flux of product gel bead/water mol s
-1
ϕ
wo
p
: mass transfer flux of product water/organic solvent mol s
-1
ϕ
wp
s
: mass transfer flux of substrate water/gel bead mol s
-1

Chapter 6
158
Appendix A. Derivation of mass balances for a 3-phase fluidized-bed
bioreactor

In a 3-phase fluidized bed three different phases are present. Mass balances are derived for
the individual phases. As conversion we took one mole of substrate giving one mole of
product. So, two components are present in the solid phase and the medium phase. Only
product is present in the organic-solvent phase.
The total reactor is divided in a number of vertically stacked ideally stirred tank
reactors, a tanks-in-series model (number of tanks is 3). We took a plug flow model for the
organic-solvent phase. A tank-in-series model was applied with a large enough number of
tanks to mimic plug flow. For simplicity reason we assumed that the gel beads did not
circulate inside the reactor. Systematically the bioreactor looks as shown in Figure 8.
Mass transfer takes place from the medium phase to the gel beads for substrate.
Product is transferred from the gel beads to the medium, and from medium to the organic
solvent. Mass balances for one tank, including the boundary and initial conditions, read as
follows:

The gel bead

Substrate
) r (
r
C
r
2
r
C
D
t
C
s
n p,
s
2
n p,
s
2
s
n p,
s
− −


+


=




t=0, ∀r, C
s
p,n
=0
r=0, centre of the gel bead 0
r
C
0 r
n p,
s
=


=

r=R
p
, edge of the gel bead ( )
R r
n p,
s
s
n int, w,
s
n w,
s s bead,
r
C
D C C k
=


= −

Product
-r r r
r
C
r
2
r
C
D
t
C
s p p
n p,
p
2
n p,
p
2
s
n p,
p
= ∧ +


+


=



t=0, ∀r, C
p
p,n
=0
r=0, centre of the gel bead 0
r
C
0 r
n p,
s
=


=

Design of a 3-phase fluidized bed
159
r=R
p
, edge of the gel bead ( )
p R r
n p,
p
p
n w,
p
n int, w,
p p bead,
r
C
D C C k
=


− = −

The medium phase

Substrate
( )
s
wp
n w,
s
1 n w,
s r c
n w,
s
r w
C C A U
dt
dC
V ϕ − − = ε


t=0, C
s
w,n
= 1, batch operation
C
s
w,n
= 0, continuous operation

Product
( )
s
wp
n w,
s
1 n w,
s r c
n w,
s
r w
C C A U
dt
dC
V ϕ − − = ε


t=0, C
p
w,n
= 0, batch operation and continuous operation




The organic solvent phase

Product
( )
p
wo
n i, o,
p
n 1, - i o,
p r d
n i, o,
p
r
d
C C A U
dt
dC
No
V
ϕ − − = ε
t=0, C
p
o,i,n
= 0


The different transfer fluxes are represented by:
( )
r s
bead
n int, w,
s
n w,
s s bead,
s
wp
V
d
6
C C k ε − = ϕ

( ) Vr
d
6
C C k
s
bead
n w,
p
n int, w,
p p bead,
p
pw
ε − = φ
No
V
d
6
m
C
C
mk
1
k
1
r
d
drop
n i, o,
p
n w,
p
1
d
p drop,
w
p drop,
i ,
p
wd
ε

















+ = φ



The mass transfer coefficients k
bead,s
, k
bead,p
, and k
drop,p
w
are calculated with the equation of
Chapter 6
160
Ranz and Marshall (1952):

33 . 0
i w
w
5 . 0
w
bead wp w
i
bead i bead,
D
d v
0.57 2
D
d k








ρ
η








η
ρ
+ =

33 . 0
p w
w
5 . 0
w
drop wo w
p
drop p bead,
D
d v
0.57 2
D
d k








ρ
η








η
ρ
+ =

with v
wp
the slip-velocity between water and gel bead, and v
wo
the slip-velocity between
water and organic phase. The droplet diameter is calculated with the equation of Kumar and
Hartland
315 . 0
2
noz d 73 . 0
noz
33 . 0
noz
noz
gd
We 0393 . 0 Eo 55 . 0
d
d









σ
ρ
+
=
with Eo
noz
=
σ
ρ
2
noz
d g∆
,and We
noz
=
σ
ρ
noz
d v
2



The rise velocity of a single droplet is calculated with Vignes' equation:















η
ρ








ρ
ρ ∆
=

6
Eo
1
g
4.2
d
v
3 / 1 3 / 2

with Eo =
σ
ρ
2
d g∆


The mass transfer coefficient k
drop,p
d
is calculated with the equation of Newman (1931).
















π −
π
− =


=1 n drop
drop p,
2 2
2 2
drop
d
p drop,
d
t
D 4n exp
n
1

6
ln
6t
d
k

t represents the contact time and is equal to: t = H
t
ε
o
/U
d



The mass balances are made dimensionless by dividing with the initial substrate
concentration, the bead diameter, and with D
s
/R
p
2
. The resulting dimensionless parameters
are: dimensionless length inside the bead z=r/R
p
, dimensionless concentration C/C
s
w,0
, and a
dimensionless time Fo=D
s
t/R
p
2
.
Design of a 3-phase fluidized bed
161
The resulting mass balances with the implementation of the transfer flux relations are:

( )
0 , w
s s
2
p
s
n p,
s
2
n p,
s
2 n p,
s
C D
r
r -
z
X
z
2
z
X
Fo
X



+


=




0 , w
s s
2
p
s
n p,
p
s
p
2
n p,
p
2
s
p
n p,
p
C D
r
) r - (
z
X
z
2
D
D
z
X
D
D
Fo
X
+


+


=




( ) ( )
n int, w,
s
n w,
s
s
2
p
s bead,
w
s
bead
n w,
s
1 n w,
s
s
2
p
t w
w
n w,
s
X X
D
r
k
d
6
X X
D
r
H
U
Fo
X

ε
ε
− −
ε
=





( ) ( )









ε
ε

+ −
ε
ε
+ −
ε
=



m
X
X
D
r
d
6
K
X X
D
r
k
d
6
X X
D
r
H
U
Fo
X
i
n w,
p
n w,
p
s
2
p
w
o
o
n w,
p
n int, w,
p
s
2
p
p bead,
w
s
bead
n w,
p
1 n w,
p
s
2
p
t w
w
n w,
p


( )








− + −
ε
=



m
X
X
D
r
d
6
K X X
D
r
H
U
Fo
X
i
n w,
p
n w,
p
s
2
p
o
i o,
p
1 i o,
p
s
2
p
i t, d
d
i o,
p



The space co-ordinate in the gel bead mass balance is discretized, making the partial
differential equation a set of ordinary differential equation (we took 31 grid points). To
mimic the plug flow of the dispersed phase, we took 15 tanks per medium tank. Per medium
tank 2 × 30 + 2 + 15 = 77 differential equations were solved simultaneously with a solver for
stiff differential equations.


162

163
7
General discussion
Chapter 7
164
The results presented in this thesis make it possible to design a liquid-liquid-solid 3-
phase fluidized-bed bioreactor. This fluidized bed consists of 1) a solid phase, i.e. gel
beads with an immobilized biocatalyst, 2) a continuous aqueous phase that was used
for fluidizing the gel beads, and for supply of substrates essential for the
biotransformation, and 3) an organic solvent, which rose through the fluidized bed of
gel beads as organic droplets. This organic phase was also used as a sink for the
biotransformation product. The design of this bioreactor for a specific
biotransformation depends on knowledge of the biotransformation itself and various
physical aspects of the bioreactor. Figure 1 summarizes the major topics to be considered for
a conceptual process design.

Figure 1. Major topics for designing a 3-phase fluidized-bed bioreactor.

In this study, we assumed the biotransformation data to be known. For the design of
a 3-phase fluidized-bed bioreactor, the understanding of its hydrodynamic behavior
is essential.


Hydrodynamics

For a 3-phase fluidized-bed bioreactor, both the aqueous phase as well as the organic
solvent is recirculated with an external pump. This means that the hydrodynamics
can be controlled with both the aqueous-phase velocity and the organic-phase
velocity. Important hydrodynamic topics are:
1) Bed stability. Stability of the bed refers to stable operation, which means that
the gel beads remain in the fluidized bed and are not washed out.
Biotransformation
⋅ kinetics

substrate consumption

product formation


⋅ biocatalyst stability







Bioreactor Design Bioreactor Design
Hydrodynamics
⋅ fluidized-bed stability
⋅ phase hold-ups
⋅ phase mixing
⋅ mass exchange

medium phase / gel beads

medium phase / organic droplets



General discussion
165
2) Phase hold-up. A phase hold-up is the fractional volume of a phase in the
fluidized bed. The droplet hold-up thus determines largely the amount of
extractive liquid present, and therefore the extraction potential. The gel-bead
hold-up determines the height of the fluidized bed at a given total volume of
gel beads. Hold-up determines also the specific area for mass exchange.
3) Phase mixing. Whether a phase shows plug-flow behavior or ideally mixed
behavior can largely influence overall reactor performance. Besides, mixing
characteristics also influence mass exchange.

In a 3-phase fluidized bed two distinct 2-phase systems are integrated: a liquid-solid
fluidized bed and a spray column (a liquid-liquid extraction column). Hydrodynamic
aspects of these constituent 2-phase systems were separately studied; Chapter 2 deals
with the liquid-fluidized bed and Chapter 3 deals with the spray column. A 3-phase
fluidized bed approaches a 2-phase system if either the amount of organic solvent
present in the 3-phase system approaches zero (a liquid-solid fluidized bed), or the
amount of gel beads present in the 3-phase system reduces to zero (a spray column).
Being able to predict the hydrodynamic characteristics of the individual 2-phase
systems, it was assumed that the two corresponding models could be combined to
describe and predict hydrodynamic characteristics of the 3-phase fluidized bed. In
Chapter 4, it is shown that this idea is only partly valid for predicting the different
phase hold-ups. With respect to mixing of the continuous phase, it is shown in
Chapter 5 that models predicting the axial dispersion coefficient in a spray column
can be reliably used for predicting the axial dispersion coefficient in a liquid-liquid-
solid 3-phase fluidized bed.

In the following paragraphs, the hydrodynamics of the individual 2-phase systems
and the integrated 3-phase system will be discussed in some detail. For the 2-phase
systems we will focus on hold-up. Operational stability of both 2-phase systems was
good, provided the water velocity did not exceed the settling velocity of a single gel
bead. Mixing characteristics of the separate 2-phase system were not determined. For
the integrated 3-phase system, we will discuss stability, hold-up and mixing below.

2-phase systems
In Chapter 2, experimental results on gel-bead hold-up in a 2-phase liquid-solid (gel
bead) fluidized bed are presented. Increasing the aqueous velocity was found to
result in a decreasing gel-bead hold-up. It was found that this hold-up could not be
Chapter 7
166
predicted with the well-established slip-velocity model of Richardson and Zaki
(1954), if standard parameter values were used. However, the experimental hold-up
data could be well described by the Richardson and Zaki equation if standard
parameter values were adopted to fit the experimental data. Further analysis showed
that the drag force between the continuous flowing phase (the medium) and gel
beads, either present in a group or as single particle, was much lower than for more
conventional solids such as glass pearls. Hence, the slip-velocity was much higher.
Gel beads consist over 95% of water, and obviously their surface properties are
different from those of conventional solids. Apparently, these differences resulted in
a lower drag force. The drag force is largely determined by a drag coefficient. In
Chapter 2, a more mechanistic model, based on the model of Foscolo et al. (1983), is
presented for the prediction of the drag coefficient of a single gel bead in a packed or
fluidized bed. This drag coefficient can successfully be used to predict pressure drop
and gel-bead hold-up in a liquid-fluidized bed.

Chapter 3 shows the influence of sparger lay-out, mode of operation (counter-current
or co-current operation) and both liquid velocities on the droplet hold-up. The
droplet hold-up in a liquid-liquid extraction spray column can be well predicted with
a Richardson and Zaki type of slip-velocity equation, featuring as model parameters
an exponent n, and a rise velocity of a single droplet. An established literature model
for rise velocities is used; for the exponent n a new model was derived. With this new
model for n, the experimental data can be well described. In Chapter 3, it is also
shown in which way the droplet hold-up can be manipulated by means of the
sparger lay-out (number of nozzles and nozzle diameter).

3-phase fluidized beds: stability
A liquid-liquid-solid (gel bead) fluidized bed is created by introduction of an organic
disperse solvent into a 2-phase liquid-solid fluidized bed. Single droplets rising
through this 3-phase fluidized bed of gel beads will hardly disturb the fluidized bed,
and wash-out of gel beads does not occur.
Through collisions, rising droplets will transfer momentum to the gel beads. If the
amount of rising droplets is substantial, gel beads will be washed out from the
bioreactor by these rising droplets. Whether this phenomenon took place depended
on the density (i.e. inertia) of the gel beads. In this study, gel beads were used with
densities of 1006 – 1065 kg m
-3
. Wash out was indeed observed for gel beads with a
density of 1010 kg m
-3
and a diameter of 2 mm, at a superficial solvent velocity of
General discussion
167
0.01

cm s
-1
. Wash out of the heavier gel beads (1065 kg m
-3
) was not observed at any
flux, and stable operation of the 3-phase fluidized bed at all practical solvent
velocities was possible.
Practical water velocities were between 0.5 and 2 cm s
-1
, while practical
organic solvent velocities were between 0.05 and 0.9 cm s
-1
. For gel beads of 1065 kg
m
-3
, a flow diagram is presented in Chapter 4. By means of the superficial velocities
of both liquid phases, this flow diagram demarcates five areas with characteristic
flow patterns: two homogeneous patterns, two heterogeneous patterns (in these four
areas, gel beads remained in the fluidized bed, and a freeboard of solids was hardly
formed), and an unstable area in which the organic droplets coalesced and gel beads
were washed out. In a 2-phase fluidized bed, the solids were homogenously
distributed over the bed, and they moved very slowly through it. At very low
organic solvent velocities (much lower than 0.05 cm s
-1
), this flow pattern of the
solids was not disturbed. Also at relatively high water velocities, the solids were
homogenously distributed. Their motion through the bed, however, was much more
rapid and more chaotic. A structured heterogeneous flow pattern of the solids was
observed at intermediate organic solvent velocities (< 0.35 cm s
-1
) and water
velocities less than 0.8 cm s
-1
. Axial concentration waves of gel beads along the
column were present throughout the fluidized bed. For this flow pattern, a radial
concentration distribution of the gel beads was not observed. An irregular chaotic
heterogeneous flow pattern of the gel beads was present at high organic solvent
velocities and intermediate water velocities. There were significant axial and radial
distributions of gel beads and these distributions rapidly and randomly changed
with time without a repetitive pattern. Coalescence of the organic solvent droplets
occurred at relatively high organic solvent velocities (> 0.35 cm s
-1
) and relatively low
water velocities (< 0.6 cm s
-1
).

So far, it was assumed that the organic solvent and medium flowed co-currently. For
gel beads with a very low density (1010 kg m
-3
), however, co-current flow does not
yield a stable bed, except at extremely low velocities of the organic solvent. At
practically all organic solvent velocities, the organic droplets push the gel beads up,
and before they can fall back again, they are pushed once more. This way, the gel
beads are transported out of the fluidized bed. For a fluidized bed of low density gel
beads (1010 kg m
-3
), the gel beads could be retained in the fluidized bed without
wash-out only if the water flux was zero. Thus, stable operation of this 3-phase
system with low density gel beads is possible. Stable operation of the analogue of
Chapter 7
168
this system, a 3-phase bubble column, filled with gel beads, is described in literature
(Santos et al. 1997). Obviously, the net upward velocity of the aqueous phase in a co-
currently operated 3-phase liquid-liquid-solid fluidized bed is, among other things,
responsible for wash out of low density gel beads. Apparently, in a 3-phase system
without external recirculation of the aqueous phase, internal circulation of the
aqueous phase (upward and downward flow) prevents the wash-out of the gel
beads.
Although a stable 3-phase fluidized bed was possible if only the organic
solvent was pumped to the fluidized bed, a gel-bead concentration gradient along
the column height was observed. This concentration gradient could be strongly
reduced by applying a downward countercurrent aqueous flow. A stable operation
of this 3-phase system was possible. The gel beads were kept in suspension, a packed
bed was not formed, and coalescence of the organic droplets was not observed. It
was observed that an increase in both flow velocities did not affect the height of the
fluidized bed. This opens the way to a really interesting 3-phase fluidized bed; if both
velocities are increased in the same way, the volume of the bioreactor and the
bioreactor activity remain the same, while the extractive power increases.

Hold-ups in a co-current 3-phase fluidized bed
For high density gel beads (1065 kg m
-3
), the influence of the organic-solvent and
medium velocities on the gel bead and droplet hold-ups is described in Chapter 4. As
the amount of gel beads in these experiments was fixed, the bed height was
depending on the two phase velocities. At water velocities below 0.75 cm s
-1
, the bed
height was smaller than in a liquid-solid fluidized bed at the same water velocity
(bed contraction occurred); at water velocities above 0.75 cm s
-1
it was larger (bed
expansion). It was also shown that at any phase velocity, the droplet hold-up was
always larger than in a 2-phase droplet column at the same phase velocity. So, the
individual models for both 2-phase systems are not additive, and a new model was
needed.
The new model for gel-bead and organic-droplet hold-up is based on the drag
forces for a gel bead and a droplet present in the 3-phase mixture (Chapter 4). The
gel- bead hold-up is well described by the model. The prediction of droplet hold-up,
however, is only fair for hold-ups below 0.06; at higher hold-ups, the droplet hold-up
is strongly underestimated. As an alternative, an empirical model was derived; it
gives a good description for the whole range of measured hold-ups (0.005 → 0.09).
General discussion
169
To calculate the drag forces, a drag coefficient is needed. Models for drag
coefficients for the individual 2-phase systems were extended to model the drag
coefficients of a gel bead and a droplet in the 3-phase fluidized bed. When the
parameters of the individual 2-phase models are used, the gel bead hold-up is very
well described, but the droplet hold-up is systematically underestimated.

Mixing
Mixing of the continuous phase in a 3-phase fluidized bed was studied with high-
density gel beads (Chapter 5). It is much more turbulent than mixing in a 2-phase
fluidized bed, which was easily verified on sight. Indeed, the axial dispersion
coefficient is much higher for a 3-phase fluidized bed and can be very well predicted
with the isotropic-turbulence theory of Baird and Rice (1975). It is concluded that
mixing of the continuous phase in a 3-phase fluidized bed is comparable with mixing
of the continuous phase of either a bubble column or the main tube of a liquid-
impelled loop reactor (Van Sonsbeek et al., 1992).

Mixing characteristics of the gel beads, i.e. the circulation velocity of individual gel
beads through the fluidized bed, were not determined. Also the mixing of the
organic solvent droplets was not determined, but a preliminary indication may be
obtained from predictions of solvent hold-up. To predict droplet hold-up, a single
droplet diameter was assumed. The successful prediction of solvent hold-up was in
support of the assumption of a single droplet diameter. Since droplets with the same
diameter will have an equal rise velocity through the 3-phase fluidized bed, plug
flow of the droplet phase may be concluded.

As stated above, mixing characteristics of the gel beads in a 3-phase fluidized bed
were not studied. For the design of the 3-phase fluidized bed, we assumed the gel
beads not to move. Obviously, this is a simplification. Future work should elucidate
the movement of gel beads throughout the bed and its impact. It may be monitored
with radio-active labeling and appropriate sensors. The results can be transformed to
generic flow patterns of the gel beads in the fluidized bed.

Mass transfer
Besides hold-up and mixing, mass exchange of substrates and products between the
different phases of the 3-phase fluidized bed is vital for the design of the bioreactor.
Mass transfer in the fluidized bed takes place at two different levels: 1) transfer in the
Chapter 7
170
continuous phase through a droplet or gel bead boundary layer, and 2) transfer
within a droplet or gel bead. The same transfer path holds for gel beads. Mass
transfer within particles or droplets has already been studied extensively in literature
and may be adequately predicted from established equations. Mass transfer through
the outer boundary layer of gel beads or droplets in a 3-phase fluidized bed, on the
contrary, has not been studied yet; also in this study, such mass transfer experiments
have not been done. In the next paragraph we will focus on these mass transfer
phenomena.
In the design of the 3-phase fluidized bed (Chapter 6), we used the well-
known equation of Ranz and Marshall (1950) for calculation of mass transfer
coefficients from the bulk liquid to the surface of either a droplet or a gel bead. This
equation uses the slip velocity between the different phases, which can be
determined with the hold-up model given in Chapter 4. It is questionable, however,
whether the empirical constants in this mass transfer correlation may be applied also
for mass transfer in the liquid-liquid-solid (gel bead) fluidized bed bioreactor.
Other models use the energy dissipation rate in the fluidized bed for
determination of mass transfer coefficients (Kikuchi et al., 1994). This method is also
widely used for the determination of mass transfer in stirred tanks and bubble
columns (Van ‘t Riet and Tramper, 1991). The energy input determines the mixing of
the aqueous phase, and mixing determines the extent of the boundary layer. The size
of the boundary layer, finally, determines the external mass transfer coefficient. So, it
is expected that the energy-input determines the external mass transfer coefficient. In
Chapter 5 it is shown that mixing of the 3-phase fluidized bed can be adequately
described with the well-established model of Baird and Rice (1975) using the energy
dissipation rate. So, external mass transfer coefficients may be adequately
determined using established correlations based on the energy dissipation rate.
Further research should substantiate this idea, as for gel beads mass transfer
in 3-phase systems (gas-liquid-solid or liquid-liquid-solid) has not been elucidated.
Experiments should yield the external mass transfer coefficients for the transfer from
the continuous liquid phase to either a droplet or a gel bead surface. In a 3-phase
fluidized bed there is simultaneous transfer from the aqueous phase to both the
organic droplet and the gel bead; one would like to determine the external transfer
coefficients independently. To determine the mass transfer coefficient from bulk
liquid to gel bead surface one could use a tracer component that is exchanged
between the bulk liquid and the gel beads, but, at the same time, does not dissolve in
the organic solvent.
General discussion
171
Determination of the external transfer coefficient from the continuous aqueous
phase to the droplet surface can use a similar set-up (Van Sonsbeek et al., 1991).
However, a tracer component that is not transferred to the gel beads might be hard to
find, as the gel beads consist for over 95% of water. An alternative would be to use a
component that is exchanged between the continuous aqueous phase and both the
gel beads and the organic droplets. Transfer experiments with such a component
would thus involve two simultaneous transfer coefficients. Nevertheless, the
coefficient for transfer into the organic phase may be calculated from experimental
data if the value for the coefficient of transfer into the gel beads would be available
from preceding experiments as described above.


Summary and Future prospects

The work in this thesis aimed at design rules for a 3-phase liquid-liquid-solid (gel
bead) 3-phase fluidized-bed bioreactor. A full design of such 3-phase fluidized bed
bioreactors is only partially possible with the results presented in this thesis.
Operational stability of the fluidized bed, hold-up of the different phases, and mixing
of the continuous phase as a function of operational variables (superficial velocities
of the organic solvent and medium) are well understood, and reliable models have
been derived. For mass transfer and mixing of the organic solvent phase, educated
guesses can be made, based on our experience with the fluidized bed. Mixing
behavior of the gel beads is still unknown. In our design, given in Chapter 6, we
assumed that the gel beads did not move. The other extreme would be an ideally
mixed bed of gel beads. The actual mixing of the bed will be in-between both
extremes. Future work should address the mixing characteristics of the gel beads,
and the correlations for mass transfer coefficients.

To elucidate the hydrodynamic behavior of the 3-phase fluidized bed, a specific type
of gel bead was used. A generalization of the behavior can thus not be given.
However, with respect to stability of the fluidized be we concluded that to prevent
wash-out of the gel beads, their terminal settling velocity should exceed 5.0 cm s
-1
.
The model for the hold-up in the 3-phase fluidized bed contains empirical constants,
which, in principle, have to be measured for each type of gel beads. Experimental
methods for determination of these constants are given. It is also shown that if an
Chapter 7
172
energy dissipation rate is available, mixing of the continuous phase may be
straightforwardly determined.

In Chapter 6, it is shown that the performance of the 3-phase fluidized-bed bioreactor
is influenced by the distribution coefficient of the product over the organic solvent
and medium phase. It appeared that a distribution coefficient larger than 2 suffices
for a better performance of this 3-phase fluidized bed. In order to demonstrate that
the newly developed bioreactor does also perform better for practical systems, a
bioconversion hindered by strong product inhibition should be performed with this
bioreactor.

In future, one could consider using a 3-phase fluidized bed bioreactor for performing
substrate or product inhibited biotransformations, as this thesis elucidated the most
important physical aspects, and showed the operational suitability of the bioreactor.

173

174
References
175
References

Adlercreutz P. 2000. In Applied Biocatalysis. Biocatalysis in non-conventional media.
Ed. Straathof AJJ and P. Adlercreutz, 2
nd
edn Harwood Academic Publishers.
(Singapore). p 295 – 316.

Almering JHJ (Ed.). 1988. Analyse. Delftse Uitgevers Maatschappij (Delft). p 356 -
359.

Baird MHI and Rice RG. 1975. Axial dispersion in large unbaffled columns. Chem
Eng J 9:171-174.

Bird RB, Stewart WE and Ligthfoot EN. 1960. Transport Phenomena. John Wiley &
Sons (Singapore). p 200.

Boeren S, Laane C and Hilborst R. 1992. In Biocatalysis in Non-conventional media.
The effect of alkanes on viability, enzyme induction and enzyme activity in
Flavobacterium dehydrogenans. Ed Tramper J, Vermuë MH, Beeftink HH and Von
Stockar U, Elsevier Science Publ (Amsterdam). p 637 – 643.

Boon MA, Van ‘t Riet K and Janssen AEM. 2000. Enzymatic synthesis of
oligosaccharides: product removal during a kinetically controlled reaction.
Biotechnol Bioeng 70:411-420.

Brink LES and Tramper J. 1985. Optimization of organic solvent in multiphase
biocatalysis. Biotechnol Bioeng 27:1258 – 1269.

Bruggink A (ed). 2001. Synthesis of beta-lactam antibiotics. Chemistry, biocatalysis &
process integration. Kluwer Academic Publishers. (Dordrecht).

Duijn G and Rietema K. 1982. Segregation of liquid-fluidized solids. Chem Eng Sci
39:365-368.

Buitelaar RM, Langenhoff AAM, Heidstra R and Tramper J. 1991. Growth and
thiophene production by hairy root cultures of tagetes patula in various two-liquid-
phase bioreactors. Enzym Microb Techn 13:487 – 494.
References
176

Chen Y-M and Fan L-S. 1990. Drift flux in gas-liquid-solid fluidized systems from the
dynamics of bed collapse. Chem Eng Sci 45:935-945.

Chisti MY (ed.). 1989. Air-lift Bioreactors. Elsevier Science Publishers BV (London).

Collins AM, Woodley JM and Lidell JM. 1995. Determination of reactor operation for
the microbial hydroxylation of toluene in a two-liquid phase process. J Ind Microbiol
14:382-388

Coutouly, http://gerard.coutouly.free.fr/fr/Origine/cdindetf.html

Dalingaros W, Kumar A and Hartland S. 1986. Effect of physical properties and
dispersed-phase velocity on the size of drops produced at a multi-nozzle distributor.
Chem Eng Process 20:95-102.

Dakshinamurty P, Veerabhadrarao K and Venkatarao AB. 1979. Bed porosities in 3-
phase (liquid-liquid) fluidized beds. Ind Eng Chem Proc Des Dev 18:638 – 640.

Dakshinamurty P, Subrahmanyam V, Prasada Rao RV and Vijayasaradhi P. 1984.
Liquid-liquid mass transfer in 3-phase fluidized beds. 3. Measurement of individual
fluid-phase resistances. Ind Eng Chem Proc Des Dev 23:132 – 137

Dallavalle JM. 1948. Micrometrics: the technology of fine particles. 2
nd
Edn. Pitman
(London).

Davison BH and Thompson JE. 1993. Continuous direct solvent extraction of butanol
in a fermenting fluidized-bed bioreactor with immobilized Clostridium acetobutylicum.
Appl Biochem Biotech 39/40: 415-426.

De Bont JAM. 1998. Solvent-tolerant bacteria in biocatalysis. Trends Biotechnol
12:493-499

Di Felice R. 1995. Hydrodynamics of liquid fluidisation. Chem Eng Sci 50(8): 1213-
1245.

References
177
Krab-Hüsken LE. 2002. Production of catechols: microbiology and technology. PhD
thesis, Wageningen Agricultural University, Wageningen, The Netherlands.

Fahim MA and Wakao N. 1982. Parameter estimation from tracer response
measurements. Chem Eng J 25:1-8.

Fan L-S. Muroyama K. and Chern S-H. 1982. Hydrodynamic characteristics of
inverse fluidization in liquid-solid and gas-liquid-solid systems. Chem Eng J 24:143 –
150.

Flaschel E, Crelier S, Schulz K, Huneke F-U and Renken A. 1992. In Biocatalysis in
non-conventional media. Process development for the optical resolution of
phenylalanine by means of chymotrypsin in a liquid-liquid-solid 3-phase reaction
system. Ed Tramper J, Vermuë MH, Beeftink HH and Von Stockar U. Elsevier
Science Publishers BV. (Amsterdam). p 637 – 643.

Foscolo PU, Di Felice R and Gibilaro IG. 1987. An experimental study of expansion
characteristics of gas fluidized beds of fine catalysts. Chem Eng Process 22:69-78.

Foscolo PU, Gibilaro LG and Waldram SP. 1983. An unified model for particulate
expansion of fluidized beds and flow in fixed porous media. Chem Eng Sci 38:1251-
1260.

Garside J and Al-Dibouni MR. 1977. Velocity-voidage relationships for fluidization
and Sedimentation in solid-liquid Systems. Ind Eng Chem Process Des Dev 16:206-
214.

Gòdia F and Solà S. 1995. Fluidized-bed bioreactors. Biotechnol Prog 11:479-497.

Gòdia F, Casas C and Solà C. 1987. A survey of continuous ethanol fermentation
systems using immobilized cells. Process Biochem 4:43-48.

Godfrey JC and Slater MJ. 1991. Slip velocity relationships for liquid-liquid extraction
columns. Trans IchemE 69:130-141.

References
178
Godfrey JC and Slater MA (Ed.) 1994. Liquid-liquid extraction equipment. John
Wiley and Sons (New York).

Gommers PJF, Christoffels LP, Kuenen JG and Luyben KChAM. 1986. Gas-phase
influence on the mixing in a fluidized bed bio-reactor. Appl Microbiol Biotechnol
25:1-7.

Grbavcic ZB, Garic RV, Hadzismajlovic DZE and Jovanvic S. 1991. Variational model
for prediction of the fluid-particle interphase drag coefficient and particulate
expansion of fluidized and sedimenting beds. Powder Technol 68:199-211.

Hartman M, Havlin V, Trnka O and Carsky M. 1989. Predicting the free-fall velocities
of spheres. Chem Eng Sci 44:1743-1745.

Heijnen JJ, Hols J., Van der Lans GJM, Van Leeuwen HLJM, Mulder A. and
Weltevrede R. 1997. A simple hydrodynamic model for the liquid circulation velocity
in a full scale two- and 3-phase internal airlift reactor operating in the gas
recirculation regime. Chem Eng Sci 52:2527 – 2540.

Hidaka N, Kakoi K and Matsumoto T. 1994. Correlation of solid hold-up in liquid-
solid fluidized bed. J Chem Eng Jpn 27:563-570.

Hiemenz PC (Ed). 1986. Principles of colloid and surface chemistry. 2
nd
Marcel
Dekker Inc (New York). p 288-293.

Hulst AC, Tramper J, Van ‘t Riet K and Westerbeek JMM. 1985. A new technique for
the production of immobilized biocatalyst in large quantities. Biotechnol Bioeng
27:870-876.

Hunik JH and Tramper J. 1993. Large Scale production of κ-carrageenan droplets for
gel bead production: theoretical and practical limitations of size and production rate.
Biotech Progress 9:186-192.

Hunik JH, Tramper J and Wijffels RH. 1994. A strategy for scale-up of nitrification
processes with immobilized nitrifying cells. Bioprocess Eng. 11:73-82.

References
179
Hüsken LE, Oomes M, Schroën CHGP, Tramper J, De Bont JAM and Beeftink HH.
2002. Membrane-facilitated bioproduction of 3-methylcatachol in an organic/water
2-phase system. J Biotechnol 96: 281-289.

Hüsken LE, Dalm MCF, Tramper J, Wery J, De Bont JAM and Beeftink HH. 2001.
Integrated bioproduction and extraction of 3-methylcatechol. J Biotechnol 88:11-19.

Kahn RA and Richardson JF. 1987. The resistance to motion of a solid sphere in a
fluid. Chem Eng Commun. 62:135-150.

Kawakami K and Yosida T. 1992. In Biocatalysis in Non-conventional media. Kinetic
study of enzymatic reaction in aequous-organic 2-phase systems – an example of
enhanced production of aldehydes by alcohol oxidase. Ed Tramper J, Vermuë MH,
Beeftink HH and Von Stockar U. Elsevier Science Publishers BV. (Amsterdam). p. 637
– 643.

Kikuchi K, Itoh E, Takahaski H and Sugawara T. 1994. Mass transfer between
particles and liquid in dilute fluidized beds containing small particles. Chem Eng J
27:674-677.

Kim SD and Kim CH. 1983. Axial dispersion characteristics of 3-phase fluidized beds.
J Chem Eng Japan 16:172 – 178.

Kim SD, Han PW and Yu YH. 1988. Phase holdups and liquid-liquid mass transfer in
three phase fluidized beds. Chem Eng Commun 68:57-68.

Kim SD, Lee MJ and Han JH. 1989. Axial dispersion characteristics of three (liquid-
liquid-solid) phase fluidized beds. Can J Chem Eng 67:276 – 282.

Kim SD, Kim DY and Han JH. 1994. Dispersed phase characteristics in 3-phase
(liquid-liquid-solid) fluidized beds. Can J Chem Eng 72: 222-228

Kim SD and Kang Y. 1997. Heat and mass transfer in 3-phase fluidized-beds reactors
– an overview. Chem Eng Sci 52:3639 – 3660.

References
180
Kim SD, Kim JS, Nam CH, Kim SH and Kang Y. 1999. Immersed heater-to-bed heat
transfer in liquid-liquid-solid fluidized beds. Chem Eng Sci 54:5173 – 5179.

Kumar A and Hartland S. 1985. Gravity settling in liquid/liquid dispersions. Can J
Chem Eng 63:368-376.

Kumar A and Hartland S. 1995. A unified correlation for the prediction of dispersed-
phase hold-up in liquid-liquid extraction columns. Ind Eng Chem Res 34:3925-3940.

Kumar A and Hartland S. 1996. Unified correlations for the prediction of drop size in
liquid liquid extraction columns. Ind Eng Chem Res 35:2682-2695.

Leenen EJTM. 2001. In Immobilized Cells. Mechanical Stability of the Support. Ed.
Wijffels RH, Springer Lab Manual (Berlin).

Leenen EJTM. 2001. In Immobilized Cells. Support Material Stability at the Process
Conditions Use. Ed. Wijffels RH, Springer Lab Manual (Berlin).

Levenspiel O (Ed.). 1972. Chemical reaction engineering. John Wiley ( New York).

Marshall CT and Woodley JM. 1995. Process synthesis for multi-step microbial
conversions. Bio/Technology 13:1072-1078.

Mateus DMR, Alves SS and Da Fonseca MMR. 1996. Model for the production of L-
tryptophan from L-serine and indole by immobilized cells in a 3-phase liquid-
impelled loop reactor. Bioprocess Eng 14:151 – 158.

Michielsen MJF, Wijffels RH, Tramper J and Beeftink HH. 2001. In Multi-phase
bioreactor design. Chapter 8 Solid-to-solid bioconversions: batch or continuous?. Ed.
Cabral JMS, Mota M and Tramper J, Taylor and Francis (London, New York). p 225-
247.

Muroyama K and Fan L-S. 1984. Fundamentals of gas-liquid-solid fluidization.
AIChE J 31:1-34.

References
181
Nelder JA and Mead R. 1965. A simplex method for function minimization.
Computer J 7: 308-313.


Newman AB. 1931. The drying of porous solids: diffusion and surface emission
equations. Trans Am Int Chem Engrs 27:203 – 220.

Paireau O and Bonn D. 1999. Drag reduction in liquid-liquid friction. Physical
Review Letters 83:5591-5594.

Qureshi N and Maddox IS. 1995. Continuous production of Acetone-butanol-ethanol
Using Immobilized Cells of Clostridium acetobutylicum and Integration with product
removal by liquid-liquid extraction. J Ferment Bioeng 80:185-189.

Ranz WE and Marshall WR. 1950. Evaporation from drop I + II. Chem Eng Progr 48:
141 – 146, 173 – 180.

Riba JP and Couderc JP. 1977. Expansion de couches fluidizées par des liquides. Can
J Chem Engng 55:118-121.

Richardson JF and Zaki WN. 1954. Sedimentation and fluidization. Part 1. Trans Inst
Chem Eng 32:35-53.

Roszak J and Gawronski R. 1979. Cocurrent liquid-liquid extraction in fluidized beds.
Chem Eng J 17:101-109.

Rowe PN. 1987. A convenient model empirical equation for estimation of the
Richardson-Zaki exponent. Chem Eng Sci. 43:2795 – 2796.

Schmid A, Dordick JS, Hauer B, Kiener A, Wubbolts M and Witholt B. 2001.
Industrial biocatalysis today and tomorrow. Nature 409:258-268.

Stieglitz B, Dicosimo R and Fallon RD. 1996. Formation of aliphatic ω-cyano-
carboxamide(s) from α,ω-dinitrile(s) – using biocatalyst having regioselective nitrile
hydratase activity derived from Pseudomonas putida. US Patent US 5728556.
References
182

Thompson VS and Worden RM. 1992. The effect of tracer diffusion on liquid
dispersion in a three phase fluidized bed bioreactor. Chem Eng Sci 47:3435 – 3441.

Tramper J, Wolters I and Verlaan P. 1987. In Biocatalysis in organic media. The
liquid-impelled loop reactor: a new type of density-difference-mixed bioreactor. Ed.
Laane C, Tramper J and Lilly MD, Elsevier Science Publishers BV (Amsterdam). p
311 – 316.

Tramper J, Vermuë MH, Beeftink HH and Von Stockar U (Eds.). 1992. Progress in
Biotechnology 8. Biocatalysis in non-conventional media. Elsevier Science Publishers
BV (Amsterdam).

Turton R and Levenspiel O. 1986. A short note on the drag coefficient for spheres.
Powder Technol 47:83-86.

Turton R and Clark NN. 1987. An explicit relationship to predict particle settling
velocity. Powder Technol 53:127-129.

Ueyama K and Miyauchi T. 1979. Properties of recirculating turbulent 2-phase flow
in gas bubble columns. AIChE J 15:258-266.

Van den Tweel WJJ, Marsman EH, Vorage MJAW, Tramper J and De Bont JAM. 1987.
In Bioreactors and biotransformations. The application of organic solvents for the
bioconversion of benzene to cis-benzene glycol. Ed. Moody GW and Baker PB,
Elsevier Science Publishers BV (London). p. 231 – 241.

Van der Meer AB. 1993. Multi-phase fermentation: reaction engineering models for
the production of optically active 1,2-epoxyoctane from 1-octene by free and
immobilized Pseudomonas oleovorans cells. PhD thesis, University of Groningen,
Groningen, The Netherlands.

Van der Wielen LAM, Van Dam MHH and Luyben KChAM. 1996. On the relative
motion of a particle in a swarm of different particles. Chem Eng Sci 51:995-1008.

References
183
Van der Wielen LAM, Van Dam MHH and Luyben KChAM. 1996a. Trickle flow of
dense particles in a fluidized bed of others. Chem Eng Sci 52: 553-565

Van der Wielen LAM, Diepen PJ, Houwers J and Luyben KChAM. 1996b. A
countercurrent adsorptive reactor for acidifying bioconversions. Chem Eng Sci
51:2315-2325.

Van der Wielen LAM. 1997. A countercurrent Adsorptive Fluidized Bed Reactor for
Heterogeneous Bioconversions. PhD thesis, Technical University Delft, Delft, The
Netherlands.

Van Sonsbeek HM, Beeftink HH and Tramper J. 1993. Two-Liquid-phase bioreactors.
Enzyme Microb Technol 15:722 – 729.

Van Sonsbeek HM, Verdurmen REM, Verlaan P and Tramper J. 1990. Hydrodynamic
model for liquid-impelled loop reactors. Biotech Bioeng 36: 940 – 946.

Van Sonsbeek HM, Gielen SJ and Tramper J. 1991. Steady state method for kla
measurements in model systems. Biotech Techn 5:157-162.
Van Sonsbeek HM. 1992a. Physical aspects of liquid-impelled loop reactors. PhD
thesis, Wageningen Agricultural University, Wageningen, The Netherlands.

Van Sonsbeek HM, De Blank H and Tramper J. 1992b. Oxygen transfer in liquid-
impelled loop reactors using perfluorocarbon liquids. Biotechnol Bioeng 40:713 – 718.

Van ‘t Riet K and Tramper J. (Ed.). 1991. Basic Bioreactor Design. 2
nd
edn. Marcel
Dekker Inc (New York).

Van Zessen E, Rinzema A and Tramper J. 2003. Fluidized-bed and packed-bed
characteristics of gel beads. submitted (Chapter 2).

Van Zessen E, Rinzema A and Tramper J. 2003. Phase hold-ups in a water-dodecane
2-phase system. submitted (Chapter 3)

References
184
Van Zessen, E., Dierendonck, F. van, Rinzema A. and Tramper J. (2003) Solid and
droplet phase hold-ups in a liquid-liquid-solid 3-phase fluidized bed bioreactor.
submitted (Chapter 4).

Van Zessen, E, Keijzer, T., Rinzema A. and Tramper J. (2003) Residence time
distribution of the water phase in a liquid-liquid-solid 3-phase fluidized bed
bioreactor. submitted (Chapter 5).

Van Zessen E, Tramper J and Rinzema A. 2001. In Immobilized Cells. Measurement
of density, particle size and shape of support. Ed Wijffels R, Springer Lab Manual,
Springer-Verlag (Berlin).

Vermuë MH, Sikkema J, Bakker R, Janssen G and Tramper J. 1995. Strategy for
selection of a suitable solvent for extractive biocatalysis in a liquid-impelled loop
reactor. Vermuë MH. In Biocatalysis in non-conventional media thermodynamic and
kinetic aspects. PhD thesis, Wageningen Agricultural University, Wageningen, The
Netherlands.

Vermuë MH and Tramper J. 1995. Biocatalysis in non-conventional media; medium
engineering aspects – a review. Pure & Applied Chem 67:345-373.

Vermuë MH, Tacken M and Tramper J. 1994. Tetralin and oxygen transfer in the
liquid-impelled loop reactor. Bioprocess Eng 11:224 – 228.

Wallis GB (Ed.). 1969. One-dimensional 2-phase flow. McGraw-Hill (New York).

Wang H, Seki M and Furusaki S. 1994. Characteristics of immobilized Lactobacillus
Delbrueckii in a liquid-solid fluidized bed bioreactor for lactic acid production. J
Chem Eng Jpn 28:198-203.

Weast RC. 1979. Handbook of Chemistry and Physics. 59th Edition. CRC Press Boca
Raton.

Wilhelm RH and Kwauk M. 1948. Fluidization of solid particles. Chem Eng Prog
44:201-217.

References
185
Willaert RG, Van Baron G and De Backer L. 1996. In Immobilized living cell systems
modeling and experimental methods. Immobilized cell systems. John Wiley & Sons
(New York). p. 67-97.

Wijffels RH, Buitelaar RM, Bucke C and Tramper J (Ed.). 1995. Immobilized Cells:
Basics and Applications. Elsevier (Amsterdam).

Yu H and Rittmann BE. 1997. Predicting bed expansion and phase holdups for 3-
phase fluidized bed reactors with and without biofilm. Wat Res 31: 2604 – 2616.

Zigrang DJ and Silvester ND. 1981. An explicit equation for particle settling velocities
in solid-liquid systems. AIChE J 27:1043-1044.

186
Summary
187
Summary

If a batchwise bioconversion and subsequent (i.e. serial) downstream processing are
adopted as a standard production strategy, a low overall productivity may result for
certain types of biotransformations. Examples may be found among reactions that
are kinetically inhibited by a substrate or a product, transformations featuring an
unfavorable thermodynamic equilibrium, and bioconversions involving poorly
soluble substrates or products.
An alternative to sequential processing would be the integration of
bioconversion and downstream processing by controlled supply of substrate to the
reaction in situ, or by controlled removal of product from the reaction in situ. Such
integration involves a multi-phase reactor in which a helper phase may serve as
substrate reservoir or as product sink.
To facilitate integration of reaction and product removal or substrate supply, a
new type of multi-phase bioreactor was developed, containing a continuous liquid
phase, a dispersed liquid phase, and a dispersed solid phase; it may be conceived of
as a 3-phase liquid-liquid-solid fluidized bed. The solid reaction phase were
biocatalytic gel beads that were fluidized by the continuous medium phase; through
this fluidized bed, organic droplets rose and served as a helper phase (extracting the
product or acting as a substrate sink).

This thesis focuses on the hydrodynamic aspects of such a 3-phase fluidized bed; a
general design strategy was developed. Within this 3-phase system, two 2-phase sub-
systems may be identified: a liquid-solid fluidized bed consisting of biocatalytic gel
beads in a continuous medium phase, and a liquid-liquid extraction column or spray
column, consisting of organic droplets in a continuous medium phase. The
hydrodynamic characteristics of these constituent 2-phase systems were studied
separately (Chapters 2 and 3). Once good descriptions and predictions of the
hydrodynamics of these individual 2-phase systems were obtained, it was assumed
that the two corresponding models could be combined to describe and predict the
hydrodynamic characteristics of the 3-phase fluidized bed. For the phase hold-ups in
the 3-phase system, this assumption was only partly valid (Chapter 4). The
continuous-phase axial dispersion coefficient of the 3-phase fluidized bed, however,
could be reliably predicted from a model for this dispersion coefficient in a 2-phase
droplet column (Chapter 5).

