Chrmic-ol Engineernrg Srirncr. Vol. 48, No. 9, pp. 1583-1592, 1993. ooo9 2509,93 16600 + 0.

00
Printed in Great Britain. G 1993 Pergamon Press Ltd

PARTICLE SEGREGATION IN FLUIDISED BINARY

MIXTURES

A. C. HOFFMANN, L. P. B. M. JANSSEN and J. PRINS

Faculty of Mathematics and Physical Sciences, University of Groningen, Nijenborgh 16,
9747 AG Groningen, The Netherlands

(Received 14 October 1991; accepted for publicnrion 28 September 1992)

Abstract-The particle segregation in fluidised beds consisting of different types of binary mixtures is
shown to be governed by the same particle transport processes. The segregation behaviour of both
“different-density mixtures” and “equal-density mixtures”, two types of system which until now largely have
been investigated and modelled separately, have been successfully simulated using the same technique, and
assuming the same transport processes to be active in both types of system. Experimental results showing
segregation profiles in both types of mixtures, but using particles very much coarser than those normally
investigated, are presented. The significance of the particle transport parameters is discussed. A new
empirical relation for the bubble wake angle in group B powders has been proposed, and it has been made
likely that the rate of material interchange between the wake of a rising bubble and the surrounding bulk is
indcpendcnt of the minimum fluidisation velocity of the bed particles. A discussion of future research seen
as useful for better prediction of particle mixing and segregation is provided.

INTRODUCTION mixtures of particles has, so far, largely been confined

Bubbling gas fluidised beds containing mixtures of to binary mixtures [see, however, Nienow et a[. (1987)
powders of different physical properties often exhibit who investigated the behaviour of ideal mixtures of
a vertically non-uniform blend of particles. Thus, al- more than two components and Hoffmann and Romp
though fluidised beds are known for their good mix- (1991) who investigated the behaviour of a powder of
ing characteristics, the solid mixing is often incom- a continuous size distribution]. In such mixtures one
plete when particles of different properties are present fraction of particles (the “jetsam” particles) tend, in the
in the bed. Knowledge of the degree and the rate of mean, to sink relative to the others (the “flotsam”
segregation is important for several reasons. In most particles), this is referred to as “segregation”. Segrega-
industrial applications of fluidised beds good mixing tion is caused by a difference in either the densities or
is required for uniform product quality or in order to the sizes of the particles present, the density has the
avoid defluidisation of parts of the bed, resulting, dominant effect. Binary systems are often referred to
perhaps, in fusion of the particles in the defluidised as “equal-density mixtures” if the two fractions differ
region. In some applications, on the other hand, the in size only, and “different-density mixtures” if the
tendency for segregation is utilised, e.g. for the con- density or both properties are different.
tinuous removal of a product. It is an empirical fact that in most binary systems
It has recently been stated (Nienow et al., 1987) a layer of high jetsam concentration builds up in the
that correlations for the prediction of the mixing/ bottom of the bed, and one obtains as steady-state
segregation behaviour of fluidised particles of differ- “segregation pattern” two distinct layers. The jetsam
ent densities are unsuitable for beds containing par- concentrations of these two layers are different, but
titles of different sizes, but the same density. The fairly uniform within each of the layers. The steady-
modelling of mixing and segregation of particles in state profiles are the result of a dynamic equilibrium
fluidised beds of binary mixtures has indeed, by and between the competing processes of mixing and segre-
large, been aimed at only one of the two types of gation (Naimer et al., 1982). For a system operating
systems. In the present article it is shown that the below the minimum fluidisation velocity of the jetsam
same particle transport processes govern the behavi- (this is often the case in equal-density systems) the
our of different-density mixtures and equal-density bottom layer is defluidised. In that case bubble forma-
mixtures, although identical equations may indeed tion will occur at the top of the defluidised layer
not be directly applicable to both. This is done with- instead of by the distributor plate.
out using adjustable parameters, although some con- Gibilaro and Rowe (1974) published extensive work
elusions about the involved parameters have been in which an analytical model predicting segregation
drawn on the basis of the experimental results. patterns in jetsam-poor binary mixtures of different-
density particles was presented. Subsequent work has
been dedicated to estimating the parameters involved
LITERATURE SURVEY in this model and to implementing it (Chiba et al.,
Modelling of the mixing/segregation behaviour of 1979; Tanimoto et al., 1981; Naimer et al., 1982).
bubbling gas fluidised beds containing non-uniform This model is based on the following concepts. Two

