Chemical Engineering Science 58 (2003) 401415

www.elsevier.com/locate/ces
Axial dispersion in the three-dimensional mixing of particles
in a rotating drum reactor
Richard G. Sherritt
a
, Jamal Chaouki
b
, Anil K. Mehrotra
c
, Leo A. Behie
c,
a
UMATAC Industrial Processes, 2540 Kensington Rd. NW, Calgary, Alta., Canada T2N 3S3
b
Department of Chemical Engineering,

Ecole Polytechnique de Montr eal, Montreal, Que., Canada H3C 3A7
c
Department of Chemical and Petroleum Engineering, 2500 University Drive NW, University of Calgary, Calgary, Alta., Canada T2N 1N4
Received 13 March 2002; received in revised form 3 October 2002; accepted 9 October 2002
Abstract
Horizontal drum reactors are widely used in industry for the processing of granular material. They are ideally suited for chemical
processes that require high temperatures at near-atmospheric pressure. However, the complexities of these reactors have resulted in
empirical design procedures that lead to very conservative and costly reactors. This study rst reviews critically the extensive literature
on experimental results obtained on rotary kilns (without ights) and proposes new design equations for the axial-dispersion coecient in
terms of rotational speed, degree of ll, drum diameter, and particle diameter. A total of 179 data points from the literature, encompassing
both the batch and the continuous operational modes, yielded design correlations for slumping, rolling/cascading and cataracting bed
behaviours. Additionally, new measurements were made on a pilot-scale rotary drum by tracking a single radioactive particle (emitting
gamma-rays) using a battery of nine scintillation counters; these data conrmed the correctness of the proposed design correlations.
? 2002 Elsevier Science Ltd. All rights reserved.
Keywords: Granular; Mixing; Dispersion; Diusion; Particle; Scale-up; Rotary kiln
1. Introduction
Horizontal rotating drums, or rotary kilns, are commonly
used for processing granular materials in the mineral, ce-
ramic, cement, metallurgical, chemical, pharmaceutical,
food, and waste industries. They are suited to processes that
require high temperatures at near-atmospheric pressure. As
chemical reactors, they are usually designed using empirical
procedures. There is signicant economic incentive to de-
velop a more fundamental understanding of a rotating drum
for processing granular solids. For example, the AOSTRA
Taciuk Processor (ATP) is a new rotary kiln, developed by
UMATAC Industrial Processes (Calgary, Canada), used to
pyrolyse oil shale to produce oil vapours (Turner, Wright,
& Taciuk, 1989). The ATP consists of four zones for pre-
heating, pyrolysing, combusting and cooling the shale. It
was successfully piloted at a feed rate of 80 tonnes of oil
shale per day. Heat transfer coecients and residence times
measured in the pilot kiln were used to scale-up to the

Corresponding author. Tel.: +1-403-220-6692;

