Mathematical modelling of pulverized coal furnaces

A. Berm udez
1
, J. L. Ferrn
1
, A. Li nan
2
and L. Saavedra
1
1
Departamento de Matematica Aplicada, Universidad de Santiago de Compostela
15782 Santiago de Compostela, Spain
2
E.T.S. Ingenieros Aeronauticos, Universidad Politecnica de Madrid
Pl. Cardenal Cisneros 3, 28040 Madrid, Spain
Monografas de la Real Academia de Ciencias de Zaragoza 34, 2750, (2010).
1 Introduction
The goal of this paper is to show how mathematics and computational science can
help to design not only the geometry but also the operation conditions of dierent parts
of a pulverized coal power plant as the one shown in Figure 1.
Figure 1. Sketch of a coal red power plant
27
In the last few years, the development of mathematical models and computational
tools, including high performance computers and commercial computational uid dynam-
ics (CFD) packages, has allowed introducing the numerical simulation as an important
tool to analyze the phenomena occurring in some dierent installations of a power plant.
For example, in Figure 2 the ow of secondary air dragging pulverized coal particles
has been modelled, whereas in Figure 3 some numerical results corresponding to the
simulation of erosion in the system of transport of ashes to silos are shown.
Figure 2. System of coal input to the boiler
Figure 3. Simulation of erosion
The main scope of this paper is to model and numerically solve the physico-chemical
28
processes occurring in the interior of the boiler, which is the part of the furnace where
combustion of pulverized coal particles takes place (it can be seen at the center of Figure
1). CFD simulations can help to achieve a better performance of the installation in order
to maximize the energetic eciency, to minimize the NO
x
formation, to control soiling
phenomena or to assure a stable and sustainable ignition.
2 Mathematical modelling
In this section we present the mathematical model for a turbulent reactive two-phase
ow taking place in the interior of the boiler. The two phases correspond to the gas
mixture and to the solid coal particles. They will be described in an Eulerian-Lagrangian
framework.
The gas phase model includes the conservation equations describing a steady, com-
pressible, turbulent, radiant and reactive uid. This uid will be a mixture of O
2
, CO
2
,
H
2
O, V
(g)
, CO, SO
2
, H
2
and N
2
.
The solid phase model must include the motion equation of a single particle and
expressions for the sources from the solid phase to the gas one.
2.1 Gas phase model
We consider the standard equations for mass and momentum conservation, for the
turbulence and for the thermal radiation, namely,
(v) = f
m
, (1)
(v v) +p = g + f
m
v
s
, (2)
=
e
_
v +v
t
_

2
3

e
( v)I, (3)
v k [( +
t
) k] = G
k
, (4)
v
__
+

t
1.3
_

_
= 1.44

k
G
k
1.92

2
k
, (5)

x
I + (a +
s
)I

s
4
_
S
2
(

, )I(x,

)d

= aI
b
, (6)
with
t
= 0.09
k
2

and G
k
=
t
v + (v)
t

2
.
Furthermore, we assume the low Mach number approximation, that is, we impose
=
p
RT
(7)
where p denotes the average ambient pressure which is supposed to be constant in this
state equation.
29
2.2 The combustion model
The combustion of coal particles, including devolatilization and char oxidation has
been treated extensively in the literature (see, for example, [1] or [7]). The BFL combus-
tion model introduced in Berm udez et al [2] considers a simplied kinetic model consisting
of the following physico-chemical process (reactions 1 to 5 within the porous particles and
reactions 6 to 8 in the gas phase):
1 CO
2
+ C
(s)
2 CO + (q
1
)
2
1
2
O
2
+ C
(s)
CO + (q
2
)
3 H
2
O + C
(s)
CO + H
2
+ (q
3
)
4 V
(s)
V
(g)
+ (q
4
)
5 H
2
O
(s)
H
2
O
(g)
+ (q
5
)
6 CO +
1
2
O
2
CO
2
+ (q
6
)
7 V
(g)
+
1
O
2

