This action might not be possible to undo. Are you sure you want to continue?
https://www.scribd.com/doc/62134091/ModellingofRotaryKilns1
07/17/2014
text
original
MODELLING OF ROTARY KILNS
PROEFSCHRIFT
ter verkrijging van
de graad van doctor aan de Universiteit Twente,
op gezag van de rector magnificus,
Prof.dr. F.A. van Vught,
volgens besluit van het College voor Promosies
in het openbaar te verdedigen
op vrijdag 26 januari 2001 te 16:45 uur
door
Michael David Heydenrych
geboren op 13 juni 1959
te Pietersburg, ZuidAfrika
ii
Dit proefschrift is goedgekeurd door de promoteren
Prof.dr.ir. G.F. Versteeg
en
Prof.dr.ir. J.A.M. Kuipers
en de assistentpromotor
Dr.ir. A.B.M. Heesink
iii
to my father
iv
Samestelling promotiecommissie:
prof.dr. W.E van der Linden, voorsitter Universiteit Twente
prof.dr.ir. J.A.M. Kuipers, promotor Universiteit Twente
prof.dr.ir. G.F. Versteeg, promotor Universiteit Twente
dr.ir. A.B.M. Heesink, ass. promotor Universiteit Twente
prof.dr.ir. C.J. Asselbergs Universiteit Twente
prof.dr.ir. L. Lefferts Universiteit Twente
prof.dr.ir. P.L. de Vaal Universiteit Pretoria
prof.dr.ir. G. Brem Universiteit Twente
Cover: Effectiveness factor vs k/ω for the passive layer model overlaid by a rotary kiln
crosssection showing evaporative mass transfer in the active layer.
Copyright © 2001 M.D. Heydenrych, Centurion, South Africa
No part of this book may be reproduced in any form by any means, nor
transmitted, nor translated into a machine language without written
permission from the author.
Heydenrych, M.D.
Modelling of Rotary Kilns
Thesis University of Twente, The Netherlands
ISBN 9036515440
Printing and binding: grafisch centrum twente, Enschede, The Netherlands
v
SUMMARY
This work primarily covers mass transfer in rotary kilns. The most
important contribution is the identification of the dimensionless numbers
that contribute to mass transfer in rotary kilns, and their relation to each
other. In this respect, this work is also applicable to heat transfer in rotary
kilns, where analogous dimensionless numbers can be identified. Another
important contribution is the recognition of the important contribution of
the passive layer in mass transfer.
Fig. 1 shows the nomenclature used in this summary, and the various zones
found within a rotary kiln. In the active layer, particles slide over each
other in granular flow. They are returned to the top part of the bed in the
passive layer, where particles move as a solid mass concentrically around
the axis of the kiln.
Figure 1. Radial cross section of a kiln, showing the nomenclature.
The first part of the work considers the reaction of the gas as it moves in
plug flow through the passive layer within the interparticle voids; the active
vi
layer is considered to be infinitely thin. The effectiveness factor η for the
bed is found to be dependent on the bed fill (represented by h/R) and the
ratio of the rate constant to the rotation speed (k/ω), shown in Fig. 2.
Comparison against experimental data shows that this relationship is only
applicable to slower reactions (k/ω < 1).
Figure 2. Effectiveness factor increases with decreasing bed fill.
The effect of diffusion within the passive layer was investigated based on a
twodimensional continuity equation for the radial cross section, again with
the assumption of an infinitely thin active layer. This requires the use of an
additional dimensionless constant: either a Peclet number (εωR
2
/D
e
) or a
Thiele modulus (
e
D k R / ). It was shown that for practical values of
Peclet number, diffusion within the passive layer can be neglected, but at
higher values of k/ω, active layer modelling is necessary to correctly
predict mass transfer.
By making the (justified) assumption that the velocity gradient of particles
(m) within the active layer is constant at any point, the active layer depth
can be estimated at any distance x along the bed surface using
vii
1 /
2 2 2
−
− −
·
ω m
h x R
d
The action of particles sliding over each other promotes movement of gas
through the interparticle voids. A correlation for an active layer dispersion
coefficient has been derived:
D
al
= 0.246fm d
p
2
+ ε
2
D
a
where f is a directionality factor; with equal dispersion in all directions,
f = 1/6. With fast reactions mass transfer can be estimated in terms of a
mass transfer coefficient on the bed surface using:
k D k
al s
ε ·
This does not fully predict the dependence of mass transfer on rotation
speed. It is also necessary to take into account the periodic renewal of gas
in the voids between the particles within the active layer. This is done by
adapting the value of the kinetic constant using
( )
m k
e m k
/
1 '
−
− ·
where k′ is the effective rate constant applicable to the active layer region.
Using m = 70ω, and the above assumptions, a good prediction of mass
transfer could be obtained for fast reactions.
Finally, a model was developed that built on the principles established for
both the active and passive layers. This combined model assumes
continuity of concentration at the interface of the active and passive layers,
where gas is entrained into the bed through rotation (Fig. 3).
viii
Figure 3. A typical concentration profile using the combined model
(clockwise rotation).
This model predicts that it is seldom possible to use either the passive layer
model alone, or the active layer model alone to accurately predict mass
transfer, and that usually both regions in a rotary kiln contribute
significantly to mass transfer. We also find that when dispersion effects
dominate over molecular diffusion effects in the active layer (expected for
particle size > 3 mm), then scaleup effects can be represented by the ratio
of kiln radius to particle diameter (R/d
p
) instead of the active layer Thiele
modulus.
An example of the predictions of the combined model is given in Fig. 4 for
a bed fill of 5.2%. For such a shallow bed, the effectiveness factor is
strongly dependent on R/d
p
. It is also clear from the shape of the curves
that the passive layer model alone is inadequate at predicting the
effectiveness factor, even for low k/ω.
ix
Figure 4. Effectiveness factors for a rotary kiln with bed fill 5.2%. The
thick line shows the prediction of the passive layer model (Chapter 2).
The value of the work lies in two aspects: firstly, the recognition of the
contribution of the passive layer towards mass transfer, and secondly, the
relative simplification of the modelling approach by identifying and using
appropriate dimensionless numbers to characterise the system. In this way,
at a given bed depth, the effectiveness factor is characterized by a single
graph such as that of Fig. 4. For heat transfer, a similar analogy should be
possible.
R/d
p
= 100
R/d
p
= 10
R/d
p
= 1000
x
SAMENVATTING
Dit proefschrift behandelt stofoverdracht in roterende ovens. Dergelijke
gasvast reactoren worden veelvuldig in de industrie toegepast, waarbij de
cement oven wellicht het bekendste voorbeeld vormt. De opschaling van
deze reactoren geschiedt vaak nog op basis van empirie omdat een gedegen
theoretische beschrijving van de processen nog ontbreekt. Met dit
proefschrift wordt getracht om deze leemte op te vullen. Er wordt met
name aandacht besteed aan de identificatie van dimensieloze kentallen die
het stoftransport in roterende ovens karakteriseren en hun onderlinge
afhankelijkheid. Door toepassing van deze dimensieloze kentallen zijn de
resultaten ook toepasbaar voor de beschrijving van het analoge warmte
transport in roterende ovens. Daarnaast vormt de analyse van de
afzonderlijke stoftransportmechanismen een belangrijk onderdeel van dit
proefschrift.
Figuur 1. Dwarsdoorsnee van een roterende oven
Figuur 1 laat de verschillende zones zien, zoals die in een roterende oven
worden onderscheiden. In de actieve laag bewegen de deeltjes langs en
over elkaar heen naar beneden. Door de roterende beweging van de oven
xi
worden de deeltjes die het eind van de actieve laag hebben bereikt, via een
roterende beweging door de onderliggende passieve laag weer naar het
bovenste gedeelte van de actieve laag getransporteerd. De passieve laag
roteert hierbij als een star lichaam om de as van de oven. Zowel in de
actieve als in de passieve laag kunnen de deeltjes reageren met het gas dat
vanuit het “lege” deel van de oven (het zogenaamde freeboard) continu
wordt aangevoerd.
Na een inleiding over de relevantie van roterende ovens voor de industrie,
wordt in het tweede hoofdstuk van dit proefschrift het transport en de
reactie van het gas in de passieve laag beschouwd. Het gas, opgesloten
tussen de deeltjes, beweegt daarbij als het ware in propstroom concentrisch
door de passieve laag. De actieve laag wordt daarbij oneindig dun
verondersteld. Er wordt een effectiviteitsfactor η gedefinieerd als de
verhouding tussen de werkelijke omzettingssnelheid in het bed en de
snelheid die zou worden bereikt wanneer overal in het bed de concentratie
van de gasvormige reactant gelijk zou zijn aan de concentratie in het
freeboard boven het bed. Deze blijkt sterk afhankelijk te zijn van de
vulhoogte van de oven (1h/R) en de verhouding tussen de reactiesnelheids
Figuur 2. Effectiviteitsfactor bij verschillende vulhoogtes.
xii
constante en de rotatiesnelheid (k/ω), zoals weergegeven in Figuur 2. Een
vergelijking met experimentele data (afkomstig uit de literatuur en eigen
data) laat zien dat de gevonden afhankelijkheid alleen toepasbaar voor
relatief langzame reacties (k/ω < 1).
In het derde hoofdstuk wordt het effect van diffusie binnen de passieve
zone bestudeerd met behulp van de continuïteitsvergelijking voor een
dwarsdoorsnee van de roterende oven. Hierbij is wederom aangenomen dat
de actieve laag oneindig dun is. Dit model resulteert in een extra
dimensieloos kental: ofwel een Pecletkental (εωR
2
/D
e
) ofwel een Thiele
modulus (
e
D k R / ). In de praktijk blijkt diffusie in de passieve zone een
verwaarloosbaar kleine bijdrage aan het stoftransport te leveren. Voor hoge
waarden van k/ω (k/ω >1) blijkt ook dit aangepaste passievezone model
niet in staat om de experimentele resultaten te beschrijven. Geconcludeerd
wordt dan ook dat een nadere beschouwing van de processen die optreden
in de actieve laag noodzakelijk is.
In hoofdstuk 4 worden de hydrodynamica van en de
stofoverdrachtsprocessen in de actieve laag nader onder de loep genomen.
Door de (gerechtvaardigde) aanname dat de snelheidsgradiënt van de
deeltjes (m) binnen de actieve laag constant is op elke positie binnen die
laag, kan de dikte van de actieve laag worden afgeschat voor elke positie x
(zie Figuur 1) langs het grensvlak volgens:
1 /
2 2 2
−
− −
·
ω m
h x R
d
De beweging van de deeltjes over en langs elkaar heen veroorzaakt
gastransport door het steeds weer creëren en opbreken van interstitiële
ruimtes. Op basis hiervan is een overall dispersiecoëfficiënt voor de gasfase
afgeleid:
xiii
D
al
= 0.246fm d
p
2
+ ε
2
D
a
In deze vergelijking verdisconteert f de richting van het transport. Voor
gelijke dispersie in elk van de orthogonale richtingen geldt f = 1/6. De
symbolen d
p
en D
a
staan respectievelijk voor de deeltjesdiameter en de
moleculaire diffusiecoëfficiënt van het reagerende gas, terwijl ε de
(gemiddelde) porositeit van de actieve laag voorstelt. In het geval van
snelle gasvast reacties kan de stofoverdracht worden geïnterpreteerd in
termen van een stofoverdrachtscoëfficiënt k
s
, betrokken op het totale bed
oppervlak:
k D k
al s
ε ·
Hierbij wordt echter het effect van de rotatiesnelheid op het stoftransport
niet volledig verdisconteerd. Het is noodzakelijk om ook het effect van de
periodieke verversing van gas in de holtes tussen de deeltjes te
verdisconteren. Dit kan worden gedaan door aanpassing van de
kinetiekconstante, volgens:
( )
m k
e m k
/
1 '
−
− ·
waarin k′ de effectieve reactiesnelheidsconstante voor de actieve zone
voorstelt. Voor m=70ω en bovengenoemde aannames wordt een goede
beschrijving van de in de literatuur gerapporteerde experimentele data
gevonden.
Tot slot wordt in hoofdstuk 5 een model ontwikkeld dat gebaseerd is op de
eerder genoemde beginselen voor de actieve en passieve zones. Dit
gecombineerde model gaat uit van een continu concentratieprofiel aan het
grensvlak tussen de actieve en de passieve zones. Het gecombineerde
xiv
Figuur 3. Een typisch concentratieprofiel volgens het gecombineerde
model (kloksgewijze rotatie)
model voorspelt dat slechts bij hoge uitzondering alleen het passieve zone
model of alleen het actieve zone model voldoet. Meestal zijn beide zones
belangrijk in het beschrijven van het stoftransport in een roterende oven.
Wanneer in de actieve zone moleculaire diffusie te verwaarlozen is ten
opzichte van gasfase dispersie (en dit is aannemelijk voor deeltjes groter
dan 3 mm), kunnen opschaaleffecten worden verdisconteerd via de
verhouding van ovenstraal ten opzichte van deeltjesdiameter (R/d
p
).
In Figuur 4 is een voorbeeld van een berekeningsresultaat van het
gecombineerde model gegeven voor een kleine vulhoogte. De effectiviteits
Figuur 4. Effectiviteitsfactor voor een roterende oven met een vulgraad
van 5.2%. Dikke lijn geeft de voorspelling van het passievezone model.
R/d
p
= 100
R/d
p
= 10
R/d
p
= 1000
xv
factor blijkt sterk af te hangen van (R/d
p
). Ook blijkt duidelijk dat het
passievezone model alleen niet in staat is de effectiviteitsfactor goed te
beschrijven, zelfs niet voor lage waarden van k/ω.
De resultaten van dit proefschrift tonen het belang aan van zowel de actieve
als de passieve laag voor het totale stoftransport. De term “passief” is zeker
niet altijd op zijn plaats. Door enkele vereenvoudigende aannamen en het
gebruik van de juiste dimensieloze kentallen kan de stofoverdracht in
roterende ovens goed worden gekarakteriseerd. Op deze wijze kan, voor
een gegeven vulgraad van de oven, de effectiviteitsfactor als functie van de
procesparameters in een enkele grafiek worden weergegeven. Voor
warmtetransport is een analoge analyse mogelijk.
xvi
CONTENTS
1. Introduction 1
The radial vs. axial direction
The flow regimes in rotary kilns
The need for modelling of rotary kilns
An overview of this work
2. Mass transfer in rolling rotary kilns: a novel approach 11
Introduction
Modelling studies
Model development
Total depletion model
Model validation
Conclusion
3. Diffusion effects in rotating rotary kilns 37
Introduction
Mathematical model
Results
Discussion
Conclusions
4. Hydrodynamics and dispersion in the active layer of
rolling rotary kilns
53
Introduction
Hydrodynamics of the active layer
Dispersion within the active layer
Concentration profiles and mass transfer
Model validation
Discussion and conclusion
xvii
5. A combined mass transfer model for both active and
passive layers in rolling rotary kilns
77
Introduction
Combined model
Model predictions
Comparison with measured data
Conclusions
Publications
101
Acknowledgements
103
Curriculum Vitae
105
Chapter 1  Introduction
1
Chapter 1  Introduction
Rotary kilns are found in many
processes that involve solids
processing. These include drying,
incineration, mixing, heating,
cooling, humidification, calcination,
reducing, sintering and gassolid
reactions (Jauhari et. al. 1998). The
most common and industrially
important application of rotary kilns
is in cement production; all major
producers use the rotary kiln as their
equipment of choice. Cement kilns
can be very large. Although this is not
a cement kiln, Fig. 1 shows a 500 tpd
lime kiln from Greer lime.
Figure 1. The scale of this 500 tpd
lime kiln can be judged from the
pickup truck to the left of the kiln.
Another important application of rotary kilns is for the incineration of
waste materials. Rotary kilns are popular for this role because of their
flexibility. They can handle a large variety of feed materials, with variable
calorific value, and removal of waste solids at the exit presents no
problems. Typically, hazardous waste incinerators operate with relatively
deep beds, and have a secondary combustion chamber after the rotary kiln
to improve the heterogeneous combustion of waste (Rovaglio et. al. 1998).
An example of such a hazardous waste incinerator is shown in Fig. 2.
Chapter 1 – Introduction
2
Figure 2. A rotary
kiln waste incinerator
with a vertical
secondary combustion
chamber.
In the mineral processing industry, there are many applications of rotary
kilns: Some of the applications that have been published are magnetite
oxidation (Davis, 1996), induration of iron ore pellets (Young et. al. 1979),
coke calcining, (Bui et. al. 1995) and drying (Friedman & Marshall, 1949).
Rotary kilns are amongst the most wellestablished unit operations in the
process industry, yet are amongst the least understood. They can be used
for 3 purposes: heating, reacting and drying of solid material, and in many
cases, they are used to achieve a combination of these aims. In the design
of kilns, there are four important aspects to consider from a process
engineering point of view, and these are heat transfer, flow of material
through the rotary kiln, gassolid mass transfer and reaction.
Heat transfer rates amongst the most important of these aspects, because in
many cases, it is the heat transfer that limits the performance of the rotary
kiln. This is reflected in the number of papers published on the modelling
of heat transfer in rotary kilns. However, the first aspect of rotary kilns that
was modelled was the movement of material through a rotary kiln. The
first such modelling was done by Saeman (1951), based on data measured
by Sullivan et. al. (1927).
Chapter 1  Introduction
3
The subject of mass transfer in rotary kilns has been neglected until Jauhari
et. al. (1998) published an article on gassolid mass transfer in a rotating
drum. The purpose of this thesis is to model mass transfer and reaction in
rotating drums, but the novel approach adopted by the author is firmly
based on the foundations of chemical engineering science, can equally well
be applied to the modelling of heat transfer.
The radial plane vs axial direction
In the literature on rotary kilns, we typically find diagrams of a radial cross
section through a rotary kiln, such as that shown in Fig. 3. This shows the
movement of particles within the bed of a rotary kiln: particles move
concentrically with the wall of the rotary kiln in the passive layer until they
reach the surface of the bed, where they slide downwards in the active
layer. Considering a single particle, this repeated process can cause the
particle to move in the axial direction every time it moves down the active
layer, as long as the rotary kiln is tilted in the axial direction. This is the
basis for models that predict the movement of particles through a rotary
kiln in an axial direction. These models can predict the bed fill (fractional
area of the bed in a radial cross section) along the length of a rotary kiln as
a function of solids feed rate.
Chapter 1 – Introduction
4
Figure 3. Radial cross section of a rotary kiln
In operation, solids heat/cool/react/dry progressively as they move through
the kiln. In this thesis, we will refer to the operation as reaction. (This
approach encompasses drying as a subset of reaction.) Simultaneously
with the movement of the particles, gas may be flowing through the
freeboard of the kiln, either cocurrently or countercurrently. Therefore, at
any point along the length of a rotary kiln, we can expect a different set of
conditions to apply to both the solid and the gas. We will assume (as most
researchers in the field do) that if we can understand and model the
processes that occur in the radial crosssection, then we can use that radial
crosssection as a differential element to model the rotary kiln as a whole in
the axial direction using established engineering practices.
Chapter 1  Introduction
5
The flow regimes in rotary kilns
The nature of the solids flow in rotary kilns can change significantly with
particle characteristics, kiln size, rotation speed and bed fill. Heinen et. al.
(1983a,b) did useful work on this subject, introducing the concept of a Bed
Behaviour Diagram (Fig. 4).
Figure 4. An example of a bed behaviour diagram; each material has its
own characteristic bed behaviour diagram.
For rotary kilns with shallow beds in particular, the bed is prone to slippage
between the wall of the kiln and the bed. For deeper beds, the motion is
either rolling, as described in Fig. 3, or at lower rotation speeds, slumping.
In slumping mode, the angle of repose of the material increases as the static
bed rotates, until a segment of material shears off and comes to rest in the
bottom half of the kiln. This process occurs repeatedly, and is less
desirable than rolling mode for heat and mass transfer. In rolling mode, the
bed surface is continuously renewed, while in slumping mode, the renewal
occurs at fixed intervals.
Chapter 1 – Introduction
6
At higher bed fills and rotation speeds, the momentum of the particles as
they reach the surface of the bed becomes important, and they fall down the
surface of the bed in a trajectory. This causes a distortion in the surface of
the bed (which is normally flat). Further increasing the rotation speed
finally results in centrifugation, where particles are flung to the wall of the
kiln. This is also an undesirable regime of operation, because the particles
do not mix, resulting in poor mass and heat transfer.
Each of these modes of operation produces different conditions for particles
within the rotary kiln, and therefore should be modelled differently. In this
thesis, and in most literature articles, modelling is applicable to the rolling
mode – the preferred mode of operation industrially.
The need for the modelling of rotary kilns
The need for a deep understanding of kilns and the need for effective
fundamental models has been expressed very well by Barr et. al. (1989):
“Rotary kilns are ubiquitous fixtures of the metallurgical and chemical
process industries. Despite challenges from newer and more specialised
gassolids reactors, they continue to find applications in the drying, heating
(or cooling), calcining, reducing, roasting and sintering of a variety of
materials. Rotary kilns can handle feed stocks with broad particle size
distributions or whose physical properties change significantly during
processing, while the long residence time of the material within the kiln
promotes uniform product quality. In addition, dirty fuels often are utilised
without serious product contamination, and multiple fuel capability is
possible. Paradoxically, this versatility, which has in the past ensured the
survival of the rotary kiln, now threatens its future. Because a thorough
understanding of the processes occurring within rotary kilns has not been a
prerequisite for their apparently satisfactory operation, research has not
Chapter 1  Introduction
7
progressed apace with competing, less tolerant, reactors. Until all the
internal processes are understood and become predictable, rotary kilns will
remain in the position of operating below their optimal performance in an
increasingly sophisticated marketplace.”
An overview of this work
This work considers mass transfer in rotary kilns, by starting in Chapter 2
with a passive layer model.
In the passive layer model, we consider the movement of gas through that
part of the bed that moves as a concentrically around the axis of the kiln.
The gas is entrained in the interstices between the particles, reacting as it is
transported through the bed. The model was based on chemical reactor
theory, because mass transfer can be considered as a subset of reactor
theory – mass transfer becomes limiting for very fast reactions. In this first
model we show that the effectiveness factor of the bed is dependent on just
two variables: the ratio of the rate constant to the angular velocity (k/ω),
and the bed fill.
In the next chapter we consider passive layer diffusion, which involves
solving a partial differential equation in the two dimensions of the radial
cross section. This model is not flexible enough to be used over a wide
range of k/ω because of the need to consider the active layer. However, it
does provide useful insight into the dimensional groups that are required to
describe scaleup effects in a rotary kiln: either a Peclet number (εωR
2
/D
e
)
or a Thiele modulus (
e
D k R / ) should be used in addition to k/ω and bed
fill.
Chapter 1 – Introduction
8
In Chapter 4, some fundamental models for the active layer are presented: a
hydrodynamic model that predicts the depth and shape of the active layer,
and a dispersion model for predicting mass transfer within the active layer.
In the final chapter, a combined model is presented that expands on the
theory established in Chapter 4 for the active layer, and uses the principles
of Chapter 2 for modelling the passive layer. We find that it is seldom
possible to use either the passive layer model alone or active layer model
alone to accurately predict mass transfer, and that usually both regions in a
rotary kiln contribute significantly to mass transfer. We also find that when
dispersion effects dominate over molecular diffusion effects in the active
layer (expected for particle size > 3 mm), then scaleup effects can be be
represented by the ratio of kiln radius to particle diameter (R/d
p
) instead of
the Thiele modulus.
Chapter 1  Introduction
9
REFERENCES
Barr, P.V., Brimacombe, J.K., & Watkinson, A.P., (1989). ‘A heattransfer
model for the rotary kiln: Part 1. Pilot plant trials’, Metall. Trans., 20B,
1989, 391402
Boateng, A.A. & Barr, P.V., (1997), ‘Granular flow behaviour in the
transverse plane of a partially filled rotating cylinder, J. Fluid Mech., 330,
233249
Bui, R.T., Simard, G., Charette, A., Kocaefe, Y. and Perron, J., (1995),
Can. J. Chem. Eng., 73, 534545.
Davis, R.A., (1996), ‘Mathematical model of magnetite oxidation in a
rotary kiln furnace’, Can. J. Chem. Eng., 74, 10041009.
Friedman, S.J. & Marshall, J.R., (1949), ‘Studies in rotary drying’, Chem.
Eng. Prog., August 1949, 482493.
Heinen, H., Brimacombe, J.K., & Watkinson, A.P. (1983a). ‘Experimental
study of transverse bed motion in rotary kilns’, Metall. Trans., 14B, 1983,
191205
Heinen, H., Brimacombe, J.K., & Watkinson, A.P. (1983b). The modelling
of transverse solids motion in rotary kilns’, Metall. Trans., 14B, 1983, 207
220
Heydenrych, M.D., Greeff, P., Heesink, A.B.M., Versteeg, G.F., (2001),
‘Mass transfer in rolling rotary kilns: A novel approach’, Submitted to
Chem. Eng. Sci, 2001.
Chapter 1 – Introduction
10
Heydenrych, M.D., Schieke, J., Heesink, A.B.M., Kuipers, J.A.M. (2001),
‘Diffusion effects in rotating rotary kilns’, Accepted for publication at the
6
th
World Congress of Chemical Engineering, Melbourne, 2001.
Jauhari, R., Gray, M.R., & Masliya, J.H. (1988). Gassolid mass transfer in
a rotating drum, Can. J. Chem. Eng, 76, 224232.
Kramers, H., & Crookewit, P. (1952). The passage of granular solids
through inclined rotary kilns. Chemical Engineering Science, 1, 259
Rovaglio, M., Manca, D. & Biardi, G., (1998). ‘Dynamic modeling of
waste incineration plants with rotary kilns: comparisons between
experimental and simulation data’, Chem. Eng. Sci., 53(15) p 27272742
Saeman, WC, “Passage of solids through rotary kilns”, Chem. Eng. Prog.,
47, 1951, 508514.
Sullivan, J.D., Maier, C.G., & Ralston, O.C., (1927). Passage of solid
particles through rotary cylindrical kilns. US Bureau of Mines, Technical
Papers 384, 142
Tscheng, S.H. and Watkinson, A.P., (1979). Convective heat transfer in a
Rotary Kiln. Can. J. Chem. Eng., 57, 433443.
Young, R.W., Cross, M., and Gibson, R.D., (1979), ‘Mathematical model
of Gratekilncooler process used for induration of iron ore pellets’,
Ironmaking Steelmaking 1, 113.
Chapter 2 – Passive layer model
11
Chapter 2  Mass transfer in rolling rotary
kilns: a novel approach
ABSTRACT
A novel approach to modeling mass transfer in rotary kilns is explored. The
movement of gas in the interparticle voids in the bed of the kiln is
considered, where particles move concentrically with the geometry of the
kiln and gas is entrained by these particles. A reactor modeling approach
has been used to derive effectiveness factors for the bed as a function of
bed fill, reaction kinetics and rotation speed. In many cases, the entrained
gas becomes depleted within the bed, leading to a simplified model for the
bed effectiveness factor. Experimental data confirms the validity of this
model for slower rates. At faster rates, mass transfer can be much higher
than the model predicts, indicating that other mechanisms, such as
dispersion or diffusion are also important in these conditions.
1. INTRODUCTION
Rotary kilns are used industrially in many applications such as drying,
incineration, mixing, preheating, humidification, calcining, reducing,
sintering and gassolid reactions (Barr, 1989; Jauhari, 1998).
For many rotary kilns, heat transfer is the limiting factor, both in the
heating section of the rotary kiln, and in the reaction zone (Barr, 1989).
Consequently, most focus in the literature has been directed at
understanding the heat transfer processes in rotary kilns. It is important
Chapter 2 – Passive layer model
12
though to understand all of the processes that occur in rotary kilns on a
fundamental level before rotary kilns can be designed and operated
optimally.
With physical processes like drying and humidification for example, mass
transfer is also important, as well as in gassolid reacting systems with high
specific reaction rates such as incineration. In this work we will examine
the mass transfer in rolling rotary kilns and propose a novel approach to
describe the phenomena that determine the rate of mass transfer.
2. MODELLING STUDIES
The first published experimental studies on rotary kilns recorded the
relationship of rotation speed and kiln inclination on bed depth and solids
residence time (Sullivan et al, 1927). A model was later developed based
on the assumption that particles in a rolling bed move in a circular motion
with the rotation of the kiln, and then fall down the surface of the bed in a
thin layer (Saeman et al, 1951). The time taken to fall down the surface was
assumed to be small compared to the time for a particle to move with the
kiln from the bottom half to the top half of the bed. Using the geometry of
an inclined rotary kiln, the angle of inclination necessary to maintain a
constant bed height over the length of the rotary kiln could be determined
for a given rotation speed. This basic model predicted the original data
(Sulllivan et al, 1927) well, and the model was further refined to predict
axial movement of particles with different bed fills, taking into account the
time for particles to fall down the surface of the bed (Kramers and
Crookewit 1952). In later work that specifically measured the movement
of particles at the surface of the kiln, the validity of the model of Kramers
and Crookewit (1952), with minor exceptions, was confirmed (Lebas et al,
1995). More recently, the same fundamental model of the path of particles
Chapter 2 – Passive layer model
13
was shown to be correct in a study of particle motion where there is no nett
axial flow of particles in a rotating drum (Spurling et al, 2000).
The conditions that determine the mode in which a rotary kiln operates 
slipping, slumping, rolling, cascading, cataracting, centrifuging  have been
extensively examined, and can be represented using a Bed Behavior
Diagram (Heinen et al, 1983a). An example of a Bed Behavior Diagram is
shown in Fig. 1. In this paper, we will focus on kilns operating in the
preferred mode for most circumstances  rolling.
Figure 1. An example of a Bed Behavior Diagram (Heinen et al 1983a).
The time for a colored particle to move down the surface of the bed was
measured with video photography (Lebas et al, 1995), and this typically
takes 15  30% of a full cycle. However, this ratio does not change with
rotation speed, and the velocity of the particles moving down the surface of
the bed was interpreted to be directly proportional to rotation speed (Lebas
et al, 1995).
Chapter 2 – Passive layer model
14
The layer of particles moving down the face of the bed leads to improved
mass transfer, therefore the surface layer is often referred to as the “active
layer” (See Fig. 2). Heinen et al (1983a) measured particle velocities
inside this layer as well as the thickness of it, and found that it is relatively
thin for beds with a depth that is much larger than the diameter of the
particles. In most cases, the active layer is typically less than 8 particle
diameters deep, except for deep beds with small particles, where the depth
of the active layer is less than 12% of the bed depth, falling to as low as 4%
for slowly rotating kilns (Heinen et al, 1983a).
Figure 2. The various zones within a rotary kiln.
3. MODEL DEVELOPMENT
For most of the volume of the bed of a rotary kiln (the passive layer),
particles move concentrically with the wall of the bed, with no slippage
between particles. As the particles approach the bed surface due to the kiln
rotation, they slip down the face of the bed, and come to rest near the
Chapter 2 – Passive layer model
15
bottom of the surface of the bed. We will attempt to quantify the
volumetric flow rate and reaction rate of the gas that is trapped in the
interparticle voids in the passive layer.
3.1 Gas volumetric flow rate
The volume of entrained gas can be determined for a single rotation of the
kiln (for convenience, the bed has been represented horizontally):
Figure 3. Overall displacement of particles at the surface of a kiln
(irrespective of the path). Note that the bed surface will normally be
inclined, but for convenience has been drawn horizontally.
The differential area within ABC can be approximated by ½ base x
perpendicular height. The perpendicular height dx can be found as follows:
dx = (sin θ) R dθ = (c/R) R dθ = cdθ (1)
dA
x
= ½ c dx = ½ c
2
dθ (2)
Chapter 2 – Passive layer model
16
The area useful for the entrainment of gas is proportional to the void
fraction of gas in the bed:
( ) θ
ε
d h R dA
xg
2 2
2
− · (3)
Integrating this expression over 1 radian, and converting to angular
velocity, the volume flow rate of gas is:
Q = ½ εωL(R
2
– h
2
) (4)
This gives the volume flow rate of entrained gas through the bed of a rotary
kiln. By estimating the concentration of the gas as it leaves the bed, the
overall reaction rate can be determined.
To understand how the active layer influences the movement of gas,
consider a line extending from the radial centre of the kiln, passing
perpendicularly through the bed surface to the deepest part of the bed. All
entrained gas will pass this line. If the active layer is infinitely thin, the
flow rate Q can easily be shown to be as in Eq. (4). A mass balance of the
particles also means that the particles will move infinitely fast over the
surface of the bed with the assumption of an infinitely thin active layer. In
practice, the particles fill move down at a finite velocity, with a finite active
layer thickness. However, for the purposes of this paper, the admittedly
unrealistic assumption of an infinitely thin layer, not affecting the
entrainment of gas perpendicular to the bed surface, will be taken through
to its conclusion.
Chapter 2 – Passive layer model
17
3.2 Nature of concentration profiles within the bed
The movement of the gas in the voids is assumed to be a circular arc,
following the geometry of the kiln, until the gas reaches the surface of the
bed again (Saeman et al, 1951). For a halffilled kiln, the retention time of
such gas will be the same, regardless of the radius of the arc. Furthermore,
this retention time for the halffull kiln will be the time it takes for half of a
revolution: τ = π/ω. If we assume that the gas following this path will
move in plug flow, we can expect a concentration profile fitting an
exponential decay. Therefore, we can expect the concentration profile
within the bed of a rotary kiln to be as follows:
Figure. 4. For a halffilled kiln, the residence time at any radius is
equal, yielding concentration profiles shown here for a firstorder reaction
with k/ω = 1.
Diffusion and dispersion will no doubt play a role in the kiln, especially in
the centre of the kiln, where several concentrations converge. In this part,
the effect of the active layer will be important, and dispersion effects will
predominate. The inclusion of dispersion will require a more complex
model, not discussed here.
Chapter 2 – Passive layer model
18
For bed fills of less than 50%, the residence time at each radius will be
different, and the expected concentration profiles can be calculated.
Figure 5. Definition of angles for the determination of isoresidence time
lines.
( )
ω
θ
ω
θ θ
θ τ
r h
R
1
0
sin
) , (
−
−
·
−
· (5)
( )
ω
θ
τ
r h
1
sin
−
−
· (6)
) sin( ωτ θ − ·
r
h
(7)
ωτ ωτ sin cos
r
x
r
y
− · (8)
) cos(
) tan(
ωτ
ωτ
h
x y + · (9)
This applies to any reaction kinetics, and different reaction kinetics will
simply determine the spacing of the lines. As an example, let us consider a
1
st
order reaction.
Chapter 2 – Passive layer model
19
P
A
= εkC
A
(10)
where P
A
is the reaction rate of A per unit of bed volume and k is the
reaction rate constant with respect to the gas phase. Then
,
`
.