fax: +1-403-284-4852.
E-mail address: behie@ucalgary.ca (L. A. Behie).
6000 t/d demonstration plant that is operating presently in
Queensland, Australia. A kiln for a 25,000 t/d commercial
plant is being designed. Because the eects of scale-up on
the complex hydrodynamic behaviour of the granular solids
cannot be accurately predicted, generous safety factors are
used to size the larger kilns. The estimated additional cost
to build these safety factors into a single commercial ATP
is in excess of $10 million.
1.1. Types of bed behaviour
Horizontal rotating drums used in industry vary greatly in
size, type of internals, rotational speed and ll level. Most
rotary kilns are inclined slightly, with the granular mate-
rial entering at the upper end and owing by gravity to the
lower end. However, some kilns, like the ATP, are not in-
clined. Some rotary drums are simple cylinders without any
internals, while others may contain ights on the inside sur-
face of the wall. The ights lift and drop the particles into
the gas stream. Lifting ights are common in rotary dryers
where increasing the contact between the particles and the
gas stream is benecial. In the ATP system, combustion and
0009-2509/02/$ - see front matter ? 2002 Elsevier Science Ltd. All rights reserved.
PII: S0009- 2509( 02) 00551- 1
402 R. G. Sherritt et al. / Chemical Engineering Science 58 (2003) 401415
Fig. 1. Possible modes of transverse bed behaviour in a partially lled, horizontal, rotating drum.
Fig. 2. Phase diagram for possible modes of transverse bed behaviour in a
partially lled, horizontal, rotating drum; based on criteria and equations
by Mellmann (2001).
cooling zones contain lifting ights, while the preheat and
pyrolysis reaction zones do not.
This paper focuses on drums without lifting ights, in
which the granular material forms a bed at the bottom of the
drum. The motion of the bed of particles in the transverse
plane (perpendicular to the drum axis) as it moves from one
end of the drum to the other can be characterised as one of
several possible behaviours. This behaviour depends on the
operating conditions (rotational speed and degree of ll) and
the friction of the particles with each other and with the sur-
face of the drum wall (Henein, Brimacombe, & Watkinson,
1983a,b). Fig. 1 contains illustrations of the possible types
of bed behaviour while Fig. 2 shows a typical shape of
the phase diagram for possible types of bed behaviour.
Mellmann (2001) provided criteria and equations to predict
the transitions between the dierent types of behaviour.
Slipping is an undesirable behaviour in which the bed
of particles slips on the surface of the drum wall. Slipping
usually occurs at low volumetric ll and can be prevented by
roughening the surface of the wall. There are several types
of slipping; but because sliding results in little or no mixing,
there is little interest in slipping behaviour.
The other extreme behaviour, centrifuging, is observed at
high rotational speeds when the centrifugal force at the wall
exceeds the gravitational force. The particles remain xed
to the wall. The speed at which centrifuging begins to occur
is referred to as the critical speed and can be determined
using the following equation.
n
c
=
60
2
_
q
R
. (1)
Centrifuging behaviour is also of little interest as rotary
kilns operate well below critical speed. However, the critical
speed is used as a reference for other types of behaviour.
Slumping, rolling, cascading, and cataracting are
non-slipping bed behaviours which occur between zero
and critical speed. The particles in most rotary kiln reac-
tors have slumping, rolling or cascading behaviour, while
cataracting is common for grinding particles in ball mills.
Slumping occurs at low rotational speeds (less than 3%
of the critical speed). The particles that form a bed at the
bottom of the drum will ride up one side of the wall due to
the rotation. The particles on the sloped surface will peri-
odically avalanche from the upper half of the surface to the
lower half.
Both rolling and cascading behaviours occur between 3%
and 30% of the critical speed. Rolling behaviour, which
occurs at lower bed depth and fraction of the critical speed
than cascading behaviour, is characterised by a at upper
surface as illustrated in Fig. 3a. The angle of the upper
surface to the horizontal plane is known as the dynamic
angle of repose of the granular material.
R. G. Sherritt et al. / Chemical Engineering Science 58 (2003) 401415 403
Fig. 3. Transverse bed motion in a rotating drum with (a) rolling behaviour and (b) cascading behaviour.
The rolling bed becomes a cascading bed at higher ro-
tational speed and ll. The upper surface is deformed, as
shown in Fig. 3b, due to the increased centrifugal forces and
increased particle acceleration and deceleration.
At still higher speeds, 30100%of critical speed, particles
reaching the upper surface of the bed are cataracting or
airborne.
1.2. Particle mixing in a rotating drum
A statistical variable can be used to describe the extent of
mixing in a non-owing bed of particles. Carley-Macauly
& Donald (1962) use an index-of-mixing obtained from the
variance (square of the standard deviation) of a number of
small samples to which a correction is applied for the random
variance owing to the sample size. Mixing is characterised as
either convective or diusive. Mixing in the axial direction
in a horizontal rotating drum is slow and is characterised
as purely diusive mixing since the direction of mixing is
perpendicular to the plane of particle circulation. Mixing in
the transverse plane is much more rapid and is a combination
of convective and diusive mixing. For both types of mixing,
however, the variance decays exponentially with time.
Although the index-of-mixing is useful in studying the
eects of variables on the mixing rate, it is not easily adapt-
able to the prediction of heat transfer and reaction conver-
sions in reactors.
2. Mixing in the axial direction
The methods used by previous authors to model mixing of
particles in the axial direction in horizontal rotating drums
can be divided into the following groups:
(1) nite stage transport,
(2) Monte Carlo simulation,
(3) discrete element modelling, and
(4) dispersion model.
For each method the mixing model can be extended to
include heat transfer and chemical reactions.
2.1. Finite stage transport
A common method used to represent a continuous reactor
with some internal mixing is to connect a number of per-
fectly mixed reactors in series. The number of mixers is set
to give the same response measured at the outlet of the re-
actor when a change in concentration is made at the inlet.
If axial mixing is small, as in rotary kilns, then the number
of mixers in series can be approximated from the mean and
standard deviation of the residence time by the following
equation:
N =
t
2
o
2
t
. (2)
From experiments on a small drum, the axial length
of one mixing unit is about 10% of the drum diameter
(Wes, Drinkenburg, & Stemerding, 1976). Similarly, Sai
et al. (1990) found the axial length to be from 2% to
10% of the drum diameter. Therefore, 100 or more per-
fect mixers in series could approximate a typical kiln with
a length-to-diameter ratio of 10. One limitation of the
mixers-in-series model is that it cannot predict the axial
mixing in a non-owing or batch reactor.
Variations of the mixers-in-series model are proposed by
Mu and Perlmutter (1980), Kelbert and Royere (1991), and
McTait, Scott, and Davidson (1998) using dierent arrange-
ments of plug ow sections and perfectly mixed sections for
each stage.
Generally, nite stage transport models can t the axial
mixing of particles in a rotating drum. However, the model
requires at least one parameter in addition to the number
of stages to t the case of a non-owing drum. Also, little
has been done to determine the eect of speed, ll, drum
diameter and particle properties on the number of stages or
other tting parameters.
404 R. G. Sherritt et al. / Chemical Engineering Science 58 (2003) 401415
2.2. Monte Carlo simulation
A number of authors used Monte Carlo simulations to
model mixing in the axial direction in a rotating drum. A
Monte Carlo simulation is a mathematical experiment by
which the expected outcome of a stochastic process is es-
timated by random sampling from the probability distribu-
tions that govern the events making up the process. The
various attempts to predict the axial mixing in rotary drums
using a Monte Carlo simulation dier in the events making
up the motion of a particle and the probability distribution
assigned to the magnitude or direction of each event. These
attempts also dier greatly in their success to predict actual
mixing behaviour.
Cahn and Fuerstenau (1967) considered a bed of particles
that circulates in a plane perpendicular to the drum axis.
The bed has a rotational speed that is greater than the drum
speed. The drum is divided into sections of equal width
perpendicular to the axis. A Monte Carlo simulation was
used to model the axial movement of a particle between the
sections. The three probability distributions are:
(1) the number of particles that leave a section per bed
revolution,
(2) the direction the particles move, and
(3) the axial distance the particles move.
Observations for an individual particle in a non-owing
or batch drum give the average bed rotational speed and
the three probability distributions for various speeds and
ll combinations used in the simulations. Axial mixing
observed in non-owing experiments is reported to agree
with mixing predicted in the simulations. Although Cahn
and Fuerstenau (1967) did not publish the measured bed
speeds or probability distributions, some discussion was
given about the eect of speed and ll on the probability
distributions. The particle is more likely to leave a section
and move a greater distance as speed and ll are increased.
Rogers and Gardner (1979) discerned the following be-
haviour for a particle in a rolling bed: The particle moves
on a xed radius until it reaches the bed surface. On the
bed surface, the particle is free to roll or tumble until it
re-enters the bed at a new radius. While on the surface, the
particle collides with other particles. Due to the collisions,
the angle at which the particle descends deviates from the
angle of maximum descent. The radial position determines
the time for the particle to complete a cycle and the distance
the particle rolls on the bed surface. The angle of descent
determines the axial movement of the particle.
The two probability distributions used in the Monte Carlo
simulation are:
(1) the radius at which the particle re-enters the bed is a
uniform distribution weighted for the number of parti-
cles at each radius, and
(2) the angle of descent has a normal distribution about
the angle of maximum descent. The standard deviation
for the angle is determined at various ll levels from
non-owing experiments by Shoji, Hogg, and Austin
(1973).
Because the simulation of experiments with continuous
ow by Shoji, Hogg, and Austin (1973) showed good agree-
ment, Rogers and Gardner (1979) concluded that axial mix-
ing in owing and non-owing rotating drums with a rolling
bed is primarily due to the same mechanism (i.e. collisions
between particles in the surface region of the bed). They also
concluded that the deviations in the angle of descent, and
not the randomness in the radial position, control the disper-
sion mechanism. Black (1988) proposed a physical model
similar to that by Rogers and Gardner (1979) except there is
no deviation from the angle of steepest descent as the parti-
cle rolls down the bed surface. The probability distribution
for movement in the radial direction is found by tracking
a single particle in non-owing experiments. Monte Carlo
simulations predict axial mixing for hypothetical, owing
drums.
Kohav, Richardson, and Luss (1995) gave six variations
of a model that, like Blacks model, have no deviation in
the axial direction from the path of steepest descent as par-
ticles roll down the bed surface. The models vary in the
way the randomness of the radial position is treated. The
simulations predict much less axial mixing than measured
by experimenters. The authors conclude that radial segre-
gation must be the reason for the larger axial mixing ob-
served in experiments. However, Black (1988) and Kohav,
Richardson, and Luss (1995) overlooked the contribution of
particle collisions in the surface region. It is clear from their
models that the predicted mixing is not sucient when one
considers the limiting case of no net ow as both would pre-
dict no mixing in the axial direction in a non-owing drum.
A Monte Carlo simulation can give good predictions of
axial mixing if the signicant events are included. One dis-
advantage of a Monte Carlo simulation is the large amount
of computing time required. Also, no attempts have been
made to relate the probability distribution of the events with
parameters such as rotational speed, volumetric ll, drum
diameter and particle properties.
2.3. Discrete element modelling
Discrete element modelling (DEM) is a simulation tech-
nique used to study the ow of granular material. Like
a Monte Carlo simulation, particles are considered indi-
vidually rather than as a continuum. However, rather than
tracking the movement of one particle, DEM tracks all the
particles simultaneously. The eect of the interstitial uid is
neglected. Until recently, most studies of granular ow us-
ing DEM have been restricted to a few hundred particles and
two dimensions due to excessive computational demands.
While the number of particles has currently been extended
to several thousand, an actual granular ow may consist of
millions or trillions of particles. As computers continue to
R. G. Sherritt et al. / Chemical Engineering Science 58 (2003) 401415 405
Table 1
Sources of experimental data from continuous-feed rotating drums from which particle dispersion coecients in the axial direction are determined
Source Drum Granular Particle
diameter (m) material diameter (mm)
Karra and Fuerstenau (1977) 0.08 Dolomite 0.357
Abouzeid, Mika, Sastry, and Fuerstenau (1974) 0.08 Dolomite 0.359
Rogers and Gardner (1979) 0.095 Dry powder 0.335
McTait, Scott, and Davidson (1998) 0.105 Ballitini 0.2
Mu and Perlmutter (1980) 0.102 Rice 2 5
Hatzilyberis and Androutsopoulos (1999) 0.12 Lignite 9.6
Ang, Tade, and Sze (1998) 0.14 Zircon+coal 1.7, 2.7
Sai et al. (1990) 0.147 Ilmenite 0.2
Rutgers (1965) 0.16 Rice, oats 2
Tscheng (1978) 0.1905 Polystyrene 1.9 3.1 3.6
Hehl, Kroger, Helmrich, and Schugerl (1978) 0.250 Soda 0.137
Sugimoto (1968) 0.255 Sand 1.5
Lebas, Hanrot, Ablitzer, and Houzelot (1995) 0.6 Coal 0.520
Wes et al. (1976) 0.6 Potato starch 0.0150.10
Rutgers (1965) 0.765 Oat groats 2
Ray et al. (1994) 0.90 Coal+ore 615
become more powerful and more ecient algorithms are
developed, the number of particles in a simulation will in-
crease and the computing time will decrease in the future.
Even with its limitations, DEM has been a useful tool for
studying granular ow in rotating drums. DEM was used
to simulate particle behaviour in a two-dimensional rotat-
ing drum by Buchholtz, P ocshel, and Tillemans (1995).
The model includes 500 square, in-elastic particles and is
capable of reproducing the transition from a slumping bed
at low rotational speeds to a rolling bed at higher rotational
speed. Baumann, J anosi, and Wolf (1995) used this tech-
nique to simulate a two-dimensional drum with about 3000
circular particles. The model was used to study the mecha-
nism of segregation at low rotational speed (slumping bed
behaviour). No comparison to a real system was given.
Kohring (1995) simulated a three-dimensional drum with
1000 spherical, in-elastic particles. A rolling behaviour
was observed but again no comparison to a real system
was given. McCarthy and Ottino (1998) were able to sim-
ulate a drum with over 10,000 particles by using DEM
for the active region only. Results were compared to real
experiments with good qualitative agreement. Yamane,
Nakagawa, Altobelli, Tanaka, and Tsuji (1998) tted sim-
ulations of a three-dimensional drum with 12,000 particles
to real experiments by adjusting a parameter that accounts
for the non-sphericity of the particles. Good agreement was
obtained for the angle of repose, velocity proles and the
thickness of the active layer.
DEM not only simulates granular ow, but also simulates
the mixing of particles in the granular ow. While DEM
has been used to study mixing in the transverse plane of
a rotating drum, the only attempt to use DEM to predict
self-diusion of particles in the axial direction in a rotating
drum is by Kohring (1995). The results for a non-owing
drumare consistent with diusive behaviour, but no compar-
ison is made to real experiments. The diusion rate is about
an order of magnitude larger than that measured by others in
real experiments and increases more with rotational speed.
The dierences may be due to the material properties spec-
ied for the simulation not being typical of most materials.
Using DEM, it would be possible to predict the eect of
rotational speed, ll, drum diameter and particle properties.
However, the number of particles would need to be limited
to several thousand.
2.4. Dispersion model
The most common method of modelling axial mixing in
both owing and non-owing, horizontal, rotating drum is
the axial-dispersion model. The bed of particles is treated
as a continuum. Mixing in the axial direction is represented
by the following one-dimensional diusion equation:
cc
ct
= D
:
c
2
c
c:
2
u
:
cc
c:
, (3)
where D
:
is the axial-dispersion coecient. The advantage
of the axial-dispersion model is that a single parameter is
used to quantify the amount of mixing in the axial direc-
tion. The number of particles in the bed is not limited. The
axial-dispersion coecient has been determined from both
continuous ow and batch experiments.
2.4.1. Continuous ow experiments
The axial-dispersion coecient D
:
from a owing drum
is almost always determined from the residence time distri-
bution (RTD). The RTD is measured at the outlet end for a
pulse of tracer injected at the feed end. Sources of contin-
uous ow RTD experimental results are listed in Table 1.
An example of RTD data from Ray et al. (1994) is plotted
in Fig. 4. The RTD is sometimes slightly skewed giving a
406 R. G. Sherritt et al. / Chemical Engineering Science 58 (2003) 401415
Fig. 4. Example of RTD in a continuous ow rotary kiln using a pulse
of tracer injected at feed end from Ray et al. (1994).
slight deviation from the dispersion model. To derive the
RTD equation, Eq. (3) is rst rewritten with dimensionless
variables as follows:
cC
c0
+
cC
c
=
1
Pe
c
2
C
c
2
. (4)
The dimensionless time, length and concentration variables
are:
0 =
t
t
, =
:
Z
, and C(0, ) =
c(t, :)
c
o
. (5)
The dimensionless Peclet number is related to the disper-
sion coecient by the following equation:
Pe =
u
:
Z
D
:
. (6)
Prior to injecting the tracer at the inlet end, there is no
tracer in the drum, which results in the following initial
condition:
C(0, ) = 0. (7)
Table 2
Sources of experimental data from batch rotating drums from which particle dispersion coecients in the axial direction are determined
Source Drum Granular Particle diameter
diameter (m) material (mm)
Rao, Bhatia, and Khakhar (1991) 0.076 Sodium 0.125, 0.214,
(bi)carbonate 0.388
Singh (1979) 0.10 Glass beads 0.680.78
Shoji, Hogg, and Austin (1973) 0.095 Silicon carbide 0.335
Hogg, Mempel, and Fuerstenau (1969) 0.102 Glass beads 0.100
0.051 Quartz 0.253
Hogg, Mempel, and Fuerstenau (1969) 0.10 Glass beads 0.090
Cahn and Fuerstenau (1967) 0.102 Lucite beads 6.35
Carley-Macauly and Donald (1962) 0.102 Sand 0.60.85
Wrightman and Muzzio (1998) 0.0106 0.066
Parker, Dijkstra, Martin, and Seville (1997) 0.136 Glass beads 1.5
0.144 Glass beads 3
0.100 Glass beads 3
The boundary conditions are (at 0 0):
C(0, 0)
1
Pe
cC(0, 0)
c
= ,(0),
cC(0, 1)
c
= 0, (8)
where ,(0) = 1 and ,(0) = 0.
Abouzeid et al. (1974) and Karra and Fuerstenau (1977)
gave the exact solution for the tracer residence time from
Eq. (4) with the conditions given by Eqs. (7) and (8). The
dimensionless standard deviation from the mean residence
time is related to the Peclet number by the following equa-
tion:
o
2
0
=
2
Pe