2
CO
2
+
3
H
2
O +
4
SO
2
+ (q
7
)
8 H
2
+
1
2
O
2
H
2
O + (q
8
)
where indices s (respectively g) refers to the solid phase (respectively, to the gas phase).
The q
i
denotes the heat released in the i-th reaction per unit of gasied mass.
Figure 4. Scheme of the combustion of a coal particle
For simplicity, in this paper we will consider the volatiles mixture represented by one
single molecule,
V
(g)
= C

1
H

2
O

3
S

4
, (8)
of molecular mass M
vol
, where coecients
1
,
2
,
3
and
4
are deduced from the ultimate
analysis of the coal.
The stoichiometric coecients
i
are calculated in terms of the volatiles composition
using the expressions:
1
= (2
1
+
2
/2 + 2
4

3
)/2,
2
=
1
,
3
=
2
/2 and
4
=
4
.
30
For char oxidation we have considered a reduced mechanism including the three overall
reactions 1, 2 and 3, of rst order with respect to the local mean concentration of CO
2
, O
2
and H
2
O, measured by the product of the local gas density,
g
, times the mass fractions.
We have considered the reactions to be of zero order with respect to the local char density,

C
. Obviously these three reaction rates must be equated to zero when the carbon of the
particle is completely consumed, and therefore a Heaviside function is also included as a
factor. More precisely, the local homogenized reaction rates per unit volume within the
particle are modelled using global Arrhenius laws of the form
w
1
= B
1
e
E
1
/RT

g
Y
CO
2
H(
C
), (9)
w
2
= B
2
e
E
2
/RT

g
Y
O
2
H(
C
), (10)
w
3
= B
3
e
E
3
/RT

g
Y
H
2
O
H(
C
), (11)
where Y
CO
2
, Y
O
2
and Y
H
2
O
are the local mass fractions of CO
2
, O
2
and H
2
O in the gas
lling the porous interstices and
C
is the density of carbon in the form of char. For the
generation of volatiles and moisture evaporation we shall use the simple laws
w
4
= B
4
e
E
4
/RT

V
, (12)
w
5
= B
5
e
E
5
/RT

H
2
O
, (13)
where
V
and
H
2
O
are the local values within the coal particle of the density of volatiles
and H
2
O remaining in condensed form. The moisture generation is, for simplicity, here
described with a kinetic model similar to the pyrolysis model of volatiles generation. We
have chosen the rates of water vapour and volatiles generation to be of rst order with
respect to
H
2
O
and
V
.
The fundamental assumption to obtain this model is the Burke-Schumann hypothesis.
Therefore, the gas phase reactions 6 to 8 are supposed to be frozen or to occur with
innitely fast velocity in a gaseous thin diusion ame that can be placed either inside the
particle, or in the gas surrounding the particle or in the gas far from the particle. The type
of combustion that will occur will depend on the temperature and local concentrations of
oxygen, carbon monoxide, volatiles and H
2
in the gas environment; the model will decide
in which situation the particle is.
The combustion model that we present in this paper consists of two coupled models:
the gas phase model including the mass and energy conservation equations and the solid
phase model. On the one hand the gas phase model determines the atmosphere in which
particles are burnt and, on the other hand, the solid phase model provides mass and
energy sources to the gas.
31
2.2.1 Gas phase Model
Let L
g
be the dierential operator dened by
L
g
(u) =
(
g
u)
t
+ (
g
uv
g
) (
g
Du), (14)
where D is a gas phase diusion coecient which, for simplicity, will be considered to be
the same for all species and equal to the thermal diusivity (Lewis number equal to one).
Then, the conservation equations of the gaseous species are given by