Summary
188
The physical behavior of gel beads was studied; they were either settling as single
beads, or present in a packed or fluidized bed (Chapter 2). For five types of gel beads
of different diameter and density, three related characteristics were analyzed that all
involved the drag force between a particle and its surroundings: the terminal settling
velocity of a single gel bead, the pressure drop over a packed bed, and the voidage in
a liquid-fluidized bed. Although gel beads showed similar characteristics as
conventional solids, established models did not quantitatively predict It is concluded
that the drag force between the gel beads and liquid was lower than for conventional
solids. Two hypotheses were put forward to explain this. The first one attributes the
drag reduction to small amounts of dissolved polymer. The second one attributes the
smaller drag force to the nature of the gel beads; gel beads contain over 95% of water
and can in that sense be regarded as ‘rigid’ water droplets. Hence, the gel bead
surface might show waterlike properties. A new model was presented that correctly
predicted the drag coefficient of a single gel bead in a packed or fluidized bed. It was
used to predict the pressure drop over a packed bed of beads and the voidage in a
fluidized bed of beads.

Droplet hold-ups in a liquid/liquid 2-phase spray column were studied with
different sparger geometries (Chapter 3). The continuous phase (water) and the
dispersed phase (dodecane droplets) were fed into the column co-currently and
counter-currently. Although water and dodecane superficial velocities were within
the same order of magnitude, the water velocity had only a minor influence on the
dodecane hold-up. The number of nozzles, however, influenced the dodecane hold-
up to a large extent: at the same dodecane superficial velocity, the sparger with the
lowest number of nozzles showed the highest hold-up. The droplet hold-up could be
well described using a model based on the slip velocity between the continuous
phase and disperse phase. The slip velocity obeyed a Richardson and Zaki type
equation. It features a parameter n, for which a new model was presented. Up to a
hold-up of 0.13, this parameter was found constant (n=4); for higher hold-ups, a
decrease in n was found. The validity of this model was demonstrated by extending
it to a 2-phase system differing in surface tension (60 mM KCl /dodecane).

The integration of a solid (gel bead)-liquid fluidized bed with a droplet spray column
results in a liquid-liquid-solid 3-phase fluidized bed reactor. Integration of the
models for a solid/liquid 2-phase system (a fluidized bed) and a liquid/liquid 2-
phase system (a droplet spray or extraction column) was assumed to yield a potential
Summary
189
description of a liquid-liquid-solid 3-phase fluidized bed reactor. The different phase
hold-ups of this 3-phase fluidized bed were studied as well as the flow behavior of
the fluidized bed (Chapter 4). The reactor featured a bed of Celite™-enforced κ-
carrageenan gel beads as solid phase (density 1065 kg m
-3
). The bed was fluidized by
a continuous aqueous phase (30 mM KCl, ‘water’); a disperse organic phase (n-
dodecane droplets) rose through the bed of particles. At various velocities of water
and n-dodecane, the flow characteristics of the gel beads in the fluidized bed were
characterized; in addition, the various phase hold-ups were studied. From visual
observation of the fluidized bed, five flow regimes were defined for the gel beads:
two stable homogeneous regimes, two stable heterogeneous regimes, and an
unstable regime in which gel beads were washed-out from the bed. Literature
models for existing 3-phase systems were tried to predict phase hold-ups. The more
fundamental literature models for existing 3-phase systems rely on parameters valid
for the individual 2-phase systems. Unfortunately, the prediction with these models
was poor using parameters derived for the individual 2-phase systems. Prediction
with other literature models was also poor. Consequently, a new model for a
fluidized bed with gel beads and dodecane droplets had to be derived, based on
stationary force balances. To this end, drag-force models for a single gel bead and for
a single droplet in a 3-phase suspension were set up. All data for gel bead hold-up
and the data for droplet hold-ups below 0.06 were well described. For the description
of the entire range of droplet hold-ups, an empirical model was adopted.

The study of the continuous phase mixing in the 3-phase fluidized bed is described
in Chapter 5. Mixing was characterized by an axial dispersion coefficient that was
determined from residence-time distribution measurements with a KCl-tracer.
Because of KCl mass transfer into the gel beads, the inherent limitations of this tracer
had to be studied as well. A pulse with a small volume was found to give reliable
results. The continuous-phase axial dispersion coefficient was found to increase with
the superficial n-dodecane velocity, while its dependency on the continuous-phase
superficial velocity showed a maximum. These results reflect the increase in
continuous-phase dispersion coefficient with the energy dissipation rate per unit
mass of continuous phase. A literature relationship for this dependency was found to
yield good predictions of the dispersion coefficient with some minor exceptions. It is
concluded that the mixing of the continuous phase in a liquid-liquid-solid 3-phase
fluidized bed is comparable with the mixing of the continuous phase of either a
bubble column or the main tube of a liquid-impelled loop reactor.
Summary
190
A simple process design for a 3-phase liquid-liquid-solid fluidized bed bioreactor
was made (Chapter 6). It was based on the results from the preceding chapters, on
educated guesses for mass exchange and flow behavior of both dispersed phases (gel
beads and droplets); a simple product-inhibited reaction was assumed. It was
demonstrated that application of a 3-phase reactor system resulted in a higher degree
of conversion than 2-phase liquid-solid fluidized bed reactors. From a practical
viewpoint, it was concluded that implementation of a 3-phase system would require
minor adaptations only as compared to a 2-phase fluidized bed bioreactor. On the
basis of the model for the 3-phase fluidized-bed bioreactor, the influence of different
parameters was discussed, such as the distribution coefficient of the product over the
medium and solvent phases, the product toxicity, and the solvent flux. Operational
conditions were established at which the 3-phase reactor would outperform a
conventional 2-phase fluidized bed. Already at quite modest values of the
distribution coefficient, the 3-phase fluidized bed performs better than the 2-phase
fluidized bed. For conversions that would benefit from in situ extraction, therefore, a
3-phase fluidized bed would be the apparatus of choice.

The thesis finally reviews the current status of the newly developed 3-phase liquid-
liquid-solid fluidized-bed bioreactor; it outlines additional research to be performed
for a full understanding of its hydrodynamic behavior. Additional steps that have to
be taken for practical implementation of the 3-phase fluidized bed as a new type of
bioreactor are emphasized.

191
Samenvatting
192
Samenvatting

Indien voor een aantal typen fermentaties een normale productiestrategie wordt
toegepast, dat wil zeggen eerst een batchgewijze fermentatie en vervolgens de
opwerking van het gewenste product, dan zal dit resulteren in een lage
productiviteit voor het gehele proces. Voorbeelden van dergelijke fermentaties zijn:
1) substraat of product geremde reacties, 2) bioconversies die sterk beperkt worden
door een ongunstige evenwichtsligging, en 3) fermentaties die gepaard gaan met een
slechte oplosbaarheid van het substraat of het product.
Een alternatief voor de normale productiestrategie kan de directe integratie in
één apparaat van bio-conversie en opwerking zijn. Hierbij is dan sprake van
enerzijds een in situ gecontroleerde toevoer van substraat naar de fase waar de
reactie plaatsvindt, of anderzijds de gecontroleerde afvoer van product vanuit de
reactiefase. Voor een dergelijke integratie is een meerfase reactor noodzakelijk,
waarbij een hulpfase gebruikt wordt als substraatreservoir of als productafvoer.
Een nieuw soort meerfasenreactor was ontwikkeld om de integratie van
reactie en productafvoer of substraattoevoer te vergemakkelijken. Deze
meerfasenreactor bestond uit een continue vloeistoffase, een disperse vloeistoffase en
een disperse vaste fase. De reactor kan worden gezien als een 3-fase vloeistof-
vloeistof-vast gefluïdiseerd bed bioreactor. De vaste fase bestaat uit biokatalytische
gel deeltjes, die worden gefluïdiseerd door de continue vloeistoffase. Door dit
gefluïdiseerde bed stijgen druppels organisch oplosmiddel op. Deze druppels treden
op als de hulpfase; zij extraheren het product of zij dienen als substraatreservoir.

In dit proefschrift worden hydrodynamische aspecten van een dergelijk 3-fase
gefluïdiseerd bed onderzocht. Daarnaast is een algemene ontwerpstrategie voor een
3-fasen gefluïdiseerd bed bioreactor ontwikkeld. In een 3-fase systeem zijn er twee 2-
fasen systemen te onderscheiden: een vast-vloeistof gefluïdiseerd bed, dat bestaat uit
de biokatalytische deeltjes in een continue vloeistoffase, en een vloeistof-vloeistof
extractie kolom, die bestaat uit druppels van een organisch oplosmiddel in een
continue vloeistoffase. Van deze twee afzonderlijke 2-fase systemen werden de
hydrodynamische karakteristieken onderzocht (Hoofdstuk 2 en 3). Nadat een goede
beschrijving en voorspelling van het hydrodynamische karakter van deze twee
afzonderlijke 2-fase systemen was verkregen, werd er verondersteld dat de
corresponderende modellen konden worden gecombineerd om het
hydrodynamische karakter van het 3-fase gefluïdiseerd bed te kunnen beschrijven en
Samenvatting
193
voorspellen. Deze veronderstelling was gedeeltelijk waar voor het voorspellen van
de verschillende fasen hold-ups in het 3-fasen systeem (Hoofdstuk 4). De
voorspelling voor de axiale dispersiecoëfficiënt van de continue vloeistoffase was
goed, indien een model gebruikt werd dat deze dispersie coëfficiënt voorspelde in
een 2-fase druppelkolom (Hoofdstuk 5).

Het fysische gedrag van gel deeltjes werd bestudeerd: een gel deeltje bezonk als
enkel deeltje in stilstaand water, of was aanwezig in een gepakt bed of gefluïdiseerd
bed (Hoofdstuk 2). Voor vijf verschillende soorten gel deeltjes (verschillend in
diameter en dichtheid) werden drie verwante karakteristieken bestudeerd. Deze
karakteristieken hadden te maken met de weerstandskracht tussen het deeltje en zijn
omringende omgeving: de eindige valsnelheid van één enkel deeltje, de drukval over
een gepakt bed van deze deeltjes, en de deeltjes fractie in een vloeistof-gefluïdiseerd
bed. Hoewel deze karakteristieken dezelfde trends lieten zijn zoals deze worden
waargenomen voor conventionele deeltjes (glas knikkers), bleek dat gerenommeerde
modellen deze karakteristieken niet kwalitatief konden voorspellen. Er werd
geconcludeerd dat de weerstandskracht tussen gel deeltje en vloeistof lager was
vergelijken met conventionele deeltjes. Twee verschillende hypothesen werden
voorgesteld. De eerste hypothese veronderstelde dat de vermindering in weerstand
veroorzaakt wordt door de aanwezigheid van een kleine hoeveelheid opgeloste
polymeer. De tweede hypothese ging uit van de aard van de gel deeltjes. Gel deeltjes
bestaan voor meer dan 95% uit water, en in dat licht bezien kunnen gel deeltjes
worden beschouwd als ‘vaste’ water druppels. Vandaar dat het oppervlak van het
gel deeltje waterachtige eigenschappen kan vertonen, met als gevolg dat de
weerstandskracht lager is. Een nieuw model dat de weerstandskracht van één gel
deeltje in een gepakt of gefluïdiseerd bed correct beschrijft, werd gepresenteerd. Dit
model werd gebruikt voor de voorspelling van de drukval over een gepakt bed en de
deeltjes hold-up in een gefluïdiseerd bed.

Voor verschillende spargerontwerpen werd de druppel hold-up in vloeistof-vloeistof
sproeikolom bestudeerd (Hoofdstuk 3). De continue waterfase en de disperse fase
(dodecaan druppels) werden zowel in mee- als in tegenstroom door de kolom geleid.
Hoewel de superficiële watersnelheid en dodecaansnelheid van dezelfde orde
grootte waren, had de watersnelheid slechts een zeer geringe invloed op de
dodecaan hold-up. Het aantal pijpjes had echter wel een grote invloed op de
dodecaan hold-up: bij een gelijke superficiële dodecaan snelheid werd de hoogste
Samenvatting
194
hold-up waargenomen voor die sparger met het laagste aantal pijpjes. De druppel
hold-up kon goed worden beschreven met een model dat gebaseerd was op het
snelheidsverschil tussen de continue fase en de dodecaan fase. Dit snelheidsverschil
volgde een Richardson en Zaki-achtige vergelijking. Deze vergelijking maakt gebruik
van een parameter n, waarvoor een nieuw model werd gepresenteerd. Tot een hold-
up van 0.13 werd gevonden dat deze parameter constant was (n=4); voor hogere
hold-ups nam deze parameter af. Dit model werd succesvol toegepast voor de
voorspelling van de hold-up van een ander 2-fase systeem, dat verschilde in
oppervlaktespanning (60 mM KCl/ dodecaan).

De integratie van een vast–vloeistof gefluïdiseerd bed met een druppel sproeikolom
heeft als resultaat een vloeistof-vloeistof-vast 3-fase gefluïdiseerd bed. Er werd
verondersteld dat de integratie van de modellen voor een vloeistof/vast 2-fasen
systeem (een gefluïdiseerd bed) met een vloeistof/vloeistof 2-fasen (een druppel
sproeikolom), een goede beschrijving zou kunnen opleveren voor het 3-fasen
systeem. De verschillende fasen hold-ups van dit 3-fasen systeem en het
stromingsgedrag van de gel deeltjes werden bestudeerd (Hoofdstuk 4). Als vaste fase
(dichtheid 1065 kg m
-3
) werden κ-carrageen deeltjes gevuld met Celite

gebruikt.
Deze deeltjes werden gefluïdiseerd door een continue waterfase (30 mM KCl,
‘water’). De disperse organische fase (dodecaan druppels) steeg op door het
gefluïdiseerd bed van gel deeltjes. Voor verschillende water en dodecaansnelheden
werden de stromingskarakteristieken van de gel deeltjes beschreven. Daarnaast
werden de verschillende fasen hold-ups bestudeerd. Op basis van visuele
waarneming van het gefluïdiseerde bed konden vijf verschillende stromingsregimes
worden onderscheiden: twee stabiele homogene stromingsregimes, twee stabiele
heterogene stromingsregimes, en één instabiel stromingsregime, waarbij uitspoeling
van de deeltjes uit het bed optrad. Verschillende literatuurmodellen voor bestaande
3-fasen systemen werden geprobeerd om de verschillende fase hold-ups te
voorspellen. Hierbij gebruiken de meer fundamentele modellen parameters, die
afgeleid zijn voor de afzonderlijke 2-fasen systemen. Helaas was de voorspelling met
deze modellen slecht, indien de parameters, afgeleid voor de afzonderlijke 2-fasen
systemen, werden gebruikt. De voorspelling met empirische literatuur modellen was
ook niet goed. Dus een nieuw model voor een gefluïdiseerd bed van gel deeltjes en
dodecaan druppels moest worden afgeleid. Dit model was gebaseerd op stationaire
krachtenbalansen. Daartoe werden modellen opgesteld voor de weerstandskrachten
voor zowel een gel deeltje als een dodecaan druppel aanwezig in de 3-fasen
Samenvatting
195
suspensie. Alle data voor de deeltjes hold-up en druppel hold-up werden goed
beschreven voor een druppel hold-up kleiner dan 0.06. Voor de beschrijving van de
gehele data set werd een empirisch model opgesteld.

Het onderzoek naar de mening van de continue fase van een 3-fase gefluïdiseerd bed
is beschreven in hoofdstuk 5. Menging wordt gekenmerkt door een axiale dispersie
coëfficiënt. Deze coëfficiënt werd bepaald op basis van metingen aan de
verblijftijdspreiding, waarbij gebruik werd gemaakt van een KCl-merkstof. KCl kan
de deeltjes in diffunderen, en daarom werden de beperkingen van het gebruik van
deze merkstof ook bestudeerd. Er werd gevonden dat een puls met een gering
volume betrouwbare resultaten gaf. Het bleek dat de axiale dispersiecoëfficiënt van
de continue fase toenam met de superficiële dodecaansnelheid, terwijl de
afhankelijkheid van deze dispersiecoëfficiënt met de superficiële watersnelheid een
maximum liet zien. Deze resultaten weerspiegelen de toename van de continue fase
dispersiecoëfficiënt met de energiedissipatie per massa-eenheid. Een model uit de
literatuur voor het verband tussen dispersiecoëfficiënt en energiedissipatie werd, op
kleine uitzonderingen na, met succes toegepast voor de voorspelling van de
dispersiecoëfficiënt. Er werd geconcludeerd, dat de menging van de continue fase
van een 3-fase gefluïdiseerd bed vergelijkbaar was met de menging van de continue
fase van zowel een bellenzuil als de hoofdbuis van een ‘liquid-impelled loop reactor’.

Een eenvoudig procesontwerp voor een 3-fase vloeistof-vloeistof-vast gefluïdiseerd
bed bioreactor werd gemaakt (Hoofdstuk 6). Dit ontwerp maakte gebruik van de
resultaten van de voorafgaande hoofdstukken, en weloverwogen schattingen voor
stofoverdracht en stromingsgedrag van beide disperse fasen (de gel deeltjes en de
dodecaan druppels). Een eenvoudige productgeremde reactiekinetiek werd
verondersteld. Er werd aangetoond, dat bij gebruik van een 3-fase reactor een hogere
conversiegraad wordt bereikt vergelijken met een 2-fase vloeistof-vast gefluïdiseerd
bed bioreactor. Er werd geconcludeerd dat, uit praktisch oogpunt, voor de inzet van
een 3-fase systeem er slechts geringe aanpassing aan een 2-fase vloeistof-vast
gefluïdiseerd bed nodig zijn. Met behulp van het procesontwerp werd de invloed
van verschillende modelparameters onderzocht, zoals 1) de verdelingscoëfficiënt van
het product over de waterfase en de organische oplosmiddelfase, 2) de toxiciteit van
het gevormde product, en 3) het debiet van het organische oplosmiddel.
Operationele voorwaarden werden opgesteld, waarvoor de prestatie van het 3-fasen
systeem beter is dan de prestatie van het 2-fasen systeem. Het bleek dat al voor
Samenvatting
196
bescheiden waarden van de verdelingscoëfficiënt het 3-fasen gefluïdiseerd bed beter
presteerde dan het 2-fasen gefluïdiseerd bed. Dus voor die bioconversies, waarvoor
in situ extractie gunstig is, is de 3-fasen gefluïdiseerd bed bioreactor een zeer geschikt
apparaat.

Dit proefschrift wordt afgesloten met een overzicht van de huidige status van deze
nieuw-ontwikkelde 3-fasen gefluïdiseerd bed bioreactor. Er wordt aangeven welk
extra onderzoek noodzakelijk is voor een volledig begrip van het hydrodynamische
gedrag van deze bioreactor. Daarnaast wordt benadrukt welke stappen ondernomen
moeten worden voor de praktische inzet van dit nieuwe type bioreactor.

197
Nawoord

Nog een paar woorden schrijven en dan is het boekje echt af. Terugdenkend aan de
afgelopen jaren bekruipt mij een gevoel dat veel lijkt op een gevoel nadat ik gefinisht
ben na een lange, zware hardloopwedstrijd. Euforische gevoelens overheersen, maar
gedachten gaan ook terug naar de onderweg veelvuldig voorkomende zware
momenten, momenten van overgave en opgave.
Allereerst wil ik jou, Hans, bedanken voor het feit dat je vertouwen in mij had
en de mogelijkheid hebt gegeven om op jouw vakgroep mijn promotie te kunnen
uitvoeren. Hoewel we de deur bij elkaar plat niet liepen, was je wel aanwezig als ik
om raad en advies vroeg. Arjen, als co-promotor en directe begeleider zagen wij
elkaar vaker; bedankt voor je inzet en goede adviezen. Rik, in het begin was je
betrokken bij dit project en vervolgens was je lange tijd uit beeld. In de laatste fase
was je echter weer prominent aanwezig. Bedankt voor je inzet en het helpen gereed
komen van mijn proefschrift.
Gedurende de eerste 4 jaren, heb ik veel hulp gehad van vier studenten,
Wouter, Jenke, Francine en Timo. Wouter, met veel plezier denk ik terug aan onze
discussies. Hoewel van jouw werk en resultaten niets is terug te vinden in dit
proefschrift, zijn een aantal gedachten toch teruggekomen in verschillende
hoofdstukken. Jenke, jij had de pech om met een meetmethode te moeten werken,
waarvan het signaal nauwelijks boven de ruis uitkwam. Jouw resultaten hebben
echter wel bijgedragen aan het tot stand komen van hoofdstuk 5 over
verblijftijdspreiding. Francine, veel heb je gemeten, geklust aan de opstelling en
goede pogingen ondernomen om de hold-up van die deeltjes te voorspellen. Je zult je
resultaten terugvinden in hoofdstuk 4. Timo, na eerst veel gestoeid te hebben met
grafiekjes om de juiste instelling te achterhalen voor het doen van goede
verblijftijdspreidingsmetingen, heb je daarna de ontwikkelde meetmethode
toegepast om het menggedrag te meten. Jouw resultaten zul je terugvinden in
hoofdstuk 5.
Promoveren is vaak een eenzame zaak. Echter er zijn altijd wel mensen,
waarmee je met gedachten kan wisselen of die je simpelweg aansporen. (Ex)-
vakgroepgenoten, als toeschouwers langs de kant was jullie aandacht en
aansporingen zeer welkom. Promoveren naast een volledige aanstelling bleek haast
onmogelijk. ECN, Hans, Johan en Huibert, bedankt dat jullie mij de ruimte hebben
gegeven om het proefschrift af te kunnen ronden.


198
Op deze plaats wil ik nog een groep mensen, van origine hardloopmaatjes, maar
tegenwoordig goede vrienden, bedanken voor hun interesse in mijn
promotieperikelen. Tijdens mijn studie in Delft begonnen als hardloopvrienden, nu
een vriendenclub die elkaar geregeld ziet. In het begin waren er jullie oprechte
vragen, hoe staat het met het boekje?, en mijn standaardantwoord was: het gaat goed,
maar de voortgangssnelheid is niet zo hoog (type duuratleet). De laatste tijd
verstomde jullie vragen, wellicht in de overtuiging dat de voortgangssnelheid tot nul
was gereduceerd. In elk geval bedankt voor de oprechte interesse. Het boekje is nu
af, en is er weer volop tijd voor trainen en mooie hardloopuitjes.

Familie, pa, Stefan, vaak hebben jullie je afgevraagd of dat boekje er nog zou komen
en misschien nog wel belangrijker, komt dat feest er nog? Het boekje ligt er nu en dat
feestje komt er ook. Bedankt voor vooral de mentale steun in de afgelopen jaren.

Als laatste bedank ik jou, Imke, vooral voor je geduld en begrip als ik weer in het
weekeinde achter de PC kroop. Toen wij elkaar leerden kennen had dit boekje
immers al af moeten zijn. De afgelopen periode is veel van onze spaarzame tijd
opgegaan aan werken achter jouw PC-tje. Het is nu af en dus veel tijd over om samen
leuke dingen te gaan doen.



Erik


199
Curriculum vitae

Erik van Zessen werd geboren op 1 december 1969 in Zoetermeer. Na het behalen
van het Gymnasium diploma aan het Stedelijk Gymnasium te Schiedam in 1988
werd in datzelfde jaar begonnen met zijn studie Chemische Technologie aan de
Technische Universiteit Delft. In november 1994 studeerde hij af in de richting
bioprocestechnologie bij het Kluverlaboratorium.

Vervolgens begon hij in februari 1995 als AIO bij de vakgroep proceskunde aan de
Wageningen Universiteit. Het onderzoek dat hij hier heeft uitgevoerd, heeft
geresulteerd in dit proefschrift.

Vanaf oktober 2000 tot oktober 2003, is hij werkzaam geweest bij het Energie
onderzoek centrum Nederland (ECN).

200

Promotor: Prof. dr. ir. J. Tramper Hoogleraar Bioprocestechnologie

Co-promotoren: Dr. ir. A. Rinzema Universitair docent Sectie proceskunde Dr. H.H. Beeftink Universitair docent Sectie proceskunde

Promotiecommissie: Prof. dr. ir. R.M. Boom (Wageningen Universiteit) Dr. ir. A.J.B. van Boxtel (Wageningen Universiteit) Dr. ir. S. Hartmans (DSM) Prof. dr. ir. L.A.M. van der Wielen (Technische Universiteit Delft)

Erik van Zessen

Hydrodynamics of a liquid-liquid-solid

fluidized-bed bioreactor

Proefschrift ter verkrijging van de graad van doctor op gezag van de rector magnificus van Wageningen Universiteit, prof. dr. ir. L. Speelman, in het openbaar te verdedigen op woensdag 1 oktober 2003 des namiddags te half twee in de Aula.

Erik van Zessen Hydrodynamics of a liquid-liquid-solid fluidized-bed bioreactor Thesis Wageningen University, The Netherlands, 2003  with Dutch Summary ISBN 90-5808-875-8

Contents Chapter 1 Chapter 2 Chapter 3 Introduction Fluidized-bed and packed-bed characteristics of gel beads Dispersed-phase hold-up in a liquid-liquid extraction spray column 7 15 43 Chapter 4 Solid and droplet hold-ups in a liquid-liquid-solid 3-phase fluidized-bed 71 Chapter 5 Mixing of the continuous phase in a liquid-liquid-solid fluidized-bed bioreactor 99 125 163 175 187 192 197 199 Chapter 6 Chapter 7 Design of liquid-liquid-solid fluidized-bed bioreactors General discussion References Summary Samenvatting Nawoord Curriculum vitae .

.

1 Introduction 7 .

After the biotransformation. Nowadays.. 2001). In case the substrate strongly inhibits the production rate. 2001). 1996). this operation strategy works satisfactorily. The use of micro-organisms to produce chemicals started more or less with the production of penicillin in 1943 (Coutouly. Only after a prolonged period of time. and production halts due to the approach of thermodynamic equilibrium. or production of 5-cyanopentazamide at DuPont (Stieglitz et al. 2001). Also for product-inhibited biotransformations the amount of product is limited: only after a long period of time. may result in a low overall production in some cases. For ages. Examples of commercial microbial production processes are the production of adipyl-7-ADCA at DSM (Bruggink.Chapter 1 The potentials of biotechnology. their productivity should be enhanced. are enormous. Mostly. Another type are those biotransformations that suffer from an unfavorable thermodynamic equilibrium. in which the desired product is also present. with a standard operation strategy and within a practical biotransformation time. A possible solution to do so is direct removal of the product from the reaction environment (in case of product-inhibited or thermodynamically controlled biotransformations). Substrate is only partially converted. in which the micro-organisms are suspended in an aqueous phase. water of doubtful quality was converted into beer and wine by micro-organisms to make it potable (Coutouly. if these biotransformations should be applied for industrial purposes. the amount of product is limited as well. production of 6-hydroxy-S-nicotine at Lonza (Schmid et al. 2001). chemicals are produced by a number of different micro-organisms (Krab-Hüsken. unless the unconverted substrate is recycled after downstream.e. In case of substrate-inhibited transformations (or low substrate solubility). These are summarized in Table 1.. In ancient days. micro-organisms have also been used for bread making (Coutouly. 2002). A last type is a biotransformation in which the applied substrate has a low solubility. most substrate will be converted. batch biotransformation and subsequent downstream processing of the desired product. most substrate will eventually be converted. the substrate conversion within a practical reaction time is limited and consequently. a controlled supply of substrate to the 8 . consequently. These chemicals are mostly produced in stirred tank reactors. the product is recovered from the mixture by subsequent downstream processing. the use of living organisms or parts thereof for the benefit of mankind in an economic manner. i. 2001). This overall production strategy. So. a small amount of product is formed. For the first three types the overall production is low.

for the biotransformations summarized in Table 1.Incomplete conversion dynamic equilibrium due to the equilibrium 9 . In the remaining part of this introduction we will focus on product-inhibited biotransformations. In this strategy. can be directly used for the other types of biotransformations. So. The same principles. A possible operation strategy for product-inhibited biotransformations is depicted in Figure 1. Supply can be arranged by using an organic solvent with dissolved substrate. Overview of different classes of biotransformations for which productivity is limited using a standard production strategy (batch biotransformation and subsequent downstream processing). integration with either substrate supply or product removal may overcome the limited overall productivity. Table I. This phase is circulated over the extraction device. or controlled supply from a solid carrier.Introduction reaction phase can overcome the limited productivity. the product is transferred to a second phase to keep the product concentration in the reaction phase low. however. (1995). Biotransformation characteristics Low solubility of substrate in water Problem Low product concentrations Possible solution Introduce a second phase (organic solvent) that acts as a large substrate reservoir Keep aqueous substrate concentration low by its controlled supply Substrate inhibition/ Low production rates toxicity Product inhibition/ toxicity High product Keep aqueous product concentration concentrations reduce the low by its in situ removal into a second production rate phase Postpone attainment of equilibrium by direct product removal into a second phase Unfavorable thermo. after Vermuë et al.

Qureshi and Maddox. 10 . the reaction mixture is recirculated over the column and the product is adsorbed onto a solid adsorption material. for example. Boon et al. The product removal device can also be an adsorption column. as both processes can be controlled and optimized independently. Operation strategy to overcome limited productivity.g. Clogging of the membrane or adsorption column by microbial cells can also be a serious problem. True integration of the biotransformation and product removal may overcome the disadvantages of dual vessel recirculation systems. Integration of both processes in a single vessel will result in a reduction in the amount of process equipment used.g. From an operational point of view this recirculation type of integration may be advantageous. This latter problem can be solved by immobilizing the cells. Integration can also be achieved by physically separating the bioreactor and product removal device. A disadvantage is the extra amount of equipment. the reaction mixture is continuously recirculated over the product removal device. who describe the production of 3-methylcatechol using octanol as the organic solvent). Hüsken et al. (2000). The product removal device can be a membrane.Chapter 1 Biotransformation Product Removal Organic Solvent Figure 1. Various examples of this latter type of integration have been presented (e. described the production of oligosaccharides and their selective adsorption onto activated carbon. 2002. and the relatively high recirculation rate between biotransformation vessel and separation device in order to keep product or substrate concentrations low. 1995).. the reaction mixture is recirculated over the membrane. True integration of the biotransformation and product removal can be realized by executing both processes in one vessel. and at the other side of the membrane an organic solvent is used to collect the product (e.

1992). An obvious disadvantage is the higher degree of complexity. (1996) for the production of L-tryptophan from indole and L-serine. they studied the deacylation of penicillin G. thus yielding a solid-solid-liquid 3-phase system. In both cases. (1995) for the degradation of tetratlin. Liquid-impelled loop reactor (A) and 3-phase fluidized-bed bioreactor (B) In a 3-phase reactor such as the liquid-impelled loop reactor (Figure 2A). 2) A bioreactor and an extraction spray column can be integrated. A B Organic solvent droplets Gel beads Figure 2.Introduction Besides. the organic solvent was used as a reservoir from which an inhibitory substrate was supplied to the aqueous reaction phase. Such a design was studied extensively by Van der Wielen (1997). but takes place instantaneously. and by Mateus et al. Both authors used a liquid-impelled loop reactor (Figure 2A) as a basis for their integrated system (Van Sonsbeek. transfer of product is not delayed. Organic solvent droplets rise through the reaction mixture that contains immobilized biocatalyst particles. and fewer degrees of freedom for process control and optimization: the two processes have to be controlled and optimized as a single system. As a model reaction. Both types of solid are fluidized in the same vessel. Such a solid-liquid-liquid 3-phase system was studied by Vermuë et al. Two examples of true integration of biotransformation and product removal could be: 1) The biocatalyst is immobilized in a solid carrier and another solid carrier is used for the adsorption of the reaction product. only a single phase (the organic solvent) is recirculated in a controlled way with an external 11 .

Chapter 1 pump. Gas production and subsequent slugging necessitated the use of a tapered column to prevent wash-out of the gel beads. with organic solvent droplets rising through this bed. Objective The objective of the study presented in this thesis is the physical behavior of a liquidliquid-solid 3-phase fluidized-bed bioreactor. the velocity of the organic phase influences directly the circulation velocity of the aqueous phase (Van Sonsbeek. In this bioreactor. 1990). It focused on the hydrodynamic behavior of the 3-phase system. As a basis. i. the most prominent operational difference between a 3-phase liquid-impelled loop reactor (Figure 2A) and a 3-phase fluidized bed (Figure 2B) is the way in which the aqueous phase is recirculated: a 3-phase fluidized bed features an extra external pump.e. A 3-phase fluidized-bed bioreactor was studied by Davison and Thompson (1993) for the production of butanol from glucose. So. the other pump recirculates the aqueous phase. It features a stationary fluidized bed of gel beads that are not subject to (external) recirculation. Hydrodynamics of this 3-phase system were not studied. These authors focused on the biotransformation. In a liquid impelled loop reactor. for a 3-phase fluidized bed more control and optimization possibilities are available with respect to the hydrodynamics. phase hold-ups. Hence. hold-up and mixing. Also for 3-phase bioreactors mass transfer. and they reported an increase of 50-90% in the butanol production rate. as two external pumps are applied. Optimization of the hydrodynamics in relation to the overall production of the desired product will result in a maximal performance of the bioreactor. and to obtain stable operation of the fluidized bed. A different type of 3-phase liquid-liquid-solid bioreactor is depicted in Figure 2B. two external pumps are used for recirculating a single liquid phase each: one pump recirculates the organic solvent. thus offering an additional control option. the 12 . and mixing characteristics are largely determined by the velocities of both liquids. both liquid velocities may be chosen independently of each other. The circulation velocities of the other two phases (aqueous phase and solids phase) are not controlled directly but are dependent on the density difference between riser and downcomer. Performance of a bioreactor is dependent on its hydrodynamics. In a 3-phase fluidized bed.

13 . the droplet hold-up in a 2-phase liquidliquid extraction spray column is reported on for different sparger lay-outs. Continuous phase mixing in the 3-phase fluidized bed is discussed in Chapter 5. The thesis concludes with a general discussion on the current status of the newly developed 3-phase fluidizedbed bioreactor. and for co-current and counter-current operation. Performance of this 3-phase fluidized bed is compared to its 2-phase counterpart. Another objective was to identify the conditions for which this new type of bioreactor would perform superior to a conventional bioreactor.Introduction hydrodynamics of the constituent 2-phase systems were studied: a liquid-solid fluidized bed and a liquid-liquid droplet spray column. the preceding results are combined into a design for a 3-phase liquidliquid-gel bead (solid) fluidized-bed bioreactor. the fluidization behavior of different types of gel beads is presented for a 2-phase liquid-fluidized bed. The results of hold-up experiments in a 3phase liquid-liquid-gel bead (solid) fluidized bed are presented in Chapter 4. In Chapter 6. In Chapter 3. Outline In Chapter 2.

14 .

Though. whether present in a packed bed or liquid-fluidized bed. Two hypotheses are suggested to explain this lower drag force. was well described. Hence. (1991) predicts the voidage well. the model of Grbavcic et al. the gel bead surface might show waterlike properties. We modeled the drag coefficient of a single gel bead in a packed bed or a fluidized bed. The drag coefficient of a single gel bead. However. Submitted for publication 15 . gel beads contain over 95% of water and can in that sense be regarded as ‘rigid’ water droplets. The second one attributes the smaller drag force to the nature of the gel beads. This opens to a better design and understanding of a packed bed and fluidized bed of gel beads. Three different but related aspects involving the drag force between particle and flowing liquid were studied for 5 types of gel beads.2 Fluidized-Bed and Packed-Bed Characteristics of Gel Beads Abstract A liquid-fluidized bed or packed bed with gel beads is attractive as an immobilizedcell bioreactor. It is concluded that the drag force between gel bead and flowing liquid is lower than that for conventional solids. differing in diameter and density. (1983). using the equation of Foscolo et al. pressure drop over a packed bed and voidage in a liquid-fluidized bed. The first one attributes the drag reduction to small amounts of dissolved polymer. The performance of such bioreactors is influenced by the physical behavior of these beads. these aspects were not correctly predicted by established models. These aspects were the terminal settling velocity of a single gel bead. For gel beads the same trends and characteristics of these aspects were observed as for conventional solids.

with the same diameter and voidage. e. Willaert et al.. In this paper pressure-drop experiments with different kinds of gel beads are described. like lead shot. predicts the data correctly. As little is known in literature on the hydrodynamics of a liquid-fluidized bed of gel beads as well as the hydrodynamics of a packed bed of gel beads. cannot be applied to predict the minimum fluidization velocity. The more fundamental model of Grbavcic et al. (1991)... like κ-carrageenan. glass pearls. 1995. These beads.. 0. This velocity is determined by the fact that pressure drop equals specific buoyant weight of the bed. 1996). alginate or agar (Hulst et al.7 for the packed-bed to zero for wash out of particles. applied in fluidized-bed bioreactors.. Gel beads differing in diameter and density were used. 1996). can be used for typical bio-processes like the production of ethanol (Gòdia et al. 1987). However. Consequently. or lactic acid (Wang et al. using independently determined parameters. We will show that pressure drop over a packed bed of gel beads is much lower than the pressure drop over a packed bed of conventional solids. To predict this gel bead hold-up several models are available in literature. the Ergun equation or the equation of Foscolo (Foscolo et al. the pressure drop equations in literature.. the data can be well predicted with the model of Wilhelm and Kwauk (1948) with parameters obtained by fitting. etc. ranging ca. In a packed bed upward liquid velocities up to the minimum fluidization velocity can be applied. The solids hold-up. The solids hold-up for different kinds of gel beads was determined as a function of the superficial velocity. we studied the behavior of these systems. making a high biocatalyst concentration possible in a bioreactor (Wijffels et al. 1983). is determined by the equilibrium between the buoyancy force and the drag force experienced by a single particle in such a fluidized bed. 1995). An elegant way of immobilization is to embed the catalyst in beads of natural gels. It will be shown that the empirical model of Wilhelm and Kwauk (1948) with parameters according to Richardson and Zaki (1954) does not correctly predict the gel bead hold-up between minimum fluidization velocity and terminal settling velocity. A liquid-fluidized bed can exists between the minimum fluidization velocity and the terminal settling velocity of a single particle. 16 .Chapter 2 Introduction A packed bed or liquid-fluidized bed has attractive characteristics for application as an immobilized-cell bioreactor (Gòdia and Solà.g. Cells can be easily immobilized in solid particles. 1985)..

solids hold-up in a fluidized bed and terminal settling velocity. In the second one it is hypothesized that the observed feature results from a different surface structure of the gel beads: in a sense they can be regarded as ‘rigid’ water droplets. Two hypotheses are suggested to explain this feature. 17 . Gel beads experience a smaller drag force than conventional particles with the same diameter and density. Many models predicting this terminal settling velocity for more conventional solids are available in literature. The common feature of pressure drop over packed bed. In the first one it is considered that drag reduction occurs by small amounts of dissolved polymer. but all these models underestimate the terminal settling velocity for gel beads. is the fact that their magnitude is reigned by the drag force experienced by a particle.bed of gel beads The terminal settling velocity of different kinds of gel beads was also measured.2-phase liquid fluidized.

Table I gives an overview of the different gel beads used. kept at 35 °C. Using plain tap water without any salt addition would result in the dissolution of the gel beads. was pressed through a nozzle. (1993). and the applied molarity of the salt solutions. 1985). All κ-carrageenan gel beads were produced using a resonance nozzle as described by Hunik et al. The coloring of the surrounding liquid 18 . The temperature in each experiment was equal to the ambient temperature.Chapter 2 Materials and Methods Different gel beads and salt solutions were used. the beads were stored in a KCl-solution.25 % κ-carrageenan filled with 10% Celite™ 8 % alginate filled with yeast cells 25 v% Salt solution 10 mM KCl 10 mM KCl 30 mM KCl 30 mM KCl 50 mM CaCl2 Characterization of the beads Diameter The diameter was determined with image-analysis (CCD-camera with a Nikon macro objective 50 mm). Different particles used for fluidization.. Table I.000.000 g mole-1). The image of the particles was digitized and analyzed with the software package Genius (Applied Imaging). To obtain spherical beads a butyl-acetate (Aldrich-Chemie) layer was brought upon the hardening solution. The specific aqueous κ-carrageenan (Genugel 0909 Copenhagen Pectin Factory) solution. Gel bead A B C D E 3% κ-carrageenan 3% κ-carrageenan 3. Mw=2. The drops were collected in a 80 mM KCl-solution for hardening of the beads. 26±2 °C. Such salt solutions are needed to prevent the elution of counter-ions from the gel beads. pressure drop and terminal settling velocity experiments.25 % κ-carrageenan filled with 5% Celite™ 3. For a higher contrast the fluid surrounding the beads (a 10 mM KCl-solution) was colored with blue dextran (Pharmacia Biotech. After hardening for approximately two hours. The alginate beads filled with yeast were prepared with a conventional dripping method (Hulst et al.

this amount of water attached. again the flask was weighted (=M3). Next the flask was filled with the blue dextran solution up to the calibrated volume of the flask. κ-carrageenan beads without any additions. 19 . The following procedure was used for determining this amount. the adsorption of the supernatant and the adsorption of the original blue dextran solution were determined with a spectrophotometer (Ultrospec 2000. has to be accounted for. Pharmacia Biotech) at 280 nm. A calibrated flask was weighted (=Mflask) and partly filled with a blue dextran solution and weighted again (=M1). The next equations show how from this procedure the exact volume of the gel beads can be determined: M 1 − M flask + M 3 − M 2 ρ water(T) Abs solution Abs supernatant Vwater added = Vwater total = Vwater added ⋅ 1 Vgel beads = Vflask − Vwater total The density of the gel beads is calculated with the next equations: M water adhering = (Vwater total − Vwater added ) ⋅ ρ water(T) M gel beads = M 2 − M 1 − M water adhering ρ gel beads = M gel beads Vgel beads 2 Diffusion of blue dextran into the gel beads was not observed after incubating the solids in a blue dextran solution for 48 hours. Volume and Density Gel beads. Sieved gel beads were added. as a consequence of their nature. do have a certain amount of water attached that cannot be removed by simply sieving. as they are transparent. Simultaneously the bead density can be calculated. and the flask was again weighted (=M2). In all experiments that rely upon the exact volume of the gel beads used.bed of gel beads was used for gel beads A and B. After shaking for over 2 hours.2-phase liquid fluidized.

The first 10 cm of this column were filled with 0. Figure 1 shows the experimental lay-out. To check whether wall effects are important. Therefore the terminal settling velocity of bead A was also determined in a rectangular vessel with dimensions of 43. Wall effects might become important as the particle to column diameter ratio is not extremely large.5 x 100. The terminal settling velocity of the beads followed from the slope between time and height.8 cm glass pearls to provide a good flow distribution. and a total height of 1 m filled with a salt solution were measured.Chapter 2 Terminal settling velocity The split-times at different heights of a particle falling in a glass column with a 6 cm inner diameter. Fluidization of gel beads Experiments were carried out in a glass column with an inner diameter of 6 cm and a total height of 3 m. A vertical distance of 50 cm from the liquid interface was used to allow a constant velocity to be attained.56 cm. Pressure drop of a packed bed of gel beads A packed bed of gel beads was formed in a column with an inner diameter of 2. the height of the bed could be determined visually. As the fluidized bed showed a quiescent behavior. Liquid was pumped from a storage vessel to the glass column. Brooks). Brooks). The first 5 cm of the column was filled with glass beads to provide a good flow distribution. and a total length of 120 cm.5 cm (l x b x h). The solids hold-up followed from the bed height and the initial volume of the gel beads. 20 . The flow rate was measured with one or more calibrated rotameters (Sho-rate. The pressure difference over a vertical distance was measured with a Validyne DP103 pressure transducer together with a CD23 digital transducer indicator (maximum pressure difference 140 Pa). This slope was determined with linear regression.5 x 29. a liquid manometer filled with water and octanoic acid (ρ=903 kg m-3) was used for this calibration. Flow rates were measured with a calibrated rotameter (Sho-rate. pressure drop experiments were also done for bead D in a 6 cm inner diameter column. In each set of pressure measurements the pressure transducer was calibrated.

2-phase liquid fluidized- bed of gel beads

5

4 6 1 2 3

Figure 1. Experimental lay-out for the fluidization experiments. 1 pump, 2 rotameter(s), 3 bed of glass pearls, 4 sieve plate, 5 gel bead fluidized bed, 6 storage vessel

The predicting quality of a model is expressed as the average deviation (AVD) defined as:   AVD =  ∑ X measured − X model X measured  N   i i i i  with Xi a model quantity, and N the number of data points.

3

21

Chapter 2

Results
First physical parameters of the different gel beads are discussed: shape, diameter and terminal settling velocity. This settling velocity is compared with literature models. Next the data of the pressure drop experiments for a packed bed are shown and compared with literature model predictions. Thereafter, we show the results of the fluidization experiments. An empirical model (Wilhelm and Kwauk, 1948), and a more fundamental model (Grbavcic et al., 1991) are tested.

Gel-bead characterization Shape and diameter The size and shape of the different gel beads were evaluated with image analysis. Two important chord sizes for the shape characterization of the gel bead are the object length (OL in insert of Figure 3), defined as the largest distance between two points on the perimeter of the object, and the object breadth (OB in insert of Figure 2), defined as the largest distance between two points on the perimeter of the object, perpendicular to the object length. Figure 2 shows a chord size distribution for both the object length and object breadth typical for all gel beads analyzed. This figure shows that the object-length distribution is shifted a little to higher values compared to the object-breadth distribution. This indicates that the beads were not completely spherical. This deviation from a perfect sphere was visually verified as most of the gel beads showed a droplet like shape.
20

Frequency (%)

16 12 8 4 0 0.7

OL OL OL OB

OB OB

1.2

1.7

2.2

2.7

Chord size (mm)
Figure 2. Object length ( ) and object-breadth ( ) distribution for gel bead A.

