CFD predictions for cement kilns including ame modelling,

heat transfer and clinker chemistry

E. Mastorakos
*
, A. Massias
1
, C.D. Tsakiroglou, D.A. Goussis, V.N. Burganos,
A.C. Payatakes
2
Institute of Chemical Engineering and High Temperature Chemical Processes, Foundation for Research and Technology,
P.O. Box 1414, 26500 Patras, Greece
Received 24 June 1997; received in revised form 20 April 1998
Abstract
Clinker formation in coal-red rotary cement kilns under realistic operation conditions has been modelled with a
commercial axisymmetric CFD code for the gaseous phase including a Monte Carlo method for radiation, a nite-
volume code for the energy equation in the kiln walls, and a novel code for the species and energy conservation
equations, including chemical reactions, for the clinker. An iterative procedure between the predictions for the tem-
perature eld of the gaseous phase, the radiative heat ux to the walls, and the kiln and clinker temperature is used to
predict the distribution of the inner wall temperature explicitly, including the calculation of heat ow to the clinker. It
was found that the dominant mode of heat transfer between the gas and the kiln walls is by radiation and that the heat
lost through the refractories to the environment is about 10% of the heat input and a further 40% is used for charge
heating and clinker formation. The predictions are consistent with trends based on experience and limited measure-
ments in a full-scale cement kiln. 1999 Elsevier Science Inc. All rights reserved.
Keywords: Industrial CFD; Rotary kilns; Clinker formation; Radiative heat transfer
1. Introduction
The cement industry is a major consumer of energy, accounting in the USA for about 1.4% of
the total fossil fuel consumption by industrial users [1], with a typical specic energy use of about
3.2 MJ/kg of clinker manufactured [2]. This is so because the reaction CaCO
3
CaO + CO
2
, the
rst step in clinker formation, is highly endothermic. Current trends to save energy in cement
kilns focus on a more ecient recuperation of energy from the exhaust gases leaving the kiln,
usually a rotating cylinder having length about 100 m length and diameter of about 5 m, and on
the use of fuels of lower caloric value. The nature of the rotary kiln, which allows ame residence
times of the order of 25 s and reaches temperatures as high as 2200 K, also makes cement kilns a
competitive alternative to commercial incinerators for organic wastes and solvents [3]. To ensure
that the formation of clinker is not aected in any way from alterations in the ame by the use of
Applied Mathematical Modelling 23 (1999) 5576
*
Corresponding author. Tel.: +30 61 965 218; fax: +30 61 965 223; e-mail: mastorak@iceht.forth.gr.
1
Also at Department of Mechanical Engineering, University of Patras, Greece.
2
Also at Department of Chemical Engineering, University of Patras, Greece.
0307-904X/99/$ see front matter 1999 Elsevier Science Inc. All rights reserved.
PII: S 0 3 0 7 - 9 0 4 X ( 9 8 ) 1 0 0 5 3 - 7
such secondary liquid fuels, and to be able to comply with current and future restrictions on the
amount of pollutants released in the atmosphere, there is a need to understand the details of the
ame structure better and the heat transfer from the ame to the solids charge.
Recently, theoretical modelling of the ow eld and coal combustion in cement kilns [47] has
been achieved by the use of computation uid dynamics (CFD) codes including radiative heat
transfer. These results were the rst of their kind for cement kilns and showed that such codes can
be used for large-scale industrial furnaces with acceptable accuracy, as they reproduced experi-
mentally observed trends concerning ame lengths, oxygen consumption etc. A limitation was
that coupling with the kiln and the clinker was not included and the boundary condition of wall
temperature, a necessary input to the calculation because it aects the gas temperature and the
heat ux out of the solutions domain, was not predicted but was based on experimental mea-
surements. CFD solutions for the gas phase are a major step forward, but alone are not adequate
to model the rotary kiln comprehensively.
Details of the heat transfer in the refractories and the charge have been examined by various
authors [811]. The charge (assumed usually to be well-mixed) is heated from the kiln walls by
conduction, in addition to the heat radiated directly onto the exposed surface. In the most
complete theoretical eorts, the 3-D heat conduction equation for the kiln walls (refractories) was
solved and coupled with a zone model for radiation heat transfer, in which radiation in the axial
coordinate was limited to two or three adjacent zones [10] or solved in detail [11]. The chemical
reactions that take place during clinker formation, however, were not taken into account and the
gas phase was xed at a uniform temperature, and hence not treated to the necessary level of
accuracy.
The solids charge, eventually the recipient of (part of) the energy released by combustion,
undergoes chemical reactions to form the nal product, the clinker. These have been the subject of
extensive investigation in the cement chemistry literature, with a simplied model for the major
features described by [12]. In that work, the o.d.e.'s describing the axial evolution of the solids
temperature and composition were formulated and solved, but the gas and wall temperatures,
which determine the heat ux to the charge and hence the rate of clinker formation, were ap-
proximated with one-dimensional models without detailed ame calculations.
It is evident that none of the above theoretical models for the ame, the walls, and the charge
are sucient in their own right as truly predictive tools for rotary kiln operation. The local gas
temperature (one of the outcomes of the CFD calculation) depends signicantly on the wall
temperature. Clinker formation absorbs energy and hence depends on the local gas and wall
temperatures, both of which depend on the ame; the wall acts as an ``intermediary'' in the net
heat transfer from the ame to the clinker, with unavoidable heat loss to the environment de-
pending on the refractory thickness. These interdependencies are crucial for the behaviour of the
kiln and are intractable by the mathematical models found in the literature because emphasis was
put on calculating individual processes in isolation, with assumptions or very crude modelling
made about the rest.