2
Pe
2
[1 exp(Pe)] , (9)
where the dimensionless standard deviation of the residence
time is the ratio of the standard deviation and the mean time.
For large Peclet numbers (Pe 50), which is the case for
low axial mixing in most rotating drums, Eq. (9) can be sim-
plied to the following equation that is used to approximate
the axial-dispersion coecient from the mean and standard
deviation of the tracer residence time.
D
:
=
Z
2
o
2
t
2t
3
. (10)
The mean residence time can also be calculated from the
measured hold-up and feed rate. The result agrees with the
mean time from the tracer response if there are no dead
zones in the drum.
2.4.2. Batch experiments
For batch or non-owing experiments, the net axial
velocity is zero, i.e. u
:
= 0. Sources of the results from
batch axial-mixing experiments are listed in Table 2. From
Eq. (3), the one-dimensional diusion equation reduces to
cc
ct
= D
:
c
2
c
c:
2
. (11)
R. G. Sherritt et al. / Chemical Engineering Science 58 (2003) 401415 407
Fig. 5. Initial tracer distributions used in a rotating drum to measure axial dispersion in batch experiments.
Axial-dispersion coecients for horizontal, rotating
drums have been determined from batch experiments using
several dierent initial conditions. Fig. 5 illustrates the ini-
tial conditions used in batch experiments to measure axial
mixing.
2.4.2.1. Half drum. Hogg, Cahn, Healy, and Fuerstenau
(1966), Rao et al. (1991), Cahn and Fuerstenau (1967),
Carley-Macauly and Donald (1962), and Wrightman and
Muzzio (1998) commenced their experiments with the
drum divided at the centre of its length. Half the drum
contained particles, which could be distinguished from the
particles in the other half of the drum as shown in Fig. 5a.
Consequently, the initial conditions are
C(0, :) = 0(Z}2 6: 0)
= 1(0 : 6Z}2). (12)
With no diusion at either end of the drum, the boundary
conditions are
cC(t, :)
c:

:=Z}2
= 0,
cC(t, :)
c:

:=Z}2
= 0.
(13)
The dispersion coecient can be determined from the
experimental results in a number of ways as shown by Hogg
et al. (1966). If the mixing time is large, then the solution
to Eq. (11) with initial and boundary conditions given by
Eqs. (12) and (13) is
C(t, :) =
1
2
+
2

exp
_

2
D
:
t
Z
2
_
sin
:
Z
. (14)
The axial-dispersion coecient at any time can be deter-
mined from the particle distribution by plotting C against
sin (:}Z). The result is a straight line with a slope equal
to (2})exp (
2
D
:
t}Z
2
). The axial-dispersion coecient is
calculated from this slope.
The standard deviation of the reference material concen-
tration at any long mixing time and the dispersion coecient
are related by the following equation:
o
2
:
(t) =
2

2
exp
_

2
2
D
:
t
Z
2
_
. (15)
A plot of !n o
:
against t gives a straight line with slope

2
D
:
}Z
2
. The axial-dispersion coecient can be calcu-
lated from this slope.
For short mixing times, the drum can be treated as an
innite cylinder and Eq. (11) can be solved to give
C(t, :) =
1
2
_
1 + erf
:
2

D
:
t
_
. (16)
A normal probability plot of C(t, :) versus : yields a straight
line of slope 0.5}(D
:
t)
0.5
.
The standard deviation of the reference material concen-
tration at short mixing time is related to the dispersion co-
ecient by the following equation:
o
2
:
(t) = o
2
:
(0)
_
1
4
Z
_
2D
:
t

_
. (17)
A plot of the variance o
2
:
versus square root of time gives
a straight line of slope (4}Z)(2D
:
})
0.5
.
For the data obtained from a DEM computer simula-
tion, the average location of the particles is more accurately
408 R. G. Sherritt et al. / Chemical Engineering Science 58 (2003) 401415
determined than the particle concentration. Kohring (1995)
gave the following solution for the average particle position:
: =
8Z

3
exp
_

2
D
:
t
Z
2
_
. (18)
A plot of !n : against t gives a straight line with slope of

2
D
:
}Z
2
. The axial-dispersion coecient from the DEM
experiment can be calculated from this slope.
2.4.2.2. Thin slice near the centre. Shoji et al. (1973)
began their batch experiments with a thin slice of tracer
loaded perpendicular to the axis at distance :
o
from the end
of the drum as shown in Fig. 5b. The initial condition is:
C(0, :) = 0 (0 6: 6Z),
C(0, :
o
) = 1. (19)
For short mixing times, the drum can be treated as having
innite length. The resulting solution of Eq. (11) for the
cumulative mass fraction of tracer is
_
:
0
C(t, :) d: =
1
2
_
1 erf
:
o
:
2

D
:
t
_
. (20)
A plot of the cumulative mass fraction of tracer on a
normal probability scale versus : gives a straight line from
which the dispersion coecient can be determined.
No method was given by Shoji et al. (1973) to obtain the
dispersion coecient from the concentration of tracer after
long mixing times when the eect of the end walls can no
longer be neglected.
2.4.2.3. Thin slice at end. Hogg et al. (1969) began batch
experiments with a thin slice of tracer loaded perpendicular
to the axis at one end of the drum as shown in Fig. 5c. The
corresponding initial conditions are
C(0, :) = 0 (0 6: 6Z),
C(0, 0) = 1. (21)
For short mixing times, the drum can be treated as a
semi-innite cylinder and the solution of Eq. (11) is given
by the following equation:
C(t, :) =
1
2

D
:
t
exp
_

:
2
4D
:
t
_
. (22)
The axial-dispersion coecient at any short time can be
determined from the experimental results by plotting (ln C)
against :
2
, which would yield a straight line with the slope
of 1}(4D
:
t). The axial-dispersion coecient is calculated
from the slope.
Hogg et al. (1969) did not specify a method to solve for
the dispersion coecient fromthe concentration of the tracer
after long mixing times when the eect of the far wall can
no longer be neglected.
2.4.2.4. Wide band. Singh (1979) began his experiments
with a wide band of tracer perpendicular to the axis near the
centre of the drum as shown in Fig. 5d. The initial conditions
are
C(0, :) = 0 (0 6: :
1
, :
2
: 6Z)
= 1 (:
1
6: 6:
2
). (23)
The eects of the end wall are included in the boundary
conditions:
cC(t, :)
c:

:=0
= 0,
cC(t, :)
c:

:=Z
= 0. (24)
Eq. (11) was solved with these initial and boundary con-
ditions using a nite dierence approximation and numeri-
cal integration. The dispersion coecient was found by trial
and error to get the best t to the measured concentration
proles.
2.4.2.5. Single particle. Parker et al. (1997) determined
the axial-dispersion coecient by tracking the random
movement of a single radioactively labelled tracer particle
in a batch drum. The tracer particle had the same proper-
ties as the other particles in the drum and is located so the
end walls do not aect its movement. At the start of each
time increment (t = 0), the tracer is at the origin : = 0,
as shown in Fig. 5e. The displacement for numerous short
time increments gives a distribution of displacements for
the specied time increment. Without any end eects on the
axial movement of the tracer particle, the drum was treated
as innitely long in the axial direction. The distribution of
the tracer displacement : was found by solving the diusion
equation with the same initial and boundary conditions as
the drum with a thin slice of tracer:
C(t, :) =
1