− ·
0
ln
A
A
C
C
kτ (11)
Substitute τ into Eq. (4) for 1
st
order isoconcentration lines, shown in
Fig. 6.
Figure 6. Concentration profiles for k/ω =1 at C
A
/C
A0
intervals of 0.1.
3.3 Effectiveness factors
For convenience, such concentration profiles can rather be expressed by an
effectiveness factor. This factor is defined as the ratio of the actual rate of
the reaction over the bed to the rate that would be obtained if the whole bed
were exposed to gas at the highest gas concentration  the concentration
above the bed.
Chapter 2 – Passive layer model
20
In the general case of kilns with a given fill, actual rate is the average rate
over the entire crosssectional area. (See Fig. 6.)
( ) ( )
dz d dr r e kC dV z r P
R
r
r h
k
A
V
bed A
]
]
]
]
·
∫ ∫ ∫
−
− −
2
1 1
1
sin
0
) , , (
θ
θ
θ
ω
θ ε θ (12)
( ) ( ) θ π θ θ sin sin sin
1
1
2
1
1
h r R h R h · − · ·
− −
(13)
From the definition of the effectiveness factor and realising that the cross
sectional area of the bed equals ½ ( ) ( )
1 2 1 2
2
sin θ θ θ θ − − − R ,
bed A
bed A
V kC
dV z r P
0
) , , (
ε
θ
η
∫
· (14)
( ) ( )
( ) ( )
∫ ∫
−
− −
− − −
·
2
1 1
1
sin
1 2 1 2
2
sin
2
θ
θ
θ
ω
θ
θ θ θ θ
η
R
r
r h
k
d dr r re
R
(15)
This function is represented graphically in Fig. 7 for different bed fillings
as a function of the k over ω ratio. It is interesting to note that the size of
the rotary kiln is irrelevant using this model. The only variables in this
model are the bed fill and k/ω. Later in the paper, we will discuss the role
of other parameters that are dependent on the scale of the rotary kiln.
Chapter 2 – Passive layer model
21
Figure 7. Effectiveness factors increase as bed fill decreases, and are
independent of bed radius R. Fill fractions shown are 0.5 (solid line), 0.37,
0.25, 0.14, 0.052, 0.019, 0.007 and 0.0017 (thin line).
As expected, for a given k/ω, the effectiveness increases for shallower beds.
For the special case of a halffilled cylinder, h = 0:
∫ ∫
−
·
π
θ
ω
θ
π
η
0 0
2
2
d dr re
R
R k
(16)
θ
π
π
θ
ω
d e
R
R
k
∫
−
·
0
2
2
2
2
(17)
,
`
.

− ·
−
ω
π
π
ω
k
e
k
1 (18)
This is also easily derived from first principles. Taking a mole balance for
a given volume of bed, where C
A
is the concentration of gas leaving the
bed,
Chapter 2 – Passive layer model
22
P
A
dV
bed
= Q(C
A0
 C
A
) (19)
ηεkC
A0
V
bed
= Q(C
A0
 C
A
) (20)
,
`
.

− ·
0
1
A
A
bed
C
C
Q L kA ηε (21)
,
`
.

− ·
−
ω
π
εω π ηε
k
e
L R L R k
1
2 2
2 2
(22)
,
`
.

− ·
−
ω
π
π
ω
η
k
e
k
1 (23)
4. TOTAL DEPLETION MODEL
The double integral of equation (15) is cumbersome to work with; a
simplified approach for estimating η would be more useful. For high k/ω
values such a simple approach is quite possible.
For high k/ω, the gas leaving the system at the surface of the bed can be
assumed to be fully depleted (or saturated, in the case of drying). In this
case, we assume complete reaction and use Eq. (4) to estimate the volume
flow rate of entrained gas.
Choosing 95% conversion as the criterion, C
A
/C
A0
= 0.05.
3 ln
0
·
,
`
.

− ·
A
A
C
C
kτ (24)
Chapter 2 – Passive layer model
23
For a halffilled kiln, all paths through the kiln have the same residence
time, and τ = π/ω. For lesser fill fractions, we will choose ½ of the
maximum angle. The angle subtended by this path is:
( ) ( ) R h
1
cos 2
2
1
−
· θ (25)
( )
ω
ω θ τ
R h
1
cos
−
· · (26)
The criterion for using the simplified model is therefore
( ) R h
k
1
cos
3
−
>
ω
(27)
Using an effectiveness factor and assuming total conversion,
0 A bed A
QC V P · − (28)
( )
0
2 2
0
2
A bed A
LC h R V kC − ·
εω
ηε (29)
( )
2 2
2
h R
A
k
x
− ·
ω
η (30)
Chapter 2 – Passive layer model
24
Table 1. The error caused by using the simplified model at the
criterion values is less than 6% in all relevant cases.
h/R Fill
fraction
k/ω ω η η
(Eq.
15)
A
x
(m
2
, basis
R = 1m)
η η, total
conversion
model
(Eq. 30)
Error
(%)
0 0.5 1.91 0.164 1.571 0.1667 1.65
0.2 0.374 2.19 0.1849 1.173 0.1867 0.97
0.4 0.252 2.59 0.2006 0.793 0.2047 2.09
0.6 0.142 3.24 0.214 0.447 0.2211 3.33
0.8 0.052 4.66 0.2259 0.164 0.2361 4.56
0.9 0.0187 6.65 0.2313 0.0587 0.2432 5.16
0.95 0.00666 9.45 0.2338 0.0209 0.2466 5.45
0.98 0.00169 14.97 0.2354 0.00532 0.2487 5.63
5. MODEL VALIDATION
5.1 Literature data
Jauhari et al. (1998) measured mass transfer in a rotating drum by
measuring the evaporation rate of decane from impregnated alumina
particles. Most of the measurements were done using shallow beds with
baffles (flights) fitted to the rotary kiln. This gives rise to a geometry that
is not well described by the model in this paper. However, one set of data
was reported for a shallow rolling bed (4,3% fill), and only this set of data
will be used here.
Chapter 2 – Passive layer model
25
Mass transfer rates were measured for various kiln conditions, assuming an
ideally mixed gas phase. The results were expressed in the form of a
volumetric mass transfer coefficient k
s
A/V
bed
, defined as:
( )
bed out decane sat decane
bed
s
out decane
V C C
V
A k
QC
, , ,
− · (31)
We can make use of the k
s
A/V
bed
data by comparing those with the
equivalent data predicted by our model.
QC
decane,out
= ∫ P
decane
dV
bed
= ηεk(C
decane,sat
–C
decane,out
)V
bed
(32)
As mass transfer from the surface of the alumina particles is the rate
determining step (as long as the pores are sufficiently filled), the first order
reaction rate constant (with respect to the gas phase), follows from
ε
ε
ε
p
gs gs
d
k a k
k
) 1 ( 6 −
· · (33)
with k
gs
being the mass transfer coefficient at the surface of the alumina
particles. We should thus compare the k
s
A/V
bed
data of Jauhari et al with
the
p gs
d k / ) 1 ( 6 ε η − data that we can produce with our model using eq. (15)
or (30). Before doing so, we must find an appropriate value for k
gs
. For
that purpose, we use the findings of Sφrensen and Stewart (1974) who
studied mass transfer in packed beds. For a stagnant fluid they derived:
p
gs
d
D Sh
k
⋅
· with Sh = 3.8 (34)
Chapter 2 – Passive layer model
26
As the gas entrained into the bed of a kiln is regarded to have no slip
velocity with regard to the particles in the bed, and the bed porosity is
comparable to that of a packed bed, eq. (34) is assumed to be appropriate.
If we now apply the precise conditions at which Jauhari et al. have
produced their k
s
A/V
bed
data, we are able to calculate the corresponding
p gs
d k / ) 1 ( 6 ε η − data with our model. These conditions are summarised in
Table 2:
Table 2: Conditions used in our model.
Total volume : 20.3 liters
Bed volume: 0.875 liters
h/R: 0.824
R: 0.145 m
h: 0.1195 m
Crosssectional bed area (A
x
): 0.00284 m
2
Particle diameter (d
p
): 3 x 10
3
m
Binary diffusion coefficient of decane
in nitrogen at 20°C (D
ab
):
6 x 10
6
m
2
s
1
Interparticle voidage (ε): 0.45 (assumed)
Sherwood number (Sh): 3.8
Gassolid mass transfer coeff. (k
gs
):
7.6 x 10
3
m s
1
k
,
`
.

·
−
p
gs
d
k
ε
ε) 1 ( 6
18.6 s
1
Below we compare our results with the data measured by Jauhari et al:
Chapter 2 – Passive layer model
27
Table 3. Comparison of measured data and model predictions.
Rotation
speed N
(rpm)
Angular
velocity ω ω
(s
1
)
Measured
data k
s
A/V
bed
(s
1
)
k
(s
1
)
Total
depletion
model: η ηε εk
(s
1
)
0.29 0.031 0.121 612 0.0164
0.59 0.062 0.198 301 0.0329
1.23 0.128 0.411 144 0.0685
1.67 0.175 0.502 106 0.0933
1.97 0.206 0.604 90 0.1101
In all cases, k/ω is much higher than the value of the criterion given by
Eqn. (29) (i.e. 0.09), meaning that saturation is easily reached within the
bed, and that the mass transfer contribution by the convective model is
practically insensitive to the calculated rate constant k.
Fig. 8. Mass transfer data measured by Jauhari et al is significantly higher
than that predicted by our model. ÷÷ Present model (total depletion)
o Data for 0.043 fill (Jauhari et. al. 1998)
Chapter 2 – Passive layer model
28
5.2 Measured data
In order to test our model at conditions where kinetics are much slower, we
performed similar experiments to Jauhari et al (1998), but using larger
particles, allowing the evaporation to continue until the particles were dry.
In this way, mass transfer within the pores of the particles became the rate
limiting process in the final drying stage.
Figure 9. Flow diagram of experimental apparatus
The rotating drum used has a diameter of 0.15 m and a length of just
0.041 m (Fig. 9). It is made of steel with a polycarbonate observation
window. It was only operated in the rolling mode at two speeds: 0,052
radians/s and 0,266 radians/s. Air was introduced through a rotameter and
blown through a 1 mm jet directly away from the bed towards the circular
wall of the rotating drum, in order to ensure good mixing of the gas above
the bed and to prevent any bypassing of gas to the central exit. Like
Jauhari et. al. (1998), hydrocarbon concentration was measured
Chapter 2 – Passive layer model
29
continuously using a sensor. The sensor was calibrated so that the sensor
reading could be transformed to a linear hydrocarbon concentration
reading. Because the hydrocarbon concentration and air flow rate were
known at all times during a run, it was possible to calculate what fraction of
hydrocarbon liquid had evaporated at any time. The fraction of liquid
remaining in the particle (liquid loading, z) has been used as a parameter to
characterise the particle at a given set of conditions.
The experimental technique to measure mass transfer rates differs from that
used by Jauhari et. al. (1998), because the sensor was placed within the
rotating drum, near the gas exit. This allowed us to directly measure the
buildup of concentration within the freeboard gas after the gas flow had
been stopped.
For the experiments reported here, we used spherical tabalumina particles
of average diameter 16 mm t 1 mm soaked in 1pentanol. The 44 pellets
that were used gave an average fill fraction of 0.27. The particles were
wiped dry before being loaded in the rotating drum. During the
experiment, air flowed through the rotating drum. At suitable intervals, the
gas flow was stopped, and the rate of increase of hydrocarbon
concentration was measured until it came close to saturation. The
characteristic rate of the system (ηεk) was calculated from the exponential
rise in the hydrocarbon concentration as a function of time:
( )
dt
dC
V V C C k
f bed sat
· − ηε (35)
which can be integrated to give:
t k
V
V
C C
C C
f
bed
sat
sat
⋅ ·
,
`
.

−
−
− ηε
0
ln (36)
Chapter 2 – Passive layer model
30
C
sat
could be accurately estimated by visually noting how the end points of
the graph of
,
`
.

−
−
−
0
ln
C C
C C
sat
sat
vs t fitted the straight line formed by the other
points. An example of such a graph is given in Fig. 10.
0.2
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
2 0 2 4 6 8 10 12
time, min
Figure 10. Mass transfer from bed to freeboard was calculated from the
rate at which concentrations increased in the freeboard after gas flow was
stopped. (Here, liquid loading was 0.57 and ω = 0.266 s
1
)
Several such measurements were performed during a run (with liquid
loading of the particles decreasing continuously), using both fast and slow
rotation speeds. The saturation concentrations inferred with this technique
did not vary substantially during the run, as can sometimes happen with
small pores, or when desorption within the particles becomes rate limiting.
The saturation concentrations varied by less than 1% throughout a run.
We used a similar technique to model the characteristic rate of the system
(ηεk) as with Jauhari’s data, except that we now added a shrinking core
,
`
.

−
−
−
0
ln
C C
C C
sat
sat
Chapter 2 – Passive layer model
31
model to the resistance term. The term L is liquid loading, and ε
part
is the
intraparticle porosity:
]
]
]
]
,
`
.

− +
−
· 1
1 1 1 ) 1 ( 6
3
1 2 2
z
Sh d
D
k
part p
ε ε
ε
(37)
The effective dispersion coefficient within the particle was estimated as
D
part
2
ε , and the value of ε
part
(0.15) was inferred from the experimental data
taken at the slowest drying rates. The parameters used in the model are
given in Table 4.
Table 4. Parameters used for the modelling of our data.
Parameter Value
Binary diffusion coefficient,
pentanolair (D)
8.25 x 10
6
m
2
/s
Bed porosity (ε) 0.636
Particle porosity (ε
part
) 0.15
Particle diameter (d
p
) 0.016 m
Sherwood number (Sh) 3.8
The total depletion model was not applicable at these experimental
conditions, so η was estimated using Eq. (15). This enabled us to plot our
model ηεk values with those that we obtained by experiment:
Chapter 2 – Passive layer model
32
Figure 11. Characteristic rate is well predicted by the model for both fast
rotation ÷x÷ (0,266 s
1
), and slow rotation ¯ (0,052 s
1
)
The predicted mass transfer rates follow the trend of the experimental data
quite well. By using the model prediction of k at the various liquid
loadings (Eq. 37), it is possible to calculate the effectiveness factor η and
k/ω at each of the experimental data points. These points (Fig. 12) follow
the expected trend quite well, but show some scatter. The experimental
data for 0.6 < k/ω < 2 were measured with wet pellets, and the values
measured for 0.02 < k/ω < 0.1 using externally dry pellets are higher than
expected, possibly due to uneven drying on the surface of the pellets.
Chapter 2 – Passive layer model
33
Fig. 12. Effectiveness factors follow the predicted trend as a function
of k/ω. ÷÷ model, × ω = 0,266 s
1
, ¯ ω = 0,052 s
1
.
6. CONCLUSION
A novel model to predict mass transfer inside a rolling rotary kiln was
presented. This model only considers convection inside the voids between
the particles inside the bed, which are assumed to be filled with entrapped
gas from the freeboard.
The model was found to underestimate mass transfer rates for fast reacting
systems (k/ω > 1), showing that other mechanisms like dispersion and/or
surface phenomena are also important for such systems. Active layer
models are to be preferred in these cases. However, for slower reactions
(k/ω < 1) the presented model appears to perform much better. It then
provides a useful alternative to active layer models, which are clearly
inappropriate if the whole bed contributes to mass transfer. For more
general applicability, it would be useful to include diffusion/dispersion
effects, which will be done in a future paper.
Chapter 2 – Passive layer model
34
SYMBOLS
a pellet outer surface area per bed volume [m
1
]
A area [m
2
]
A
x
radial crosssectional area of the bed [m
2
]
C concentration [mol m
3
]
D binary diffusion coefficient [m
2
s
1
]
d
p
particle diameter [m]
F solid fill fraction: bed volume/kiln volume
g gravitational acceleration [m s
2
]
h perpendicular distance from radial center of the kiln to the bed
surface [m]
k reaction rate constant [s
1
]
k
gs
mass transfer coefficient on the surface of particles [m s
1
]
k
s
mass transfer coefficient of active layer [m s
1
]
L length of the rotary kiln in axial direction [m]
P
A
reaction rate of A [moles m
3
(bed) s
1
]
Q volumetric gas flow rate [m
3
s
1
]
R radius of rotary kiln [m]
r distance from radial center of bed to the surface of the bed at angle
2 [m]
s half of the chord length formed by the bed surface [m]
V volume [m
3
]
V
bed
volume of bed [m
3
]
V
f
volume of freeboard [m
3
]
z liquid loading in particles: volume fraction of pores filled by liquid
Chapter 2 – Passive layer model
35
Greek symbols
ε bed voidage [m
3
(gas) m
3
(bed)]
η effectiveness factor of reaction
τ residence time of gas in the bed [s
1
]
θ angle [rad]
φ gas flow rate [m
3
(gas) s
1
]
ω angular velocity [s
1
]
REFERENCES
Heinen, H., Brimacombe, J.K., & Watkinson, A.P. (1983a). Experimental
study of transverse bed motion in rotary kilns, Metall. Trans., 14B, 1983,
191205
Heinen, H., Brimacombe, J.K., & Watkinson, A.P. (1983b). The modelling
of transverse bed motion in rotary kilns”, Metall Trans., 14B, 1983, 207
220
Jauhari, R., Gray, M.R., & Masliya, J.H. (1988). Gassolid mass transfer in
a rotating drum, Can. J. Chem. Eng, 76, 224232.
Kramers, H., & Crookewit, P. (1952). The passage of granular solids
through inclined rotary kilns. Chemical Engineering Science, 1, 259
Lebas, E.F., Hanrot, F., Ablitzer, D. & Houzelot, J.L. (1995).
“Experimental study of residence time, particle movement and bed depth
profile in rotary kilns”, Can. J. Chem. Eng, 73, 173179.
Chapter 2 – Passive layer model
36
Saeman, WC, “Passage of solids through rotary kilns”, Chem. Eng. Prog.,
47, 1951, 508514.
Sørensen, JP, Stewart, WE, 1974, Effect of fluid dispersion coefficients on
particletofluid mass transfer coefficients in packed beds”, Chem. Eng. Sci,
33, 13741384
Spurling, R.J., Davidson, J.F., & Scott, D.M. (2000). The noflow problem
for granular material in rotating kilns and dish granulators. Chemical
Engineering Science, 55, 23032313
Sullivan, J.D., Maier, C.G., & Ralston, O.C., (1927). Passage of solid
particles through rotary cylindrical kilns. US Bureau of Mines, Technical
Papers 384, 142
Tscheng, S.H. and Watkinson, A.P., (1979). Convective heat transfer in a
Rotary Kiln. Can. J. Chem. Eng., 57, 433443.
Chapter 3 – Diffusion in the passive layer
37
Chapter 3  Diffusion effects in rotating
rotary kilns
ABSTRACT
A novel approach to the modeling of mass transfer in rotary kilns has been
described (Heydenrych et al, 2001). It considers the mass transfer to occur
by the inclusion of gas in the interparticle voids in between the particles
that move concentrically with the kiln. By doing so, the rate of mass
transfer was found to be dependent on bed fill and the ratio of reaction rate
constant to angular velocity (k/ω). The model was found to be valid at
slow to medium fast reactions. For fast reactions it underpredicted mass
transfer. Therefore in this paper, the model will be extended to include
diffusion effects. An additional dimensionless number is necessary then to
describe the system. This can either be a Peclet number (ωR
2
/D
e
) or a
Thiele modulus (kR
2
/D
e
)
½
. The solution of the 2dimensional partial
differential equations that describe the extended model gives a handle on
the effect of scaleup in rotary kilns. For industrialscale kilns, the Peclet
number is large, which means that diffusion within the lower (passive)
layer of the bed is unimportant for slower rates. With high reaction rates,
isoconcentration lines are closely stacked near the surface of the bed,
implying that it is important to model the active layer rather than the bed as
a whole in these circumstances. However, the stiff differential equations
are not easily solved then, and other methods of solution are advisable.
Chapter 3 – Diffusion in the passive layer
38
1. INTRODUCTION
Traditionally, mass transfer in rotary kilns has been modelled as a mass
transfer coefficient at the surface of the bed of a rotary kiln (Jauhari et. al.
1998). Recently, a novel approach to the modelling of mass transfer has
been suggested though (Heydenrych et al, 2001). This approach takes into
account the inclusion of gas in the interparticle voids in between the
concentrically moving particles in the passive layer of the bed. In that
work, the role of the active layer where particles move along the surface of
the bed was ignored by assuming that the thickness of the active layer
region is infinitely thin. They concluded that their concentric movement
model has merit, but that the effect of diffusion and dispersion within the
bed must also be taken into account in the case of fast reactions. In this
paper we will explore the effect of diffusion within the concentric flow
region, and how it affects the overall mass transfer, again under the
assumption of an infinitely thin active layer.
Figure 1. Shows the concentric movement and active layer regions
Chapter 3 – Diffusion in the passive layer
39
2. MATHEMATICAL MODEL
To take diffusion into account, consider the mass balance for a differential
element within the bed (i.e. concentric flow region):
A
A
r
t
C
+ ⋅ −∇ ·
∂
∂
A
N (1)
After substituting equations for the transport of gaseous A (Fick’s diffusion
plus convection) into the above equation, the concentration profile of A
follows from
( )
A A A e
A
r c C D
t
C
+ − ∇ ⋅ ∇ ·
∂
∂
V ε (2)
A A e A
A
r C D C
t
C
+ ∇ · ∇ ⋅ +
∂
∂
2
V ε (3)
Sharp concentration gradients are expected in the radial crosssection of
kilns [Heydenrych et al, 2000]. In the axial direction, relatively low
concentration gradients can be expected due to the high length to diameter
ratio found in typical rotary kilns. This makes it a reasonable assumption
to simplify the governing equation to the two dimensions that represent the
radial crosssection. Although polar coordinates may be a more natural
choice of coordinate system, we have chosen rectangular coordinates
because of software restrictions. Assuming steadystate operation:
0
2
2
2
2
· +
∂
∂
+
∂
∂
+
∂
∂
+
∂
∂
A
A
e
A
e
A
y
A
x
r
y
C
D
x
C
D
y
C
V
x
C
V ε ε (4)
Chapter 3 – Diffusion in the passive layer
40
Here, D
e
is the effective diffusivity and e is interparticle voidage. Both are
assumed to be constant. V
x
and V
y
are the x and y components respectively
of the velocity of particles (and gas) in the bed, and are described by the
concentric motion of the bed. Zero flux at the solid walls gives a simple
boundary condition: 0 · ∇ ⋅
A
C n
v
where n
v
denotes the normal vector at the
kiln wall. For the bed surface, various boundary conditions can be used as
appropriate, discussed later in the paper.
Rewriting Eqn. 4 in terms of dimensionless variables, for an n
th
order
reaction with
n
A A
kC r ε · − :
0
1 1
1
,
2
2
2
2
· Ψ +
∂
Ψ ∂
+
∂
Ψ ∂
+
∂
Ψ ∂
+
∂
Ψ ∂
−
n
n
s A
y
y
x
x
y x
kC
Pe Pe ω
ε
λ
υ
λ
υ
λ λ
(5)
where
s A
A
C
C
,
· Ψ dimensionless concentration,
R
x
x
· λ ,
R
y
y
· λ ,
R
V
x
x
ω
υ · ,
R
V
y
y
ω
υ · ,
e
D
R
Pe
2
ω ε
· , a Peclet number modified for rotary kilns. The
reaction term
ω
1
,
− n
s A
kC
is similar to the k/ω term used by Heydenrych et. al.
(2001) to describe concentration terms within the bed of a rotary kiln in
which a first order reaction takes place.
Noting that υ
x
is simply the xcomponent of the velocity divided by the
velocity at the wall, it follows from the concentric movement geometry that
y x
λ υ · and
x y
λ υ − · using the coordinate system given in Fig. 2. See the
Appendix for more detail.
Chapter 3 – Diffusion in the passive layer
41
Figure 2: Applied coordinate system. We assumed that the active layer
was infinitely thin.
In addition to the noflux condition at the wall, the Danckwerts boundary
condition for nonideal flow in a plug flow reactor was used as the
boundary condition at the bed surface:
,
`
.