In the present paper, we present a comprehensive and modular model for most of the processes
occurring in a rotary cement kiln, accomplished by coupling three separate models of the types
outlined above. Our work treats the gas phase with a multi-dimensional CFD code for coal
combustion as in [47]; it solves the 3-D heat conduction equation in the kiln walls as in [10,11];
and it solves for the well-mixed charge temperature and composition with a model similar to
[9,12]. The three modules (ame, walls, charge) are iteratively calculated until convergence is
achieved for the axial distribution of heat ux incident on the refractories, which is the quantity
that (directly or indirectly) determines the gas temperature, the kiln temperature, and the charge
temperature and composition. The major advantage of the present over previous work lies
56 E. Mastorakos et al. / Appl. Math. Modelling 23 (1999) 5576
therefore in its completeness. This allows a better understanding of kiln operation, through the
quantication of energy ows in various parts of the gas-kiln-clinker system and makes the
methods used here applicable to other rotary kiln applications such as drying and incineration.
The particular objectives of this paper are to present a comprehensive model for the rotary
cement kiln and to quantify the energy uxes from the ame to the clinker and the heat loss from
the kiln. The rest of this paper is organised as follows. In Section 2, the various models and
solution methods are outlined and the results are presented and discussed in Section 3. Included is
a comparison with limited measurements from a full-scale rotary cement kiln. The paper closes
with a summary of the most important conclusions.
2. Model formulation
2.1. An overview
Fig. 1 shows a cross-section of a cement rotary kiln. The rotation of the kiln induces a motion
of the charge which keeps the latter approximately well mixed [10], an assumption we will adopt
here. We will allow for the presence of clinker attached onto the kiln refractories by modelling this
coating as a solid material of physical properties similar to those of clinker. We assume that this
layer's thickness is uniform along the kiln. This is probably an over-simplication, but no relevant
data are available. It is important to understand the various energy ows in a rotary kiln,
schematically shown in Fig. 2, before proceeding with a description of the models.
The energy (per unit time) released by the combustion of coal (Q
coal
) ows out of the kiln with
the exhaust gases (DH
gas
) and is radiated (Q
rad
) and convected (Q
conv
) onto the kiln walls and
directly onto the charge, while a small part is lost by radiation in the axial direction to the feed
and exhaust ductworks (Q
radY1
+ Q
radY2
). The charge-kiln system absorbs energy (DH
cl
) from the
Fig. 1. Denition of geometrical parameters and dimensions of the rotary kiln.
E. Mastorakos et al. / Appl. Math. Modelling 23 (1999) 5576 57
gas by radiation (Q
rad
) and convection (Q
conv
) and loses heat to the surroundings (Q
loss
). The
calculation of the individual terms in the overall energy balances, i.e.
Q
coal
= DH
gas
Q
rad
Q
conv
Q
radY1
Q
radY2
Y (1a)
Q
rad
Q
conv
= DH
cl
Q
loss
(1b)
is performed here by solving, locally, the discretized energy for the gas, kiln, and clinker, as
outlined in the following sections.
2.2. The CFD code
The Favre-averaged equations of gas momentum, species concentrations, and energy are
solved by an axisymmetric commercially available CFD code (FLOW-3D) using a standard ke
model, in conjunction with a radiation module (RAD-3D) [13]. Although the geometry is really
Fig. 2. Schematic indication of the separate parts comprising the rotary kiln model, together with the energy uxes from
Eq. (1).
58 E. Mastorakos et al. / Appl. Math. Modelling 23 (1999) 5576
three-dimensional and there exists an angular distribution of the wall temperature, we treat the
gas phase as axisymmetric so as to be able to obtain solutions in reasonable amounts of time and
in present-day workstations, as a full 3-D calculation to the required resolution would necessitate
resort to supercomputers. Some features of the various sub-models used in FLOW-3D and of a
special treatment for the wall boundary condition are given below.
2.2.1. Coal combustion
Coal particles following a Rossin-Rammler size distribution (45 lm mean diameter, 1.3 ex-
ponent [6]) were tracked deterministically in the solution domain (i.e. without a stochastic
component) to reduce CPU time, but with the drawback of under-estimating dispersion [14]. The
particles were allowed to devolatilise according to a two-reaction model and volatile combustion
was taken as innitely fast. It was found that details of the description of coal combustion did not
produce appreciable dierences in the temperature proles obtained, with the ``standard'' set of
constants of coal devolatilisation and pyrolysis in FLOW-3D giving reasonable results and similar
to those in Ref. [5].
2.2.2. Radiation and convection
The radiation intensity in the gas was calculated with a Monte-Carlo method with 80,000
particles with the RAD-3D module. The kiln was divided in non-uniform zones, seven in the
radial direction and 19 in the axial (of size between 0.1 1.0 m and 0.2 5.0 m), in each one of
which the radiation intensity was assumed uniform. The output of the radiation module was
interfaced to the uid calculation internally to FLOW-3D, and externally to the codes for the
walls and clinker (see Sections 2.3 and 2.4). It is important to realize that the discretised zones of
the radiation package should be much larger than the cells of the CFD grid, so that the zone will
contain many cells over which a temperature average can be formed. The relatively coarse
resolution of the radiation zones and the statistical nature of the Monte-Carlo method implied
that the radiation heat ux in a zone could be inaccurate by up to about 10%, as determined by a
few runs with ner zoning and more particles. Note also that the heat ux incident on the
bounding surfaces cannot be determined to a resolution ner than the zone size, and so the
radiative heat ux is calculated only at the centre of each of the 19 zones adjacent to the wall. An
absorption factor of 0.15 m
1
was used for the gas throughout the whole domain, following Ref.