1583
1584 A. C. HOFFMANN et al.
mixing mechanisms were envisaged:
(a) a dispersive mixing resulting from the disturb-
ance of the bed material by rising fluidisation
bubbles, and
(b) a “convective” mixing (often referred to as “cir-
culation”) due to the elevation of bed material
in the wake of bubbles.
When a bubble rises through the bed there is ex-
change of material between the wake and the sur-
rounding bulk. The wake material is splashed onto
the surface of the bed as the bubble reaches the top.
The mixing caused by the action of rising fluidisa-
tion bubbles, mentioned under (a) above, arises from
the fact that particles initially positioned in a horizon-
tal plane in the bed will drift axially due to the disturb-
ance of a rising bubble (Tanimoto et al., 1981). The
drift of bulk particles due to the rise of one fluidisation
bubble was measured by Tanimoto et al. whose ex-
perimental points fell in a cloud around the curve
shown in Fig. 1. The drift can be seen not to be
symmetrical with respect to the horizontal plane. The
cumulative effect, however, of many bubbles rising
through the bed will be to disperse the particles in the
bed. This dispersion will give rise to a net transport of
any sort of “marked” particles in their down-gradient
direction. This is what is meant when the word “dis-
persive” is used about this mixing mechanism.
The segregation mechanism envisaged was that of
individual jetsam particles settling relative to flotsam
particles. This has been observed only to occur in
regions being excited by passing bubbles. The cumu-
lative effect of many bubbles rising through the bed is
to cause a net downwards flow of jetsam particles
proportional to the total volume of bubbles (including
their wakes) flowing upwards.
Detailed measurements of the events taking place in
a segregating fluidised bed have been performed
(Chiba et al., 1979; Tanimoto et al., 1977, 1981;
Naimer et al., 1982). Among other the paths of indi-
z” 1.0
Initial position
t\ /
___L___ - corresponded exactly with the cross-sectional area of
0 -+___-
t tance of three bubble radii from the centre of the rising
-1.0 ’
0 1.0 2.0 3.0
N, bubble radii
Fig. 1, The drift of bulk material due to the passage of one
bubble according to Tanimoto et al. (1981). The drift curve of
jetsam material will fall below this curve due to the segrega-
tion.
vidual flotsam and jetsam particles have been
followed in two-dimensional beds as bubbles rise
through the bed, and the segregation mechanism has
been observed. It has become clear from these
measurements that the particle transport mechanisms
proposed by Gibilaro and Rowe are consistent with
observation.
Five models have been proposed aimed at pre-
dicting the behaviour of equal-density binary mix-
tures: Gelperin et al. (1977), Yoshida et al. (1977),
Burgess et nl. (1980), Chen (1981) and Nienow et al.
(1987). The models of Gelperin et al. (1977) and
Nienow et al. (1987) are purely empirical. The numer-
ical model of Burgess et al. (1980) has also been
applied to different-density mixtures. The particle
transport processes envisaged in the three other
models are all different from the processes described
above, and these models all contain adjustable para-
meters used for fitting the predictions to experimental
results.
The present work was undertaken in order to
demonstrate, without the use of adjustable para-
meters, whether the particle transport processes de-
scribed above also can account for results obtained in:
(a) jetsam-rich different-density mixtures and (b)
equal-density mixtures with defluidisation.
IMPLEMENTATION OF THE MODEL CONCEPTS
Gibilaro and Rowe (1974) formulated conservation
equations for jetsam and solved these analytically for
simple systems. For various reasons the choice was
made not to follow this procedure but to implement
the model concepts purely numerically. The advant-
ages of this approach are mentioned later. How the
different transport parameters were estimated is
briefly described below. Here, as in the other works
referred to, radial variations in bed composition or
rates of particle transport due to, for instance, maldis-
tribution of bubbles or wall effects are neglected.
Tanimoto et al. (1981) measured the drift (settling
distance) of jetsam particles relative to the bulk par-
ticles due to the rise of one single fluidisation bubble.
They determined profiles of the drift as a function of
the distance from the centre of the rising bubble in
various mixtures. They defined a “segregation dis-
tance” by integrating the drift profile radially from the
centre of the bubble to infinity and dividing this
integral by the cross-sectional area of the bubble. The
distance thus obtained would be equal to the mean
settling distance if the region wherein drift occurred
1
the bubble (in fact, significant drift occurs up to a dis-
I bubble). The segregation distance was made dimen-
4.0 sionless by normalising with the radius of the bubble.
They found empirically that the dimensionless segre-
gation distance was related to the properties of the
jetsam and flotsam in the following manner:
(1)
 Particle segregationin
where p and d are densities and particle diameters,
respectively, and subscripts j and frefer to the jetsam
and flotsam fractions, respectively. Hoffmann and
Romp (1991) have given reasons for replacing this
relation with
~=0.8(~)(+J~*3-0.8 (2)
which they proved fitted the experimental results of
Tanimoto et al. almost as well as eq. (1). Defining the
properties of the bulk material in mixtures containing
significant amounts of all the constituents present is
no straightforward matter. A discussion of this prob-
lem in a powder of a continuous size distribution is
given in Hoffmann and Romp (1990). For the present
work in binary mixtures, the bulk properties of the
mixtures have been estimated by linear interpolation
according to volume fraction between the properties
of the pure constituents. The bulk properties, thus,
vary axially in the bed.
In order to determine the circulation due to mater-
ial flow in bubble wakes, the values of two parameters
must be known: (a) the bubble diameter at the time of
formation (by the distributor plate or the defluidised
layer) and the rate of growth due to coalescence and
(b) the fractional wake volume as a function of the
bubble diameter. The relation of Geldart (1972) was
used for the bubble diameter in the present work:
d = (U - Uln,Y
ll
J,,0.4g0.2 (&>.-”
2.OS(U - U,J)O.94h
+ (3)
where N is the number of holes per unit area in the gas
distributor plate; this is by Geldart set to 1000 in the
case of a porous plate. U and U,, are the superficial
fluidisation velocity and the minimum fluidisation
velocity, respectively, fv the bubble wake fraction,
h the height in the bed and g is the gravitational
acceleration. The constant 2.05 has the units