∂
∂
+
∂
∂
− ·
y
C
x
C
V
D
C C
A A
x
e
A s A
ε
,
(6)
The condition may also be written in dimensionless form as
x x
y x
Pe Pe υ ψ υ
λ
ψ
λ
ψ
− · −
∂
∂
+
∂
∂
where
ω
υ
R
V
x
x
· . (7)
The noflux condition at the wall is given in dimensionless form by
0
1
·
∂
Ψ ∂
+
∂
Ψ ∂
· λ
λ λ
y
y
x
x
n n (8)
Chapter 3 – Diffusion in the passive layer
42
where n
x
and n
y
are the components normal to the wall confining the
particulate solids.
The model described here (Eqns. 5, 7 and 8) is dependent on two
dimensionless variables: Pe and
ω
1
,
− n
s A
kC
. Note that the square of the Thiele
modulus
ω
φ
1
, 2
−
⋅ ·
n
s A
kC
Pe . This means that the system can be described by
any two of the dimensionless groups Pe, φ or
ω
1
,
− n
s A
kC
. In this paper we have
chosen to use
ω
1
,
− n
s A
kC
because it is the only variable that is required to
describe the nodiffusion case, and Pe. The rate constant k typically
changes significantly along the length of the kiln, as the solid phase
approaches full conversion (or total drying). Pe only needs to be calculated
once for a given kiln because it does not typically change along the length
of a rotary kiln.
Fraction fill is the third independent dimensionless variable that is
necessary to describe this model. It is also interesting to note that the
fraction fill does not affect the governing equations (Eqn. 5), but only the
geometry of the boundary conditions.
Solution of the model
This model was solved using Matlab’s partial differential equation toolbox.
Although it would be possible to model n
th
order reactions using the same
toolbox, in this case, we have restricted our analysis to 1
st
order reactions.
For future reference, we will refer to the second dimensionless variable in
its form for firstorder reaction, k/ω.
Chapter 3 – Diffusion in the passive layer
43
The solution gives the concentration profile within the bed of the kiln, and
the effectiveness factor of the bed is calculated from that concentration
profile. The effectiveness factor of a rotary kiln is defined analogously to
the effectiveness factor of a catalyst pellet.
bed
A
n
bed
n
S A
A
n
A
n
S A
V
n
A
V
S A
V
A
A
dA
A C
dA C
V kC
dV kC
dV P
dV P
∫ ∫ ∫
∫
∫
Ψ
· ·
−
−
·
−
−
·
, ,
,
ε
ε
η (6)
where –P
A
is the local reaction rate in the bed and –P
A,s
is the reaction rate
that would occur if the entire cross section of the bed was exposed to the
conditions at the surface of the bed.
For 1
st
order reactions,
bed
A
A
dA
∫
·
ψ
η ,
where area A is calculated in terms of λ
x
and λ
y
. (7)
Given that the solution of the partial differential equation (Eq. 5) is the
concentration profile over the bed, the effectiveness factor can be
determined from the average dimensionless concentration over the bed
cross section.
3. RESULTS
The effectiveness factor curves for a 50%, 20% and 10% filled kiln that
were generated using the model are shown in figures 3 to 6. The curve for
an infinite Peclet number represents the nodiffusion case reported earlier
by Heydenrych et. al. (2001).
Chapter 3 – Diffusion in the passive layer
44
Figure 3: Effectiveness factors as a function of k/ω for a 50% filled kiln;
1
st
order reaction.
As expected, the inclusion of diffusion effects in the model increases the
effectiveness factor in all conditions. It should also be noted that for a
given Peclet number (i.e. kiln size and rotation speed), the effect of
diffusion as a fraction of the contribution by gas inclusion, is far higher at
high values of k/ω, where high concentration gradients occur in the bed
(predominantly near the surface of the bed). For commercial scale rotary
kilns, Peclet numbers are typically greater than 1000, discussed later in the
paper.
At higher Peclet numbers (Pe > 10), the numerical method was unable to
converge on a solution. The dashed lines on Figure 3 are extrapolations,
where the model failed to solve. A possible reason for the poor
convergence of the model at large Peclet numbers, is that the numerical
methods used to solve the model were based on techniques suited for an
elliptical (secondorder) partial differential equation, with the firstorder
term simply being handled as a source term. At high Peclet numbers, the
secondorder terms of the partial differential equation become less
important, and the numerical method is unsuited to the solution of what
effectively becomes a firstorder hyperbolic pde that dominates the model.
Chapter 3 – Diffusion in the passive layer
45
Similar solutions have been found for kilns with bed fills of 20% and 10%
(Figs. 4 and 5 respectively)
Figure 4. Effectiveness factors as a function of k/ω for a 20% filled kiln;
1
st
order reaction
.
Figure 5: Effectiveness factors as a function of k/ω for a 10% filled kiln; 1
st
order reaction.
As the fill fraction decreases, the effect of diffusion appears to become
more important. The concentration profiles generated at certain Pe and k/ω
values are interesting, and are shown in Fig. 6 for a halffilled kiln:
Chapter 3 – Diffusion in the passive layer
46
P
e
=
1
Fill = 10% Fill = 20% Fill = 50%
k
/
ω
=
0
.
1
k
/
ω
=
1
k
/
ω
=
1
0
k
/
ω
=
1
0
0
Figure 6: Concentration profiles at Pe = 1; 1
st
order reaction.
The straight isoconcentration lines predicted for the nodiffusion case
(Heydenrych et. al, 2001), Fig. 7, are only readily evident for the diagram
Chapter 3 – Diffusion in the passive layer
47
where k/ω = 0.1. For the other cases, diffusion effects dominate the shape
of the concentration profile, and the rotation of the kiln does not distort the
profile significantly.
Figure 7. If diffusion is not taken into account, straight isoconcentration
lines are expected.
4. DISCUSSION
Scaleup effects
The models have shown that for Peclet numbers less than 100, diffusion
effects are important. But under what circumstances can we expect such
low Peclet numbers, and what Peclet numbers are typical for rotary kilns?
The diffusivities of a gas at atmospheric pressure vary between 10
4
10
6
m
2
/s (Knudsen et. al., 1997). Values of 2 x 10
4
m
2
/s, 2 x 10
5
m
2
/s and 2 x
10
6
m
2
/s were used to for the diffusivity of the reactant gas D
A
.
The effective diffusivity in the bed of the kiln can be estimated as follows:
τ
ε
A
e
D
D · (Knudsen et. al., 1997)
Chapter 3 – Diffusion in the passive layer
48
Using a tortuosity (τ) of 2 and voidage (ε) of 0.5, we can expect effective
diffusivities within the bed of 5 x 10
5
m
2
/s (max.), 5 x 10
6
m
2
/s and
5 x 10
7
m
2
/s (min.).
Kiln rotation speeds vary less, because at lower rotation speeds, slumping
of the bed can occur. At higher rotation speeds, centrifugation begins. For
this reason, only two angular velocities are considered, 0,5 rad/s and
0,1 rad/s.
The effect of scale is largely dependent on the radius, because Pe ∝ R
2
.
Table 1 shows Peclet numbers for selected D
e
and ω values as a function of
radius.
Table 1. Peclet numbers for typical rotation speeds and effective diffusion
coefficients. Cells with Pe < 1000 are not shaded to indicate that this only
occurs in laboratoryscale kilns with high D
e
values.
D
e
, m
2
/s 5 x 10
5
5 x 10
6
5 x 10
7
ω, rad/s 0.1 0.5 0.1 0.5 0.1 0.5
0.05 5 25 50 250 500 2500
0.1 20 100 200 1000 2000 10000
0.5 500 2500 5000 25000 50000 250000
1 2000 10000 20000 100000 200000 1000000
r
a
d
i
u
s
(
m
)
5 50000 250000 500000 2500000 5000000 25000000
As an example, the Peclet numbers for the rotating drum used by Jauhari et.
al. (1998), with a rotating drum of radius 0.145 m, varies from 530 to 3600,
depending on rotation speed. It is obvious that diffusion as a possible extra
transport mechanism to gas entrapment can safely be neglected for practical
applications as well as for most lab experiments.
Chapter 3 – Diffusion in the passive layer
49
5. CONCLUSIONS
A mathematical model has been presented that improves on the previous
model (Heydenrych et al, 2001) by adding diffusion effects to gas
entrapment in the concentric flow region of a rotary kiln. The technique,
but not the model, is limited to moderate Pe numbers. For practical values
of Pe, a modified solution technique is required. Heydenrych et al, 2001
have shown that k/ω is the parameter that describes mass transfer at low to
moderate reaction rates. In this paper, Peclet number was implemented (in
addition to k/ω) to model diffusion effects. This number is defined as
ωR
2
/D
e
. The Peclet number provides a tool for predicting scaleup effects,
and even though the model presented here is oversimplified due to the
assumption of an infinitely thin active layer, it nevertheless provides an
insight to the use of the Peclet number for the scaleup of rotary kilns.
Diffusion in the concentric flow region was found not to add much to the
mass transfer at realistic conditions where the Peclet number is much
higher than 1. At high k/ω values, however, concentration gradients
become fully concentrated near the surface of the bed, and here the
assumption that the active layer is infinitely thin becomes a poor
assumption as all mass transfer is occurring within this layer. This model
shows that the entrapment of gas perpendicular to the bed is relatively
unimportant compared to diffusion effects at these higher k/ω values.
Further work should be concentrated at developing a suitable model for the
active layer, where a 1dimensional dispersion model would be a fair
assumption, based on the results of this work. It will also be important to
determine the depth of the active layer, and how the movement of particles
over each other affects the dispersion of the gas in the direction
perpendicular to the bed surface. This information will give the boundary
Chapter 3 – Diffusion in the passive layer
50
conditions for the concentric flow region, with the aim of generating a
comprehensive model including all mechanisms for mass transfer.
SYMBOLS
A Cross sectional area of the bed of the rotary kiln [m
2
]
C
A
Concentration of reactant A in the bed of the rotary kiln [kmol/m
3
]
C
A,s
Concentration of reactant A at the surface of the bed [kmol/m
3
]
D
e
Effective diffusivity of reactant A [m
3
/s]
Pe Peclet number modified to suit a rotary kiln
R Radius of kiln [m]
r Reaction rate of a reactant A [kmol/m
3
/s]
V Velocity of gas within the bed [m/s]
x Coordinate in x direction
y Coordinate in y direction
Greek Letters
ε Voidage of the bed of the rotary kiln
φ Thiele modulus
η Effectiveness factor
λ Dimensionless distance
τ Tortuosity
υ Dimensionless velocity
ω Angular velocity of kiln [rad/s]
ψ Dimensionless concentration
Chapter 3 – Diffusion in the passive layer
51
REFERENCES
Heydenrych, M.D., Greeff, P., Heesink, A.B.M., Versteeg, G.F., (2001),
‘Mass transfer in rolling rotary kilns: A novel approach’, Submitted to
Chem. Eng. Sci, 2001.
Jauhari, R., Gray, M.R., & Masliya, J.H. (1988). Gassolid mass transfer in
a rotating drum, Can. J. Chem. Eng, 76, 224232.
Knudsen J.G., Hottel, H.C., Sarofim, A.F., Wankat, P.C. and Knaebel, K.S.
(1997) in Perry’s Chemical Engineer’s Handbook, Seventh Edition, D.W.
Green (Ed.), McGrawHill, New York, 51 to 579.
Thiele, E.W. (1939) ‘Relation between catalytic activity and size of
particle’, Industrial and Engineering Chemistry, 31 (7), 916920
Chapter 3 – Diffusion in the passive layer
52
APPENDIX
Proof that υ
x
= λ
y
and υ
y
= λ
x
Consider a point on the radius, which is at λ = 1 in dimensionless
coordinates. By definition,
R
V
x
x
ω
υ ·
Substituting for V
x
,
θ
ω
θ ω
υ sin
sin
· ·
R
R
x
From the diagram,
x y
υ θ λ · · sin
At points away from the wall, υ
x
and λ
y
decrease proportionately. It can
similarly be shown that υ
y
= λ
x
.
Chapter 4 – Active layer models
53
Chapter 4  Hydrodynamics and dispersion
in the active layer of rolling rotary kilns
ABSTRACT
In previous works (Heydenrych et. al. 2001a,b), it was shown that the top
layer (or active layer) of a rolling bed is the only active place in the case of
a fast reaction. The bottom part of the bed (also referred to as the
concentric flow region) is almost depleted and therefore hardly reacting.
Therefore in this work we focus on that active layer. A hydrodynamic
model is prepared to predict its shape. A mass balance on particles moving
within a rotary kiln gives the shape of the active layer, defined as that part
of the bed where particles move past each other. The derivation is based on
the assumption that there is a linear velocity gradient of particles in the
direction perpendicular to the surface of the bed that is constant at any
place on the bed surface. Published data was used to establish a
relationship between this velocity gradient on the one hand, and bed fill and
rotation speed on the other. A dispersion model was derived to predict the
mass transfer of gaseous reactant within the active layer. As void spaces
are created and destroyed due to the action of particles moving over each
other, mixing of gas occurs. A dispersion coefficient was derived, based on
this gas mixing, and was found to be proportional to the velocity gradient
and to the square of particle diameter. Applying these relationships to a
1dimensional concentration gradient model gave a correlation for the mass
transfer coefficient at the bed surface. This model does not correctly
predict the proportionality of mass transfer with bed speed reported in the
literature. We conclude that another effect, probably the granular
temperature of the particles, is important for modeling of dispersion and
mass transfer in the active layer of rolling rotary kilns.
Chapter 4 – Active layer models
54
1. INTRODUCTION
The movement of particles in a rotary kiln is shown in Figure 1. Particles
move concentrically with the wall of the kiln (passive layer) until they
reach a point near the surface, at which point the particles slide down the
top surface of the bed. This differentiation of the two zones within a rotary
kiln has been recognized from the earliest published data on rotary kilns
(Sullivan et. al. 1927, Saeman, 1951).
Fig. 1. Particles slide over each other to the lower half of the bed in the
active layer.
This motion of particles relative to each other leads to much improved heat
and mass transfer rates compared to what one would expect from a
stationary bed (Heinen et. al. 1983).
Traditionally, heat and mass transfer rates have been modeled using a
coefficient based on the surface area of the bed (Tscheng & Watkinson,
1979). Recently, Heydenrych et. al. (2001) have shown that in certain
Chapter 4 – Active layer models
55
cases, particularly for slow processes, the active layer, and the volume of
the bed can also become important. Those models were developed for
mass transfer, and in this paper we will continue to model mass transfer as
an example of transfer processes that occur in rotary kilns.
In this paper, we explore the physical processes that occur within the active
layer in order to model the depth and shape of the active layer, and to
predict the dispersion that can be expected. With these models, we
estimate concentration profiles and finally calculate the effective mass
transfer coefficient (based on bed surface area) for the active layer for a
given rotation speed, particle size and bed depth.
2. HYDRODYNAMICS OF THE ACTIVE
LAYER
2.1 The shape of the active layer
The movement of particles sliding over each other in the active layer causes
dispersion of the gas in the voids. This dispersion effect is typically much
greater than the effect of gaseous diffusion in the active layer, and
concentration gradients in this layer are much less than one would predict
by the diffusion model presented by Heydenrych et. al. (2001).
The most important parameter for predicting the degree of dispersion will
be the gradient of the velocity profile near the surface of the bed.
If we take any line from a point on the bed surface to a point on the wall
within the bed, the mass on either side of that line must remain constant in
Chapter 4 – Active layer models
56
time. We will use a line extending from 1 to 2 to 3 on Fig. 1; we expect the
mass within the volume 1, 2, 3, 4 to remain constant.
Figure 1. Mass balance over envelope described by 1234
The volume flow rate out of the control volume (using solids in this case)
has earlier been shown (Heydenrych et. al., 2001) to be:
( )
2 2 2
) (
2
) 1 (
d h x R
L
Q + − −
−
·
ε ω
( 1)
Boateng & Barr (1997) have measured the velocity profiles in the active
layers of rotary kilns, and found that for certain materials, the gradient of
the velocity profile is substantially linear in the direction perpendicular to
the bed surface (12) on Fig. (1). They also found that this gradient does
not change along the xdirection in Fig. 1. We will make this assumption
to calculate the volume flow rate into the control volume through the
envelope extending from 1 to 2 in Fig. 1. It would be useful to understand
through granular dynamics simulations (Hoomans, 2000) how the choice of
materials affects these assumptions.
Chapter 4 – Active layer models
57
Firstly, let us define the gradient of the velocity profile (called shear rate by
Boateng and Barr (1997)) as the variable m. Also, defining the velocity of
particles at the surface and at depth d as υ
s
and υ
d
respectively, it follows
that
dm
d S
· −υ υ ( 2)
and assuming that υ
d
is determined by the xcomponent velocity of the
passive layer,
ω θ ω υ ) ( sin d h r
d
+ − · − · ( 3)
Substituting for υ
d
into Eq. (2),
ω υ ) ( d h dm
S
+ − · ( 4)
The volume flow rate of solids between points 1 and 2 can now be
calculated, using the average velocity between these two points:
2
) ( ) 1 (
d S
Ld
Q
υ υ ε + −
· ( 5)
Eliminating Q using Eqns. (1) and (5), we get:
( )
2 2 2
) ( d h x R
d
d S
+ − − · +
ω
υ υ ( 6)
Chapter 4 – Active layer models
58
Substituting Eq. (3) for υ
d
and Eq. (4) for υ
S
and simplifying,
1 /
2 2 2
−
− −
·
ω m
h x R
d ( 7)
Note that with d and m/ω known, υ
S
can be determined as a function of x
from Eq. (4).
The m/ω term is interesting. It means that m has to be greater than ω to
have any physical significance. If we choose a linear relationship for m =
bω, (b is a constant), it means that b must be greater than 1. Literature data
suggests b is about 15. Also, if (as the data suggests), m = bω+ c, then the
only requirement is that c>ω, or b>1. In reality, this constraint is easily
met.
A constraint that is not as easily met is the constraint that the active layer
depth d must be less than the bed depth H, where H = R  h. This constraint
can be represented by
d(x=0) < R – h ( 8)
Substituting Eqn. (6) for maximum active layer thickness at x = 0, we can
show that this constraint simplifies to
H
R m 2
>
ω
( 9)
The relationship between m and ω has been measured by Boateng & Barr
(1997). They measured particle velocities at various depths using an
optical probe. Their measurements indicate that for a given material and
Chapter 4 – Active layer models
59
bed depth, m increases linearly with ω. Their data for beds consisting of
polyethylene beads and limestone are shown in Fig. 2. The data shows
some scatter, and simulation using granular dynamics (Hoomans, 2000)
might shed some light on where the true relationships lie. We will look for
a relationship describing m not just as a function of rotation speed ω, but
also as a function of bed fill, based on the experimental data.
Figure 2. The linear relationship m = 19ω + 7.5 is shown by the solid line.
Data points marked PE represent polyethylene beads, and LS represents
limestone.
Noting that the constraint for the value of m imposed by Eqn. (9) also
follows a linear relationship with ω, it would be interesting to see if m also
follows a linear relationship against R/H. We have plotted the data of
Boateng & Barr (1997) against R/H in Fig. 3
Chapter 4 – Active layer models
60
Figure 3. Shear stress m shown as a function of R/H for polyethylene
particles.  ¯  5 rpm ««««r«««« 3 rpm ÷ o ÷ 2 rpm
The data appears to correlate reasonably well with R/H, indicating that the
shear stress is indeed dependent on R/H. In fact for shallower beds (high
R/H), there appears to be a linear relationship, and the slope of the lines
correlates well with the slope of 2ω suggested by the criterion of Eqn. (9).
In fact, a suggested relationship for small bed depths
c
H
R
m + · ω 2 ( 10)
is shown on Fig.3, with c = 9.5 s
1
. This model fits the data for R/H > 3
very well. The intercept is apparently independent of rotation speed, and
represents the shear stress that will result from particles sliding down the
surface of an inclined chute with an infinitely deep bed, and will depend on
particle characteristics only. In this case, one can expect a layer of finite
thickness, and constant m as flow increases – i.e. bed depth proportional to
flow rate of particles.
Chapter 4 – Active layer models
61
Noting from the data of Boateng & Barr (1997) that a typical value for
shear stress m/ω is 10, we conclude that the constraint (Eqn. 9) will become
important when R/H < 5. In terms of bed fill, it means that for bed fill less
than 5.2%, m must be larger than the narrow range expected from the data
shown in Fig. 2.
A further implication of the bed height expression, Eqn. (7), is that
ω
ω
−
∝
m
d . Substituting for m using Eqn. (10),
( ) 5 . 9 1 / 2 + − ∝ ω ω H R d . This means that at low rotation speeds,
d ∝ ω
0.5
. As rotation speed increases, d becomes less dependent on rotation
speed.
Our model does not predict the shear stress well at lower R/H values (fuller
beds). In this case, geometric constraints and perhaps centrifugal forces
may affect the shear stress profile.
Substituting Eqn. (10) into Eqn. (7), we can determine the shape of the
active layer at any point in the active layer, at any given kiln rotation speed.
The active layer profile predicted by this model for various bed depths and
rotation speeds is shown in Fig. 4.
Chapter 4 – Active layer models
62
Figure 4. Active layer profile is an elliptical shape, shown here for H/R
values of 0.2, 0.5 and 0.8. The depth of the active layer does not increase
linearly with rotation speed (ω=0,1 s
1
dotted and ω=0,2 s
1
dash shown).
Heinen et. al. (1983) have published a good deal of data showing observed
active layer thickness as a function of bed depth and rotation speed. Fig. 5
shows how the measured data at the deepest part of the active layer
compares with our model for predicting bed depth (at x = 0).
Chapter 4 – Active layer models
63
Figure 5. Our model predicts a deeper active layer than visually measured
data at x = 0, probably due to wall effects that were confirmed by Boateng
& Barr (1997).
Some of the scatter in the data can be attributed to the fact that Heinen et.
al. (1983) reported rotation speeds within limits, rather than the exact
values. We used the average values within those limits.
2.2 The volume of the active layer
In some cases, it is useful to know the volume of the active layer, and
Jauhari et. al. (1998) presented a model for predicting mass transfer rates
based on the volume of the active layer. To calculate the volume of the
active layer, it is useful to show that, using our model, the shape of the
active layer is described by an ellipse with the central axis located on the
bed surface.
Chapter 4 – Active layer models
64
An ellipse (Fig. 6) is described by
1
) (
2
2
2
2
· +
−
a
x
b
c y
( 11)
Figure 6. Definition of parameters describing an ellipse.
We would like to rewrite the expression for active layer depth in the form
shown in Eqn. (11).
2 2 2 2
1 h x R d
m
− − ·
,
`
.

−
ω
( 12)
Substitute for d
2
using the relationship y =  d – h
2 2 2
2
1
) (
h R x
m
h y
− · +
−
+
ω
( 13)
( )( )
1
1
) (
2 2
2
2 2
2
·
−
+
− −
+
h R
x
h R m
h y
ω
( 14)
Chapter 4 – Active layer models
65
Writing in terms of the dimensions of Fig (6),
2 2
h R a − ·
1
2 2
− − · ω m h R b
h c − ·
Because the axis of the ellipse lies along the surface of the bed, the area of
the active layer is half of the area of the ellipse. Therefore the volume of
the active layer is
( )
1 2
2
2 2
−
−
· ⋅ ·
ω
π π
m
h R L
L ab V
al
( 15)
Consider the points on the ellipse where y = h. The slope of the tangent =
∞ for all h. Because the slope of the wall is finite for h < 0, it implies that
there is always a point where the active layer theoretically extends beyond
the bed (for h < 0, all m/ω). For practical purposes, this is usually
insignificant, and should not detract from the usefulness of the model.
3. DISPERSION WITHIN THE ACTIVE
LAYER
As particles slide over each other, the size of interparticle voids changes.
This gas moves to adjoining voids, where it mixes with gas at another
concentration, assuming that a concentration gradient exists. Consider
Figure 7, where successive layers of particles are separated by a distance
Chapter 4 – Active layer models
66
approximately equal to the particle diameter d
p
. In a full cycle, the top
layer of particles moves horizontally by a distance 2d
p
relative to the
bottom layer. Noting that the top and bottom layers are separated by a
vertical distance of 2d
p
, the time it takes for this cycle will be 1/m seconds
(where m is the velocity gradient, in m/s per m active layer depth).
Start of cycle ½cycle full cycle
Figure 7. Description of voids in one cycle of particles sliding over each
other.
Assume also that a steadystate concentration gradient exists with molar
flux W
A
, where we need to determine dispersion coefficient D,
dy
dC
D W
A
− · ( 16)
Noting that at steady state the concentrations remain constant at a given x, a
nett mole balance from position 1 to position 2 over a full cycle gives:
( ) ( )
2 1 2 2 1 1
V V f C V V f C N − − − · ( 17)
The parameter f is used to take into account that the gas can move in any of
6 orthogonal directions, and therefore will have a minimum value of 1/6.
If the particles slide over each other as rigid layers, then the movement of
Chapter 4 – Active layer models
67
the gas will primarily be in the ydirection (perpendicular to the bed
surface), and f could have a value as large as ½. We will use f = 1/6 as our
default value.
Substitute α = V
2
/V
1
– this will later be estimated from the change of
voidage from a loosely packed solid to a tightly packed solid.
) )( 1 (
2 1 1
C C f V N − − · α ( 18)
The time required for one cycle is 1/m. It follows then that the molar flow
rate F
A
= Nm, and from the definition of molar flux W
A
, W
A
= F
A
/A.
) )( 1 (
2 1
1
C C
A
mfV
W
A
− − · α ( 19)
p
p
d
A
A d
A
V
1
1
1
ε
ε
· · (ε
1
is voidage for loosepacked solid.) ( 20)
) )( 1 (
2 1 1
C C mf d W
p A
− − · α ε ( 21)
p
d
C C
dy
dC
2 1
−
· ( 22)
D is determined by dividing Eqn. (21) by Eqn. (22):
) 1 (
1
2
α ε − · mf d D
p
( 23)
α can be determined from the ratio of the voidage of loosely packed
spheres vs tightly packed spheres.
Chapter 4 – Active layer models
68
1
1
1 1 1
1
1
− · ⇒
+
·
ε
ε
V
V
V V
V
S
S
( 24)
1
2
2
1
2
1
1
2
1
1
1
1
1
1
ε
ε
ε
ε
ε
ε
α
,
`
.