[11]. Strictly speaking, the absorption factor is a function of the local gas composition and
temperature and hence should not be uniform. However, the gas composition varies only inside
the ame, which occupies only a part of the kiln (see Section 3), and so the assumption of a
uniform absorption factor is reasonable. In addition, the present version of the software does not
allow the automatic re-adjustment of this factor during the iteration process in FLOW-3D,
where the local gas properties are calculated, and hence a constant and uniform absorption
factor was necessary.
Finally, the convective heat transfer at the walls, a standard output from the FLOW-3D
package, was based on an expression for a turbulent boundary layer incorporated in a standard
``wall function'' formulation, similar to the velocity boundary condition in the ke model. The
various quantities used as input and calculated as output from FLOW-3D and RAD-3D are
shown schematically in Fig. 3.
2.2.3. Grid
A non-uniform grid with 47 nodes in the radial and 155 in the axial direction was used and it
was considered sucient after performing grid-independence studies. Grids of similar magnitude
have also been found adequate in Refs. [47]. Extra care was taken to resolve the region close to
E. Mastorakos et al. / Appl. Math. Modelling 23 (1999) 5576 59
the coal burner well, due to the very high axial and swirl velocities there. Typical CPU times for
this grid on an HP 715/100 MHz workstation were 10 h.
2.2.4. Boundary conditions
It is important to realize that a boundary condition for the wall temperature is needed both for
the gas phase and the radiation solvers. Below, we outline an approximate method for predicting,
rather than prescribing [47], this temperature distribution for an axisymmetric grid.
Assume the angular distribution of the inner wall temperature T
w
(R
in
, x, u) were known. Then,
to reduce the full 3-D problem to an ``equivalent'' axisymmetric one, a ctitious inner wall
temperature, T
RAD
(x), calculated from
2pe
w
1
4
RAD
(x) = e
cl

H
0
1
4
cl
(x) du e
w

2p
H
1
4
w
(R
in
Y xY u) du (2)
can be used as an ``eective'' boundary condition. T
RAD
(x) is a ``radiation-weighted'', azimuthally
averaged temperature. The use of this temperature as the necessary boundary condition is mo-
tivated by the expectation that radiation dominates heat transfer (conrmed a posteriori; Sec-
tion 3.4). It is hence expected that the overall energy ows from the gas to the kiln in the full 3-D
problem and in this ``eective'' axisymmetric one, are approximately equal. The use of Eq. (2)
allows the calculation of the gas phase and radiation eld with an axisymmetric code, hence
making the problem tractable using engineering workstations.
At the scale and temperatures of the kilns considered, the gas can be considered as optically
thick. Solutions (presented later) show that a mean path length (i.e. the mean length before all the
energy of a ``photon'' has been absorbed) is about 3.2 m, which is less than the kiln inner diameter
of 4.1 m. This then implies that the radiation ux incident on the inner wall depends on the local
wall and gas temperatures strongly and only weakly on the temperatures of the walls at farther
Fig. 3. The structure of the commercially available FLOW-3D RAD-3D package. Shown are also the input and
output quantities.
60 E. Mastorakos et al. / Appl. Math. Modelling 23 (1999) 5576
axial or azimuthal locations. This gives credence to the axisymmetric approximation used for the
gas phase. It also implies that the ``zonal method'' of Refs. [810] was probably sucient, where
the radiation incident on a surface was taken to be independent of axial locations farther than 12
zone lengths.
2.3. Kiln temperature
The inner kiln surface temperature, T
w
(R
in
, x, u), is needed in Eq. (2) and is part of the
outcome of the solution of the kiln wall energy equation. In a coordinate system rotating with an
angular velocity x, the latter becomes [10]
x
o(.
w
c
p
1
w
)
ou
=
1
r
o
or
k
w
r
o1
w
or

1
r
2
o
ou
k
w
o1
w
ou

o
ox
k
w
o1
w
ox

(3)
subject to the boundary conditions
r = R
in
Y H ` uT2pX k
o1
w
or
= q
rad
(x) q
conv
(x)Y (3a)
r = R
in
Y 0 ` uTHX k
o1
w
or
= q
wcl
(xY u) = h
wcl
1
cl
(x) 1
w
(R
in
Y xY u) [ [Y (3b)
r = R
out
Y 0 ` uT2pX k
o1
w
or
= h
sh
1
sh
1

( ) e
sh
1
4
sh
1
4

X (3c)
Values for the conductivity, density, and specic heat capacity are taken from manufacturer's
information for the material of the refractories used in a real kiln [15]. The outer shell temperature
T
sh
=T
w
(R
out
, x, u) may depend on x and u.
Some comments on the above equations are due. In Eq. (3a), the azimuthal dependence of the
heat ux has been removed. As was mentioned previously, the ow is considered optically thick,
i.e., a wall surface element does not ``see'' the opposite wall across the ame because the ame is
too radiant and too wide. A calculation of q
rad
(x, u) therefore could be based on T
gas
(r, x) and the
local T
w
(R
in
, x, u), without need to include radiation from other angular positions. Since here
q
rad
(x) is taken from the axisymmetric RAD-3D solution, which was based on the azimuthally
averaged temperature from Eq. (2), the resulting q
rad
(x) is, in a sense, an azimuthally averaged
heat ux. Eq. (3a) hence implies that we equi-distribute this heat ux to all u. Note that neglect of
the angular variation of q
rad
in Eq. (3a) is probably a bad approximation for kilns of small di-
ameter or for colder ames, where the full method of Refs. [10] or [11] should be preferred.