(s m-1)o.94. The first term in eq. (3) gives the bubble
diameter at the time of formation, the second ac-
counts for the growth due to coalescence. Equation (3)
is not the only empirical relation for the size of
bubbles in ffuidised beds proposed in the literature.
The bubble size at formation is, due to the strong
variation of wake fraction with size for small bubbles
and the high concentration of jetsam in the wake
upon formation, a critical parameter for the predic-
tion of segregation profiles. Which of the proposed
relations is chosen does therefore influence the predic-
tions significantly (see also the Discussion).
The minimum fluidisation velocity of mixtures of
particles, U,,, m,were, following Cheung et nl. (1974),
calculated from those of the jetsam and flotsam frac-
tions as
u mf,i y;
u mf.m -- unf,f
( >U MT/
where yj is the weight fraction jetsam. This relation
was originally intended to model only equal-density
fluid&d binary mixtures 1585
mixtures, but has been shown also to model the beha-
viour of different-density mixtures quite well (Naimer
et al., 1982). It was, in the course of the present work,
verified that this relation also for these powders gave
realistic predictions (Prins, 1988).
The fractional wake volume can be calculated if the
wake angle, 8, (see Fig. 2), is known. This angle was
estimated by Naimer et al. (1982) who gave an empiri-
cal reIation for the variation of the wake angle with
the bubble diameter, which, however, was founded on
somewhat scant experimental evidence, Better experi-
mental data are available in Hoffmann (1983) (see Fig.
3), who measured the wake angle directly on the basis
of X-ray images generated in three-dimensional beds.
The following relation, which is also shown in Fig. 3,
is a fit to these data:
6r,,,= 160 - 160exp (-
6Od,) (5)
where db is the bubble diameter. The constant 60 has
the units m - ‘. The powders used to generate the data
shown in the figure fell, like the ones used in the
present work, well within group B of the Geldart
(1972) classification. Equation (5) can, therefore, be
used with some confidence for the present work and
for group B particles in general. Indications are, how-
ever, that the wake fraction becomes a function of the
particle properties in very fine powders.
The correlation coefficient (defined as the fraction
of the standard deviation of the data accounted for by
the correlation) between eq. (5) and the experimental
data is 0.504, which is considerably better than that
obtained when using the relation suggested by
Cloud of
eW ,/-----._\
Fig. 2. Schematicof fluidisationbubble.
1
0 0.02 0.04 0.06
Bubble dlau&er [ml
Fig. 3. Experimental values of wake angle for bubbles of
various sizes (Hoffmann, 1983) with eq. (5) (each experi-
mental point represents an average of around 10 individual
measurements).
1586 A. C. HOFFMANN
Naimer et al. (1982), which correlates the data shown
in Fig. 3 worse than simply using the mean.
Equation (5) indicates that the wake fraction of
a fluidisation bubble tends to zero as the bubble
diameter tends to zero. The relation of Naimer et al.
(1982) gives a zero wake fraction for all bubbles of
diameter less than 0.01 m. In fact, bubbles of a dia-
meter smaller than 0.01 m are very rarely seen in
fluidised beds of the Geldart group B powders.
Hoffmann (1983) observed that a minimum stable
bubble size existed below which the bubble would
rapidly decrease in size and be assimilated in the bed if
it did not grow by coalescence with other bubbles.
The results shown in Fig. 3 certainly indicate that the
wake fraction decreases with decreasing bubble dia-
meter for small bubbles and this is also consistent
with the generally held view. There is, however, no
surface tension in Ruidised beds tending to make
small bubbles more spherical, so analogies with
bubbles rising through liquids cannot be drawn, and
the exact form of empirical relations for the wake
angle in the region of very small bubbles is, therefore,
somewhat arbitrary.
Details about the calculations of the wake fraction
of each bubble from 0, are available (Prins, 1988).
The jetsam concentration in the bubble wake upon
formation of the bubble was assumed to be that of the
bulk material just above the distributor or if a de-
fluidised layer was present that of this layer.
The rate of exchange of material between the wake
of an individual rising bubble and the surrounding
bulk was initially calculated following Chiba and
Kobayashi (1977) and Chiba et al. (1979), a brief
summary of the methodology can be found in Naimer
et al. (1982). The main assumption is that the rate of
material entering and exiting the wake of the bubble is
proportional to the rate of material entering the
bubble cloud at the front. The flow into the cloud is
calculated based on the assumption of potential flow
of the bed material relative to the bubble and a well-
known estimate of the radius of the bubble cloud.
The formula derived by Chiba and Kobayashi
(1977) is an expression for the change of the volume
fraction of jetsam in the wake of a rising bubble, C,,
with height in the bed:
GV 3 unls
-=
a2
- 2 (cw - Cbuld where K, = -.
2&,f db
(6)
K, is an exchange coefficient based on the volume of
the wake (i.e. net rate ofjetsam flow from the wake per
unit of concentration difference per volume of wake),
z is the axial coordinate, CbUlr is the jetsam concentra-
tion in the bulk, ub is the rise velocity of the bubble
and E,,,/ is the voidage at minimum fluidisation condi-
tions.
The total flow of gas in the bubble phase, which in
combination with eq. (3) for the volume of the indi-
vidual bubbles gives the bubble frequency at different
heights in the bed, was estimated from the so-called
et al.
two-phase theory. This theory states that all gas in
excess of that required to fluidise the bed incipiently
goes through the bed in the bubble phase. The two-
phase theory is likely to be a good approximation to
the real situation in the case of Geldart group B pow-
ders.
Many of the profiles obtained consisted of two
layers each with an approximately uniform composi-
tion. In these cases it was found that instead of calcu-
lating the local CJ,,,, in each computational slice of the
bed it was sufficient to calculate two values, one for
each segregation layer, and this resulted in significant
reduction of the time required for the simulations.
The rise velocity of the bubbles is needed for evalu-
ation of the exchange of material between the bubble
and the bulk. This was evaluated from the well-known
equation (Nicklin, 1962);
ub = (U - U,,) + 0.711 (gdb) i’2. (7)
THE NUMERICAL IMPLEMENTATION OF THE MODEL
CONCEPTS
It was found that the analytical solution of the
conservation equation proposed by Gibilaro and
Rowe (1974) was too inflexible and difficult to extend
to complicated systems. Moreover, it is not possible to
apply the analytical solution to equal-density systems