g
t
+ (
g
v
g
) = f
m
, (15)
L
g
(Y
g
O
2
) = f
m
O
2

4
7
w
6

32
1
M
vol
w
7
8w
8
, (16)
L
g
(Y
g
CO
2
) = f
m
CO
2
+
11
7
w
6
+
44
2
M
vol
w
7
, (17)
L
g
(Y
g
H
2
O
) = f
m
H
2
O
+
18
3
M
vol
w
7
+ 9w
8
, (18)
L
g
(Y
g
SO
2
) = f
m
SO
2
+
64
4
M
vol
w
7
, (19)
L
g
(Y
g
CO
) = f
m
CO
w
6
, (20)
L
g
(Y
g
V
) = f
m
V
w
7
, (21)
L
g
(Y
g
H
2
) = f
m
H
2
w
8
, (22)
L
g
(h
g
T
) = f
e
+ q
6
w
6
+ q
7
w
7
+ q
8
w
8
q
rg
, (23)
where w
6
, w
7
and w
8
denote, respectively, the mass consumption rates per unit volume of
the volatiles, CO and H
2
, due to the chemical reactions 6, 7 and 8, taking place in the gas
phase; h
g
T
is the gas phase specic thermal enthalpy, which we shall for simplicity write
as c
p
T
g
considering the gas phase specic heat at constant pressure, c
p
, to be constant,
and q
rg
is the radiant heat ux vector.
The local homogenized reaction rates per unit volume within the particle could be
modelled by using global Arrhenius laws of the form:
w
6
=
2
g
Y
1/2
O
2
Y
CO
Y
1/2
H
2
O
B
6
e
E
6
/RT
,
w
7
=
1+
1
g
Y

1
O
2
Y
V
B
7
e
E
7
/RT
,
w
8
=
3/2
g
Y
1/2
O
2
Y
H
2
B
8
e
E
8
/RT
.
Nevertheless, as we are considering the Burke-Schumann hypothesis of innitely fast
gas phase reactions, which implies the non coexistence of CO, V and H
2
with O
2
, we do
not need to provide expressions for the gas phase reaction rates. We can obtain equations
32
without the gas phase reaction terms considering the following conserved scalars of Shvab-
Zeldovich type:

g
1
= Y
g
O
2

4
7
Y
g
CO

32
1
M
vol
Y
g
V
8Y
g
H
2
, (24)

g
2
= Y
g
CO
2
+
11
7
Y
g
CO
+
44
2
M
vol
Y
g
V
, (25)

g
3
= Y
g
H
2
O
+
18
3
M
vol
Y
g
V
+ 9Y
g
H
2
, (26)

g
4
= Y
g
SO
2
+
64
4
M
vol
Y
g
V
, (27)
H
g
= h
g
T
+ q
6
Y
g
CO
+ q
7
Y
g
V
+ q
8
Y
g
H
2
. (28)
Then from (15)-(23) we have the equations
L
g
(
g
1
) = f
m
O
2