The heat transfer coecient between the kiln and the charge, h
wcl
, is important for the correct
prediction of energy ow and temperatures in the kiln, and is not very well known. A typical value
of 500 W/m
2
K has been used for the results presented here [8]. The calculated T
w
(r, x, u) and
T
RAD
(x) depend on the choice of this coecient, but the predictions do not change qualitatively.
The heat transfer from the shell to the atmosphere occurs, apart from radiation, through forced
and natural convection. The heat transfer coecient for natural convection was calculated in the
manner of Ref. [11], with typical values about 5 W/m
2
K for the present conditions. However,
external fans that blow onto the shell are often employed in industry to avoid overheating, and
such eects were modelled here by using a total h
sh
=30 W/m
2
K. The following values were used
for emissivities: e
w
= e
cl
= 0X9 and e
sh
= 0X8X
Eq. (3) was solved by an explicit nite volume method, after neglecting azimuthal and axial
conduction because the radial gradients in the kiln are much steeper. A non-uniform grid with 50
nodes in the radial and 19 in the axial direction was used, with adequate radial resolution at the
inner surface to follow the fast temperature changes there caused by the periodic surface exposure
E. Mastorakos et al. / Appl. Math. Modelling 23 (1999) 5576 61
to the ame due to rotation. The small step (Du=p/100) that was used for the same reason, along
with the large thermal inertia of the kiln, required CPU times of the order of 2 h to ensure that the
kiln wall temperature converged.
2.4. Clinker
The nal part of the model for the kiln concerns the clinker temperature and composition
conservation equations, which become, considering only axial gradients and neglecting conduc-
tion:
c
p
J
cl
d(.
cl
1
cl
)
dx
=
L
wcl
HA
cl

H
0
q
wcl
(xY u) du
L
gcl
A
cl
q
rad
(x) q
conv
(x) [ [

N
sp
i
r
i
h
f
0Yi
c
p
1

Y (4)
J
cl
d(.
cl
Y
i
)
dx
= r
i
Y (5)
J
cl
d.
cl
dx
= r
CO
2
Y (6)
where A
cl
is the cross-sectional area covered by the charge owing at a velocity V
cl
, both assumed
constant, L
gcl
=2 R
in
sin(H/2) the chord of the sector covered by the charge (Fig. 1), L
wcl
=H R
in
,
N
sp
the number of chemical species, and r
i
the net rate of formation (in kg/m
3
s) of species i. The
density of the charge decreases Eq. (6) due to mass loss to CO
2
during calcination, and its initial
value was chosen such that the total mass ow rate is equal to .
cl
V
cl
A
cl
. Note that .
cl
is not the
material density of the charge, as the charge can be thought to be composed of loosely-packed
particles, but a bulk density. We also neglect conduction in the charge because we believe that
anything less than examination of the actual granular ow patterns in the charge [10] is an ad hoc
description that does not add much to the reliability of the model. The charge is assumed to be
composed of CaCO
3
, CaO, SiO
2
, Al
2
O
3
, Fe
2
O
3
, C2S, C3S, C3A, and C4AF, with the last four
species denoted by a special notation used by cement chemists for the complex salts formed
during clinkerisation. We assume the following chemical reactions [12]:
The pre-exponential factors and activation temperatures used in the Arrhenius expressions
above are chosen by trial and error to give the expected composition of the charge at the exit of
the kiln, with the exception of the activation energy for CaCO
3
decomposition which was taken
from Ref. [16]. We emphasize that these reactions are ad hoc models. In reality, chemical re-
actions in the solid phase occur in the interface between crystals of the various species and are
diusion limited [17], but detailed treatment of clinker chemistry is beyond the scope of this
work.
Reaction (III), considered to be the nal step in clinker formation, occurs at high temperatures
and only at the presence of liquid phase. Melting of the clinker was modelled by also solving for
the liquid fraction, Y
fus
,
(I) CaCO
3
CaOCO
2
k = 10
8
exp(175728aRT)
(II) CaO2SiO
2
C2S k = 10
7
exp(240000aRT)
(III) CaOC2S C3S k = 10
9
exp(420000aRT)
(IV) 3CaOAl
2
O
3
C3A k = 10
8
exp(310000aRT)
(V) 4CaOAl
2
O
3
Fe
2
O
3
C4AF k = 10
8
exp(330000aRT)
62 E. Mastorakos et al. / Appl. Math. Modelling 23 (1999) 5576
L
fus
J
cl
d(.
cl
Y
fus
)
dx
= RXHXSX of EqX (4) [ [ (7)
if T
cl
becomes greater than or equal to the fusion temperature T
fus
, with T
fus
=1560 K. An upper
limit of Y
fus
=0.3 was taken [17], above which Eq. (7) was neglected.
The ordinary dierential equations, Eqs. (4)(7), were integrated with the Gear method. The
nite volume code for the kiln temperature (Section 2.3) and the code for the clinker were solved
iteratively (Fig. 4), a procedure that is necessary because of the coupling through the wall-clinker
heat ux q
wcl
(x, u).
2.5. Final coupling
The full problem (gas, kiln, charge) was solved by an iteration scheme. Starting from a
uniform distribution of T
RAD
, the gas phase was solved to give the axial distributions of q
rad
(x)
and q
conv
(x). These were then used in Eqs. (3)(7), whose solutions resulted in a new estimate of
T
RAD
(x) through Eq. (2). The FLOW3D-RAD3D run was then repeated with the new boundary
condition, introduced into the program in the form of coecients of a sixth order polynomial t.