with defluidisation. Naimer et al. (1982) also found
that the analytical solution proposed by Gibilaro and
Rowe ( 1974) created a need for oversimplification, and
proposed to apply the analytical solution only over
small segments of the bed where the properties could
be assumed to be constant and matching the solutions
in these segments numerically.
In the present work it was decided to simulate the
transport processes described above purely numer-
ically. Such a simulation is applicable to all systems,
and this method has the great advantage that the
basic parameters can be changed easily and their
significance investigated directly. A simulation also
carries the potential of predicting transient phe-
nomena.
The strategy was to divide the bed into layers,
adjusting the concentration of jetsam in each as the
fluidisation bubbles travelled past them accounting
for the segregation and the dispersive mixing (causing
exchange of material between the layers), the circula-
tion (removing material from the bottom of the bed
lowering the position of the layers and adding the
wake material to the top of the bed), the material
exchange between the wake and the bulk (adjusting
the concentration in each layer and in the wake) and
the increase in the wake fraction with the average size
of the growing bubbles (removing material from
layers, lowering the position of the layers above and
adjusting the wake concentration).
When studying the computer simulations it soon
became clear-and this is consistent with the asser-
tion of Naimer et al. (1982ethat the dispersive mix-
ing of the particles, arising from the disturbance of the
bed material due to the action of the rising bubbles,
 Particle segregation in fluidisedbinary
did not have any measurable influence on the form of
the segregation profiles obtained. This allowed a con-
siderable simplification of the numerical simulations:
the slices into which the bed was partitioned could be
made much thicker than was the case when the details
of the drift profile of the jetsam had to be simulated
faithfully. The limit to the thickness of the slices was
then only dictated by the need to avoid “numerical
dispersion” resulting from the averaging of the con-
centrations in the slices after the passage of each
bubble.
As mentioned below, the value of K, turned out
only to be of significance for the different-density
mixtures. It moreover became clear that the variation
of this parameter around its mean value due to the
growth of the bubbles in the bed had a negligible
influence on the shape of the profiles predicted. It was,
therefore, in most cases possible to calculate K, on
basis of the mean bubble diameter in the upper part of
the bed, which resulted in a considerable reduction in
the time needed for the simulations.
EXPERIMENTAL
The segregation patterns obtained were, as one
might expect, found to be very sensitive to static
electricity effects on the one hand and to surface
moisture on the particles on the other. A glass column
(diameter 0.146 m) was used as fluidisation vessel to
eliminate static as far as possible, and a humidifi-
cation system was used to bring the gas, which was air
from a high-pressure supply, to a relative humidity of
between 50 and 60%. It was found that if the fluidising
gas was kept within this band of relative humidities,
the obtained profiles were almost constant. A sum-
mary of the properties of the powders used for the
experiments is given in Table 1. The powders all fall
well within the Geldart group B class (Geldart, 1972),
i.e. they fluidise easily and start to bubble immediately
above the minimum fluidisation velocity. Each pow-
der fraction was created by sieving in order to obtain
the low spreads in particle diameter. The height of the
settled bed was always 16cm.
All the experiments were carried out after initial
mixing of the powder by fluidising at a very high
velocity. The superficial velocity was then reduced to
the desired value, and the powder left to fluidise for
15 min. It was determined experimentally that
a fluidisation time of 5 min was sufficient for the bed
to reach its steady-state segregation profile. The
fluidising gas was suddenly turned off in order to
Table 1. Properties of the test powders
Mean
diameter
Powder (pm)
Glass ballotini (I) 565
Glass ballotini (II) 365
Glass ballotini (III) 285
Bronze 235
mixtures 1587
“freeze” the bed composition. The bed was then sec-
tioned by the use of a vacuum technique, also used by
other workers, for the removal of one layer of the bed
at a time (in total 10 layers), and each of the layers was
analysed by sieving.
EXPERIMENTAL RESULTS COMPARED WITH
SIMULATIONS
Below the simulations are compared with the ex-
perimental segregation patterns obtained. The differ-
ent types of binary systems are considered separately.
Jetsum-poor different-density mixtures
The main object of the present paper is to show that
particle transport mechanisms proposed by Gibilaro
and Rowe, and applied to jetsam-poor different-den-
sity mixtures in the past, also can account for the
behaviour of jetsam-rich different-density mixtures
and equal-density mixtures. However, preliminary ex-
perimental work was done also on jetsam-poor differ-
ent-density mixtures. The particles used for this work
have minimum fluidisation velocities an order of mag-
nitude higher than those used by other workers in the
past for this sort of measurement Some interesting
observations were therefore made, the most import-
ant are briefly summarised below.
The system used was glass ballotini (I) and bronze.
It can be seen in Table 1, that the U,, of the jetsam is
considerably lower than that of the flotsam. This
makes the behaviour of the system very “volatile” and
especially difficult to simulate accurately. A small
decrease, for instance, in the fluidisation velocity will
cause less bubbling, which will result in less mixing
and, therefore, a lower concentration of jetsam in the
upper part of the bed. This in turn will lead to an
increase in the local U,, in the upper part of the bed,
which again causes less bubbling, etc. In spite of the
system being so difficult the general impression was
that the simulations agreed quite well with the experi-
mental results.
Example results obtained in this sort of system are
shown in Fig. 4. The figure shows the composition of
each individual layer, in which the bed was frac-
tionated, as a function of the dimensionless axial posi-
tion of that layer (traditionally plotted on the y-axis).
A 90% confidence interval for the observed composi-
tion of each layer of the bed was calculated on basis of
duplicate experiments. A convenient formula for es-
timation of the variance of observations which are in
Range of Particle
diameters density V ml
(pm) (kg m -3) (ms-‘)
65 2510 0.285
55 2510 0.135
25 2510 0.080
25 8750 0.190