22

2-phase liquid fluidized- bed of gel beads

Two shape factors could be determined with the image analysis software: the elongation and the object circularity, see equation 4.

Elongation =

object length object breadth

object area Circularity = perimeter

4

These shape factors are given in Table II together with the standard deviation. It can be concluded that the beads approach the spherical shape closely. Thus an equivalent spherical diameter, based on the measured area of the gel bead, was calculated. These diameters yielded the Sauter mean diameters in Table II.
Table II. Physical parameters of the different gel beads mentioned in Table I. Gel bead A B C D E Size and Shape Mean (±σ) Median mm mm 1.90 (0.28) 1.99 2.99 (0.45) 3.03 3.14 (0.17) 3.14 2.90 (0.42) 2.83 4.25 (0.24) 4.27
i

Sauter mean* mm 1.97 3.12 3.16 2.76 4.27

Shape factor 1* 1.03 0.98 1.01 1.02 1.00

Shape factor 2* 1.18 1.10 1.06 1.15 1.13

Density kg m-3 1007.4 1005.4 1029.8 1065.1 1039.9

* definition of the Sauter mean d 32 =

∑ di

3

∑ di
i

2

,

shape factor 1: circularity, shape factor 2 elongation.

Terminal Settling Velocity The mean and median of the settling-velocity distribution for the different types of gel beads are given in Table III. As the mean and median values are almost equal, it is concluded that the distribution is symmetrical for each type of gel bead. The terminal settling velocity is the largest liquid superficial velocity for which a fluidized bed exists. So, it is a key parameter in modeling fluidization. As the fluidization experiments were done in a 6 cm inner diameter column, the terminalsettling-velocity experiments were also done in this column. However, the ratio between column diameter and gel bead diameter is not large, and consequently wall effects will be present. In order to compare our measurements with literature models, these wall effects have to be accounted for. Therefore, the terminal settling velocity

23

Chapter 2

for gel bead A was also measured in a large rectangular vessel, in which wall effects are negligible. The mean terminal velocity was equal to 1.518 (± 0.065) cm s-1, the value between brackets is the standard deviation. The ratio of the terminal settling velocity with and without wall effects is equal to 0.90. This value is close to the value calculated with the correlation suggested by Richardson and Zaki (1954) to account for wall effects: v ∞ with wall effects = 10 − dp v ∞ without wall effects
D − 2.02 60

= 10

= 0.92

This correlation is used to correct the measured terminal settling velocity of the other gel beads for wall effects. This corrected settling velocity is also given in Table III.
Table III. Terminal Settling velocity of the different gel beads mentioned in Table I in cm s-1. Gel bead A B C D E Mean (±σ) 1.37 1.96 4.41 5.12 5.22 (±0.08) (±0.25) (±0.17) (±0.24) (±0.11) Median 1.38 2.02 4.44 5.12 5.22 Corrected for wall effects 1.52 1 2.22 4.98 5.69 6.15 Model of Dallavalle 0.99 1.42 3.49 4.89 5.62

1 Experimental

value

The settling velocity of a single particle in a stagnant liquid can be calculated from a force balance for this particle. This force balance accounts for the gravity force, the buoyancy force and the drag force. This balance reads in dimensionless form:
Ga =
3 4

C d Re p 2

5

where Ga is the Galileo number, Cd is the drag coefficient for a single particle and Rep is the particle Reynolds number, which is based on the particle dimensions. To calculate the terminal settling velocity with equation 5, a model for the drag coefficient is needed. Models relating the drag coefficient to the particle Reynolds number are abundantly available in literature (Dallavalle, 1948; Turton and Levenspiel, 1986; Hidaka et al., 1994). Other models relate the Galileo number directly to the particle Reynolds number (Zigrang and Silvester, 1981; Turton and

24

(1983) revised the Ergun model by 25 . All these literature models predict more or less the same terminal settling velocity. 1948. 1989.5) and the drag coefficient calculated with literature models. 1981. A commonly used model to predict pressure drop over a packed bed is the model of Ergun (Bird et al. 1989). and the maximum pressure drop is highest for the gel-bead type with the largest density. Khan and Richardson. calculated with the equation of Dallavalle (1948). 7 6 5 4 3 2 1 0 0 50 100 150 200 250 Cd Rep Figure 3. Table III shows the terminal settling velocity for the different gel beads. Experimental values and model predictions: data . Table III shows that the calculated terminal settling velocity is considerably lower than the measured terminal settling velocity of the gel beads corrected for wall effects. 1987. Foscolo et al. Drag coefficient for gel bead particles settling in stagnant water as a function of the particle Reynolds number (physical properties given in Table I). a smaller particle diameter gives a higher pressure drop at the same Reynolds number. ─┼─ Model of Turton and Clarke. Hartman et al. This figure clearly shows features commonly observed in packed-bed pressure-drop experiments: pressure drop increases with an increasing Reynolds number.bed of gel beads Clark.. ▬▬ model of Silvester and Zigrang. Figure 3 shows the drag coefficient calculated from the experimental values for the settling velocity (eq. 1960). ─ ─ model of Dallavalle. 1987. 1979). From this figure it can be concluded that published models overestimate the drag coefficient for gel beads. Packed-Bed Pressure Drop Figure 4 shows the measured pressure drop over a fixed bed length for the different gel beads as a function of the Reynolds number (Re = ρUdp/η).2-phase liquid fluidized. using the physical properties of the gel beads given in Table II and the physical properties of water at 24 °C (Weast.

Column diameter is 2. gel bead C. Pressure drop over a packed bed of gel beads as a function of the Reynolds number (see Table I for physical properties of the different gel beads). 400 (Pa m ) ∆P/L 300 200 100 0 0 5 10 15 20 -1 Re Figure 4. which results in: ∆P  ε 3  d p   L  1 − ε  ρU 2   Cd = 4 3 6 The drag coefficients derived from Ergun's and Foscolo's model are given by equation 7.3 Foscolo C = 4  + 0.336  ε −1. Appendix A gives a detailed derivation. gel bead A. This drag coefficient can be related to the pressure drop. gel bead D. gel bead B.Chapter 2 introducing a voidage dependent tortuosity.ε)  7 26 . This model predicts drag forces acting on an individual particle. A good measure for this drag force is the drag coefficient. either present in a packed bed or settling in a stagnant liquid.75  d 3  Re (1 . Both models overestimate the pressure drop over the packed bed of gel beads (predictions not shown).54 cm.8  d 3  Re   Ergun C  100  = 4 + 1. see for details Appendix A: 17.▲ gel bead E The large discrepancy between measured and predicted pressure drop can be elucidated by considering the drag force experienced by a single particle in the packed bed.

Drag coefficient of a single particle in a packed bed of identical particles as function of the hydraulic Reynolds number. is shown in Figure 5 as a function of the hydraulic Reynolds number ( Re h = 2 3 (1 − ε)ρUd p η ). As the particle diameter to column diameter ratio is less than 10 for gel beads B and C.0 Reh 10. Foscolo. a wall effect might by expected. 27 . ▲ gel bead D. Indeed.. column diameter 2. a small wall effect is observed for gel beads D.2-phase liquid fluidized.0 Figure 5. …. This figure clearly shows that the drag experienced by a single gel bead in a packed bed is smaller than the drag predicted by Ergun's and Foscolo's equations.54 cm. the Cd values for data corresponding with the 6 cm diameter column are a little lower than those for data corresponding with the 2.bed of gel beads The experimental drag coefficient for gel beads A and D. together with the model predictions calculated with equation 7. column diameter 6 cm. Experimental data: gel bead A. Model predictions: ── Foscolo. Therefore the drag coefficient for these gel beads is not shown in Figure 5. For gel bead A this ratio is larger than 10.25.38. 300 200 Cd 100 0 0.1 1. calculated with equation 6.54 cm diameter column. ▬▬ Ergun. voidage is 0. For gel bead D this ratio is smaller but close to 10. voidage is 0.. gel bead D.

ε is the voidage. The solids hold-up as a function of the superficial velocity is shown in Figure 6.2 0 0 1 U 2 (cm s ) -1 3 4 Figure 6. Experimental data for particulate bed expansion were correlated by Wilhelm and Kwauk (1948) with the following empirical equation: U = kε n 8 where U is the superficial liquid velocity. ◊ gel bead C.Chapter 2 Fluidization of gel beads The different gel beads mentioned in Table I.8 0. and k and n are empirical parameters. Table IV gives the parameters in equation 8 determined from fitting this equation to the experimental data. Data and description with the Wilhelm and Kwauk model (for fitted parameters see Table IV). 0. at the same U gel beads with the highest terminal settling velocity show the highest solids holdup.4 0.6 εs 0. ▲ gel bead D. at the same U the solids hold-up is higher for gel beads with the same diameter but larger density (type C versus type B). Figure 6 shows that the experimental data are well described. Gel bead hold-up in a fluidized bed as a function of the superficial velocity. gel bead A. O gel bead B. were fluidized by the appropriate salt solutions. □ gel bead E Features commonly seen in liquid fluidization of solid particles are also observed for the fluidization of gel beads: the solids hold-up decreases with an increasing superficial liquid velocity (U). 28 .

67 (±0. Grbavcic et al.10 (±0. As the voidage approaches 1.40 (±0. Extrapolation of this straight line to ε=1 results in a ratio U/v∞ smaller than 1. because the bed height could not be clearly determined. In our data this second regime is hardly present.03) 2. For higher voidages. Gel bead A B C D E 1 Fitted Parameters1 k n 1.61 2.08) 2.35 (±0. 29 . Fitted parameters in Wilhelm and Kwauk’s model for liquid fluidization of different gel beads. U/v∞ . see Figure 7.05) 1.77 (±0. the voidage as a function of U/v∞ shows also a straight line on a log-log scale but the slope is smaller. but extrapolation gives a lower value.04) n from Rowe’s model 3. This kind of expansion behavior is one of the four different types of liquid-fluidized-bed expansion behavior mentioned by Di Felice (1994).. For lower voidage.08) 2. 1977. it is apparent that the parameter k is always smaller than the terminal settling velocity. the superficial velocity should approach the terminal settling velocity.31 (±0.2-phase liquid fluidized. the voidage as a function of the ratio superficial velocity to terminal settling velocity.29 (±0.33 (±0. 1977.03) 3. If we compare parameter k in Table IV with v∞ in Table III.06) 4. Riba and Couderc.07) 2.16 (±0. The observed expansion behavior is characterized by two different regimes. 1991).06) 3. We did not do experiments at higher voidages. shows a straight line on a log-log scale.29 (±0. see the dotted line in Figure 7. This fact has also been reported by different authors for liquid fluidization of glass pearls (Garside and Al-Dibouni.79 2.06 2.55 value between brackets gives the 95% confidence interval Richardson and Zaki (1954) showed that the parameter k is equal to the terminal settling velocity and that the power n is a function of the particle Reynolds number. the extrapolation of this line to ε=1 gives a ratio equal to 1.bed of gel beads Table IV.61 2.04) 3.

1991 (Umf = 0. provided the empirical constants are known.75 1 + 0. the fitted parameter n is smaller than the calculated parameter. the 30 .358 cm s-1.7 0. v∞ = 5.4 0.20 U/v∞ 0. (1991). 1948 with fitted constants.38). As parameters this model uses only the minimum fluidization velocity (Umf).model of Grbavcic.38. except type B.70 1.9 ε 0.4112 ⋅ Re p 0. model of Wilhelm and Kwauk. εmf = 0.08 0.76 mm.8) is the more fundamental model of Grbavcic et al. So. The relationship between voidage and superficial velocity as described by equation 8 does predict our data satisfactorily. An alternative to the Wilhelm and Kwauk approach (eq. These authors describe a model predicting the voidage in liquid fluidized beds without any adjustable constants. dp=2. 1987): n= 4.1 0.0 Figure 7. Voidage for gel bead D as a function of the dimensionless velocity. Table IV shows the fitted parameter n in equation 8.175 ⋅ Re p 0.5 0.6 0.8 0.Chapter 2 1. the parameter n in equation 8 cannot be estimated from literature models in order to predict the gel bead hold-up.0 0.75 9 For all gel beads. O experimental data.7 + 0.50 0. and parameter n calculated with a well-established model from literature (Rowe.12 cm s-1. εmf = 0. The superficial velocity (U) was divided by the terminal settling velocity of gel bead D (v∞). However. model of Grbavcic 1991 conventional solid (Umf = 0. ρs=1065 kg m-3). v∞ = 4. in the preceding paragraphs it was demonstrated that these constants cannot be determined from literature models.18 cm s-1. ── ---.4 cm s-1.30 0.

2-phase liquid fluidized- bed of gel beads

voidage at the onset of fluidization (εmf) and the terminal settling velocity; parameters that are easy to measure. The equation is:
 ε 3 ( 1 − ε) C d , mf U=U  mf  ε 3 1 − ε C  d mf  mf     
0.5

(

)

10

The drag coefficient ratio is given by:
  ε−ε d = 1 − c + 1 ⋅ 1 −  λ  mf 2 λ   1−ε C d, mf mf     C
2

  +c   1  

    

0.5

11

The constants, c1, c2, and λ, in equation 11 are functions of the minimum fluidization velocity, the voidage at the onset of fluidization, and the terminal settling velocity, see Nomenclature for the corresponding correlation. An interesting aspect of this model, equations 10 and 11, is the fact that it uses as one parameter a single particle characteristic, the terminal settling velocity. The other two parameters are packed bed characteristics: packed bed voidage at the onset of fluidization and the minimum fluidization velocity, which is equal to the maximum velocity through a packed bed. The packed bed voidage has to be determined experimentally, regardless of the kind of particles used. Literature models are abundantly available for adequately predicting v∞ and Umf for conventional solids. However, as shown in preceding paragraphs, these models fail to predict v∞ and Umf for gel beads. These parameters, however, can be easily determined. The predicting quality of this model, equations 10 and 11, expressed as the AVD (eq.3), is given in Table V for the different gel beads mentioned in Table I. This table also gives the minimum fluidization velocity, the voidage at this velocity, the terminal settling velocity of the different gel beads, and the corresponding constants c1, c2, and λ. As a typical example, Figure 7 shows the measured voidage, and the voidage predicted by equations 10 and 11 with the constants given in Table V, as a function of the ratio of superficial velocity to v∞ for type D gel beads. Considering that the AVD of all gel beads in Table V is less than 0.05, we can conclude that the model of Grbavcic et al. (1991) predicts the experimental data well. This conclusion is supported by Figure 7.

31

Chapter 2

Two models, equation 8 and Grbavcic’s model, were used to describe the gel bead hold-up as a function of the superficial liquid velocity. Figure 7 shows that both models describe the experimental data well, although Grbavcic’s model slightly overestimates the voidage in the intermediate regime. The advantage of equation 8 over Grbavcic’s model is the less complicated mathematical functionality. However, fluidization data are necessary to get the empirical constants. For application of Grbavcic’s model estimates of terminal settling velocity, minimum fluidization velocity and the voidage at this velocity suffice.
Table V. Constants for calculating the voidage in a liquid fluidized bed as described by equation 10 and 11, together with its predicting quality expressed as the average deviation (AVD). Bead Characteristics Gel bead A B C D E εmf Umf Ut εmf 0.25 0.35 0.31 0.38 0.32 Umf 0.065 0.080 0.216 0.358 0.40 Ut 1.38 2.02 4.44 5.12 5.22 Constants in equation 11 10C1 9.90 9.99 9.96 9.96 9.83 -100C2 16.4 3.8 8.6 9.5 23.1 -10λ 8.50 9.63 9.18 9.10 7.98 0.040 0.050 0.019 0.022 0.018 AVD

: voidage at minimal fluidization velocity : minimal fluidization velocity cm s-1 : measured terminal settling velocity cm s-1

General Discussion
In this section we show that the drag coefficient for a gel bead in a fluidized bed is smaller than that of conventional solid with the same density and diameter. Next it is concluded that a smaller drag coefficient is observed in three related aspects, terminal settling velocity, pressure drop over a packed bed, and voidage of a liquid fluidized bed. Two hypotheses are discussed to explain the difference between gel beads and conventional solids. This sections ends with the derivation of a model to predict the drag coefficient of an individual gel bead either present in a packed-bed or a fluidized bed of identical gel beads. The drag force experienced by a particle in a fluidized bed of identical particles depends on the voidage in such a fluidized bed. Appendix A shows how the drag coefficient is related to the voidage:
32

2-phase liquid fluidized- bed of gel beads

C drag =

4 ρs − ρ ε3 d pg 3 ρ u2

12

So, to compare the drag coefficient for a gel bead and a more conventional solid particle the voidage has to be known. For the gel beads we used the measured voidage. For the conventional solid particle, the voidage was calculated with the model of Grbavcic et al. (1991). This model is superior to all other literature models, as discussed above. As the model of Grbavcic et al. (1991) uses the minimal fluidization velocity, we used Foscolo’s pressure-drop equation for calculating this velocity. The terminal settling velocity, the other parameter, was calculated with Dallavalle’s model (1948). Figure 7 shows the voidage in the fluidized bed of conventional solids as a function of a dimensionless superficial velocity (the superficial velocity divided by the terminal settling velocity of gel bead D). Obviously, the voidage is always smaller for the gel beads at an equal velocity ratio, and consequently their drag coefficient is also smaller. Above, it has been shown that the drag coefficient is smaller for a single gel bead settling in a stagnant liquid, see Figure 3. The same conclusion has been drawn for the drag coefficient in a packed bed, see Figure 5. Thus, for the three related aspects it is concluded that the drag coefficient for a gel bead in an array of identical gel beads is smaller than that of a conventional solid, e.g. a glass pearl. Two hypotheses are discussed below to explain this observation. The observed difference between gel beads and conventional solids might be explained by the following hypothesis. It is known that extremely low amounts of dissolved polymer can reduce the drag force between a solid surface and a flowing liquid (Paireau and Bonn, 1999). In our experimental set-up small amounts of polymer, κ-carrageenan, will be present in the liquid. Consequently, the drag force between gel bead and flowing water is reduced. A different hypothesis comes from considering the nature of these gel beads. Gel beads are made by dripping a solution of a natural polymer like κ-carrageenan or alginate into a salt solution; the liquid droplet transforms to a solid due to exchange of ions. Consequently, gel beads consist over 95% of water and in a sense can be considered as ‘rigid’ water droplets. Water streaming alongside the gel bead surface will experience a water-like surface. Consequently, the drag force is less, compared to the case in which water flows alongside a non-water-like surface. In the next part of this section a model will be derived that describes the drag
33

Chapter 2

coefficient of a gel bead in an array of other identical gel beads, whether present in a packed-bed configuration or in a fluidized bed. Most literature models do not predict the drag coefficient for gel beads correctly, except the model of Grbavcic et al. (1991) which predicts the drag coefficient in fluidized beds. Unfortunately, this model cannot predict the drag coefficient for a gel bead in a packed bed of gel beads. As the drag coefficient is related to the pressure drop over a bed of particles, see appendix A, a pressure-drop relation is used to model the drag coefficient for gel beads. Pressure drop is well correlated by the Ergun equation (Bird et al., 1960). In the derivation of this equation it is not required that the particles are in contact with each other. Therefore, this equation should also be applicable to a fluidized bed with a voidage ranging from the packed bed condition to the fully expanded state. Unfortunately the Ergun equation does not describe the pressure drop over a particulate fluidized bed well (Foscolo et al., 1983). Foscolo et al. (1983) made a reexamination of the basis for the Ergun equation, which resulted in an equation describing pressure drop over a packed bed as well as a fluidized bed. Incorporating this pressure drop equation in equation 6 resulted in equation 7 for the drag coefficient. This equation does not predict the drag coefficient for gel beads correctly, see Figure 8. We used the form of this equation and introduced two unknown parameters:

c −c C d = 4  3 ε 4 + 0.336ε −1.8    3  Re 

13

Equation 13 consists of two parts: a part describing the drag coefficient at laminarflow conditions and a part describing this coefficient at turbulent-flow conditions. Under turbulent flow conditions the dominant contribution to energy dissipation in a bed of particles comes from the numerous expansions and contractions of the fluid streams in their passage around the particles. As fluid streams do not follow the surface of the particles strictly in the turbulent regime, we believe that in this regime the surface structure does not influence the energy dissipation. In the laminar-flow regime, however, fluid streams do follow the surface of the particles. So the structure of the surface will influence the drag force between flowing liquid and a single particle in a packed or fluidized bed. Consequently, the turbulent part of the model of Foscolo, equation 13, was used unchanged but the constants appearing in the laminar part of the model of Foscolo were fitted to the experimental data.

34

especially when the sensitivity of the model for this constant is considered. a constant c4 of 0. (1983) with fitted constants. Table VI shows that constant c3 for gel bead A is almost equal 35 . In this column with a small diameter a wall effect was observed for gel beads B.024.5 results in a AVD of 0. • Model of Foscolo et al. Parity plot for the drag coefficient of a single gel bead in a bed of identical gel beads. C and D. but constant c4 does not differ much for gel beads A – D. The resulting constants are given in Table VI. For gel bead D pressure-drop experiments carried out in a 6-cm inner diameter column were extended above the minimal fluidization velocity. ∆ gel bead C. Therefore. we minimized the average deviation (AVD). a common voidage dependency was assumed for gel beads A to D. For example. O gel bead D. The fitted constants are given in Table VI. but also the drag coefficients calculated from the fluidization experiments. with Xi is the drag coefficient. This common voidage dependency was implemented into fitting equation 13 to the experimental drag coefficient. Model of Foscolo et al. Not only the drag coefficient calculated from the pressure-drop experiments were used. (1983) with original constants.bed of gel beads 100 Cd predicted 10 1 1 Cd experimental 10 100 Figure 8. a wall effect was not observed.2-phase liquid fluidized. for gel bead B. see also appendix A. So for gel beads B and C. pressure drop over a fluidized bed is straightforwardly correlated to voidage. ◊ gel bead A. Figure 8 shows that the experimental drag coefficient is well described by equation 13. To find the constants c3 and c4. and the constants from Table VI based on the common voidage-dependency. we only used the fluidization experiments done in a 6cm inner diameter column to fit the constants in equation 13. gel bead B. Table VI shows that constant c3 differs considerably for each gel-bead type. These experiments were used also to fit the constants in equation 13. equation 3. For gel bead A pressure-drop experiments done in a small column could be used.

55 0.079 Common voidage dependency C3 C4 AVD 16. Gel bead A B C D E Each gel bead type individually fitted C3 C4 AVD 16.77 0.43 14 The influence of the voidage on the pressure drop is much smaller for gel beads than for conventional solids. pressure drop over a packed bed.073 166. Although the experiments with gel beads show the same trends and characteristics as previously reported for conventional solids. the pressure drop is lower.00 0.96 2.D: conventional solid : ∆P ~ (1 . pressure drop and terminal settling velocity are not correctly predicted by established models.549 0. Constants for equation 13 describing the drag coefficient of a single particle in a packed or fluidized bed of identical particles.0235 0.95 0. We have no explanation for this difference.571 0.05 0.84 0.8 gel beads A . and voidage of a liquid fluidized bed. as well as the average deviation (AVD) of the model from the experimental data.19 0. For a packed bed and fluidized bed the effect of the voidage on the pressure drop is obtained by combining equations 6 and 7 for conventional solids and by combining equations 6 and 13 for gel beads A . but for the other gel beads this constant is increasingly larger.051 95. The terminal settling velocity is higher.571 0.Chapter 2 to the value in the original model of Foscolo.073 For gel bead E the constants given in this table are fitted on the fluidization experiments as the pressure drop experiments could not be satisfactorily fitted on equation 13.D : ∆P ~ (1 .ε) ⋅ ε −4.571 0.46 0.104 33.075 54.571 0.47 0.ε) ⋅ ε −2.085 25.041 92. Voidage of a liquid fluidized bed is also not well predicted by the empirical 36 .45 -0. equation 7. Table VI. Conclusions Three different properties have been measured for five different types of gel beads: terminal settling velocity of a single gel bead.004 66. The fitted constants for gel bead E are quite different from the fitted constants for gel bead A – D.85 0.

shows a different but common voidage dependency for the gel beads. So. a different model is presented for predicting the drag coefficient of a gel bead present in a packed bed or fluidized bed of other identical gel beads. Unfortunately.. The first considers a lower drag coefficient by the presence of small amounts of dissolved polymer. This work was financially supported by the Ministry of Economic Affairs. The laminar part. this model is not applicable for a packed bed. and hence the voidage in a liquid fluidized bed.2-phase liquid fluidized. 37 . a gel bead consists over 95% of water and can be regarded as a ‘rigid’ water droplet.bed of gel beads model of Wilhelm and Kwauk with parameters according to Richardson and Zaki.. whether settling or present in a packed bed or fluidized bed of other identical gel beads. however. derived for more conventional solids. predicts the voidage well. It can be concluded that the drag coefficient for a single gel bead. the Ministry of Education. Nature Management and Fishery in the framework of a research program of the Netherlands Association of Biotechnology Centers (ABON). The other considers that the difference between gel bead and conventional solid results from the difference in surface structure between gel beads and more conventional particles. The more fundamental model of Grbavcic et al. The drag force experienced by a gel bead in a liquid fluidized bed. the Ministry of Agriculture. is always smaller than the drag coefficient of a conventional solid having the same diameter and density. is well predicted for the different gel beads by the model of Grbavcic et al. Two hypotheses are presented to explain the features above. however. This model is based on the model of Foscolo (1983) for conventional solids. Culture and Science. using a measured minimal fluidization velocity and a measured terminal settling velocity. The turbulent part is equal for gel beads and conventional solids. Acknowledgements We thank Arend Mulder for supplying the alginate beads.

6 Pressure drop over a packed bed due to friction between fluid and particles can be described 38 . . and consequently the hydrostatic pressure. and no work (W) is done by the surroundings on the fluid:  p 2 dp  φm ⋅ ∫ + g (h 2 − h 1 ) + E v ⋅ φ m = 0  p1 ρ  A.3 The force exerted by the fluid on a single particle.4 relates the drag coefficient for a single particle in a bed of identical particles to the pressure drop: 3 P1 − P2 = ρE v + ρg∆h = 4 C drag ρρu 2 1 − εε L + ρg∆h εε 3 d p A.4 Combining equation A.3 and A. is cancelled.1 This energy balance simplifies to equation A. 1969):  p 2 dp 1 φm ⋅ ∫ + g (h 2 − h 1 ) + 2 v 2 p1 ρ  ( 2 − v1 2 ) − W + E   v ⋅ φm = 0 A. So equation A. A steady-state macroscopic mechanical-energy balance from point 1 to 2 reads (Bird et al. this gives the total power dissipated. This v ⋅ N p = E vρ uA Np = L(1 − ε)A 3 1 6 πd p 1−ε L ε3 dp u=ε v 3 1 E v = 2 C drag 2 u 2 A. as the velocity of the fluid at point 1 equals the velocity at point 2 (continuity).2 (the fluid density is supposed to be independent of the pressure) results in: (P2 − P1 ) + ρg(h 2 − h1 ) + ρE v = 0 dissipated power is equal to Ev φm (Stammers et al.1) the pressure difference between point 1 and 2 is measured. ρ g∆h ..Chapter 2 Appendix A.5 simplifies to: P1 − P 2 3 1−ε 1 = 4 C drag ρu 2 3 L ε dp by the Ergun equation: A.2 with Ev the amount of mechanical energy irreversibly converted to thermal energy due to friction.1986): 1 1 Fdrag ⋅ v ⋅ N p = C drag 4 πd p 2 2 ρ v 2 A. Solving the integral in equation A.5 In the set-up (Figure A. is multiplied with the velocity of the fluid and the total amount of particles present. Relationship between the drag coefficient of a single particle in packed bed or fluidized bed of other identical particles and the pressure drop Water flows through a bed of particles.2.

bed of gel beads ∆P 150ηu (1 − ε)2 ρu 2 1 − ε = + 1.6 gives the drag coefficient of single particle in a fluidized C drag = 4 3 ρs − ρ ε3 d pg ρ u2 A.6 can be combined with equations A.9 and A.336   L  Re  dp − 4. ∆P  17.8 At a voidage of 0.75 L dp2 ε3 dp ε3 A.3 Cd = 4  + 0.7 or A.2-phase liquid fluidized. (1983).7 Pressure drop can also be described with the equation of Foscolo et al. Liquid Fluidization The pressure drop in liquid fluidization is equal to the specific buoyant weight of the bed (Foscolo et al.8 A. These authors used the same basic elements but a porosity dependent tortuosity instead of a constant tortuosity.4 both models give the same pressure drop.8 to give an equation for the drag coefficient for a single particle in a packed bed of other identical particles: Foscolo Ergun 17.3 ρ U2 (1 − ε)ε = + 0.336 ε −1.10 39 .9 Combining equation A.8   3  Re  Cd = 133 + 2. Equation A. 1983): ∆P = (1 − ε) ⋅ (ρ s − ρ )g L bed of identical particles: A.3 Re h 2 3 with Reh the hydraulic Reynolds number defined as (1 − ε)ρUd p η .

Chapter 2 Nomenclature c1 c2 c3 c4 Cd dp g Ga k L n ∆P Re U v∞ : constant in equation 11 : constant in equation 11 : constant in equation 13 : constant in equation 13 (1 + (U /(v ε )) ) (1 .c ) − c ) mf 2 ∞ 2 mf 3 2 −0.5 1 : drag coefficient for a single particle : particle diameter : gravity constant : Galileo number : constant in equation 8 : length : constant in equation 8 : pressure drop : Reynolds number : superficial liquid velocity : terminal settling velocity ρ c Ud η d p 3 (ρ s − ρ c )ρ c g η2 m N m-2 ρc v ∞d η m s-1 m s-1 Rep : particle Reynolds number Greek ε η λ ρ ∆ρ : voidage : viscosity : constant in equation 11 : density : density difference kg m-1 s-1 (1 − c ) 1 2 0.c ) /((1 .5 1 2 0.5 m m2 s-1 1 2 0.5 − c1 kg m-3 kg m-3 Subscript mf s c : at minimal fluidization conditions : gel bead : liquid 40 .

41 .

42 .

in order to support the development of an extractive fermentation process. A model that predicts the dispersed-phase hold-up in a spray column regardless of the mode of operation is presented. A Richardson and Zaki type equation is used for the slip velocity. The proposed model incorporates the nozzle geometry. This model uses the slip velocity between the continuous phase and the dispersed phase to determine the dispersedphase hold-up. Different sparger geometries were used. Validation of the model was obtained with a 60 mM KCl solution/dodecane 2-phase system. influenced the hold-up to a large extent. the water flux had a minor influence on the dodecane hold-up. The model describes the hold-up data for the water/dodecane system well. The nozzle geometry. This 2-phase system differs only in surface tension. Water and dodecane were introduced in cocurrent and counter-current flow. however. We present a new model for the power n in the Richardson and Zaki model.3 Dispersed-phase hold-up in a liquid-liquid extraction spray column Abstract Phase hold-ups were studied in a spray column with a water/dodecane 2-phase system. the sparger with the lowest number of nozzles showed the highest hold-up at the same dodecane flux. Although water and dodecane fluxes were of the same order of magnitude. the physical properties of the 2-phase system and the fluxes of the continuous and dispersed phase. 43 . in which n depends on the dimensionless drift flux.

1996). This system can be a suitable reactor set-up for biotransformations using immobilized biocatalysts and applying in situ extraction.Chapter 3 Introduction Kinetic or thermodynamic product inhibition in bioconversion processes can be reduced by integrating the conversion with an extraction process in one apparatus. 1995). 1954) for solid fluidization can be extended to predict the hold-up of the different components in a multi-component fluidized bed (Van der Wielen et al. This holds especially for solid biocatalysts immobilized in a gel matrix (Wijffels et al. This article presents a model that predicts the dispersed-phase hold-up in a 44 . This reactor set-up can offer high biocatalyst hold-up as well as high mass transfer rates. 1991. The (bio)catalyst can be freely suspended or immobilized in a solid support. This paper describes the work on the spray column.. In order to understand the behavior of the 3-phase system we have studied the behavior of the spray column with two liquids (the extractor) and a fluidized bed with one liquid and suspended solids (the reactor) separately. has been demonstrated for the prediction of the gas hold-up in a liquid-solid-gas fluidized bed (Chen and Fan. 1994). 1990). We believe that the prediction of droplet hold-up in such a 3-phase system can be based on the prediction of droplet hold-up in a spray column. a choice has to be made between systems with and without moving internals. The protection against direct contact with extractant droplets is an important advantage of immobilized biocatalyst systems (Vermuë and Tramper. columns without moving internals are more appropriate for immobilized biocatalysts. With regard to the extraction. We are currently developing a fluidized bed with extractant droplets rising through the bed of biocatalytic solids. Our first aim is to predict the hold-up of the two liquids and the solids phase in this system. Therefore. the models resulting from these studies cannot be extended straightforwardly to models predicting the dispersed-phase hold-up in the above-mentioned 3-phase fluidized bed. 1994). Godfrey and Slater. but are likely to cause inadmissible damage to the solid biocatalysts. so-called in situ extraction. Such an extension of a 2-phase model to a 3-phase model predicting the dispersed-phase hold-up. Richardson and Zaki's model (Richardson and Zaki. 1995). as these are important variables with respect to the conversion and extraction rates. Although the dispersed-phase hold-up in spray columns has been studied several times (Kumar and Hartland. Moving internals enhance the extraction process (Godfrey and Slater.

Ud: superficial velocity dispersed phase. Theory The dispersed-phase hold-up can be predicted according to a model presented below. Uc: superficial velocity continuous phase. In this paper we will propose a new model for this power n. This figure shows how the combination of a definition of the slip velocity and a model for this slip velocity can be used to predict the dispersed-phase hold-up. d: droplet diameter.4) physical properties v∞ (eq.2) hold-up Figure 1. This model can be easily extended to predict the dispersed-phase hold-up in a 3-phase system. n: power n in equation 2.Droplet spray-column spray column regardless of the mode of operation. The slip velocity is described with Richardson and Zaki's model for a fluidized system of mono-sized particles. 45 . the terminal rise velocity of a single droplet and a power n.13 ⇒ n (eq.3) Uc Ud hold-up < 0. This model uses two parameters. Schematic representation of the model for the prediction of the dispersed-phase hold-up.1) ≡ model (eq. A schematic overview of this model is shown in Figure 1. and hence the dispersed-phase hold-up. sparger lay out (eq. viz. Droplet Hold-ups are less than 0.23. v∞ : terminal rise velocity of a single droplet.13 ⇒ n = constant hold-up > 0.8) Ud Uc Ud d Slip velocity Definition (eq. The slip velocity between the continuous phase and the dispersed phase is used to determine the dispersed-phase hold-up. The terminal velocity can be calculated with models from literature. analogous to models presented in literature. This figure also shows how the various operating variables (superficial velocity of both phases and lay-out of the sparger) as well as the physical properties of the 2-phase system influence the slip velocity. but models for determining the power n in liquid-liquid 2phase systems have not yet been published.

in most cases equal to the terminal rise velocity of a single droplet (v∞).α)n 2 with vk a characteristic velocity. The power n indicates the influence of liquid flow around a droplet. 1995). Upward flow is taken as the positive direction for both flows. Different relations for slip velocity can be found in literature (Kumar and Hartland. Godfrey and Slater. also a model is needed to predict this diameter (see also Figure 1). Empirical models that predict directly the dispersed-phase hold-up have been reviewed (Kumar and Hartland. 1995. based on physical properties of both phases. Terminal rise velocity Godfroy and Slater (1994) give a selection chart. The selection is mainly based on the Morton number of the 2-phase system.Chapter 3 Slip velocity The slip velocity (v12) is defined as the difference between the velocities of both phases taking into account the direction of flow: v 12 r r U sd U sc = − α 1− α 1 with Usd is the superficial velocity of the dispersed phase. The relationship between slip velocity and hold-up is implicit: the dispersedphase hold-up is determined by the slip velocity and the slip velocity depends on the dispersed-phase hold-up.2). models for the terminal rise velocity and power n are needed. Godfrey and Slater state that the best equation describing the slip velocity is the equation of Kumar and Hartland (Kumar and Hartland. these authors demonstrated that those empirical models are less accurate than slip velocity models. 1991). However. with the Morton-number 46 . To calculate the slip velocity (eq. 1985). for choosing the best-suited terminal rise-velocity equation. Godfrey and Slater (1991) also concluded that most of the equations describing slip velocity can be summed up to one equation based on Richardson and Zaki's equation for fluidized systems of mono-sized rigid particles: v 12 = v k (1 . As the terminal rise velocity depends on the droplet diameter. Usc is the superficial velocity of the continuous phase and α is the dispersed-phase hold-up.

We have studied a water/dodecane 2-phase system. Eonoz is the Eötvös number based on the nozzle dimensions.0393 We noz 0. At a of a nozzle.315 4 where dnoz is the nozzle diameter.2  ρ c    2 /3  ρc    η   c 1/3  1 − Eo    6   3 where d is the droplet diameter. based on the selection chart. ρd is the density of the dispersed phase. Above a nozzle Weber number of 8. ρc is the density of the continuous phase. ∆ρ is the density difference between water and dodecane. 47 . The 2 ρc σ 4 4 surface tension between water and dodecane is high.73  d noz    σ   −0. ηc is the viscosity of the continuous phase.64 has been derived by Kumar and Hartland (1996): d= d noz  ρ gd 2  0. which results in a low Morton number. single drops are formed at the tip ρ u noz 2 d noz . Eo is the Eötvös-number defined as g∆ρd 2 .33 + 0. An overview of different droplet formation mechanisms is given elsewhere (Dalingaros et al.. σ representing the ratio of the buoyancy force over the resistive force defined by the interfacial tension.55Eo noz 0.Droplet spray-column defined as Mo = g ∆ρ ηc .64. g is the gravity constant.64 the liquid jet starts to atomize and a broad droplet-size distribution is formed. At a nozzle Weber number less than 2. This relation was used here for calculating the droplet diameter. A single relation covering nozzle Weber numbers up to 8. Therefore. 1986). The Weber σ number represents the ratio between inertial force and resistive interfacial force. Vignes' equation was used: d  g∆ρ    v∞ = 4. Droplet diameter Liquid droplets can be made by pumping liquid through one or more nozzles. with the nozzle Weber number defined as nozzle Weber number between 2 and a critical nozzle Weber number 8. drops are formed by uniform break-up of the liquid jet existing in this range. and σ is the interfacial tension between water and dodecane.

we will show that these models for the power n do not describe the slip velocity for a water-dodecane 2-phase system accurately. In that case relations for the power n derived for water and solid particles can be used. the dispersed-phase hold-up is also not well predicted. It is concluded by Godfrey and Slater (1991) that this power n has a value between 0 and 4.2). and the presented equations for droplet diameter (eq. An overview of different relations and their accuracy can be found elsewhere (Hartman et al. has not yet been published for liquid-liquid 2-phase systems. liquid droplets can be regarded as solid particles. Applying the total model as given in Figure 1. the experimental dispersed-phase hold-up is well described.4) and terminal rise velocity (eq. for reasons of clarity this model is integrated in the Results and Discussion section.Chapter 3 Power n A model predicting the power n used in the slip velocity model (eq. As slip velocity and hold-up are closely related. the new model for the power n. see Figure 1. In this study we will propose a new model for the power n.. i. However. 48 .3). 1994). the combination of equation 1 and 2.e. Data from a water/dodecane 2-phase system with a different surface tension are well predicted by the model presented in Figure 1. When the interfacial tension is high.

Tap water was introduced at the bottom of the column in co-current operation and at the top in counter-current operation. 2 phase separator.Droplet spray-column Materials and Methods Wageningen tap water was used as continuous phase and n-dodecane (Aldrich. 99%+ pure) as dispersed phase. 6 tap water storage vessel. On top of this column a phase separator was placed with a diameter of 180 mm. 1 column. 3 sparger. temperature could be maintained at 25±1°C by circulating water of appropriate temperature. it was introduced in all experiments at the bottom of a 2 m high and 60 mm internaldiameter column equipped with sample ports to make measurements at different heights. As water and n-dodecane are highly immiscible. As the n-dodecane is lighter than tap water. Experimental set-up for co-current operation. Both tap water and ndodecane were introduced through a sparger at the bottom of the column. phase separation was easily achieved for the applied fluxes. As the storage vessels were jacketed. 4 rotameters. 5 dodecane storage vessel. 2 7 1 5 4 3 4 6 Figure 2. No temperature gradient was observed between top and bottom of the column. 7 inverse water manometer 49 . The separated phases at the top of the column were sent back to the storage vessels to allow recirculation of both liquids. Co-Current Operation Figure 2 shows schematically the experimental set-up.

The flows through these openings were adjusted with throttle valves in such a way that they equaled each other. The distance between both cones was constant. This distributor consisted of an inner and an outer cone. that the growing droplets did not touch each other. 3 sparger. The second adaptation was made on the tube used for the water introduction at the top of the column. see Figure 3c1 and 3c2. except for the introduction and recirculation of the tap water. and both ends were directed to the outer wall and bottom of the annulus made up between column and enlarged column. This tube was split in two parts. Both cones had a 10° angle and were filled with glass beads to provide a good flow distribution. The first adaptation was made in the glass column wall. The liquid distributor is schematically shown in Figure 3a. 4 nozzles Figure 3b. 4 nozzles. diameter 10 mm. Alternative liquid distributor in counter-current operation. 4 }3 4 5 }3 2 1 2 1 2 Figure 3a. which consisted of a number of nozzles. and the tap water through the outer one. located in one plane at 90° relative to each other. These openings were 1 cm above the liquid distributor. the inner cone had a 30 mm maximum diameter. The nozzles were evenly distributed over the area of a circle with a 30 mm diameter. which were hammered through a Teflon plate. 1 dodecane inlet. Figure 3c gives details of the changes made in the co-current experimental set-up to make counter-current water recirculation possible. 1 dodecane inlet. In this way plug flow was assured through the 50 .Chapter 3 Liquid Distributor. 5 glass pearls. 2 water inlet. Liquid distributor in co-current operation. with dodecane flowing through the inner cone. the wall had four circular openings. see also Figure 3c3. On top of the inner cone a sparger was placed. 2 water inlet. 3 sparger. An overview of the different spargers used is given in Table I. The maximum diameter of the outer cone was 60 mm. Figure 3c2. see also Figure 3c3. Figure 3c1. The nozzles were made of hollow stainless-steel tube with a length of 3 cm. The distance between the nozzles was arranged in such a way. Counter-Current Operation The same experimental set-up was used as described in the co-current operation part.

This liquid distributor consisted of a sparger with 209 nozzles (5 cm length and 1 mm diameter). diameter 6 cm. With these colorpulse experiments a flat front was still observed at the bottom of the column. see Figure 3b.4 1.0 1. Although only the dodecane storage vessel was maintained at 25±1°C by circulating water of appropriate temperature.0 only in counter-current operation 51 . Details on adaptations made on the co-current experimental set-up for counter-current operation Liquid Distributor.0 1. This was visually verified by color-pulse experiments. Different spargers used in experiments. 2 1 2 1 3c1 3c2 3c3 Figure 3c.0 1. Table I. The same liquid distributor was used for dodecane dispersal as described with the co-current operation. Also another sparger was used. On top of a cone a 10 cm high glass column. sparger 1 sparger 2 sparger 3 sparger 4 sparger 51 1 # nozzles 12 23 37 54 209 nozzle diameter (mm) 1. The cone had an angle of 10° and a maximum diameter of 60 mm. a temperature gradient was not observed in the column. Both cone and column were filled with glass beads to provide a good flow distribution over the nozzles that were placed on top of the glass column. was placed.Droplet spray-column column. evenly distributed over the area of a circle with a 60 mm diameter.

From these data (not shown) it can be concluded that an axial hold-up profile is not significant for any of the combinations. Radial profiles may exist (Ueyama and Miyauchi. In this figure experimental data are shown for spargers 3. To increase the accuracy. for the sake of simplicity. However.7 cm s-1 and the dodecane flux ranged from 0 to 1. Dodecane holdup measurements have also been done with a slightly different 2-phase system: a 60 mM KCl solution as continuous phase and the same dodecane as dispersed phase. The average deviation (AVD) between measurement and model is defined as: AVD = ∑ α meas − α mod α meas N−1 5 Results and Discussion The dispersed-phase hold-up was measured for different sections of the column and different combinations of dodecane and water fluxes. a uniform distribution of the droplets across the column diameter is assumed here. These measurements were used to validate the new model.25 cm s-1. 1979). The static interfacial tension between continuous phase and dispersed phase was determined with the Wilhelmy-plate method. Zero water flux The effect of the sparger lay-out on the measured dodecane hold-up is shown in Figure 4.Chapter 3 Measurements The hold-up of n-dodecane at different flow rates of tap water and n-dodecane was determined with a reversed U-tube water manometer. These measurements were used to determine the parameters in the model for the power n.and counter-current hold-up data. Hold-up increases with an increasing superficial dodecane velocity. 4 and 5 at zero water flux. the water manometer was filled with hexanoic acid (ρ=921 kg m-3 ). The same trends were observed for the co. The flux of tap water ranged from 0 to 0. The fluxes of tap water and dodecane were measured with calibrated rotameters (Sho-Rate). Figure 4 52 .