An under-relaxation factor of less than 0.5 between the successive estimates of T
RAD
(x) was
found to be necessary, due to the large sensitivity of the radiation heat ux to the wall tem-
perature. Typically, about 10 iterations between these steps were needed to achieve convergence
of the inner wall temperature (dened when successive iterates did not change by more than 40
K), amounting to a total CPU time of the order of 10 days on an HP 715 workstation. Fig. 5
shows the convergence history of T
RAD
(x) starting from a uniform value (1600 K) to the nal
prole.
Fig. 4. The KILN-CLINKER package for wall temperature and charge temperature and composition.
E. Mastorakos et al. / Appl. Math. Modelling 23 (1999) 5576 63
2.6. Kiln conditions
Operating conditions and values for some of the parameters used are given in Tables 13.
These values are typical of cement rotary kilns.
3. Results and discussion
In this section we present, rst, details of the predicted aerodynamic patterns together with
information on the ame structure. Presentation of the temperature distributions in the kiln re-
fractories and the evolution of clinker composition follows. The section closes with a discussion
on overall energy balances for the kiln and on possible model improvements.
3.1. Flame structure
Fig. 6 shows a contour plot of temperature, with the radial coordinate exaggerated for clarity.
The coal ignites about 1 m from injection and slightly o-axis and the maximum ame temper-
ature (of about 2400 K) is reached about 40 m downstream, in accordance with experience [15].
The longest time for complete particle burnout was about 1.4 s, corresponding to 45 m from the
burner. Radial proles of temperature (Fig. 7) show that there is signicant non-uniformity of the
Fig. 5. The distributions of T
RAD
for successive iterations between the FLOW-3D RAD-3D and KILN-CLINKER
code solutions.
64 E. Mastorakos et al. / Appl. Math. Modelling 23 (1999) 5576
temperature, which however diminishes downstream with the exit prole becoming essentially
at. It is also evident that there exists a thermal boundary layer, with the gas adjacent to the wall
being colder than the wall for up to 30 m from the coal nozzle. This implies that in this region heat
transfer by convection is negative (i.e. towards the gas), a point discussed in more detail in
Section 3.4.
Table 2
Clinker composition (mass fractions)
Farine Kiln inlet Kiln exit
m (kg/s) 50.374 39.815 32.775
T (K) A 1100 1785
CaCO
3
0.7947 0.40218 0
CaO 0 0.33801 0.0229
SiO
2
0.1434 0.18143 0
Al
2
O
3
0.0349 0.0442 0
Fe
2
O
3
0.0270 0.03416 0
C2S 0 0 0.1808
C3S 0 0 0.5981
C3A 0 0 0.0731
C4AF 0 0 0.1242
Calcination factor 0 0.6 1.0
The calcination factor is the percentage of CaCO
3
in the farine that has been transformed into CaO.
Table 1
Air and coal-particle inlet conditions
Axial Transport Swirl Secondary Coal
m (kg/s) 2.253 1.759 2.910 45.930 4.0
U (m/s) 77.1 36.5 76.1 12.73 36.5
V (m/s) A20.7 0 63.9 0 0
W (m/s) 0 0 112.8 0 0
T (K) 318 383 318 1273 383
Table 3
Material properties and other parameters
x (rad/s) 0.5
V
cl
(m/s) 0.035
T

(K) 300
h
sh
(W/m
2
K) 30
h
wcl
(W/m
2
K) 500
e
w
Y e
cl
0.9
e
sh
0.8
c
p
(clinker) (kJ/kg K) 1.5
.
cl
(kg/m
3
) 1200
L
fus
(kJ/kg) 418.4
c
p
(wall) (kJ/kg K) 1.5
.
w
(kg/m
3
) 16003000
k
w
(W/m K) 0.63.0
Coal heat release (kJ/kg) 25475
E. Mastorakos et al. / Appl. Math. Modelling 23 (1999) 5576 65
There exists a recirculation zone with length about one burner diameter immediately down-
stream of the burner exit (Fig. 8), which does not serve for ame stabilisation as in swirl-stabilised
ames [7], because here the ame ignites further downstream. However, it may be important when
liquid fuel is being used, and so the ne detail of the grid close to the burner should be em-
phasized. The rst few metres from the burner involve very high shear and high turbulent energy
Fig. 7. Radial proles of the gas temperature at various locations along the kiln.
Fig. 6. Gas temperature contours. The radial coordinate has been exaggerated for clarity.
66 E. Mastorakos et al. / Appl. Math. Modelling 23 (1999) 5576
production and this is yet another reason to emphasize grid renement. Typical predicted values
of k and e in the region of volatiles combustion (x =10 m, r =1 m) are 24.3 and 142 m
2
/s
3
, re-
spectively. The large-scale turbulent time is then 171 ms and the Kolmogorov time scale 1.1 ms.
Combustion of volatiles occurs at a time scale of the order of 0.1 ms (typical of hydrocarbon
fuels), which is an order of magnitude shorter than the small-scale turbulent time in the kiln.
Therefore it does not seem necessary to include nite-rate kinetics for combustion in this ow and
the ``mixed-is-burnt'' approximation is reasonable.
3.2. Kiln temperature distributions
The calculated gas temperature at the centreline, the temperature T
RAD
(x), and the clinker
temperature are compared in Fig. 9. The peak gas temperature is located between 25 and 40 m,
Fig. 8. Velocity vectors in the near-burner region. The shade of each vector denotes the magnitude of the swirl com-
ponent of velocity.