4
7
f
m
CO

32
1
M
vol
f
m
V
8f
m
H
2
, (29)
L
g
(
g
2
) = f
m
CO
2
+
11
7
f
m
CO
+
44
2
M
vol
f
m
V
, (30)
L
g
(
g
3
) = f
m
H
2
O
+
18
3
M
vol
f
m
V
+ 9f
m
H
2
, (31)
L
g
(
g
4
) = f
m
SO
2
+
64
4
M
vol
f
m
V
, (32)
L
g
(H
g
) = f
e
+ q
6
f
m
CO
+ q
7
f
m
V
+ q
8
f
m
H
2
q
rg
. (33)
Owing to the Burke-Schumann hypothesis we can determine two regions of the ow
eld: one region with oxygen,
O
, and another one without oxygen,
F
. The gas phase
reactions 6 to 8 take place in a diusion-controlled ame where the volatiles, CO and H
2
are burning with oxygen separating
O
from
F
. This surface is called
F
(see Figure 5
for details).
Figure 5. Diusion ame
Once we have solved (29),
g
1
will determine the regions of the domain and therefore
the way to recover the mass fractions and the enthalpy from the Shvab-Zeldovich variables:
33
1. If
g
1
> 0 we are in the domain
O
, this implies that Y
g
V
= Y
g
H
2
= Y
CO
= 0 and using
(24)-(28) we obtain Y
g
O
2
=
g
1
, Y
g
CO
2
=
g
2
, Y
g
H
2
O
=
g
3
, Y
g
SO
2
=
g
4
and h
g
T
= H
g
.
2. If
g
1
< 0 we are in the domain
F
so Y
g
O
2
= 0 and w
6
= w
7
= w
8
= 0. In this case
to determine the mass fractions and the enthalpy we need to solve two additional
equations of (21)-(23) without gas phase reaction terms.
2.2.2 Particle Gasification Model
The solid phase model is based on a Lagrangian computation of the temperature and
density of each coal particle throughout its trajectory. We assume that the particles are
spherical and its radius and its ash density remain constant during the combustion. So
the model will be valid for particles with high content of ash.
The density of a coal particle is determined by:

p
=
H
2
O
+
V
+
C
+
ash
. (34)
The evolution of
H
2
O
,
V
and
C
, with radial coordinate r and time t, will be given by

V
t
= w
4
,

H
2
O
t
= w
5
,

C
t
= w
C
, (35)
in terms of the mass rates, per unit volume and time, of generation of volatiles w
4
, water
vapour w
5
and char gasication w
C
, given by equations (9)-(13).
To write the particle temperature equation we take into account that the heat conduc-
tion time across the particle is short compared with the diusion time so we can consider
the temperature spatially uniform. This temperature is given by the equation
4
3
a
3

p
c
s
dT
p
dt
= 4a
2
(q

p
+ q

r
) +
_
a
0
_
8

i=1
q
i
w
i
_
4r
2
dr, (36)
where 4a
2
q

p
and 4a
2
q

r
are the heat ux by conduction and by radiation, respectively,
which have the following expressions
q

p
= k
dT
dr |
r=a
+
, q

r
=
p
_
1
4
_
S
2
I(x, )d T
4
p
_
, (37)
where I(x, ) is the radiation intensity in the direction at the position of the particle,
denoted by x,
p
is the particle emissivity and S
2
is the unit sphere.
Due to the assumed uniformity of the temperature within the particle and the expres-
sions (12) and (13), the rates of generation of water vapour and volatiles are spatially
uniform and known functions of T
p
and we can write the evolution with time of
V
and

H
2
O
as
d
V
dt
= B
4
e
E
4
/RTp

V
, (38)
d
H
2
O
dt
= B
5
e
E
5
/RTp

H
2
O
, (39)
34
We cannot write a similar expression for the third equation in (35) because the depen-
dence of
1
,
2
and
3
on Y
CO
, Y
O
2
and Y
H
2
O
, which are not uniform within the particle.
In fact, if we want to know the evolution of
C
, the values of Y
CO
, Y
O
2
, Y
H
2
O
have to be
calculated using mass conservation equations for the gas phase within the pores of the
particle for each given value of T
p
and the mass fractions on the surface of the particle.
In order to simplify the treatment of the eects of the gasication char reactions 1, 2
and 3, Damk ohler numbers have been dened for these reactions as
Da
i
=
a
2
D
e
B
i
e
E
i
/RTp
, i = 1, 2, 3. (40)
We consider that the activation energies of these heterogeneous reactions are large enough,
so that Da
i
are of order unity only a very short time. Damk ohler numbers will determine
the stage of char gasication:
First stage (Da
i
1, i = 1, 2, 3): kinetically-controlled reactions. Char gasication
reactions can be considered frozen (
1
=
2
=
3
= 0) and there is no production
of CO or H
2
within the particle.
So, to determine the particle density in the rst stage, we only have to solve the
equations (38) and (39). In this stage, the equation for the particle temperature
(36) which can be simplied using that the gas phase reaction 7 is assume to be
innitely fast. Thus, if we compute the value
1
at the particle surface