Droplet spray-column clearly shows that the increase in hold-up is larger for a sparger with a lower number of nozzles. then the true velocity equals the rise velocity of a single droplet.006 0. the superficial velocity goes beyond the velocity corresponding with the critical nozzle 53 . Dodecane hold-up versus dodecane superficial velocity for sparger 3 to 5 (Table I). Above this nozzle Weber number atomization of the liquid jet starts.009 -1 0. 0. and the true velocity (vd) of the dodecane phase.16 0. At an equal superficial dodecane velocity there is a difference in droplet diameters generated by the different spargers.003 0. A smaller number of nozzles per sparger will result in a higher velocity through one nozzle. Moreover.64. as the diameter depends on the dodecane velocity through one nozzle (see also the section on the droplet diameter). This holds especially at dodecane fluxes larger than 3 mm s-1. a broad droplet-size distribution will occur. ● sparger 3. Figure 5 shows data for spargers 1 and 2. as it can not be achieved by a single-stage phase-separator that is based on gravity. 54 nozzles. ▲ sparger 5.04 0 0 0.12 α 0. Figure 4 shows data up to a nozzle Weber number of 8. phase separation becomes less straightforward. In this figure.08 0. If it is assumed that the droplets do not influence each other. 37 nozzles.012 Ud (m s ) Figure 4.3) and hence a higher hold-up. and the slip velocity is not described by equation 2. ◊ sparger 4. and hence a smaller diameter. A smaller diameter results in a lower terminal rise velocity (eq. contrary to Figure 4. 209 nozzles The difference between these spargers can be explained by considering the onedimensional continuity equation for the dodecane phase: Ud = α vd ⇒ α = Ud vd So the dodecane hold-up (α) is defined by the ratio between the superficial velocity (Ud).

4) and terminal rise velocities (eq.4 mm and 0. Figure 5 shows that for sparger two at each superficial dodecane velocity. Density kg m-3 998 742. Physical Properties of the liquids applied at 25±1°C. The true dodecane velocity is almost constant above a certain value of the superficial dodecane velocity. and 2. This Weber number is indicated by the vertical lines in Figure 5. Ud α . In Figure 5 this velocity is shown as a function of the superficial dodecane velocity. with a diameter of 1. So above the critical nozzle Weber number the true dodecane velocity is independent of the hold-up. calculated from the critical nozzle Weber number. Below the critical nozzle Weber number this is explained in exactly the same way as above: at a dodecane superficial velocity of 3 mm s-1 the velocity through one nozzle equals 0. Obviously the relationship between hold-up and superficial dodecane velocity becomes different above the critical nozzle Weber number. the hold-up is expected to be higher for sparger 2.470 m s-1 for sparger 2 (23 nozzles.103 cm s-1 for sparger 2.5 24 Surface tension2 mN/m 40 Tap water Dodecane 1surface tension against water. Table II.3). 54 .123 cm s-1 for sparger 1. and 0. At this superficial velocity the droplet diameters (eq.00093 0. the hold-up is equal or larger compared with data for sparger 1. are 3 mm and 0. Consequently.7 Viscosity Pa s 0. (The physical properties used for calculating droplet diameter.459 m s-1 for sparger 1 (12 nozzles.Chapter 3 Weber number. with a diameter of 1 mm). and terminal rise velocity are given in Table II). This means that the holdup is linearly proportional to the superficial dodecane velocity. For both spargers this value corresponds with the critical superficial dodecane velocity.0012 Surface tension1 mN/m 68-63.4 mm). as indicated by the dotted lines in Figure 5. 2surface tension between tap water and dodecane At a zero water flux the slip velocity reduces to the true velocity.

4 mm. respectively. This can be illustrated by comparing the true water velocity with the true dodecane velocity: at a superficial water velocity of 7.14 ⋅ −1 = 5.05 0.09 0. Under the conditions tested the influence of the superficial water velocity is marginal. ▲sparger 2.002 0. Obviously. ● sparger 1. when the dodecane flux is changed at a constant water flux (data not shown). i. Superficial water velocity is zero.006 0. 1 mm. At higher superficial dodecane fluxes.07 0.8 55 .51   0. i.and counter current operation.14  0. and a superficial dodecane velocity of 7. At low superficial dodecane fluxes. With these values the true dodecane velocity becomes 5. no effect of the water flux can be observed. Co.and counter-current water flux The above-described characteristics for dodecane hold-up at zero water flux are also observed for co.200 0.15 0.e.100 0. there is a small effect of the water flux. For co-current operation an increase in water flux resulted in a slightly lower hold-up at the same superficial dodecane flux.e. low hold-ups.Droplet spray-column v12 (m s-1) 0.000 0. higher hold-ups. whereas for counter-current operation a slightly higher hold-up was measured.e. 23 nozzles.51 mm s-1.008 Ud (m s-1) Figure 5.8 times higher than the true water velocity: U d ⋅  Uc    α 1−α −1 =  7. i.or co-current operation. this is straightforwardly explained by the reduction or increase of the true dodecane velocity at counter. i. 12 nozzles. 1.86  7.004 0.100 7 5 2 0.010 7 5 0.14 mm s-1.14 was measured in counter-current operation using sparger 4 (54 nozzles).12 α 0. a dodecane hold-up of 0. Dodecane hold-up (closed symbols) and slip velocity (open symbols) versus dodecane superficial velocity for sparger 1 and 2.e.

with this velocity determined from the experimental hold-up. however. This model. However. can be easily extended. Obviously. especially at low hold-ups.53 +  d v 12 ρ c    6 However. at the higher applied water fluxes satellite-formation was observed. Modeling droplet hold-up The droplet hold-up can be predicted from an accurate slip velocity model (Kumar and Hartland. Figure 1 shows how from such a model the droplet hold-up can be predicted. this model can not be extended easily to systems with more than two phases.56 α 0. flooding will occur at a certain water flux. 1985): 4 3  24 ηc   d g ∆ρ (1 − α) = ρ c v 12 2 (1 + 4. Less nozzles on a sparger give a higher hold-up at the same superficial dodecane flux. As these satellites form emulsions. However. Udc (adapted from Wallis. Also.64. the water flux cannot be increased unlimitedly. the relative error in the hold-up is large. This scatter largely disappears when the slip velocity is transformed to a drift flux. and consequently there is large scatter in the slip velocity. the dodecane flux corresponding with a nozzle Weber number of 8.64 is lower compared with a sparger equipped with more nozzles. Under our conditions the water flux only slightly influenced the dodecane hold-up.73 )⋅  0. analogous to the Richardson and Zaki model (1954).2). the power n can be obtained from fitting equation 2 to the slip velocity. and phase separation of emulsions is not straightforward.Chapter 3 Dodecane hold-up can be manipulated to a large extent with the sparger lay-out and to a minor extent with the mode of operation. A larger influence can be expected with increasing water flux. A slip velocity model (eq. Below a model will be proposed to predict this power n. This model holds for dispersed-phase velocities up to velocities corresponding with the critical nozzle Weber number of 8. depends on a model for the power n. 1969): U  U U dc =  d − c α ⋅ (1 − α) = v 12 α ⋅ (1 − α) = v ∞ α ⋅ (1 − α)n + 1  α 1−α 7 56 . However. in counter-current operation. for a sparger equipped with few nozzles. which has not been developed yet. In counter-current operation this would yield a higher hold-up.

Next. equation 8. Up to a hold-up of 0.13 the dimensionless drift flux is well described with a constant power n. The hold-up for which Udc* becomes constant.00 with a 95% confidence interval of 0. and is also shown in Figure 6.0975 with a 95% confidence interval of 0. is applicable for hold-ups up to αboundary. a definition and a model of the slip velocity transformed to a drift flux. the dimensionless drift-flux.07. as can be seen in Figure 6. a model for the power n.13. So. 1969). At higher hold-ups this model systematically underestimates the dimensionless drift flux. 57 .8. is shown in Figure 6. This figure shows that the dimensionless drift flux is virtually independent of the sparger lay-out.0004. The total model for predicting the dimensionless drift flux consists of two parts: • for hold-ups below αboundary. the model is represented by Figure 1: ⋅ ⋅ ⋅ nozzle lay-out and physical properties of the 2-phase system determine the terminal rise velocity of a droplet. and equal to 0. An underestimation of the dimensionless drift flux means that the power n in equation 7 is too large. For all the spargers applied. • for holds-ups above αboundary. It appears that the power n. defined as the drift flux divided by the terminal rise velocity of a single droplet. the drift flux is made dimensionless by dividing this drift flux by the terminal rise velocity (eq. is called αboundary. eq. eq.and counter-current operation.7. decreases monotonically with the dimensionless drift-flux. So a new model is proposed for the power n: U  n = n 0 − c 1  dc  v   ∞  c2 8 The dimensionless drift flux. is shown in Figure 6 as a function of the dodecane hold-up. the dimensionless drift flux is constant. Equation 7 was fitted to the dimensionless drift flux data up to a dispersedphase hold-up of 0.Droplet spray-column A drift flux represents the volumetric flow rate of the dispersed-phase relative to a surface moving at the volumetric average velocity (Ud+Uc) (Wallis. This figure shows that the dimensionless drift flux (Udc*) becomes constant for higher hold-ups. This gave a power n of 4. derived from the measured hold-up and equations 3 and 7.3). the new model for the power n. derived from experiments for co.

… total model with n=4 sparger 3. ◊ sparger 5. At each αboundary a corresponding nboundary is calculated.3 0.075 α boundary 0.0975). a corresponding power n can be calculated: this is nboundary. using a downhill simplex method (Nelder and Mead.predicted i ( ) 2 10 In appendix A it is shown that the parameters no and c1 cannot have an arbitrary value. c1.025 0 0 0.35 Figure 6. and c2.05 0. O sparger 1. ▲ Both parts of the model hold at the αboundary.0975)  n 0 − n boundary = c 1 ⋅ 0. gives the following relationship between the coefficients in equation 8.1 Udc * 0. Dimensionless drift flux versus dodecane hold-up.  total model with n = f(Udc).Chapter 3 0.0975 9 If coefficient c2 is determined with equation 9 at any given n0. sparger 4. with the hold-up equal to αboundary. but are subjected to some constraints. and the coefficients no and c1 are fitted to equations 3. which gives an additional equation for the coefficients n0.1 0. fitting was 58 .15 α 0. A clear value for αboundary cannot be determined unambiguously from Figure 6. ● sparger 2. c1 and nboundary both parts of the model for the dimensionless drift flux will hold at αboundary. Applying these constraints.25 0.05 0.2 0. Using equation 7.measured − α i. 1965): SSres = ∑ α i. 8 and 9 by minimizing the residual sum of squares of measured and predicted droplet hold-up (equation 10). Substituting this nboundary in equation 8. and the dimensionless drift flux equal to its constant value (0. Therefore αboundary has been varied on the interval 0.19…0. c2  n 0 − n boundary ⇒ c 2 = ln   c1   1 ⋅  ln(0. 7.24.

B : slip velocity model of Kumar and Hartland 1985. Coefficient c2 follows from equation 9 and is equal to 0.1 0.2 Measured α 0.1 0.2 Measured α 0. Parity-plot between measured and predicted dodecane hold-up for different models. D : slip velocity model with n according Richardson and Zaki 1954 A complete analysis of the confidence intervals of the fitted parameters is beyond the scope of this paper.2. 8 and 9 and the fitted constants. Predicted α 0.3 0. A : presented model.23.3 0.2 0. This figure also shows the relation between the parameters n0 and c1 that resulted from the constraint 59 . C : explicit model of Kumar and Hartland 1995. that the prediction of the model is good.Droplet spray-column successful with the next set of parameters yielding the lowest sum of squares of residuals: αboundary=0.55 and c1=33.3 B Predicted α 0.3 A 0. It is concluded from Fig 6.1 0.3 0.2 Measured α 0.88.3 Predictedα 0.2 Predicted α 0. Figure 8 shows contour lines at 1. the SSres was calculated at different combinations of parameters n0 and c1 for αboundary=0. n0=6.3 0.2 0.233. as well as the experimentally determined dimensionless drift flux. 7.2 Measured α 0.1 0 0 0.3 0 0 Figure 7.1 0 0 0.2 0. This is also supported by the parity plot in Figure 7.0244.1 D 0 0.1 0. which shows the dimensionless drift flux predicted with equations 3.1 C 0 0. 5 and 10 % higher values than the minimal SSres. to get some insight in the sensitivity of the model to its parameters. However. The SSres for the fitted parameters is equal to 0.

and for c1 these values are 21. as can be seen in Figure 8. At a high interfacial tension droplets indeed tend to behave like solids.13 a constant n of 4 suffices to describe the data (see Fig 6). all referring to the 5% contour line.8.5.5). This behavior results from the constraint violation. At the left. n=3. This is in agreement with the results of Godfrey and Slater (1991). the contours are discontinuous and behave not at all smooth. For values of n0 and c1 in the area above the constraint line. 60 . n=3. n=3.4 (0.1 (1. The interfacial tension between water and dodecane is 40 mN m-1. Foscolo (1987). see also appendix A.82 × fitted value) and 11. the end of this constraint line is (25. n=4. This power n also compares well with values of n derived from literature models for liquid fluidization of solid spheres at laminar flow conditions (Richardson and Zaki (1954). who concluded that this power is between 0 and 4. for lower c1-values the constraint function does not exist. For n0 the minimum and maximum values are 5. This Figure 8 ends at c1=60.66 × fitted value) and 60 (1.81 × fitted value). So the solidslike behavior of the droplets is to be expected.75 (0. as for higher c1 values the constraint is violated for the contour lines. For values of n0 and c1 in the area below the constraint line.09). the contours are continuous and smooth. Such an interfacial tension is regarded as high. Rowe (1987). The large range might suggest that the model seems insensitive to its parameters.7.Chapter 3 used in the fitting procedure. the dodecane droplets behave as solid spheres at laminar flow conditions. However. The 5 and 10% contour lines cover a broad interval for n0 and c1.71 × fitted value). Apparently. Garside and Al-Dibouni (1977). Up to a hold-up of 0. 8. at a given c1 the allowed range for n0 is limited.65.

but to an endless set of Reynolds numbers. and two column diameters. 61 . this transition corresponds with a certain dimensionless drift flux (0. Figure 1 shows how this can be achieved. Sensitivity analysis for parameter n0 and c1 of the presented model. instead of a single Reynolds number. --.2 (●). Contour lines of 1% (black). For this type of droplet flow. that will result in a hold-up of 0.constraint line The transition from a constant power n of 4 to a decreasing power n might indicate a deviation from laminar flow to more turbulent flow.Droplet spray-column Figure 8. This is in contrast to what is suggested in literature (Wallis. This table clearly shows that there is not a constant Reynolds number for the transition from a constant power n to a decreasing power n. 5% (dark grey) and 10% (grey) higher values than the SSres for n0=6. we calculated the superficial dodecane velocity at different nozzle geometries. 1969).13. To illustrate that this transition does not correspond to a single Reynolds number.065) or dispersed-phase hold-up (0. the results are shown in Table III.33 and c1=33.13).

3 3. Up to a hold-up of 5% the prediction is good. 62 .3 7. the liquid velocity through one nozzle to obtain a hold-up of 0. predicts the measured hold-ups well. This model is based on a force balance around a single droplet and a hold-up dependent drag coefficient to account for the presence of other droplets.Chapter 3 Table III.7 4. The results are given in Table IV.2 and 4. Droplet characteristics at the transition from laminar to more turbulent flow for different sparger geometries (nozzle diameter is 1 mm). Their prediction is expressed as the average deviation (AVD).120 0. It can be concluded from Table IV and Figure 7 that. and parity plots for the different models are shown in Figure 7 (the physical constants used are given in Table II). At higher hold-ups no solution for the hold-up is possible. only the slip-velocity model (Kumar and Hartland.0).0 11.1 2.6.2 d (mm) D2 2. as defined by equation 5.094 0. For hold-ups up to 15% the hold-up is well predicted with a slip velocity according to equation 2 and a power n according to Richardson and Zaki (1954).3 (the value of the power n obtained from fitting the hold-up data up to 13% is equal to 4. and two column diameters: D1=6 cm.64. So no calculations of terminal droplet rise velocity and droplet diameter could be done. eq.13 using a column with diameter D2 should be larger than the maximum velocity corresponding with a nozzle Weber number of 8.8 9. The good prediction of the Richardson and Zaki model is not surprising as the power n is between 4.3 9. D2=9 cm. at larger hold-ups the model slightly underestimates the experimental data.124 0.0 3. # nozzles 501 100 200 500 1 Ud (mm s-1) D1 D2 7.098 0. for the whole range of measured hold-ups.9 3. This model was derived for a large number of different 2-phase systems with a wide range of physical constants.145 0.147 D1 2.8 11.7 D1 254 425 607 764 Re D2 230 394 631 for this sparger. Comparison with literature models Different models from literature were evaluated with all the measured droplet holdups (N=912).0 v∞ (m s-1) D1 0.0 10. 1985).158 D2 0.

in contrast to 40 mN m-1 for the first 2-phase system. at higher hold-ups no solution of the model for the hold-up Good prediction of the data. 63 . Figure 9A shows the measured dimensionless drift-flux. At these higher hold-ups the model of Kumar and Hartland overestimates the dimensionless drift flux.Droplet spray-column Table IV. our model predicts the dimensionless drift-flux and hold-up better. Figure 9B shows the parity plot for predicted and measured droplet hold-up for both models. as well as the predictions of our proposed model.6. eq.2748 0.0797 An explicit model Kumar and Hartland (1995) Presented model 0. that both models predict the droplet hold-up well. 72% of the data show a deviation less than 10%. and 0. is 0. which is equal to 29 mN m-1. the continuous phase is a 60 mM KCl-solution instead of tap water and dodecane is again used as dispersed phase. Comparison of measured data with models from literature. except for the interfacial tension. 68% of the data show a deviation less than 10%. the AVD as defined by equation 5. however. a different 2-phase system was used to validate the newly proposed model. 95% of the data has a deviation less than 20% All data points are underestimated Droplets behave like solids Richardson and Zaki (1954) A slip velocity model Kumar and Hartland (1985) 0. It can be concluded from Figure 9A-B. The physical properties are equal to those for a tap-water/dodecane 2-phase system. At higher hold-ups. AVD 0. and underestimates the hold-up. Physical constants are given in Table II.0750 Good prediction of the data. and those from Kumar and Hartland (1985). as a function of the dodecane hold-up.115 for the presented model.119 for the model of Kumar and Hartland (1985). 97% of the data has a deviation less than 20% Validation of the presented model As mentioned in the experimental part.0728 Remarks Good prediction of the hold-up up to 15%.

The average deviation between data and model (=AVD) is for the model of Kumar and Hartland 0.06 0. a constant n of 4 in the slip-velocity model suffices to describe the hold-up adequately. The model of Kumar and Hartland (1985) is applicable for the whole range of measured hold-ups. At hold-ups higher than 0.06 0. This model uses the definition of the slip velocity and a Richardson and Zaki type equation for the slip velocity: an equation for the droplet diameter and an equation for the terminal rise velocity of a single droplet.23 the dimensionless drift flux becomes constant.13 the parameter n decreases with an increasing dimensionless drift. using the same superficial dodecane velocity. 1985.075. which predicts the dispersed-phase hold-up from zero to 0.03 A 0 0.09 0. or a dimensionless drift-flux of 0.15 the slip velocity model of Richardson and Zaki (1954) for solids predicts the data well. Parity plot between predicted and measured hold-up. O : model of Kumar and Hartland. Up to 0. ● : presented model.12 α measured Figure 9 A.05 Udc* 0.09 B 0. at low hold-ups the prediction is good. at higher hold-ups the model underestimates the hold-up. ● data.12 0.09 0.Chapter 3 0. B.065. but for higher hold-ups this model gives no solution.075 α predicted 0. A higher hold-up is obtained with a sparger with less nozzles and the same nozzle diameter. — presented model.03 0.13.flux. 64 . Continuous phase is a 60 mM KCl solution.model of Kumar and Hartland. A model for this parameter based on the dimensionless drift-flux is used in the hold-up model. --.12 Conclusion The sparger lay-out influences the dodecane hold-up.06 0.03 0. The presented hold-up model predicts well the dispersed-phase hold-up and related properties like slip velocity and drift-flux. Dimensionless drift flux versus dodecane hold-up.23 well. At hold-ups larger than 0.025 0 0 0 0.080. α 0. Up to a hold-up of 0. 1985. whereas our model gives an AVD of 0.

the Ministry of Agriculture. Culture and Science.0975   ln 1 − α boundary n boundary =  ln    α boundary     ( ) − 1    A. Acknowledgements This work was financially supported by the Ministry of Economic Affairs.1: f = U dc * − α ⋅ (1 − α) n(Udc *) + 1 = 0 A. except for the interfacial tension (29 mN m-1 for the new data versus 40 mN m-1 for the old data). Appendix A. the Ministry of Education.3   0.1 yields an implicit equation for Udc*. Nature Management and Fishery in the framework of a research program of the Netherlands Association of Biotechnology Centres (ABON).0975)  c1   with nboundary given by : A.5 The constants n0 and c1 were fitted with a downhill simplex method as suggested by Nelder 65 . Substituting equation A.4 Parameter c2 is constant at a given n0. all physical properties were the same. The new data were well predicted. Constraints applied in the fitting procedure The equations describing the dimensionless drift flux (Udc*) are: U dc = U dc * = α ⋅ (1 − α)n + 1 v∞ with n given by: A.1 n = n 0 − c 1 (U dc * ) c2 A.2 The parameter c2 is given by:  n 0 − n boundary  1 ⋅ c 2 = ln   ln(0.Droplet spray-column Our model was validated with measured data using a different dodecane/water 2-phase system. c1 and αboundary.2 into equation A. A new function f is defined by rewriting equation A.

At αboundary a discontinuity can be observed.1.2 shows f as a function of Udc* (equation A.1 0.5.5 still gives two roots for Udc*.08 0.09 0.19. This gave a value for n0 equal to 4. 0.001 -0. and for c1 equal to 685. --.0015 0. It is only demanded that Udc* is smaller than 0.0975) at αboundary. these fitted constants do not give a good fit at higher hold-ups. Udc*=0. c1 = 6.15 0.25 α 0.5.37. This figure shows that at a given α two roots exist for Udc* that satisfy equation A. None of these roots are near the constant dimensionless drift flux.3 0.0975. … hold-up is 0.Chapter 3 and Mead (1969) without any constraint on the fitting procedure.4 demands that the root of equation A. The smallest of both roots is taken as the correct solution of equation A.1 Udc* 0.15 0.11 Udc* Figure A. c1 = 5. This means that the derivative of function f with respect to the dimensionless drift flux is less than or equal to zero. Figure A.2 α 0.1 Udc* 0.  n0 = 685. hold-up is 0.0975. However.20.n0 = 69. equation A. f=0.0005 0 0.25 0.075 0.075 0. When α approaches αboundary equation A. The dimensionless drift flux as a function of the hold-up for different combinations of the parameters n0 and c1.35 Figure A. Udc* decreases at higher hold-ups and is estimated far too low.1 0. This explains the discontinuity in Figure A.1.21 .025 0 0 0.21. c1 = 4. we demand that the first root of the function f at αboundary is equal to the constant value of the dimensionless drift flux. function f 0. × hold-up is 0. The add-in figure is an enlargement. as shown in Figure A.05 0. Function f versus the dimensionless drift flux for different values of the hold-up  holdup is 0.37.2.  .2. n0 = 33. However.05 0.48.07 -0.55. In order to avoid the discontinuity.001 0.0005 -0.1.1.0015 0. 66 .5).5 is equal to the constant dimensionless drift flux (0.2 0.002 0.

This figure clearly shows that the fit is much better for the higher hold-ups.48 and c1=69.7 Equation A. This resulted in a n0 equal to 5.6 c2   ( c −1 ) df = 1 − α (1 − α) 1+ n 0 −c1 (Udc * )  ⋅ ln(1 − α) ⋅ − c 1 c 2 (U dc * ) 2 ≤0   * dU dc ( ) Equation A. and αboundary.Droplet spray-column From equation A. This feature of the relationship between the dimensionless drift flux and the hold-up has no physical meaning.21.8 is applied.8 If the numerical operation given in equation A. around a hold-up of 0. suggest a monotonic increase in drift-flux with the hold-up.6.21. another more severe constraint has to applied to the fitting procedure. This leads to: α≤ 1 n+2 67 .6 a maximum value for n0 can be determined at a given c1. The dimensionless drift flux as a function of the hold-up is given in Figure A. 1988) : d U dc * (1 − α)n + 1 − (n + 1) ⋅ α ⋅ (1 − α)n A.5 shows that Udc* is an implicit function of the hold-up.1. An obvious constraint would be demanding that the first derivative of Udc* with respect to α is equal or larger than 0: d U dc * ≥ 0 dα A.1 at αboundary = 0. equation A.9 = ) ( dα 1 + α ⋅ (1 − α)n + 1 ⋅ c1 c2 ⋅ U dc * c2 − 1 ⋅ln (1 − α) This derivative should be larger than or equal to zero. However. Our data.9 is obtained (Almering. however. together with the constraint given by equation A. the increase of the dimensionless drift flux with the hold-up is really large. and shows an inflexion point. U dc − α ⋅ (1 − α) * n +1 =0 d (U dc * ( α) − α ⋅ (1 − α)n +1 ) ⇒ =0 dα A. see the enlargement in Figure A. To obtain a relationship between the dimensionless drift and the hold-up which does not show the sharp increase.8). To obtain the first derivative of Udc* with respect to α implicit differentiation has to be applied (equation A.1.1 and A.2 were fitted on the data. df dU dc * d  U * − α ⋅ (1 − α) 1+ n 0 −c1 (Udc * )   dc =  d U dc *  c2       boundary ≤0 ⇒ A. So the nominator should be larger than or equal to zero.

At a higher αboundary dUdc*/ dα is negative. αboundary. The dimensionless drift flux as a function of the hold-up.2 were fitted on the data.55 and c1=33. this means that the denominator should not be equal to zero. together with the constraint discussed above (the denominator in equation A. This resulted in an n0 equal to 6.0975 n boundary + 2 Solving these equations gives a maximum for αboundary: 0.2 with the fitted constants. This figure shows that the obtained constants give a good fit. However. 68 .9 equals zero). Equation A.0975.Chapter 3 This equation gives no extra information for the parameters c1 and n0.9 a maximum n0 can be determined at a given c1.23. The first derivative should also exist.1 and A.2334.1. a maximum αboundary can be derived. α boundary 1 − α boundary α boundary = 1 ( ) boundary n +1 = 0. From the denominator of equation A. equation A.2 at αboundary = 0. At the boundary the following two equations hold. is given in Figure A. and Udc*=0.1 and A.

8) : number of data points : power n : parameter in equation 6 : drift flux : superficial velocity of the continuous phase : superficial velocity of the dispersed phase : drift velocity : terminal rise velocity single droplet : characteristic velocity : slip velocity ρ d v ∞ 2 d noz σ : nozzle Weber number. - Greek α αmeas αmod ∆ρ ρc ρd ηc σ : dispersed-phase hold-up : measured dispersed-phase hold-up : predicted dispersed-phase hold-up : density difference continuous and dispersed phase : density continuous phase : density disperse phase : viscosity continuous phase : interfacial tension kg m-3 kg m-3 kg m-3 Nm s-1 N m-1 69 . c2 d dnoz Eo Eonoz g N n n0 Udc Usc Usd Vdj v∞ vk v12 Wenoz : parameter in equation 6 : diameter droplet : diameter nozzle : Eötvös number. σ : gravity constant (9.Droplet spray-column Nomenclature c1. g∆ρd σ 2 m m m2 s-1 m3 m-2 s-1 m3 m-2 s-1 m3 m-2 s-1 m s-1 m s-1 m s-1 m s-1 g∆ρd noz 2 : Eötvös nozzle number.

70 .

based on stationary force balances. the type of gel bead flow inside the 3-phase fluidized bed was characterized. The bed was fluidized by a continuous aqueous phase (30 mM KCl as continuous aqueous phase). All data for gel bead hold-up and for droplet hold-up up to 0. characterized by the gel beads flow: two homogeneous flow regimes. Literature models for different types of 3-phase systems could not be extended to the system under study and a new model for gel bead and dodecane droplet hold-up in the fluidized bed was derived. and a disperse organic phase (n-dodecane droplets) rose through the bed of particles.06 were well described. and the various phase hold-ups were measured and studied. 71 . Visual observations of the 3-phase fluidized bed revealed five different flow regimes. An empirical model was derived to describe the droplet hold-ups for the entire range of measured hold-ups. the reactor featured a bed of κ-carrageenan gel beads as solid phases. A drag-force model for a single gel bead and a single droplet in the 3-phase mixture was derived. two heterogeneous flow regimes and a region with wash-out of gel beads. At different aqueous and n-dodecane velocities.4 Solid and Droplet hold-ups in a liquid-liquid-solid 3phase fluidized-bed Abstract The hydrodynamics of a 3-phase fluidized bed reactor were studied. In view of biocatalytic purposes.

1994.e. who studied a 3-phase system of water. 1998. and a solid like glass beads. the solids are fluidized by a rising liquid. especially if the biological reaction is inhibited by toxicity of the reaction substrate or reaction product (De Bont. As a result of such an exchange between the liquid phases. The density of gel beads is much lower. (1979). have been reported by Kim et al.. enzymes or micro-organisms. Here. who also reviewed the disperse phase characteristics (Kim and Kang. thus improving the reaction 72 . while gas bubbles rise through the bed. and with kerosene and glass pearls as disperse phases. has received little attention in literature. The physical characteristics (density and composition) between both types of solid differ largely. 2001. they found the mass transfer rate to be higher than in a 2-phase droplet column.g. An overview and classification of 3-phase systems is given in Table I. Application of a liquid-liquid-solid fluidized bed as extraction column was firstly studied by Roszak and Gawronski (1979) and Dakshinamurty et al. Marshall and Wooley. Dakshinamurty et al. i. The liquid analogue of these systems. as they frequently are used to immobilize a biocatalyst. Most frequently. In support of Roszak and Gawronski (1979). Roszak and Gawronski (1979) focused on mass transfer and found it to be faster than in a fluidized bed. with water as continuous phase. (1988. 1992). 1995. and with n-dodecane droplets and polymeric gel beads as disperse phases. Such a 3-phase liquid-fluidized bed may have an advantage over more conventional bioreactors. and gel beads consist over 95% percentage of water. a liquid-liquid-solid fluidized bed with liquid droplets instead of gas bubbles. studied solids hold-up (1979). gel beads versus glass pearls.. Tramper et al. 1995. Collins et al. and mass transfer (1984). with water as the continuous phase. 1999). Physical aspects of a similar 3-phase system. of which the liquid-liquid-solid 3-phase systems are discussed in more detail below. organic solvent droplets.. Hüsken et al. 1997). This 3-phase system differs from the above described 3-phase liquid-liquid-solid system in the type of solid used. e.. thus allowing a large amount of biocatalyst to be retained in the reactor (Wijffels et al. 1989. In a liquid-liquid-solid fluidized bed. Gel beads are an interesting type of solid. Different modes of operation were classified by Muroyama and Fan (1984). the organic solvent droplets may serve as a substrate reservoir or a product sink. the concentration of substrates and products are kept at sub-inhibitory levels. 1996). we report on a different type of 3-phase liquid-fluidized bed.Chapter 4 Introduction Gas-liquid-solid fluidized beds have been abundantly studied and are widely used in industry.

Hold-up in 3-phase fluidized bed

rate significantly. Application of such a liquid-liquid-solid fluidized bed as an extractive bioreactor was studied by Davison and Thompson (1993). Clostridium acetobutylicum, immobilized in gel beads, was used to ferment glucose to acetone and butanol, while an organic solvent was used for in situ removal of inhibitory butanol. A substantial increase in butanol production rate was reported. These authors, however, did not study physical aspects of the 3-phase fluidized bed. To our knowledge these are the only 3-phase liquid-liquid-solid fluidized bed systems studied. In this paper we therefore focus on the physical aspects of such 3phase liquid fluidized beds. To this end, a 3-phase fluidized bed reactor has been developed on a pilot scale. This paper describes observed phenomena, like the bed stability and the type of the solids flow inside the fluidized bed. Subsequently, droplet and solids hold-ups at different velocities of water and dodecane are discussed. Hydrodynamics of our 3phase fluidized bed have not been studied before. Since available models for the hold-ups in more or less similar 3-phase systems (gas-liquid-solid: Yu and Rittmann, 1997; liquid-solid-solid fluidized beds: Van der Wielen et al., 1996a) gave an unsatisfactorily description of our data, a new model was developed. The solids hold-up followed from the drag force of a single gel bead in the 3-phase mixture. For this drag force, an existing 2-phase model was successfully adopted with minor adaptations. For the droplet hold-up, an empirical model was derived from experimental data. The combination of these models gave a satisfactory description of the solid-liquid-liquid 3-phase system over a wide range of hold-up values.

73

Chapter 4

Table I. Overview of different 3-phase systems Reference Muroyama and Fan, 1984 Hunik et al., 1994 Disperse phase gas Disperse phase solid Continuous phase liquid Remarks hydrodynamics

gas (air bubbles)

solid (reactive gel bead)

liquid

aerobic conversions

Roszak and Gawronski 1979, Dakshinamurty et al. 1979, Kim et al. 1994 Davison and Thompson, 1993

liquid (organic droplets) liquid (organic droplets) solid

solid liquid (glass beads) (water)

hydrodynamics

solid (reactive gel beads) solid

liquid (water)

bioreaction

Van der Wielen et al., 1996b Michielsen et al. 2001

liquid

reaction and hydrodynamics reaction not performed in a packed or fluidized bed

solid (substrate crystals) liquid (organic droplets)

solid (product crystals) solid (gel beads)

liquid (water)

this study

liquid

Material and Methods
Gel beads Celite™-enforced κ-carrageenan gel beads were used as disperse solid phase; they were produced with a resonance nozzle (Hunik and Tramper, 1993). An aqueous 3.25 % κ-carrageenan solution (Genugel 0909 Copenhagen Pectin Factory), mixed with 10 w/w% Celite™, kept at 35 °C, was forced through a 0.8-mm nozzle at 202 Hz and 1.5 bar. Drops were collected in a 80-mM KCl solution for hardening. To obtain spherical beads, butyl acetate (Aldrich) was layer upon the hardening solution. After hardening for approximately two hours, the beads were stored in a 30-mM KCl solution. The same KCl solution was used as continuous phase; it prevented the
74

Hold-up in 3-phase fluidized bed

elution of counter-ions from the gel beads. Density and diameter of the gel beads were determined as described by Van Zessen et al. (2001). Physical properties of the gel beads are given in Table II.
Table II. Phase properties in the 3-phase fluidized bed gel beads density (kg ) 1065 -1) viscosity (mN s diameter ( mm ) 2.80 settling velocity (cm s-1) 5.12 surface tension 30 mM KCl/ n-dodecane m-3 n-dodecane 742.7 flux dependent flux dependent (mN m-1) 30 mM KCl 998 0.93 40.0

Fluidized bed The fluidized bed (Figure 1) consisted of a 2-m high and 6-cm diameter glass column. On top of it, a widened section facilitated coalescence of the organic phase droplets and separate discharge of the organic solvent (n-dodecane, Aldrich 99%+ pure) and aqueous (30 mM KCl) phases. n-Dodecane was introduced at the bottom of the column through a liquid distributor (Figure 2). It consisted of a 10 cm high glass column filled with 8 mm glass pearls for a good distribution of the flow over the flow-through column area, and with a sparger on top. The sparger consisted of 209 nozzles protruding through a Teflon plate with a diameter of 6 cm. The nozzles (5 cm length and 1 mm internal diameter), were placed sufficiently far apart to prevent contact between the growing droplets. As continuous phase, a 30-mM KCl solution was introduced through four openings in the column wall, located below the nozzles outlets (Figure 2). An equal flow rate through each of these continuous phase inlets was obtained with throttle valves; this was verified with color-pulse experiments. The storage vessels of both liquids (Figure 2) were maintained at 26 °C. A temperature gradient over the column was not observed. Solids hold-up Solids hold-up (εs ) was calculated from the measured bed height (Hbed, see also Figure 1), the known total solids volume (Vs), and the known column cross-sectional area (A).

εs =

Vs H bed A

1

75

Chapter 4

Bed height was easily determined as there was a distinct transition between the 3phase and 2-phase regions in the upper part of the column (Figure 2). Total solids volume was determined as described by Van Zessen (2001).
2

Hbed

3
1 7

5 4

P2 ∆P
6

P1
5 4 3

4

2 1
Figure 1. Experimental set-up. 1 column; 2 phase separator; 3 sparger; 4 rotameters; 5 dodecane storage; 6 30 mM KCl storage; 7 pressure measurement device. • gel beads, • droplets; Hbed height of the 3-phase fluidized bed Figure 2. Sparger lay-out. 1 dodecane inlet, 2 glass pearls, 3 Teflon™ plate, 4 nozzles, 5 water inlet

Dodecane hold-up At steady state conditions, the pressure drop (∆P) between two points a distance ∆z along the column apart is equal to the total weight per unit of cross-sectional area of the bed inbetween these points (Muroyama and Fan, 1984):
∆P = (ρ s ε s + ρ d ε d + ρ c ε c ) g ∆z

2

where ρs, ρd and ρc are the densities of the solid gel beads, the dodecane droplets and the continuous phase (30 mM KCl). Measurement of the pressure difference between both points involved an external loop filled with continuous phase only (Figure 1). The actually measured pressure difference (∆Pexp) is therefore equal to: ∆Pexp = P1 − P2 = (ρ s ε s + ρ d ε d + ρ c ε c ) g ∆z − ρ c g ∆z As the sum of the various phase hold-ups is unity,
76

3

1986).00 ml. at 24 cm and 49 cm above the sparger. The static interfacial tension between the continuous phase and n-dodecane phase was determined by the Wilhelmy plate method (Hiemenz. Similar pressure measurements were also done in the 2-phase solids-free top section of the column above the 3-phase fluidized beds (positions at 175 cm and 190 cm above the sparger). 77 . Pressure difference was measured over a vertical distance of 25 cm. Table II summarizes the physical properties of both liquids. As rising droplets lead to pressure fluctuations. A Validyne DP103 pressure transducer with a CD23 digital transducer indicator (maximum pressure difference 140 Pa) was used. For each set of measurements. total volume 25. Values for model parameters were obtained from a Gauss-Newton algorithm that minimized the residual sum of squares. the analogue signal was monitored with a personal computer and an A/D converter until a total of 1000 data points were collected that were normally distributed.Hold-up in 3-phase fluidized bed ε s + εd + ε c = 1 4 the droplet hold-up follows from solving equations 3 and 4.91 cm s-1) were measured with calibrated rotameters (Sho-Rate). Density of both liquids was measured with a pyknometer. The mean of this distribution was used as a measure for ∆Pexp. Miscellaneous Flow rates of the KCl solution (0.58 to 2. the transducer was calibrated with a liquid manometer with water and octanoic acid (ρ =903 kg m-3).06 cm s-1) and n-dodecane (0 to 0.

the gel beads formed a packed bed. observations on the nature of the 3-phase fluidized bed will be described and discussed. A new model describing the hold-ups correctly will be presented. below this flux. the hold-ups of the different phases will be shown. Visual Observations A large variety in water and dodecane fluxes was applied to study the behavior of the fluidized bed. the flow pattern of gel beads was not disturbed by the droplets. a minimal water flux (0.Chapter 4 Results First. The gel beads. but the transition from one region to another region was not sharp. At large dodecane fluxes (not applied in our experiments). only 3-phase fluidized bed regions are briefly discussed. It will be demonstrated that literature models for comparable 3-phase systems do not predict these hold-ups correctly. little bead ‘eruptions’ were observed when dodecane droplets escaped. the gel beads were washed out and a 2-phase water/ dodecane system remained. fell back in the fluidized bed. 78 . Homogeneous flow pattern: mimic of a 2-phase fluidized bed (region 1) In this region. and the dispersed phase will be water. the continuous phase will be dodecane. In the absence of dodecane. At a water flux exceeding 5 cm s-1. Inspired by visual observations on the behavior of the 3-phase fluidized bed. Thereafter. five different regions were identified (Figure 3).36 cm s-1) was required to obtain fluidization. phase inversion will occur. however. At the top of the bed. Below. The gel bead movement did not differ from its movement in a 2-phase fluidized bed. and a so-called free board was not formed.

The overall picture was not influenced by the dodecane flux.8 mm).4 0.5 4 s p r a y c o l u m n 1. Heterogeneous flow pattern: structured solids flow (region 3) A clear characteristic of this region is bed contraction: the bed height of the 3-phase fluidized bed was smaller as compared with the 2-phase fluidized bed. 2 Homogeneous flow pattern: turbulent solids flow. The velocity of the solid groups was much higher in the 3-phase fluidized bed as compared to a 2-phase bed. (ρs=1067 kg m-3 and ds=2.2 0 0.Hold-up in 3-phase fluidized bed phase inversion 1 Ud cm s-1 P 0. at higher water fluxes these aggregates became increasingly smaller. a significant freeboard of solids was formed.0 0. due to a continuous change in direction the group motion was much more chaotic (turbulent). Flow map for a 3-phase fluidized bed with gel beads. At none of the fluxes. although the solids group flow became more pronounced for the lower dodecane fluxes. 3 Heterogeneous flow pattern: structured solids flow.8 a c k 0. 4 Heterogeneous flow pattern: unstructured chaotic solids flow. like aggregates. until at a water flux of ~ 2 cm s-1 the gel beads moved individually.75 1 1. However. and dodecane droplets (ρd=743 kg m-3) in an aqueous continuous phase. Another characteristic for this region is the gel bead flow.5 5.6 e d 0. The gel bead flow is characterized by repetitive waves of solids concentrations traveling through the fluidized bed 79 . 5 Droplet coalescent region: slug formation Homogeneous flow pattern: turbulent solids flow (region 2) Characteristic for this region is bed expansion. The height of the 3-phase fluidized bed always exceeded the 2-phase fluidized-bed.25 b e d Uc cm s-1 Figure 3. 1 Homogeneous flow pattern: mimic of a 2-phase fluidized bed. The size of these groups did not alter in time. Gel beads moved randomly through the fluidized bed in little groups.0 5 2 3 1 0.

Wash-out of the gel beads did not occur. Heterogeneous flow pattern: unstructured chaotic solids flow (region 4) Behavior in this region could not be characterized by a single key feature. white areas are low solids concentrations.8 cm s-1. A subsequent swarm of droplets arrives and the cycle is repeated. dodecane droplets rose very slowly through the fluidized bed. Droplets entering into such pockets rose very fast. As the droplet velocity is larger than the solids velocity. This gel bead flow behavior can be explained with the droplet flow. increase in time Figure 4. droplet coalesce was observed to some extent. and the gel beads fall back (decrease of the solids concentration). and the droplets start to coalesce. they stopped moving and fell back. dodecane droplets are released group wise from the tip of the nozzles.8 cm s-1 and water fluxes below 0. Eventually at a certain height. and a freeboard was not formed. No other swarm of droplets follows immediately. as smaller slugs were rapidly disintegrated. was the very heterogeneous axial and radial distribution of the gel beads. At dodecane fluxes above 0. The flow pattern of both gel beads and droplets was random and chaotic. These slugs pushed the 80 . As opposed to full coalescence. Droplet coalescent region: slug formation (region 5) At the bottom of the column. An overall characteristic. tumbling chaotically. full slug formation was not observed. Locally. dodecane slugs spanning the column diameter were formed. thus creating high-voidage pockets. the swarm of droplets passes this high-concentration area. Once gel beads approached each other closely. the solvent droplets pushed the gel beads towards one side of the column. and entailed some gel beads. For the dodecane fluxes in this region. Schematic gel bead flow pattern through a 3-phase fluidized bed in heterogeneous flow region: structured solids flow. however. Dark areas represent high solids concentration.Chapter 4 (Figure 4). Such a swarm of droplets rises then through the fluidized bed and pushes the gel beads upwards (thus increasing the solids concentration).

75 cm s-1 the bed height decreased with the dodecane flux (bed contraction. 1989). Kim et al. data not shown). who reviewed gas-liquid-solid fluidization. At water fluxes above 0.25 cm s-1.Hold-up in 3-phase fluidized bed full bed of gel beads upwards out of the column. Particle and droplet hold-ups Dodecane droplet hold-up as a function of the water flux at various dodecane fluxes is shown in Figure 5a. also reported on bed contraction. 1979. hold-ups could not be measured unequivocally in this region.75 cm s-1 and dodecane fluxes above 0. who studied a gas-liquid-solid fluidized bed with and without biofilms. which was found to increase smoothly with the decreasing water flux. Similar results have been reported for a water-kerosene-glass pearl fluidized bed (Dakshinamurty et al. and thus features prominent axial and radial distributions of solids and droplets. hold-ups were measured in region 2 and region 3 (heterogeneous structured solids flow). This phenomenon is not unique: Muroyama and Fan (1984). and to decrease with the water flux. which is in full agreement with the findings of Kim et al. cite various authors who observed it.. Droplet hold-up The droplet hold-up was found to increase with the dodecane flux. Solids hold-up The solids hold-up was found to decrease with increasing water fluxes (Figure 5b). At water fluxes below 0. the bed was higher) than in a 2-phase system (Ud=0) at the same water flux. In this region a fluidized bed could not exist.e. hold-ups were only measured in region 2 (homogeneous turbulent solids flow). both in 2-phase and 3-phase fluidized beds. Yu and Rittmann (1997). (Figure 5a). The transition from region 3 to region 2 had no effect on the dodecane hold-up. Consequently. the solids hold-up in the 3-phase system was always smaller (i.36 cm s-1 and water fluxes below 1.42 cm s-1. (1999). At dodecane fluxes above 0. For the lower dodecane velocities.44 cm s-1). gel bead hold-up as a function of the water flux at various dodecane fluxes is shown in Figure 5b. For the higher dodecane velocities (Ud ≥ 0. For liquid-liquid-solid 81 . the 3-phase fluidized-bed is in the unstructured chaotic flow region 4 (Figure 3).