E. Mastorakos et al. / Appl. Math. Modelling 23 (1999) 5576 67
where the refractory inner surface temperature also peaks. The clinker is, surprisingly, hotter than
the wall for the last few metres before it exits the kiln. The heat ux incident on the refractory by
radiation is between 1 and 2 orders of magnitude higher than that by convection (Fig. 10) and the
total heat transfer for the rst 10 m of the kiln is towards the gas. The signicance of this ob-
servation is discussed later.
The energy ux due to convection is found to be a very small part of the overall energy ux
from the ame to the kiln (Fig. 10). The small contribution of convection predicted here is
contrary to the conclusion of Ref. [11]. In that work, the gas temperature used for the calculation
of convective heat transfer was an average over the kiln cross section, and hence much higher than
the temperature in the vicinity of the wall due to the contribution of the ame which lies close to
the axis. Here, we have used the more accurate heat transfer coecient based on local values of
temperatures and gas velocity and turbulent kinetic energy, and we therefore expect our result to
be more exact.
The predicted wall temperature is inhomogeneous in all directions. Fig. 11 shows that, as
the kiln rotates, the wall is successively heated by the gas when exposed to the ame and cools
by supplying heat to the clinker. This is so for most of the kiln length with a typical angular
temperature variation of about 100 K. In contrast, close to the burner, the wall gains heat
from the clinker for (0 ` u ` p/2), and loses heat to the gas when exposed (p/2 ` u ` 2p).
Fig. 9. Axial distributions of the gas temperature at the centreline, the ctitious temperature T
RAD
(Eq. (2)), and the
clinker temperature.
68 E. Mastorakos et al. / Appl. Math. Modelling 23 (1999) 5576
It may happen that the wall and clinker temperatures are equal and then there is no azimuthal
variation (Fig. 11, x =17.5 m). Note that even though this temperature variation may be
considered small, one needs to calculate the full 3-D inner wall temperature distribution to
be in a position to calculate the heat ux from the wall to the clinker (which occurs in the
range 0 ` u ` p/2).
Fig. 12 shows that the region of variations in the angular direction extends only about 1 cm
into the wall (Fig. 12(b)) because of the large specic heat capacity of the solid, with the outer
(shell) temperature independent of u. The wall temperature radial distribution depends on the
axial distance because the gas temperature, the incident radiation, and the properties of the re-
fractory material change. Accurate predictions require an accurate knowledge of the thickness of
the clinker coating attached on the refractories; here, we have treated this coating as a material of
properties similar to those of clinker and with uniform thickness. The reliability of the predictions
will be improved, however, by using experimental data concerning the real physical properties and
the thickness distribution of this coating layer.
Finally, Fig. 13 shows that the outer shell temperature peaks at the ame region and follows,
approximately, experimental trends [15]. An exact comparison is not meaningful because the
outer kiln shell has variable steel thickness, emissivity (due to shell colouration), and heat transfer
coecient (due to the spacing of the blowers), and the kiln has variable refractory thickness (due
to erosion by the clinker), contrary to the assumptions used here. The full-scale kiln also shows
Fig. 10. Axial distributions of the radiative and convective heat ux from the gas to the kiln-clinker system calculated by
FLOW3D RAD3D.
E. Mastorakos et al. / Appl. Math. Modelling 23 (1999) 5576 69
angular uctuations of the shell temperature up to 100 K attributed to dierential coating and
refractory erosion [15]. We believe therefore that the agreement in Fig. 13 is as good as can be
expected considering the complexity of the real kiln.
The predictions presented in this section support the following picture of heat transfer path-
ways inside the kiln. For most of the kiln length, the charge is being heated by radiation from the
ame and by heat conduction from the walls, as the charge is colder than the refractories. As the
kiln rotates, the refractory gains heat when exposed to the ame and then supplies it to the clinker
(Fig. 11). This ``regenerative'' action of the wall has been postulated in Refs. [9,10] and was re-
produced in the present results. However, an opposite energy ow occurs close to the burner
(Fig. 11, small x). The gases here are not yet suciently hot and hence do not supply energy to
neither the refractories nor the charge. In this region, the walls gain heat by the oncoming charge
and hence stay hotter than they would be in the absence of the solids. The walls and the charge
supply heat to the gas by both convection and radiation. We can identify this as a ``negative
regenerative'' action, and is sustained by the energy absorbed by the clinker in the higher tem-
perature region of the kiln (x > 15 m). Overall, the clinker absorbs heat at x > 15 m, and gives a
small part of it back to the gas at 0 ` x ` 15 m.
This interaction is not straightforward and could not have been anticipated easily. It implies
that modications to the ame, for example by substituting solid by liquid fuel, may have a
profound eect on the nal clinker temperature by altering the heat ux distributions. The detail
oered by the present comprehensive model will be helpful in evaluating such changes on the kiln.
Fig. 11. Azimuthal distributions of the inner surface wall temperature for various axial locations.
70 E. Mastorakos et al. / Appl. Math. Modelling 23 (1999) 5576
Fig. 12. (a) Radial distributions of the wall temperature at x =22.5 m for various u. (b) The region close to the inner
surface shown in more detail.