−
−
·
−
−
· ·
V
V
( 25)
For loosepacked spheres, 1  ε
1
= π/6
ε
1
= 0,4764
For tightpacked spheres,
ε
2
= 0,32
2
246 . 0
p
fmd D · ( 26)
The effect of binary diffusion must be adjusted to take account of the
reduced area for diffusion due to the solid particles, and the tortuosity of
the path. This diffusion contribution D
e
is commonly estimated by D
e
=
ε
2
D
a
. The overall dispersion coefficient for the active layer D
al
will be the
sum of D and D
e
:
D
al
= 0.246fm d
p
2
+ ε
2
D
a
( 27)
4. CONCENTRATION PROFILES AND
MASS TRANSFER
Having established the dispersion coefficient applicable to the active layer,
we now need to establish the concentration profile that exists within the
active layer in order to predict mass transfer. For the purposes of this
paper, we will assume that the gas is fully depleted at the boundary of the
Chapter 4 – Active layer models
69
active and passive layers. If the thickness of the layer containing
undepleted gas is much smaller than the width of the bed surface, then wall
effects can be neglected, and we have the situation of gas dispersion and
reaction in a flat plate. This analysis has already been done by Thiele
(1937) for diffusion in a catalyst.
Thiele (1937) showed that the concentration profile in a flat plate can be
determined (using our notation) as
φ φ
λφ λφ
−
−
+
+
· Ψ
e e
e e
, where
al
D
k
d
ε
φ · ( 28)
and the effectiveness factor is then
φ
φ
η
) tanh(
· ( 29)
The above equations (28) and (29) are valid for d being measured from the
catalyst surface to the center of the plate, where there is no flux at the
center due to symmetry. In our case, we have assumed more stringent
conditions, needing to assume that there is no flux at the bottom of the
active layer due to depletion of the gas (we do not have symmetry). In this
case, Eq. (29) reduces to
η = 1/φ ( 30)
For this model, with reaction only occurring at the surface of the bed, we
will find it convenient to express the reaction rate for the system in terms of
a mass transfer coefficient at the bed surface k
s
:
Chapter 4 – Active layer models
70
al s A s A s
V
A
V kC AC k dV P
al
⋅ · ·
∫
, ,
ηε ( 31)
Neglecting end effects, d = V
al
/A. Then we can rewrite Eqn. (31) in terms
of k
s
, and substitute Eqn. (29) to get
k D kd k
al s
ε ηε · · ( 32)
The rate constant εk can be calculated (Heydenrych et. al. 2001) using
2
) 1 ( 6
p
a
d
D Sh
k
⋅ −
·
ε
ε , Sh = 3.8 ( 33)
5. MODEL VALIDATION
Jauhari et. al. (1998) reported their mass transfer coefficients as k
s
A,
enabling us to make a direct comparison of our model with their results, by
multiplying Eqn. (33) by bed surface area A. We note that the results of
Jauhari et. al. (1998), measured at a bed fill of 4.3%, are substantially
proportional to rotation speed (Fig. 8), whereas our model predicts a much
smaller dependency on rotation speed, although the predicted values are in
the correct range.
Chapter 4 – Active layer models
71
Figure 8. Our model does not correctly predict the linearity of k
s
A with
rotation speed, but predicts values in the correct order of magnitude.
f = 1/6, d
p
= 3 mm.
The very small dependency of m on rotation speed (using Eqn. (10) to
estimate m, with c = 9.5 s
1
), results in a similarly low dependency of D
al
on
rotation speed. Using Eqn. (33) we calculated εk = 8.36 s
1
.
Clearly, this model does not accurately predict mass transfer rates for the
conditions used by Jauhari et. al. (1998). There may be two reasons for
this. The first is that mass transfer at the top surface of the bed, where the
freeboard gas slip velocity becomes important, has not been taken into
account in this model. In their measurements, Jauhari et. al. (1998)
operated under the constraint that the superficial axial gas velocity was less
than the drum wall velocity. This could lead to particularly low slip
velocities on the outer surface of the bed at the slower rotation speeds. This
could lead to the lower than expected mass transfer rates that were
measured at rotation speeds lower than 1 rpm.
The second possible reason for deviation from our model could be due to
the nature of the movement of particles in the active layer. Our model
Chapter 4 – Active layer models
72
assumes that the particles slide over each other, creating and destroying
voids in an orderly fashion. Boateng and Barr (1997) have measured the
granular temperature of particles in the active layer: a term that describes
the random movement of particles due to collisions within the active layer.
The frequency of these collisions may be much higher than the frequency
predicted by our model. Moreover, the granular temperature measured by
Boateng and Barr (1997) is dependent on depth (d) within the active layer,
which means that D
al
changes with bed depth, giving a different
dependency of k
s
A on rotation speed. It may also affect the mass transfer
resistance at the surface of the bed.
6. DISCUSSION AND CONCLUSIONS
A relatively simple model has been derived to predict bed depth and the
surface velocity of particles, based on the assumption that the shear stress
in the active layer is constant for a given bed depth and rotation speed.
A correlation was presented to predict the shear stress within the active
layer as a function of bed depth and rotation speed, m = 2(R/H)ω + c. The
constant c is expected to depend on the properties of the bed material; data
for polyethylene beads gives c = 9.5 s
1
.
By examining the effect of particles sliding over each other within the
active layer, where void spaces are filled and formed again, a correlation
was established for prediction of the effective dispersion coefficient D
al
.
Our model predicts that as long as the reactant does not occur deeper than
the active layer in significant concentrations, the mass transfer coefficient
for the bed surface can be calculated as k D k
al s
ε · .
Chapter 4 – Active layer models
73
These attempts to fundamentally predict the mass transfer coefficient were
not successful, based on the data measured by Jauhari et. al. (1998). The
most probable reason for the poor correlation of our model with experiment
was the fact that neither bed surface effects nor the granular temperature of
the bed material were considered in the determination of the effective
dispersion coefficient D
e
. Future work should be directed at a means of
predicting k
s
A based on granular temperatures within the active layer.
Chapter 4 – Active layer models
74
SYMBOLS
A bed surface area, m
2
a, b, c symbols used to describe an ellipse (Fig. 6)
C
A
gasphase concentration, mol m
3
C
A0
gasphase concentration at t = 0, mol m
3
C
A,s
gasphase concentration at the bed surface. mol m
3
d active layer depth, m
D binary diffusion coefficient, m
2
s
1
D
al
total dispersion coefficient for the active layer, m
2
s
1
h perpendicular distance from radial center of the kiln to the bed
surface, m
H bed depth at the deepest point (H = Rh), m
k gasphase reaction constant, s
1
k’ gasphase reaction constant for use in active layer, s
1
k
s
mass transfer coefficient, m s
1
L axial distance, m
m shear stress in the active layer (velocity gradient of particles) s
1
P
A
reaction rate, mol s
1
m
3
Q volumetric flow rate, m
3
s
1
R kiln radius, m
V volume, m
3
f factor for the directionality of dispersion,
2
1
6
1
< < f
W molar flux, mol s
1
Chapter 4 – Active layer models
75
Greek letters
η effectiveness factor, dimensionless
ε bed voidage, dimensionless
λ dimensionless distance; depth below bed surface/active layer depth
φ thiele modulus, dimensionless
ψ concentration, dimensionless
REFERENCES
Boateng, A.A. & Barr, P.V., (1997), ‘Granular flow behaviour in the
transverse plane of a partially filled rotating cylinder, J. Fluid Mech., 330,
233249
Heydenrych, M.D., Greeff, P., Heesink, A.B.M., Versteeg, G.F., (2001),
‘Mass transfer in rolling rotary kilns: A novel approach’, Submitted to
Chem. Eng. Sci, 2001.
Heydenrych, M.D., Schieke, J., Heesink, A.B.M., Kuipers, J.A.M. (2001),
‘Diffusion effects in rotating rotary kilns’, Accepted for publication at the
6
th
World Congress of Chemical Engineering, Melbourne, 2001.
Heinen, H., Brimacombe, J.K., & Watkinson, A.P. (1983a). Experimental
study of transverse bed motion in rotary kilns, Metall. Trans., 14B, 1983,
191205
Hoomans, B.P.B. (2000), ‘Granular dynamics of gassolid twophase
flows’, Doctoral thesis, University of Twente.
Chapter 4 – Active layer models
76
Jauhari, R., Gray, M.R., & Masliya, J.H. (1988). Gassolid mass transfer in
a rotating drum, Can. J. Chem. Eng, 76, 224232.
Kramers, H., & Crookewit, P. (1952). The passage of granular solids
through inclined rotary kilns. Chemical Engineering Science, 1, 259
Lebas, E.F., Hanrot, F., Ablitzer, D. & Houzelot, J.L. (1995).
“Experimental study of residence time, particle movement and bed depth
profile in rotary kilns”, Can. J. Chem. Eng, 73, 173179.
Saeman, WC, “Passage of solids through rotary kilns”, Chem. Eng. Prog.,
47, 1951, 508514.
Sullivan, J.D., Maier, C.G., & Ralston, O.C., (1927). Passage of solid
particles through rotary cylindrical kilns. US Bureau of Mines, Technical
Papers 384, 142
Tscheng, S.H. and Watkinson, A.P., (1979). Convective heat transfer in a
Rotary Kiln. Can. J. Chem. Eng., 57, 433443.
Chapter 5 – Combined model
77
Chapter 5  A combined mass transfer
model for both active and passive layers in
rolling rotary kilns
ABSTRACT
A mass transfer model for rotary kilns that incorporates both the active and
passive layer concentrations, is applicable over a wide range of rates.
Dispersion effects are included in the active layer model, and convection
(gas entrainment) effects are included in the passive layer model. At high
k/ω, it is not sufficient to use the contribution of the active layer alone. Gas
entrained into the passive layer at the wall of the kiln makes an important
contribution to mass transfer. At low k/ω, the passive layer model alone
can be used only for deep beds and low Thiele modulus φ, otherwise the
combined model is recommended.
By making two assumptions: 1) that molecular diffusion is unimportant
compared to dispersion, and 2) that the particle velocity profile in the active
layer (m) is proportional to rotation speed, then the effectiveness factor for
the bed is dependent on k/ω, h/R and R/d
p
. Of these, only R/d
p
is dependent
on the scale of the kiln.
Chapter 5 – Combined model
78
1. INTRODUCTION
In rolling rotary kilns, there are two distinct regions within the bed: the
passive layer, where particles are transported concentrically with the kiln
until they reach a point near the surface of the bed, and the active layer,
where these particles cascade down the surface of the bed. It has
previously been assumed that the active layer is the most important area for
mass or heat transfer. Heydenrych et. al. (2001a) have shown that for
slower reactions (or drying), mass transfer in the passive layer can also play
an important role. In this paper, we explore the transition from mass
transfer domination in the active layer, to kinetic domination in the passive
layer.
Figure 1. The regions found in a rotary kiln.
The work of Heydenrych et. al. (2001a) explores how gas entrapped in the
voids between particles in the active layer, gets depleted as it moves
concentrically in plug flow through the passive layer with the particles.
Chapter 5 – Combined model
79
They used an effectiveness factor to characterise the radial concentration
profile, and a gasphase volumetric rate constant to include mass transfer
limitation, for example, for 1
st
order processes:
P
A
= ηεkC
A
(1)
By rewriting Eqn. (1) in terms of dimensionless units, they showed that the
effectiveness factor depends only on k/ω and the bed fill. In later work
(Heydenrych et. al. 2001b), the effect of diffusion within the passive layer
was investigated. They showed that diffusion effects in the passive layer
are unimportant for slower reactions (k/ω < 1), but importantly, they also
showed that an additional dimensionless number must be used to describe
mass transfer in faster reactions when diffusion effects are to be taken into
account. This can either be a Thiele modulus, or (as they chose), a Peclet
number, defined as
e
D
R
Pe
2
εω
· (2)
Jauhari et. al. (1998) have measured mass transfer rates for the drying of
relatively small silicaalumina particles soaked with decane. These results
gave far greater mass transfer rates than the passive layer model
(Heydenrych et. al. 2001a) predicts. Jauhari et. al. (1998) used an
empirical model based on the volume of the active layer to predict mass
transfer as a function of rotation speed.
Heydenrych et. al. (2001c) developed a more fundamental model to predict
the shape and volume of the active layer. This model was largely based on
the assumption that the velocity gradient (m) of particles within the active
layer is constant at any point within the active layer, given a set of
Chapter 5 – Combined model
80
conditions such as bed depth and rotation speed. They predicted that bed
depth could be described by
1 /
2 2 2
−
− −
·
ω m
h x R
d (3)
This model was further extended to predict the dispersion coefficient in the
active layer caused by the change of void volumes that mix the gas as
particles slide over each other. If successive layers of particles in the active
layer move as rigid layers, then the dispersion of gas will occur
predominantly in the direction perpendicular to the surface of the bed, and
the volume of gas dispersed in this way can be calculated from the
difference in voidage of a loosepacked structure and a tightpacked
structure. Using these assumptions, they derived the following expression
to calculate dispersion in the active layer:
D
al
= 0.123 m d
p
2
+ ε
2
D
a
( 4)
They further showed that a relatively simple 1dimensional model could be
derived to predict the concentration gradients in the active layer in the case
of fast reactions (k/ω > 10), giving rise to a correlation for a mass transfer
coefficient for the surface of the bed:
k D k
al s
ε · ( 5)
Using published data to predict the change of the velocity gradient m with
rotation speed, they showed that the model represented by equations (3), (4)
and (5) is inadequate for predicting the mass transfer rates measured by
Jauhari et. al. (1998). In this paper, we would like to adapt the fundamental
approach adopted by Heydenrych et. al. (2001c) to one that better predicts
Chapter 5 – Combined model
81
the measured data of Jauhari et. al. (1998), for use as a design tool for
rotary kilns and applicable to reaction rates that cover a large range. As a
design tool, we need a model that is relatively easy to use, and some
simplifying assumptions will be made so that the model can easily be
solved on a spreadsheet.
2. COMBINED MODEL
We continue with the approach of Heydenrych et. al. (2001a), but we now
explore whether the rate constant k from Eqn. (5) is affected by the
continuous mixing of gas as the layers of particles slide over each other.
On the basis of Sørensen & Stewart (1974) the “rate constant” describing
the rate of evaporation of gas from the surface of wet particles in a stagnant
packed bed can be formulated:
ε
ε −
·
1 6
2
p
a
d
ShD
k , Sh = 3.8 (6)
Heydenrych et. al. (2001a) have shown that estimation of k using Eqn. (6)
works well to predict the slow rate of evaporation from relatively large
particles in the passive layer of rotary kilns, where there is a batchwise
saturation of gas in the voids as the gas rotates with the kiln. The gas
concentration in the voids changes according to
( )
dt
dC
C C k
A
A sat
· − (7)
Chapter 5 – Combined model
82
Integrating Eqn. (7),
kt
A sat
A sat
e
C C
C C
−
·
−
−
0
(8)
In the active layer, the same process can be expected to occur, except that
at regular intervals, the gas volume is flushed out with fresh gas. The
frequency of this cycle is given by the shear rate m, and therefore, time (t)
in Eqn. (8) is the inverse of m.
We wish to determine the rate of generation within a control volume V in
the active layer:
( ) ( )( ) dV m e C C dV m C t C dV P
m k
A sat A A A
ε ε
/
0 0
1 ) (
−
− − · − · (9)
Replacing the generation term P
A
with a more conventional rate expression,
as we have used in previous work,
( ) ( )( ) mdV e C C dV C C k
m k
A sat A sat
ε ε
/
0 0
1 '
−
− − · − (10)
( )
m k
e m k
/
1 '
−
− · (11)
Lastly, we assume an alternative correlation to describe the relationship of
m with rotation speed. This is essentially an empirical correlation, chosen
to reflect the mass transfer rates measured by Jauhari et. al. (1998):
m = 70ω (12)
It is interesting to note that by substituting m in Eqn. (12) into Eqn. (3), we
find that the shape and volume of the active layer is not dependent on
Chapter 5 – Combined model
83
rotation speed. Fig. 2 shows how this model predicts the maximum active
layer depth (x = 0) compared to the maximum active layer depths that were
measured for various materials and drum sizes, and reported by Heinen et.
al. (1983). For each material type, the active layer depth was measured at a
range of rotation speeds.
Figure 2. Surprisingly, relatively the steep particle velocity gradient of
Eqn. (12) predicts deeper active layer depths than those measured visually.
Using this correlation for m and replacing k with k′ of Eqn. (11), we predict
the mass transfer rates measured by Jauhari et. al. (1998) as follows:
Chapter 5 – Combined model
84
Figure 3: By using m = 70ω and ( )
m k
e m k
/
1 '
−
− · we can reasonably
predict measured mass transfer rates from ' k D k
al s
ε · . The data of
Jauhari et. al. (1998) is represented by ¿, and the line represents model
predictions.
Within the active layer, we predict concentration profiles at the deepest part
(x = 0), and assume that at any other part of the bed (x ≠ 0) the same
concentrations will be found at a given depth y from the surface. The
dimensionless concentration is estimated as:
φ φ
λφ λφ
λ
−
−
+
+
· · Ψ
e e
e e
C
C
A
A
0
) (
, where
al
D
k
d
'
max
ε
φ · (13)
Heydenrych et. al. (2001b) have shown that k/ω and either Pe or φ can be
used to describe concentration profiles in a rotary kiln. In this case, we
prefer to use φ because we have assumed that all diffusion effects take
place in the active layer. When binary diffusion is negligible, we can have
D
al
= 0.123 m d
p
2
(14)
Chapter 5 – Combined model
85
We can now determine φ from Eqn. (13) by substituting for d
max
using
Eqn. (3), D
al
using Eqn. (14), k′ using Eqn. (11) and finally m using
Eqn. (11):
]
]
]
]
,
`
.

−
]
]
]
]
− ·
,
`
.

,
`
.

−
2
70
1
1 1 72 . 9
R
h
e
d
R
k
p
ω
ε φ (15)
This reduces the model to the same dimensionless numbers that were
introduced by Heydenrych et. al. (2001a), but with the addition of the third
term R/d
p
. This is the only term that takes dispersion (and therefore scale
up effects) into account, bearing in mind that Eqn. (15) is only valid when
dispersion effects are much greater than molecular diffusion.
We will solve the model using the Thiele modulus φ as the third
independent variable (rather than R/d
p
), because it increases the validity of
the simulation results: it is possible to include binary diffusion effects, or to
use any other model to determine the Thiele modulus φ.
The active layer model described above predicts the concentrations that
occur at the interface between the active and passive layers of a rotary kiln.
We use these concentrations as the starting concentrations for the gas as it
moves concentrically around the kiln in the passive layer, getting depleted
as it moves as it would in a plug flow reactor. This is similar to the model
described by Heydenrych et. al. (2001a), except that the surface boundary
is not flat, but in the shape of the active layer interface, and the starting
concentrations vary along the boundary. This is not easily solved
analytically, but it can be quite easily solved numerically.
Both the active layer model and the concentric flow model described above
give a concentration profile over the radial crosssection of the bed. The
Chapter 5 – Combined model
86
effectiveness factor for a bed area (active or passive layer) is found by
integrating the dimensionless concentrations over that area as described by
Heydenrych et. al. (2001b):
al
al A
al
A
dA
∫
·
) (
ψ
η and
pl
pl A
pl
A
dA
∫
·
) (
ψ
η with
0 A
A
C
C
· ψ (16)
We use the same C
A0
for both the active and passive layers; C
A0
is the
freeboard gas concentration. We need to calculate the effectiveness of the
active layer η
al
separately from the effectiveness of the passive layer η
pl
in
order to get an overall effectiveness factor η, defined in terms of k.
pl pl al al bed
kA A k kA ε η ε η ηε + · ' (17)
bed
pl
pl
bed
al
al
A
A
A
A
k
k
η η η + ·
'
(18)
The term k′/k is determined by substituting m (Eqn. 12) in Eqn. (11):
,
`
.