91 cm s-1 Figure 6 shows the dodecane hold-up and the gel bead hold-up for various dodecane fluxes. ε s 3 − phase ≥ ε s 2−phase (region 3).5 ε 0.75 1 1. εs 3 − phase ≤ ε s 2−phase (region 2).42 cm s-1. 0. (1979) and Kim et al.75 1 Uc cm s 1. According to figure 4. × Ud = 0.29 cm s-1. ◊ Ud = 0. Numbers refer to the different flow regions.1 d 0. O Ud = 0.25 -1 1. This transition was indeed observed (data not shown). ● Ud = 0.55 0. ▲ Ud = 0.27 cm s-1).18 cm s-1. ▲ Ud = 0.75 2 Figure 5a.25 2 0.06 0.5 1.15 Uc cm s -1 1. □ Ud = 0.75 cm s-1.73 cm s-1.05 cm s-1. When compared to a 2-phase droplet column of dodecane in water (εs=0). when the water flux decreased) the dodecane hold-up increased as well. O Ud = 0.e.10 cm s-1. When the gel bead hold-up increased (i.54 cm s-1.5 0. Gel bead hold-up as a function of the water and dodecane flux.08 0.5 1.91 cm s-1 ♦ Ud = 0. □ Ud = 0.27 cm s-1.35 ε s 4 3 1 0. ● Ud = 0 cm s-1. Figure b.75 2 0.25 0.10 cm s-1. to a contracted state. ■ Ud = 0.Chapter 4 fluidization the phenomenon was observed before by Dakshinamurty et al. the mere presence of gel beads in a 3-phase fluidized bed had a positive effect on droplet hold-up: ε d ε s >0 > εd ε s =0 in all cases. (1989).02 0 0.54 cm s-1. ◊ Ud = 0. upon decreasing the water flux. ♦ Ud = 0.44 cm s-1.45 0. ■ Ud = 0. the bed will transform from an expanded state. Droplet hold-up as a function of water and dodecane fluxes. at lower dodecane fluxes (Ud≤ 0.04 0. 82 .

Droplet hold-up (εd) as a function of solids hold-up (εs) and dodecane fluxes. although not for liquid-liquidsolid (gel bead) fluidized beds.54 cm s-1 ♦ Ud = 0. which were either empirical or based on the wake model. Van der Wielen et al.06 0. For gas-liquid-solid fluidized beds.44 cm s-1. based on the wake model and with an empirical model for the gas hold-up.10 cm s-1 ■ Ud = 0.04 0. 83 . Muroyama and Fan (1984) gave an extensive overview of models.Hold-up in 3-phase fluidized bed 0. the fundamental force balances are the same. and overestimated dodecane hold-ups above 0. It appeared that both models failed in predicting the droplet hold-up. The model of Chen and Fan largely underestimated the dodecane hold-ups below 0. A more fundamental model for the gas hold-up in a liquid-solid-gas fluidized bed was derived by Chen and Fan (1990). Yu and Rittmann (1997) presented a model for all three hold-ups in a gas-solid-liquid fluidized bed. In these calculations we used the measured solid hold-ups. ● Ud = 0. Although gas bubbles differ in physical characteristics from dodecane droplets.27 cm s-1 ▲ Ud = 0. Points on the εd-axis are calculated from equation 14 (assuming a water flux of 2 cm s-1).06.05 cm s-1. ◊ Ud = 0.91 cm s-1 Available models for solids and droplet hold-up 3-Phase hold-up models are available in literature.03. □ Ud = 0. O Ud = 0. The solids holdup was assumed to be known.4 0. Droplet hold-up We tested the model of Chen and Fan (1990) and Yu and Rittmann (1997) for the droplet hold-up.02 0 0 0.1 0. For a solid-solid-liquid 3-phase system.18 cm s-1.73 cm s-1.6 s Figure 6.2 d ε 0.08 ε 0. (1996a) presented a model for predicting all phase hold-ups. × Ud = 0.

03 – 0. When the model parameters were fitted to our data. 84 . which has no physical meaning. (1996a) presented a model for all hold-ups. the description did not improve. This model uses parameters of the individual 2-phase systems. a solid-solid-liquid fluidized bed. For a different 3-phase system. the hold-ups were well predicted (81% of all data showed a deviation less than 25%). Van der Wielen et al. When model parameters were fitted on our data. the prediction was fair. The model of Yu and Rittmann largely overestimated the hold-up. but the higher hold-ups were systematically underestimated (measured parameters of the 2phase liquid-solid fluidized bed were used). When the model parameters were fitted to our data. Solids hold-up When the liquid-liquid-solid model of Van der Wielen was used to predict the solid hold-up in our 3-phase fluidized bed. hold-ups up to 0.02) were well predicted. but the higher hold-ups were largely overestimated.3 were well predicted. Using these parameters.Chapter 4 In the intermediate region (0. low droplet hold-ups (<0. the fitted parameters resulted in a negative wake volume. However. The model of Yu and Rittmann highly overestimated all hold-ups. the description improved.06).

i /N ) 10% is the percentage of data with an AVD less or equal to 10%.15. 1 For εs → 0 the 3-phase fluidized bed simplifies to a droplet column. 25% be the percentage of data with an AVD less or equal to 25%.N (ε measured. Coefficient m2 is equal to 5 for droplet hold-ups less than 0. Thus.. for droplet hold-ups less than 0.Hold-up in 3-phase fluidized bed Table III and IV summarize the above described results. 3 The average deviation (AVD) is defined as: AVD = (∑ i = 1. (2003) 2 For εs → 0 the 3-phase fluidized bed simplifies to a droplet column.i ) ε measured.i − ε model. this model describes the hold-up in a droplet column. and coefficient nd=5. 85 . (2003). as shown by Van Zessen et al. 4 The parameters in this model are fitted together with the parameter in the gel bead hold-up model based on the total gel bead drag coefficient model in Table IV.15 as shown by Van Zessen et al.

1016 + 1.d .06 .8     −0.45 d  1−ε  1 − εs εd s   Fitting results in a much better description.Table III. m2=51 26 21 54 with constant m1 fitted on m1=0.06. In the intermediate region.8/n d 86 71 6.488 d  1−ε  1 − εs εd s   0.045.02 the hold-ups are well predicted. ~ 0.d v cd = 5.88 ⋅ ε d . nd=52 ⋅ ε c − 3.d   ε +ε  1 − εs εd  c s  εc    ε +ε   c d m2 Results3 AVD 10 % 24 32 with m1=1. up to 0.76 v ∞ .5 Lower hold-ups are underestimated. At higher hold-ups the data are underestimated All hold-ups are well described All hold-ups are well described Empirical –24 v cd = 6. Van der Wielen et al. higher hold-ups are less overestimated all hold-ups are highly overestimated with constants fitted on our data 14 51 81 4.4 1. whereas higher hold-ups are largely overestimated.d 13 12 55 63 89 89 The numbers in results column refer to a hold-up up to 0.76 v ∞ .06 This paper: Drag coefficient based 16. 1996   Ud Uc   εc   ε − ε   ρ d − ρ mix   d  c  = v∞d  ρd − ρc       and ρ mix = ε d ρ d + ε s ρ s + ε c ρc .15 U d 0.03 – 0.5 0. All hold-ups are well described.5 our data Yu and Rittmann 1997 150 0.2  U  Ud Ud + U c = + 0. The higher hold-ups are largely overestimated.5 34 81 Empirical v ρ − ρc εs s + 1 − ε d =  cd v ρc − ρd  ∞ .23 + 3.90 ⋅ ε d -0. the prediction is fair Fitting does not improve the description: lower hold-ups are underestimated.0. Models for the droplet hold-up in a 3-phase fluidized bed Remarks 25% 64 Droplet Hold-up Reference m1 Equation Chen and Fan 1990  εc  Ud Ud + U c = + v∞ .3 16  U  Ud Ud + U c = + 4.37 U d 0.

s   ∞.s ⋅ ε c − 3.51 7. 1996 4. v∞s=3. v∞s=3. are systematically underestimated.26 and ρ mix = ε d ρ d + ε s ρ s + ε c ρ c .s   v  cs  v   = 1. but the data are still overestimated.77 cm/s This paper drag coefficient based ρ mix − ρ s ρc − ρs  v ∞ .s  ⋅ (1 − εd − kεd ) 1 − 1/n + kεd 33 k = 3. The higher hold-ups.8/n s ρ s − ρ mix  U = (ρ s − ρ c ) ⋅  c v  ∞ .8     7 75 100 Fitting results in a better description of the data. Models for the gel bead hold-up in a 3-phase fluidized bed. Results1 AVD 1/n Gel bead Hold-up Remarks 10% 13 42 the gel bead hold-up is highly overestimated 25% Reference Equation Yu and Rittmann 1997  U U  εc =  c − k d  v v∞ .2 Hold-ups are well predicted over the whole range . n=2. 87 2 -1. with the fitted constants a negative volume ratio is predicted.352.17 εd ) data Van der Wielen et al.3 are well predicted.5 97. A negative volume ratio has no physical meaning. ns=2.77 cm/s 16 35 82 with constants fitted on our k = −5.2 76.01 εc 3 exp( −34.5 εc 3 exp( −5. Moreover. Hold-ups up to 0.35. s     (1 − ε d ) − 6.24 ⋅  cs  v   ∞.08 εd ) .Table IV.

It was assumed that the time-averaged velocity of the gel beads is zero. ε d ) 7a 88 . in which one species of particles settled at constant velocity in a fluidized bed of a second species of particles (Van der Wielen et al.Chapter 4 None of the available models thus gave a fair description of the whole range of both solvent and solids hold-up. a total drag force for a single particle is used (Van der Wielen et al. 1996a). This drag force is the sum of all liquid-particle and particleparticle drag forces. a model was developed to describe these drag coefficient data. New model for gel bead and droplet hold-up As a basis to a new model for all three phase hold-ups. Fgravity = Fbuoyancy + Fdrag 5 where Fdrag is the total drag force experienced by a single particle from its mixed 3phase surroundings.. finally. This model was based on a steady state force balance for a single particle in a multi-component system. The total drag force for a single particle in a 3-phase mixture is given by a general equation: 1 1 Fdrag = 4 π d 2 ⋅ 2 ρ c v ci 2 ⋅ C d (ε s . drag coefficients for single droplets or gel beads in a 3-phase mixture were calculated from the experimental data on hold-ups and liquid fluxes.εd) is the total drag coefficient for a single particle in a 3-phase mixture. including the drag forces originating from like particles. Combining equations 5 and 6 and evaluating gravity and buoyancy force gives for a gel bead: 1 6 π d s 3 g (ρ s . 1996a). ε d ) 6 where d is the particle diameter. Consequently. was used to describe the droplet and gel bead hold-up.ρ mix ) = 1 4 1 π d s 2 2 ρ c v cs 2 C d. Subsequently.s (ε s . As in general the contributions of both fractions cannot be distinguished. vci is the slip velocity between water and particle and Cdi(εs. a new model was derived.. This model.

73  ρ d gd noz     = noz noz    d noz  σ     −1 10 and the rise velocity can be calculated with Vignes' equation (Godfrey and Slater. Based on continuity.55Eo 0.0393We 0. ε d ) 7b The bulk density ρmix is determined by all phases (Duijn and Rietema.33 + 0. Water and droplet flows were assumed to be strictly vertical. Based on the validity of these hold-up predictions. assuming the droplet diameter to be the same throughout the whole column.d (ε s . However.315   dd  0. 1982): ρ mix = ε d ρ d + ε sρ s + ε cρ c 8 The slip velocities between water/ particle or water/ droplet are defined as the velocity differences between water and particle and between water and droplet.Hold-up in 3-phase fluidized bed and for a dodecane droplet: 1 6 π d d 3 g (ρ d . as they were developed for a droplet column. the slip velocities thus should equal: Slip velocity droplets/water Slip velocity gel beads/water : : v cd = v cs = Ud Uc − εd εc Uc εc 9a 9b The gel bead diameter (ds) is given in Table II. the hold-up in the column section above the 3-phase bed (a droplet column equivalent) could be well predicted with equations 9a. 11 and 14.2  ρ c    2 /3  ρc    η   c 1/3  1 − Eo    6   11 The validity of equations 10 and 11 for the present case might be argued. while the solids flow was assumed zero. The droplet diameter (dd) is calculated with the equation of Kumar and Hartland (1996): 2 −0.ρ mix ) = 1 4 1 π d d 2 2 ρ c v cd 2 C d. equations 10 89 . 1994): v ∞ .d d  g∆ρ    = 4. 10.

s for the friction between a single gel bead and its surroundings (droplets. The total drag coefficient in a 3-phase mixture for a droplet (Cd.s on εd: a larger droplet hold-up was found to yield a relatively high Cd.Chapter 4 and 11 were adopted for the present 3-phase fluidized bed. gel beads and 30 mM KCl) was not fully accounted for by variations in vcs’ (Figure 7b). In. 90 . Figure 7a shows the total droplet drag coefficient in a 3-phase mixture as a function of the dimensionless slip velocity between droplets and water. Drag coefficients were classified according to the measured solids hold-up.d) and for a single gel bead (Cd. droplets and 30 mM KCl mixture) was found to be fully accounted for by its dimensionless slip velocity vcd`. Apart from vcs’=f (εd).s) were calculated from the measured hold-ups and the water and dodecane fluxes by equations 4 and 7-11.s value. The behavior of the total drag coefficient (Cd. its value ranges from zero to one. There is no influence of the solids hold-up εs other than vcd`= f (εs). addition the total droplet drag coefficient in a 2-phase water/dodecane droplet column follows directly the behavior of the total droplet drag coefficient in a 3-phase fluidized-bed (Figure 7a. Figure 7a reveals two interesting features. which was defined as the actual slip velocity divided by the rise velocity of a single droplet in stagnant water (equation 11). the behavior of the drag coefficient Cd. there is an additional dependency of Cd. As opposed to the behavior of the total droplet drag coefficient Cd. εs=0).d) for the friction between a droplet and its surroundings (gel beads.d.

0725 ◊ εd = 0. Single gel bead drag coefficient in a 3-phase mixture of gel beads.0125 O εd = 0.30 ▲ εs = 0.45 – 0.35 □ εs = 0. Single droplet drag coefficient in a 3-phase mixture of gel beads.d = c 1 (v cd ') c2 v  = c 1  cd  v   ∞ .20 – 0. dodecane droplets and 30 mM KCl as function of the dimensionless slipvelocity.0475 □ εd = 0. gel beads and continuous liquid. (1996a).17 – 0. A model was developed for the total drag coefficients of droplets and gel beads in a 3-phase mixture. ♦ εd = 0 ● εd = 0.20 ♦ εs = 0.4 0. the gel bead hold-up may be calculated directly.d  c2 = C d. This model used features demonstrated in Figure 7.s 4 1 3 2 0. Model for the droplet hold-up In a 3-phase fluidized bed. O εs = 0 ◊ εs = 0.56 The drawn line is the prediction with equation 12.25 – 0.30 – 0.d 10 Cd. however.6 0.50 Figure 7b. the gel bead hold-up is assumed to be known in this paragraph. as the correlations of Van der Wielen et al.40 ■ εs = 0.2 0. and. the model was based on force balances for a single droplet or gel bead in a mixture of droplets.55 Vcd' Vcs' Figure 7a. (1996a) did not gave satisfactory predictions.35 – 0.45 0.40 – 0. Different correlations for the various total drag coefficients were used.8 1 0.0275 ■ εd = 0.Hold-up in 3-phase fluidized bed 100 6 5 Cd.45 ● εs = 0.09 × εs = 0.35 0.25 0. Therefore. the bed height is easily measured. Similar to Van der Wielen et al. from a known gel bead volume.1 0 0. Figure 7a shows that the total drag coefficient for a single droplet in the 3-phase mixture may be described by a general equation: C d.∞d (v cd ') c2 12 91 . dodecane droplets and 30mM KCl as function of the dimensionless slip-velocity. The newly derived models for the drag coefficients were evaluated for predicting the various hold-ups.50 – 0.25 εs = 0.

can also be described by equation 12. which resulted in c2=-2. The force balance for a droplet in the 3-phase mixture (equation 7a) is divided by the force balance for a single droplet rising in stagnant water (an equivalent of equation 7a).d (1 − εd ) 4 14 Substituting equation 14 in 13 and solving for parameter c2 (εs=0).e. 92 . An adequate model for the slip velocity between droplet and water in a 2-phase system has been presented by Van Zessen et al.75. which results in the drag coefficient ratio. the slip velocity is equal to the terminal rise velocity of a single droplet (v∞.∞d).13: v cd = v ∞ . the prediction of droplet drag coefficients in a 2-phase water/ droplet mixture is not very sensitive to parameter c2.Chapter 4 For εs → 0 and εd → 0. and rewriting the resulting equation (with equation 8 for the mixture density).d). a 2-phase droplet column. parameter c2 was fitted on the 3-phase droplet hold-up data.d/Cd. (2003) for hold-ups below 0.06. the total droplet drag coefficient in the droplet/water 2-phase mixture.∞d (cf. Figure 7a shows that for εs = 0. Although the droplet hold-up and drag coefficient (equation 12) for the 2-phase droplet water mixture are well described. results in c2 = -1. Substituting this relation for the drag coefficient ratio in equation 12. The total droplet drag coefficient for both a 2-phase and 3phase system is now well described (Figure 7a). Consequently. equation 12) may also be derived from force balances. The drag coefficient ratio Cd. yields an implicit correlation for the droplet hold-up: v  ρ − ρc εs s + 1 − ε d =  cd  v  ρ c − ρd  ∞ . the 3-phase experimental data (drag coefficient. it will be shown how parameter c2 can be derived from a 2phase droplet hold-up model. and droplet hold-up) are systematically underestimated (data not shown). parameter c1 is equal to the drag coefficient for a single droplet rising in stagnant water (Cd. Apparently. i.d  c2 + 2 13 In the next paragraph. Therefore.

o the model fitted on the data is equation 12. 93 . ● the model fitted on the data is equation 15. 0. c2=-2.06 the prediction is not good (Figure 8).02 0 0 0. 89% of the model predictions show a deviation less than 25%.06 0.02 0.06. and 76% of the model predictions show a deviation less than 25%.76.5%.04 0. an empirical model was tested as an alternative.06 is 16. the average deviation is 17%. but underestimation of the higher droplet hold-ups). Parity plot for the droplet hold-up in a 3-phase fluidized bed. the description of the data (equation 13. The average deviation between data and model for hold-ups less than 0.d 15 A fit of this equation to the data in figure 5a yielded c3=5. For holdups larger than 0.04 0.08 0. The average deviation between data and model is 13%. 81% of the model predictions show a deviation less than 25%.88 and c5=-0. In view of this partial inadequacy of the total droplet drag coefficient-based model (good description of the drag coefficient.Hold-up in 3-phase fluidized bed Up to a hold-up of about 0.1 measured εd Figure 8.08 predicted ε d 0. A power-law dependency of the dimensionless slip-velocity on both the solvent phase velocity and its hold-up were assumed according to: v cd = v cd ' = c 3 ⋅ U d c4 ⋅ ε d c5 v ∞ . The prediction of the drag coefficient with this empirical model is fair.4. A parity plot between predicted and measured droplet hold-ups (Figure 8) shows that droplet hold-up is well described over the entire range of measured hold-ups.06 0.1 0. c4=0.06) is fair (Figure 8).

now both hold-ups were assumed unknown and will be predicted simultaneously. contrary to the preceding paragraph.s). 9b) ∑εi = 1 (eq. the slip velocity (vcs) is equal to the terminal settling velocity of a gel bead in stagnant water (v∞. The ratio of drag coefficients (Cd.10) ρs ρd ρ ρmixture (eq. 12) / Empirical equation (eq. showing how physical parameters. 8) v12 gel beads Uc (eq. 16) ρd ρ Drag ratio droplets (eq. Figure 7b suggests that the total drag coefficient of a single gel bead in mixture of gel beads. 9a) Ud Uc εs εd Figure 9. nozzle geometry and liquid fluxes determine the three hold-ups. the gel bead hold-up may be predicted by: 94 .s/Cd∞.s). s  c7 C d. parameter c6 should be equal to the drag coefficient for a single gel bead settling in stagnant water (Cd ∞. in which a known gel bead hold-up was assumed. Hold-up calculations in a 3-phase fluidized bed.Chapter 4 Model for droplet and gel bead hold-up In the experimental set-up.11) d droplets (eq. s (1 − ε d ) c8 = c 6 (v c.s ρs Drag ratio gel beads ρ (eq.s ') c7 (1 − ε d ) c8 = C d ∞ . s c 9 (v c. This overall model for hold-up prediction in a 3-phase fluidized bed is outlined in Figure 9.s) may be obtained by dividing the force balance for a single gel bead in the 3-phase mixture (equation 7b) with the force balance for a single gel bead settling in stagnant water (an equivalent of equation 7b). 4) εw v12 droplets (eq.s ') c7 (1 − ε d ) c8 16 For εd → 0 and εs → 0. droplets and 30 mM KCl can be described by:  v  = c 6  cs  v   ∞. So. the droplet hold-up was determined from the gel bead hold-up (equations 1-4). together with equations 8 and 16. So. v∞.15) v∞ droplets (eq. Using this drag coefficient ratio.

s  c7 (1 − εd ) c8 17 In the next paragraph.12 cm s-1) but equals 3.15 0. also c7=f(Uc).08 0. However.s    v  = c9   cs  2  v cs   ⋅ v   ∞. Substituting equation 18 in equation 17. c9=1 and c8=0) resulted in c7 = f(εc). This resulted in c7=-1. 0. Van Zessen et al.04 0. it will be shown how parameters c7 and c9 can be derived from the 2-phase hold-up model of a liquid fluidized bed.45 0..24. and solving for parameter c7 (εd=0. Therefore. 95 . it was fitted together with parameter c9 to the solids hold-up data for a liquid-fluidized bed using equation 17 (data in figure 5b. there was a systematic underestimation of the higher solids hold-up.45 0.08 0. ● 2-phase fluidized bed Figure 10a.06 0. a parity plot between measured and predicted gel bead hold-up.1 0.1 predicted εs predicted εd 0. shows that the data are well described.55 0.35 εc 18 where parameter k is not equal to the terminal settling velocity of a gel bead (5.02 0.04 0. and the relative residuals were not randomly distributed.77 cm s-1. as for c9=1.Hold-up in 3-phase fluidized bed ε d ρ d + ε sρ s + ε cρ c − ρ s ρc − ρs  v ∞ .25 0.06 0. Models fitted on the data are equations 15 and 17.35 0. Ud=0 cm s-1). Figure 10b. 2003 concluded that for a 2-phase fluidized bed of gel beads the slip velocity can be described by: v cs = Uc = k ε c 1. as they should be. Parity plot for gel bead hold-up o 3-phase fluidized bed.55 measured εd measured εs Figure 10b.02 0 0 0. Parity plot for droplet hold-up in a 3-phase fluidized bed.15 0.35 0. Figure 7b suggests that parameter c7 might have a single value.25 0.26 and c9=1. and since εc = f(Uc). Parameter c9 had to be fitted on the data.

while at the same time gel bead hold-up decreased. the dodecane hold-up increased with the dodecane velocity. At a constant water velocity. c4 = 0.90. Equations 15 and 17 were now used to describe the droplet hold-up and gel bead hold-up simultaneously (cf. Its predictions were good for the entire range of measured droplet hold-ups. five flow patterns were discerned: two homogeneous regions. could not predict correctly the entire range of hold-ups in the present system. however. An empirical model for droplet hold-up was developed as an alternative. Figure 9). Characterized by the gel beads flow. For all fluxes. Gel bead hold-up and dodecane hold-up were measured. for the gel bead hold-up data the average deviation is 7. a gas-liquid-solid 3-phase fluidized bed and a solid-solid-liquid fluidized bed.2% (97% of all data show a deviation less than 25%). Conclusion The liquid analogue of a gas-liquid-solid fluidized bed. the dodecane hold-up was always higher than the dodecane hold-up in a 2-phase droplet column. this model was adopted as a part of the simultaneous prediction of all hold-ups. Visual observations of the 3-phase fluidized bed showed how the behavior of the fluidized bed depended on the water and dodecane fluxes. Literature models for related 3-phase systems.Chapter 4 As equation 15 was found adequate for the entire range of droplet hold-ups. droplets and 30 mM KCl) was developed for predicting gel bead hold-up and dodecane droplet hold-up. Parameters c3. two heterogeneous regions and one unstable region in which droplets coalesced and gel beads were washed out from the fluidized bed.76 and c8=-6. We used for c7 and c9 the parameter values of the 2-phase fluidized bed. but droplet hold-ups were described adequately for hold-ups up to 0. a liquid-liquid-solid (gel bead) fluidized bed was studied. Our data are well described by this model (Figure 10). The average deviation between model and droplet hold-up data is 12% (89% of all data show a deviation less than 25%).06 only.15 s m-1. based on steady state force balances for both a single gel bead and a single droplet in a 3-phase mixture (gel beads. At a constant dodecane velocity. c5 and c8 were obtained from fitting the model equations on the hold-up data (Figure 5). 96 .51. the droplet hold-up decreased with the water velocity. The gel bead hold-up was well described for all data. resulting in c3 = 6. c5 = -0. c4. A new model.

d 2 d noz σ : height between two points Nm s-1 kg m-3 kg m-3 N m-1 subscript c d mix s : continuous : dispersed (organic solvent droplet) : mixture : gel bead 97 .8) : bed height : pressure drop : superficial velocity : solids volume : terminal rise/settling velocity : slip velocity between water and component i ∆Pexp : measured pressure difference Wenoz : nozzle Weber number. ρ c v ∞ . g∆ρd noz 2 σ g Hbed ∆P U Vs v∞ vci ∆z Greek ε η ρ ∆ρ σ : hold-up : viscosity : density : density difference continuous and disperse phase : interfacial tension : gravity acceleration (9. g∆ρd 2 σ m-2 s m-1 m m m m2 s-1 m N m-2 N m-2 m s-1 m3 m s-1 m s-1 m Eonoz : Eötvös nozzle number.Hold-up in 3-phase fluidized bed Nomenclature A c1.c2 c3 c4.c5 c6-9 Cdi d di dnoz Eo : cross-sectional column area : parameters in equation 12 : parameter in equation 15 : parameter in equation 15 : parameters in equation 16 : total drag coefficient for particle i : diameter droplet : diameter particle i : nozzle diameter : Eötvös number.

98 .

8 cm s-1 the Dax increased with an increasing Uc.5 Mixing of the continuous phase in a liquid-liquid-solid fluidized-bed bioreactor Abstract A liquid-liquid-gel bead (solid) fluidized bed is a new type of bioreactor to overcome biokinetic or solubility limitations present in different kinds of biotransformations. A KCl-tracer was used. the limitations using this tracer were studied as well. The axial dispersion coefficient in the continuous phase (Dax) increased with an increasing superficial n-dodecane velocity. Due to mass transfer to the gel beads. and 30 mM KCl as the continuous phase) are separately introduced to the bioreactor. In this work the mixing of the continuous phase was studied by measuring the residence-time distribution for different superficial velocities of both liquid phases. An axial dispersion coefficient for the continuous phase was determined from inlet and outlet residence time distribution curves. The mixing of the continuous phase of this 3-phase fluidized bed was comparable with the mixing of either the continuous phase of a bubble column or the continuous phase in the main tube of a liquidimpelled loop reactor. In this 3-phase fluidized bed both liquid phases (n-dodecane dispersed as droplets. A pulse volume of 3 ml (4 M KCl) gave reliable results. For a superficial velocity of the continuous phase (Uc) smaller than ~ 0. 99 . whereas for Uc larger than ~ 0.8 cm s-1 the Dax decreased with an increasing Uc. These results were easily explained by the isotropic turbulence theory of Baird and Rice (1975).

As such. A suitable tracer can be a blue dextran solution (Gommers et al. For a biotransformation in which the production rate is strongly reduced by the product itself. can be transferred to the gel beads (Thompson and Worden. this tracer must not be transferred to either of the two dispersed phases.g. a new type of bioreactor has been studied: a liquid-liquid-solid 3-phase fluidized bed bioreactor.Chapter 5 Introduction The potentials of using an organic solvent as an extraction aid to overcome biokinetic or solubility limitations in different kinds of biotransformations have been shown by many authors (e. 2000. however. To measure a residence-time distribution. 2003). however. 3) another liquid phase which is continuous. 1995). as the signal is full of noise. Vermuë. i. A residence time distribution can be related to mixing characteristics of the bioreactor. Adlercreutz. rising through the fluidized bed. In this paper the residence time distribution of the continuous phase is studied. In the second part of the results section this validated measuring method was applied for measuring the residence-time distribution of the continuous phase of the 3-phase fluidized bed. it serves as an important design parameter. the interpretation of this signal can not be performed reliably. a tracer is injected and its concentration is followed in time at a fixed point in the bioreactor downstream from the point of injection. In this bioreactor there are three different phases: 1) the solid phase that consists of gel beads with biocatalyst. an organic solvent. extract the formed product. and this phase is used for the supply of nutrients to the immobilized biocatalyst. 1986). the concentration of the blue dextran comes close to the detection limit of the analysis equipment. viz. 2003). So. which is dispersed as droplets. Measurements have been performed in the structured solids flow and in the turbulent solids flow regimes. In our experimental set-up. we chose a different tracer. As mixing of the continuous phase is studied. a KCl solution. A dispersion coefficient of the continuous phase is calculated from the 100 . 1992) and this measuring method had to be validated. Different flow patterns regimes have been distinguished in the 3-phase fluidized bed (Van Zessen et al. whereas the organic-solvent droplets. 2) one liquid phase.. Consequently.e. This salt. the gel beads and the organic droplets. This continuous phase fluidizes the gel beads. In the first part of the results section the validation of this measuring method is discussed. The hold-ups of the different phases in a liquid-liquid-solid fluidized bed bioreactor have been studied before (Van Zessen et al.

Mixing of the 3-phase fluidized bed residence-time distribution curve. E(t). C(t). a tracer is injected and its concentration is followed in time. Based on the magnitude of this dispersion coefficient it can be concluded that mixing of the continuous phase in a liquid-liquid-solid 3-phase fluidized bed bioreactor is more comparable to an ideally mixed flow pattern than to a plug flow pattern. Unless the shape of the inlettracer profile is known. see Figure 1. It will be shown that the dispersion coefficient of the continuous phase is well described by the model of Baird and Rice (1975). to a residence time distribution curve.1972): E(t) = Φ C(t) m 1 with Φ is the flux of the continuous phase and m is the total amount of tracer injected. this inlet-tracer profile should be measured. using a suitable flow model. This amount is equal to: m = Φ ∫ C(t) dt 0 ∞ 2 The shape of the distribution curve at the outlet is influenced by the tracer profile of the incoming fluid and the system itself. The shape of this time-distribution can be translated to flow characteristics of the bioreactor. To measure this distribution. follows from (Levenspiel. 101 . time-profiles of the tracer concentration at the outlet as well as time-profiles of the tracer concentration at the inlet have been measured. using the energy-dissipation rate per unit mass for a 3-phase fluidized bed. in this case the continuous aqueous phase in the bioreactor. So. Theoretical aspects Residence-time distribution represents the time-distribution of volume elements in a specific part of a device. The translation from tracer profile.

Models for describing the flow behavior of the continuous phase can range from simple to highly complicated. e. methods based on 1) the moments of the distribution curve.03 0 1 System i N 0.g.09 0.09 0. Different methods were analyzed. They concluded that the most reliable parameter values are obtained from the time-domain analysis. Thus in this paper.06 E(t) 0. To use time-domain analysis for parameter estimation. 2) the application of transfer functions and 3) time-domain analysis. a dynamic model describing flow behavior of the continuous phase is needed.06 E(t) 0.03 0 0 15 30 time (s) 45 60 Outlet 0 15 30 time (s) 45 60 Figure 1. Fahim and Wakao (1982) presented a review on the parameter estimation from tracer-response measurements. Transformation of the residence-time distribution curve at the inlet of a (reactor)system to the residence-time distribution curve at the outlet of a (reactor)system by the (reactor)system itself between the inlet and the outlet. time-domain analysis is used for parameter estimation.Chapter 5 Inlet 0. Two of the most common and simple models are the tanks-in-series model and the plug-flow-with-dispersion model (axial dispersion model). The tanks-in-series model reads: 102 .

t represents time and τ is the mean residence time given by: τ= ε cΦ N V 4 in which εc is the continuous phase hold-up.e.Mixing of the 3-phase fluidized bed d C1 1 = (C in ( t ) − C 1 ) dt τ d Ci 1 = (C i -1 − C i ) dt τ d C out 1 = (C N -1 − C out ) dt τ ∧ i = 2KN . these parameters are related to each other. C(t > 0. the number of tanks in the tanks-in-series model and the axial-dispersion coefficient in the latter model. i. x = 0) = C in ( t ) When Uc is given and εc is known. and V is the total volume of the bioreactor section considered in this analysis. Ci is the tracer profile in tank i. ∀x) = 0 . as pointed out by Levenspiel (1972): D ax = LU c 2 Nε c 6 with L is the height of the bioreactor section. The initial conditions are: C(t = 0. both models contain only one parameter. Cout is the outlet tracer profile. For a flow pattern comparable to plug flow. 103 .1 3 in which Cin is the inlet tracer profile. The axial dispersion model reads: U ∂C ∂2 C ∂C =− c + D ax ∂x 2 ∂t ε c ∂x 5 in which Uc is the superficial velocity of the continuous phase and the Dax is the axial dispersion coefficient. or a small axial-dispersion coefficient. a large number of tanks-in-series.

the limitation of using only discrete values might result in a less accurate description of the flow pattern.. Materials The gel beads used in all experiments were κ-carrageenan gel beads filled with 10% Celite®. Materials and Methods Two sets of different experiments were performed. as the axial dispersion coefficient can take continuous values. whereas the number of tanks can only be a discrete value. but a small part was pumped through a home-made flow-through conductivity cell connected to a conductivity meter (WTW). the liquid flew through a flow104 . A 30 mM KCl solution was pumped through the column with a flow rate high enough to establish a fluidized bed.0 kg m-3). A tracer pulse was introduced to the column by switching a three-way valve. (2001). 1989). However. Development and validation of the measuring method The experimental set-up for developing and validating the measuring method is shown in Figure 2. The axial dispersion model was also successfully applied to describe axial mixing in a liquid-liquid-solid fluidized bed (Kim et al.Chapter 5 For a flow pattern comparable to an ideally mixed tank.76 mm. The organic solvent used in the 3-phase fluidized bed was n-dodecane (Aldrich. A 30mM KCl solution was used as the continuous phase (density is 998. In this paper both models were used to describe the experimental data. 2003). the basic idea of the axial dispersion model is violated. Most of the liquid was purged. The density and diameter were determined as described by Van Zessen et al. As tracer solution a mixture of a blue-dextran solution and a KCl-solution was used. The density of the gel beads was 1065. The preparation of these gel beads can be found elsewhere (Van Zessen et al. density is 742. 99%+ pure). Gommers et al. In the second set of experiments this validated method was used to determine the mixing characteristics of the continuous phase of the 3-phase fluidized bed.7 kg m-3. Next. (1986) showed the validity of applying the axial dispersion model to describe axial mixing for gas-liquid-solid fluidized beds. Especially for a small number of tanks.1 kg m-3 and the diameter was 2. Applying the axial dispersion model instead of the tanks-in-series models is tempting. The first set of experiments aimed at developing and validating the measuring method.

11 A/D converter and personal computer. 7 flow-through conductivity cell. 10 Ultrospec 2000 spectrophotometer. Residence time measurements Measurements were performed in an experimental set-up as shown in Figure 3. 2 glass pearls to provide a good flow distribution. and as both liquids were highly immiscible. At different flow rates tracer concentrations ranging from 40 mM up to 1 M and pulse volumes from 1 ml up to 7 ml were used to study whether mass transfer took place. One column was 20 cm high. 6 storage vessel for 30 mM KCl. The spectrophotometer and conductivity meter were connected to an A/D converter (Hewlett Packet) and a personal computer for data acquisition. 4 fluidized bed of gel beads. Experimental set-up for developing and validating the measuring method. in which the absorption at 280 nm was measured with a spectrophotometer (Ultrospec 2000. was also used. Two different columns were used. Pharmacia Biotech). Water and n-dodecane flew co-currently upwards. they were easily separated with a one-stage gravity settler. 9 flow-through cuvette. 3 sieve-plate. the set-up for measuring absorption and conductivity. the other was 130 cm high. 8 WTW Conductivity Meter.Mixing of the 3-phase fluidized bed through cuvette. in combination with the experimental set-up of the 3-phase fluidized bed. The column was 2 m high with an internal diameter of 6 cm. 12 tracer solution vessel. 2001). as described above. 105 . To verify whether mass transfer took place in the 3-phase fluidized bed. 1 gear pump. A detailed description of this set-up is given elsewhere (Van Zessen et al.. Both columns had an internal diameter of 2 cm. 5 7 4 9 8 11 10 3 6 5 1 2 12 Figure 2. see below. 5 three-way valve.

While a 30 mM KCl-solution was circulated over the column. b data acquisition device). The conductivity of the solution was measured at two different column heights: 25cm and 75cm. 2 separator for 30 mM KCl and dodecane. 8 device for introducing the tracer solution. 1 3-phase fluidized bed. 2 1 7 c a1 b1 a b a 3 a 8 b 5 4 4 6 Figure 3. In the residence-time experiments the tube length was 6 cm that resulted in a pulse volume of 3 ml (internal diameter of the tube is 0. c A/D converter and personal computer). dodecane droplets. the gel beads. Experimental set-up of the 3-phase fluidized bed for measuring the residence time distribution. 3 conductivity measurement (a conductivity cells. the pulse was introduced to the column. the tracer volume could be changed. b1 Ultrospec 2000 spectrophotometer. 6 storage vessel for 30 mM KCl-solution. By switching both three-way valves simultaneously. 5 storage vessel for dodecane. a1 WTW conductivity meter.Chapter 5 Figure 4 shows the set-up used for introducing a tracer pulse to the column. Residence-time distribution measurements were performed for the two (liquid-solid) and three phase (liquid-liquid-solid) fluidized bed. b flow-though cuvette. a tube between two three-way valves was filled with the tracer solution.8 cm). 106 . By adjusting the length of this tube. 4 rotameter. 7 absorbance measurement (a flow-through conductivity cell.

8 – 1. Set-up for the introduction of the tracer pulse. 1 mm thick and 1 cm long. see also Figure 5.05 – 0.8 1 0. Uc (cm s-1) 3-phase fluidized bed Structured solids flow regime Turbulent solids flow regime Ud (cm s-1 ) 0. 4 valve Figure 5. a) By-passing the tracer pulse section.Mixing of the 3-phase fluidized bed Conductivity cell The conductivity cell was home made. The different flow rates applied in the experiments performed with either the 3-phase fluidized bed or the liquid fluidized bed are summarized below.55 – 0. The distance between both wires was equal to the column diameter.6 – 1. Home-made conductivity cell. 2 tube for tracer pulse (6 ml). It consisted of two parallel placed platinum wires. i. b) flow through the tracer pulse section. 6 cm. and Ud represents the superficial velocity of the dodecane phase. 2 (glass)holder for platinum wire. The cell was connected to a conductivity meter (WTW). 3 copper wire for connecting to a conductivity meter. Using an A/D converter (Hewlett Packard) and a personal computer both conductivity signals were stored in time. Uc represents the superficial velocity of the continuous phase.7 - Liquid fluidized bed 107 .4 0.e.1 – 0. B) flow to the fluidized bed.8 0. A) flow from the 30mM KCl-storage vessel. 1 platinum wire. Flow rates were measured with calibrated rotameters (Sho-Rate). 4 column wall. B 1 2 4 b 4 a 1 b 3 A 4 3 1 2 1 2 4 3 Figure 4. 1 three-way valve. 3 storage vessel tracer solution.

For the tanks-in-series model the resulting set of equations is:  d E1 1  S in (t) =  − E1   dt τ  ∫ S in (t) dt   d Ei 1 = (E i -1 − E i ) ∧ i = 2KN . these signals are linearly proportional with the KCl concentration and blue dextran concentration. the tracer concentration-profile has to be known. respectively: C(t) = k i S(t) 7 in which S is the measured signal and ki is a proportionality constant for component i. This was done by multiplying the respective equations with the water flux and by dividing the equations by the total mass of the injected tracer. Although conductivity and UV-absorption are measured. equation 3 and 5. Implementing equation 7 into equation 1 and 2 gives the residence-time distribution based on a measured signal: E(t) = S(t) ∫ S(t) dt 8 The measured outlet signal is transformed to a residence-time distribution using equation 8.1 dt τ d E out 1 = (E N -1 − E out ) dt τ 9 For the axial dispersion model the resulting equation is: Uc ∂ E ∂2 E ∂E =− + D ax ∂t ε c ∂x ∂x 2 10 108 . are transformed in order to calculate a residence-time distribution.Chapter 5 Fitting procedure To obtain the parameters used by both presented flow models. Both models proposed for describing the mixing of the continuous phase.

m) was compared with the measured outlet profile (Eout.m (t) − E out. ∀x) = 0 . Equation 10 was then solved by discretization of the axial coordinate.Mixing of the 3-phase fluidized bed The initial conditions are transformed as well E(t = 0. Both parameters were determined by minimizing the residual sum of squares as defined by equation 11: SS res = ∑( E out. E(t > 0.exp (t) )2 t 11 109 . the calculated outlet profile (Eout. equations 14 and 15a-c. 2003). To determine the number of ideally stirred tanks-in-series in equation 3 or the axial dispersion coefficient in equation 5..exp). x = 0) = ∞ S in ( t ) in ∫S 0 (t)dt The hold-up of the continuous phase followed from a previous presented model predicting the different hold-ups in this type of fluidized bed (Van Zessen et al.

Chapter 5

Results and Discussion
First the results of the validation experiments are shown. These results are applied for performing reliable residence time distribution measurements in the 3-phase fluidized bed. Next, the results of these measurements are shown. Development and validation of the measuring method For a column without gel beads, the residence time distribution, determined with a KCl and blue dextran solution, is shown in Figure 6. As the molecular weight of blue dextran is ~ 2,000,000 g mol-1, diffusion of blue dextran into the gel beads can be neglected. In different experiments diffusion was indeed not observed. Figure 6 demonstrates that both tracers yield identical response curves. So, in order to evaluate whether mass transfer takes place, both response curves can be compared to each other. When mass transfer of KCl takes place, the salt-E(t) curve will differ from the blue dextran E(t) curve remarkably at two points: 1) the top of the salt-curve is lower and 2) the tail is elongated over a longer period of time. Consequently, the KCl-response curve crosses the blue dextran curve, see Figure 7.
0.01
0.04

E(t)
0.0075 0.005 0.0025 0 0 100 200 300

E(t)
0.03 0.02 0.01 0

Time (s)

400

500

600

700

0

40

Time (s)

80

120

160

Figure 6. Residence time distribution curve of a 20 cm high column filled with water. ···· blue dextran tracer; — KCl tracer.

Figure 7. Residence time distribution curve for a 130 cm high column filled with gel beads. Pulse volume 2.9 ml (2 g/l blue dextran, 1 M KCl), . ▬ blue-dextran tracer; ── KCl tracer.

From the experiments performed in the small column (20 cm high) it was observed that mass transfer was only important for a large pulse volume (6 ml) (data not shown). A better understanding whether mass transfer took place, was obtained from experiments in the larger column (130 cm). For a low flow rate (1 cm/s) mass transfer was absent for any salt concentration applied and each pulse volume used (data not shown). At a high flow rate (1.9 cm/s) mass transfer was observed for the
110

Mixing of the 3-phase fluidized bed

larger pulse volumes used, regardless of the salt concentrations applied (data not shown). By reducing the pulse volume, mass transfer almost disappeared (data not shown). So, a KCl-solution as tracer for E(t)-measurements is well possible, provided the pulse volume is not too large. For determining the E(t) in a 3-phase fluidized bed, a pulse volume of 3 ml was used and a tracer concentration of 4 M KCl. This high concentration was used as experiments showed that a high salt concentration did not result in mass transfer, provided the pulse volume is small. Whether mass transfer took place, was verified by using the blue dextran/ KCl-salt tracer pulse for a dodecane flux of 0.1 cm s-1 and a water flux of 1.2 cm s-1. Although the UV-absorption of the blue dextran was close to the detection limit, and consequently the signal is full of noise, it could be concluded that mass transfer was not significant (results not shown). Applying the same tracer pulse (3 ml, 4M KCl) in a 2-phase liquid-fluidized bed, it was concluded that even for high water fluxes mass transfer is not detectable (see Figure 8).
0.02

E(t)
0.015 0.01 0.005 0 0 50 100 150 200 250 300

Time (s)
Figure 8. Residence time distribution curve for a liquid-solid (gel beads) fluidized bed. Puls volume 3 ml, ▬ 4M KCl tracer, ── 2 g l-1 blue dextran tracer.

111

Chapter 5

Residence time distribution experiments First, a typical residence-time distribution curve for a 2-phase and a 3-phase fluidized bed will be shown. Thereafter, the dispersion coefficient derived from the residence time distribution curve, will be given as a function of the water and dodecane flux. Figure 9 shows a typical measured E(t)-curve at the inlet and the outlet of a liquidsolid fluidized bed (figure 9a), as well as of a liquid-liquid-solid fluidized bed (figure 9b). It appears that the inlet E(t)-curve for a liquid-liquid-solid fluidized bed is much broader and the top is much lower. Regarding the outlet E(t)-curves for both fluidized beds it can be said that the E(t)-curve for the 3-phase fluidized bed is much broader. At equal superficial water velocity the differences between the E(t)-curves of both fluidized beds shown in figure 9, were always observed.
1 0.75 1 0.75

E(t) 0.5
0.25 0 0 30 60 90 120

E(t) 0.5
0.25 0 0 30 60 90 120

time (s)

time (s)

A B Figure 9. Residence time distribution for a liquid-solid fluidized bed (A) and a liquid-liquid-solid fluidized bed (B). 30 mM KCl-solution superficial velocity is 1.2 cm s-1, dodecane flux is 0.4 cm s-1.  inlet, ▬ outlet.