E. Mastorakos et al. / Appl. Math. Modelling 23 (1999) 5576 71
3.3. Clinker temperature and composition
The clinker temperature (Fig. 9) peaks approximately at the axial location where the highest
ame temperature is attained, and the clinker exists the kiln at about 1780 K, a value close to
experimental measurements in cement kilns [15]. The initial and nal clinker compositions are
shown in Table 2 and are close to operational values in the real kiln [15]. Axial distributions of the
various clinker constituents (Fig. 14) show that it takes about 25 m from the solids inlet for
complete calcination and a further 10 m for the creation of C2S, C3A, and C4AF. The rst liquid
phase is found at x =50 m and liquefaction is completed very shortly afterwards, a prediction that
is consistent with experience [17]. The clinkerisation reaction (R-III) has been assumed in the
model to begin when liquid appears, and it is evident from Fig. 14 that clinkerisation requires
almost the whole of the remaining length to be completed. The predictions are, on the whole,
consistent with experience from cement kiln operation, which implies that the chemical and
physical parameters used here have realistic values.
3.4. Global energy balances
Global energy balances can be readily calculated from the solutions for the gas phase (by
FLOW-3D and RAD-3D) and the kiln-charge system, and are presented in Table 4. The CFD
code, in conjunction with the radiation module, conserve energy to about 2%. This error, con-
sidered small, is attributed mostly to the error caused by the zone discretisation in RAD-3D and
Fig. 13. The outer kiln temperature compared with measurements from a full-scale cement kiln [15].
72 E. Mastorakos et al. / Appl. Math. Modelling 23 (1999) 5576
the nite number of particles in the Monte-Carlo calculation and can be improved at the expense
of CPU time. The accuracy of the kiln-clinker calculation is worse, with an error of about 10% of
the energy input to the kiln-clinker system (Q
rad
+ Q
conv
). This may be due to the cumulative error
Fig. 14. Mass fraction distributions of the various chemical constituents and the liquid fraction of the charge.
Table 4
Global energy balances for the CFD grid and the kiln-clinker combination
Gas (MW)
Q
rad
,1 A2.47
Q
rad
,2 A2.72
Q
rad
A57.12
Q
conv
0.04
Q
coal
101.2
DH
gas
41.25
Balance 2.32
Kiln-clinker
Q
rad
57.12
Q
conv
A0.04
Q
loss
A10.45
DH
cl
40.99
Balance 5.64
Refer to Fig. 2 for the denition of the energy ows.
E. Mastorakos et al. / Appl. Math. Modelling 23 (1999) 5576 73
caused by the long transient followed until a converged solution is obtained to Eq. (3), and may
be improved by a more accurate implicit solver.
Some interesting conclusions concerning energy use in the cement rotary kiln can be formed by
the results in Table 4. About 40% of the energy released by combustion is needed for charge
heating and clinker formation and about 10% is lost through the refractories to the atmosphere.
Most of the remaining essentially ows out of the kiln with the exhaust gases, a part of which is
recovered in the pre-calciner and cyclones outside the kiln. It is clear that it is impossible to
predict accurately the local gas temperature, and hence pollutant formation in future studies,
without resort to the comprehensive model presented here that treats in detail the inner kiln wall
and charge temperatures.
3.5. Discussion
The results of the comprehensive model for rotary cement kiln operation presented here are
reasonable and reproduce experimentally observed trends. The major advantage of this work over
previous modelling eorts lies in the inclusion of most physical processes occurring in the kiln. In
particular, the inner wall temperature, a quantity of the utmost importance in determining the gas
temperature and the heat ux to the clinker and hence clinker formation, was predicted, rather
than prescribed using experimental data. This particular feature makes the present model truly
predictive.
We have included a method to reduce the full 3-D problem to an ``equivalent'' axisymmetric
one for the aerodynamics (Eq. (2)), which allows solutions to be obtained with current work-
stations. The modular nature of the model, i.e. separate codes for the aerodynamics, radiation,
kiln, and charge, implies that it can be used also for other rotary kiln applications (e.g. incin-
eration and drying) with modications only to the appropriate module. For example, incineration
of solid waste may be modelled with the present codes but with appropriate chemistry introduced
in Eqs. (5) and (6).
Detailed comparison with experimental data is very dicult because there are hardly any
available measurements and was hence limited to empirical data provided by a local cement
company [15]. The comparison, as indicated previously, showed that the exit clinker composition
and temperature are within industrial experience (Table 2), the exhaust gas composition is close to
what is measured at the factory stack (after allowing for ``false air'' dilution), and the kiln shell
temperature is within the range of the measurements (Fig. 13). This agreement suggests that the
model is an accurate representation of the process.
Improvements are needed in various parts of the model to achieve predictions of higher ac-
curacy. The correct thickness of the refractories (i.e. refractory plus attached clinker) should be
established, because this aects the heat loss through the kiln walls to the surroundings and hence
the outer shell temperature. Detailed comparison should be made with shell temperature mea-
surements in a kiln with new refractories (so that non-uniformities is the coating thickness are
avoided). Proper values of the physical properties (heat capacities, densities, conductivities) of the
wall materials should be used; the largest uncertainty concerns the assumed properties of the
clinker coating. The emissivity of the inner wall surface and the absorption coecient of the gases
should be examined more closely, as they aect the radiative heat ux incident on the kiln, and
hence can aect the wall temperature. The temperature of the clinker depends on the specic heat
capacity used and so extra care should be taken at its accurate evaluation. Knowledge of in-ame
local temperature and species composition would also be very useful to validate the CFD code,
although measurements in such a hostile environment are clearly very dicult to accomplish.
Finally, clinker chemistry and charge motion are areas where improvements could be made.