s
1
= Y
s
O
2

32
1
M
vol
Y
s
V
, (41)
we have two dierent cases:
If
s
1
< 0 the reaction 7 takes place in a diusion ame outside the particle and
prevents the oxygen to reach the surface. In this case equation (36) becomes
4
3
a
3

p
c
s
dT
p
dt
= 4a
2
(q

p
+ q

r
) +
_
a
0
(q
4
w
4
+ q
5
w
5
) 4r
2
dr, (42)
If
s
1
> 0 the oxygen reaches the particle and reacts with the volatiles which
cannot leave the particle. In this case the energy equation simplies to
4
3
a
3

p
c
s
dT
p
dt
= 4a
2
(q

p
+ q

r
) +
_
a
0
(q
4
+ q
7
)w
4
+ q
5
w
5
4r
2
dr, (43)
Second stage (Da
i
1, i = 1, 2, 3): diusion-controlled reactions. Char gasication
reactions are very fast and they occur in a diusion ame at r = r
c
which implies that
char and CO
2
, O
2
and H
2
O cannot coexist. This limit case leads to the shrinking
core model.
The procedure which allows us to deduce the equations to be solved to determine

1
,
2
and
3
in the second stage, as well as the expressions for the sources to the
35
gas phase, has been done in [2]. There are only two cases in which
2
= 0, one in the
limit case in which the gasication of char reactions are kinetically controlled and
the other one when the oxygen reaches the particle core surface because of the small
particle size based Damk ohler numbers. Since we consider that this two cases are not
relevant in our simulations, we will suppose, for simplicity of the implementation,
that reaction 2 never takes place. Next, they are summarized taking into account
that they are dierent depending on the domain in which the particle is and the
position of the diusion ame with respect to the particle:
1. Particle in
F
(
g
1
< 0). In this case we have to solve
11
3

=
_
Y
g
CO
2
+
11
3

_
e

D
De
(1a/rc)
,
3
2

=
_
Y
g
H
2
O
+
3
2

_
e

D
De
(1a/rc)
,
(44)
where

i
=
m
i

g
aD
, i = 1, 3, 4, 5, (45)
and D
e
is the diusion coecient of the gas mixture inside the particle pores.
The temperature equation is
4
3
a
3

p
c
s
dT
p
dt
= 4a
2
(q

p
+ q

r
) + 4
g
aD(q
1

1
+ q
3

3
+ q
4

4
+ q
5

5
), (46)
where
q

p
=
k
ac
p
(h
g
T
h
s
T
)

e

1
. (47)
2. Particle in
O
(
g
1
> 0) and Y
s
O
2
> 0. The diusion ame moves into the
particle (i.e. r
c
< r
f
a). The equations to solve are
e

D
De
(a/r
f
1)+
= + 1,
11
3

_
22
3

+
11
3

+
44
2
M
vol

_
e

D
De
(a/r
f
a/rc)
=
_
Y
g
CO
2

11
3
_

+

3

44
2
M
vol

_
e

D
De
(1a/rc)
,
3
2

_
3
2

+
18
3
M
vol

_
e

D
De
(a/r
f
a/rc)
=
_
Y
g
H
2
O


5

18
3
M
vol

_
e

D
De
(1a/rc)
,
(48)
with given by
=
Y
g
O
2
8
3
_

+

3

_
+
32
1
M
vol

. (49)
36
The temperature equation is
4
3
a
3

p
c
s
dT
p
dt
= 4a
2
(q

p
+ q

r
) + 4
g
aD(q
1

1
+ q
3

3
+ q
4

4
+ q
5

5
+
14
3
q
6

1
+
7
3
q
6

3
+ q
7

4
+
1
6
q
8

3
) (50)
where q

p
is given by (47).
3. Particle in
O
(
g
1
> 0) and Y
s
O
2
= 0. The diusion ame is surrounding the
particle (i.e. r
f
> a). In this case we have to solve
11
3