− ·
−
ω
ω
70
1
70 '
k
e
k k
k
(19)
Note that k′/k is only a function of k/ω. Therefore the model for predicting
the dependent variable η is a function of three independent variables (we
consider ε as a constant): k/ω, h/R and φ. We could replace φ with R/d
p
using Eqn. (15), but then the results we show would be limited to the model
represented by Eqn. (15), so we have chosen φ.
Chapter 5 – Combined model
87
3. MODEL PREDICTIONS
A contour plot of the concentrations found using the model developed here
is shown in Fig. 4, where h/R = 0.5 and k/ω = 1 for a clockwise rotating
kiln. This shows the continuity of concentrations on the right hand side
active/passive layer interface, where the gas moves into the bed. On the
left hand side of Fig. 4, there is a very steep concentration gradient, which
is unlikely to occur in practice due to diffusion effects there, and the
transport of depleted gas from the active/passive layer interface into the
active layer. We expect, from the findings of Heydenrych et. al. (2001b,c),
that the diffusion effects in the active layer will predominate over the
transport effects in the passive layer. If this is the case, then the model
should be a reasonable indication of what happens in practice.
Figure 4. A contour plot of dimensionless concentrations using the
combined active/passive layer model with h/R = 0.5 and k/ω = 1.
Note how we find a high concentration at the wall of the kiln, because at
the wall, the active layer depth is very small, and gas enters the concentric
region at almost the maximum concentration.
Effectiveness factors at various φ are shown as a function of k/ω in Fig. 4.
Chapter 5 – Combined model
88
Figure 5. Effectiveness factors at various φ for a bed fill of 37% (h = 0.2).
The thick line represents the passive layer model (Heydenrych et. al.
2001a).
Not surprisingly, for larger kilns (larger φ) we always have a lower
effectiveness factor. It is interesting to note that the active layer does not
simply add to the mass transfer predicted by the passive layer model
(Heydenrych et. al. 2001a), but predicts lower values than that model. This
is due to the depletion of gas within the active layer, effectively blocking
reactive gas from entering the passive layer.
Fig. 6 shows lines of constant R/d
p
. These are useful because typically we
find that k varies along the length of a kiln as the solid dries (intraparticle
diffusion limiting) or as the solid material becomes converted, in the case
of reaction. This means that φ is not constant along the length of a kiln, and
as reaction proceeds, we move from right to left on Figure 3 along a line of
constant R/d
p
.
φ = 0.1 and 0.3
φ = 1
φ = 10
φ = 30
φ = 3
Chapter 5 – Combined model
89
Figure 6. Lines of constant R/d
p
are useful because typically R/d
p
remains
constant over the length of a rotary kiln.
Fig. 7 shows a similar plot to Fig. 5, but with a shallow bed, with bed fill of
5.2% (h = 0.8). The trends are similar, but the effectiveness factor here is
less dependent on k/ω because of the increased ratio of bed surface area
(active layer) to bed volume. Here, the passive layer model of Heydenrych
et. al. (2001a) is poor at predicting the shape of the curves, and cannot be
used even for slow reactions. As bed fill decreases, the active layer volume
increases relative to the passive layer volume. Therefore, even with slow
reactions (low k/ω), the active layer predominates and we do not benefit
fully from the entrainment effect of the passive layer as it draws gases into
the bed.
R/d
p
= 10
R/d
p
= 100
R/d
p
= 1000
Chapter 5 – Combined model
90
Figure 7. Effectiveness factors at various φ for a bed fill of 5.2%
(h = 0.8). The thick line represents the passive layer model (Heydenrych et.
al. 2001a).
In both Fig. 5 and Fig. 7, we find low effectiveness factors even at low k/ω.
This is unlikely in practice, because when the reaction rate is slow, usually
the Thiele modulus is also low. The lines of constant R/d
p
can be used as a
guide. If we consider a rotary kiln using pelletised particles of 10 mm
diameter, then a 4 m diameter kiln would give R/d
p
= 200. It would be
quite unusual to find a rotary kiln of a scale larger than R/d
p
= 1000,
because then molecular diffusion would predominate, and we can
approximate this using the dispersion model with an appropriate value of
d
p
. We find that d
p
= 2 mm gives D
al
values that are consistent with
molecular diffusion, and this gives R/d
p
= 1000 for the 4m diameter kiln.
To examine the contribution of active layer at various bed heights, we
determine what fraction the active layer contributes towards the total
effectiveness of the bed. This is calculated using the 1
st
term of Eqn. (18)
divided by the total effectiveness factor:
φ = 0.1 and 0.3
φ = 1
φ = 3
φ = 10 and 30
Chapter 5 – Combined model
91
η
η
bed
al
al
al
A
A
k
k
f
'
· (20)
Fig. 8 shows for φ = 3 how the active layer is always more important for
shallower beds. It is also interesting to note that for deep beds, the passive
layer continues to play a significant part in mass transfer even at high
reaction rates.
Figure 8. The contribution of the active layer is always more for shallower
beds (φ = 3). From top to bottom, curves show bed fills of 5.2%, 9.4%,
14%, 20%, 25%, 31% and 37%, corresponding to h = 0.8, 0.7, 0.6, 0.5,
0.4, 0.3 and 0.2.
The fraction contributable to the active layer does not approach unity for a
combination of two reasons. The first is that even for fast reactions that
occur very close to the surface of the bed, there is always an appreciable
amount of gas that moves into the passive layer at the wall of the kiln,
where the depth of the active layer tends to zero. The second effect is that
Bed fill 5.2%
Bed fill 37%
Chapter 5 – Combined model
92
the weighting due to the term k′/k comes into play, lessening the active
layer contribution. For k/ω = 100, for instance, k′/k = 0.23.
Noting that we typically get higher effectiveness factors using shallower
beds, and that commercially operated kilns typically use shallow beds, it
would be interesting to see if there is a maximum mass transfer that can be
achieved for a given kiln radius. Fig. 9 shows an effectiveness factor based
on total volume, rather than the bed volume that the previous effectiveness
factor was based on. We see that there may be a maximum at the highest
reaction rates (k/ω = 100), but in general, a deeper kiln gives better mass
transfer.
Figure 9. For fast reactions, it is possible to get an optimum bed fill, but in
most cases better mass transfer will be obtained with a deeper bed.
k/ω = 100
k/ω = 0.01, 0.1
k/ω = 10
k/ω = 1
Chapter 5 – Combined model
93
Figure 10. Active layer contribution for a deep bed (fill 37%) does not
change much as dispersion increases. φ (top to bottom) = 0.1, 0.3, 1, 3
and 10.
Figure 11. Active layer contribution for shallow beds (fill = 5.2%) is
highest for the larger kilns.
φ = 0.1  1
φ = 0.1 and 0.3
φ = 1
φ = 10
φ = 3
φ = 10
Chapter 5 – Combined model
94
For a shallow bed, we see a reversal in the order (Fig. 11), and we find that
the active layer contribution becomes increasingly important as φ increases.
4. COMPARISON WITH MEASURED DATA
Applying the combined model to Jauhari’s data (Fig. 12) gives small but
significant changes to the active layer model that was fitted using
Eqn. (12).
Figure 12. The combined model ⋅⋅⋅⋅⋅⋅⋅⋅⋅ predicts improved mass transfer
rates at the higher rotation speeds compared to the active layer model
÷÷. Jauhari’s data is represented by ♦.
The increase in the slope of the line predicted by the combined model is
due to two effects: Firstly, at lower rotation speeds, dispersion is low, and
the omission of binary diffusion from the model results in the prediction of
lower mass transfer rates. The second effect is the additional effect of the
passive layer. Figure 11 shows that for shallow beds, even at high k/ω,
there is still a significant contribution by the passive layer that the active
layer model does not account for.
Chapter 5 – Combined model
95
The data of Heydenrych et. al. (2001a) is shown in Fig. 13 against their
passive layer model (thick line) and the combined model (thin line), using
R/d
p
= 4.7. Although the scatter of the data does not allow definite
conclusions to be made, it appears that the combined model fits the data
better than the passive model alone.
Figure 13. The combined model (thin line) fits the data of Heydenrych et.
al. (2001a) somewhat better than the passive layer model (thick line).
5. CONCLUSIONS
In attempting to simplify the relationship for the velocity gradient in the
active layer to m = 70ω, we arrive at some useful relationships that provide
insight into the scaleup of rotary kilns. In particular, we showed that the
dimensionless number R/d
p
describes the dispersion effect within a rotary
kiln, and therefore it is the parameter that is important for scaling of rotary
kilns.
Chapter 5 – Combined model
96
Practically, we find a large range of R/d
p
in rotary kilns. An active layer
layer model that calculates the mass transfer coefficient k
s
at the surface of
the kiln can be used when the fraction contribution of the active layer is
high. In many cases, this is not true, and the combined model presented
here should be used. Also, the passive layer model used alone will in most
cases incorrectly predict the mass transfer, having a totally different form
(as a function of k/ω) to that predicted by the combined model.
It would also be incorrect to assume that the mass transfer predicted by an
active layer model can be added to the mass transfer from a passive layer
model: we have shown that in many cases the depletion of gas in the active
layer prevents activity in the passive layer.
Chapter 5 – Combined model
97
SYMBOLS
A crosssectional area [m
2
]
C
A
gasphase concentration, [mol m
3
]
C
A0
gasphase conc. at t = 0 or freeboard gas conc., [mol m
3
]
C
A,s
gasphase concentration at the bed surface. [mol m
3
]
d active layer depth, [m]
d
max
active layer depth at x = 0, [m]
D
a
binary diffusion coefficient, [m
2
s
1
]
D
al
total dispersion coefficient for the active layer, [m
2
s
1
]
D
e
effective dispersion coefficient for a packed bed [m
2
s
1
]
d
p
particle diameter [m]
f
al
fraction contribution of active layer towards total effectiveness
factor []
h perpendicular distance from radial center of the kiln to the bed
surface [m]
k gasphase reaction rate constant [s
1
]
k′ gasphase reaction constant for use in active layer, s
1
k
gs
mass transfer coefficient on the surface of particles [m s
1
]
k
s
mass transfer coefficient of active layer [m s
1
]
L length of the rotary kiln in axial direction [m]
m shear stress in the active layer (velocity gradient of particles) s
1
P
A
reaction rate of A [moles m
3
(bed) s
1
]
Q volumetric gas flow rate [m
3
s
1
]
R internal radius of the rotary kiln [m]
V volume [m
3
]
V
bed
volume of bed [m
3
]
V
al
volume of the active layer [m
3
]
x direction parallel to the surface of the bed in the radial cross
section; x = 0 is at the centre [m]
y direction perpendicular to the surface of the bed [m]
Chapter 5 – Combined model
98
Greek symbols
ε bed voidage [m
3
(gas) m
3
(bed)]
λ dimensionless distance; depth below bed surface/active layer depth
η effectiveness factor of reaction
φ Thiele modulus
ψ concentration, dimensionless
ω angular velocity [s
1
]
REFERENCES
Boateng, A.A. & Barr, P.V., (1997), ‘Granular flow behaviour in the
transverse plane of a partially filled rotating cylinder, J. Fluid Mech., 330,
233249
Heydenrych, M.D., Greeff, P., Heesink, A.B.M., Versteeg, G.F., (2001),
‘Mass transfer in rolling rotary kilns: A novel approach’, Submitted to Can.
J. Chem. Eng., 2001.
Heydenrych, M.D., Schieke, J., Heesink, A.B.M., Kuipers, J.A.M. (2001),
‘Diffusion effects in rotating rotary kilns’, Accepted for publication at the
6
th
World Congress of Chemical Engineering, Melbourne, 2001.
Heydenrych, M.D., Heesink, A.B.M., Kuipers, J.A.M. (2001), Versteeg,
G.F., ‘Hydrodynamics and dispersion in the active layer of rolling rotary
kilns’, Submitted to Can. J. Chem. Eng., 2001.
Chapter 5 – Combined model
99
Heinen, H., Brimacombe, J.K., & Watkinson, A.P. (1983a). Experimental
study of transverse bed motion in rotary kilns, Metall. Trans., 14B, 1983,
191205
Jauhari, R., Gray, M.R., & Masliya, J.H. (1988). Gassolid mass transfer in
a rotating drum, Can. J. Chem. Eng, 76, 224232.
Kramers, H., & Crookewit, P. (1952). The passage of granular solids
through inclined rotary kilns. Chemical Engineering Science, 1, 259
Lebas, E.F., Hanrot, F., Ablitzer, D. & Houzelot, J.L. (1995).
“Experimental study of residence time, particle movement and bed depth
profile in rotary kilns”, Can. J. Chem. Eng, 73, 173179.
Saeman, WC, “Passage of solids through rotary kilns”, Chem. Eng. Prog.,
47, 1951, 508514.
Sørensen, JP, Stewart, WE, 1974, Effect of fluid dispersion coefficients on
particletofluid mass transfer coefficients in packed beds”, Chem. Eng. Sci,
33, 13741384
Sullivan, J.D., Maier, C.G., & Ralston, O.C., (1927). Passage of solid
particles through rotary cylindrical kilns. US Bureau of Mines, Technical
Papers 384, 142
Tscheng, S.H. and Watkinson, A.P., (1979). Convective heat transfer in a
Rotary Kiln. Can. J. Chem. Eng., 57, 433443.
Publications
101
Publications
Van Vuuren, D.S. and Heydenrych, M.D., (1985), ‘Multicomponent
modeling of FischerTropsch slurry reactors’, CSIR rep. CENG581, 46 pp.
Van Vuuren, D.S., Hunter, J.R. and Heydenrych, M.D., (1988), ‘The
solubility of various gases in FischerTropsch reactor wax’, Chem. Eng.
Sci., 43(6) 12911296.
Heydenrych, M.D. and Van Vuuren, D.S. (1989) ‘Catalytic process for the
oligomerisation of olefins’, South African patent ZA 89 02,236.
Heydenrych, M.D. (1989) ‘Intrinsic kinetics of olefin oligomerisation on a
supported nickel catalyst’, Thesis, University of the Witwatersrand.
Heydenrych, M.D., Scurrel, M.S. and Nicolaides, C.P., (2001),
‘Oligomerization of ethene in a slurry reactor using a nickel(II)exchanged
silicaalumina catalyst’, Applied catalysis, Accepted for publication
Heydenrych, M.D., Reid, C. and Manzini, D. (2001), ‘Pressure drop over
fluidised bed distributor caps’, S.A.J. Chem. Eng., Accepted for publication
Moolman, F.S. and Heydenrych, M.D. (2001), ‘Fluorination of zircon in a
fluidized bed’, Can. J. Chem. Eng., Accepted for publication
Heydenrych, M.D., Greeff, P., Heesink, A.B.M., Versteeg, G.F., (2001),
‘Mass transfer in rolling rotary kilns: A novel approach’, Submitted to Can.
J. Chem. Eng., 2001.
Publications
102
Heydenrych, M.D., Schieke, J., Heesink, A.B.M., Kuipers, J.A.M. (2001),
‘Diffusion effects in rotating rotary kilns’, Accepted for publication at the
6
th
World Congress of Chemical Engineering, Melbourne, 2001.
Heydenrych, M.D., Heesink, A.B.M., Kuipers, J.A.M. (2001), Versteeg,
G.F., ‘Hydrodynamics and dispersion in the active layer of rolling rotary
kilns’, Submitted to Can. J. Chem. Eng., 2001.
Acknowledgements
103
Acknowledgements
At first glance, a doctoral thesis would seem to be an individual task, but it
ultimately involves many people. I would like to acknowledge the support
and contributions of the following people in particular:
My wife René, who supported my move to the academic world, and indeed
supported me in this doctorate all the way. My whole family needs to be
thanked, because an endeavour such as this inevitably encroaches on the
time spent with them. And when I am with them physically, they will attest
that I often seem far away . . .
To the University of Pretoria, and in particular Prof. Uys Grimsehl, who
did many things to provide the environment for completing this thesis. He
ensured that, apart from the core educational tasks that I was responsible
for, my time was free to concentrate on my research. He supported my
application to complete my studies at the University of Twente in every
way. In that respect, I also owe a debt of gratitude to the other members of
the Department of Chemical Engineering at the University of Pretoria for
taking up some of my duties when I left the Department on study leave to
the University of Twente.
Finally, to the University of Twente and my promotors Prof. Geert
Versteeg and Prof. Hans Kuipers for initiating and funding my stay at the
University of Twente. Special thanks go to my assistant promotor Dr. Bert
Heesink who taught me how to add value to papers in review – a skill that I
hope to be able apply much more effectively in my career in future. He
spent a lot of his time not only on the thesis, but both he and Prof. Geert
Versteeg spent a lot of time hosting me and my family in our stay in
Enschede – something I value greatly.
Acknowledgements
104
My hope is that in future we will be able to keep regular contact, and
strengthen the friendships we have established here, by actively promoting
the cooperation agreement between our respective Universities.
Curriculum Vitae
105
Curriculum vitae
Mike Heydenrych spent his primary and secondary school years in
Rustenburg, a town about 100 km west of Pretoria. He obtained his
bachelors degree at the University of the Witwatersrand in 1982 and then
completed two years of compulsory National Service. From 1984 he
worked at CSIR (a contract research organisation) as a team member,
determining the hydrodynamics of FischerTropsch slurry reactors. The
information was used by SASOL to scale up their slurry reactors, and these
are now used industrially.
Still working for CSIR, he completed an MSc (at the University of the
Witwatersrand) on the determination of the intrinsic kinetics of the
oligomerisation of ethene on a supported nickel catalyst. He then spent two
years as a consulting engineer for Megkon, a company specialising in
systems engineering. In 1993 he returned to CSIR to work on fluidised bed
combustors: fullscale design, supplying units to industry. After that, he
headed a large minerals processing development project that that sparked
his interest in mass transfer in rotary kilns.
In 1997, he joined the University of Pretoria where he is responsible for the
undergraduate and postgraduate courses in Reactor design and for the Final
year plant design project.
He is married to René with 2 children, Nicole and Graeme.
Dit proefschrift is goedgekeurd door de promoteren
Prof.dr.ir. G.F. Versteeg
en
Prof.dr.ir. J.A.M. Kuipers
en de assistentpromotor
Dr.ir. A.B.M. Heesink
ii
to my father
iii
dr.dr.L. P. South Africa No part of this book may be reproduced in any form by any means. M.D.J. The Netherlands ISBN 9036515440 Printing and binding: grafisch centrum twente. promotor prof.dr. L. G. nor translated into a machine language without written permission from the author. Heesink. nor transmitted.ir. A. Kuipers.M.dr.dr. Lefferts prof.ir.dr. Heydenrych. Enschede. ass. Modelling of Rotary Kilns Thesis University of Twente. J. G. Brem Universiteit Twente Universiteit Twente Universiteit Twente Universiteit Twente Universiteit Twente Universiteit Twente Universiteit Pretoria Universiteit Twente Cover: Effectiveness factor vs k/ω for the passive layer model overlaid by a rotary kiln crosssection showing evaporative mass transfer in the active layer.Samestelling promotiecommissie: prof.M.dr.ir.D.E van der Linden.ir. W.A.ir. promotor dr. Centurion. Versteeg. promotor prof. C.F. Copyright © 2001 M.ir. voorsitter prof.B.ir. The Netherlands iv . Asselbergs prof. de Vaal prof. Heydenrych.
the active v . Fig. The first part of the work considers the reaction of the gas as it moves in plug flow through the passive layer within the interparticle voids. and the various zones found within a rotary kiln. where particles move as a solid mass concentrically around the axis of the kiln. They are returned to the top part of the bed in the passive layer. and their relation to each other. where analogous dimensionless numbers can be identified. 1 shows the nomenclature used in this summary. particles slide over each other in granular flow. Radial cross section of a kiln. In the active layer. Figure 1. this work is also applicable to heat transfer in rotary kilns. In this respect. The most important contribution is the identification of the dimensionless numbers that contribute to mass transfer in rotary kilns.SUMMARY This work primarily covers mass transfer in rotary kilns. showing the nomenclature. Another important contribution is the recognition of the important contribution of the passive layer in mass transfer.
shown in Fig. the active layer depth can be estimated at any distance x along the bed surface using vi . again with the assumption of an infinitely thin active layer. The effect of diffusion within the passive layer was investigated based on a twodimensional continuity equation for the radial cross section.layer is considered to be infinitely thin. Figure 2. but at higher values of k/ω. active layer modelling is necessary to correctly predict mass transfer. Comparison against experimental data shows that this relationship is only applicable to slower reactions (k/ω < 1). The effectiveness factor η for the bed is found to be dependent on the bed fill (represented by h/R) and the ratio of the rate constant to the rotation speed (k/ω). 2. This requires the use of an additional dimensionless constant: either a Peclet number (εωR2/De) or a Thiele modulus ( R k / De ). diffusion within the passive layer can be neglected. It was shown that for practical values of Peclet number. By making the (justified) assumption that the velocity gradient of particles (m) within the active layer is constant at any point. Effectiveness factor increases with decreasing bed fill.
d= R 2 − x2 − h2 m /ω −1 The action of particles sliding over each other promotes movement of gas through the interparticle voids. Finally. It is also necessary to take into account the periodic renewal of gas in the voids between the particles within the active layer. a good prediction of mass transfer could be obtained for fast reactions.246fm dp2 + ε2Da where f is a directionality factor. where gas is entrained into the bed through rotation (Fig. and the above assumptions. This combined model assumes continuity of concentration at the interface of the active and passive layers. f = 1/6. vii . Using m = 70ω. a model was developed that built on the principles established for both the active and passive layers. 3). with equal dispersion in all directions. A correlation for an active layer dispersion coefficient has been derived: Dal = 0. This is done by adapting the value of the kinetic constant using k ' = m 1 − e −k / m ( ) where k′ is the effective rate constant applicable to the active layer region. With fast reactions mass transfer can be estimated in terms of a mass transfer coefficient on the bed surface using: k s = Dalεk This does not fully predict the dependence of mass transfer on rotation speed.
An example of the predictions of the combined model is given in Fig.2%. It is also clear from the shape of the curves that the passive layer model alone is inadequate at predicting effectiveness factor. the viii . the effectiveness factor is strongly dependent on R/dp. For such a shallow bed. even for low k/ω. and that usually both regions in a rotary kiln contribute significantly to mass transfer. 4 for a bed fill of 5.Figure 3. This model predicts that it is seldom possible to use either the passive layer model alone. A typical concentration profile using the combined model (clockwise rotation). or the active layer model alone to accurately predict mass transfer. We also find that when dispersion effects dominate over molecular diffusion effects in the active layer (expected for particle size > 3 mm). then scaleup effects can be represented by the ratio of kiln radius to particle diameter (R/dp) instead of the active layer Thiele modulus.
the recognition of the contribution of the passive layer towards mass transfer. ix . the relative simplification of the modelling approach by identifying and using appropriate dimensionless numbers to characterise the system. Effectiveness factors for a rotary kiln with bed fill 5. In this way. The value of the work lies in two aspects: firstly. at a given bed depth.R/dp = 10 R/dp = 100 R/dp = 1000 Figure 4. the effectiveness factor is characterized by a single graph such as that of Fig. and secondly. 4. a similar analogy should be possible. The thick line shows the prediction of the passive layer model (Chapter 2). For heat transfer.2%.
De opschaling van deze reactoren geschiedt vaak nog op basis van empirie omdat een gedegen theoretische beschrijving van de processen nog ontbreekt. Er wordt met name aandacht besteed aan de identificatie van dimensieloze kentallen die het stoftransport in roterende ovens karakteriseren en hun onderlinge afhankelijkheid.SAMENVATTING Dit proefschrift behandelt stofoverdracht in roterende ovens. Met dit proefschrift wordt getracht om deze leemte op te vullen. Dergelijke gasvast reactoren worden veelvuldig in de industrie toegepast. Figuur 1. waarbij de cement oven wellicht het bekendste voorbeeld vormt. zoals die in een roterende oven worden onderscheiden. Dwarsdoorsnee van een roterende oven Figuur 1 laat de verschillende zones zien. Daarnaast vormt de analyse van de afzonderlijke stoftransportmechanismen een belangrijk onderdeel van dit proefschrift. Door de roterende beweging van de oven x . Door toepassing van deze dimensieloze kentallen zijn de resultaten ook toepasbaar voor de beschrijving van het analoge warmte transport in roterende ovens. In de actieve laag bewegen de deeltjes langs en over elkaar heen naar beneden.
worden de deeltjes die het eind van de actieve laag hebben bereikt. Er wordt een effectiviteitsfactor η gedefinieerd als de verhouding tussen de werkelijke omzettingssnelheid in het bed en de snelheid die zou worden bereikt wanneer overal in het bed de concentratie van de gasvormige reactant gelijk zou zijn aan de concentratie in het freeboard boven het bed. opgesloten tussen de deeltjes. beweegt daarbij als het ware in propstroom concentrisch door de passieve laag. Het gas. Deze blijkt sterk afhankelijk te zijn van de vulhoogte van de oven (1h/R) en de verhouding tussen de reactiesnelheids Figuur 2. Na een inleiding over de relevantie van roterende ovens voor de industrie. Effectiviteitsfactor bij verschillende vulhoogtes. De actieve laag wordt daarbij oneindig dun verondersteld. via een roterende beweging door de onderliggende passieve laag weer naar het bovenste gedeelte van de actieve laag getransporteerd. Zowel in de actieve als in de passieve laag kunnen de deeltjes reageren met het gas dat vanuit het “lege” deel van de oven (het zogenaamde freeboard) continu wordt aangevoerd. xi . wordt in het tweede hoofdstuk van dit proefschrift het transport en de reactie van het gas in de passieve laag beschouwd. De passieve laag roteert hierbij als een star lichaam om de as van de oven.
Een vergelijking met experimentele data (afkomstig uit de literatuur en eigen data) laat zien dat de gevonden afhankelijkheid alleen toepasbaar voor relatief langzame reacties (k/ω < 1).constante en de rotatiesnelheid (k/ω). kan de dikte van de actieve laag worden afgeschat voor elke positie x (zie Figuur 1) langs het grensvlak volgens: d= R 2 − x2 − h2 m /ω −1 De beweging van de deeltjes over en langs elkaar heen veroorzaakt gastransport door het steeds weer creëren en opbreken van interstitiële ruimtes. In de praktijk blijkt diffusie in de passieve zone een verwaarloosbaar kleine bijdrage aan het stoftransport te leveren. Geconcludeerd wordt dan ook dat een nadere beschouwing van de processen die optreden in de actieve laag noodzakelijk is. Door de (gerechtvaardigde) aanname dat de snelheidsgradiënt van de deeltjes (m) binnen de actieve laag constant is op elke positie binnen die laag. Dit model resulteert in een extra dimensieloos kental: ofwel een Pecletkental (εωR2/De) ofwel een Thiele modulus ( R k / De ). Op basis hiervan is een overall dispersiecoëfficiënt voor de gasfase afgeleid: xii . In het derde hoofdstuk wordt het effect van diffusie binnen de passieve zone bestudeerd met behulp van de continuïteitsvergelijking voor een dwarsdoorsnee van de roterende oven. In hoofdstuk 4 worden de hydrodynamica van en de stofoverdrachtsprocessen in de actieve laag nader onder de loep genomen. Voor hoge waarden van k/ω (k/ω >1) blijkt ook dit aangepaste passievezone model niet in staat om de experimentele resultaten te beschrijven. zoals weergegeven in Figuur 2. Hierbij is wederom aangenomen dat de actieve laag oneindig dun is.
Dit gecombineerde model gaat uit van een continu concentratieprofiel aan het grensvlak tussen de actieve en de passieve zones. Voor m=70ω en bovengenoemde aannames wordt een goede beschrijving van de in de literatuur gerapporteerde experimentele data gevonden. In het geval van snelle gasvast reacties kan de stofoverdracht worden geïnterpreteerd in termen van een stofoverdrachtscoëfficiënt ks. terwijl ε de (gemiddelde) porositeit van de actieve laag voorstelt. Voor gelijke dispersie in elk van de orthogonale richtingen geldt f = 1/6. De symbolen dp en Da staan respectievelijk voor de deeltjesdiameter en de moleculaire diffusiecoëfficiënt van het reagerende gas.246fm dp2 + ε2Da In deze vergelijking verdisconteert f de richting van het transport. Het gecombineerde xiii . Dit kan worden gedaan door aanpassing van de kinetiekconstante. Het is noodzakelijk om ook het effect van de periodieke verversing van gas in de holtes tussen de deeltjes te verdisconteren. volgens: k ' = m 1 − e −k / m ( ) waarin k′ de effectieve reactiesnelheidsconstante voor de actieve zone voorstelt.Dal = 0. betrokken op het totale bedoppervlak: k s = Dalεk Hierbij wordt echter het effect van de rotatiesnelheid op het stoftransport niet volledig verdisconteerd. Tot slot wordt in hoofdstuk 5 een model ontwikkeld dat gebaseerd is op de eerder genoemde beginselen voor de actieve en passieve zones.
Dikke lijn geeft de voorspelling van het passievezone model. xiv . Meestal zijn beide zones belangrijk in het beschrijven van het stoftransport in een roterende oven. De effectiviteits R/dp = 10 R/dp = 100 R/dp = 1000 Figuur 4.2%. Een typisch concentratieprofiel volgens het gecombineerde model (kloksgewijze rotatie) model voorspelt dat slechts bij hoge uitzondering alleen het passieve zone model of alleen het actieve zone model voldoet. kunnen opschaaleffecten worden verdisconteerd via de verhouding van ovenstraal ten opzichte van deeltjesdiameter (R/dp).Figuur 3. Effectiviteitsfactor voor een roterende oven met een vulgraad van 5. Wanneer in de actieve zone moleculaire diffusie te verwaarlozen is ten opzichte van gasfase dispersie (en dit is aannemelijk voor deeltjes groter dan 3 mm). In Figuur 4 is een voorbeeld van een berekeningsresultaat van het gecombineerde model gegeven voor een kleine vulhoogte.
De resultaten van dit proefschrift tonen het belang aan van zowel de actieve als de passieve laag voor het totale stoftransport. Op deze wijze kan. voor een gegeven vulgraad van de oven. zelfs niet voor lage waarden van k/ω. de effectiviteitsfactor als functie van de procesparameters in een enkele grafiek worden weergegeven.factor blijkt sterk af te hangen van (R/dp). xv . Ook blijkt duidelijk dat het passievezone model alleen niet in staat is de effectiviteitsfactor goed te beschrijven. Door enkele vereenvoudigende aannamen en het gebruik van de juiste dimensieloze kentallen kan de stofoverdracht in roterende ovens goed worden gekarakteriseerd. Voor warmtetransport is een analoge analyse mogelijk. De term “passief” is zeker niet altijd op zijn plaats.
CONTENTS 1. Diffusion effects in rotating rotary kilns Introduction Mathematical model Results Discussion Conclusions 37 4. Hydrodynamics and dispersion in the active layer of rolling rotary kilns Introduction Hydrodynamics of the active layer Dispersion within the active layer Concentration profiles and mass transfer Model validation Discussion and conclusion 53 xvi . axial direction The flow regimes in rotary kilns The need for modelling of rotary kilns An overview of this work 1 2. Mass transfer in rolling rotary kilns: a novel approach Introduction Modelling studies Model development Total depletion model Model validation Conclusion 11 3. Introduction The radial vs.
A combined mass transfer model for both active and passive layers in rolling rotary kilns Introduction Combined model Model predictions Comparison with measured data Conclusions 77 Publications 101 Acknowledgements 103 Curriculum Vitae 105 xvii .5.
.
hazardous waste incinerators operate with relatively deep beds. Rotary kilns are popular for this role because of their flexibility. mixing. Although this is not a cement kiln. al.Introduction Chapter 1 . Figure 1. 1998). Another important application of rotary kilns is for the incineration of waste materials. Fig. incineration. Cement kilns can be very large. The scale of this 500 tpd lime kiln can be judged from the pickup truck to the left of the kiln. The most common and industrially important application of rotary kilns is in cement production.Chapter 1 . heating. calcination. An example of such a hazardous waste incinerator is shown in Fig. 1998).Introduction Rotary kilns are found in many processes processing. all major producers use the rotary kiln as their equipment of choice. 1 . and removal of waste solids at the exit presents no problems. sintering and gassolid reactions (Jauhari et. 1 shows a 500 tpd lime kiln from Greer lime. They can handle a large variety of feed materials. cooling. al. 2. reducing. with variable calorific value. Typically. and have a secondary combustion chamber after the rotary kiln to improve the heterogeneous combustion of waste (Rovaglio et. humidification. that involve solids These include drying.
the first aspect of rotary kilns that was modelled was the movement of material through a rotary kiln. A rotary kiln waste incinerator with a vertical secondary combustion chamber. 1979). Rotary kilns are amongst the most wellestablished unit operations in the process industry. and these are heat transfer. (Bui et. In the design of kilns. there are many applications of rotary kilns: Some of the applications that have been published are magnetite oxidation (Davis. This is reflected in the number of papers published on the modelling of heat transfer in rotary kilns. induration of iron ore pellets (Young et. flow of material through the rotary kiln. coke calcining. Heat transfer rates amongst the most important of these aspects. al. The first such modelling was done by Saeman (1951). yet are amongst the least understood. (1927).Chapter 1 – Introduction Figure 2. there are four important aspects to consider from a process engineering point of view. they are used to achieve a combination of these aims. it is the heat transfer that limits the performance of the rotary kiln. 1996). reacting and drying of solid material. based on data measured by Sullivan et. because in many cases. 1949). In the mineral processing industry. al. However. and in many cases. al. They can be used for 3 purposes: heating. 2 . gassolid mass transfer and reaction. 1995) and drying (Friedman & Marshall.