Two different, but comparable models for describing the mixing of the continuous phase were used: the tanks-in-series model (eq. 9) and the axial dispersion model (eq. 10). Using the experimental data for a superficial continuous phase velocity of 1.4 cm s-1 and a superficial dodecane velocity ranging from 0.05 to 0.7 cm s-1, the number of tanks-in-series and the axial dispersion coefficient were fitted. The number of tanks-in-series can be transformed to a dispersion coefficient using equation 6. Figure 10 shows the fitted dispersion coefficient, using the axial dispersion model, and the dispersion coefficient based on the tanks-in-series model. From this figure it can be concluded that both models give the same relationship between the superficial dodecane velocity and the dispersion coefficient. From duplicate experiments (the same superficial dodecane velocity) it can be
112

Mixing of the 3-phase fluidized bed

concluded that the experimental method gives reproducible results (see Figure 10). Using the tanks-in-series model, the number of tanks-in-series can only take discrete values. As a result, for a liquid-liquid-solid fluidized bed, in which the number of tanks-in-series is between 2 and 9, the accuracy of the transformed dispersion coefficient is small: for N=2, Dax is between 21 and 63 cm2 s-1; for N=9, Dax is between 5.5 and 6.9 cm2 s-1. As the dispersion coefficient in the axial dispersion model can take any value, the axial dispersion model was used for fitting the dispersion coefficient on the other experimental data.
35 30

(cm2 s-1) Dax

25 20 15 10 5 0 0 0.2 0.4 0.6
-1

0.8

Ud (cm s )
Figure 10. Dispersion coefficient as a function of the dodecane flux for a liquid-liquid-solid fluidized bed, 30 mM KCl-solution superficial velocity is 1.4 cm s-1. ● dispersion coefficient fitted with the axial dispersion model, o dispersion coefficient based on the number of tanks-in-series fitted with the tanks-in-series model. Ud (cm s-1) 0.05 N 9 0.1 4 0.1 3 0.26 0.26 3 2 0.4 3 0.4 3 0.55 0.58 0.71 0.71 4 2 2 3

In this type of liquid-liquid-solid fluidized bed different distinct flow regimes can exist (Van Zessen et al. 2003) depending on the superficial velocities of both phases. Measurements have been performed in the structured solids flow and turbulent solids flow regime. The transition from structured solids flow regime to turbulent solids flow regime occurred at a superficial velocity of the continuous phase of about 1 cm s-1. For these two different flow regimes the dispersion coefficients of the continuous phase are discussed separately. A clear characteristic of the structured solids flow regime is bed contraction: the bed height of the 3-phase fluidized bed is smaller than the bed height of the 2-phase fluidized bed at the same superficial water velocity. Another distinctive characteristic is the movement of the gel beads: the gel beads flow a short distance in upward
113

2003). at a high velocity the gel beads move individually and randomly through the fluidized bed.7 cm s-1 the dispersion coefficient increases with an increase in both superficial velocities. the dispersion coefficient increases. The decrease of the dispersion coefficient at an Uc higher than 0. This transition from structured solids flow to turbulent solids flow regime is not sharp. At higher Uc. and the solids are a little diluted (Van Zessen et al. Large groups of gel beads move up and down the fluidized bed. and hence the mixing of the continuous phase is better. but in such a large group the individual gel beads stick together. The structured solids flow regime is characterized by the flow of the gel beads (Van Zessen et al. At higher Ud.2 cm s-1). and then they stop moving. Increasing Uc. Structured solids flow regime In figure 11 the dispersion coefficient of the water phase fitted on the measured E(t)curve at the outlet. at a lower velocity the gel beads move in small groups. the up-and-down movement of the large groups of gel beads intensifies. In flowing downwards the gel beads are a little further apart.and they stop moving again. the large groups of gel beads fall apart gradually into smaller groups and the mixing of the continuous phase becomes more comparable to mixing characteristics of the turbulent solids flow regime.. 114 . The movement of the gel beads depends on the superficial water velocity. the movement of the individual gel beads inside the large groups of gel beads intensifies.Chapter 5 direction. 2003). the dispersion coefficient increases as well. 2003). using the measured E(t)-curve at the inlet and equation 10 and 11 is shown as a function of the superficial water velocity (Uc) for two superficial dodecane velocities (Ud) (0. and these small groups move randomly through the fluidized bed (Van Zessen et al. This cycle is continuously repeated in time. This figure clearly shows that up to an Uc of 0. At an Uc of about 1 cm s-1. so the solids are a little concentrated. the flow pattern of the 3-phase fluidized bed is in turbulent solids flow. In flowing upwards the gel beads approach each other.7 cm s-1 follows the trends observed for the turbulent solids flow regime. At a higher Uc the dispersion coefficient decreases. and hence mixing of the continuous phase increases. Next they flow a short distance in downward direction – falling back . A clear characteristic of the turbulent solids flow regime is bed expansion: the bed height of the 3-phase fluidized bed is always larger compared with the 2-phase fluidized-bed at the same water flux.1 cm s-1 and 0.

is always much larger than the dispersion coefficient of the continuous phase in a liquid-solid fluidized bed.8 1 Dax cm2 s-1 Uc cm s-1 Figure 11.1 cm s-1. Structured solids flow regime. for a liquid-liquid-solid fluidized the dispersion coefficient is virtually constant or decreases with an increasing Uc. However. ● Ud = 0.4 0.Mixing of the 3-phase fluidized bed 40 30 20 10 0 0. From figure 12 it follows that the dispersion coefficient of the continuous phase in a liquid-liquid-solid fluidized bed. For a liquid-solid fluidized bed the dispersion coefficient increases with Uc (the dispersion coefficient at Uc is 1. Axial dispersion coefficient as a function of the 30 mM KCl superficial velocity. This holds for all superficial dodecane velocities applied. Turbulent solids flow regime Figure 12 shows the dispersion coefficient of the water phase fitted on the measured E(t)-curve at the outlet.6 0. 115 .8 cm s-1 is 84% higher than the dispersion coefficient at Uc is 1 cm s-1). regardless of the superficial water velocity applied.2 cm s-1. ∆ Ud = 0. using the measured E(t)-curve at the inlet and equation 10 and 11 as a function of Uc at different values of Ud.

▲Ud= 0. ● Ud= 0 cm s-1 (liquid-solid fluidized bed).05 cm s-1. ■ Uc = 1.8 cm s-1.3 cm s-1. Axial dispersion coefficient as a function of the 30 mM KCl superficial velocity at different superficial dodecane velocity.8 Ud cm s Figure 13.1 cm s-1.0 cm s-1.4 cm s-1. ● Uc = 1.3 1.6 cm s-1. These observations hold for all the different superficial water velocities applied. Axial dispersion coefficient as a function of the superficial dodecane velocity at different 30 mM KCl-solution superficial velocities.Chapter 5 40 cm2 s-1 Dax 30 20 10 0 0. When dodecane droplets are introduced into a fluidized bed of gel beads.4 cm s-1 the dispersion coefficient is hardly increasing with an increasing Ud.4 0. In figure 13 the dispersion coefficient of the continuous phase for the 3-phase fluidized bed is shown as a function of Ud at different values of Uc. At Ud higher than 0.2 0. this will result in 116 . 40 30 20 10 0 0 0. ∆ Ud = 0.5 1. ■ Ud=0.6 -1 Dax cm2 s-1 0. o Uc = 1.1 1. o Ud=0.9 1.7 Uc cm s-1 Figure 12. The experimental results (Figure 12 and 13) can be explained as follows. This figure 13 shows that at up to an Ud of 0.4 cm s-1 the dispersion coefficient increases with an increasing Ud.

An increase in Ud always results in an increase in the movement of the gel beads. The effect of the superficial dodecane velocity on the axial dispersion coefficient. and the mixing of the continuous phase increases as well. or it will increase only slightly. at a constant Ud. which is in complete contradiction with our results. (1989). However. Such an increase of the movement of the gel beads in comparison with a 2-phase (liquid. the dispersion coefficient increases with an increasing dodecane superficial velocity. As a result. Our results could be explained in the same way. (1989) could well explain their data with the energy dissipation rate per unit mass (Pm). Kim et al. and a further increase in Ud will hardly result in better mixing of the continuous phase. kerosene and glass beads. as will be shown in the next paragraph. and consequently the dispersion coefficient is lower. the dispersion coefficient of the continuous phase increases. A higher Uc resulted for their system in a higher Pm. a general relationship between Uc and the dispersion coefficient does not exist. Obviously. As 3-phase system these authors used water. it depends on the physical constants of the 3-phase system. can be explained by considering the energy dissipation rate per unit mass of the continuous phase. the density of the glass beads (2500 kg m-3) differs very much from the density of the gel beads. 117 . Consequently. (1989) reported that the dispersion coefficient increases with an increasing Uc. This large difference in density can explain the different dependence of the dispersion coefficient on Uc.gel beads) fluidized bed results in a better mixing of the continuous phase of a 3-phase fluidized bed. which is to our knowledge the only study on the dispersion coefficient for liquidliquid-solid 3-phase fluidized bed. at a certain Ud the continuous phase is very well mixed. So. found in this study is in complete agreement with the results of Kim et al. this change in behavior of the fluidized bed could be easily observed). For the relationship between Uc and the dispersion coefficient Kim et al. A decrease in dispersion coefficient with an increasing superficial continuous phase velocity. as will be shown below. and consequently they can be moved easily by the rising solvent droplets. Apparently. A higher superficial dodecane velocity results in more dodecane droplets rising through the fluidized bed. This energy-dissipation rate decreases with an increasing superficial water velocity. The density of the gel beads is relatively low.Mixing of the 3-phase fluidized bed a much more intense movement of the solids (visually. The movement of the gel beads increases. the dispersion coefficient is constant.

and Pm is the energy-dissipation rate per unit mass. (1992) the Pm for a liquid-liquid-solid 3-phase fluidized bed is given by. 1975) has been successfully applied to predicting the axial-dispersion coefficient for two.24 ⋅  c c  U ε   v   c c  ∞. 1983) as well as for liquid-impelled loop reactors (Van Sonsbeek et al. In analogy to Van Sonsbeek et al. and εc is the hold-up of the continuous phase. is given by: ρ mix = ρ c ε c + ρ d ε d + ρ s ε s 14 in which ρs is the density of the gel beads. εs and εd are the hold-ups of the solid phase and dodecane phase.15 ⋅ U d 0. see also appendix A: ρ − ρd ρ mix − ρ c ⋅ U c + g mix ⋅ Ud ε cρ c ε cρ c Pm = g 13 in which g is the gravitational constant. ρc and ρd are the densities of the continuous and dodecane phase.9 ⋅ ε d -0.d εd εc ε c + εd + ε s = 1 118 15b 15c . (2003): ρ mix − ρ s ρc − ρs U ε   v ∞ .s    = 1.51 15a Ud Uc − = 6. The mixture’s density.76 ⋅ v ∞ . s  2 -1..26 (1 − εd )−6. The hold-ups of the three different phases in a liquidliquid-solid 3-phase fluidized bed are given by Van Zessen et al. L is the length scale of the largest eddy (L equals the column diameter). The dispersion coefficient is given by: 4 1 3 D ax = c 1 ⋅ L 3 ⋅ Pm 12 in which c1 is a dimensionless constant. ρmix. 1992).and 3-phase fluidized beds (Kim and Kim.Chapter 5 Modeling the dispersion coefficient of the continuous phase The isotropic turbulence theory (Baird and Rice.

for the different applied superficial velocities. 14 and 15. Figure 14 shows the axial-dispersion coefficient as a function of the energydissipation rate per unit mass. Figure 14 also shows equation 12 with the constant c1 fitted on the experimental value of the axial dispersion coefficient: c1 is equal to 0. and the droplet diameter depends on the superficial dodecane velocity and geometry of the sparger.  isotropic turbulence model with c1=0. at a constant Ud.6 and 1. the dispersion coefficient decreases with an increasing Uc. results in a decrease of Pm.Mixing of the 3-phase fluidized bed in which v∞. The measured axial dispersion coefficients are well described by the model of Baird and Rice (1975).9 % of the data.5 cm2 s-1 and 30 cm2 s-1 have a deviation less than 25%. equations 12-15: the average deviation is 27. 40 30 20 10 0 0 0. It can be calculated that an increase in Uc.04 Figure 14.8 cm s-1.33.02 Dax cm2s-1 Pm J (kg s) -1 0. see also Figure 15.78).017 (R=0.12 cm/s).25% higher and lower predictions with the isotropic turbulence model. ---. A measured axial dispersion coefficient less than 12. It is shown in Figure 15 that most of the measured axial dispersion coefficients between 12. see also Van Zessen et al. An increase in Pm obviously results in a higher dispersion coefficient. 1994).03 0.5 cm2 s-1 is found for Ud=0. and for the droplet diameter the equation of Kumar and Hartland (1996).9%. The rise velocity of a single droplet depends on the droplet diameter. and for Ud =0.05 cm s-1 and Uc between 1 and 1. For calculating the single rise velocity we used the Vignes' equation (Godfroy and Slater.6 cm s-1.01 0. A deviation less than 25% holds for 67. (2003). calculated with equations 13. At these flow conditions it could be visually verified that the fluidized bed of gel beads is only slightly disturbed by the rising 119 .1 cm s-1 and Uc between 1.330 ± 0. and v∞. Axial dispersion coefficient as a function of the energy-dissipation rate per unit mass of the continuous phase.d is the terminal rise velocity of a single dodecane droplet. ● data.s is the terminal settling velocity of a single gel bead (5.

a c1 value of 0. see Figure 15. So.Chapter 5 dodecane droplets. 120 . Absolute deviation between the measured axial dispersion coefficient and the predicted axial dispersion coefficient as a function of the measured axial dispersion coefficient. and for the main tube of a liquid-impelled loop reactor. 150 abs ( (data .35 was also found (Van Sonsbeek. For a measured axial dispersion coefficient larger than 30 cm2 s-1. However. 1975) describes well the relationship between axial dispersion coefficient and energy dissipation rate per unit mass for a liquid-liquid-solid fluidized bed. based on the theory of Baird and Rice (1974). if those combinations of superficial velocities are excluded for which the solids flow in the 3phase fluidized bed is not disturbed by the rising droplets.44m).06m and length 0. Apparently. at these conditions the relationship between energy dissipation rate and axial dispersion coefficient does not hold. the isotropic turbulence theory (Baird and Rice. there is always a duplicate experiment (using the same superficial velocities of both phases) that gives an axial dispersion coefficient with a deviation less than 25%. So. it can be said that the mixing behavior of the continuous phase of a liquid-liquid-solid 3-phase fluidized bed is comparable with both the mixing behavior of the continuous phase of a bubble column and the mixing behavior of the continuous phase in the main tube of a liquid-impelled loop reactor. 1992). at labscale (diameter 0.model) / data) x 100% 125 100 75 50 25 0 0 10 20 30 cm s 40 2 -1 measured axial dispersion coefficient Figure 15. For a bubble column (length/diameter ratio > 5) a c1 value of ~ 0. the deviation is always larger than 25%.35 was found in the absence of slugging.

because the number of tanks can only have discrete values. two different but comparable models. the dispersion coefficient slightly decreases with an increasing continuous phase superficial velocity and increases with an increasing dodecane superficial velocity. These results are well explained by the energy dissipation rate per unit mass of the continuous phase (Pm). For flow conditions that belong in turbulent solids flow regime. For this study a pulse volume of 3 ml 4M KCl resulted in reliable results. show the same relationship between dispersion coefficient and superficial velocities of both phases. It is concluded that. 121 . regardless of the flow pattern. It is concluded that the dispersion coefficients determined with both models. the accuracy of the dispersion coefficient calculated with the tanks-in-series model is much lower.1 cm s-1 the continuous phase superficial velocity is smaller than 1. an increase in Pm results in a higher dispersion coefficient and a decrease in Pm results in a lower dispersion coefficient.Mixing of the 3-phase fluidized bed Conclusion A method to determine the residence time distribution of the continuous phase of a liquid-liquid-solid (gel beads) 3-phase fluidized bed bioreactor. It is concluded that the use of a KCl-tracer gives a reliable residence-time distribution curve provided the pulse volume is not too large. Using the residence-time distribution curves measured for different superficial velocities of the continuous phase (30 mM KCl) and dispersed phase (n-dodecane). With these restrictions the total deviation between experiment and model is 17. were applied to calculate the dispersion coefficient of the continuous aqueous phase. using a KCl-tracer.05 cm s-1. The relationship between dispersion coefficient and Pm as suggested by Baird and Rice (1975) describes the experimental dispersion coefficients of the continuous phase well. The dispersion coefficient of the continuous phase increases with the superficial velocities of both phases.6 cm s-1. for flow conditions that belong in the structured solids flow regime. the mixing of the continuous phase in liquid-liquid-solid (gel beads) fluidized bed is comparable with the mixing of the continuous phase of either a bubble column or the main tube of a liquid-liquid impelled loop reactor. provided the dodecane superficial velocity is larger than 0. was evaluated and validated. the axial dispersion model and the tanks-inseries model.5%. However. and for a dodecane superficial velocity of 0.

2 ρ mix = ρ c ε c + ρ d ε d + ρ s ε s 14 For a liquid-liquid-solid fluidized bed. Energy dissipation rater per unit mass for a 3-phase fluidized-bed bioreactor. as described in the experimental part.1 ∆Phydrostatic = g H ρ mix with the mixture’s density given by: A.3 and A. (1992) we assume that the power input is totally dissipated in liquid motion of the continuous phase.1 and dividing by the mass of the continuous phase (εcρcAH).Chapter 5 Appendix A.3 ∆E potential = A g H (ρ d U d + ρ c U c ) A. The power input is equal to the difference between the gain in potential energy and loss in pressure energy: E disp = ∆E pressure − ∆E potential The hydrostatic pressure gradient is given by A. After rewriting the resulting equation for Pm is: Pm = g ρ − ρd ρ mix − ρ c ⋅ U c + g mix ⋅ Ud ε cρ c ε cρ c 13 122 . In analogy to Van Sonsbeek et al. both liquid phases loose energy when they flow through this pressure gradient: ∆E pressure = A g H ρ mix (U c + U d ) Both liquid phases gain potential energy A.4 in A.4 The energy dissipation rate per unit mass follows from implementing A.

out mix s : continuous phase : dodecane phase : inlet.i x greek Φ εi ρi τ subscript c d in. outlet : mixture of the different phases : solid (gel beads) phase 123 : area of the column : concentration : constant : residence time distribution : dissipated energy : change in energy : gravitational constant : height of the fluidized bed : proportionality constant : length scale : mass : total number of tanks-in-series : hydrostatic pressure gradient : energy dissipation per unit mass : signal : time : superficial velocity of phase i : volume of a section of the bioreactor : terminal rise velocity of phase i : axial coordinate m2 mol m-3 s-1 J s-1 J s-1 m2 s-1 m mol mA-1 m mol N m-2 J kg-1 s-1 mA s m s-1 m3 m s-1 m : axial dispersion coefficient of the continuous phase m2 s-1 : volume flux of the continuous phase : hold-up of phase i : density of phase i : residence time m3 s-1 kg m-3 s .Mixing of the 3-phase fluidized bed Nomenclature A C c1 Dax E Edisp ∆E g H ki L m N ∆Phydrostatic Pm S t Ui V v∞.

124 .

The aim of this paper is to show that the application of such a 3-phase reactor system for biotransformations strongly inhibited by product.g. like the different possibilities with a loop reactor. In the second part of this paper a simple reactor model was developed for a 3-phase liquid-liquid-solid fluidized-bed bioreactor. At a given maximum substrate conversion rate. different reactor configurations are discussed. but also more fancy options. Mota M and Tramper J. Taylor and Francis (London. New York). In Multi-phase bioreactor design. The influence of different parameters are discussed. Chapter 12 Design of liquid-liquid-solid fluidized-bed bioreactors. It appears that already a low distribution coefficient (larger than 1 but less than 2) suffices for a better operation.6 Design of liquid-liquid-solid fluidized-bed bioreactors Abstract A liquid-liquid-solid 3-phase system is automatically formed. or less simple 3-phase bioreactors are a good choice. 125 . Ed. p 225-247. For that. when in one apparatus an immobilized biocatalyst is used for a biotransformation and an organic solvent for extraction. will result in a higher degree of conversion. the distribution coefficient of product over medium/organic solvent. and the flux of the organic solvent. the distribution coefficient is determined for which the 3-phase fluidized bed performs equally well as the 2-phase fluidized bed. In the last part of this paper conditions have been established for which this 3-phase reactor performs better than a conventional 2phase fluidized bed. the toxic product concentration. Provided in situ extraction is needed for such a biotransformation. It is concluded that building and operating a 3-phase system is nothing more than a minor extension of conventional bioreactors. published as: Erik van Zessen. Cabral JMS. e. 2001. a 3-phase fluidized bed. Arjen Rinzema and Johannes Tramper. compared to a conventional reactor system. These include a simple 3-phase fluidized bed.

− there is product inhibition.g. κ-carrageenan. 1992. Adlercreutz 2000). Immobilization may also reduce the amount of surface-active agents. For example. by using an organic solvent the product concentration in the medium is lowered. Inhibition due to direct contact between micro-organisms and the organic solvent can be reduced by immobilizing the micro-organism. whether excreted by the micro-organism. Furthermore. 1993. Depending on the nature of the biotransformations. the equilibrium products dissolve in the organic solvent and the degree of conversion can be enhanced. in any bioconversion with micro-organisms involved. or use a plant set-up with 126 . and consequently prevent a stable emulsion (Vermuë et al. Examples of different biotransformations carried out in a 2-phase system and the advantages of using such a system are given by Vermuë and Tramper (1995a). stable emulsions that are difficult to separate (Vermuë et al. the reason for applying a liquid-liquid 2-phase system is different. using an organic solvent in bioconversions also leads to disadvantages. In general one can profitably choose for a liquid-liquid 2-phase system when: − the substrate dissolves poorly in the medium. surface active agents are present. An elegant way of immobilizing micro-organisms is a method that uses a natural gel solution (e. or present due to cell lysis. In operating an extractive bioconversion one should decide whether to use a bioreactor with bioconversion and extraction integrated.Chapter 6 Introduction Advantages of using a liquid-liquid 2-phase system in biotransformations have been discussed (Van Sonsbeek et al. the combination of bioconversion with the first unit operation in the downstream processing. Broth is mixed with the gel solution and gel beads are made as described by for instance Hunik and Tramper (1993). − the reaction equilibrium is unfavorable. 1995b). 1995b). the toxicity of the solvent might reduce the biocatalytic activity and stability (Vermuë et al. Another reason for using in situ extraction. Obviously. 1995b). immobilization will only work when there is hardly any outgrowth of the micro-organisms. the organic solvent acts as a reservoir. This can result in non-desirable. and little excretion of proteins. Tramper et al. Immobilization of micro-organism has been studied extensively (Wijffels et al. 1996). alginate). might be a facilitated product recovery. Vermuë and Tramper 1995a. Obviously.

The hydrodynamics of this reactor have been studied: the hold-up of the different phases as a function of the fluxes of both phases (Van Zessen et al. Buitelaar et al. Next. the focus is on one of the reactor configurations: a 3-phase fluidized-bed bioreactor. 1987. 1987) and hydrodynamically characterized by Van Sonsbeek (1992a). the liquid-impelled loop reactor. a 3-phase system is always favorable. In the latter case the medium has to be circulated between both apparatus at a high velocity. The aim of this chapter is to show that the application of a 3-phase reactor system for bioconversions strongly inhibited by product. Thereafter. 1995. and physical parameters (distribution coefficient and inhibition constant). First. Another simple bioreactor is the liquid-liquid-solid 3-phase fluidized bed. but the water flow is controlled with a pump and not induced by a density difference. As apparatus. Using in one apparatus immobilized cells for a bioconversion and an organic solvent for extraction. A stronger immobilization matrix would be required than the natural gels mentioned above. Not all conditions favor a 3-phase system. the conditions for which a 3phase fluidized-bed gives a better performance than a liquid-solid fluidized bed are demonstrated. For this purpose a liquid analogue of the air-lift loop reactor. although favorable for good mixing and a high interfacial area between medium and organic solvent. Another option is using column reactors. stirred tank reactors. 2003). Mateus et al. as well as the mixing of the medium phase (Van Zessen et al. and characteristics of this model are discussed. compared to a conventional reactor system. see also the section on loop reactors for more detail. 2003). the harsh conditions in the tank will destroy the gel beads. are not very useful. 1996). a 3-phase system is automatically formed. will result in a higher degree of conversion. a reactor model is developed. Then. This type of reactor has good mixing characteristics. Van den Tweel et al. and seems promising for application in an in situ extraction bioprocess. was designed (Tramper et al.Design of a 3-phase fluidized bed both processes separated. Different biotransformations have been executed in this type of reactor (Vermuë et al. Figure 1 summarizes different aspects looked upon in this paper. The reactor is almost identical to a liquid-impelled loop reactor. but it will be shown that with the right combination of operating conditions (fluxes of both liquids). The application of this type of reactor is discussed in the section on loop reactors. possible process lay-outs and different reactor configurations for a liquid-liquid-solid 3-phase system are discussed. 127 . 1991.

keeping the substrate 128 . pH is controlled by keeping the pH of the substrate storage vessel constant. This column is fed by a continuous medium flow. both liquids flow co-currently . water flow counter-currently .3 Liquid-impelled loop reactors Model derivation for reactor configuration 1 Simulations for a batch operation Profiles inside a gel bead Different distribution coefficients Different toxic product concentrations Different fluxes of organic solvent Comparison between a conventional reactor and reactor configuration 1 Evaluation of the distribution coefficient for which reactor configuration 1 performs equally Batch operation is studied extensively Continue operation is briefly discussed • • Figure 1.3-phase liquid-liquid-solid fluidized-beds. The core of the plant is the bioreactor. If separation between both liquids is not achieved by a simple one-stage gravity settler.e. This up-flowing stream is also used for fluidizing the biocatalytic particles. which might be a prerequisite for good operation. Short overview of different aspects looked upon Process lay-out and operation A schematic presentation of a possible plant is shown in Figure 2.3-phase liquid-liquid-solid fluidized-beds. Temperature control of the bioreactor can be done by keeping the temperature of both storage vessels constant. If the biocatalyst requires oxygen. An immiscible solvent is pumped through a sparger.Chapter 6 • Reactor Configuration . and the formed droplets rise through the fluidized bed. accumulation of the product in the organic phase and depletion of the substrate will occur. i. If coalescence of the droplets is easily achieved. than aeration of one or both storage vessels will be an adequate method. A continuous operation with respect to the substrate phase can be achieved by adding a high substrate concentration to the storage vessel. a 3-phase fluidized bed. Operating the bioreactor in the above-described manner is a batch operation. introduced at the bottom of this column. the 2-phase mixture can be separated outside the column. both liquids are separated at the top of the column and sent back to their respective storage vessels.

and both liquids have to be separated at the bottom of the column. Applying gel beads with a density of 1010 kg m-3 and a diameter of 2 mm in a 3phase fluidized bed resulted in the wash out of the beads at any organic flow rate larger than 10-4 m3 (m2 s)-1. particles remain in the column and the droplets do not coalesce in the fluidized bed. Operating the organic phase continuously is easily done by sending the total out-flowing organic phase to the recovery section. This is true when the organic phase has a smaller density than the medium phase. It is assumed that both liquid phases flow upward co-currently. at least. which means that the fluidized bed is in a hydrodynamic equilibrium. Operating the bioreactor with a continuous supply of substrate will still result in the accumulation of the product in the organic phase. see Figure 2. 129 . When the organic-phase density is larger. Coalescence of droplets can be prevented by carefully adjusting both flows. Experiments with different kinds of gel beads have shown that a stable fluidized bed exists. below the equilibrium concentration.3 mm.0 cm s-1. This strategy will be disadvantageous when the product concentration in the organic phase is low. see also Figure 2. It can be concluded that a stable 3-phase fluidized bed is possible for particles with a settling velocity larger than 5. and to send part of the out-flowing organic phase to the recovery section. and pumping purified organic phase into the column.Design of a 3-phase fluidized bed concentration constant. To prevent overflow of the storage vessel part of the circulating medium flow has to be purged. this liquid has to be introduced at the top of the column. Another assumption is that the 3-phase fluidized-bed bioreactor is stable. a large volume of organic solvent has to be clarified. Purge Substrate to purification section Purified organic solvent DO pH Buffer vessel 1 T Buffer vessel 2 air air Figure 2. Process lay-out for a 3-phase fluidized-bed bioreactor Above a plant lay-out and a process operation is discussed for a 3-phase fluidizedbed bioreactor. A better strategy will be to allow accumulation of the product to a certain concentration. for gel beads with a density of 1060 kg m-3 and a diameter of 2.

to maintain the gel beads in suspension a minimal solvent flux was required.by the rising gas bubbles. droplets rise from bottom to top and medium flows counter-currently. A limit situation of this strategy is the case in which there is no medium flow.0 cm s-1. only visual observations were made on this operating strategy.97 mm and 3.12 mm. see Figure 3. i. but for particles with a velocity smaller than 2 cm s-1 a stable bed could not be obtained (Van Zessen. We do not know whether a stable fluidized bed is possible for particles with a terminal settling velocity between 2 and 5. For particles with a terminal settling velocity larger than 5. For bubble columns. it should be possible to create a droplet column with solids kept in suspension. this situation is shortly discussed.Chapter 6 Alternatives Below two different alternatives are given for operating a 3-phase liquid-liquid-solid system. unpublished). see also Figure 1.4 kg m-3 and 1005. unpublished). A stable 3-phase fluidized bed is possible for those particles with a settling velocity smaller than 2 cm s-1. the counter-current operation is discussed.kept in suspension . were not determined. such as hold-up. As a liquid-liquid-solid 3-phase fluidized bed is the liquid analogue of a gas-liquid-solid 3-phase fluidized bed. These gel beads had a density of 1007. from top to bottom. it is known that gel beads can be fluidized . Experimental data on hydrodynamic parameters. 130 . this 3-phase system was stable for at least 24 hours. Indeed. if a different strategy is followed. However. In operating this strategy.e.4 kg m-3 and a diameter of 1. Firstly. it was possible to keep two different kinds of κ-carrageenan gel beads in suspension in a droplet column (Van Zessen. Provided the solvent flux was higher than a minimal value. Thereafter. Counter-current water flow A 3-phase fluidized bed with both liquids flowing co-currently is not possible for each type of particle.0 cm s-1 a stable bed is possible.

as there is always a higher driving force for mass transfer. Again. In extraction processes counter-current operation is always better than co-current operation. i. Alternative process lay-out: water flows counter-current with respect to the droplets flow As the droplet column with suspended gel beads worked well without any water flow. the organic solvent flux has to have a minimal value in order to make this system work. At higher values the system will not work. Obviously.e. i. This operating strategy gives a couple of advantages over a co-current strategy. The water flux puts a downward directed force on the gel beads. 131 .e. when the water flux becomes too high. the water flux can be increased to a maximum value. we only made visual observations for this set-up. whereas the force put on the gel beads by the organic solvent flux is directed upwards. higher than the minimum flux required for stable operation.Design of a 3-phase fluidized bed Buffer vessel 1 Buffer vessel 2 Figure 3. see Figure 3. a lower gel-bead hold-up.at a fixed water flux. In co-current operation the gel-bead hold-up decreases with an increasing medium flux. . gel beads settle. It is also observed that at a higher organic solvent flux a higher water flux can be applied.at a fixed organic solvent flux. Increasing the organic solvent flux resulted in a larger bed height. the gel beads will settle. and droplets coalesce. this means that the throughput of substrate can be higher in counter-current operation. whereas in counter-current operation the gel-bead hold-up increases with an increasing medium flux. These observations can be summarized as follows: . we also tried to operate the set-up with water flowing counter-currently.

Figure 4 shows four possible reactor configurations for the application of a 3phase system in a loop reactor. has been used by Mateus et al. In part C and D. This type of reactor with particles. Particles circulating between both tubes are discussed below. droplets instead of gas bubbles. 1989). a bed of biocatalytic particles and a spray extraction column are separated. A density difference results in a pressure difference and water will flow between both tubes: the direction of this water flow goes from a high to a low density. internal and external designs exist (Van ‘t Riet and Tramper. (1996) and Vermuë et al. κcarrageenan gel beads. The liquid analogue. This principle has been applied in air-lift reactors (Chisti. and water flows from top to bottom in the column with particles. it is assumed that the particles remain in one column. pressure is lower in the spray column. has been described by Van Sonsbeek (1992a). 132 . there has to be a difference in density of the mixture in both tubes.Chapter 6 Loop reactors Another type of bioreactors are the so-called loop reactors: these reactors consist of two tubes connected to each other. A B C D Figure 4. and are not circulating. In parts A and B of this figure. In order to function. conversion and extraction are fully integrated. 1991). (1995) for biotransformations. Consequently. Loop reactors (see text for explanation) In part A droplets of an organic solvent rise from the bottom to the top of the column as their density is lower than the water density. First. the so-called liquid-impelled loop reactor.

a packed bed will be formed. (1982). Consequently. For example. and the water flux is larger than the rise velocity of the swarm of particles.08 was measured. In that case water will flow from bottom to top in the 3-phase column.8×10-2 m s-1 and a dodecane flux of 0. in our laboratory experiments have been done on the hold-ups in a 3phase system. water. The density of the organic solvent is less than the medium density and solvent droplets will rise. This gives a mixture density of 992 kg m-3. but in this case the density of the organic solvent is larger than the density of water. the aeration must be carefully carried out. If the density of the particles is lower than the water density. However. hence droplets settle. The column with only medium can be aerated. A 3-phase fluidized bed will only exists when the density of this mixture is less than the density of the medium in the other column. i. In part B. In this case water will flow from top to bottom. Water will always flow in the direction indicated in this figure. Part D of Figure 4 is the same as part C. At this smaller medium velocity a 3-phase fluidized bed must still exist. due to a higher solids hold-up or a lower organic solvent hold-up. droplets of an organic solvent settle from the top to the column.8×10-2 m s-1 (Van Sonsbeek et al.91×10-2 m s-1. from top to bottom in the 3- 133 . A fluidized bed will be observed.e. Aeration results in an air hold-up. and the circulation velocity of the medium between both columns is changed. Obviously.20 and a dodecane hold-up of 0. as the density is higher than the medium density. if the water flow is higher than the minimum fluidization velocity.Design of a 3-phase fluidized bed If the density of the particles is higher than the water density. At a water flux of 1. aeration results in the decrease of the mixture density in this column. Provided the density of the solids is close to water. A different situation exists when the density in the 3-phase system is larger than the density of the medium in the other column. this configuration can also work. and the circulation velocity decreases. and hence a smaller mixture density in the aerated column. Part C of Figure 4 shows a 3-phase mixture in one column and medium in the other column of the loop reactor. a gel bead hold-up of 0. see the preceding paragraph ‘Counter-current water flow’. The density difference between both columns is in this case large enough to obtain this water flux of 1. if the particle density is higher than the water density. a fluidized bed will be formed. dodecane and gel beads (density equals 1060 kg m-3). the density difference between both columns becomes less. This so-called inverse fluidization has been studied by Fan et al. This set-up will work. In order a 3-phase fluidized bed exists. 1990). water flows from bottom to top in the column with particles.

In that case particles flow together with the water. This model consists of three major parts: • • • hydrodynamics of the reactor kinetics of the bioconversion mass transfer of compounds between the different phases 134 . that it exceeds the terminal settling velocity of a single particle.e. This principle is widely used in air-lift bioreactors (Heijnen et al. one should consider a fair comparison. with a 3-phase liquid-liquid-solid fluidized-bed bioreactor. Whether such a 3-phase reactor will work better than a conventional reactor. i. they rise in the column. Model Derivation Below. the model we used to calculate the performance of both bioreactors is briefly discussed. One can also compare both reactors on their performance. Only if the biocatalytic particles have a density smaller than water. Making a total process design and optimizing this design for both reactors. the water circulation velocity can be so high. and are present in both columns. the model used for the calculations is described. Before we go into a detailed comparison of both bioreactors. Comparing two different reactors. i. (1997). A general total model for the design of a reactor is schematically shown in Figure 5.e.e. cost of a total production plant. is the topic of this section.Chapter 6 phase column. the conventional reactor. this configuration will work. So far it is assumed that the particles remain in a fluidized state. We decided to compare a 2-phase fluidized-bed bioreactor. but in that case one should consider total cost. i. or synthesis rate of product. goes beyond the scope of this paper. degree of conversion of the substrate. This comparison can be based on cost. Performance of a conventional bioreactor and a 3-phase fluidized bed bioreactor In the preceding section possible reactor configurations have been discussed. However.

For a fluidized bed of 1 m high. This system is used as the model system. in which more than one substrate and product influence the kinetics. Ud Uc Hydrodynamics Hold-ups Mixing phases Cs Cp Biokatalysator Kinetics Substrate consumption rate Product formation rate Mass transfer between phases Figure 5. the number of tanks has been experimentally determined. We have chosen a very simple one-to-one conversion: substrate S gives product P. It appears that for medium fluxes the number of ideally mixed tanks is equal to 13 for the 2-phase fluidized bed and 3 for the 3-phase fluidized bed. transported to the gel bead and there converted to product. the fluidized bed is divided in a number of ideally mixed tanks. dodecane and gel beads. To model the medium flow. In order to make the reactor model less complicated. A model has been developed for predicting the hold-ups in a 3-phase fluidized bed (Van Zessen.5×10-2 and 2 ×10-2 m s-1. highly complicated bioconversions can be involved in producing industrially interesting products. The product itself is transported out of the gel bead to the medium phase and subsequently to the organic solvent phase. but we chose to use here the experimentally 135 . the total model is complete. It is further assumed that substrate is dissolved in the medium and insoluble in the organic solvent. 2003).Design of a 3-phase fluidized bed Together with mass balances over each phase for the different components. Hydrodynamics This part of the model considers the hold-up and mixing of the different phases present. it is assumed that gel beads remain in one tank and do not circulate between the different tanks. Obviously. Model outline Reactor Performance Cp Cs Investigated system In our laboratory hydrodynamic studies have been done on a 3-phase system consisting of water. and for medium fluxes between 0.

44 cm s-1 □ Ud = 0.75 2 0.25 -1 1.5 1.15 ε s Uc cm s -1 1. For describing the mass 136 .02 0 0.5 0.75 cm s-1 □ Uc = 0.5 0. ● Uc = 0. To describe mass transfer inside the gel bead we used Fickian diffusion of substrate and product.10 cm s-1 ■ Uc = 0.06 0. see Figure 6 U c = v ∞ε w n Mass transfer Mass transfer between the different phases in a 2.25 0.54 cm s-1 ▲ Uc = 0. ● Ud = 0.04 0.91 cm s-1 ♦ Ud = 0. 0. Table I.91 cm s-1 Table I gives an overview of the assumptions made and the models used for determining the hold-up and mixing characteristics of both reactors.5 1.75 1 Uc cm s 1.1 d 0.25 1. The dodecane hold-up and gel bead hold-up as a function of medium flux and dodecane flux is shown in Figure 6.54 cm s-1 Figure b.55 0.27 cm s-1 ▲Ud = 0. Gel bead hold-up as a function of the water and dodecane flux.Chapter 6 determined hold-ups as they are more accurate.75 2 Figure 6a.or 3-phase system can schematically be presented as shown in Figure 7.73 cm s-1 × Ud = 0.29 cm s-1 ♦ Uc = 0. Hydrodynamic characteristics for a liquid-fluidized bed and a liquid-liquid-solid fluidized bed 2-phase fluidized bed mixing water phase gel bead phase droplet phase hold-up number of tanks-in-series (15) ideally mixed per tank Richardson and Zaki model with experimentally determined parameters 3-phase fluidized bed number of tanks-in-series (3) ideally mixed per tank plug flow Experimentally determined. Droplet hold-up as a function of the water and dodecane flux.00 cm s-1 ◊ Uc = 0.75 1 ε 0.45 0.35 0.10 cm s-1 ■ Ud = 0.08 0.18 cm s-1 O Ud = 0.42 cm s-1 O Uc = 0.05 cm s-1 ◊ Ud = 0.

the mass transfer coefficient kdrop. Further it is assumed that the distribution coefficient between gel bead and surrounding liquid is equal to one. ··· substrate profile The mass transfer coefficient for the transfer of product from medium to organic phase (K) is a combination of the transfer coefficient from bulk to interface at both sides of the interface:   1 1  + K=  k drop. p o    −1 The mass transfer coefficients kbead. p w mk drop. The equilibrium distribution at the interface can be described with a constant distribution coefficient (m): C p o.po is calculated with the equation of Newman (1931). respectively.s. it is assumed that there is thermodynamic equilibrium at the interface between droplet and medium. the rates are a combination of mass transfer per gel bead or droplet. The resulting rates are given per volume reactor. As 137 . Profiles in a multi-phase system: — product profile. kbead. drop Cbulk water Cinterface gel bead Ci Cbulk component profile Figure 7.Design of a 3-phase fluidized bed transfer from gel bead interface to the bulk of the medium.p. and kdrop.pw are calculated with the equation of Ranz and Marshall (1952). and the number of gel beads or droplets present. we used the film model for substrate as well as product. see appendix A for equations.int = mC p w. Mass transfer from this interface to the bulk of the organic-solvent droplet was also described by the film model. Next.int The mass transfer rate between the different phases can be described with the equations given in Table II. The film model was also used for the mass transfer of the product from the bulk to the interface between medium and organic-solvent droplet.

bulk − C p w. we used the residence time of a droplet ( t = Hbed εo/Ud ). In each tank the medium is ideally mixed.bulk − C s w. The gel beads remain in each tank. see also Table 1. which dissolves well in medium and not in the organic solvent. The organic-solvent phase is in each tank divided in 15 ideally mixed tanks-in-series to mimic plug flow behavior.k bead.Chapter 6 contact time in the latter equation. The total 3-phase fluidized bed is divided in a number of tanks.int )⋅ 6 d bead ε s medium / droplet no transfer to the droplet product .bulk −  m  d drop ε o   Kinetics Let us consider bioconversions carried out by either non-growing cells or enzymes immobilized in κ-carrageenan gel beads.s (C s w.int ⋅ ( )d 6 bead s ε  C p d. Medium is water containing all other essential nutrients.p C p w. 138 . Mass balances Figure 8 shows how the liquid-liquid-solid 3-phase fluidized bed is divided in ideally stirred tanks to represent flow characteristics. It is further assumed that there is only one limiting substrate. bulk  ⋅ 6 K  C p w. Straightforward MichaelisMenten kinetics with first order product inhibition are considered: − rs = X v max C s K m + Cs  Cp   1 − tox   Cp    When the product concentration reaches the toxic concentration (Cptox) the substrate consumption rate is zero. Table II. but this is the simplest model to demonstrate the benefits for in situ extraction. More complicated kinetics can be used. Substrate and product transfer rates between the different phases medium / gel bead substrate k bead.

the mass transfer coefficients. and the diffusion coefficients of the compounds. parameters that can be easy manipulated by the designer. parameters that are difficult to manipulate.g. i. the physical parameters like density and viscosity of the pure liquids.Design of a 3-phase fluidized bed i=15 3 i=15 2 i=15 Ud Uc 1 Figure 8. The total derivation of these balances is given in appendix A. e. 139 . The resulting dimensionless mass balances with the initial and boundary conditions are: Within this model two different kinds of parameters can be discerned: 1. 2. Schematic representation of the 3-phase fluidized bed Mass balances are derived for single tanks. operating variables like the fluxes of the medium and organic solvent.e. and hence the hold-up of the different phases. initial concentration of substrate and product in the different phases. and reactor geometry.

n − X p w. D s dX s rp dz z =1 product ∂ X p p. z = 0.rs ) D s C s w . z = 1. X p = 0 140 .n = ∀t. t = 0.n − X s w.0 Ds z ∂ z ∀t.0 ∂ Fo z ∂z ∂ z2 ( ) ∀t. X = 0 ∀t.Chapter 6 Particle mass balances substrate rp 2 ∂ X s p . t = 0.p X p w.s (X s w. k bead.n p + + ( . dX p dz z =0 ( ) D p dX p rp dz z =1 =0 ∀z.n ) = dX s =0 dz z =0 ∀z.int.n Ds ∂ z2 r 2 D p 2 ∂ X p p.int.n = + − -r s D s C s w . z = 0. z = 1.n ∂ Fo = D p ∂ 2 X p p. k bead.n ∂ 2 X s p .n 2 ∂ X s p .

int.s D (X s w.n ) = ∂ Fo εwHt Ds w s bead batch operation continuous operation t=0. i Xp = − Xp + K ∂ Fo ε d H t.Design of a 3-phase fluidized bed Medium-phase mass balances substrate 2 rp 2 ∂ X s w.n − X p w.i D s do Ds ( )  X w.p X p w. n (X s − X s ) − d ε k bead. n − 1 w.n U w rp 6 εs w. n −1 w. n = − Xp + Xp k bead. Xpo.n − p  p m  i     batch or continuous operation t=0.i=0 for each tank i 141 .n 2 rp 2 U w rp 6 εs w.n − p  p m  i     batch or continuous operation t=0. Xsw=1 for each tank n t=0.n  X w. Xsw=0 for each tank n product ∂ X p w.n + ∂ Fo εw H t Ds d bead ε w Ds ( ) ( ) 2 6 ε o rp −K do εw Ds  X w. Xpw=0 for each tank n Organic-solvent-phase mass balances ∂ X p o. i −1 o.i 2 2 U d rp 6 rp o.n  X w.n − X s w.int.