=
_
Y
g
CO
2
+
11
3

+
_
22
3

+
11
3

+
44
2
M
vol

_
e

D
De
(1a/rc)
,
3
2

=
_
Y
g
H
2
O
+
3
2

+
_
3
2

+
18
3
M
vol

_
e

D
De
(1a/rc)
.
(51)
The temperature equation is
4
3
a
3

p
c
s
dT
p
dt
= 4a
2
(q

p
+ q

r
) + 4
g
aD(q
1

1
+ q
3

3
+ q
4

4
+ q
5

5
+
14
3
q
6

1
+
7
3
q
6

3
+ q
7

4
+
1
6
q
8

3
) (52)
where
q

p
=
k
ac
p

1
_
h
g
T
h
s
T
+
_
q
6
_
14
3

+
7
3

_
+ q
7

+ q
8
1
6

_
. (53)
In all the three cases, the position of the shrinking core is given by

0
C

g
aD
r
2
c
dr
c
dt
= (
1
+
3
). (54)
2.2.3 Sources to the gas phase
The purpose of solving the solid phase model is to obtain the sources of mass and
energy to the gas phase from coal particles gasication. Once we know this sources we
can solve equations (29)-(33) and obtain the temperature and composition of the gas
mixture.
The expressions for the sources of mass of each of the species due to one single particle
are:
37
1. In
F
(Y
g
O
2
= 0):
F
m
O
2
= F
m
SO
2
= 0, (55)
F
m
CO
2
=
4ak
c
p
_

11
3

1
_
, (56)
F
m
H
2
O
=
4ak
c
p
_

3
2

3
_
, (57)
F
m
CO
=
4ak
c
p
_
14
3

1
+
7
3

3
_
, (58)
F
m
V
=
4ak
c
p

4
, (59)
F
m
H
2
=
4ak
c
p
1
6

3
. (60)
2. In
O
(Y
g
O
2
> 0):
F
m
O
2
=
4ak
c
p
_

8
3
(
1
+
3
)
32
1
M
vol

4
_
, (61)
F
m
CO
2
=
4ak
c
p
_
11
3
(
1
+
3
) +
44
2
M
vol

4
_
, (62)
F
m
H
2
O
=
4ak
c
p
_

5
+
18
3
M
vol

4
_
, (63)
F
m
SO
2
=
4ak
c
p
64
4
M
vol

4
, (64)
F
m
CO
= F
m
V
= F
m
H
2
= 0. (65)
Both in
O
and in
F
the total source of mass is
F
m
=
4ak
c
p
(66)
and the source of energy is
F
e
= 4ak
_
(
c
s
c
p
T
p
T
g
)

e

1
+ (1 )
_
q
6
c
p
_
14
3

1
+
7
3

3
_
+
q
7
c
p

4
+
q
8
c
p
1
6

3
__
c
s
T
p
dm
p
dt
,
(67)
where
=
_

_
1 when
s
1
> 0 (ame inside the particle) or

g
< 0 (there is no oxygen in the vicinity),

1
when
g
1
> 0 and
s
1
< 0 (ame outside the particle),
The homogenized sources in the gas phase per unit volume and time, at point x, are
computed adding the contributions of each single particle that at some instant t is at
point x by the expression
f