Introduction The subject of mass transfer in rotary kilns has been neglected until Jauhari et. this repeated process can cause the particle to move in the axial direction every time it moves down the active layer. Considering a single particle.Chapter 1 . (1998) published an article on gassolid mass transfer in a rotating drum. al. 3. 3 . These models can predict the bed fill (fractional area of the bed in a radial cross section) along the length of a rotary kiln as a function of solids feed rate. can equally well be applied to the modelling of heat transfer. we typically find diagrams of a radial crosssection through a rotary kiln. where they slide downwards in the active layer. The purpose of this thesis is to model mass transfer and reaction in rotating drums. but the novel approach adopted by the author is firmly based on the foundations of chemical engineering science. This is the basis for models that predict the movement of particles through a rotary kiln in an axial direction. The radial plane vs axial direction In the literature on rotary kilns. such as that shown in Fig. as long as the rotary kiln is tilted in the axial direction. This shows the movement of particles within the bed of a rotary kiln: particles move concentrically with the wall of the rotary kiln in the passive layer until they reach the surface of the bed.
Therefore. (This approach encompasses drying as a subset of reaction. we will refer to the operation as reaction. Radial cross section of a rotary kiln In operation. then we can use that radial crosssection as a differential element to model the rotary kiln as a whole in the axial direction using established engineering practices. 4 .Chapter 1 – Introduction Figure 3. at any point along the length of a rotary kiln. gas may be flowing through the freeboard of the kiln. solids heat/cool/react/dry progressively as they move through the kiln. either cocurrently or countercurrently. we can expect a different set of conditions to apply to both the solid and the gas. In this thesis.) Simultaneously with the movement of the particles. We will assume (as most researchers in the field do) that if we can understand and model the processes that occur in the radial crosssection.
b) did useful work on this subject. al. Figure 4.Introduction The flow regimes in rotary kilns The nature of the solids flow in rotary kilns can change significantly with particle characteristics. the bed surface is continuously renewed. An example of a bed behaviour diagram. while in slumping mode. This process occurs repeatedly. the angle of repose of the material increases as the static bed rotates. or at lower rotation speeds. kiln size. the motion is either rolling. 4). In rolling mode. 5 . the renewal occurs at fixed intervals. For deeper beds. slumping. introducing the concept of a Bed Behaviour Diagram (Fig. rotation speed and bed fill. the bed is prone to slippage between the wall of the kiln and the bed.Chapter 1 . 3. Heinen et. as described in Fig. and is less desirable than rolling mode for heat and mass transfer. (1983a. each material has its own characteristic bed behaviour diagram. For rotary kilns with shallow beds in particular. In slumping mode. until a segment of material shears off and comes to rest in the bottom half of the kiln.
Because a thorough understanding of the processes occurring within rotary kilns has not been a prerequisite for their apparently satisfactory operation. and multiple fuel capability is possible. and they fall down the surface of the bed in a trajectory. now threatens its future. Rotary kilns can handle feed stocks with broad particle size distributions or whose physical properties change significantly during processing. which has in the past ensured the survival of the rotary kiln. Each of these modes of operation produces different conditions for particles within the rotary kiln. and in most literature articles. and therefore should be modelled differently. roasting and sintering of a variety of materials. In addition. dirty fuels often are utilised without serious product contamination. The need for the modelling of rotary kilns The need for a deep understanding of kilns and the need for effective fundamental models has been expressed very well by Barr et. while the long residence time of the material within the kiln promotes uniform product quality. heating (or cooling). reducing. This causes a distortion in the surface of the bed (which is normally flat). where particles are flung to the wall of the kiln. they continue to find applications in the drying. Further increasing the rotation speed finally results in centrifugation. resulting in poor mass and heat transfer. Paradoxically. research has not 6 . Despite challenges from newer and more specialised gassolids reactors. because the particles do not mix. In this thesis. This is also an undesirable regime of operation. this versatility. (1989): “Rotary kilns are ubiquitous fixtures of the metallurgical and chemical process industries. the momentum of the particles as they reach the surface of the bed becomes important. modelling is applicable to the rolling mode – the preferred mode of operation industrially.Chapter 1 – Introduction At higher bed fills and rotation speeds. al. calcining.
Introduction progressed apace with competing. This model is not flexible enough to be used over a wide range of k/ω because of the need to consider the active layer. The model was based on chemical reactor theory.” An overview of this work This work considers mass transfer in rotary kilns. less tolerant. reacting as it is transported through the bed. In the passive layer model. rotary kilns will remain in the position of operating below their optimal performance in an increasingly sophisticated marketplace. which involves solving a partial differential equation in the two dimensions of the radial cross section. by starting in Chapter 2 with a passive layer model. However. The gas is entrained in the interstices between the particles. In the next chapter we consider passive layer diffusion. Until all the internal processes are understood and become predictable. it does provide useful insight into the dimensional groups that are required to describe scaleup effects in a rotary kiln: either a Peclet number (εωR2/De) or a Thiele modulus ( R k / De ) should be used in addition to k/ω and bed fill. we consider the movement of gas through that part of the bed that moves as a concentrically around the axis of the kiln. In this first model we show that the effectiveness factor of the bed is dependent on just two variables: the ratio of the rate constant to the angular velocity (k/ω).Chapter 1 . reactors. because mass transfer can be considered as a subset of reactor theory – mass transfer becomes limiting for very fast reactions. and the bed fill. 7 .
We also find that when dispersion effects dominate over molecular diffusion effects in the active layer (expected for particle size > 3 mm). a combined model is presented that expands on the theory established in Chapter 4 for the active layer. then scaleup effects can be be represented by the ratio of kiln radius to particle diameter (R/dp) instead of the Thiele modulus.Chapter 1 – Introduction In Chapter 4. and uses the principles of Chapter 2 for modelling the passive layer. some fundamental models for the active layer are presented: a hydrodynamic model that predicts the depth and shape of the active layer. We find that it is seldom possible to use either the passive layer model alone or active layer model alone to accurately predict mass transfer. 8 . and that usually both regions in a rotary kiln contribute significantly to mass transfer. and a dispersion model for predicting mass transfer within the active layer. In the final chapter.
and Perron. Brimacombe. J.A. Sci. 14B.J.. Eng.P. ‘Mass transfer in rolling rotary kilns: A novel approach’. Trans.. 73.. A. & Marshall. G. Metall. (1997). ‘Granular flow behaviour in the transverse plane of a partially filled rotating cylinder. Simard. J. 10041009.A. Metall. Can.. J.B.. 2001. Chem.. Pilot plant trials’. P.P. J.. 14B. Friedman.T. Y. Can. Kocaefe. Heinen.. Greeff. A..V. H.V. & Watkinson..R.M. A.K. 233249 Bui.K. Heesink.P. ‘A heattransfer model for the rotary kiln: Part 1. (2001). 1983. 20B.. S.K. Chem. R.. 191205 Heinen. A. H.. G. (1996). R. The modelling of transverse solids motion in rotary kilns’. & Barr. A. 74. Eng.. 482493. Metall. Davis. (1983a). J. P. M.F.. (1949).. & Watkinson. Trans. Eng.. Trans. 9 . Eng.. Brimacombe.. 534545. J... (1989). Chem.Chapter 1 . P. A. Versteeg. J. ‘Mathematical model of magnetite oxidation in a rotary kiln furnace’.. Brimacombe. ‘Studies in rotary drying’. Prog. Submitted to Chem. August 1949. J.Introduction REFERENCES Barr. (1983b). 1983.. 330.. ‘Experimental study of transverse bed motion in rotary kilns’.D. & Watkinson. (1995). Charette. 391402 Boateng.. Fluid Mech. 1989. 207220 Heydenrych.
259 Rovaglio.R. Cross. Can.P..M. M. Jauhari.. ‘Diffusion effects in rotating rotary kilns’..H. H. Heesink.. Maier. (2001). WC. M.D.H. The passage of granular solids through inclined rotary kilns.. R. J.G. Kuipers.D. ‘Dynamic modeling of waste incineration plants with rotary kilns: comparisons between experimental and simulation data’. (1998). Schieke. C.W. Sullivan. G. 47. and Gibson. J.D.A. A. Accepted for publication at the 6th World Congress of Chemical Engineering. Eng. US Bureau of Mines. R. Can. J. M.. P. D. J. and Watkinson. 76. J. 508514. Manca. Gray.. 57. (1988). 433443. “Passage of solids through rotary kilns”. J. (1979). Chem. Eng. & Masliya. 2001. (1927).C. Technical Papers 384. R. Chem. O.... (1979).M. Sci. Kramers. Eng. 142 Tscheng.. M. 224232. Melbourne. Prog. Chemical Engineering Science. 1951.Chapter 1 – Introduction Heydenrych. Eng. Young. Gassolid mass transfer in a rotating drum. & Crookewit.. A. 10 ... Convective heat transfer in a Rotary Kiln. S.. Ironmaking Steelmaking 1. Chem. ‘Mathematical model of Gratekilncooler process used for induration of iron ore pellets’. 1. & Biardi.. & Ralston... 113. Passage of solid particles through rotary cylindrical kilns. (1952). Chem. 53(15) p 27272742 Saeman.B.
preheating. Consequently. Experimental data confirms the validity of this model for slower rates. incineration. INTRODUCTION Rotary kilns are used industrially in many applications such as drying. heat transfer is the limiting factor. 1. humidification. A reactor modeling approach has been used to derive effectiveness factors for the bed as a function of bed fill. For many rotary kilns. 1989. It is important 11 . and in the reaction zone (Barr. reaction kinetics and rotation speed. sintering and gassolid reactions (Barr. calcining. where particles move concentrically with the geometry of the kiln and gas is entrained by these particles. The movement of gas in the interparticle voids in the bed of the kiln is considered.Mass transfer in rolling rotary kilns: a novel approach ABSTRACT A novel approach to modeling mass transfer in rotary kilns is explored. leading to a simplified model for the bed effectiveness factor. reducing. such as dispersion or diffusion are also important in these conditions. 1998). mass transfer can be much higher than the model predicts. mixing. 1989). At faster rates.Chapter 2 – Passive layer model Chapter 2 . the entrained gas becomes depleted within the bed. In many cases. indicating that other mechanisms. both in the heating section of the rotary kiln. most focus in the literature has been directed at understanding the heat transfer processes in rotary kilns. Jauhari.
MODELLING STUDIES The first published experimental studies on rotary kilns recorded the relationship of rotation speed and kiln inclination on bed depth and solids residence time (Sullivan et al. Using the geometry of an inclined rotary kiln. 2. was confirmed (Lebas et al. In later work that specifically measured the movement of particles at the surface of the kiln. 1951). the validity of the model of Kramers and Crookewit (1952). 1995). and the model was further refined to predict axial movement of particles with different bed fills. A model was later developed based on the assumption that particles in a rolling bed move in a circular motion with the rotation of the kiln.Chapter 2 – Passive layer model though to understand all of the processes that occur in rotary kilns on a fundamental level before rotary kilns can be designed and operated optimally. 1927) well. the angle of inclination necessary to maintain a constant bed height over the length of the rotary kiln could be determined for a given rotation speed. More recently. 1927). mass transfer is also important. as well as in gassolid reacting systems with high specific reaction rates such as incineration. This basic model predicted the original data (Sulllivan et al. In this work we will examine the mass transfer in rolling rotary kilns and propose a novel approach to describe the phenomena that determine the rate of mass transfer. taking into account the time for particles to fall down the surface of the bed (Kramers and Crookewit 1952). and then fall down the surface of the bed in a thin layer (Saeman et al. with minor exceptions. With physical processes like drying and humidification for example. The time taken to fall down the surface was assumed to be small compared to the time for a particle to move with the kiln from the bottom half to the top half of the bed. the same fundamental model of the path of particles 12 .
have been extensively examined. 1983a). The time for a colored particle to move down the surface of the bed was measured with video photography (Lebas et al. and can be represented using a Bed Behavior Diagram (Heinen et al. this ratio does not change with rotation speed. In this paper.Chapter 2 – Passive layer model was shown to be correct in a study of particle motion where there is no nett axial flow of particles in a rotating drum (Spurling et al. 2000). 13 . we will focus on kilns operating in the preferred mode for most circumstances . and this typically takes 15 . 1.30% of a full cycle.rolling. 1995). cataracting. However. slumping. An example of a Bed Behavior Diagram (Heinen et al 1983a). centrifuging . and the velocity of the particles moving down the surface of the bed was interpreted to be directly proportional to rotation speed (Lebas et al. Figure 1. rolling. 1995). The conditions that determine the mode in which a rotary kiln operates slipping. cascading. An example of a Bed Behavior Diagram is shown in Fig.
As the particles approach the bed surface due to the kiln rotation. except for deep beds with small particles. falling to as low as 4% for slowly rotating kilns (Heinen et al. and found that it is relatively thin for beds with a depth that is much larger than the diameter of the particles. where the depth of the active layer is less than 12% of the bed depth. 2).Chapter 2 – Passive layer model The layer of particles moving down the face of the bed leads to improved mass transfer. The various zones within a rotary kiln. In most cases. 3. with no slippage between particles. Figure 2. the active layer is typically less than 8 particle diameters deep. 1983a). MODEL DEVELOPMENT For most of the volume of the bed of a rotary kiln (the passive layer). particles move concentrically with the wall of the bed. Heinen et al (1983a) measured particle velocities inside this layer as well as the thickness of it. they slip down the face of the bed. and come to rest near the 14 . therefore the surface layer is often referred to as the “active layer” (See Fig.
The perpendicular height dx can be found as follows: dx = (sin θ) R dθ = (c/R) R dθ = cdθ dAx = ½c dx = ½c2 dθ (1) (2) 15 . Overall displacement of particles at the surface of a kiln (irrespective of the path). The differential area within ABC can be approximated by ½ base x perpendicular height. the bed has been represented horizontally): Figure 3. We will attempt to quantify the volumetric flow rate and reaction rate of the gas that is trapped in the interparticle voids in the passive layer.1 Gas volumetric flow rate The volume of entrained gas can be determined for a single rotation of the kiln (for convenience.Chapter 2 – Passive layer model bottom of the surface of the bed. but for convenience has been drawn horizontally. 3. Note that the bed surface will normally be inclined.
All entrained gas will pass this line. and converting to angular velocity. (4). A mass balance of the particles also means that the particles will move infinitely fast over the surface of the bed with the assumption of an infinitely thin active layer. with a finite active layer thickness.Chapter 2 – Passive layer model The area useful for the entrainment of gas is proportional to the void fraction of gas in the bed: dAxg = ε 2 (R − h 2 )dθ 2 (3) Integrating this expression over 1 radian. If the active layer is infinitely thin. By estimating the concentration of the gas as it leaves the bed. will be taken through to its conclusion. the particles fill move down at a finite velocity. not affecting the entrainment of gas perpendicular to the bed surface. passing perpendicularly through the bed surface to the deepest part of the bed. consider a line extending from the radial centre of the kiln. the admittedly unrealistic assumption of an infinitely thin layer. for the purposes of this paper. In practice. the overall reaction rate can be determined. However. the volume flow rate of gas is: Q = ½εωL(R2 – h2) (4) This gives the volume flow rate of entrained gas through the bed of a rotary kiln. 16 . To understand how the active layer influences the movement of gas. the flow rate Q can easily be shown to be as in Eq.
the effect of the active layer will be important. and dispersion effects will predominate. we can expect the concentration profile within the bed of a rotary kiln to be as follows: Figure. yielding concentration profiles shown here for a firstorder reaction Diffusion and dispersion will no doubt play a role in the kiln. the retention time of such gas will be the same. not discussed here. the residence time at any radius is equal. where several concentrations converge. For a halffilled kiln. 17 . In this part. we can expect a concentration profile fitting an exponential decay. Furthermore.Chapter 2 – Passive layer model 3. following the geometry of the kiln. For a halffilled kiln. until the gas reaches the surface of the bed again (Saeman et al. with k/ω = 1. 1951). this retention time for the halffull kiln will be the time it takes for half of a revolution: τ = π/ω. regardless of the radius of the arc. especially in the centre of the kiln.2 Nature of concentration profiles within the bed The movement of the gas in the voids is assumed to be a circular arc. Therefore. 4. If we assume that the gas following this path will move in plug flow. The inclusion of dispersion will require a more complex model.
and different reaction kinetics will simply determine the spacing of the lines. As an example. and the expected concentration profiles can be calculated. Definition of angles for the determination of isoresidence time lines. the residence time at each radius will be different. R) = = ω ω θ − sin −1 (h r ) ω (5) τ= (6) h = sin(θ − ωτ ) r (7) = y x cos ωτ − sin ωτ r r (8) y = x tan(ωτ ) + h cos(ωτ ) (9) This applies to any reaction kinetics. Figure 5. let us consider a 1st order reaction. θ − θ 0 θ − sin −1 (h r ) τ (θ .Chapter 2 – Passive layer model For bed fills of less than 50%. 18 .
Chapter 2 – Passive layer model PA = εkCA (10) where PA is the reaction rate of A per unit of bed volume and k is the reaction rate constant with respect to the gas phase. such concentration profiles can rather be expressed by an effectiveness factor. 3. 6.1. (4) for 1st order isoconcentration lines. Then C kτ = − ln A C A0 (11) Substitute τ into Eq. Concentration profiles for k/ω =1 at CA/CA0 intervals of 0. 19 . Figure 6. This factor is defined as the ratio of the actual rate of the reaction over the bed to the rate that would be obtained if the whole bed were exposed to gas at the highest gas concentration . shown in Fig.the concentration above the bed.3 Effectiveness factors For convenience.
z )dVbed ε kC A0Vbed θ R (14) 2 − (θ −sin −1 ( h r ) ) 2 η= 2 re ω r dr dθ ∫r R (θ 2 − θ 1 − sin (θ 2 − θ 1 )) θ1 ∫ 1 k (15) This function is represented graphically in Fig.) k θ 2 R − (θ −sin −1 ( h r )) PA ( r . z )dVbed = ∫ ∫ εkC A0 e ω r dr dθ dz ∫ θ1 r1 V (12) θ 1 = sin −1 (h R ) θ 2 = π − sin −1 (h R ) r1 = h sin θ (13) From the definition of the effectiveness factor and realising that the crosssectional area of the bed equals ½R 2 (θ 2 − θ 1 − sin (θ 2 − θ 1 )) . Later in the paper. we will discuss the role of other parameters that are dependent on the scale of the rotary kiln. θ . actual rate is the average rate over the entire crosssectional area. 7 for different bed fillings as a function of the k over ω ratio. η= ∫P A (r . The only variables in this model are the bed fill and k/ω. It is interesting to note that the size of the rotary kiln is irrelevant using this model. 20 .Chapter 2 – Passive layer model In the general case of kilns with a given fill. (See Fig.θ . 6.
0.5 (solid line).052.007 and 0.019.0017 (thin line).37. For the special case of a halffilled cylinder. 0.Chapter 2 – Passive layer model Figure 7. 0. 0. 21 . Taking a mole balance for a given volume of bed. 0. where CA is the concentration of gas leaving the bed. Effectiveness factors increase as bed fill decreases. 0. h = 0: π R 2 η= 2 πR ∫ ∫ re 0 0 k − θ ω dr dθ k (16) 2 = 2 πR = π R 2 −ω θ ∫ 2 e dθ 0 (17) ω kπ kπ − 1 − e ω (18) This is also easily derived from first principles. and are independent of bed radius R. As expected.14. the effectiveness increases for shallower beds.25. for a given k/ω. Fill fractions shown are 0.
Chapter 2 – Passive layer model PAdVbed = Q(CA0 . in the case of drying).05.CA) (19) ηεkCA0Vbed = Q(CA0 . CA/CA0 = 0.CA) C ηεkAbed L = Q1 − A C A0 kπ − ηεkπR 2 L εωR 2 L 1 − e ω = 2 2 (20) (21) (22) ω η= kπ kπ − 1 − e ω (23) 4. For high k/ω values such a simple approach is quite possible. In this case. For high k/ω. TOTAL DEPLETION MODEL of equation (15) is cumbersome to work with. C kτ = − ln A C A0 =3 (24) 22 . Choosing 95% conversion as the criterion. we assume complete reaction and use Eq. (4) to estimate the volume flow rate of entrained gas. a The double integral simplified approach for estimating η would be more useful. the gas leaving the system at the surface of the bed can be assumed to be fully depleted (or saturated.
all paths through the kiln have the same residence time. − PAVbed = QC A0 (28) ηεkC A0Vbed = εω 2 (R − h 2 )LC A0 2 ω (R 2 − h 2 ) 2 Ax (29) ηk = (30) 23 . we will choose ½ of the maximum angle. For lesser fill fractions. The angle subtended by this path is: θ= 1 2 cos −1 (h R ) 2 ( ) (25) cos −1 (h R ) τ =θ ω = ω (26) The criterion for using the simplified model is therefore k 3 > −1 ω cos (h R ) (27) Using an effectiveness factor and assuming total conversion.Chapter 2 – Passive layer model For a halffilled kiln. and τ = π/ω.
00169 1.98 0. k/ω ω η (Eq.164 0. 15) 2 h/R Fill fraction Ax (m . However.59 3.24 4.447 0.793 0.95 0.6 0.66 6.5 0.56 5.0209 0. 30) Error (%) 0 0. (1998) measured mass transfer in a rotating drum by measuring the evaporation rate of decane from impregnated alumina particles.2 0.1849 0.252 0.65 9.2259 0.45 14.173 0.2466 0.91 2.00532 0.1867 0.2047 0.4 0.052 0. basis R = 1m) η.2361 0. and only this set of data will be used here.8 0.63 5.3% fill).19 2.16 5.9 0.2338 0. The error caused by using the simplified model at the criterion values is less than 6% in all relevant cases.142 0.1 MODEL VALIDATION Literature data Jauhari et al.00666 0.2432 0.09 3.2354 1.164 0. 5.Chapter 2 – Passive layer model Table 1.33 4.97 0.2487 1. This gives rise to a geometry that is not well described by the model in this paper.2006 0.214 0.571 1.374 0.45 5.0587 0. total conversion model (Eq.2211 0.65 0. Most of the measurements were done using shallow beds with baffles (flights) fitted to the rotary kiln. one set of data was reported for a shallow rolling bed (4.2313 0.1667 0.97 2.0187 0. 24 .
out)Vbed (32) As mass transfer from the surface of the alumina particles is the rate determining step (as long as the pores are sufficiently filled).8 (34) 25 .out )Vbed Vbed QC decane . For that purpose. We should thus compare the ksA/Vbed data of Jauhari et al with the 6ηk gs (1 − ε ) / d p data that we can produce with our model using eq. defined as: ks A (Cdecane.sat –Cdecane.sat − Cdecane. we must find an appropriate value for kgs.Chapter 2 – Passive layer model Mass transfer rates were measured for various kiln conditions. Before doing so. The results were expressed in the form of a volumetric mass transfer coefficient ksA/Vbed. assuming an ideally mixed gas phase. QCdecane. follows from 6k gs (1 − ε ) d pε k= k gs a ε = (33) with kgs being the mass transfer coefficient at the surface of the alumina particles.out = ∫ PdecanedVbed = ηεk(Cdecane. we use the findings of Sφrensen and Stewart (1974) who studied mass transfer in packed beds. (15) or (30). the first order reaction rate constant (with respect to the gas phase). For a stagnant fluid they derived: Sh ⋅ D dp k gs = with Sh = 3.out = (31) We can make use of the ksA/Vbed data by comparing those with the equivalent data predicted by our model.
3 liters 0. (kgs): k 20. (34) is assumed to be appropriate.145 m 0.00284 m2 3 x 103 m 6 x 106 m2 s1 0. These conditions are summarised in Table 2: Table 2: Conditions used in our model.45 (assumed) 3.824 0.875 liters 0. Total volume : Bed volume: h/R: R: h: Crosssectional bed area (Ax): Particle diameter (dp): Binary diffusion coefficient of decane in nitrogen at 20°C (Dab): Interparticle voidage (ε): Sherwood number (Sh): Gassolid mass transfer coeff.Chapter 2 – Passive layer model As the gas entrained into the bed of a kiln is regarded to have no slip velocity with regard to the particles in the bed. have produced their ksA/Vbed data. If we now apply the precise conditions at which Jauhari et al.6 x 103 m s1 6k gs (1 − ε ) εd p 1 = 18.8 7. we are able to calculate the corresponding 6ηk gs (1 − ε ) / d p data with our model. eq.6 s Below we compare our results with the data measured by Jauhari et al: 26 . and the bed porosity is comparable to that of a packed bed.1195 m 0.
09).128 0.121 0.Chapter 2 – Passive layer model Table 3. meaning that saturation is easily reached within the bed.198 0.97 0.175 0. 0. (29) (i.604 612 301 144 106 90 0.0329 0.29 0. Present model (total depletion) o Data for 0.67 1. Fig. Rotation speed N (rpm) Angular velocity ω (s 1) Measured data ksA/Vbed (s 1) k (s 1) Total depletion model: ηεk (s1) 0.043 fill (Jauhari et. Comparison of measured data and model predictions.59 1.e. al. Mass transfer data measured by Jauhari et al is significantly higher than that predicted by our model.0164 0. 8.411 0.0933 0.0685 0.23 1. and that the mass transfer contribution by the convective model is practically insensitive to the calculated rate constant k. k/ω is much higher than the value of the criterion given by Eqn.031 0.062 0.502 0.206 0. 1998) 27 .1101 In all cases.
2 Measured data In order to test our model at conditions where kinetics are much slower. Like Jauhari et. In this way.Chapter 2 – Passive layer model 5. Figure 9. we performed similar experiments to Jauhari et al (1998). al.266 radians/s. Flow diagram of experimental apparatus The rotating drum used has a diameter of 0. Air was introduced through a rotameter and blown through a 1 mm jet directly away from the bed towards the circular wall of the rotating drum.041 m (Fig.15 m and a length of just 0. (1998). It was only operated in the rolling mode at two speeds: 0. allowing the evaporation to continue until the particles were dry. in order to ensure good mixing of the gas above the bed and to prevent any bypassing of gas to the central exit. hydrocarbon concentration was measured 28 .052 radians/s and 0. 9). It is made of steel with a polycarbonate observation window. but using larger particles. mass transfer within the pores of the particles became the rate limiting process in the final drying stage.
because the sensor was placed within the rotating drum. al. The experimental technique to measure mass transfer rates differs from that used by Jauhari et. At suitable intervals. The sensor was calibrated so that the sensor reading could be transformed to a linear hydrocarbon concentration reading.27. it was possible to calculate what fraction of hydrocarbon liquid had evaporated at any time. For the experiments reported here.Chapter 2 – Passive layer model continuously using a sensor. This allowed us to directly measure the buildup of concentration within the freeboard gas after the gas flow had been stopped. and the rate of increase of hydrocarbon concentration was measured until it came close to saturation. The characteristic rate of the system (ηεk) was calculated from the exponential rise in the hydrocarbon concentration as a function of time: ηεk (C sat − C )Vbed = V f dC dt (35) which can be integrated to give: C −C − ln sat C −C 0 sat Vbed = V ηεk ⋅ t f (36) 29 . the gas flow was stopped. we used spherical tabalumina particles of average diameter 16 mm ± 1 mm soaked in 1pentanol. The particles were wiped dry before being loaded in the rotating drum. near the gas exit. During the experiment. (1998). The fraction of liquid remaining in the particle (liquid loading. Because the hydrocarbon concentration and air flow rate were known at all times during a run. The 44 pellets that were used gave an average fill fraction of 0. air flowed through the rotating drum. z) has been used as a parameter to characterise the particle at a given set of conditions.
Mass transfer from bed to freeboard was calculated from the rate at which concentrations increased in the freeboard after gas flow was stopped. The saturation concentrations inferred with this technique did not vary substantially during the run.2 1 0.6 1. liquid loading was 0.6 0.266 s1) Several such measurements were performed during a run (with liquid loading of the particles decreasing continuously). min Figure 10.8 1.2 0 2 4 6 8 10 12 time. using both fast and slow rotation speeds. C −C − ln sat C −C 0 sat 2 1.2 0 2 0. except that we now added a shrinking core 30 .Chapter 2 – Passive layer model Csat could be accurately estimated by visually noting how the end points of C −C the graph of − ln sat C −C 0 sat vs t fitted the straight line formed by the other points.4 1.8 0. The saturation concentrations varied by less than 1% throughout a run.57 and ω = 0. 10. An example of such a graph is given in Fig. We used a similar technique to model the characteristic rate of the system (ηεk) as with Jauhari’s data.4 0. (Here. or when desorption within the particles becomes rate limiting. as can sometimes happen with small pores.
15) was inferred from the experimental data taken at the slowest drying rates.636 0. pentanolair (D) Bed porosity (ε) Particle porosity (εpart) Particle diameter (dp) Sherwood number (Sh) Value 8. The parameters used in the model are given in Table 4. (15). This enabled us to plot our model ηεk values with those that we obtained by experiment: 31 .016 m 3. so η was estimated using Eq. and the value of εpart (0.8 The total depletion model was not applicable at these experimental conditions. The term L is liquid loading. and εpart is the intraparticle porosity: 1 1 + 2 Sh ε part 1 1 − 1 3 z k= 6(1 − ε ) D 2 ε dp (37) The effective dispersion coefficient within the particle was estimated as 2 ε part D . Table 4.Chapter 2 – Passive layer model model to the resistance term. Parameter Binary diffusion coefficient.25 x 106 m2/s 0. Parameters used for the modelling of our data.15 0.
32 .1 using externally dry pellets are higher than expected.(0. possibly due to uneven drying on the surface of the pellets. but show some scatter.052 s1) The predicted mass transfer rates follow the trend of the experimental data quite well. Characteristic rate is well predicted by the model for both fast rotation x (0.02 < k/ω < 0. The experimental data for 0. it is possible to calculate the effectiveness factor η and k/ω at each of the experimental data points. 12) follow the expected trend quite well. and the values measured for 0. 37).266 s1).6 < k/ω < 2 were measured with wet pellets.Chapter 2 – Passive layer model Figure 11. and slow rotation ¯. These points (Fig. By using the model prediction of k at the various liquid loadings (Eq.
Active layer models are to be preferred in these cases. It then provides a useful alternative to active layer models. Effectiveness factors follow the predicted trend as a function model. 33 . it would be useful to include diffusion/dispersion effects.266 s1. For more general applicability. × ω = 0. which are clearly inappropriate if the whole bed contributes to mass transfer. for slower reactions (k/ω < 1) the presented model appears to perform much better. 12. 6.Chapter 2 – Passive layer model Fig. which will be done in a future paper. of k/ω. CONCLUSION A novel model to predict mass transfer inside a rolling rotary kiln was presented. This model only considers convection inside the voids between the particles inside the bed. which are assumed to be filled with entrapped gas from the freeboard. The model was found to underestimate mass transfer rates for fast reacting systems (k/ω > 1). ¯ ω = 0. However. showing that other mechanisms like dispersion and/or surface phenomena are also important for such systems.052 s1.
Chapter 2 – Passive layer model SYMBOLS a A Ax C D dp F g h pellet outer surface area per bed volume [m1] area [m2] radial crosssectional area of the bed [m2] concentration [mol m3] binary diffusion coefficient [m2 s1] particle diameter [m] solid fill fraction: bed volume/kiln volume gravitational acceleration [m s2] perpendicular distance from radial center of the kiln to the bed surface [m] k kgs ks L PA Q R r reaction rate constant [s1] mass transfer coefficient on the surface of particles [m s1] mass transfer coefficient of active layer [m s1] length of the rotary kiln in axial direction [m] reaction rate of A [moles m3(bed) s1] volumetric gas flow rate [m3 s1] radius of rotary kiln [m] distance from radial center of bed to the surface of the bed at angle 2 [m] s V Vbed Vf z half of the chord length formed by the bed surface [m] volume [m3] volume of bed [m3] volume of freeboard [m3] liquid loading in particles: volume fraction of pores filled by liquid 34 .
. Metall Trans. Gray. J. E. “Experimental study of residence time. & Watkinson. (1983b). 76. D. The modelling of transverse bed motion in rotary kilns”. Eng. 73. J. 207220 Jauhari. Hanrot. particle movement and bed depth profile in rotary kilns”.. & Masliya. M. A.P.. & Crookewit. Gassolid mass transfer in a rotating drum. Brimacombe.P. 1983. (1983a).K. Experimental study of transverse bed motion in rotary kilns.H. 1983. H.F. 14B... R. P. 14B. (1995).. Kramers. Trans. Can. Ablitzer. H. 259 Lebas. Chemical Engineering Science. H.. (1952). Eng. A.R.. Can. 173179.. J.L. Chem. F. J. & Watkinson..Chapter 2 – Passive layer model Greek symbols ε η τ θ φ ω bed voidage [m3(gas) m3(bed)] effectiveness factor of reaction residence time of gas in the bed [s1] angle [rad] gas flow rate [m3(gas) s1] angular velocity [s1] REFERENCES Heinen. Brimacombe. J. (1988).K. 1. 224232. 35 . Metall. 191205 Heinen. & Houzelot. Chem. J. The passage of granular solids through inclined rotary kilns..