Next the influence of the distribution coefficient and toxic product concentration is shown.Chapter 6 Calculations Before we compare a 3-phase fluidized-bed bioreactor with its 2-phase analogue. Figure 1 gives a short overview of the different simulations done. 142 . see also Figure 2. we will first show some typical results of the model. The parameters that were used throughout the simulations are summarized in Table III. These parameters are system dependent. and can hardly be influenced. the distribution coefficient depends on the two liquids used. First we show a typical time course of substrate and product concentration inside the gel bead. a toxic product concentration depends on the micro-organism used. but can be influenced by adding specific compounds to the organic solvent for enlarging the distribution coefficient. The influence of the dodecane flux is discussed at the end of this section. For these simulations we took a batch operation strategy for both liquids. and both liquids flew co-currently.

0377 n v∞ Kinetics Michaelis-Menten constant toxic product concentration maximum substrate conversion rate Operation parameters medium flux dodecane flux initial substrate concentration initial product concentration m s-1 1 simulation depended simulation depended mol m-3 mol m-3 mol (m3 s)-1 simulation depended simulation depended 1000 0 m s-1 m s-1 mol m-3 mol m-3 143 .1 0.7 1065.0009325 0.001 1 0.06 m m cm 998 742.Design of a 3-phase fluidized bed Table III Overview of parameters used in simulations Reactor Constants # nozzles diameter nozzle reactor height reactor diameter Physical parameters density medium density dodecane density gel bead viscosity medium viscosity dodecane surface tension between water dodecane diameter gel bead diffusion coefficient substrate diffusion coefficient product distribution coefficient 209 0.00135 and 0.040 2.35 fluidization 0.76 10-9 10-9 simulation depended kg m-3 kg m-3 kg m-3 Nm s-1 Nm s-1 N m-1 mm m2 s-1 m2 s-1 Hydrodynamics hold-ups 3-phase fluidized bed simulation depended Number of tanks per m reactor 3-phase fluidized bed 3 2-phase fluidized bed 13 Richardson and Zaki constants 2-phase 2.

consequently a maximum concentration is found inside the gel bead. for which the substrate is not completely converted.29×10-2 m s-1 and a dodecane flux of 0. total conversion of substrate is reached.32 and a dodecane hold-up of 0. The maximum substrate conversion rate was 0. At larger times the concentration decreases as result of consumption. product is formed. the product concentration inside the gel bead will decrease. the toxic product concentration was 10 mol m-3. As time progresses the concentration increases. after some time product concentration is highest in the centre. Figure 9b: As substrate is converted. With the set of parameters used for calculating the profiles in Figure 9a.Chapter 6 In Figure 9 a typical time course of substrate and product concentration inside the gel bead is shown. Figure 9a: Substrate diffuses into the gel bead and gradually penetrates towards the centre of the gel bead. This transfer of product will continue until equilibrium is reached between medium and organic phase. Early in the process little product is present. Substrate will be converted to product. the profile flattens. until all substrate is converted into product. product not only diffuses out of the gel bead but also towards the centre. a combination of parameters is easily found. 144 . This maximum concentration progresses towards the centre of the gel bead in time. hence substrate concentration inside the gel bead is not equal to zero. Gradually product appears throughout the whole gel bead.91×10-2 m s-1. If at this point of time. We took a medium flux of 1. As product is initially formed near the surface of the gel bead. there is still mass transfer between the organic and medium phase. Hence.093.01 mol (m3 s)-1. which resulted in gel bead hold-up of 0. as long as the product concentration at any place inside the gel bead is less than the toxic product concentration. and the distribution coefficient was 1000. and a maximum concentration is reached. However. After some time also a maximum concentration profile is reached.

Design of a 3-phase fluidized bed 0. • • Complete conversion of substrate can be reached with some combinations of distribution coefficient and toxic product concentration. Xvmax=0.4 0. • At the point in time where substrate is not longer converted.006 0.4 0.6 0. The second process is the gradual conversion of substrate to product.91 cm s-1.004 0.002 0 0. ♦ t = 0:0:3 □ t = 19:6:0 O t = 0:0:33.2 0.32.01 mol (m3 s)-1. Cptox=10 mol m-3. Profile of substrate (A) and product (B) concentration in a gel bead. Uc=1. ▲t = 0:1:53. ■ t = 66:0:0.6 0. the first process is completed after roughly 100 seconds.8 1 dimensionless place B dimensionless place Figure 9.2 0 A 0 0. × t = 98:0:0. ● t = 0:19:0.8 1 0 0.6 dimensionless product concentration 0.4 0.008 dimensionless substrate concentration 0. and is more or less independent of the distribution coefficient and the toxic product concentration. first some general remarks: • The time course profile of the substrate concentration in the water phase shows two processes with a different time constant. and the time course of product concentration in the medium and organic phase. m=1000. εo=0. Before discussing the influence of the distribution coefficient and toxic product concentration on the time course of the substrate concentration in the medium phase.29 cm s-1 Ud=0. there is equilibrium for the product concentration between the medium phase and the organic phase. It can be seen in Figure 9b that around that point in time the gel bead is completely filled with substrate. If substrate is not completely converted. εs=0. ◊ t = 0:6:40. than the end-product concentration inside the gel bead and the medium phase is equal to the toxic product concentration. + t = 111:0:0.0925. and is dependent on the distribution coefficient and the toxic product concentration. 145 .2 0.

when substrate is not converted totally. and the toxic product concentration was 10 mol m-3. Figures 11a and 11b show the concentrations in the medium phase. This is observed for a distribution coefficient of 1000. than at the end of the batch operation the product concentration in the medium phase will be lower than the toxic product concentration.29×10-2 m s-1 and a dodecane flux of 0. The maximum substrate conversion rate was 0. which resulted in a gel bead hold-up of 0.Chapter 6 The influence of the distribution coefficient (m) is shown in Figure 10a. At Cptox equal to 100. Figure 10b. Figure 10a clearly shows that a larger m results in the end in a lower substrate concentration in the medium phase. If the m is high enough. Figure 10a and 10b shows the concentrations in the medium phase. Naturally. Obviously. substrate is completely converted. We took a medium flux of 1. The end of the batch operation is reached at an earlier time for a lower m. hence substrate is converted to a higher degree. hence a more complete conversion and the product concentration goes to Cptox. Figure 10c. Consequently. the product concentration in the organic phase is higher for a higher m. a higher Cptox results in a lower substrate concentration.91×10-2 m s-1. whereas Figure 11c shows the product concentration in the dodecane phase. whereas Figure 10c shows the product concentration in the dodecane phase.32 and a dodecane hold-up of 0. 146 . substrate is totally converted. the product concentration in the medium phase is equal to the toxic product concentration. and Cptox is not reached in the medium phase.01 mol (m3 s)-1.093. As there is equilibrium between the medium and organic phase. The influence of the toxic product concentration (Cptox) is shown in Figure 11. This concentration is reached earlier for a lower Cptox. 10b and 10c.

εo=0. ▬ m=1. Uc=1.004 0. O m=1000 1 0.04 147 0.06 6 dimensionless substrate concentration A 1 10 100 time (s) 100 1000 1000 10000 100000 1000000 0 B time (s) 10000 100000 1000000 dimensionless product concentration organic phase 0 dimensionless product concentration 0.Ud=0.8 0.1 0.-.4 0. Profile of product concentration in the dodecane phase (C) at different distribution coefficients.008 0.8 0. Xvmax=0. ● Cptox=50 mol m-3.29 cm/s.32.29 cm s-1.2 2 0 100 1000 time (s) 10000 100000 1000000 Design of a 3-phase fluidized bed Figure 11. Profile of product concentration in the dodecane phase (C) at different toxic product concentrations. Cptox=10 mol m-3. O Cptox=100 mol m-3 .01 mol m-3 s-1. .4 dimensionless substrate concentration dimensionless product concentration 0 A 1 10 100 1000 time (s) B 10000 100000 1000000 time (s) C 10000 100000 1000000 dimensionless product concentration in organic phase 100 0.32.91 cm s-1.-.0925. Profile of substrate (A) and product (B) concentration in the medium phase at different distribution coefficients. Ud=0.6 4 0.91 cm/s. ▬ Cptox=1 mol m-3 . .02 C 0. εo=0.2 1000 time (s) 10000 100000 1000000 Figure 10. Profile of substrate (A) and product (B) concentration in the medium phase at different toxic product concentrations.01 mol/m3s.002 0 0. Uc=1. εs=0.6 0.0925. ● m=100.01 7 6 5 4 3 2 1 0 1 10 100 1000 0.. Xvmax=0..006 0. εs=0. Cptox=10 mol m-3. m=100. m=10.

εo=0. The dodecane flux influences the hold-up of the different phases.08 0.26. the product concentration in the medium phase is equal to the toxic product concentration. This concentration is reached at an earlier point of time for the lowest Ud. Xvmax=0. ▬ Ud=0. Obviously. Figure 12b.4 0. a higher dodecane flux gives a higher dodecane hold-up. Figures 12a and 12b show the concentrations in the medium phase.04 0.26. although the gel-bead hold-up is the lowest (Ud=0. ● Ud=0. 1 0.75×10-2 m s-1.012. εs=0. whereas Figure 12c shows the product concentration in the dodecane phase. εs=0.75 cm s-1. Profile of product concentration in the organic phase (C) at different fluxes of the organic phase. εs=0.30. Figure 12b. the product concentration in the medium phase is always below the toxic product concentration.14×10-2 m s-1.069. ..8 0. Cptox=100 mol m-1.30).2 0 1 10 100 dimensionless product concentration 1000 time (s) 10000 100000 1000000 0. The time course of product concentration in the organic phase follows the time course for the product concentration in the medium phase.. 148 .06 0. at the highest dodecane flux. Uc=1. m=100. We kept the medium flux constant and changed the dodecane flux. At a constant medium flux.01 mol m-3s-1. but a lower gel bead hold-up.14 cm s-1. Ud=0.27.49 cm s-1. εs=0. Figure 12a shows that complete conversion of substrate is reached for the highest Ud.02 0 100 1000 A B time (s) 10000 100000 1000000 dimensionless product concentration organic phase 10 8 6 4 2 0 100 C 1000 10000 100000 1000000 time (s) Figure 12.6 0.Chapter 6 The influence of the organic solvent flux is shown in Figure 12.036.Ud=0. For the other dodecane fluxes applied in the simulations. εo=0.1 dimensionless substrate concentration 0. εs=0. εo=0.36 cm s-1. Profile of substrate (A) and product (B) concentration in the medium phase at different fluxes of the organic phase.

In a 2phase fluidized bed substrate is not totally converted due to product inhibition.14×10-2 m s-1. Obviously.Design of a 3-phase fluidized bed We kept the medium flux constant and changed the dodecane flux.75×10-2 m s-1. at the highest dodecane flux. At a constant medium flux. So. An economic evaluation must show whether a total reflux yields the lowest cost. We used in this strategy a total reflux as a fair comparison. At this moment. εs=0. the total amount of moles of product produced is equal to the toxic product concentration times the total reactor volume.30).26. The time course of product concentration in the organic phase follows the time course for the product concentration in the medium phase. see Figure 2. When the product concentration reaches the toxic product concentration in the medium phase the conversion stops. but this is not the topic of this paper. a higher dodecane flux gives a higher dodecane hold-up. This concentration is reached at an earlier point of time for the lowest Ud. Figure 12b. In order to compare a 3-phase fluidized bed with a 2phase fluidized bed we calculated the time necessary to reach 99% of the total amount that can be converted in a 2-phase fluidized bed. Figure 12b. εs=0. At the end some remarks are made about operating the bioreactor continuously. Obviously. 149 . but a lower gel bead hold-up. the product concentration in the medium phase is equal to the toxic product concentration. there is a lot of freedom in totally or partly recirculating the organic solvent in the 3phase fluidized-bed bioreactor. The dodecane flux influences the hold-up of the different phases. the product concentrations inside the gel bead and in the medium phase are the same and equal to the toxic product concentration. although the gel-bead hold-up is the lowest (Ud=0. Comparison between a 2-phase and a 3-phase fluidized-bed bioreactor In our comparison we did the simulations for batch operations: substrate is circulated until a certain amount of substrate is converted. Ud=0. Figure 12a shows that complete conversion of substrate is reached for the highest Ud. For the other dodecane fluxes applied in the simulations. the product concentration in the medium phase is always below the toxic product concentration.

Introducing an organic solvent in a 3-phase fluidized bed. that yields the same time to reach 99% of the total amount of moles converted in a 2-phase fluidized bed. The results are shown in Figures 13a-c. resulting in a higher substrate conversion rate. to conclude whether a 3-phase fluidized bed performs better than a 2-phase fluidized bed. Two different toxic product concentrations were used. 150 . i.Chapter 6 Batch operation The comparison between both bioreactors was made for different combinations of medium and dodecane fluxes. at a given maximum substrate conversion rate. a higher distribution coefficient gives a higher transfer rate. i. The extraction rate is influenced by the distribution coefficient. a distribution coefficient is determined. whereas a lower distribution coefficient gives a worse performance.e. and an initial substrate concentration of 1 kmol m-3. This is achieved by extracting the product into the organic solvent. and for different toxic product concentrations. the maximum substrate conversion rate (X⋅vmax) and the distribution coefficient (m). will in most cases result in a lower gel-bead hold-up. Thus. So. The distribution coefficient at a given maximum substrate conversion rate is thus calculated for which a 3-phase fluidized bed performs the same as 2-phase fluidized bed. We chose to evaluate the performance of the bioreactor with two parameters. This was done for a number of organic solvent fluxes and medium fluxes. at a given maximum substrate conversion rate. and the distribution coefficient might be influenced by adding specific compounds to the organic solvent for enlarging the distribution. The maximum substrate conversion rate can be influenced by using more or less biocatalyst. the time to reach a certain degree of conversion can be manipulated by the distribution coefficient (m): a higher m results in a shorter time. This loss of activity has to be compensated for by lowering the product concentration in the gel beads. As can be deduced from the equation in Table II. A higher distribution coefficient results in a better performance. This means loss of biocatalytic activity per unit volume of bioreactor.e. see also Figure 10. and to keep the other parameters constant. 1 mol m-3 and 100 mol m-3 .

14 cm s -1 Design of a 3-phase fluidized bed X v max -3 -1 mol m s 0.1 2.01 X vmax ( mol m s ) -3 -1 0.95 Cptox = -3 100 mol m 15545 1560 319.1 1 Cptox = -3 1 mol m 166.1 0.3 22.91 cm s tbatch (99%) (min) -1 Figure 13.94 .3 1.14 cm/s.9 1.001 0.5 227.29 cm s -1 -1 = 0.1 1 C A B X vmax ( mol m-3 s-1 ) X vmax ( mol m-3 s-1 ) 0.9 32.3 1.001 0. ♦ Ud = 0.9 1. closed symbols Cptox = 100 mol m-3 C Uc = 2 cm s -1 ■ Ud =0.05 0. Distribution coefficient as a function of the maximum substrate conversion rate for which a 3-phase fluidized bed performs equally well as a liquid fluidised bed.001 0.1 1 0. ♦ Ud = 0.14 cm s .54 cm s -1 A Uc ■ Ud = 0.63 17.7 1.1 1.5 1.7 1.47 14.13 15. ♦ Ud = 0.81 19.9 0.3 1.1 1 151 B Uc ■ Ud ● Ud = 1.4 167.9 0.9 0.1 1.4 Cptox = -3 100 mol m 20950 2108 435.01 0.7 m 1.5 m 1.1 0.02 Cptox = -3 100 mol m 32504 3278 699 361.5 1.2.55 cm s -1 = 0.1 2.8 55.2 21.28 cm s -1 ● Ud = 0.3 Cptox = -3 1 mol m 355.01 0.40 cm s . Open symbols Cptox = 1 mol m-3.01 0.9 1.4 55.04 15.75 cm s -1 -1 = 0.001 m 1.31 23.

.4 3. Especially.6 6. The distribution coefficient at each X⋅vmax in favor of the 3-phase fluidized bed is relatively low. 1985 Brink and Tramper. 1992 Boeren. et al..52 120 > 5000 reference oxygen FC40/water octene/water toluene/water butylacetate/water toluene/water hexadecane/water FC40/water dodecene/water Sonsbeek et al. 1993 Brink and Tramper. et al.. 1985 Kawakami. 1992 isobutyl methyl 123 keton / water Flaschel.. 1992 Boeren.9 1. in which distribution coefficients for different solutes for different liquidliquid 2-phase systems are given. Table IV. et al.. . et al.17–dione dehydro-epi androsterone phenylalanine propyl ester Hexane/water Hexane/water Hexane/water 1500 8.5 4. see Figures 13a-c. et al. whereas the results in Figure 13c – the highest water flux 152 .1992a Meer.. 1992 Looking in more detail to Figure 13a to c.. Distribution coefficients for solute for different liquid-liquid 2-phase systems Solute 2-phase system distribution coefficient 12 11. et al. Based on this table one might suggest that at any pre-set medium flux. 1992 Vermuë. et al.5 24 Boeren.. when we consider the examples in Table IV. 1994 propenoxide butanal tetraline Steroids Androstenoloneactate 4-Androstene3. 1994 Vermuë. a 3-phase fluidized bed performs better than a conventional 2phase fluidized bed.Chapter 6 A 3-phase fluidized bed performs better for combinations of the maximum substrate conversion rate (X⋅vmax) and distribution coefficient in the area above the lines in these figures. the results in Figure 13a and Figure 13b – a lower water flux applied show that a lower dodecane flux requires a higher distribution coefficient. et al. 1992 Kawakami.

although dodecane hold-up is higher for a higher dodecane flux.1 % of the initial substrate concentration). there is a significant decrease in gel-bead hold-up with an increasing dodecane flux (compared to the 2-phase fluidized-bed analogue. as the dodecane flux determines largely the dodecane hold-up and the mass transfer coefficient. At the lower water flux. a higher distribution coefficient gives also a higher transfer rate. At the lower water fluxes. In most cases the dodecane flux decreases the gel-bead hold-up and. For the highest water flux applied. Hence. When a lower distribution coefficient is used for the higher dodecane fluxes. as above for the lower water fluxes. at these water fluxes.e. as explained above. see also the equation in Table II. to establish a high enough transfer rate. less than 15% for all dodecane fluxes applied. the transfer rate must be higher for the higher dodecane fluxes. Figure 13a. i. 0. the difference between both dodecane fluxes is almost insignificant. both observations can be easily explained. A higher dodecane flux results in a higher transfer rate.Design of a 3-phase fluidized bed applied – show that a higher dodecane flux requires also a higher distribution coefficient. Consequently. Thus. Figure 13c. Figure 13a and Figure 13b. this decrease in gel-bead hold-up has to be compensated for with a high enough transfer rate to the dodecane phase. A general remark about the difference between a high and low toxic product concentration can hardly be made: at X⋅vmax the distribution necessarily for giving an equal performance is sometimes lower other times higher. The transfer rate is also influenced by the distribution coefficient. At the higher water flux. the decrease in gel-bead hold-up is not that large. a higher distribution coefficient is needed for the lower dodecane flux. are more or less the same. The open symbols in Figure 13a and c show the distribution coefficient as a function of the X⋅vmax for which the 3-phase fluidized bed performs equally well for a low toxic product concentration (1 mol m-3. the resulting transfer rate is not enough for establishing the required transfer rate. the transfer rate in the 3-phase fluidized beds. this difference is a little more pronounced. the distribution coefficient has to be higher for the higher dodecane fluxes. Figure 13b. So. The transfer rate itself is influenced by the dodecane flux. 39% for the highest dodecane flux and 7% for the lowest). 153 . necessary to perform equally well as the 2-phase fluidized bed. and there is hardly any difference in gel-bead hold-up between the highest and lowest dodecane flux. Although the latter observation might seem unexpected.

Application of an organic solvent. little substrate is converted. The dimensionless concentrations for a 2-phase fluidized bed and 3-phase fluidized bed are calculated and given in Table V for different recycles. a medium velocity of 0. As the product concentration increases with an increasing recycle.02 m s-1. and hence a small amount of product is made.9994 0. dodecane flux 0. hence εs=0.g. εo=0. Only for fast conversions or high reactors.72e-4 2.99 2-phase fluidized bed substrate product 0.07e-4 0.9994 0. using a 3-phase fluidized bed. i.75×10-2 m s-1. Consequently the residence time is short and the conversion rate is low.5e-4 Obviously.48. As an example. Table V. Otherwise one should use a set-up with a partial recycle and fresh substrate supply. recycle of medium and/or organic solvent.9994 6. a high degree of substrate conversion is possible. 154 . as we calculated for the next case (the organic solvent was not recycled): medium flux 0. the initial substrate concentration.Chapter 6 Continuous Operation Executing a biotransformation continuously in a 2-phase fluidized bed without a recycle is highly inefficient.50e-4 0. The introduction of the organic solvent results in a smaller product concentration in the medium phase.40×10-2 m s-1. Dimensionless substrate and product concentrations in a continuously operated fluidized bed Recycle 0.98e-4 8. Cptox = 100 mol m-3. X⋅vmax = 0.9 0.0061 0. So.1 0. and a reactor height of 1 m results in conversion of 50 mol. assuming zero-order kinetics. e. substrate will be less converted.01 mol s-1. The incoming substrate concentration is kept constant. there is lot of freedom.0019 0.9993 6. and m=100. and hence a fluidized bed might be attractive. will result in a higher conversion. This is 50% of the attainable conversion at a toxic product concentration of 100 mol m-3.0396 3-phase fluidized bed substrate product 0.e.9996 0.045. whether a continuous operation is feasible depends on a total cost calculation of the plant. The product concentration in the medium phase will accumulate until steady state is reached.9994 0. as the medium flux is high compared to the maximum substrate conversion rate. a substrate conversion of 1 mol s-1. In optimizing a continuous operation.005 product organic solvent 0.

for which the 3-phase fluidized bed performs equally well.Design of a 3-phase fluidized bed Conclusion Different aspects of liquid-liquid-solid 3-phase systems have been discussed. It can be concluded that building and operating a 3-phase system is nothing more than a minor extension of conventional bioreactors. However. like the different possibilities with a loop reactor. are good options for operating specific biotransformations. The benefits of such a system for typical biotransformations have been discussed in literature. the ideal reactor configuration can still be argued about. it can be concluded that a 3-phase fluidized bed. For a liquid-liquid-solid 3-phase fluidized-bed bioreactor conditions have been established for which this reactor performs better than a conventional 2-phase fluidized bed. So. 155 . the distribution coefficient is determined. Keeping all parameters constant. but also more fancy options. at given maximum substrate conversion rate. Different reactor configurations are discussed here. including a simple 3-phase fluidized bed. changing the operation variables. provided in situ extraction is needed for such a specific biotransformation. medium flux and organic solvent flux. or less simple 3-phase bioreactors. It appears that already a low distribution coefficient (larger than 1 but less than 2) suffices for a better performance.

Chapter 6 Nomenclature Ar Ci j Di : area reactor : concentration component i in phase j : diffusion coefficient component i m2 mol m-3 mol m-3 m2 s-1 m m m m2 s-1 m m m m3 (m2int s)-1 mol m-3 m3 (m2int s)-1 m3 (m2int s)-1 m m s m3 (m2s)-1 m3 (m2s)-1 m3 m s-1 mol (gcats)-1 m s-1 m s-1 gcat m-3 - Csw.i : mass transfer coefficient component i water/gel bead kdropw : mass transfer coefficient bulk water/interface droplet m Rp r t Uc Ud Vr v∞ vmax vwp vwo X Xij z : distribution coefficient organic/water : gel bead radius : space-coordinate in gel bead : time : medium phase flux : organic solvent flux : reactor volume : terminal rise velocity droplet : maximum substrate conversion rate : slip velocity water particle : slip velocity water droplet : concentration biocatalyst : dimensionless concentration component i in phase j : dimensionless space coordinate in gel bead kdropo : mass transfer coefficient interface droplet/bulk droplet m3 (m2int s)-1 N_dis : number of tanks in disperse phase per one medium tank 156 .0 : substrate concentration in medium phase at t=0 dbead : gel bead diameter ddrop : droplet diameter dnoz Fo g Ht Ht.i K Km : nozzle diameter : dimensionless time : gravity constant : height of one medium tank : height of disperse phase tank : overall mass transfer coefficient water/droplet : michaelis-menten constant Hbed : height fluidized bed kbead.

Design of a 3-phase fluidized bed subscripts p s : product : substrate superscripts bulk : bulk phase int n o p tox w greek ηw ∆ρ εo εs εw ρ σ : viscosity medium phase : densitydifference : organic solvent phase hold-up : gel bead phase hold-up : medium phase hold-up : density : interfacial tension Nm s-1 kg m--3 kg m-3 N m-1 mol s-1 mol s-1 mol s-1 : at the interface between two phases : tank number in medium phase : organic solvent phase : gel bead phase : toxic : medium phase ϕpwp : mass transfer flux of product gel bead/water ϕwop : mass transfer flux of product water/organic solvent ϕwps : mass transfer flux of substrate water/gel bead 157 .

Product is transferred from the gel beads to the medium. A tank-in-series model was applied with a large enough number of tanks to mimic plug flow.n = D s ( ) ∂C s ∂r p. So. As conversion we took one mole of substrate giving one mole of product.n r =R Product ∂ C p p.n 2 ∂ C s p. Systematically the bioreactor looks as shown in Figure 8.int. ∀r. For simplicity reason we assumed that the gel beads did not circulate inside the reactor. read as follows: The gel bead Substrate ∂ C s p. including the boundary and initial conditions. edge of the gel bead k bead. Only product is present in the organic-solvent phase. ∀r.n ∂ r2 2 ∂ Cp + r ∂r p. n ∂t = Ds ∂ 2 C p p.n = Ds + − ( − rs ) ∂t ∂ r2 r ∂r t=0. Mass balances are derived for the individual phases. centre of the gel bead ∂C s p.n ∂r =0 r =0 158 . The total reactor is divided in a number of vertically stacked ideally stirred tank reactors. n + rp ∧ rp = -rs t=0. a tanks-in-series model (number of tanks is 3).n=0 r=0. We took a plug flow model for the organic-solvent phase. Csp. centre of the gel bead ∂C s p.Chapter 6 Appendix A.s C s w. Mass balances for one tank.n=0 r=0.n ∂ 2 C s p.n ∂r =0 r =0 r=Rp. Mass transfer takes place from the medium phase to the gel beads for substrate. two components are present in the solid phase and the medium phase. Derivation of mass balances for a 3-phase fluidized-bed bioreactor In a 3-phase fluidized bed three different phases are present. Cpp.n − Cs w. and from medium to the organic solvent.

i.n = 0 The different transfer fluxes are represented by: ϕ wp s = k bead.n ) − ϕ wp s dt Csw. n − C p o. kbead. n −1 − C s w.pw are calculated with the equation of 159 .n = 0. n ) = −D ∂C p p. n − Cp w. batch operation and continuous operation The organic solvent phase Product V dC p εd r No dt o. n −1 − C s w.n p ∂r r=R p The medium phase Substrate ε w Vr dC s w.int. continuous operation t=0. Cpw.p d   w.n = 1.p. i. batch operation Product ε w Vr dC s w.i  1 1 = + w  k drop.n C p o.1. i.s (C s w. Csw. n φ wd p ( )d 6     −1 ε s Vr  6 V  εd r  d drop No  bead .int. n = U c A r (C s w. n − C s w. edge of the gel bead k bead.p C p w.Design of a 3-phase fluidized bed r=Rp.s. n = U d A r C p o. and kdrop.n − C p w.p C p ( w.p mk drop. i .n = 0.int. Cpo.i.n = U c A r (C s w.n Cp −  m  The mass transfer coefficients kbead.n ) 6 d bead ε s Vr φ pw p = k bead. n − ϕ wo p ( ) t=0. n ) − ϕ wp s dt t=0.

33 with vwp the slip-velocity between water and gel bead. and a dimensionless time Fo=Dst/Rp2.57    ηw    ρ w v wo d drop = 2 + 0.pd is calculated with the equation of Newman (1931).0393We noz 0.p d drop Dp  ρ w v wp d bead   = 2 + 0.73  ρ d gd noz 2   σ  2     − 0.33 d noz + 0.5  ηw  ρ D  w i         0. and vwo the slip-velocity between water and organic phase.2  ρ    with Eo = g∆ρd 2 σ ρ    η    1 − Eo    6   The mass transfer coefficient kdrop.55Eo noz with Eonoz = 0. 160 .drop d  n =1 drop  ∞     t represents the contact time and is equal to: t = Htεo/Ud The mass balances are made dimensionless by dividing with the initial substrate concentration. The resulting dimensionless parameters are: dimensionless length inside the bead z=r/Rp.and Wenoz = σ 2 /3 1/3 ρ v ∞ d noz σ The rise velocity of a single droplet is calculated with Vignes' equation: d  g∆ρ  v∞ =   4.i d bead Di k bead. dimensionless concentration C/Csw.5  ηw   ρwDp  0.57   ηw      0.315 g∆ρd noz 2 .0. k drop.33 0.p d = − d drop  6 ln  2  π 6t   1 t ∑ n 2 exp − 4n 2 π 2 D p. and with Ds/Rp2. The droplet diameter is calculated with the equation of Kumar and Hartland d= 0. the bead diameter.Chapter 6 Ranz and Marshall (1952): k bead.

n Ds ∂ z2 D p 2 ∂ X p p. To mimic the plug flow of the dispersed phase.i X = − Xp + K − Xp p ∂ Fo εd H t.n + ∂ Fo εw H t Ds d bead ε w Ds ( ) ( ) 2 6 ε o rp −K do εw Ds  X w.s = d bead ε w Ds ∂ Fo εwHt Ds ∂ X p w.n U w rp 6 εs w.n ) k bead.i 2 2 X p w.0 ∂ X p p. n = − Xp + Xp k bead.0 2 = 2 rp 2 ∂ X s w. n (X s − X s ) − (X s w. making the partial differential equation a set of ordinary differential equation (we took 31 grid points).n − X p w. n −1 w.n o.int.n ∂ Fo D p ∂ 2 X p p. 161 .n U d rp 6 rp  w.int.n = + − (. we took 15 tanks per medium tank.rs ) 2 ∂ Fo ∂z z ∂z D s C s w .n  X w.i −1 o.n rp 2 + + ( .n 2 rp 2 U w rp 6 εs w.n − X s w.i D s do Ds  m  ( ) i The space co-ordinate in the gel bead mass balance is discretized.p X p w.n − p  p m  i         ∂ X p o.n 2 ∂ X s p.Design of a 3-phase fluidized bed The resulting mass balances with the implementation of the transfer flux relations are: rp ∂ X s p. n −1 w.rs ) Ds z ∂ z D s C s w . Per medium tank 2 × 30 + 2 + 15 = 77 differential equations were solved simultaneously with a solver for stiff differential equations.n ∂ 2 X s p.

162 .

7 General discussion 163 .

both the aqueous phase as well as the organic solvent is recirculated with an external pump. and 3) an organic solvent. Bioreactor Design Hydrodynamics ⋅ fluidized-bed stability ⋅ phase hold-ups ⋅ phase mixing ⋅ mass exchange medium phase / gel beads medium phase / organic droplets Biotransformation ⋅ kinetics substrate consumption product formation ⋅ biocatalyst stability Figure 1. Stability of the bed refers to stable operation. Major topics for designing a 3-phase fluidized-bed bioreactor. which means that the gel beads remain in the fluidized bed and are not washed out. 2) a continuous aqueous phase that was used for fluidizing the gel beads. For the design of a 3-phase fluidized-bed bioreactor. This means that the hydrodynamics can be controlled with both the aqueous-phase velocity and the organic-phase velocity.Chapter 7 The results presented in this thesis make it possible to design a liquid-liquid-solid 3phase fluidized-bed bioreactor. i. gel beads with an immobilized biocatalyst. and for supply of substrates essential for the biotransformation. Hydrodynamics For a 3-phase fluidized-bed bioreactor. This fluidized bed consists of 1) a solid phase. In this study. 164 .e. Figure 1 summarizes the major topics to be considered for a conceptual process design. Important hydrodynamic topics are: 1) Bed stability. which rose through the fluidized bed of gel beads as organic droplets. the understanding of its hydrodynamic behavior is essential. The design of this bioreactor for a specific biotransformation depends on knowledge of the biotransformation itself and various physical aspects of the bioreactor. This organic phase was also used as a sink for the biotransformation product. we assumed the biotransformation data to be known.

With respect to mixing of the continuous phase. hold-up and mixing below. experimental results on gel-bead hold-up in a 2-phase liquid-solid (gel bead) fluidized bed are presented. and therefore the extraction potential. A phase hold-up is the fractional volume of a phase in the fluidized bed. In a 3-phase fluidized bed two distinct 2-phase systems are integrated: a liquid-solid fluidized bed and a spray column (a liquid-liquid extraction column). It was found that this hold-up could not be 165 . 2-phase systems In Chapter 2. 3) Phase mixing. it is shown in Chapter 5 that models predicting the axial dispersion coefficient in a spray column can be reliably used for predicting the axial dispersion coefficient in a liquid-liquidsolid 3-phase fluidized bed. In Chapter 4. Being able to predict the hydrodynamic characteristics of the individual 2-phase systems. Increasing the aqueous velocity was found to result in a decreasing gel-bead hold-up. the hydrodynamics of the individual 2-phase systems and the integrated 3-phase system will be discussed in some detail. For the integrated 3-phase system. we will discuss stability. Mixing characteristics of the separate 2-phase system were not determined. Operational stability of both 2-phase systems was good. The gel-bead hold-up determines the height of the fluidized bed at a given total volume of gel beads.General discussion 2) Phase hold-up. it is shown that this idea is only partly valid for predicting the different phase hold-ups. Chapter 2 deals with the liquid-fluidized bed and Chapter 3 deals with the spray column. Besides. For the 2-phase systems we will focus on hold-up. or the amount of gel beads present in the 3-phase system reduces to zero (a spray column). Hold-up determines also the specific area for mass exchange. The droplet hold-up thus determines largely the amount of extractive liquid present. In the following paragraphs. Hydrodynamic aspects of these constituent 2-phase systems were separately studied. mixing characteristics also influence mass exchange. provided the water velocity did not exceed the settling velocity of a single gel bead. Whether a phase shows plug-flow behavior or ideally mixed behavior can largely influence overall reactor performance. A 3-phase fluidized bed approaches a 2-phase system if either the amount of organic solvent present in the 3-phase system approaches zero (a liquid-solid fluidized bed). it was assumed that the two corresponding models could be combined to describe and predict hydrodynamic characteristics of the 3-phase fluidized bed.

In Chapter 2. and a rise velocity of a single droplet. was much lower than for more conventional solids such as glass pearls. Hence. This drag coefficient can successfully be used to predict pressure drop and gel-bead hold-up in a liquid-fluidized bed. 3-phase fluidized beds: stability A liquid-liquid-solid (gel bead) fluidized bed is created by introduction of an organic disperse solvent into a 2-phase liquid-solid fluidized bed. the slip-velocity was much higher. The drag force is largely determined by a drag coefficient. Chapter 3 shows the influence of sparger lay-out. rising droplets will transfer momentum to the gel beads. Apparently. is presented for the prediction of the drag coefficient of a single gel bead in a packed or fluidized bed. However. An established literature model for rise velocities is used. the experimental data can be well described. either present in a group or as single particle. If the amount of rising droplets is substantial. The droplet hold-up in a liquid-liquid extraction spray column can be well predicted with a Richardson and Zaki type of slip-velocity equation. featuring as model parameters an exponent n. gel beads were used with densities of 1006 – 1065 kg m-3. Wash out was indeed observed for gel beads with a density of 1010 kg m-3 and a diameter of 2 mm. it is also shown in which way the droplet hold-up can be manipulated by means of the sparger lay-out (number of nozzles and nozzle diameter).Chapter 7 predicted with the well-established slip-velocity model of Richardson and Zaki (1954). With this new model for n.e. for the exponent n a new model was derived. inertia) of the gel beads. Further analysis showed that the drag force between the continuous flowing phase (the medium) and gel beads. (1983). at a superficial solvent velocity of 166 . a more mechanistic model. the experimental hold-up data could be well described by the Richardson and Zaki equation if standard parameter values were adopted to fit the experimental data. In this study. if standard parameter values were used. Single droplets rising through this 3-phase fluidized bed of gel beads will hardly disturb the fluidized bed. mode of operation (counter-current or co-current operation) and both liquid velocities on the droplet hold-up. gel beads will be washed out from the bioreactor by these rising droplets. Gel beads consist over 95% of water. and obviously their surface properties are different from those of conventional solids. and wash-out of gel beads does not occur. these differences resulted in a lower drag force. Whether this phenomenon took place depended on the density (i. In Chapter 3. based on the model of Foscolo et al. Through collisions.

and before they can fall back again. two heterogeneous patterns (in these four areas. For gel beads with a very low density (1010 kg m-3). and they moved very slowly through it. was much more rapid and more chaotic. stable operation of this 3-phase system with low density gel beads is possible. while practical organic solvent velocities were between 0.35 cm s-1) and water velocities less than 0. however. For a fluidized bed of low density gel beads (1010 kg m-3). There were significant axial and radial distributions of gel beads and these distributions rapidly and randomly changed with time without a repetitive pattern. For this flow pattern. the solids were homogenously distributed over the bed. and stable operation of the 3-phase fluidized bed at all practical solvent velocities was possible. So far. A structured heterogeneous flow pattern of the solids was observed at intermediate organic solvent velocities (< 0. For gel beads of 1065 kg m-3. a radial concentration distribution of the gel beads was not observed. Coalescence of the organic solvent droplets occurred at relatively high organic solvent velocities (> 0. they are pushed once more. co-current flow does not yield a stable bed. the solids were homogenously distributed. this flow diagram demarcates five areas with characteristic flow patterns: two homogeneous patterns. Practical water velocities were between 0.01 cm s-1. At very low organic solvent velocities (much lower than 0. Axial concentration waves of gel beads along the column were present throughout the fluidized bed. this flow pattern of the solids was not disturbed.35 cm s-1) and relatively low water velocities (< 0.8 cm s-1. Stable operation of the analogue of 167 . and a freeboard of solids was hardly formed). the gel beads are transported out of the fluidized bed. a flow diagram is presented in Chapter 4. Thus. it was assumed that the organic solvent and medium flowed co-currently. and an unstable area in which the organic droplets coalesced and gel beads were washed out.6 cm s-1). the gel beads could be retained in the fluidized bed without wash-out only if the water flux was zero.05 and 0. In a 2-phase fluidized bed. By means of the superficial velocities of both liquid phases. gel beads remained in the fluidized bed. This way. except at extremely low velocities of the organic solvent. the organic droplets push the gel beads up. Their motion through the bed.05 cm s-1). At practically all organic solvent velocities. however. Also at relatively high water velocities.9 cm s-1. Wash out of the heavier gel beads (1065 kg m-3) was not observed at any flux. An irregular chaotic heterogeneous flow pattern of the gel beads was present at high organic solvent velocities and intermediate water velocities.General discussion 0.5 and 2 cm s-1.

and a new model was needed. is described in literature (Santos et al. At water velocities below 0. the droplet hold-up was always larger than in a 2-phase droplet column at the same phase velocity. it gives a good description for the whole range of measured hold-ups (0. if both velocities are increased in the same way. It was also shown that at any phase velocity. at water velocities above 0. As the amount of gel beads in these experiments was fixed. So. Obviously. while the extractive power increases.75 cm s-1. internal circulation of the aqueous phase (upward and downward flow) prevents the wash-out of the gel beads. This opens the way to a really interesting 3-phase fluidized bed. the net upward velocity of the aqueous phase in a cocurrently operated 3-phase liquid-liquid-solid fluidized bed is. Although a stable 3-phase fluidized bed was possible if only the organic solvent was pumped to the fluidized bed. the influence of the organic-solvent and medium velocities on the gel bead and droplet hold-ups is described in Chapter 4. the bed height was depending on the two phase velocities. The prediction of droplet hold-up. the volume of the bioreactor and the bioreactor activity remain the same. a gel-bead concentration gradient along the column height was observed. the bed height was smaller than in a liquid-solid fluidized bed at the same water velocity (bed contraction occurred).09). It was observed that an increase in both flow velocities did not affect the height of the fluidized bed. and coalescence of the organic droplets was not observed. at higher hold-ups. however. is only fair for hold-ups below 0.75 cm s-1 it was larger (bed expansion). Apparently. 168 . the droplet hold-up is strongly underestimated. in a 3-phase system without external recirculation of the aqueous phase.bead hold-up is well described by the model. an empirical model was derived. Hold-ups in a co-current 3-phase fluidized bed For high density gel beads (1065 kg m-3). a 3-phase bubble column. filled with gel beads. The gel.Chapter 7 this system. A stable operation of this 3-phase system was possible. responsible for wash out of low density gel beads. As an alternative.005 → 0. This concentration gradient could be strongly reduced by applying a downward countercurrent aqueous flow. among other things. The new model for gel-bead and organic-droplet hold-up is based on the drag forces for a gel bead and a droplet present in the 3-phase mixture (Chapter 4).06. the individual models for both 2-phase systems are not additive. a packed bed was not formed. 1997). The gel beads were kept in suspension.

Since droplets with the same diameter will have an equal rise velocity through the 3-phase fluidized bed. the axial dispersion coefficient is much higher for a 3-phase fluidized bed and can be very well predicted with the isotropic-turbulence theory of Baird and Rice (1975).General discussion To calculate the drag forces. Models for drag coefficients for the individual 2-phase systems were extended to model the drag coefficients of a gel bead and a droplet in the 3-phase fluidized bed. Future work should elucidate the movement of gel beads throughout the bed and its impact. mass exchange of substrates and products between the different phases of the 3-phase fluidized bed is vital for the design of the bioreactor. As stated above. but a preliminary indication may be obtained from predictions of solvent hold-up. Indeed. When the parameters of the individual 2-phase models are used. 1992). which was easily verified on sight. this is a simplification. plug flow of the droplet phase may be concluded. Also the mixing of the organic solvent droplets was not determined. For the design of the 3-phase fluidized bed. It is concluded that mixing of the continuous phase in a 3-phase fluidized bed is comparable with mixing of the continuous phase of either a bubble column or the main tube of a liquidimpelled loop reactor (Van Sonsbeek et al. To predict droplet hold-up. the circulation velocity of individual gel beads through the fluidized bed. The successful prediction of solvent hold-up was in support of the assumption of a single droplet diameter. It may be monitored with radio-active labeling and appropriate sensors. Mass transfer Besides hold-up and mixing. the gel bead hold-up is very well described. mixing characteristics of the gel beads in a 3-phase fluidized bed were not studied. a drag coefficient is needed. were not determined. The results can be transformed to generic flow patterns of the gel beads in the fluidized bed. but the droplet hold-up is systematically underestimated. It is much more turbulent than mixing in a 2-phase fluidized bed. Mass transfer in the fluidized bed takes place at two different levels: 1) transfer in the 169 . Obviously.e. Mixing characteristics of the gel beads. Mixing Mixing of the continuous phase in a 3-phase fluidized bed was studied with highdensity gel beads (Chapter 5). we assumed the gel beads not to move. a single droplet diameter was assumed. i..

Mass transfer through the outer boundary layer of gel beads or droplets in a 3-phase fluidized bed. determines the external mass transfer coefficient. 1991). The same transfer path holds for gel beads. which can be determined with the hold-up model given in Chapter 4. Further research should substantiate this idea. In Chapter 5 it is shown that mixing of the 3-phase fluidized bed can be adequately described with the well-established model of Baird and Rice (1975) using the energy dissipation rate. but. Experiments should yield the external mass transfer coefficients for the transfer from the continuous liquid phase to either a droplet or a gel bead surface. and mixing determines the extent of the boundary layer. and 2) transfer within a droplet or gel bead.Chapter 7 continuous phase through a droplet or gel bead boundary layer. To determine the mass transfer coefficient from bulk liquid to gel bead surface one could use a tracer component that is exchanged between the bulk liquid and the gel beads. does not dissolve in the organic solvent. So. as for gel beads mass transfer in 3-phase systems (gas-liquid-solid or liquid-liquid-solid) has not been elucidated. we used the wellknown equation of Ranz and Marshall (1950) for calculation of mass transfer coefficients from the bulk liquid to the surface of either a droplet or a gel bead. Other models use the energy dissipation rate in the fluidized bed for determination of mass transfer coefficients (Kikuchi et al. The energy input determines the mixing of the aqueous phase.. This method is also widely used for the determination of mass transfer in stirred tanks and bubble columns (Van ‘t Riet and Tramper. at the same time. It is questionable. In the next paragraph we will focus on these mass transfer phenomena. 170 . The size of the boundary layer. it is expected that the energy-input determines the external mass transfer coefficient. Mass transfer within particles or droplets has already been studied extensively in literature and may be adequately predicted from established equations. also in this study. 1994). finally. In the design of the 3-phase fluidized bed (Chapter 6). This equation uses the slip velocity between the different phases. So. whether the empirical constants in this mass transfer correlation may be applied also for mass transfer in the liquid-liquid-solid (gel bead) fluidized bed bioreactor. In a 3-phase fluidized bed there is simultaneous transfer from the aqueous phase to both the organic droplet and the gel bead. external mass transfer coefficients may be adequately determined using established correlations based on the energy dissipation rate. has not been studied yet. on the contrary. one would like to determine the external transfer coefficients independently. such mass transfer experiments have not been done. however.

a tracer component that is not transferred to the gel beads might be hard to find. In our design. and reliable models have been derived. hold-up of the different phases. Operational stability of the fluidized bed. However. and mixing of the continuous phase as a function of operational variables (superficial velocities of the organic solvent and medium) are well understood. their terminal settling velocity should exceed 5. Transfer experiments with such a component would thus involve two simultaneous transfer coefficients. with respect to stability of the fluidized be we concluded that to prevent wash-out of the gel beads. based on our experience with the fluidized bed. Experimental methods for determination of these constants are given. However. a specific type of gel bead was used. The other extreme would be an ideally mixed bed of gel beads. Nevertheless. It is also shown that if an 171 . A generalization of the behavior can thus not be given. as the gel beads consist for over 95% of water. Summary and Future prospects The work in this thesis aimed at design rules for a 3-phase liquid-liquid-solid (gel bead) 3-phase fluidized-bed bioreactor. An alternative would be to use a component that is exchanged between the continuous aqueous phase and both the gel beads and the organic droplets. Mixing behavior of the gel beads is still unknown. For mass transfer and mixing of the organic solvent phase. the coefficient for transfer into the organic phase may be calculated from experimental data if the value for the coefficient of transfer into the gel beads would be available from preceding experiments as described above. we assumed that the gel beads did not move. The actual mixing of the bed will be in-between both extremes. To elucidate the hydrodynamic behavior of the 3-phase fluidized bed. educated guesses can be made. which. and the correlations for mass transfer coefficients. given in Chapter 6. A full design of such 3-phase fluidized bed bioreactors is only partially possible with the results presented in this thesis.General discussion Determination of the external transfer coefficient from the continuous aqueous phase to the droplet surface can use a similar set-up (Van Sonsbeek et al. Future work should address the mixing characteristics of the gel beads. The model for the hold-up in the 3-phase fluidized bed contains empirical constants. 1991).0 cm s-1. in principle.. have to be measured for each type of gel beads.