(x) =
Ne

j=1
Np

i=1
q
j
p
ij
100
_
t
ij
f
0
F

ij
(t)(x x
ij
p
(t))dt (68)
38
where F

ij
(t) is the source of mass or energy of one particle of type i introduced through
inlet j, at instant t, x
ij
p
(t) is the position occupied by this particle at instant t, (x) is the
Dirac measure at point 0, t
ij
f
is the time needed for the particle to be completely burned
or to leave the furnace, q
j
is the mass ow of coal through the inlet j, p
ij
is the percentage
of particles of the type i through inlet j, and N
e
and N
p
are the number of inlets and
types of particles, respectively.
2.3 Particle Motion Model
To solve the solid phase model we need to follow each single particle along the domain.
We will suppose that the forces aecting the particle motion are only the drag force and
the gravity. We obtain the particle velocity by solving the ordinary dierential equation:
dv
p
dt
= F
A
(v
g
v
p
) +g, (69)
v
p
(0) = v
p0
, (70)
where F
A
is the drag force per mass unit which can be written as
F
A
=
3
16

p
a
2
C
D
Re. (71)
Here v
g
is the gas mixture velocity, v
p
the particle velocity, the gas viscosity, Re is the
Reynolds number relative to the particle, namely,
Re =
g
| v
g
v
p
|
2a

, (72)
and C
D
is the drag coecient which can be deduced from
C
D
=
_
1+0.15Re
0.687
Re/24
if Re 1000,
0.44 in other case.
2.3.1 Stochastic Particle Dispersion Modelling
In the particle motion model that we propose, coal particles only respond to the mean
uid velocity. This is correct if the ow is laminar but if we consider a turbulent ow
we must take into account the eect of a random uctuating velocity. For this purpose
we have coupled the particle motion model with a discrete random walk model. This
type of models specify the velocity as the sum of the mean uid velocity and a Gaussian
distributed random velocity uctuation with zero mean and a variance related to the
turbulent velocity scale computed from the turbulence model used. The random value of
the velocity is kept constant over an interval of time given by the characteristic lifetime
of the eddies, as made by [8].
39
Since we use the k turbulence model, the instantaneous uid velocity is calculated
as
v
g
= v
g
+
_

1
_
2k
3
,
2
_
2k
3
,
3
_
2k
3
_
, (73)
where
i
, i = 1, 2, 3, are normally distributed random numbers. The eddy time scale that
we consider is

e
= 0.15
k

log(r), (74)
where r is a uniform random number between 0 and 1. The particle eddy crossing time
is given by

c
= log
_
1
L
e
v
g
v
p

_
, (75)
where = 1/F
A
is the particle relaxation time and
L
e
= 0.09
k
1.5

(76)
is the eddy length scale.
As many random walk models, this model produces reasonable behavior in ows with
homogeneous turbulence. The suitability of many random walk models and others particle
dispersion models is analyzed in [6] or in [9] to include the eect of anisotropy.
We compute the trajectory, the density, the temperature and the sources to the gas
phase for a sucient number of representative particles (number of tries) with dierent
sizes, dropped from each cell of each inlet (as seen in Figure 6).
Figure 6. Trajectories of particles from a single cell
3 Algorithm and numerical methods
In order to numerically solve the dierent equations of the coupled two-phases problem
we implement a segregated procedure, leading to the algorithm shown in Figure 7.
40
Figure 7. Global algorithm
In the next subsections we describe the numerical methods used to solve the equations
of each problem in the sequential algorithm of Figure 7.
3.1 Momentum conservation equations
For solving the steady-state Navier-Stokes equations we use a P
1
-bubble/P
1
mixed
nite element combined with the method of characteristics and proceed to the elimination
of bubbles prior to the assembly by static condensation. Thus, we have to solve the
41
following boundary-value problem:
_

3v 4v X
t
+v X
2t
2t
+p = f in
v =
f
m

1
2t
(3 ln 4 ln X
t
+ ln X
2t
) in
v = v
e
on
e
v = 0 on
p
n p

n = 0 on
s
where X
t
(x) = (x, t; 0) and X
2t
(x) = (x, 2t; 0).
For that, we propose the following algorithm:
Step 0. Input v
0
y p
0
Step 1. Given v
0
and p
0
, v
1
and p
1
are the solution of the system:
_