R. C. Chem. Sørensen.. and Watkinson. D.G.. & Ralston. Stewart. JP. WE. & Scott. Eng.. Chem. 33.M. Technical Papers 384. The noflow problem for granular material in rotating kilns and dish granulators. “Passage of solids through rotary kilns”. 1951.Chapter 2 – Passive layer model Saeman.. 1974. Engineering Science.. S. J. WC. Can. 57. J. 55.D.F. (1979). 47. US Bureau of Mines.. Eng.H.. 36 .P.J. J. Convective heat transfer in a Rotary Kiln. 433443. Maier. Effect of fluid dispersion coefficients on particletofluid mass transfer coefficients in packed beds”. Eng.. O. 13741384 Spurling.C. Sci. (1927). 23032313 Chemical Sullivan. 142 Tscheng. Passage of solid particles through rotary cylindrical kilns. 508514. Prog. (2000). Davidson. Chem. A.
This can either be a Peclet number (ωR2/De) or a Thiele modulus (kR2/De)½. 37 . the stiff differential equations are not easily solved then. 2001). The solution of the 2dimensional partial differential equations that describe the extended model gives a handle on the effect of scaleup in rotary kilns. and other methods of solution are advisable. With high reaction rates. An additional dimensionless number is necessary then to describe the system. By doing so. For industrialscale kilns.Diffusion effects in rotating rotary kilns ABSTRACT A novel approach to the modeling of mass transfer in rotary kilns has been described (Heydenrych et al. isoconcentration lines are closely stacked near the surface of the bed. For fast reactions it underpredicted mass transfer. which means that diffusion within the lower (passive) layer of the bed is unimportant for slower rates.Chapter 3 – Diffusion in the passive layer Chapter 3 . It considers the mass transfer to occur by the inclusion of gas in the interparticle voids in between the particles that move concentrically with the kiln. the model will be extended to include diffusion effects. implying that it is important to model the active layer rather than the bed as a whole in these circumstances. However. Therefore in this paper. the rate of mass transfer was found to be dependent on bed fill and the ratio of reaction rate constant to angular velocity (k/ω). The model was found to be valid at slow to medium fast reactions. the Peclet number is large.
This approach takes into account the inclusion of gas in the interparticle voids in between the concentrically moving particles in the passive layer of the bed. a novel approach to the modelling of mass transfer has been suggested though (Heydenrych et al. Recently. In this paper we will explore the effect of diffusion within the concentric flow region. but that the effect of diffusion and dispersion within the bed must also be taken into account in the case of fast reactions. again under the assumption of an infinitely thin active layer.Chapter 3 – Diffusion in the passive layer 1. They concluded that their concentric movement model has merit. INTRODUCTION Traditionally. al. In that work. Shows the concentric movement and active layer regions 38 . mass transfer in rotary kilns has been modelled as a mass transfer coefficient at the surface of the bed of a rotary kiln (Jauhari et. the role of the active layer where particles move along the surface of the bed was ignored by assuming that the thickness of the active layer region is infinitely thin. 2001). and how it affects the overall mass transfer. Figure 1. 1998).
Assuming steadystate operation: ∂C A ∂C A ∂ 2C A ∂ 2C A + ε Vy + De + De + rA = 0 ∂x ∂y ∂x 2 ∂y 2 ε Vx (4) 39 . In the axial direction. we have chosen rectangular coordinates because of software restrictions. concentric flow region): ∂C A = −∇ ⋅ N A + rA ∂t (1) After substituting equations for the transport of gaseous A (Fick’s diffusion plus convection) into the above equation.Chapter 3 – Diffusion in the passive layer 2. the concentration profile of A follows from ∂C A = ∇ ⋅ (De ∇C A − c AεV ) + rA ∂t ∂C A + εV ⋅ ∇C A = De ∇ 2 C A + rA ∂t (2) (3) Sharp concentration gradients are expected in the radial crosssection of kilns [Heydenrych et al. consider the mass balance for a differential element within the bed (i. MATHEMATICAL MODEL To take diffusion into account. Although polar coordinates may be a more natural choice of coordinate system. 2000]. relatively low concentration gradients can be expected due to the high length to diameter ratio found in typical rotary kilns. This makes it a reasonable assumption to simplify the governing equation to the two dimensions that represent the radial crosssection.e.
s V x y V .Chapter 3 – Diffusion in the passive layer Here. for an nth order n reaction with − rA = εkC A : 1 ∂ 2Ψ 1 ∂ 2Ψ ∂Ψ ∂Ψ εkC A. al. For the bed surface. Zero flux at the solid walls gives a simple v v boundary condition: n ⋅ ∇C A = 0 where n denotes the normal vector at the kiln wall. a Peclet number modified for rotary kilns. Rewriting Eqn. ωR ωR R R ε ωR 2 . Both are assumed to be constant. υ y = y . C A. it follows from the concentric movement geometry that υ x = λ y and υ y = −λ x using the coordinate system given in Fig. 40 . υx = x . See the Appendix for more detail. and are described by the concentric motion of the bed. De n kC A−s1 . Noting that υx is simply the xcomponent of the velocity divided by the velocity at the wall.s n + +υx +υy + Ψ =0 2 2 Pe ∂λx Pe ∂λ y ω ∂λ x ∂λ y n −1 (5) where Ψ= λx = Pe = CA dimensionless concentration. λy = . The reaction term ω is similar to the k/ω term used by Heydenrych et. Vx and Vy are the x and y components respectively of the velocity of particles (and gas) in the bed. De is the effective diffusivity and e is interparticle voidage. 2. discussed later in the paper. (2001) to describe concentration terms within the bed of a rotary kiln in which a first order reaction takes place. various boundary conditions can be used as appropriate. 4 in terms of dimensionless variables.
We assumed that the active layer was infinitely thin. ∂λ x ∂λ y Rω (7) The noflux condition at the wall is given in dimensionless form by nx ∂Ψ ∂Ψ + ny ∂λ x ∂λ y =0 λ =1 (8) 41 . In addition to the noflux condition at the wall. the Danckwerts boundary condition for nonideal flow in a plug flow reactor was used as the boundary condition at the bed surface: ∂C A ∂C A ∂x + ∂y C A. s = C A − De ε Vx (6) The condition may also be written in dimensionless form as V ∂ψ ∂ψ + − Peυ xψ = − Peυ x where υ x = x .Chapter 3 – Diffusion in the passive layer Figure 2: Applied coordinate system.
any two of the dimensionless groups Pe. in this case. This means that the system can be described by n kC A−s1 . 5). changes significantly along the length of the kiln. 42 . but only the geometry of the boundary conditions. ω . For future reference. The model described here (Eqns. In this paper we have chosen to use ω because it is the only variable that is required to The rate constant k typically describe the nodiffusion case. ω . 5. and Pe. Although it would be possible to model nth order reactions using the same toolbox. we have restricted our analysis to 1st order reactions. Fraction fill is the third independent dimensionless variable that is necessary to describe this model. Note that the square of the Thiele ω . we will refer to the second dimensionless variable in its form for firstorder reaction. Solution of the model This model was solved using Matlab’s partial differential equation toolbox. n kC A−s1 . φ or n kC A−s1 . It is also interesting to note that the fraction fill does not affect the governing equations (Eqn. k/ω. 7 and 8) is dependent on two dimensionless variables: Pe and modulus φ = Pe ⋅ 2 n kC A−s1 . as the solid phase approaches full conversion (or total drying). Pe only needs to be calculated once for a given kiln because it does not typically change along the length of a rotary kiln.Chapter 3 – Diffusion in the passive layer where nx and ny are the components normal to the wall confining the particulate solids.
η = st ∫ψ dA A Abed .Chapter 3 – Diffusion in the passive layer The solution gives the concentration profile within the bed of the kiln. S V η= − ∫ PA. 43 . 20% and 10% filled kiln that were generated using the model are shown in figures 3 to 6. and the effectiveness factor of the bed is calculated from that concentration profile.s is the reaction rate that would occur if the entire cross section of the bed was exposed to the conditions at the surface of the bed. S Abed Abed where –PA is the local reaction rate in the bed and –PA. The curve for an infinite Peclet number represents the nodiffusion case reported earlier by Heydenrych et. (2001). The effectiveness factor of a rotary kiln is defined analogously to the effectiveness factor of a catalyst pellet. − ∫ PA dV V n − ∫ ε kC A dV V n − ε kC A. (7) where area A is calculated in terms of λx and λy. 3. For 1 order reactions. RESULTS The effectiveness factor curves for a 50%. 5) is the concentration profile over the bed. S dV V = = ∫C A n A dA = ∫Ψ A n dA (6) n C A. the effectiveness factor can be determined from the average dimensionless concentration over the bed cross section. Given that the solution of the partial differential equation (Eq. al.
Peclet numbers are typically greater than 1000. kiln size and rotation speed). is that the numerical methods used to solve the model were based on techniques suited for an elliptical (secondorder) partial differential equation. For commercial scale rotary kilns. the secondorder terms of the partial differential equation become less important. and the numerical method is unsuited to the solution of what effectively becomes a firstorder hyperbolic pde that dominates the model. 1st order reaction.Chapter 3 – Diffusion in the passive layer Figure 3: Effectiveness factors as a function of k/ω for a 50% filled kiln. The dashed lines on Figure 3 are extrapolations. the effect of diffusion as a fraction of the contribution by gas inclusion. It should also be noted that for a given Peclet number (i. where high concentration gradients occur in the bed (predominantly near the surface of the bed). A possible reason for the poor convergence of the model at large Peclet numbers. 44 . At high Peclet numbers.e. with the firstorder term simply being handled as a source term. the inclusion of diffusion effects in the model increases the effectiveness factor in all conditions. discussed later in the paper. As expected. where the model failed to solve. is far higher at high values of k/ω. the numerical method was unable to converge on a solution. At higher Peclet numbers (Pe > 10).
Chapter 3 – Diffusion in the passive layer Similar solutions have been found for kilns with bed fills of 20% and 10% (Figs. The concentration profiles generated at certain Pe and k/ω values are interesting. 6 for a halffilled kiln: 45 . Effectiveness factors as a function of k/ω for a 20% filled kiln. 4 and 5 respectively) Figure 4. and are shown in Fig. As the fill fraction decreases. Figure 5: Effectiveness factors as a function of k/ω for a 10% filled kiln. 1st order reaction . the effect of diffusion appears to become more important. 1st order reaction.
The straight isoconcentration lines predicted for the nodiffusion case (Heydenrych et. Fig. are only readily evident for the diagram 46 k/ω = 100 k/ω = 10 k/ω =1 k/ω = 0.Chapter 3 – Diffusion in the passive layer Pe = 1 Fill = 10% Fill = 20% Fill = 50% Figure 6: Concentration profiles at Pe = 1. al. 1st order reaction. 2001).1 . 7.
al. and what Peclet numbers are typical for rotary kilns? The diffusivities of a gas at atmospheric pressure vary between 104 106 m2/s (Knudsen et. and the rotation of the kiln does not distort the profile significantly. Figure 7. 1997) τ 47 . If diffusion is not taken into account.Chapter 3 – Diffusion in the passive layer where k/ω = 0. For the other cases. diffusion effects are important. al. 4.. 2 x 105 m2/s and 2 x 106 m2/s were used to for the diffusivity of the reactant gas DA. Values of 2 x 104 m2/s. diffusion effects dominate the shape of the concentration profile. But under what circumstances can we expect such low Peclet numbers.. straight isoconcentration lines are expected. DISCUSSION Scaleup effects The models have shown that for Peclet numbers less than 100. 1997). The effective diffusivity in the bed of the kiln can be estimated as follows: De = D Aε (Knudsen et.1.
with a rotating drum of radius 0. De. 0.5 1 5 5 x 105 0. we can expect effective diffusivities within the bed of 5 x 105 m2/s (max. only two angular velocities are considered. slumping of the bed can occur. al.5. For this reason.05 radius (m) 0.1 rad/s. Kiln rotation speeds vary less.5 2500 10000 250000 1000000 50000 250000 500000 2500000 5000000 25000000 As an example. At higher rotation speeds.145 m. Cells with Pe < 1000 are not shaded to indicate that this only occurs in laboratoryscale kilns with high De values.). 5 x 106 m2/s and 5 x 107 m2/s (min.1 0.Chapter 3 – Diffusion in the passive layer Using a tortuosity (τ) of 2 and voidage (ε) of 0. because Pe ∝ R2.5 rad/s and 0. varies from 530 to 3600. The effect of scale is largely dependent on the radius. Table 1.1 5 20 500 2000 0. Table 1 shows Peclet numbers for selected De and ω values as a function of radius. It is obvious that diffusion as a possible extra transport mechanism to gas entrapment can safely be neglected for practical applications as well as for most lab experiments. 48 . the Peclet numbers for the rotating drum used by Jauhari et.). m2/s ω. because at lower rotation speeds.1 50 200 5000 20000 5 x 106 0.5 25 100 2500 10000 0.1 500 2000 50000 200000 5 x 107 0. Peclet numbers for typical rotation speeds and effective diffusion coefficients.5 250 1000 25000 100000 0. depending on rotation speed. centrifugation begins. rad/s 0. (1998).
2001) by adding diffusion effects to gas entrapment in the concentric flow region of a rotary kiln. The Peclet number provides a tool for predicting scaleup effects. 2001 have shown that k/ω is the parameter that describes mass transfer at low to moderate reaction rates. Peclet number was implemented (in addition to k/ω) to model diffusion effects. and how the movement of particles over each other affects the dispersion of the gas in the direction perpendicular to the bed surface. This number is defined as ωR2/De. This information will give the boundary 49 . CONCLUSIONS A mathematical model has been presented that improves on the previous model (Heydenrych et al. At high k/ω values. but not the model. For practical values of Pe. and here the assumption that the active layer is infinitely thin becomes a poor assumption as all mass transfer is occurring within this layer. Heydenrych et al.Chapter 3 – Diffusion in the passive layer 5. a modified solution technique is required. and even though the model presented here is oversimplified due to the assumption of an infinitely thin active layer. is limited to moderate Pe numbers. Further work should be concentrated at developing a suitable model for the active layer. In this paper. The technique. based on the results of this work. This model shows that the entrapment of gas perpendicular to the bed is relatively unimportant compared to diffusion effects at these higher k/ω values. concentration gradients become fully concentrated near the surface of the bed. it nevertheless provides an insight to the use of the Peclet number for the scaleup of rotary kilns. It will also be important to determine the depth of the active layer. where a 1dimensional dispersion model would be a fair assumption. however. Diffusion in the concentric flow region was found not to add much to the mass transfer at realistic conditions where the Peclet number is much higher than 1.
s De Pe R r V x y Cross sectional area of the bed of the rotary kiln [m2] Concentration of reactant A in the bed of the rotary kiln [kmol/m3] Concentration of reactant A at the surface of the bed [kmol/m3] Effective diffusivity of reactant A [m3/s] Peclet number modified to suit a rotary kiln Radius of kiln [m] Reaction rate of a reactant A [kmol/m3/s] Velocity of gas within the bed [m/s] Coordinate in x direction Coordinate in y direction Greek Letters ε φ η λ τ υ ω ψ Voidage of the bed of the rotary kiln Thiele modulus Effectiveness factor Dimensionless distance Tortuosity Dimensionless velocity Angular velocity of kiln [rad/s] Dimensionless concentration 50 .Chapter 3 – Diffusion in the passive layer conditions for the concentric flow region. with the aim of generating a comprehensive model including all mechanisms for mass transfer. SYMBOLS A CA CA.
.R. 224232. R. Can. J. Greeff. Sarofim.M. McGrawHill.B. Industrial and Engineering Chemistry. Sci.W. ‘Mass transfer in rolling rotary kilns: A novel approach’. K. 51 to 579. Jauhari. J. Heesink. Submitted to Chem.. Wankat. H.. G. (1939) ‘Relation between catalytic activity and size of particle’. & Masliya.D... and Knaebel... New York. A. A.F. 76. Thiele.F. Gray. Hottel.). P.C.W. Gassolid mass transfer in a rotating drum.C. E..H. Knudsen J.Chapter 3 – Diffusion in the passive layer REFERENCES Heydenrych. 916920 51 . (2001). Versteeg. 31 (7). Eng. M.S..G. P. 2001. Green (Ed. Seventh Edition. Chem. D. (1997) in Perry’s Chemical Engineer’s Handbook. M. (1988). Eng.
Chapter 3 – Diffusion in the passive layer
APPENDIX
Proof that υx = λy and υy = λx
Consider a point on the radius, which is at λ = 1 in dimensionless coordinates. By definition,
υx =
Vx ωR
Substituting for Vx,
υx =
ωR sin θ = sin θ ωR
From the diagram,
λ y = sin θ = υ x
At points away from the wall, υx and λy decrease proportionately. It can similarly be shown that υy = λx.
52
Chapter 4 – Active layer models
Chapter 4  Hydrodynamics and dispersion in the active layer of rolling rotary kilns
ABSTRACT
In previous works (Heydenrych et. al. 2001a,b), it was shown that the top layer (or active layer) of a rolling bed is the only active place in the case of a fast reaction. The bottom part of the bed (also referred to as the
concentric flow region) is almost depleted and therefore hardly reacting. Therefore in this work we focus on that active layer. A hydrodynamic model is prepared to predict its shape. A mass balance on particles moving within a rotary kiln gives the shape of the active layer, defined as that part of the bed where particles move past each other. The derivation is based on the assumption that there is a linear velocity gradient of particles in the direction perpendicular to the surface of the bed that is constant at any place on the bed surface. Published data was used to establish a
relationship between this velocity gradient on the one hand, and bed fill and rotation speed on the other. A dispersion model was derived to predict the mass transfer of gaseous reactant within the active layer. As void spaces are created and destroyed due to the action of particles moving over each other, mixing of gas occurs. A dispersion coefficient was derived, based on this gas mixing, and was found to be proportional to the velocity gradient and to the square of particle diameter. Applying these relationships to a 1dimensional concentration gradient model gave a correlation for the mass transfer coefficient at the bed surface. This model does not correctly
predict the proportionality of mass transfer with bed speed reported in the literature. We conclude that another effect, probably the granular
temperature of the particles, is important for modeling of dispersion and mass transfer in the active layer of rolling rotary kilns.
53
Chapter 4 – Active layer models
1.
INTRODUCTION
The movement of particles in a rotary kiln is shown in Figure 1. Particles move concentrically with the wall of the kiln (passive layer) until they reach a point near the surface, at which point the particles slide down the top surface of the bed. This differentiation of the two zones within a rotary kiln has been recognized from the earliest published data on rotary kilns (Sullivan et. al. 1927, Saeman, 1951).
Fig. 1. Particles slide over each other to the lower half of the bed in the active layer.
This motion of particles relative to each other leads to much improved heat and mass transfer rates compared to what one would expect from a stationary bed (Heinen et. al. 1983).
Traditionally, heat and mass transfer rates have been modeled using a coefficient based on the surface area of the bed (Tscheng & Watkinson, 1979). Recently, Heydenrych et. al. (2001) have shown that in certain
54
Chapter 4 – Active layer models cases. HYDRODYNAMICS OF THE ACTIVE LAYER 2. and concentration gradients in this layer are much less than one would predict by the diffusion model presented by Heydenrych et.1 The shape of the active layer The movement of particles sliding over each other in the active layer causes dispersion of the gas in the voids. In this paper. With these models. and to predict the dispersion that can be expected. Those models were developed for mass transfer. and in this paper we will continue to model mass transfer as an example of transfer processes that occur in rotary kilns. the active layer. al. If we take any line from a point on the bed surface to a point on the wall within the bed. 2. (2001). and the volume of the bed can also become important. the mass on either side of that line must remain constant in 55 . we estimate concentration profiles and finally calculate the effective mass transfer coefficient (based on bed surface area) for the active layer for a given rotation speed. we explore the physical processes that occur within the active layer in order to model the depth and shape of the active layer. The most important parameter for predicting the degree of dispersion will be the gradient of the velocity profile near the surface of the bed. particularly for slow processes. This dispersion effect is typically much greater than the effect of gaseous diffusion in the active layer. particle size and bed depth.
3. 2000) how the choice of materials affects these assumptions. 4 to remain constant.. It would be useful to understand through granular dynamics simulations (Hoomans. They also found that this gradient does not change along the xdirection in Fig. 1. Mass balance over envelope described by 1234 The volume flow rate out of the control volume (using solids in this case) has earlier been shown (Heydenrych et. We will make this assumption to calculate the volume flow rate into the control volume through the envelope extending from 1 to 2 in Fig. We will use a line extending from 1 to 2 to 3 on Fig. and found that for certain materials. the gradient of the velocity profile is substantially linear in the direction perpendicular to the bed surface (12) on Fig. 2. 1. we expect the mass within the volume 1. 2001) to be: Q= ωL(1 − ε ) 2 (R − x 2 − (h + d ) 2 ) 2 ( 1) Boateng & Barr (1997) have measured the velocity profiles in the active layers of rotary kilns. (1).Chapter 4 – Active layer models time. 56 . Figure 1. al. 1.
Chapter 4 – Active layer models Firstly. using the average velocity between these two points: (1 − ε ) Ld (υ S + υ d ) 2 Q= ( 5) Eliminating Q using Eqns. ( 3) υ S = dm − (h + d )ω ( 4) The volume flow rate of solids between points 1 and 2 can now be calculated. let us define the gradient of the velocity profile (called shear rate by Boateng and Barr (1997)) as the variable m. defining the velocity of particles at the surface and at depth d as υs and υd respectively. (1) and (5). Also. it follows that υ S − υ d = dm ( 2) and assuming that υd is determined by the xcomponent velocity of the passive layer. υ d = − rω sin θ = −(h + d )ω Substituting for υd into Eq. we get: υS + υd = ω 2 (R − x 2 − (h + d ) 2 ) d ( 6) 57 . (2).
h. This constraint can be represented by d(x=0) < R – h ( 8) Substituting Eqn. Their measurements indicate that for a given material and 58 . if (as the data suggests). If we choose a linear relationship for m = bω. m = bω+ c. then the only requirement is that c>ω. (4). (6) for maximum active layer thickness at x = 0. (b is a constant). (4) for υS and simplifying. υS can be determined as a function of x from Eq. we can show that this constraint simplifies to m 2R > ω H ( 9) The relationship between m and ω has been measured by Boateng & Barr (1997). A constraint that is not as easily met is the constraint that the active layer depth d must be less than the bed depth H. this constraint is easily met. or b>1. In reality.Chapter 4 – Active layer models Substituting Eq. They measured particle velocities at various depths using an optical probe. It means that m has to be greater than ω to have any physical significance. Also. where H = R . it means that b must be greater than 1. d= R 2 − x2 − h2 m /ω −1 ( 7) Note that with d and m/ω known. The m/ω term is interesting. Literature data suggests b is about 15. (3) for υd and Eq.
(9) also follows a linear relationship with ω.Chapter 4 – Active layer models bed depth. 2. The linear relationship m = 19ω + 7. The data shows some scatter. Noting that the constraint for the value of m imposed by Eqn. 2000) might shed some light on where the true relationships lie. We will look for a relationship describing m not just as a function of rotation speed ω. 3 59 . but also as a function of bed fill. and simulation using granular dynamics (Hoomans.5 is shown by the solid line. We have plotted the data of Boateng & Barr (1997) against R/H in Fig. Data points marked PE represent polyethylene beads. based on the experimental data. it would be interesting to see if m also follows a linear relationship against R/H. Figure 2. m increases linearly with ω. Their data for beds consisting of polyethylene beads and limestone are shown in Fig. and LS represents limestone.
The intercept is apparently independent of rotation speed. there appears to be a linear relationship.5 rpm ««««r«««« 3 rpm o 2 rpm The data appears to correlate reasonably well with R/H.¯. In fact. and represents the shear stress that will result from particles sliding down the surface of an inclined chute with an infinitely deep bed. and will depend on particle characteristics only. indicating that the shear stress is indeed dependent on R/H. In this case.5 s1. This model fits the data for R/H > 3 very well. and constant m as flow increases – i.Chapter 4 – Active layer models Figure 3. one can expect a layer of finite thickness. In fact for shallower beds (high R/H). (9). 60 .. Shear stress m shown as a function of R/H for polyethylene particles. . with c = 9.3.e. a suggested relationship for small bed depths m = 2ω R +c H ( 10) is shown on Fig. and the slope of the lines correlates well with the slope of 2ω suggested by the criterion of Eqn. bed depth proportional to flow rate of particles.
(10). d ∝ ω0. As rotation speed increases. geometric constraints and perhaps centrifugal forces may affect the shear stress profile. 61 . 4. This means that at low rotation speeds. In this case. at any given kiln rotation speed. d∝ ω (2 R / H − 1)ω + 9. we can determine the shape of the active layer at any point in the active layer. A further implication of the bed height expression. Our model does not predict the shear stress well at lower R/H values (fuller beds). (7).Chapter 4 – Active layer models Noting from the data of Boateng & Barr (1997) that a typical value for shear stress m/ω is 10. 2. Eqn.2%. Substituting Eqn. In terms of bed fill. 9) will become important when R/H < 5. d becomes less dependent on rotation speed. it means that for bed fill less than 5. (7). is that d∝ ω . (10) into Eqn.5. m −ω Substituting for m using Eqn.5 . we conclude that the constraint (Eqn. The active layer profile predicted by this model for various bed depths and rotation speeds is shown in Fig. m must be larger than the narrow range expected from the data shown in Fig.
Fig.5 and 0. 5 shows how the measured data at the deepest part of the active layer compares with our model for predicting bed depth (at x = 0).8.Chapter 4 – Active layer models Figure 4. 0. Active layer profile is an elliptical shape.1 s1 dotted and ω=0. 62 . Heinen et.2. al. (1983) have published a good deal of data showing observed active layer thickness as a function of bed depth and rotation speed. The depth of the active layer does not increase linearly with rotation speed (ω=0. shown here for H/R values of 0.2 s1 dash shown).
To calculate the volume of the active layer. al. it is useful to know the volume of the active layer. Our model predicts a deeper active layer than visually measured data at x = 0. (1998) presented a model for predicting mass transfer rates based on the volume of the active layer. (1983) reported rotation speeds within limits.2 The volume of the active layer In some cases. al. 63 . rather than the exact values. using our model. 2. it is useful to show that. the shape of the active layer is described by an ellipse with the central axis located on the bed surface.Chapter 4 – Active layer models Figure 5. Some of the scatter in the data can be attributed to the fact that Heinen et. We used the average values within those limits. and Jauhari et. probably due to wall effects that were confirmed by Boateng & Barr (1997).
m 2 2 2 2 − 1d = R − x − h ω Substitute for d2 using the relationship y = . 6) is described by ( y − c) 2 x 2 + 2 =1 b2 a ( 11) Figure 6.Chapter 4 – Active layer models An ellipse (Fig. We would like to rewrite the expression for active layer depth in the form shown in Eqn. (11).d – h ( y + h) 2 + x2 = R2 − h2 m ω −1 ( y + h) 2 x2 + 2 =1 (m ω − 1) R 2 − h 2 R − h 2 ( 12) ( 13) ( ) ( 14) 64 . Definition of parameters describing an ellipse.
where successive layers of particles are separated by a distance 65 . Consider Figure 7. it implies that there is always a point where the active layer theoretically extends beyond the bed (for h < 0. 3. this is usually insignificant. DISPERSION LAYER WITHIN THE ACTIVE As particles slide over each other. all m/ω).Chapter 4 – Active layer models Writing in terms of the dimensions of Fig (6). Therefore the volume of the active layer is π πL(R 2 − h 2 ) Val = ab ⋅ L = 2 2 m ω −1 ( 15) Consider the points on the ellipse where y = h. and should not detract from the usefulness of the model. Because the slope of the wall is finite for h < 0. For practical purposes. The slope of the tangent = ∞ for all h. assuming that a concentration gradient exists. the size of interparticle voids changes. the area of the active layer is half of the area of the ellipse. a = R2 − h2 b = R2 − h2 c = −h m ω −1 Because the axis of the ellipse lies along the surface of the bed. This gas moves to adjoining voids. where it mixes with gas at another concentration.
where we need to determine dispersion coefficient D. Noting that the top and bottom layers are separated by a vertical distance of 2dp. Description of voids in one cycle of particles sliding over each other. the top layer of particles moves horizontally by a distance 2dp relative to the bottom layer. then the movement of 66 . In a full cycle. Start of cycle ½cycle full cycle Figure 7. dC dy WA = −D ( 16) Noting that at steady state the concentrations remain constant at a given x. Assume also that a steadystate concentration gradient exists with molar flux WA. and therefore will have a minimum value of 1/6. the time it takes for this cycle will be 1/m seconds (where m is the velocity gradient.Chapter 4 – Active layer models approximately equal to the particle diameter dp. in m/s per m active layer depth). a nett mole balance from position 1 to position 2 over a full cycle gives: N = C1 f (V1 − V2 ) − C 2 f (V1 − V2 ) ( 17) The parameter f is used to take into account that the gas can move in any of 6 orthogonal directions. If the particles slide over each other as rigid layers.
default value. and from the definition of molar flux WA. (22): 2 D = d p mfε 1 (1 − α ) ( 23) α can be determined from the ratio of the voidage of loosely packed spheres vs tightly packed spheres.) ( 20) A A W A = d p mfε 1 (1 − α )(C1 − C 2 ) ( 21) dC C1 − C 2 = dy dp ( 22) D is determined by dividing Eqn. Substitute α = V2/V1 – this will later be estimated from the change of voidage from a loosely packed solid to a tightly packed solid. WA = mfV1 (1 − α )(C1 − C 2 ) A ( 19) V1 ε 1d p A = = ε 1d p (ε1 is voidage for loosepacked solid. 67 . WA = FA/A. It follows then that the molar flow rate FA = Nm. N = V1 f (1 − α )(C1 − C 2 ) ( 18) The time required for one cycle is 1/m. and f could have a value as large as ½ We will use f = 1/6 as our . (21) by Eqn.Chapter 4 – Active layer models the gas will primarily be in the ydirection (perpendicular to the bed surface).
This diffusion contribution De is commonly estimated by De = ε2Da. For the purposes of this paper.ε1 = π/6 ε1 = 0. we will assume that the gas is fully depleted at the boundary of the 68 .32 2 D = 0. CONCENTRATION MASS TRANSFER PROFILES AND Having established the dispersion coefficient applicable to the active layer. and the tortuosity of the path.246 fmd p ( 26) The effect of binary diffusion must be adjusted to take account of the reduced area for diffusion due to the solid particles.4764 For tightpacked spheres. The overall dispersion coefficient for the active layer Dal will be the sum of D and De: Dal = 0.246fm dp2 + ε2Da ( 27) 4.Chapter 4 – Active layer models ε1 = V1 VS + V1 ⇒ VS 1 = −1 V1 ε 1 ε2 ε 1 ( 24) 1 −1 1 − ε1 V2 ε1 =α = = 1− ε 1 V1 2 −1 ε2 ( 25) For loosepacked spheres. ε2 = 0. we now need to establish the concentration profile that exists within the active layer in order to predict mass transfer. 1 .
we will find it convenient to express the reaction rate for the system in terms of a mass transfer coefficient at the bed surface ks: 69 . then wall effects can be neglected. Eq. (29) reduces to η = 1/φ ( 30) For this model.Chapter 4 – Active layer models active and passive layers. needing to assume that there is no flux at the bottom of the active layer due to depletion of the gas (we do not have symmetry). If the thickness of the layer containing undepleted gas is much smaller than the width of the bed surface. with reaction only occurring at the surface of the bed. Thiele (1937) showed that the concentration profile in a flat plate can be determined (using our notation) as εk e λφ + e − λφ Ψ= φ . In this case. In our case. where φ = d −φ Dal e +e ( 28) and the effectiveness factor is then tanh(φ ) φ η= ( 29) The above equations (28) and (29) are valid for d being measured from the catalyst surface to the center of the plate. This analysis has already been done by Thiele (1937) for diffusion in a catalyst. and we have the situation of gas dispersion and reaction in a flat plate. where there is no flux at the center due to symmetry. we have assumed more stringent conditions.
al. and substitute Eqn. s ⋅ Val ( 31) Neglecting end effects. al. by multiplying Eqn. whereas our model predicts a much smaller dependency on rotation speed.Chapter 4 – Active layer models Val ∫ P dV = k A s AC A. Sh = 3. (33) by bed surface area A. (1998) reported their mass transfer coefficients as ksA. d = Val/A. al. (1998). although the predicted values are in the correct range. We note that the results of Jauhari et. enabling us to make a direct comparison of our model with their results.3%. (29) to get k s = ηεkd = Dal εk ( 32) The rate constant εk can be calculated (Heydenrych et. 70 . measured at a bed fill of 4. (31) in terms of ks. MODEL VALIDATION Jauhari et. 8). 2001) using 6(1 − ε ) Sh ⋅ Da . s = ηεkC A. Then we can rewrite Eqn. are substantially proportional to rotation speed (Fig.8 2 dp εk = ( 33) 5.