Chapter 7 energy dissipation rate is available. a bioconversion hindered by strong product inhibition should be performed with this bioreactor. mixing of the continuous phase may be straightforwardly determined. one could consider using a 3-phase fluidized bed bioreactor for performing substrate or product inhibited biotransformations. it is shown that the performance of the 3-phase fluidized-bed bioreactor is influenced by the distribution coefficient of the product over the organic solvent and medium phase. 172 . It appeared that a distribution coefficient larger than 2 suffices for a better performance of this 3-phase fluidized bed. In Chapter 6. In order to demonstrate that the newly developed bioreactor does also perform better for practical systems. as this thesis elucidated the most important physical aspects. and showed the operational suitability of the bioreactor. In future.

173 .

174 .

Heidstra R and Tramper J. biocatalysis & process integration. Ed Tramper J. (Dordrecht). Straathof AJJ and P. Analyse. Boon MA. Boeren S. Delftse Uitgevers Maatschappij (Delft). Buitelaar RM. Bird RB. Bruggink A (ed). The effect of alkanes on viability. Almering JHJ (Ed. Optimization of organic solvent in multiphase biocatalysis. 2001. Axial dispersion in large unbaffled columns. Enzymatic synthesis of oligosaccharides: product removal during a kinetically controlled reaction. 2nd edn Harwood Academic Publishers. Kluwer Academic Publishers. Langenhoff AAM. Vermuë MH. 1975. enzyme induction and enzyme activity in Flavobacterium dehydrogenans. Brink LES and Tramper J. Beeftink HH and Von Stockar U. 1988. (Singapore). 1982. Stewart WE and Ligthfoot EN. 175 . 2000. Baird MHI and Rice RG. p 295 – 316. Van ‘t Riet K and Janssen AEM.References References Adlercreutz P. Biotechnol Bioeng 70:411-420. Elsevier Science Publ (Amsterdam). Segregation of liquid-fluidized solids. Transport Phenomena. 1985. 2000. p 637 – 643. In Applied Biocatalysis.). Chemistry. Synthesis of beta-lactam antibiotics. Laane C and Hilborst R. In Biocatalysis in Non-conventional media. Enzym Microb Techn 13:487 – 494. John Wiley & Sons (Singapore). 1991. p 200. Adlercreutz. Duijn G and Rietema K. Chem Eng Sci 39:365-368. Chem Eng J 9:171-174. Biotechnol Bioeng 27:1258 – 1269. 1992. p 356 359. 1960. Growth and thiophene production by hairy root cultures of tagetes patula in various two-liquidphase bioreactors. Ed. Biocatalysis in non-conventional media.

Air-lift Bioreactors.fr/fr/Origine/cdindetf. Micrometrics: the technology of fine particles. Subrahmanyam V. Pitman (London).html Dalingaros W. 1990. Davison BH and Thompson JE. 1979. 1986. De Bont JAM. Chem Eng Sci 45:935-945. Chisti MY (ed. Bed porosities in 3phase (liquid-liquid) fluidized beds. Measurement of individual fluid-phase resistances. Elsevier Science Publishers BV (London). Liquid-liquid mass transfer in 3-phase fluidized beds. 1989.References Chen Y-M and Fan L-S. Effect of physical properties and dispersed-phase velocity on the size of drops produced at a multi-nozzle distributor. Prasada Rao RV and Vijayasaradhi P. Collins AM.coutouly. Appl Biochem Biotech 39/40: 415-426. Ind Eng Chem Proc Des Dev 23:132 – 137 Dallavalle JM. J Ind Microbiol 14:382-388 Coutouly. Hydrodynamics of liquid fluidisation. Drift flux in gas-liquid-solid fluidized systems from the dynamics of bed collapse. 1993. 2nd Edn. Chem Eng Process 20:95-102. Trends Biotechnol 12:493-499 Di Felice R. http://gerard. Chem Eng Sci 50(8): 12131245. Ind Eng Chem Proc Des Dev 18:638 – 640. Solvent-tolerant bacteria in biocatalysis. Determination of reactor operation for the microbial hydroxylation of toluene in a two-liquid phase process. Kumar A and Hartland S. 1995. Dakshinamurty P. Woodley JM and Lidell JM. 176 .). Continuous direct solvent extraction of butanol in a fermenting fluidized-bed bioreactor with immobilized Clostridium acetobutylicum. 1984. 1995. Veerabhadrarao K and Venkatarao AB.free. 1998. 3. Dakshinamurty P. 1948.

1991. Wageningen Agricultural University. p 637 – 643. 1982. 150. Chem Eng Process 22:69-78. Crelier S. Slip velocity relationships for liquid-liquid extraction columns. 1982. The Netherlands. An unified model for particulate expansion of fluidized beds and flow in fixed porous media. Wageningen. Fluidized-bed bioreactors. 1992. Foscolo PU. Garside J and Al-Dibouni MR. 1977. Parameter estimation from tracer response measurements. A survey of continuous ethanol fermentation systems using immobilized cells. Velocity-voidage relationships for fluidization and Sedimentation in solid-liquid Systems. Process Biochem 4:43-48. Beeftink HH and Von Stockar U. Vermuë MH. An experimental study of expansion characteristics of gas fluidized beds of fine catalysts. Gibilaro LG and Waldram SP. Hydrodynamic characteristics of inverse fluidization in liquid-solid and gas-liquid-solid systems. In Biocatalysis in non-conventional media. Di Felice R and Gibilaro IG. Elsevier Science Publishers BV. Gòdia F and Solà S. Ind Eng Chem Process Des Dev 16:206214. 1983. 1987. Muroyama K.References Krab-Hüsken LE. Casas C and Solà C. Fahim MA and Wakao N. Chem Eng Sci 38:12511260. Chem Eng J 25:1-8. Process development for the optical resolution of phenylalanine by means of chymotrypsin in a liquid-liquid-solid 3-phase reaction system. PhD thesis. Huneke F-U and Renken A. (Amsterdam). Production of catechols: microbiology and technology. Godfrey JC and Slater MJ. Biotechnol Prog 11:479-497. Gòdia F. Trans IchemE 69:130-141. Foscolo PU. and Chern S-H. Ed Tramper J. Schulz K. Fan L-S. Flaschel E. 1987. 1995. 2002. Chem Eng J 24:143 – 177 .

Hadzismajlovic DZE and Jovanvic S. and Weltevrede R. Hunik JH and Tramper J. Biotech Progress 9:186-192. 1991. 1985. Principles of colloid and surface chemistry. John Wiley and Sons (New York). 1986. Hulst AC. Bioprocess Eng. Kakoi K and Matsumoto T. Chem Eng Sci 52:2527 – 2540.and 3-phase internal airlift reactor operating in the gas recirculation regime. Hunik JH.) 1994. Variational model for prediction of the fluid-particle interphase drag coefficient and particulate expansion of fluidized and sedimenting beds. Appl Microbiol Biotechnol 25:1-7. p 288-293. 11:73-82. Van der Lans GJM. Hidaka N. 178 . Predicting the free-fall velocities of spheres. Tramper J. Correlation of solid hold-up in liquidsolid fluidized bed. Liquid-liquid extraction equipment. Mulder A. J Chem Eng Jpn 27:563-570. Christoffels LP. Hartman M. 1989. Heijnen JJ. A strategy for scale-up of nitrification processes with immobilized nitrifying cells. Hols J. 1997. Garic RV. Grbavcic ZB. 2nd Marcel Dekker Inc (New York). 1986. 1994. Tramper J and Wijffels RH. Gas-phase influence on the mixing in a fluidized bed bio-reactor. Hiemenz PC (Ed). Trnka O and Carsky M. Biotechnol Bioeng 27:870-876. A new technique for the production of immobilized biocatalyst in large quantities.. 1993. Van ‘t Riet K and Westerbeek JMM. Chem Eng Sci 44:1743-1745. Van Leeuwen HLJM. Powder Technol 68:199-211. Kuenen JG and Luyben KChAM. Large Scale production of κ-carrageenan droplets for gel bead production: theoretical and practical limitations of size and production rate.References Godfrey JC and Slater MA (Ed. 1994. A simple hydrodynamic model for the liquid circulation velocity in a full scale two. Havlin V. Gommers PJF.

Chem Eng Sci 52:3639 – 3660. Itoh E. Tramper J. Dalm MCF. Chem Eng Commun 68:57-68. Integrated bioproduction and extraction of 3-methylcatechol. Phase holdups and liquid-liquid mass transfer in three phase fluidized beds. Tramper J. Axial dispersion characteristics of three (liquidliquid-solid) phase fluidized beds. J Biotechnol 96: 281-289. 2001. Kawakami K and Yosida T. 1994. Kim SD. Kim SD and Kim CH. Heat and mass transfer in 3-phase fluidized-beds reactors – an overview. 1989. Kikuchi K. Vermuë MH. Wery J. The resistance to motion of a solid sphere in a fluid. Membrane-facilitated bioproduction of 3-methylcatachol in an organic/water 2-phase system. Axial dispersion characteristics of 3-phase fluidized beds. Hüsken LE. Han PW and Yu YH. Kinetic study of enzymatic reaction in aequous-organic 2-phase systems – an example of enhanced production of aldehydes by alcohol oxidase. Schroën CHGP. Can J Chem Eng 67:276 – 282. 1994. Kim DY and Han JH. Ed Tramper J. 637 – 643. Chem Eng J 27:674-677. 179 . p. Kim SD. 1992. Kim SD. Takahaski H and Sugawara T. J Chem Eng Japan 16:172 – 178. (Amsterdam). 1987. Can J Chem Eng 72: 222-228 Kim SD and Kang Y. De Bont JAM and Beeftink HH. Dispersed phase characteristics in 3-phase (liquid-liquid-solid) fluidized beds. Elsevier Science Publishers BV. 1983. J Biotechnol 88:11-19. 1988. Kahn RA and Richardson JF.References Hüsken LE. Oomes M. De Bont JAM and Beeftink HH. Chem Eng Commun. 2002. Beeftink HH and Von Stockar U. 1997. In Biocatalysis in Non-conventional media. 62:135-150. Mass transfer between particles and liquid in dilute fluidized beds containing small particles. Lee MJ and Han JH.

Cabral JMS. Leenen EJTM. Michielsen MJF. Bio/Technology 13:1072-1078. Kumar A and Hartland S. Leenen EJTM. Marshall CT and Woodley JM. Ed.). Taylor and Francis (London. A unified correlation for the prediction of dispersedphase hold-up in liquid-liquid extraction columns. Wijffels RH. Can J Chem Eng 63:368-376. Alves SS and Da Fonseca MMR. Wijffels RH. 180 . Fundamentals of gas-liquid-solid fluidization. 1985. 1999. Muroyama K and Fan L-S. Mota M and Tramper J. Chemical reaction engineering. In Immobilized Cells. Springer Lab Manual (Berlin). Mechanical Stability of the Support. Chapter 8 Solid-to-solid bioconversions: batch or continuous?. Ed. Kumar A and Hartland S. Nam CH. Kumar A and Hartland S. Immersed heater-to-bed heat transfer in liquid-liquid-solid fluidized beds. In Immobilized Cells. AIChE J 31:1-34. John Wiley ( New York). 1995. Ind Eng Chem Res 35:2682-2695. Wijffels RH. Mateus DMR. Ed. Gravity settling in liquid/liquid dispersions. 2001. 1984. Springer Lab Manual (Berlin). Model for the production of Ltryptophan from L-serine and indole by immobilized cells in a 3-phase liquidimpelled loop reactor.References Kim SD. 1995. 2001. Support Material Stability at the Process Conditions Use. Kim SH and Kang Y. Kim JS. New York). Process synthesis for multi-step microbial conversions. 1972. In Multi-phase bioreactor design. 1996. Unified correlations for the prediction of drop size in liquid liquid extraction columns. Ind Eng Chem Res 34:3925-3940. Bioprocess Eng 14:151 – 158. 2001. Chem Eng Sci 54:5173 – 5179. Tramper J and Beeftink HH. 1996. Levenspiel O (Ed. p 225247.

US Patent US 5728556. 1977. Trans Inst Chem Eng 32:35-53. 173 – 180. Cocurrent liquid-liquid extraction in fluidized beds. 1950. A simplex method for function minimization. 1979. Evaporation from drop I + II.ω-dinitrile(s) – using biocatalyst having regioselective nitrile hydratase activity derived from Pseudomonas putida. Wubbolts M and Witholt B. Newman AB. 1996. Chem Eng J 17:101-109. 181 . Schmid A. 43:2795 – 2796. Stieglitz B. J Ferment Bioeng 80:185-189. Kiener A. Riba JP and Couderc JP. Ranz WE and Marshall WR.References Nelder JA and Mead R. Richardson JF and Zaki WN. Part 1. A convenient model empirical equation for estimation of the Richardson-Zaki exponent. Sedimentation and fluidization. Roszak J and Gawronski R. 1987. Dordick JS. Dicosimo R and Fallon RD. The drying of porous solids: diffusion and surface emission equations. 2001. Expansion de couches fluidizées par des liquides. Formation of aliphatic ω-cyanocarboxamide(s) from α. Hauer B. Computer J 7: 308-313. Industrial biocatalysis today and tomorrow. Continuous production of Acetone-butanol-ethanol Using Immobilized Cells of Clostridium acetobutylicum and Integration with product removal by liquid-liquid extraction. Can J Chem Engng 55:118-121. Paireau O and Bonn D. Trans Am Int Chem Engrs 27:203 – 220. Drag reduction in liquid-liquid friction. Nature 409:258-268. Rowe PN. Chem Eng Sci. Chem Eng Progr 48: 141 – 146. 1954. Qureshi N and Maddox IS. 1999. 1995. Physical Review Letters 83:5591-5594. 1965. 1931.

The effect of tracer diffusion on liquid dispersion in a three phase fluidized bed bioreactor. Tramper J and Lilly MD. Tramper J and De Bont JAM. Turton R and Clark NN. Ed. In Bioreactors and biotransformations. 1996. Moody GW and Baker PB. Tramper J. 1992. 231 – 241. Chem Eng Sci 47:3435 – 3441. A short note on the drag coefficient for spheres. Wolters I and Verlaan P. Vermuë MH. Van den Tweel WJJ. PhD thesis. 1993. Vorage MJAW. 182 . 1987. Ueyama K and Miyauchi T. Ed. Van Dam MHH and Luyben KChAM. Marsman EH. Multi-phase fermentation: reaction engineering models for the production of optically active 1. 1992.2-epoxyoctane from 1-octene by free and immobilized Pseudomonas oleovorans cells. Tramper J. Turton R and Levenspiel O. Elsevier Science Publishers BV (Amsterdam). Laane C.References Thompson VS and Worden RM. Elsevier Science Publishers BV (Amsterdam). Chem Eng Sci 51:995-1008. Progress in Biotechnology 8. AIChE J 15:258-266. The liquid-impelled loop reactor: a new type of density-difference-mixed bioreactor. Biocatalysis in non-conventional media. On the relative motion of a particle in a swarm of different particles. Van der Wielen LAM. p. Beeftink HH and Von Stockar U (Eds. Properties of recirculating turbulent 2-phase flow in gas bubble columns.). The application of organic solvents for the bioconversion of benzene to cis-benzene glycol. Elsevier Science Publishers BV (London). 1979. 1986. 1987. 1987. Van der Meer AB. Groningen. In Biocatalysis in organic media. The Netherlands. University of Groningen. p 311 – 316. Powder Technol 53:127-129. An explicit relationship to predict particle settling velocity. Powder Technol 47:83-86.

Delft. Van der Wielen LAM. Rinzema A and Tramper J. PhD thesis. The Netherlands. Basic Bioreactor Design. Biotech Techn 5:157-162. 2nd edn. Van Dam MHH and Luyben KChAM. Van Sonsbeek HM. 1992a. Wageningen. Biotechnol Bioeng 40:713 – 718. Physical aspects of liquid-impelled loop reactors. 1996a. 1991. Wageningen Agricultural University. Van Sonsbeek HM. Chem Eng Sci 52: 553-565 Van der Wielen LAM. A countercurrent adsorptive reactor for acidifying bioconversions. Chem Eng Sci 51:2315-2325. 2003. De Blank H and Tramper J. Marcel Dekker Inc (New York). Van Zessen E. PhD thesis. Phase hold-ups in a water-dodecane 2-phase system. Van Zessen E. 1991. Oxygen transfer in liquidimpelled loop reactors using perfluorocarbon liquids. Houwers J and Luyben KChAM. Trickle flow of dense particles in a fluidized bed of others.). Van ‘t Riet K and Tramper J. submitted (Chapter 3) 183 . Verdurmen REM. submitted (Chapter 2). 2003. Enzyme Microb Technol 15:722 – 729. Gielen SJ and Tramper J. 1992b. 1990. 1993. (Ed. Technical University Delft. Hydrodynamic model for liquid-impelled loop reactors. The Netherlands. Rinzema A and Tramper J. Beeftink HH and Tramper J. 1997. Van Sonsbeek HM. Van Sonsbeek HM. Steady state method for kla measurements in model systems. A countercurrent Adsorptive Fluidized Bed Reactor for Heterogeneous Bioconversions. 1996b. Van Sonsbeek HM. Diepen PJ. Verlaan P and Tramper J. Fluidized-bed and packed-bed characteristics of gel beads. Biotech Bioeng 36: 940 – 946.References Van der Wielen LAM. Two-Liquid-phase bioreactors.

Ed Wijffels R. Wageningen. T. Fluidization of solid particles. Vermuë MH and Tramper J. In Immobilized Cells. and Tramper J. Wallis GB (Ed. Pure & Applied Chem 67:345-373. Springer-Verlag (Berlin). E. E. 59th Edition. Bakker R. 1969. Wilhelm RH and Kwauk M. Biocatalysis in non-conventional media. In Biocatalysis in non-conventional media thermodynamic and kinetic aspects.. 1994. PhD thesis. Chem Eng Prog 44:201-217. Van Zessen E. 2001. submitted (Chapter 5). 1995. Measurement of density. Strategy for selection of a suitable solvent for extractive biocatalysis in a liquid-impelled loop reactor.. Tetralin and oxygen transfer in the liquid-impelled loop reactor. Tramper J and Rinzema A. Keijzer. 184 .References Van Zessen. One-dimensional 2-phase flow. The Netherlands. van. particle size and shape of support. Vermuë MH. Dierendonck. McGraw-Hill (New York). Handbook of Chemistry and Physics. Van Zessen. submitted (Chapter 4). medium engineering aspects – a review. Weast RC. Tacken M and Tramper J. Wang H. 1995. Seki M and Furusaki S. (2003) Solid and droplet phase hold-ups in a liquid-liquid-solid 3-phase fluidized bed bioreactor. Wageningen Agricultural University. 1994. Characteristics of immobilized Lactobacillus Delbrueckii in a liquid-solid fluidized bed bioreactor for lactic acid production. F.). Vermuë MH. CRC Press Boca Raton. Vermuë MH. Rinzema A. Rinzema A. 1979. and Tramper J. (2003) Residence time distribution of the water phase in a liquid-liquid-solid 3-phase fluidized bed bioreactor. 1948. Janssen G and Tramper J. J Chem Eng Jpn 28:198-203. Bioprocess Eng 11:224 – 228. Springer Lab Manual. Sikkema J.

Van Baron G and De Backer L. 1995. AIChE J 27:1043-1044. An explicit equation for particle settling velocities in solid-liquid systems. Bucke C and Tramper J (Ed.References Willaert RG. Wijffels RH. Immobilized cell systems. Yu H and Rittmann BE. p. 67-97. Wat Res 31: 2604 – 2616. John Wiley & Sons (New York). In Immobilized living cell systems modeling and experimental methods. 1981. 185 .). Immobilized Cells: Basics and Applications. Buitelaar RM. Elsevier (Amsterdam). Zigrang DJ and Silvester ND. 1996. Predicting bed expansion and phase holdups for 3phase fluidized bed reactors with and without biofilm. 1997.

186 .

however. The solid reaction phase were biocatalytic gel beads that were fluidized by the continuous medium phase. transformations featuring an unfavorable thermodynamic equilibrium. consisting of organic droplets in a continuous medium phase. To facilitate integration of reaction and product removal or substrate supply. a new type of multi-phase bioreactor was developed. a low overall productivity may result for certain types of biotransformations. through this fluidized bed. a dispersed liquid phase. and a liquid-liquid extraction column or spray column. containing a continuous liquid phase. two 2-phase subsystems may be identified: a liquid-solid fluidized bed consisting of biocatalytic gel beads in a continuous medium phase. 187 . For the phase hold-ups in the 3-phase system. An alternative to sequential processing would be the integration of bioconversion and downstream processing by controlled supply of substrate to the reaction in situ. serial) downstream processing are adopted as a standard production strategy. Within this 3-phase system. a general design strategy was developed. The hydrodynamic characteristics of these constituent 2-phase systems were studied separately (Chapters 2 and 3). Examples may be found among reactions that are kinetically inhibited by a substrate or a product. Once good descriptions and predictions of the hydrodynamics of these individual 2-phase systems were obtained. Such integration involves a multi-phase reactor in which a helper phase may serve as substrate reservoir or as product sink. it was assumed that the two corresponding models could be combined to describe and predict the hydrodynamic characteristics of the 3-phase fluidized bed. and bioconversions involving poorly soluble substrates or products. it may be conceived of as a 3-phase liquid-liquid-solid fluidized bed. could be reliably predicted from a model for this dispersion coefficient in a 2-phase droplet column (Chapter 5). and a dispersed solid phase.e. This thesis focuses on the hydrodynamic aspects of such a 3-phase fluidized bed. or by controlled removal of product from the reaction in situ. The continuous-phase axial dispersion coefficient of the 3-phase fluidized bed. organic droplets rose and served as a helper phase (extracting the product or acting as a substrate sink).Summary Summary If a batchwise bioconversion and subsequent (i. this assumption was only partly valid (Chapter 4).

Although water and dodecane superficial velocities were within the same order of magnitude. The droplet hold-up could be well described using a model based on the slip velocity between the continuous phase and disperse phase. A new model was presented that correctly predicted the drag coefficient of a single gel bead in a packed or fluidized bed. For five types of gel beads of different diameter and density. established models did not quantitatively predict It is concluded that the drag force between the gel beads and liquid was lower than for conventional solids. three related characteristics were analyzed that all involved the drag force between a particle and its surroundings: the terminal settling velocity of a single gel bead. Two hypotheses were put forward to explain this. Hence. a decrease in n was found. the gel bead surface might show waterlike properties. for which a new model was presented.13. they were either settling as single beads. or present in a packed or fluidized bed (Chapter 2). this parameter was found constant (n=4). The continuous phase (water) and the dispersed phase (dodecane droplets) were fed into the column co-currently and counter-currently.Summary The physical behavior of gel beads was studied. the pressure drop over a packed bed. Integration of the models for a solid/liquid 2-phase system (a fluidized bed) and a liquid/liquid 2phase system (a droplet spray or extraction column) was assumed to yield a potential 188 . Droplet hold-ups in a liquid/liquid 2-phase spray column were studied with different sparger geometries (Chapter 3). It was used to predict the pressure drop over a packed bed of beads and the voidage in a fluidized bed of beads. however. Although gel beads showed similar characteristics as conventional solids. the sparger with the lowest number of nozzles showed the highest hold-up. the water velocity had only a minor influence on the dodecane hold-up. for higher hold-ups. influenced the dodecane holdup to a large extent: at the same dodecane superficial velocity. Up to a hold-up of 0. The first one attributes the drag reduction to small amounts of dissolved polymer. The validity of this model was demonstrated by extending it to a 2-phase system differing in surface tension (60 mM KCl /dodecane). The number of nozzles. The slip velocity obeyed a Richardson and Zaki type equation. It features a parameter n. and the voidage in a liquid-fluidized bed. The second one attributes the smaller drag force to the nature of the gel beads. The integration of a solid (gel bead)-liquid fluidized bed with a droplet spray column results in a liquid-liquid-solid 3-phase fluidized bed reactor. gel beads contain over 95% of water and can in that sense be regarded as ‘rigid’ water droplets.

Consequently. A literature relationship for this dependency was found to yield good predictions of the dispersion coefficient with some minor exceptions. 189 . Mixing was characterized by an axial dispersion coefficient that was determined from residence-time distribution measurements with a KCl-tracer. To this end. a new model for a fluidized bed with gel beads and dodecane droplets had to be derived. A pulse with a small volume was found to give reliable results. The study of the continuous phase mixing in the 3-phase fluidized bed is described in Chapter 5.06 were well described. The different phase hold-ups of this 3-phase fluidized bed were studied as well as the flow behavior of the fluidized bed (Chapter 4). a disperse organic phase (ndodecane droplets) rose through the bed of particles. the various phase hold-ups were studied. Unfortunately. in addition. From visual observation of the fluidized bed. All data for gel bead hold-up and the data for droplet hold-ups below 0. Because of KCl mass transfer into the gel beads. based on stationary force balances. For the description of the entire range of droplet hold-ups. while its dependency on the continuous-phase superficial velocity showed a maximum. an empirical model was adopted. ‘water’). It is concluded that the mixing of the continuous phase in a liquid-liquid-solid 3-phase fluidized bed is comparable with the mixing of the continuous phase of either a bubble column or the main tube of a liquid-impelled loop reactor. Prediction with other literature models was also poor. the inherent limitations of this tracer had to be studied as well. The reactor featured a bed of Celite™-enforced κcarrageenan gel beads as solid phase (density 1065 kg m-3). the flow characteristics of the gel beads in the fluidized bed were characterized. and an unstable regime in which gel beads were washed-out from the bed. The more fundamental literature models for existing 3-phase systems rely on parameters valid for the individual 2-phase systems. the prediction with these models was poor using parameters derived for the individual 2-phase systems.Summary description of a liquid-liquid-solid 3-phase fluidized bed reactor. The continuous-phase axial dispersion coefficient was found to increase with the superficial n-dodecane velocity. These results reflect the increase in continuous-phase dispersion coefficient with the energy dissipation rate per unit mass of continuous phase. two stable heterogeneous regimes. Literature models for existing 3-phase systems were tried to predict phase hold-ups. drag-force models for a single gel bead and for a single droplet in a 3-phase suspension were set up. The bed was fluidized by a continuous aqueous phase (30 mM KCl. At various velocities of water and n-dodecane. five flow regimes were defined for the gel beads: two stable homogeneous regimes.

Summary A simple process design for a 3-phase liquid-liquid-solid fluidized bed bioreactor was made (Chapter 6). therefore. the influence of different parameters was discussed. On the basis of the model for the 3-phase fluidized-bed bioreactor. a simple product-inhibited reaction was assumed. the 3-phase fluidized bed performs better than the 2-phase fluidized bed. From a practical viewpoint. Operational conditions were established at which the 3-phase reactor would outperform a conventional 2-phase fluidized bed. It was based on the results from the preceding chapters. 190 . it was concluded that implementation of a 3-phase system would require minor adaptations only as compared to a 2-phase fluidized bed bioreactor. Additional steps that have to be taken for practical implementation of the 3-phase fluidized bed as a new type of bioreactor are emphasized. The thesis finally reviews the current status of the newly developed 3-phase liquidliquid-solid fluidized-bed bioreactor. For conversions that would benefit from in situ extraction. on educated guesses for mass exchange and flow behavior of both dispersed phases (gel beads and droplets). the product toxicity. It was demonstrated that application of a 3-phase reactor system resulted in a higher degree of conversion than 2-phase liquid-solid fluidized bed reactors. such as the distribution coefficient of the product over the medium and solvent phases. it outlines additional research to be performed for a full understanding of its hydrodynamic behavior. Already at quite modest values of the distribution coefficient. a 3-phase fluidized bed would be the apparatus of choice. and the solvent flux.

191 .

dat bestaat uit de biokatalytische deeltjes in een continue vloeistoffase. Een alternatief voor de normale productiestrategie kan de directe integratie in één apparaat van bio-conversie en opwerking zijn. die bestaat uit druppels van een organisch oplosmiddel in een continue vloeistoffase. 2) bioconversies die sterk beperkt worden door een ongunstige evenwichtsligging. Nadat een goede beschrijving en voorspelling van het hydrodynamische karakter van deze twee afzonderlijke 2-fase systemen was verkregen. werd er verondersteld dat de corresponderende modellen konden worden gecombineerd om het hydrodynamische karakter van het 3-fase gefluïdiseerd bed te kunnen beschrijven en 192 . zij extraheren het product of zij dienen als substraatreservoir. dat wil zeggen eerst een batchgewijze fermentatie en vervolgens de opwerking van het gewenste product. Voor een dergelijke integratie is een meerfase reactor noodzakelijk. In een 3-fase systeem zijn er twee 2fasen systemen te onderscheiden: een vast-vloeistof gefluïdiseerd bed.Samenvatting Samenvatting Indien voor een aantal typen fermentaties een normale productiestrategie wordt toegepast. Door dit gefluïdiseerde bed stijgen druppels organisch oplosmiddel op. Deze druppels treden op als de hulpfase. De vaste fase bestaat uit biokatalytische gel deeltjes. die worden gefluïdiseerd door de continue vloeistoffase. en een vloeistof-vloeistof extractie kolom. dan zal dit resulteren in een lage productiviteit voor het gehele proces. waarbij een hulpfase gebruikt wordt als substraatreservoir of als productafvoer. Een nieuw soort meerfasenreactor was ontwikkeld om de integratie van reactie en productafvoer of substraattoevoer te vergemakkelijken. Hierbij is dan sprake van enerzijds een in situ gecontroleerde toevoer van substraat naar de fase waar de reactie plaatsvindt. of anderzijds de gecontroleerde afvoer van product vanuit de reactiefase. De reactor kan worden gezien als een 3-fase vloeistofvloeistof-vast gefluïdiseerd bed bioreactor. Daarnaast is een algemene ontwerpstrategie voor een 3-fasen gefluïdiseerd bed bioreactor ontwikkeld. een disperse vloeistoffase en een disperse vaste fase. Van deze twee afzonderlijke 2-fase systemen werden de hydrodynamische karakteristieken onderzocht (Hoofdstuk 2 en 3). en 3) fermentaties die gepaard gaan met een slechte oplosbaarheid van het substraat of het product. Voorbeelden van dergelijke fermentaties zijn: 1) substraat of product geremde reacties. In dit proefschrift worden hydrodynamische aspecten van een dergelijk 3-fase gefluïdiseerd bed onderzocht. Deze meerfasenreactor bestond uit een continue vloeistoffase.

Voor vijf verschillende soorten gel deeltjes (verschillend in diameter en dichtheid) werden drie verwante karakteristieken bestudeerd. De continue waterfase en de disperse fase (dodecaan druppels) werden zowel in mee. met als gevolg dat de weerstandskracht lager is. Twee verschillende hypothesen werden voorgesteld.als in tegenstroom door de kolom geleid. De eerste hypothese veronderstelde dat de vermindering in weerstand veroorzaakt wordt door de aanwezigheid van een kleine hoeveelheid opgeloste polymeer.Samenvatting voorspellen. De tweede hypothese ging uit van de aard van de gel deeltjes. Dit model werd gebruikt voor de voorspelling van de drukval over een gepakt bed en de deeltjes hold-up in een gefluïdiseerd bed. Een nieuw model dat de weerstandskracht van één gel deeltje in een gepakt of gefluïdiseerd bed correct beschrijft. en in dat licht bezien kunnen gel deeltjes worden beschouwd als ‘vaste’ water druppels. Hoewel deze karakteristieken dezelfde trends lieten zijn zoals deze worden waargenomen voor conventionele deeltjes (glas knikkers). Gel deeltjes bestaan voor meer dan 95% uit water. had de watersnelheid slechts een zeer geringe invloed op de dodecaan hold-up. werd gepresenteerd. Voor verschillende spargerontwerpen werd de druppel hold-up in vloeistof-vloeistof sproeikolom bestudeerd (Hoofdstuk 3). Deze karakteristieken hadden te maken met de weerstandskracht tussen het deeltje en zijn omringende omgeving: de eindige valsnelheid van één enkel deeltje. de drukval over een gepakt bed van deze deeltjes. Deze veronderstelling was gedeeltelijk waar voor het voorspellen van de verschillende fasen hold-ups in het 3-fasen systeem (Hoofdstuk 4). Het aantal pijpjes had echter wel een grote invloed op de dodecaan hold-up: bij een gelijke superficiële dodecaan snelheid werd de hoogste 193 . en de deeltjes fractie in een vloeistof-gefluïdiseerd bed. Er werd geconcludeerd dat de weerstandskracht tussen gel deeltje en vloeistof lager was vergelijken met conventionele deeltjes. indien een model gebruikt werd dat deze dispersie coëfficiënt voorspelde in een 2-fase druppelkolom (Hoofdstuk 5). Hoewel de superficiële watersnelheid en dodecaansnelheid van dezelfde orde grootte waren. of was aanwezig in een gepakt bed of gefluïdiseerd bed (Hoofdstuk 2). Het fysische gedrag van gel deeltjes werd bestudeerd: een gel deeltje bezonk als enkel deeltje in stilstaand water. bleek dat gerenommeerde modellen deze karakteristieken niet kwalitatief konden voorspellen. Vandaar dat het oppervlak van het gel deeltje waterachtige eigenschappen kan vertonen. De voorspelling voor de axiale dispersiecoëfficiënt van de continue vloeistoffase was goed.

die afgeleid zijn voor de afzonderlijke 2-fasen systemen. Daartoe werden modellen opgesteld voor de weerstandskrachten voor zowel een gel deeltje als een dodecaan druppel aanwezig in de 3-fasen 194 . Deze deeltjes werden gefluïdiseerd door een continue waterfase (30 mM KCl. Tot een holdup van 0. en één instabiel stromingsregime. Er werd verondersteld dat de integratie van de modellen voor een vloeistof/vast 2-fasen systeem (een gefluïdiseerd bed) met een vloeistof/vloeistof 2-fasen (een druppel sproeikolom). werden gebruikt. De voorspelling met empirische literatuur modellen was ook niet goed. De druppel hold-up kon goed worden beschreven met een model dat gebaseerd was op het snelheidsverschil tussen de continue fase en de dodecaan fase. voor hogere hold-ups nam deze parameter af. Daarnaast werden de verschillende fasen hold-ups bestudeerd. Op basis van visuele waarneming van het gefluïdiseerde bed konden vijf verschillende stromingsregimes worden onderscheiden: twee stabiele homogene stromingsregimes. dat verschilde in oppervlaktespanning (60 mM KCl/ dodecaan). Hierbij gebruiken de meer fundamentele modellen parameters. Helaas was de voorspelling met deze modellen slecht. waarvoor een nieuw model werd gepresenteerd. Als vaste fase (dichtheid 1065 kg m-3) werden κ-carrageen deeltjes gevuld met Celite™ gebruikt.Samenvatting hold-up waargenomen voor die sparger met het laagste aantal pijpjes. twee stabiele heterogene stromingsregimes. Dit model werd succesvol toegepast voor de voorspelling van de hold-up van een ander 2-fase systeem. Verschillende literatuurmodellen voor bestaande 3-fasen systemen werden geprobeerd om de verschillende fase hold-ups te voorspellen.13 werd gevonden dat deze parameter constant was (n=4). Dus een nieuw model voor een gefluïdiseerd bed van gel deeltjes en dodecaan druppels moest worden afgeleid. ‘water’). Dit snelheidsverschil volgde een Richardson en Zaki-achtige vergelijking. indien de parameters. een goede beschrijving zou kunnen opleveren voor het 3-fasen systeem. De verschillende fasen hold-ups van dit 3-fasen systeem en het stromingsgedrag van de gel deeltjes werden bestudeerd (Hoofdstuk 4). afgeleid voor de afzonderlijke 2-fasen systemen. De disperse organische fase (dodecaan druppels) steeg op door het gefluïdiseerd bed van gel deeltjes. De integratie van een vast–vloeistof gefluïdiseerd bed met een druppel sproeikolom heeft als resultaat een vloeistof-vloeistof-vast 3-fase gefluïdiseerd bed. waarbij uitspoeling van de deeltjes uit het bed optrad. Dit model was gebaseerd op stationaire krachtenbalansen. Deze vergelijking maakt gebruik van een parameter n. Voor verschillende water en dodecaansnelheden werden de stromingskarakteristieken van de gel deeltjes beschreven.

en daarom werden de beperkingen van het gebruik van deze merkstof ook bestudeerd. Alle data voor de deeltjes hold-up en druppel hold-up werden goed beschreven voor een druppel hold-up kleiner dan 0. Het bleek dat de axiale dispersiecoëfficiënt van de continue fase toenam met de superficiële dodecaansnelheid. Voor de beschrijving van de gehele data set werd een empirisch model opgesteld. en 3) het debiet van het organische oplosmiddel. waarbij gebruik werd gemaakt van een KCl-merkstof. dat de menging van de continue fase van een 3-fase gefluïdiseerd bed vergelijkbaar was met de menging van de continue fase van zowel een bellenzuil als de hoofdbuis van een ‘liquid-impelled loop reactor’. Er werd gevonden dat een puls met een gering volume betrouwbare resultaten gaf. dat bij gebruik van een 3-fase reactor een hogere conversiegraad wordt bereikt vergelijken met een 2-fase vloeistof-vast gefluïdiseerd bed bioreactor. uit praktisch oogpunt. terwijl de afhankelijkheid van deze dispersiecoëfficiënt met de superficiële watersnelheid een maximum liet zien. KCl kan de deeltjes in diffunderen. met succes toegepast voor de voorspelling van de dispersiecoëfficiënt. zoals 1) de verdelingscoëfficiënt van het product over de waterfase en de organische oplosmiddelfase. Dit ontwerp maakte gebruik van de resultaten van de voorafgaande hoofdstukken. Deze coëfficiënt werd bepaald op basis van metingen aan de verblijftijdspreiding. Een eenvoudig procesontwerp voor een 3-fase vloeistof-vloeistof-vast gefluïdiseerd bed bioreactor werd gemaakt (Hoofdstuk 6). Een model uit de literatuur voor het verband tussen dispersiecoëfficiënt en energiedissipatie werd. Menging wordt gekenmerkt door een axiale dispersie coëfficiënt.06. Er werd aangetoond.Samenvatting suspensie. op kleine uitzonderingen na. 2) de toxiciteit van het gevormde product. Het bleek dat al voor 195 . Een eenvoudige productgeremde reactiekinetiek werd verondersteld. Operationele voorwaarden werden opgesteld. Er werd geconcludeerd. Deze resultaten weerspiegelen de toename van de continue fase dispersiecoëfficiënt met de energiedissipatie per massa-eenheid. Met behulp van het procesontwerp werd de invloed van verschillende modelparameters onderzocht. en weloverwogen schattingen voor stofoverdracht en stromingsgedrag van beide disperse fasen (de gel deeltjes en de dodecaan druppels). waarvoor de prestatie van het 3-fasen systeem beter is dan de prestatie van het 2-fasen systeem. Er werd geconcludeerd dat. Het onderzoek naar de mening van de continue fase van een 3-fase gefluïdiseerd bed is beschreven in hoofdstuk 5. voor de inzet van een 3-fase systeem er slechts geringe aanpassing aan een 2-fase vloeistof-vast gefluïdiseerd bed nodig zijn.

Samenvatting bescheiden waarden van de verdelingscoëfficiënt het 3-fasen gefluïdiseerd bed beter presteerde dan het 2-fasen gefluïdiseerd bed. Dus voor die bioconversies. Daarnaast wordt benadrukt welke stappen ondernomen moeten worden voor de praktische inzet van dit nieuwe type bioreactor. 196 . waarvoor in situ extractie gunstig is. Dit proefschrift wordt afgesloten met een overzicht van de huidige status van deze nieuw-ontwikkelde 3-fasen gefluïdiseerd bed bioreactor. is de 3-fasen gefluïdiseerd bed bioreactor een zeer geschikt apparaat. Er wordt aangeven welk extra onderzoek noodzakelijk is voor een volledig begrip van het hydrodynamische gedrag van deze bioreactor.

Je zult je resultaten terugvinden in hoofdstuk 4. Hans. momenten van overgave en opgave. Hans. was je wel aanwezig als ik om raad en advies vroeg. Echter er zijn altijd wel mensen. Promoveren naast een volledige aanstelling bleek haast onmogelijk. Euforische gevoelens overheersen. bedanken voor het feit dat je vertouwen in mij had en de mogelijkheid hebt gegeven om op jouw vakgroep mijn promotie te kunnen uitvoeren. als toeschouwers langs de kant was jullie aandacht en aansporingen zeer welkom. ECN. Gedurende de eerste 4 jaren. Promoveren is vaak een eenzame zaak. veel heb je gemeten. Hoewel we de deur bij elkaar plat niet liepen. zijn een aantal gedachten toch teruggekomen in verschillende hoofdstukken. Wouter. bedankt voor je inzet en goede adviezen. heb ik veel hulp gehad van vier studenten. geklust aan de opstelling en goede pogingen ondernomen om de hold-up van die deeltjes te voorspellen. heb je daarna de ontwikkelde meetmethode toegepast om het menggedrag te meten. In de laatste fase was je echter weer prominent aanwezig.Nawoord Nog een paar woorden schrijven en dan is het boekje echt af. Hoewel van jouw werk en resultaten niets is terug te vinden in dit proefschrift. (Ex)vakgroepgenoten. Rik. bedankt dat jullie mij de ruimte hebben gegeven om het proefschrift af te kunnen ronden. Allereerst wil ik jou. Francine en Timo. Jouw resultaten zul je terugvinden in hoofdstuk 5. waarvan het signaal nauwelijks boven de ruis uitkwam. Francine. Wouter. Jenke. in het begin was je betrokken bij dit project en vervolgens was je lange tijd uit beeld. na eerst veel gestoeid te hebben met grafiekjes om de juiste instelling te achterhalen voor het doen van goede verblijftijdspreidingsmetingen. jij had de pech om met een meetmethode te moeten werken. Jouw resultaten hebben echter wel bijgedragen aan het tot stand komen van hoofdstuk 5 over verblijftijdspreiding. 197 . als co-promotor en directe begeleider zagen wij elkaar vaker. waarmee je met gedachten kan wisselen of die je simpelweg aansporen. zware hardloopwedstrijd. maar gedachten gaan ook terug naar de onderweg veelvuldig voorkomende zware momenten. Bedankt voor je inzet en het helpen gereed komen van mijn proefschrift. Timo. Arjen. met veel plezier denk ik terug aan onze discussies. Johan en Huibert. Jenke. Terugdenkend aan de afgelopen jaren bekruipt mij een gevoel dat veel lijkt op een gevoel nadat ik gefinisht ben na een lange.

Het boekje is nu af. hoe staat het met het boekje?.Op deze plaats wil ik nog een groep mensen. De afgelopen periode is veel van onze spaarzame tijd opgegaan aan werken achter jouw PC-tje. komt dat feest er nog? Het boekje ligt er nu en dat feestje komt er ook. nu een vriendenclub die elkaar geregeld ziet. en is er weer volop tijd voor trainen en mooie hardloopuitjes. maar tegenwoordig goede vrienden. Als laatste bedank ik jou. Imke. Familie. Toen wij elkaar leerden kennen had dit boekje immers al af moeten zijn. en mijn standaardantwoord was: het gaat goed. De laatste tijd verstomde jullie vragen. Tijdens mijn studie in Delft begonnen als hardloopvrienden. In het begin waren er jullie oprechte vragen. wellicht in de overtuiging dat de voortgangssnelheid tot nul was gereduceerd. In elk geval bedankt voor de oprechte interesse. maar de voortgangssnelheid is niet zo hoog (type duuratleet). vaak hebben jullie je afgevraagd of dat boekje er nog zou komen en misschien nog wel belangrijker. vooral voor je geduld en begrip als ik weer in het weekeinde achter de PC kroop. Het is nu af en dus veel tijd over om samen leuke dingen te gaan doen. van origine hardloopmaatjes. Bedankt voor vooral de mentale steun in de afgelopen jaren. Erik 198 . bedanken voor hun interesse in mijn promotieperikelen. pa. Stefan.

Het onderzoek dat hij hier heeft uitgevoerd. heeft geresulteerd in dit proefschrift. 199 .Curriculum vitae Erik van Zessen werd geboren op 1 december 1969 in Zoetermeer. Vervolgens begon hij in februari 1995 als AIO bij de vakgroep proceskunde aan de Wageningen Universiteit. Vanaf oktober 2000 tot oktober 2003. is hij werkzaam geweest bij het Energie onderzoek centrum Nederland (ECN). Na het behalen van het Gymnasium diploma aan het Stedelijk Gymnasium te Schiedam in 1988 werd in datzelfde jaar begonnen met zijn studie Chemische Technologie aan de Technische Universiteit Delft. In november 1994 studeerde hij af in de richting bioprocestechnologie bij het Kluverlaboratorium.

200 .