1.0
# 0.8 1 0 u = 0.29 m/m
A u - 0.33 m/s
1 I
0.B .
\
E 0.4 j \
_\
0.2 -
0 0
0 o.a 0.4 0.8
weight frauon
Fig. 4. Experimentalsegregation profiles, jetsam-poor dif-
f&ent-densitymixtures compared with numerical simula-
tions. Jetsam:bronze, flotsam: glass ballotini (I). Overall
weight fraction of jetsam: 0.50.Ninety per cent confidence
interval: f 0.03 weight fraction of jetsam.
duplicate and do not have the same mean is
g = i=1
2N
where x are the individual observations, the subscripts
1 and 2 refer to the two duplicate observations (weight
fraction jetsam in this case), subscript i denotes the
(duplicated) experiment, in this case the bed layer, and
N is the total number of (duplicated) experiments,
here equal to the number of layers in the bed; sz is the
estimate of the variance of the observations. This
estimate of the variance was translated into a 90%
confidence interval under the assumption of a normal
distribution of the randomly scattered observations.
In no case did this exceed +0.03 weight fraction of
jetsam, and this interval is, therefore, given as a 90%
confidence interval in Fig. 4.
Numerical simulations of jetsam-poor different-
density mixtures were performed both following the
method described by Naimer et al. (1982) and using
the simulation scheme developed in the course of the
present work, The two simulations yielded very sim-
ilar predictions, but neither yielded profiles
remotely consistent with the experimental results (see
the broken curves in Fig. 4) when using the particle
transport parameters calculated from the formulae
given above. This disagreement is mainly due to the
estimate of the wake/bulk exchange parameter, which,
as mentioned, is the main parameter determining the
deviation of the profile of the top segregation layer
from the vertical. This discrepancy can be resolved in
the following way.
Equation (6) shows the exchange rate coefficient to
be proportional to the 7J,,,fof the bed. This is a result
of the theoretical analysis of Chiba and Kobayashi
(1977), but was not verified experimentally in the
course of that work. The powders used in the present
study had values of U,, an order of magnitude
A. C. HOFFMANNet al.
* 1.0
sp 0.8
Onpalmntsl
E
0.8 - slmuiaubM
0.4
0.2
0.8 Lb 0 b.2 0.4 0.8 0.8 1.0
paam wd&t fracnon jetsam
Fig. 5. Experimentalsegregation profile, jetsam-rich differ-
ent-density mixture compared with simulations. Jetsam:
bronze, flotsam: glass ballotini (I). Superficial fluidisation
velocity: 0.525 m s -I. Overall weight fraction of jetsam: 0.75.
Ninety per cent confidence interval: + 0.03 weight fraction
of jetsam.
greater than those used by Chiba et al. (1979) for
experimental verification of eq. (6), and it would ap-
pear from the results that the proportionality of the
exchange rate parameter with U,, does not hold. The
minimum fluidisation velocities of the two powders
used by Chiba et al. were 0.026 and 0.028 m s - ’ and, if
a value of 0.027 m s - ’ is used in eq. (6) instead of the
U,, of our powders, the equation for K v becomes
K, = (0.081)/(2&,f d,) (9)
then the simulated profiles agree very well with the
experimental ones (the solid curves in Fig. 4). The
results, therefore, indicate that the exchange coeffi-
cient is independent of CJ,,,,.That this may, in fact,
also be the case is made likely by physical arguments
in the Discussion.
Jetsam-rich different-density mixtures
Naimer et al. (1982) report that jetsam-rich differ-
ent-density mixtures could not be modelled by their
method. An experimental segregation pattern of
a mixture containing 75% jetsam by weight is shown
in Fig. 5. The profile is clearly well modelled by the
numerical simulation program, using the exchange
rate parameter independent of U,,,, given by eq. (9).
Equal-density mixtures with dejluidisation
When performing experiments, it soon became
clear that the jetsam concentration in the upper
(fluidised) part of the bed always was very close to
uniform. It would, therefore, seem that rhe wake/bulk
interchange in these systems is less than that seen in
different-density systems. It seems plausible on reflec-
tion that the rate of material exchange should depend
on the difference between the densities of the two
fractions as such a difference will cause the bulk den-
sity of the powder in a wake richer in jetsam to be
higher than that in the surrounding material. The
 Particle segregationin fluid&d binary mixtures
0.8 - from their respective means and dividing by the num-
- --Model ofChen
0.6 -
J O.4 -
0.2 - assumption of normal distributions, translated into
Fig. 6. Equal-density mixtures, jetsam concentration in up-
per and lower segregation layers vs fluidisation velocity.
Jetsam: glass ballotini (I), flotsam: glass ballotini (II). Com-
pared with numerical simulations and with the model of
Chen (1981). Ninety per cent confidence intervals indicated.
..”
L The model concepts of Gibilaro and Rowe, which
I
0.8
0.6
0.4 L bubbles forming on top of the defluidised layer consist
0.2
0 L
0
Fig. 7. Equal-density mixtures, jetsam concentration in up
per and lower segregation layers vs fluidisation velocity.
Jetsam: glass ballotini (I), flotsam, glass ballotini (III). Com-
pared with numerical simulations and with the model of
Chen (1981).Ninety per cent confidence intervals indicated.
exact quantification of such a dependency is a possible
subject for further research.
The profiles obtained, thus, consisted of two layers
each with a uniform jetsam concentration. The com-
position of the two layers were clearly independent of
the initial concentration of jetsam (unless insufficient
jetsam was present for the formation of the lower
segregation layer). The results for all initial composi-
tions are, therefore, plotted together in the following
form: on the horizontal axis the jetsam concentrations
of each of the two segregation layers is shown and on
the vertical axis the fluidisation velocity (Figs 6 and 7).
The predictions of the model of Chen (1981) are also
indicated in these figures.
Each of the pairs of points representing
umetric concentration of jetsam in the upper and
lower segregation layers are averages of a few (mostly
three or four) repeat experiments. The 90% confi-
dence intervals indicated in the figures were calculated
in the following manner. A global estimate of the
1589
variance of the individual observations was obtained
by summing the squared errors of the observations
ber of degrees of freedom (the number of observations
less the number of means determined). Estimates of
the variances of the means were then obtained by
dividing by the number of repeat observations at each
experimentaI condition. These were then, under the
90% confidence intervals.
It can be seen in the figures that there is good
agreement between the observations and the model
predictions especially in view of the approximations
made (the adoption of the two-phase theory and the
relations for the bubble diameter and the wake frac-
tion). Some discrepancy is evident in Fig. 6, where the
difference in compositions between the upper and
lower segregation layers is slightly higher than the
model predicts. The agreement in Fig. 7 is very good,
the model predictions lie within the 90% confidence
intervals of the experimental points throughout the
range of fluidisation velocities.