4tD
:
exp
_

:
2
4tD
:
_
. (25)
The axial-dispersion coecient is related to the standard
deviation of the tracer displacement in the axial direction
after time interval t by the following equation:
D
:
=
o
2
:
2t
. (26)
A plot of the variance of the tracer displacement and the
time interval gives a straight line with a slope of 2D
:
.
2.4.3. Empirical correlations for dispersion coecient
From their experimental results, several researchers have
made observations of the eect of various system parameters
on the axial-dispersion coecient in rotating drums. The
eects of rotational speed, volumetric ll, feed rate, drum
incline, and particle properties have been studied. Except
for one experiment by Rutgers (1965), the eect of drum
diameter on the axial dispersion has not been reported.
R. G. Sherritt et al. / Chemical Engineering Science 58 (2003) 401415 409
With all other variables held constant, increasing the
rotational speed causes the axial dispersion to increase.
Some experimenters such as Rutgers (1965) found that the
axial-dispersion coecient is proportional to the square
root of the speed, while others, such as Rao et al. (1991),
found the coecient to be directly proportional to speed.
Generally, the dispersion coecient decreases as the vol-
umetric ll is increased. Rutgers (1965) found the coe-
cient to be inversely proportional to the square root of the
volumetric ll fraction.
Some experimenters have studied the eects of rotational
speed, feed rate, and drum incline in continuous ow drums
(Abouzeid et al., 1974; Hehl et al., 1978). Because changing
any one of these variables in a owing drum will also change
the drum ll, the eects observed by these experimenters are
a combination of the eect of changing the ll in addition
to the manipulated variable.
A number of researchers have investigated the eects of
particle properties on the dispersion coecient. Abouzeid
et al. (1974) reported that doubling the particle diameter has
no eect on the axial dispersion, while Rao et al. (1991)
found that increasing the particle diameter decreases the dis-
persion coecient and Singh (1979) found increasing parti-
cle diameter increases the coecient. Rutgers (1965) found
that long particles give lower coecients than spherical par-
ticles and that sticky particles give larger coecients than
non-cohesive particles. Rao et al. (1991) found that increas-
ing the particle roughness increases the dispersion coe-
cient.
Attempts have also been made to correlate experimental
values of axial dispersion coecients with the operating pa-
rameters and the properties of granular material. Moriyama
and Suga (1974) gave the following equation based on re-
sults from continuous ow experiments:
D
:
u
:
Z
= 9.46 10
5
_
F
(2R)
3
jn
_
0.516
_
Z
2R
_
0.524

_
d

2R
_
0.604
(tan [)
5.55
. (27)
Sai et al. (1990) gave the following equation, also based
on continuous ow experiments in which the two tracer
materials were substantially dierent than the bulk material:
D
:
u
:
Z
= 0.000562
[
0.79

0.67
n
1.06
j
0.25
, (28)
where the angle of repose and bulk density j are properties
of the tracer material.
Sze, Ang & Tade (1995) developed the following equa-
tion for the axial dispersion of coal in a kiln fed a coal and
zircon mixture:
_
D
:
u
:
Z
_
c
=
0.0034

0.11
n
0.29
B
0.94
_
d
:
d
c
_
0.28
, (29)
where B is the zircon-to-coal mass feed ratio.
It is noted that Eqs. (27)(29) predict axial-dispersion
coecients in batch drums to be zero, i.e. no axial mixing,
which is not in agreement with experimental results that
indicate the axial-dispersion coecients to be about the same
in batch and continuous ow drums.
3. Radioactive particle tracer (RPT) experiments
In the previous section, published experimental studies for
the axial-dispersion coecients were summarised. In none
of the experiments were the coecients in the transverse
plane determined in the same drum for the same conditions.
Also, important variables such as the cycle time, dimensions
and residence times of the active and static regions, and
particle properties were seldom reported.
This section describes experiments involving a non-
invasive technique to measure the mixing of solids in a
rotating drum. The objective of the experiments was to
determine dispersion coecients for the same conditions
in the same drum by tracking a single radioactive particle
(Sherritt, 2001).
The experiments were similar to those performed by
Parker et al. (1997); hence, the axial-dispersion coecients
were determined in the same manner. At the same time,
data to determine dispersion coecient in the radial and
angular directions were collected. Other variables such as
the cycle time and the dimensions of the active and static
regions were to be determined from the same data.
3.1. Apparatus
A schematic of the apparatus is shown in Fig. 6. The
transparent plastic, horizontal drum used for the experiments
was 200 mm in diameter and 900 mm in length. The drum
had two tires that sat on four support rollers. A common
axle that was driven by a motor with a variable speed trans-
ducer connected two of the rollers. The inside surface of
the drum had several thin ridges of plastic to prevent slip-
ping between the bed of solids and the wall. The drum con-
tained a bed of uniform3 mmdiameter, spherical glass beads
(j
s
= 2500 kg}m
3
).
A facility for tracking a single radioactive particle in
multiphase uidised beds, developed at the

Ecole Polytech-
nique de Montr eal (Larachi, Chaouki, and Kennedy, 1995;
Cassanello, Larachi, Marie, Gey, and Chaouki, 1995) was
adapted to perform the same function for the rotary drum.
Nine 76 mm76 mm sodium iodide (NaI) scintillation de-
tectors were supported around the drum on sliding rails.
The detectors were placed at 90

spacing on both sides and

above the drum. The detectors were staggered at dierent
axial locations along the drum. The distance between the
front of the detector and the drum wall was between 72 and
184 mm, in order to achieve the greatest sensitivity without
exceeding the maximum count-rate of the detectors.
410 R. G. Sherritt et al. / Chemical Engineering Science 58 (2003) 401415
Fig. 6. Schematic diagram of rotating drum and detectors used in radioac-
tive particle tracking experiments.
The tracer particle was a sphere formed by melting a
mixture of soda lime powder and scandium oxide at high
temperatures. With 1012% Sc, the density and size of the
tracer particle was within 5% of that of a glass bead in the
drum. The tracer particle was coated with a diamond-like
carbon layer in order to prevent its rupture due to attrition
in the drum. The particle was activated in the institutions
SLOW-POKE nuclear reactor.
45
Sc captures a neutron to
produce
46
Sc which has a half-life of 83.8 d.
Nuclear decay of the tracer particle emits gamma-rays
that are detected by the scintillation detectors. With the
count-rate being approximately inversely proportional to
the square of the distance, the count-rate in each detector
depends on the distance from the tracer particle to the de-
tector. However, the amount of absorbing and scattering
material between the particle and the detector and the ef-
ciency of the detector also has a signicant eect on the
count-rate.
3.2. Procedure
A Monte Carlo simulation of gamma-ray emission and
detection was used to generate a map that related the
gamma-ray count-rates from the nine detectors to each po-
sition on a three-dimensional grid inside the drum (Larachi,
Kennedy, & Chaouki, 1994). Before an actual experiment,
the detection system was calibrated. A row of aluminium
probe ports along the drum wall allowed insertion of a
Plexiglas rod used to place a tracer particle at a number
of calibration locations in the stationary drum. The drum
Table 3
Operating parameters for radioactive particle tracking experiments
Test no. 1 2 3 4 5
Rotational speed (rpm) 15 15 5 25 15
Fill (vol %) 10 20 20 20 30
was turned to locate the tracer particle at dierent angular
locations. While the drum could not be rotated during the
calibration, the drum was turned so that the stationary bed
was in the approximate position that it would be during an
experiment. The calibration was used to adjust the linear
attenuation assumed for the contents of the drum.
In an experiment, a single tracer particle was placed in
the bed in the rotating drum. Gamma-rays were counted si-
multaneously from the nine detectors for 30 ms. After 4096
counting intervals over 128 s, the acquisition was momen-
tarily interrupted and the data were transferred and stored
in temporary computer memory. Data acquisition was trig-
gered to resume after the data transfer. After a tracking ex-
periment, which lasted about 1.53 h, the acquired data were
transferred to permanent computer memory and then con-
verted to ASCII format. These les were transferred to a
faster computer where the tracer co-ordinates were recon-
structed from the gamma-ray counts. A least squares method
was used to nd the grid point for the best agreement with
the count-rates for all nine detectors. A second least squares
search was used to nd the position of the tracer in the neigh-
bourhood of the grid point. The three-dimensional position
of the tracer particle was calculated for each 30 ms interval
for the 3 h period. Experiments using a three-phase uidised
bed indicated that this technique allowed the position of the
tracer particle to be determined with a typical precision of
5 mm (Larachi et al., 1994).
Five tests were performed using various rotational speeds
and ll levels. Table 3 gives the rotational speed and level
of ll for each test. All the tests exhibited rolling or cascad-
ing bed behaviour. Calibration of the detection system was
performed prior to the rst, second and fth tests.
4. Experimental results
4.1. Axial dispersion
Fig. 7 shows the axial position of the tracer particle in the
drum during a test. The particle randomly migrated along the
drums length. When the tracer particle migrated within one
drum diameter of the end wall, it was relocated to the centre
of the drum during an interruption in the data acquisition.
Assuming that the particle migration in the axial direc-
tion approximates diusion in an innite one-dimensional
medium, then the distribution of the particles displacements
in the axial direction after time interval t is related to the
R. G. Sherritt et al. / Chemical Engineering Science 58 (2003) 401415 411
Fig. 7. Random migration of tracer particle along the length of the rotating
drum during Test 1.
axial-dispersion coecient by the following equation:
C(:, t) =
1