74 E. Mastorakos et al. / Appl. Math. Modelling 23 (1999) 5576
Notwithstanding all these potential improvements and the uncertainties involved in the
models, the overall behaviour is at least qualitatively correct because all the energy paths in the
kiln have been modelled to adequate detail. Important quantities such as clinker exit composition,
shell temperature, and exhaust gas composition are predicted to acceptable accuracy. It is ex-
pected that the model will prove very useful to industry in many situations, such as with alter-
ations in the burner, fuel type, quality and quantity, in the level of pre-calcination (Table 2) or
solids ow rate, etc. The kiln operator must ensure that the nal clinker composition is still ac-
ceptable and the present comprehensive model can be of help in this direction.
4. Conclusions
Clinker formation in coal-red rotary cement kilns under realistic operation conditions has
been modelled with special emphasis given to the heat exchange between the coal ame and the
kiln, the heat exchange between the kiln and the counter-owing solids, and the chemical reac-
tions that transform the solids into the nal product (the clinker). An axisymmetric CFD code
(the commerical package FLOW-3D) that includes a Monte-Carlo method for radiation has been
used for the gaseous phase. The temperature of the kiln wall has been calculated with a nite-
volume heat conduction code, and the species and energy conservation equations for the clinker
have also been formulated and solved. An iterative procedure between the predictions for the
gaseous temperature eld, the radiative heat ux to the wall, and the kiln and clinker tempera-
tures was used to calculate the distribution of the inner wall temperature explicitly, in contrast to
previous modelling eorts that used such information from experiments. Integration over the
tangential coordinate resulted in an axial distribution of an ``eective'' inner wall temperature,
used as the necessary boundary condition to the CFD code. This procedure makes the calculation
of heat ow to the clinker possible, and also allows the approximate treatment of a 3-D problem
with an axisymmetric CFD code.
The results show that radiation accounts for most of the heat transfer between the gas and the
kiln walls, while the heat loss through the refractories to the environment accounts for about 10%
of the heat input. The chemical reactions and heating of the charge absorb about 40% of the
energy of combustion; it is hence imperative that these are accounted for. The predictions are
consistent with trends based on experience and available measurements from a full-scale cement
kiln.
Acknowledgements
This work has been funded by the Greek Secretariat for Science and Technology, Project
EPET-II/649. Sincere thanks are due to Mr.P. Panagiotopoulos of TITAN S.A., Greece, for
guidance and data concerning cement kilns.
References
[1] S.R. Turns, An Introduction to Combustion, Concepts and Applications, McGraw-Hill, New York, 1996.
[2] V. Johansen, T.V. Kouznetsova, Clinker formation and new processes, Presented at the Ninth International
Congress on the Chemistry of Cement, India, 1992; also RAMBOLL Bulletin No. 42, 1993.
[3] Basel Convention, UNEP Document No. 93-7758, 1993.
[4] N.C. Markatos, Mathematical modelling of single and two-phase ow problems in the process industries, Revue
de l'Institut Franc ais du Petrole 48 (1993) 631.
E. Mastorakos et al. / Appl. Math. Modelling 23 (1999) 5576 75
[5] T. Avgeropoulos, J.P. Glekas, C. Papadopoulos, Numerical simulation of the combustion aerodynamics inside a
rotary cement kiln, in: Pilavachi (Ed.), Energy Eciency in Process Technology, Elsevier, London, 1993, p. 767.
[6] F.C. Lockwood, B. Shen, T. Lowes, Numerical study of petroleum coke red cement kiln ames, Presented at the
Third International Conference on Combustion Technologies for a Clean Environment, Lisbon, 1995.
[7] F.C. Lockwood, B. Shen, Performance predictions of pulverised-coal ames of power station furnace and cement
kiln types, Twenty-Fifth Symposium International on Combustion, The Combustion Institute, 1994 p. 503.
[8] P.V. Barr, J.K. Brimacombe, A.P. Watkinson, A heat-transfer model for the rotary kiln: Part II, development of
the cross-section model, Metallurgical Transactions B 20B (1989) 403.
[9] V. Frisch, R. Jeschar, Possibilities for optimizing the burning process in rotary cement kilns, Zement-Kalk-Gips 36
(1983) 549.
[10] A.A. Boateng, P.V. Barr, A thermal model for the rotary kiln including heat transfer within the bed, Int. J. Heat
Mass Transfer 39 (1996) 2131.
[11] M.G. Carvahlo, T. Farias, A. Martius, A three-dimensional modelling of the radiative heat transfer in a cement
kiln, in: Carvahlo et al. (Eds.), Combustion Technologies for a Clean Environment, Gordon and Breach, London,
1995, p. 146.
[12] H.A. Spang, A dynamic model of a cement kiln, Automatica 8 (1972) 309.
[13] CFDS, FLOW-3D Users Manual, AEA Harwell, UK.
[14] E. Mastorakos, J.J. McGuirk, A.M.K.P. Taylor, The origin of turbulence acquired by heavy particles in a round,
turbulent jet, Part. Part. Syst. Charact. 7 (1990) 203.
[15] P. Panagiotopoulos, TITAN S.A. Cement Company, Personal communication, 1996.
[16] M.S. Murthy, B.R. Harish, K.S. Rajanandam, K.Y. Ajoy Pavan Kumar, Investigation on the kinetics of thermal
decomposition of calcium carbonate, Chem. Eng. Sci. 49 (1996) 2198.
[17] V. Johansen, Cement production and chemistry, Presented at the Symposium on Cement Manufacturing and
Chemistry, Anaheim, November 1989; also RAMBOLL Bulletin No. 41, 1993.
76 E. Mastorakos et al. / Appl. Math. Modelling 23 (1999) 5576