in the past have only been applied to different-density
mixtures, have, thus, successfully been applied to the
behaviour of equal-density mixtures. We may con-
clude that, if it is assumed that the wakes of the
of material from this lower layer, the distribution of
jetsam between the two segregation layers can be
predicted by applying these model concepts, and
equal-density mixtures have, thus, been shown to be
governed by the same particle transport mechanisms
as different density mixtures.
DISCUSSION
It has been possible to demonstrate that the mixing
and segregation behaviour of particles in all classes of
binary mixtures can be modelled based on the same
concepts. The numerical values for the involved para-
meters have all been determined separately, and in
this sense adjustable parameters have not been re-
quired. The results of the experimental work, how-
ever, has given rise to some conclusions about the
magnitude of these parameters: a new empirical rela-
tion for the bubble wake fraction has been given, and
the coefficient describing the material exchange be-
tween the bubble wake and the bulk, was found only
to fit the results if it was assumed to be independent of
the value of U,, for the bed particles. That this may
well be the case, however, will be made plausible by
comparing the flow of solids around fluidisation
bubbles of like volume in fluidised beds consisting of
large or small particles, respectively.
Consider the bubble with its associated wake
the vol- shown in Fig. 8. This model of a “sphere completing
wake” is consistent with the photographic images of
rising fluidisation bubbles (Reuter, 1966; Stewart,
1968). The velocity of the rising bubble is given by eq.
(7), and, since we are comparing bubbles of like vol-
ume, we are, according to eq. (3), comparing beds
 1590 A. C. HOFFMANN et al
Fig. 8. Particles travelling along the surface of the bubble.
fluidised at like values of (U - V,,). Thus, the bubble
‘rise velocity is independent of the size of the bed
particles. The size of the “cloud” of recirculating gas
surrounding the bubble, however, increases with in-
creasing size of the bed particles (Davidson, 1961).
The particles most likely to be taken up in the wake
travel along the surface of the bubble. Consider a vol-
ume of particles travelling towards the bubble wake as
indicated in the figure, the velocity with which they
travel is independent of the size of the bed particles
since the bubble velocity is. The mass of a given
volume of particles is also independent of the particle
size if the voidage and the particle density is the same.
The reaction of the travelling particles to a given force
is, therefore, independent of particle size. If, therefore,
the force acting on the particle assembly arising from
the gas percolating inwardly towards the bubble void
is independent of the size of the bed particles as well
we may conclude that so is, probably, the rate of
assimilation of particles in the bubble wake.
All models for the flow of the fluidising gas around
fluidisation bubbles (this flow is assumed to be steady
in a coordinate system moving with the bubble) are
based on the adoption of Darcy’s law, which in its
three-dimensional form is
u, = k grad (p) (10)
where II, is a vector describing the fluid velocity rela-
tive to the solids, k is a proportionality constant, and
p is the pressure. For an incompressible fluid this
leads to
div [grad (p)] = 0. (11)
The boundary conditions for this differential equation
are independent of the size of the bed particles (con-
stant pressure at the surface of the bubble and the
gradient of the pressure equal to the buoyant weight
of unit volume of the particles far from the bubble).
Thus, the gradient of the pressure and, therefore, the
force per unit volume acting on the particles every-
where, also the particles indicated in Fig. 8, is inde-
pendent of the size of the particles. We may thus, as
pointed out above, expect the rate of assimilation of
particles in the bubble wake to be independent of the
size of the bed particles and, therefore, their U,,.
Equation (11) and the associated boundary condi-
tions show that even though the fluid flow pattern is
different for different cloud volumes (and different bed
particles), the pressure field and the force acting on the
particles per unit volume is the same. Therefore,
whether particles enter the bubble cloud at the front
of the bubble or not does not affect the likelihood of
their entering the bubble wake.
Equation (9) indicates that the wake/bulk material
transfer coefficient is inversely proportional to
E,,,~ and to de. The former varies very little in fluidised
beds, and this dependency is, therefore, difficult to
verify experimentally. For geometrically similar
bubbles it is easy to show that inverse proportionality
of K w with db simply means that the total volumetric
flow rate of material in and out of the bubble wake is
proportional to the size of the wake/bulk interface,
which on reflection is very reasonable. The bubble rise
velocity is a weak function of db [see eq. (7)] and
one might expect that this would give rise to another
dependence of K, on db. It is not intuitively obvious,
however, when studying the photographs of Reuter
(1966) whether the rate of interchange should be dir-
ectly or inversely related to the rise velocity of the
bubble.
It is likely that other mechanisms of mixing in
fluidised beds have to be taken into account in other
systems. While the mixing processes considered here
should account for variations with type of gas dis-
tributor plate through the variation of the size and
wake fraction of bubbles at formation, there is little
doubt that gross solid circulation is likely to play an
increasing role the larger the bed, especially at height
to diameter ratios close to unity.
Although it has been possible to obtain good agree-
ment between simulations and experiment using only
published or separately fitted relations for the para-
meters, a considerable amount of empirical and theor-
etical research work is necessary before the mixing
and segregation processes considered here can be
quantified with confidence. In the first place it was
mentioned above that a number of relations for the
bubble size and growth rate exist in the literature. The
predictions of these models are too different to allow
exact predictions of segregation patterns. In the sec-
ond place, although the relation for the bubble wake
fraction as a function of size proposed in the present
article is somewhat better founded in experiment
than earlier proposed relations, a thorough empirical
investigation of this aspect is called for. Finally, also
the rate of material exchange between the wake of
a rising bubble and the surrounding bulk should be
determined directly in a wider range of powders to
confirm the independence of U,, of the bed particles
suggested by the present results.
Not only empirical but also theoretical work is
needed. Up to the present time, the research into the
mixing and segregation of particles in fluidised beds
has been limited to empirical determination of the
magnitudes of the parameters involved. Theoretical
research into the following areas would help in the
 Particle segregation in fluidisedbinary mixtures 1591