4tD
:
exp
_

:
2
4tD
:
_
. (30)
Note that the normal distribution is dened as
C(:, t) =
1

2o
:
exp
_

:
2
2o
2
:
_
. (31)
Therefore, the axial-dispersion coecient is related to the
variance of the tracer displacement in the axial position after
time interval t by the following equation:
D
:
=
o
2
:
2t
, (32)
where the variance of the tracer displacement is determined
by the following equation:
o
2
:
=

:
2
(

:)
2
N(N 1)
. (33)
The variance of the measured displacement is the sum of
the variance due to the tracer particle dispersion and the
variance due to measurement error:
o
2
:
= o
2
D
+ o
2
R
. (34)
The displacement variance due to dispersion increases as
the time interval increases but the variance due to measure-
ment error is constant. For short time intervals, the error
in measuring the axial location increases the calculated dis-
persion coecient. As the size of the time increment is in-
creased the measurement error becomes less signicant. By
determining the dispersion coecient for dierent time in-
tervals the eect of the measurement error can be determined
and eliminated.
Using the time and position data collected during a test,
the variance of the displacement in the :-direction was de-
termined for time intervals ranging from 2 to 128 s. An
axial-dispersion coecient was calculated for each pair of
Fig. 8. Axial-dispersion coecient determined from tracer displacement
distribution for various time intervals in ve experiments listed in Table 3.
variance and time interval using Eq. (32). The process was
repeated for each of the ve tests, with the results plotted in
Fig. 8. As expected, the calculated axial-dispersion coe-
cient decreases as the time interval increases. That is, as the
size of the time interval increases, the measurement error
becomes less signicant and the true axial-dispersion co-
ecient is approached. The axial-dispersion coecient ap-
proached a value of the order of 10
5
m
2
}s, but it was still
decreasing slightly for a maximum time interval of 128 s.
The dispersion coecient values corresponding to the max-
imum time interval of 128 s were used for comparing our
results with literature data.
5. Correlations for axial-dispersion coecient
As mentioned previously, the sources of the literature
results of RTD experiments on continuously fed rotat-
ing drums are listed in Table 1. For the cases where the
axial-dispersion coecient is not reported, it was deter-
mined by one of the following methods:
1. if the Peclet number is given, then the dispersion coe-
cient was determined using Eq. (6),
2. if the mean and standard deviation of the residence time
are given, then the dispersion coecient was determined
using Eq. (10), and
3. if only the RTD curve is given, then the mean and stan-
dard deviation of the residence time were calculated and
the dispersion coecient was determined using Eq. (10).
Following the above approach, a total of 124 coecient
values were obtained from 15 sources. Details of the operat-
ing parameters such as drum diameter, rotational speed, ll
and particle type and size are reported elsewhere (Sherritt,
2001). Most researchers varied the rotational speed and ll,
but considered only one type of material and drum diam-
eter. Drum diameters ranged from 0.08 to 0.90 m. Values
412 R. G. Sherritt et al. / Chemical Engineering Science 58 (2003) 401415
Table 4
Parameters (regression coecients) for axial-dispersion coecients in Eq. (35), X 0.5
Bed behaviour ln k a b c d
All modes
n}n
c
1
6.81 0.33 0.39 0.06 1.15 0.12 0.46 0.04 0.43 0.11
Slumping
n}n
c
0.03
12.36 1.69 0.35 0.18 0.85 0.50 0.21 0.07 0.50 0.19
Rolling and
cascading
0.03 n}n
c
0.3
7.45 0.53 0.44 0.14 1.29 0.16 0.35 0.06 0.55 0.16
Cataracting
0.3 n}n
c
1.0
3.15 0.96 0.90 0.35 1.22 0.22 0.87 0.10 0.02 0.36
of axial-dispersion coecients, encompassing all modes of
bed behaviour, range from 1 10
7
to 1 10
4
m
2
}s.
Sources of experimental data from batch rotating drums
from which the particle axial-dispersion coecients were
determined are listed in Table 2. The nine sources provide
55 coecient values. The coecients and most of the op-
erating parameters such as drum diameter, rotational speed,
ll and particle type and size are given in Sherritt (2001).
Most researchers varied the rotational speed and ll, but con-
sidered only one type of material and drum diameter. The
drum diameters range from 0.076 to 0.144 m. The dispersion
coecients range from 1 10
7
to 1 10
4
m
2
}s.
5.1. Correlating published data
The following equation was used for multiple regressions
to relate the combined 179 data points for axial-dispersion
coecients (summarised in Tables 1 and 2) with respect to
the relative speed, diameter, particle size and volumetric ll
(Sherritt, 2001).
D
:
= k(n}n
c
)
c
(2R)
b
d
c