al.5 s1). (10) to estimate m. This could lead to particularly low slip velocities on the outer surface of the bed at the slower rotation speeds. Clearly. The second possible reason for deviation from our model could be due to the nature of the movement of particles in the active layer. In their measurements. where the freeboard gas slip velocity becomes important. The first is that mass transfer at the top surface of the bed. (33) we calculated εk = 8. Jauhari et. There may be two reasons for this. this model does not accurately predict mass transfer rates for the conditions used by Jauhari et. f = 1/6. (1998). Our model does not correctly predict the linearity of ksA with rotation speed.36 s1. has not been taken into account in this model. dp = 3 mm. results in a similarly low dependency of Dal on rotation speed. Our model 71 . Using Eqn. with c = 9. This could lead to the lower than expected mass transfer rates that were measured at rotation speeds lower than 1 rpm.Chapter 4 – Active layer models Figure 8. (1998) operated under the constraint that the superficial axial gas velocity was less than the drum wall velocity. but predicts values in the correct order of magnitude. The very small dependency of m on rotation speed (using Eqn. al.
The constant c is expected to depend on the properties of the bed material. where void spaces are filled and formed again. which means that Dal changes with bed depth. By examining the effect of particles sliding over each other within the active layer.Chapter 4 – Active layer models assumes that the particles slide over each other. 6. creating and destroying voids in an orderly fashion. DISCUSSION AND CONCLUSIONS A relatively simple model has been derived to predict bed depth and the surface velocity of particles. Moreover. The frequency of these collisions may be much higher than the frequency predicted by our model. A correlation was presented to predict the shear stress within the active layer as a function of bed depth and rotation speed. based on the assumption that the shear stress in the active layer is constant for a given bed depth and rotation speed. data for polyethylene beads gives c = 9. the granular temperature measured by Boateng and Barr (1997) is dependent on depth (d) within the active layer. the mass transfer coefficient for the bed surface can be calculated as k s = Dal εk . m = 2(R/H)ω + c. Our model predicts that as long as the reactant does not occur deeper than the active layer in significant concentrations.5 s1. a correlation was established for prediction of the effective dispersion coefficient Dal. 72 . Boateng and Barr (1997) have measured the granular temperature of particles in the active layer: a term that describes the random movement of particles due to collisions within the active layer. It may also affect the mass transfer resistance at the surface of the bed. giving a different dependency of ksA on rotation speed.
(1998). 73 .Chapter 4 – Active layer models These attempts to fundamentally predict the mass transfer coefficient were not successful. al. Future work should be directed at a means of predicting ksA based on granular temperatures within the active layer. based on the data measured by Jauhari et. The most probable reason for the poor correlation of our model with experiment was the fact that neither bed surface effects nor the granular temperature of the bed material were considered in the determination of the effective dispersion coefficient De.
m volume. molar flux. m2 gasphase concentration. m gasphase reaction constant. m shear stress in the active layer (velocity gradient of particles) s1 reaction rate. m3 factor for the directionality of dispersion. m s1 axial distance. b. c symbols used to describe an ellipse (Fig. m2 s1 total dispersion coefficient for the active layer. mol m3 gasphase concentration at the bed surface. m3 s1 kiln radius. 6) CA CA0 CA. mol m3 gasphase concentration at t = 0. m binary diffusion coefficient. m H k k’ ks L m PA Q R V f W bed depth at the deepest point (H = Rh). m2 s1 perpendicular distance from radial center of the kiln to the bed surface. s1 gasphase reaction constant for use in active layer. mol s1 m3 volumetric flow rate. s1 mass transfer coefficient. mol s1 1 1 < f < 6 2 a.s d D Dal h 74 . mol m3 active layer depth.Chapter 4 – Active layer models SYMBOLS A bed surface area.
. A. Eng. ‘Diffusion effects in rotating rotary kilns’.P. Trans. Heesink... Heesink. A. dimensionless concentration.P. 233249 Heydenrych. Brimacombe. Greeff. J. ‘Granular dynamics of gassolid twophase flows’. M. (2000).. Experimental study of transverse bed motion in rotary kilns. dimensionless bed voidage. 2001. 330. 1983. J.A. Heydenrych.B..K. (2001). (1983a). A. Schieke.A. 75 ..M.. ‘Granular flow behaviour in the transverse plane of a partially filled rotating cylinder. J. depth below bed surface/active layer depth thiele modulus..B. G.D.B. P.M. 2001.F. & Barr. dimensionless REFERENCES Boateng. Versteeg. Doctoral thesis. Accepted for publication at the 6th World Congress of Chemical Engineering.Chapter 4 – Active layer models Greek letters η ε λ φ ψ effectiveness factor. Fluid Mech. H.M.V.. Heinen. ‘Mass transfer in rolling rotary kilns: A novel approach’. 14B.D. Kuipers. (1997). J. A.. Melbourne. University of Twente... Sci. P. (2001). dimensionless dimensionless distance. B. Metall. & Watkinson. Submitted to Chem. M. 191205 Hoomans.
..H. S. Maier.C. WC. H. (1995). 173179. J. Chemical Engineering Science. D.. (1952). Chem. 1. C. & Ralston. J. particle movement and bed depth profile in rotary kilns”. & Houzelot... 142 Tscheng. Technical Papers 384. Eng.R. 259 Lebas. (1979). Eng. Gray.. J.. E. & Crookewit. “Experimental study of residence time. Eng. 433443. J. “Passage of solids through rotary kilns”. 73. Chem. (1988). Can.. 57. and Watkinson. O.D.G. F. Saeman. J. R. Can. Can. Gassolid mass transfer in a rotating drum.P. Eng.. US Bureau of Mines.. 224232.L. A. Passage of solid particles through rotary cylindrical kilns. (1927). Kramers. The passage of granular solids through inclined rotary kilns. 508514. 76. Chem. Prog. Chem. & Masliya. 76 .H. J. 47.F. P. Ablitzer. Convective heat transfer in a Rotary Kiln.Chapter 4 – Active layer models Jauhari. 1951. Sullivan. M. Hanrot..
At low k/ω. Dispersion effects are included in the active layer model. the passive layer model alone can be used only for deep beds and low Thiele modulus φ.A combined mass transfer model for both active and passive layers in rolling rotary kilns ABSTRACT A mass transfer model for rotary kilns that incorporates both the active and passive layer concentrations. Of these. and convection (gas entrainment) effects are included in the passive layer model. Gas entrained into the passive layer at the wall of the kiln makes an important contribution to mass transfer. By making two assumptions: 1) that molecular diffusion is unimportant compared to dispersion. h/R and R/dp. otherwise the combined model is recommended. then the effectiveness factor for the bed is dependent on k/ω. only R/dp is dependent on the scale of the kiln. At high k/ω. it is not sufficient to use the contribution of the active layer alone.Chapter 5 – Combined model Chapter 5 . and 2) that the particle velocity profile in the active layer (m) is proportional to rotation speed. is applicable over a wide range of rates. 77 .
Chapter 5 – Combined model
1.
INTRODUCTION
In rolling rotary kilns, there are two distinct regions within the bed: the passive layer, where particles are transported concentrically with the kiln until they reach a point near the surface of the bed, and the active layer, where these particles cascade down the surface of the bed. It has
previously been assumed that the active layer is the most important area for mass or heat transfer. Heydenrych et. al. (2001a) have shown that for slower reactions (or drying), mass transfer in the passive layer can also play an important role. In this paper, we explore the transition from mass transfer domination in the active layer, to kinetic domination in the passive layer.
Figure 1. The regions found in a rotary kiln.
The work of Heydenrych et. al. (2001a) explores how gas entrapped in the voids between particles in the active layer, gets depleted as it moves concentrically in plug flow through the passive layer with the particles. 78
Chapter 5 – Combined model
They used an effectiveness factor to characterise the radial concentration profile, and a gasphase volumetric rate constant to include mass transfer limitation, for example, for 1st order processes: PA = ηεkCA
(1)
By rewriting Eqn. (1) in terms of dimensionless units, they showed that the effectiveness factor depends only on k/ω and the bed fill. In later work (Heydenrych et. al. 2001b), the effect of diffusion within the passive layer was investigated. They showed that diffusion effects in the passive layer are unimportant for slower reactions (k/ω < 1), but importantly, they also showed that an additional dimensionless number must be used to describe mass transfer in faster reactions when diffusion effects are to be taken into account. This can either be a Thiele modulus, or (as they chose), a Peclet number, defined as
εωR 2 Pe = De
(2)
Jauhari et. al. (1998) have measured mass transfer rates for the drying of relatively small silicaalumina particles soaked with decane. These results gave far greater mass transfer rates than the passive layer model (Heydenrych et. al. 2001a) predicts. Jauhari et. al. (1998) used an
empirical model based on the volume of the active layer to predict mass transfer as a function of rotation speed.
Heydenrych et. al. (2001c) developed a more fundamental model to predict the shape and volume of the active layer. This model was largely based on the assumption that the velocity gradient (m) of particles within the active layer is constant at any point within the active layer, given a set of
79
Chapter 5 – Combined model
conditions such as bed depth and rotation speed. They predicted that bed depth could be described by
d=
R 2 − x2 − h2 m /ω −1
(3)
This model was further extended to predict the dispersion coefficient in the active layer caused by the change of void volumes that mix the gas as particles slide over each other. If successive layers of particles in the active layer move as rigid layers, then the dispersion of gas will occur predominantly in the direction perpendicular to the surface of the bed, and the volume of gas dispersed in this way can be calculated from the difference in voidage of a loosepacked structure and a tightpacked structure. Using these assumptions, they derived the following expression to calculate dispersion in the active layer: Dal = 0.123 m dp2 + ε2Da
( 4)
They further showed that a relatively simple 1dimensional model could be derived to predict the concentration gradients in the active layer in the case of fast reactions (k/ω > 10), giving rise to a correlation for a mass transfer coefficient for the surface of the bed: k s = Dal εk
( 5)
Using published data to predict the change of the velocity gradient m with rotation speed, they showed that the model represented by equations (3), (4) and (5) is inadequate for predicting the mass transfer rates measured by Jauhari et. al. (1998). In this paper, we would like to adapt the fundamental approach adopted by Heydenrych et. al. (2001c) to one that better predicts
80
and some simplifying assumptions will be made so that the model can easily be solved on a spreadsheet. As a design tool. but we now explore whether the rate constant k from Eqn. Sh = 3. where there is a batchwise saturation of gas in the voids as the gas rotates with the kiln.8 2 ε dp k= (6) Heydenrych et. al. (6) works well to predict the slow rate of evaporation from relatively large particles in the passive layer of rotary kilns. (1998). The gas concentration in the voids changes according to k (C sat − C A ) = dC A dt (7) 81 . al. On the basis of Sørensen & Stewart (1974) the “rate constant” describing the rate of evaporation of gas from the surface of wet particles in a stagnant packed bed can be formulated: 6ShDa 1 − ε . COMBINED MODEL We continue with the approach of Heydenrych et. (2001a) have shown that estimation of k using Eqn. al. (2001a). 2.Chapter 5 – Combined model the measured data of Jauhari et. (5) is affected by the continuous mixing of gas as the layers of particles slide over each other. for use as a design tool for rotary kilns and applicable to reaction rates that cover a large range. we need a model that is relatively easy to use.
The frequency of this cycle is given by the shear rate m. time (t) in Eqn. This is essentially an empirical correlation. C sat − C A = e − kt C sat − C A0 (8) In the active layer. (1998): m = 70ω (12) It is interesting to note that by substituting m in Eqn. the same process can be expected to occur. chosen to reflect the mass transfer rates measured by Jauhari et. We wish to determine the rate of generation within a control volume V in the active layer: PA dV = (C A (t ) − C A0 )mεdV = (C sat − C A0 ) 1 − e − k / m mεdV (9) ( ) Replacing the generation term PA with a more conventional rate expression. we find that the shape and volume of the active layer is not dependent on 82 .Chapter 5 – Combined model Integrating Eqn. the gas volume is flushed out with fresh gas. and therefore. al. (3). εk ' (C sat − C A0 )dV = (C sat − C A0 )(1 − e − k / m )εmdV k ' = m 1 − e −k / m (10) ( ) (11) Lastly. (8) is the inverse of m. (12) into Eqn. we assume an alternative correlation to describe the relationship of m with rotation speed. as we have used in previous work. except that at regular intervals. (7).
al. (1983). Using this correlation for m and replacing k with k′ of Eqn. al. Figure 2. For each material type. (11). relatively the steep particle velocity gradient of Eqn. Surprisingly. the active layer depth was measured at a range of rotation speeds. we predict the mass transfer rates measured by Jauhari et. Fig. 2 shows how this model predicts the maximum active layer depth (x = 0) compared to the maximum active layer depths that were measured for various materials and drum sizes. (12) predicts deeper active layer depths than those measured visually. (1998) as follows: 83 .Chapter 5 – Combined model rotation speed. and reported by Heinen et.
al. In this case. The data of Jauhari et. The dimensionless concentration is estimated as: C A (λ ) e λφ + e − λφ ε k' .123 m dp2 (14) 84 .Chapter 5 – Combined model Figure 3: By using m = 70ω and k ' = m 1 − e − k / m ( ) we can reasonably predict measured mass transfer rates from k s = Dal εk ' . (2001b) have shown that k/ω and either Pe or φ can be used to describe concentration profiles in a rotary kiln. and the line represents model predictions. we can have Dal = 0. Within the active layer. we prefer to use φ because we have assumed that all diffusion effects take place in the active layer. When binary diffusion is negligible. and assume that at any other part of the bed (x ≠ 0) the same concentrations will be found at a given depth y from the surface. (1998) is represented by ¿. we predict concentration profiles at the deepest part (x = 0). al. where φ = d max = φ −φ Dal C A0 e +e Ψ= (13) Heydenrych et.
This is not easily solved analytically. and the starting concentrations vary along the boundary.72ε 1 − e 1 − R (15) This reduces the model to the same dimensionless numbers that were introduced by Heydenrych et. al. (15) is only valid when dispersion effects are much greater than molecular diffusion. We use these concentrations as the starting concentrations for the gas as it moves concentrically around the kiln in the passive layer. (2001a). (13) by substituting for dmax using Eqn. (11) and finally m using Eqn. but it can be quite easily solved numerically. Both the active layer model and the concentric flow model described above give a concentration profile over the radial crosssection of the bed. The active layer model described above predicts the concentrations that occur at the interface between the active and passive layers of a rotary kiln. (2001a). but in the shape of the active layer interface. (14). bearing in mind that Eqn. We will solve the model using the Thiele modulus φ as the third independent variable (rather than R/dp). The 85 . k′ using Eqn. This is the only term that takes dispersion (and therefore scaleup effects) into account. except that the surface boundary is not flat. but with the addition of the third term R/dp. getting depleted as it moves as it would in a plug flow reactor. Dal using Eqn. because it increases the validity of the simulation results: it is possible to include binary diffusion effects. (3). This is similar to the model described by Heydenrych et. (11): R φ= dp k 1 2 − h ω 70 9. al. or to use any other model to determine the Thiele modulus φ.Chapter 5 – Combined model We can now determine φ from Eqn.
defined in terms of k. 12) in Eqn. CA0 is the freeboard gas concentration. (2001b): η al = A ( al ) ∫ ψ dA Aal and η pl = A ( pl ) ∫ψ dA A pl with ψ = CA C A0 (16) We use the same CA0 for both the active and passive layers. We could replace φ with R/dp using Eqn. Therefore the model for predicting the dependent variable η is a function of three independent variables (we consider ε as a constant): k/ω. h/R and φ. ηεkAbed = η al εk ' Aal + η pl εkApl (17) η = η al A pl k ' Aal + η pl k Abed Abed (18) The term k′/k is determined by substituting m (Eqn. (11): k − k ' 70ω 1 − e 70ω = k k (19) Note that k′/k is only a function of k/ω. so we have chosen φ. but then the results we show would be limited to the model represented by Eqn. (15). al. We need to calculate the effectiveness of the active layer ηal separately from the effectiveness of the passive layer ηpl in order to get an overall effectiveness factor η. 86 . (15).Chapter 5 – Combined model effectiveness factor for a bed area (active or passive layer) is found by integrating the dimensionless concentrations over that area as described by Heydenrych et.
and gas enters the concentric region at almost the maximum concentration. 4. Figure 4. MODEL PREDICTIONS A contour plot of the concentrations found using the model developed here is shown in Fig. al. We expect. Effectiveness factors at various φ are shown as a function of k/ω in Fig. If this is the case. there is a very steep concentration gradient. because at the wall.5 and k/ω = 1 for a clockwise rotating kiln. This shows the continuity of concentrations on the right hand side active/passive layer interface. and the transport of depleted gas from the active/passive layer interface into the active layer. which is unlikely to occur in practice due to diffusion effects there. Note how we find a high concentration at the wall of the kiln. where the gas moves into the bed. where h/R = 0. 87 . from the findings of Heydenrych et. 4. On the left hand side of Fig. that the diffusion effects in the active layer will predominate over the transport effects in the passive layer. (2001b. A contour plot of dimensionless concentrations using the combined active/passive layer model with h/R = 0.c).5 and k/ω = 1. 4. then the model should be a reasonable indication of what happens in practice. the active layer depth is very small.Chapter 5 – Combined model 3.
The thick line represents the passive layer model (Heydenrych et. in the case of reaction. we move from right to left on Figure 3 along a line of constant R/dp.1 and 0. This means that φ is not constant along the length of a kiln. Effectiveness factors at various φ for a bed fill of 37% (h = 0. These are useful because typically we find that k varies along the length of a kiln as the solid dries (intraparticle diffusion limiting) or as the solid material becomes converted. effectively blocking reactive gas from entering the passive layer. 2001a). for larger kilns (larger φ) we always have a lower effectiveness factor. It is interesting to note that the active layer does not simply add to the mass transfer predicted by the passive layer model (Heydenrych et. Fig. Not surprisingly.Chapter 5 – Combined model φ=1 φ=3 φ = 10 φ = 30 φ = 0. 2001a). 6 shows lines of constant R/dp. but predicts lower values than that model. al. al.2). and as reaction proceeds. This is due to the depletion of gas within the active layer. 88 .3 Figure 5.
Chapter 5 – Combined model R/dp = 100 R/dp = 1000 R/dp = 10 Figure 6. Lines of constant R/dp are useful because typically R/dp remains constant over the length of a rotary kiln. with bed fill of 5. but with a shallow bed. Therefore. Here. the active layer predominates and we do not benefit fully from the entrainment effect of the passive layer as it draws gases into the bed. 5. but the effectiveness factor here is less dependent on k/ω because of the increased ratio of bed surface area (active layer) to bed volume. Fig. As bed fill decreases.2% (h = 0. the passive layer model of Heydenrych et. even with slow reactions (low k/ω). (2001a) is poor at predicting the shape of the curves. the active layer volume increases relative to the passive layer volume. 89 . al.8). 7 shows a similar plot to Fig. and cannot be used even for slow reactions. The trends are similar.
2% (h = 0. The thick line represents the passive layer model (Heydenrych et. This is unlikely in practice. To examine the contribution of active layer at various bed heights. Effectiveness factors at various φ for a bed fill of 5. 2001a). al. we determine what fraction the active layer contributes towards the total effectiveness of the bed. This is calculated using the 1st term of Eqn.3 φ=3 φ = 10 and 30 Figure 7. we find low effectiveness factors even at low k/ω. and this gives R/dp = 1000 for the 4m diameter kiln.1 and 0. If we consider a rotary kiln using pelletised particles of 10 mm diameter. In both Fig. The lines of constant R/dp can be used as a guide. because when the reaction rate is slow. usually the Thiele modulus is also low. (18) divided by the total effectiveness factor: 90 .Chapter 5 – Combined model φ=1 φ = 0. We find that dp = 2 mm gives Dal values that are consistent with molecular diffusion. 5 and Fig.8). then a 4 m diameter kiln would give R/dp = 200. and we can approximate this using the dispersion model with an appropriate value of dp. It would be quite unusual to find a rotary kiln of a scale larger than R/dp = 1000. 7. because then molecular diffusion would predominate.
25%.2. 9. It is also interesting to note that for deep beds. 20%. 8 shows for φ = 3 how the active layer is always more important for shallower beds.7. corresponding to h = 0. The contribution of the active layer is always more for shallower beds (φ = 3). 0.8.2%.4. From top to bottom. 0.Chapter 5 – Combined model η al f al = k ' Aal k Abed η (20) Fig.2% Bed fill 37% Figure 8. The fraction contributable to the active layer does not approach unity for a combination of two reasons.5.6. The second effect is that 91 . The first is that even for fast reactions that occur very close to the surface of the bed. 14%. the passive layer continues to play a significant part in mass transfer even at high reaction rates. where the depth of the active layer tends to zero. Bed fill 5. curves show bed fills of 5. 0.3 and 0.4%. there is always an appreciable amount of gas that moves into the passive layer at the wall of the kiln. 0. 0. 31% and 37%.
lessening the active layer contribution. but in most cases better mass transfer will be obtained with a deeper bed.1 k/ω = 1 k/ω = 10 k/ω = 100 Figure 9. For fast reactions. 0. Noting that we typically get higher effectiveness factors using shallower beds. for instance. but in general. 9 shows an effectiveness factor based on total volume.Chapter 5 – Combined model the weighting due to the term k′/k comes into play. it is possible to get an optimum bed fill.01. For k/ω = 100. and that commercially operated kilns typically use shallow beds. 92 . it would be interesting to see if there is a maximum mass transfer that can be achieved for a given kiln radius. Fig. rather than the bed volume that the previous effectiveness factor was based on. We see that there may be a maximum at the highest reaction rates (k/ω = 100). k′/k = 0. k/ω = 0. a deeper kiln gives better mass transfer.23.
0.1 and 0.Chapter 5 – Combined model φ = 10 φ = 0. 1.1 Figure 10. φ (top to bottom) = 0. and 10. Active layer contribution for shallow beds (fill = 5.1 . 3 φ = 10 φ=3 φ=1 φ = 0. Active layer contribution for a deep bed (fill 37%) does not change much as dispersion increases.1.3 Figure 11.3. 93 .2%) is highest for the larger kilns.
94 . Figure 11 shows that for shallow beds. COMPARISON WITH MEASURED DATA Applying the combined model to Jauhari’s data (Fig. 11). The increase in the slope of the line predicted by the combined model is due to two effects: Firstly. there is still a significant contribution by the passive layer that the active layer model does not account for. at lower rotation speeds. and we find that the active layer contribution becomes increasingly important as φ increases. even at high k/ω. The combined model ⋅⋅⋅⋅⋅⋅⋅⋅⋅ predicts improved mass transfer rates at the higher rotation speeds compared to the active layer model . The second effect is the additional effect of the passive layer. and the omission of binary diffusion from the model results in the prediction of lower mass transfer rates.Chapter 5 – Combined model For a shallow bed. dispersion is low. 12) gives small but significant changes to the active layer model that was fitted using Eqn. 4. Jauhari’s data is represented by ♦. we see a reversal in the order (Fig. (12). Figure 12.
Chapter 5 – Combined model
The data of Heydenrych et. al. (2001a) is shown in Fig. 13 against their passive layer model (thick line) and the combined model (thin line), using R/dp = 4.7. Although the scatter of the data does not allow definite
conclusions to be made, it appears that the combined model fits the data better than the passive model alone.
Figure 13. The combined model (thin line) fits the data of Heydenrych et. al. (2001a) somewhat better than the passive layer model (thick line).
5.
CONCLUSIONS
In attempting to simplify the relationship for the velocity gradient in the active layer to m = 70ω, we arrive at some useful relationships that provide insight into the scaleup of rotary kilns. In particular, we showed that the dimensionless number R/dp describes the dispersion effect within a rotary kiln, and therefore it is the parameter that is important for scaling of rotary kilns.
95
Chapter 5 – Combined model
Practically, we find a large range of R/dp in rotary kilns. An active layer layer model that calculates the mass transfer coefficient ks at the surface of the kiln can be used when the fraction contribution of the active layer is high. In many cases, this is not true, and the combined model presented here should be used. Also, the passive layer model used alone will in most cases incorrectly predict the mass transfer, having a totally different form (as a function of k/ω) to that predicted by the combined model.
It would also be incorrect to assume that the mass transfer predicted by an active layer model can be added to the mass transfer from a passive layer model: we have shown that in many cases the depletion of gas in the active layer prevents activity in the passive layer.
96
Chapter 5 – Combined model
SYMBOLS
A CA CA0 CA,s d dmax Da Dal De dp fal crosssectional area [m2] gasphase concentration, [mol m3] gasphase conc. at t = 0 or freeboard gas conc., [mol m3] gasphase concentration at the bed surface. [mol m3] active layer depth, [m] active layer depth at x = 0, [m] binary diffusion coefficient, [m2 s1] total dispersion coefficient for the active layer, [m2 s1] effective dispersion coefficient for a packed bed [m2 s1] particle diameter [m] fraction contribution of active layer towards total effectiveness factor [] h perpendicular distance from radial center of the kiln to the bed surface [m] k k′ kgs ks L m PA Q R V Vbed Val x gasphase reaction rate constant [s1] gasphase reaction constant for use in active layer, s1 mass transfer coefficient on the surface of particles [m s1] mass transfer coefficient of active layer [m s1] length of the rotary kiln in axial direction [m] shear stress in the active layer (velocity gradient of particles) s1 reaction rate of A [moles m3(bed) s1] volumetric gas flow rate [m3 s1] internal radius of the rotary kiln [m] volume [m3] volume of bed [m3] volume of the active layer [m3] direction parallel to the surface of the bed in the radial crosssection; x = 0 is at the centre [m] y direction perpendicular to the surface of the bed [m] 97
J. Heesink. Greeff. J.D.. ‘Hydrodynamics and dispersion in the active layer of rolling rotary kilns’. 233249 Heydenrych. M.V. (2001).A... (1997). A. Versteeg. 2001. ‘Granular flow behaviour in the transverse plane of a partially filled rotating cylinder. J.B. 98 .. depth below bed surface/active layer depth effectiveness factor of reaction Thiele modulus concentration. A.M. G. A...M. Heydenrych...M. G.. M. Accepted for publication at the 6th World Congress of Chemical Engineering. 2001.M. Eng. J. J. Heydenrych.. P. ‘Mass transfer in rolling rotary kilns: A novel approach’.. J.. (2001). Eng.Chapter 5 – Combined model Greek symbols ε λ η φ ψ ω bed voidage [m3(gas) m3(bed)] dimensionless distance. 2001. 330.. & Barr. (2001). dimensionless angular velocity [s1] REFERENCES Boateng.M. Schieke. Kuipers. Heesink. Kuipers. Submitted to Can. Melbourne. ‘Diffusion effects in rotating rotary kilns’. P. A.B. Chem.A. Submitted to Can.D.F. Versteeg.B.D. Chem.. M. Fluid Mech. Heesink.F.A.
C. 76..G.. O.R.. (1995). Eng. 14B. J. E. (1927). Chem.P. J. H. Ablitzer. The passage of granular solids through inclined rotary kilns. 191205 Jauhari. R. Hanrot.F. J. “Passage of solids through rotary kilns”. 99 . US Bureau of Mines. J.C. Prog. & Masliya.. Eng.. Effect of fluid dispersion coefficients on particletofluid mass transfer coefficients in packed beds”. particle movement and bed depth profile in rotary kilns”. 57. (1983a). 224232.. Trans.P. Maier. J.. Technical Papers 384. “Experimental study of residence time. S. H. Saeman. 33. 173179. Eng.H.. WC.Chapter 5 – Combined model Heinen.D.L.. Passage of solid particles through rotary cylindrical kilns. Sørensen. Chem. 13741384 Sullivan. & Houzelot. 1951. Chem.. Experimental study of transverse bed motion in rotary kilns. JP.K.. 1. Chem. Chem. Metall. 142 Tscheng. 47.. M. Chemical Engineering Science. (1988). WE. Kramers. 508514. 433443.. Stewart. Brimacombe. Eng. A. J. (1952). (1979). Can. 1974.. Gassolid mass transfer in a rotating drum. Gray. F. J. Can. & Ralston. D. Eng. Sci. 259 Lebas. 1983. A. and Watkinson. P.H. 73. & Crookewit. & Watkinson. Convective heat transfer in a Rotary Kiln. Can.
.
. G. M.236. Accepted for publication Moolman.. C. Hunter. D. (1989) ‘Intrinsic kinetics of olefin oligomerisation on a supported nickel catalyst’.. Chem. M.. 2001.M.. J. (2001). (2001). M. M. Scurrel. 43(6) 12911296. Chem.. J. D.P. and Nicolaides. South African patent ZA 89 02.D. Applied catalysis.S. D. CENG581. and Heydenrych. Eng. Chem. Sci. (1985). M. ‘Pressure drop over fluidised bed distributor caps’. Van Vuuren. CSIR rep.S.F. Heydenrych. Heydenrych.S.. F. and Manzini.D. D. Eng.A. and Van Vuuren. J.S.S.... M.Publications Publications Van Vuuren. 101 . Chem. ‘The solubility of various gases in FischerTropsch reactor wax’. Accepted for publication Heydenrych. Eng. Submitted to Can. Reid. (2001).D. Eng. Thesis. ‘Fluorination of zircon in a fluidized bed’.D. Accepted for publication Heydenrych. ‘Mass transfer in rolling rotary kilns: A novel approach’. Versteeg. University of the Witwatersrand.D. (1988).. M. A. Greeff.B... Heydenrych.. P. (2001). ‘Multicomponent modeling of FischerTropsch slurry reactors’.D.J.R. Can. ‘Oligomerization of ethene in a slurry reactor using a nickel(II)exchanged silicaalumina catalyst’. Heesink. (1989) ‘Catalytic process for the oligomerisation of olefins’. C. and Heydenrych.D. S. and Heydenrych.D. 46 pp. M. M.
D. J.B. A. ‘Diffusion effects in rotating rotary kilns’. J. J.F.M. 2001. Eng.M..Publications Heydenrych. Heydenrych. Heesink. 2001..... 102 . ‘Hydrodynamics and dispersion in the active layer of rolling rotary kilns’.A.M.. M. Accepted for publication at the 6th World Congress of Chemical Engineering. Chem..D. G.A. Versteeg. Melbourne. Kuipers. (2001).M. (2001). A. Heesink. M. J. Schieke. Submitted to Can.B. Kuipers.
To the University of Pretoria. they will attest that I often seem far away . Geert Versteeg and Prof. He spent a lot of his time not only on the thesis. . Uys Grimsehl. And when I am with them physically.Acknowledgements Acknowledgements At first glance. . a doctoral thesis would seem to be an individual task. He ensured that. In that respect. I also owe a debt of gratitude to the other members of the Department of Chemical Engineering at the University of Pretoria for taking up some of my duties when I left the Department on study leave to the University of Twente. I would like to acknowledge the support and contributions of the following people in particular: My wife René. to the University of Twente and my promotors Prof. Hans Kuipers for initiating and funding my stay at the University of Twente. apart from the core educational tasks that I was responsible for. Finally. and in particular Prof. and indeed supported me in this doctorate all the way. my time was free to concentrate on my research. Special thanks go to my assistant promotor Dr. who did many things to provide the environment for completing this thesis. Bert Heesink who taught me how to add value to papers in review – a skill that I hope to be able apply much more effectively in my career in future. because an endeavour such as this inevitably encroaches on the time spent with them. He supported my application to complete my studies at the University of Twente in every way. but it ultimately involves many people. who supported my move to the academic world. My whole family needs to be thanked. 103 . Geert Versteeg spent a lot of time hosting me and my family in our stay in Enschede – something I value greatly. but both he and Prof.
by actively promoting the cooperation agreement between our respective Universities. and strengthen the friendships we have established here.Acknowledgements My hope is that in future we will be able to keep regular contact. 104 .
105 . In 1993 he returned to CSIR to work on fluidised bed combustors: fullscale design. a town about 100 km west of Pretoria. he joined the University of Pretoria where he is responsible for the undergraduate and postgraduate courses in Reactor design and for the Final year plant design project. and these are now used industrially. he completed an MSc (at the University of the Witwatersrand) on the determination of the intrinsic kinetics of the oligomerisation of ethene on a supported nickel catalyst. He then spent two years as a consulting engineer for Megkon. supplying units to industry. The information was used by SASOL to scale up their slurry reactors. From 1984 he worked at CSIR (a contract research organisation) as a team member. He obtained his bachelors degree at the University of the Witwatersrand in 1982 and then completed two years of compulsory National Service.Curriculum Vitae Curriculum vitae Mike Heydenrych spent his primary and secondary school years in Rustenburg. he headed a large minerals processing development project that that sparked his interest in mass transfer in rotary kilns. a company specialising in systems engineering. determining the hydrodynamics of FischerTropsch slurry reactors. In 1997. After that. He is married to René with 2 children. Still working for CSIR. Nicole and Graeme.
This action might not be possible to undo. Are you sure you want to continue?