prediction of particle mixing and segregation in NOTATION

fluidised beds. c volume fraction, dimensionless
How can the segregation behaviour of particles diameter, m
different from the bulk particles be explained in terms ;i wake fraction, dimensionless
of the forces acting on them: the hydrodynamic forces 9 gravitational acceleration, m s - z
and the forces (collisional, sliding or adhesive) arising h height (in bed), m
from point contacts with other particles? k constant of proportionality in Darcy’s law,
A better understanding of the nature of the wake m3 skg-’
carried by fluidisation bubbles could lead to more K coefficient for the material exchange wake/
confident predictions of wake fractions, A promising bulk, s - 1
development is recent work wherein the so-called N number of holes per unit area in distributor
two-fluid equations of fluidisation are being solved plate, m -’
numerically (Kuipers, 1990). This work has already P pressure, N m - ’
been successful in that simulated rising bubbles u superficial gas velocity, m s - l
exhibit a wake-like indentation. The nature of this u, vector denoting the gas velocity relative to
“wake”, however, is not entirely clear, whether it the solids phase, m s ’
contains indentifiable wake material following the 0 velocity, m s - r
motion of the bubble, or if bed material flows freely in x individual observation, dimensionless
and out of this identation. Work in the segregation V weight fraction, dimensionless
I_
behaviour of binary mixtures, the present work in- YS dimensionless segregation distance, dimen-
cluded, indicates that the wake material is carried sionless
with the bubble with only very little exchange with the Z vertical coordinate, m
surrounding bulk and this is confirmed by the photos
taken by Reuter (1966). This would have to be re- Greek letters
flected in the numerical simulations for a satisfactory E voidage fraction, dimensionless
description of the wake. 8 angle, dimensionless
A theoretical description of the mechanism of P density, kg m - ’
formation of fluidisation bubbles would aid in the
prediction of bubble sizes at formation. The theory of Subscripts
formation of bubbles in liquids is well advanced. In b bubble
fluidised beds, however, where there is no surface bulk bulk
tension, the problem of detachment of the bubble f flotsam
from the orifice at which it is formed is not so straight- j jetsam
forward. The numerical solutions of the two-fluid m mixture
equations of fluidisation seem to represent progress mf at minimum fluidisation conditions
also in this respect. P particle
w wake