X
d
, (35)
where D
:
is the axial-dispersion coecient (m
2
}s), n}n
c
the
relative speed, fraction of critical, R the drum radius (m),
d

the particle diameter (m), and X the ll level, volume

fraction.
Optimum values of the tted parameters k, a, b, c, and d
in Eq. (35) were determined from a regression analysis for
all the data together as well as for each individual type of
bed behaviour. Because some experimenters did not report
the type of bed behaviour, drums rotating at less than 3% of
the critical speed were assumed to have slumping behaviour.
Drums rotating between 3% and 30% of critical speed were
assumed to have rolling or cascading behaviour. Drums ro-
tating at more than 30% of critical speed were assumed to
have cataracting behaviour. The eect of bed depth on the
bed behaviour was neglected. However, only drums that are
less than 50% full were used in the regressions. The results
of regression calculations are summarised in Table 4.
The test for signicance (F-test) of the regression model
gave satisfactory statistical indicators for all modes together
Fig. 9. Comparison of axial-dispersion coecients given by Eq. (35)
with tting parameters for all modes of bed behaviour to experimental
coecients.
and for individual bed behaviour modes. In Table 4, the
exponent for ll volume (X) in the cataracting mode has
a large standard error; however, its contribution to the cal-
culated axial-dispersion coecient is small due to its low
value, i.e. d = 0.02. Axial-dispersion coecients pre-
dicted by Eq. (35) with parameters from Table 4 for all
bed behaviour modes are compared to the measured data in
Fig. 9. The error between the predicted and experimental
values of axial-dispersion coecients is attributed to dif-
ferences in the properties (stickiness, roughness, shape)
of the variety of materials used, sliding behaviour, and
experimental error. It was noted that the coecients from
batch experiments are slightly lower than the coecients
from continuous ow experiments at the same speed, ll
and diameters (Sherritt, 2001). This suggests that some
axial mixing in continuous ow drums could be due to the
non-uniform axial velocity from the core to the periphery
of the bed.
R. G. Sherritt et al. / Chemical Engineering Science 58 (2003) 401415 413
5.1.1. Eect of drum and particle diameters
The values of parameters in Table 4 indicate that the
axial-dispersion coecient increases when the drum or the
particle diameter is increased. The regression results indi-
cate that, for all types of behaviour, the coecient is approx-
imately proportional to the drum diameter, i.e. (D
:
R).
The majority of tests used to derive the correlations utilised
drums with a diameter of about 0.1 m, although the largest
drum was 0.90 m in diameter. This is much smaller than the
diameter of a typical industrial drum that is 35 m in di-
ameter. Therefore, considerable extrapolation is involved in
applying Eq. (35) to industrial-size drums.
For all types of behaviour, the axial-dispersion coecient
increases proportionally with the square root of the particle
diameter (D
:
d
0.5

).
5.1.2. Eect of rotational speed
For the slumping behaviour, the axial-dispersion coef-
cient tends to decrease with an increase in the rotational
speed. For rolling and cascading behaviours, the coecient
is approximately proportional to the square root of the
speed (D
:
n
0.5
). For cataracting behaviour, the coecient
is nearly directly proportional to the rotational speed (i.e.,
D
:
n).
5.1.3. Eect of ll
For slumping, rolling and cascading behaviours, the
axial-dispersion coecient decreases as the volumetric ll
is increased. The coecient is approximately inversely pro-
portional to the square root of the ll fraction (D
:
X
0.5
).
As mentioned previously, for the cataracting behaviour, the
ll has negligible eect on the coecient, i.e. (D
:
X
0
).
The test of signicance on individual regression coe-
cients (t-test) indicated that there is a poor relationship
between the coecient and the ll for the cataracting be-
haviour. All other regression coecients for both individ-
ual bed behaviour and all modes together are statistically
signicant.
5.1.4. Comparison to drums with lifting ights
Our laboratory has reported axial-dispersion coecients
based on experiments with rotating drums with ights at-
tached to the wall that lift and spill the particles (Sherritt,
Behie, and Mehrotra, 1996, Chapter 18). The coecients
for drums with lifting ights are about two orders of magni-
tude larger than those without lifting ights, and they range
between 1 10
5
and 1 10
3
m
2
}s. This is expected con-
sidering the additional energy imparted to the particles in
drums with lifting ights.
With or without lifting ights, the eect of drum diameter
on the coecient is about the same (D
:
R). The eect
of rotational speed is similar to that in a cataracting drum
(D
:
n). This is likely due to the particles bouncing after
made airborne in both a cataracting drum and a drum with
lifting ights. Unlike drums without ights, the dispersion
coecient appears to increase with hold-up in drums with
ights (D
:
X).
5.2. Comparison with radioactive tracer experiments
Finally, axial-dispersion coecients predicted from
Eq. (35) for four of the ve radioactive tracer experiments
(i.e. Expts 14) range from 6 10
6
to 1 10
5
m
2
}s.
These predictions agree well with the measured values
corresponding to the largest time interval in Fig. 8.
6. Conclusions
Mixing in the axial direction is purely diusive and is
caused by random collisions of particles in the active region.
Axial mixing can be described by a one-dimensional diu-
sion equation. The axial-dispersion coecient is the same
in continuous ow and batch drums for the same condi-
tions. The axial-dispersion coecient ranges from 10
7
to
10
4
m
2
}s. The coecient is proportional to the drum diam-
eter for all types of bed behaviour and proportional to the
square root of the particle diameter. The dependence of the
coecient on the rotational speed depends on the type of bed
behaviour. For rolling and cascading bed behaviour, the co-
ecient is proportional to the square-root of the speed. For
scale-up, the coecient increases proportionally with the
drum diameter assuming that the fraction of critical speed,
ll and particle size remain constant.
Notation
a, b, c, d tting parameters in Eq. (35)
ATP AOSTRA Taciuk Processor
B zircon-to-coal mass feed ratio
c tracer concentration, kg/m
3
c
o
maximum tracer concentration, kg/m
3
C fraction of maximum tracer concentration
d

particle diameter, m
D
r
radial-dispersion coecient, m
2
}s
D
:
axial-dispersion coecient, m
2
}s
DEM discrete element modelling
q gravitational acceleration, 9.81 m}s
2
k tting parameter in Eq. (35), m}s
2
n rotational speed, rpm
n
c
critical rotational speed, rpm
N number of stages or samples
Pe Peclet number
R drum radius, m
RTD residence time distribution
t time, s
t mean residence time, s
u
:
mean axial velocity, m}s
X volumetric ll fraction
414 R. G. Sherritt et al. / Chemical Engineering Science 58 (2003) 401415
: axial distance, m
Z drum length, m
Greek letters
angle of drum incline to horizontal, deg.
,(0) concentration function
[ dynamic angle of repose, rad
0 dimensionless time
j bulk density, kg}m
3
o
t
residence time standard deviation, s
o
0
dimensionless residence time standard deviation
o
:
standard deviation of displacement in axial di-
rection, m
fraction of drum length
Subscripts
c critical
particle
: axial direction
Acknowledgements
The authors acknowledge the generous nancial support
of UMATAC Industrial Processes and the Natural Sciences
and Engineering Research Council of Canada (NSERC).
References
Abouzeid, A. S. M. A., Mika, T. S., Sastry, K. V., & Fuerstenau,
D. W. (1974). The inuence of operating variables on the residence
time distribution for material transport in a continuous rotary drum.
Powder Technology, 10, 273288.
Ang, H. M., Tade, M. D., & Sze, M. W. (1998). Residence time
distribution for a cold model rotary kiln. The Australian IMM
Proceedings, 1, 1116.
Baumann, G., J anosi, I. M., & Wolf, D. E. (1995). Surface properties and
ow of granular material in a two-dimensional rotating-drum model.
Physical Review E, 51(3), 18791888.
Black, J. M. (1988). Particle motion and heat transfer in rotary drums.
Ph.D. thesis, Edinburgh University.
Buchholtz, V., P oschel, T., & Tillemans, H. (1995). Simulation of rotating
drum experiments using non-circular particles. Physica A, 216(3),
199212.
Cahn, D. S., & Fuerstenau, D. W. (1967). Simulation of diusional mixing
of particulate solids by Monte Carlo techniques. Powder Technology,
1, 174182.
Carley-Macauly, K., & Donald, M. (1962). The mixing of solids in
tumbling mixersI. Chemical Engineering Science, 17, 493506.
Cassanello, M., Larachi, F., Marie, M-N., Guy, C., & Chaouki, J. (1995).
Experimental characterization of the solid phase chaotic dynamics
in three-phase uidization. Industrial and Engineering Chemistry
Research, 34(9), 29712980.
Hatzilyberis, K. S., & Androutsopoulos, G. P. (1999). An RTD study
for the ow of lignite particles through a pilot rotary dryer. Drying
Technology, 17(4&5), 745757.
Hehl, M., Kroger, H., Helmrich, H., & Schugerl, K. (1978). Longitudinal
mixing in horizontal rotary drum reactors. Powder Technology, 20,
2937.
Henein, H., Brimacombe, J., & Watkinson, A. P. (1983a). The modeling
of transverse solids motion in rotary kilns. Metallurgical Transactions
B, 14B, 207220.
Henein, H., Brimacombe, J., & Watkinson, A. P. (1983b). Experimental
study of transverse bed motion in rotary kilns. Metallurgical
Transactions B, 14B, 191205.
Hogg, R., Cahn, D., Healy, T., & Fuerstenau, D. (1966). Diusional
mixing in an ideal system. Chemical Engineering Science, 21,
10251038.
Hogg, R., Mempel, G., & Fuerstenau, D. W. (1969). The mixing of trace
quantities into particulate solids. Powder Technology, 2, 223228.
Karra, V. K., & Fuerstenau, D. W. (1977). Material transport in a
continuous rotating drum. Eect of discharge plate geometry. Powder
Technology, 16, 2328.
Kelbert, F., & Royere, C. (1991). Lateral mixing and heat transfer in a
rolling bed. International Chemical Engineering, 31(3), 441449.
Kohav, T., Richardson, J., & Luss, D. (1995). Axial dispersion of
solid particles in a continuous rotary kiln. A.I.Ch.E. Journal, 41(11),
24652476.
Kohring, G. A. (1995). Studies of diusional mixing in rotating drums
via computer simulations. Journal of Physics I France, 5, 15511561.
Larachi, F., Chaouki, J., & Kennedy, G. (1995). 3-D mapping of solids
ow elds in multiphase reactors with RPT. A.I.Ch.E. Journal, 41(2),
439443.
Larachi, F., Kennedy, G., & Chaouki, J. (1994). A -ray detection system
for 3-D particle tracking in multiphase reactors. Nuclear Instruments
and Methods in Physics Research, A, 338, 568576.
Lebas, E., Hanrot, F., Ablitzer, D., & Houzelot, J.-L. (1995). Experimental
study of residence time, particle movement and bed depth prole
in rotary kilns. Canadian Journal of Chemical Engineering, 73(2),
173180.
McCarthy, J. J., & Ottino, J. M. (1998). Particle dynamics simulation:
A hybrid technique applied to granular mixing. Powder Technology,
97, 9199.
McTait, G. E., Scott, D. M., & Davidson, J. F. (1998). Residence time
distribution of particles in rotary kilns. Fluidization IX, Proceedings
of the ninth Engineering Foundation Conference on Fluidization,
Durango, CO, May 1722.
Mellmann, J. (2001). The transverse motion of solids in rotating cylinders
forms of motion and transition behavior. Powder Technology, 118,
251270.
Moriyama, A., & Suga, T. (1974). Axial dispersion and residence time
distribution of spherical particles in rotary kiln. Tetsu-to-Hague, 60(9),
12831288.
Mu, J., & Perlmutter, D. (1980). The mixing of granular solids in a rotary
cylinder. A.I.Ch.E. Journal, 26(6), 928934.
Parker, D. J., Dijkstra, A. E., Martin, T. W., & Seville, J. P. K. (1997).
Positron emission particle tracking studies of spherical particle motion
in rotating drums. Chemical Engineering Science, 52(13), 20112022.
Rao, S., Bhatia, S., & Khakhar, D. (1991). Axial transport of granular
solids in rotating cylinders. Part 2: Experiments in a non-ow system.
Powder Technology, 67, 153162.
Ray, A. K., Prasad, H. S., Ray, H. S., & Sen, P. K. (1994). Residence
time in a rotary kiln for sponge iron making prediction from kinetic
data. Steel India, 17(1), 1620.
Rogers, R. S. C., & Gardner, R. P. (1979). A Monte Carlo method
for simulating dispersion and transport through horizontal rotating
cylinders. Powder Technology, 23, 159167.
Rutgers, R. (1965). Longitudinal mixing of granular material owing
through a rotating cylinder Part II. Experimental. Chemical
Engineering Science, 20, 10891100.
Sai, P. S. T., Surender, G. D., Damodaram, A. D., Suresh, V., Philip,
Z. G., & Sankaran, K. (1990). Residence time distribution and material
ow studies in a rotary kiln. Metallurgical Transactions B, 21B, 1005.
Sherritt, R. G. (2001). Three-Dimensional Particle Diusion in a
Rotating Drum Reactor. Ph.D. thesis, University of Calgary, Calgary,
Canada.
R. G. Sherritt et al. / Chemical Engineering Science 58 (2003) 401415 415
Sherritt, R. G., Behie, L. A., & Mehrotra, A. K. (1996). The movement
of solids in ighted rotating drums. In N. P. Cheremisino (Ed.),
Mixed-ow hydrodynamics, advances in engineering uid mechanics
series, pp. 421441, Texas: Gulf Publishing Co.
Shoji, K., Hogg, R., & Austin, L. G. (1973). Axial mixing of particles
in batch ball mills. Powder Technology, 7, 331336.
Singh, D. (1979). A fundamental study of the mixing of solid particles.
Ph.D. thesis, University of Rochester.
Sugimoto, M. (1968). An estimation of the residence time distribution of
particles owing through a rotary cylinder. Kagaku Kogaku, 32(3),
291.
Sze, M. W., Ang, H. M., & Tade, M. O. (1995). Particle dynamics in a
cold rotary kiln. Chemeca 95 Proceedings, 4, 2732.
Tscheng, S. (1978). Convective heat transfer in a rotary kiln. Ph.D.
thesis, University of British Columbia, Vancouver, Canada.
Turner, R., Wright, B. C., & Taciuk, W. (1989). Oil shale processing
with the AOSTRA Taciuk Processor. Proceedings of the Fourth
UNITAR/UNDP Conference, Calgany, Canada, Vol. 5, No. 237.
Wes, G., Drinkenburg, A., & Stemerding, S. (1976). Solids mixing and
residence time distribution in a horizontal rotary drum reactor. Powder
Technology, 13, 177184.
Wrightman, C., & Muzzio, J. F. (1998). Mixing of granular material in
a drum mixer undergoing rotational and rocking motions I. Uniform
particles. Powder Technology, 98, 113124.
Yamane, K., Nakagawa, M., Altobelli, S. A., Tanaka, T., & Tsuji, Y.
(1998). Steady particulate ows in a horizontal rotating cylinder.
Physics of Fluids, 10(6), 14191427.