CONCLUSIONS
It has been shown to be possible, without the use
of adjustable parameters, to model the mixing/
segregation behaviour of all classes of binary systems
using model concepts along the lines first proposed by
Gibilaro and Rowe (1974) and later experimentally
verified.
New light has been thrown upon the significance
and the magnitude of the different parameters in-
volved in the model.
The rate of exchange of material between wake and
bulk proposed by Chiba and Kobayashi (1977) has
been found only to be consistent with the present
results if the dependence on the U,, of the bed par-
ticles is dropped. The relation for the interchange
coefficient thus obtained is consistent with both the
experimental results of Chiba et al. (1979) and those
presented here.
The assumption that the wakes of bubbles formed
on top of a defluidised layer contain material from
this layer yields predicted profiles consistent with ex-
perimental results.
REFERENCES
Burgess, J. M., Fane, A. G. and Fell, C. J. D.. 1980, Measure-
ment and prediction of the mixing and segregation of
solids in gas fluidised beds in Fluidization (Edited by
J. R. Grace and J. M. Matsen), pp. 1405-1412, New York.
Plenum Press.
Chen, J. L.-P., 1981, A theoretical model for particle segrega-
tion in a fluid&d bed due to size difference. Chem. Engng
Commun. 9, 303-396.
Cheung L. Y.-L., Nienow, A. W. and Rowe, P. N., 1974,
Minimum fluidisation velocitv in a binarv mixture of
different sized solids. Chem. E&tag Sci. 29, i3Oll1303.
Chiba, S. and Kobayashi, H., 1977, Solid exchange between
the bubble wake and the emulsion phase in a gas fluid&d
bed. J. them. Engng Japan 10, 206-210.
Chiba, S., Tanimoto, H. and Kobayashi, H., 1979, Measure-
ment of solid exchange between the bubble wake and the
emulsion phase in a three-dimensional gas-fluidised bed.
J. them. Engng Japan 12, 43-45.
Geldart, D., 1972, The effect of particle size and size distribu-
tion on the behaviour of gas-fluidised beds. Powder Tech-
nol. 6, 201-215.
Gelperin, N. I., Zakharenko, V. V. and Ainshtein, V. G.,
1977, Segregation of solid particles in a fluid&d bed and
the equilibrlum distribution_ Fheoreticheskie Osnovy
Khimicheskoi Technologii 11(4), 572-578 (English transla-
tion, Plenum Publishing Corporation 1978).

CES48:9-o
1592 A. C. HOFFMANNet al.
Gibilaro, L. G. and Rowe, P. N., 1974, A model for a seg-
regating gas fluidised bed. Chum. Engng Sci. 29,
l&3-1412:
Hoffmann. A. C., 1983, Ph.D. thesis, University College
London. Department of Chemical and Biochemical En-
gineering.
Hoffmann, A. C. and Romp, E., 1991, Segregation in
a fluid&d powder of a continuous size distribution. Pow-
der Technol. 66, 119-126.
Kuipers, J. A. M., 1990, A two-fluid micro balance model of
fluidized beds. Ph.D. thesis, Technical University of
Twente, The Netherlands.
Naimer, N. S., Chiba, T. and Nienow, A. W., 1982, Para-
meter estimation for a solids mixing/segregation model for
gas fluidised beds. Chem. Engng Sci. 37, 1047-1057.
Nicklin, D. J., 1962, Two-phase bubble flow. Chem. Engng
Sci. 17, 693-702.
Nienow, A. W., Naimer, N. S. and Chiba, T., 1987, Studies of
segregation/mixing in fluidised beds of different size par-
ticles. Chem. Engng Commun. 62, 53-66.
Prins, J., 1988, Unpublished Internal Report, University of
Groningen, The Netherlands.
Reuter, Hz 1966, On the nature of bubbles in gas and liquid
fluidized beds. Chem. Engng Prog. Symp. Ser. 62, 92-99.
Stewart, P. S. B., 1968, Isolated bubbles in fluid&d bcds-
theory and experiment. Trans. lnstn Chem. Engrs 46,
T60-T66.
Tanimoto, H., Chiba, S., Chiba, T. and Kobayashi, H., 1977,
Mechanism of solid segregation in gas fluidised beds, in
Proceedings PACHEC, Vol. II, pp. 381-388. Denver,
Colorado.
Tanimoto, H., Chiba, S., Chiba, T. and Kobayashi, H., 1981,
Jetsam descent induced by a single bubble passage in
three-dimensional gas-fluidised beds. J. them. Engng
Japan 14, 273-276.
Yoshida, K., Kameyama, H. and Shimizu, F., 1977, Mechan-
ism of particle mixing and segregation in gas fluidized
beds, in Proceedings PACHEC, Vol. II, pp. 389-396.
Denver. Colorado.