_
_

1
t
v
1
zdx +
_

e
(v
1
+ (v
1
)
T
) : zdx
_

p
1
zdx =
_

f
1
zdx+
+
_

1
t
v
0
X
0
zdx
_

v
1
qdx =
_

g qdx
v
1
= v
e
on
e
v
1
= 0 on
p
with
g =
f
m

1
t
(ln ln X
0
),
p = p +
2
3

e
g.
Step n (n 1). Given v
n1
, v
n
, p
n1
and p
n
, v
n+1
and p
n+1
are the solution of the
system:
_

_
_

3
2t
v
n+1
zdx +
_

e
(v
n+1
+ (v
n+1
)
T
) : zdx
_

p
n+1
zdx =
_

f
n+1
zdx+
+
_

2
t
v
n
X
n
zdx
_

1
2t
v
n1
X
n1
zdx
_

v
n+1
qdx =
_

g qdx
v
n+1
= v
e
on
e
v
n+1
= 0 on
p
42
with
g =
f
m

1
2t
(3 ln 4 ln X
n
+ 3 ln X
n1
),
p = p +
2
3

e
g.
Discretizing and eliminating the bubbles for all steps, this leads to a linear system
with the following structure:
_
_
_
_
_
_
A 0 0 B
(1)
T
0 A 0 B
(2)
T
0 0 A B
(3)
T
B
(1)
B
(2)
B
(3)
C
_
_
_
_
_
_
_
_
_
_
_
_
v
1
v
2
v
3
p
_
_
_
_
_
_
=
_
_
_
_
_
_
S
(1)
1
S
(2)
1
S
(3)
1
S
2
_
_
_
_
_
_
3.2 Radiation
For the equation of the intensity of radiation we use a six-ux method for the semidis-
cretization in and choose between a nite element method and a nite dierence method
(the latter is only valid for a structured mesh), for spatial discretization.
The equations to be solved are
_

_

x
I + (a +
s
)I

s
4
_
S
2
(

, )I(x,

)d

= aI
b
in ,
I(x) =
1
w

w
q
r
n +
T
4
w

on .
3.2.1 Six-flux method
Let

I(x, ) W :=< w
1
, . . . , w
6
> and
I
+
1
(x) =

I(x,
1
) I

1
(x) =

I(x,
2
)
I
+
2
(x) =

I(x,
3
) I

2
(x) =

I(x,
4
)
I
+
3
(x) =

I(x,
5
) I

3
(x) =

I(x,
6
)
Then

I(x, ) = I
+
1
(x)w
1
() + I

1
(x)w
2
() + + I
+
3
(x)w
5
() + I

3
(x)w
6
()
because of
w
j
(
i
) =
ij
, 1 i, j 6.
By introducing F
i
:= I
+
i
+ I

i
and q
i
:= I
+
i
I

i
, for 1 i 3, the six-ux method
for the semidiscretization in leads to the boundary-value problem,
_

x
i
_

F
i
x
i
_
+ KF
i
2
s
3

j=1
j=i
F
j
=
2aT
4

in , 1 i 3
F
i
+
n
i
|n
i
|

F
i
x
i
=
1
w

w
_

i=1
F
i
x
i
n
i
_
+ 2
T
4
w

on , 1 i 3
43
3.2.2 Finite elements
The algorithm proposed is the following:
Step 0. Input F
0
i
, 0 i 3
Step n+1. Given F
n
i
, F
n+1
i
are solution of the problems:
_

F
n+1
i
x
i
z
i
x
i
dx +
_

KF
n+1
i
z
i
dx +
_

1 + C
2
C
3
C
1
F
n+1
i
z
i
|n
i
|d
_

C
2
C
1
(
3

j=1
j=i
F
n
j
|n
j
|)z
i
|n
i
|d

1
C
1
2T
4
w

z
i
|n
i
|d
_

2
s
s(
3

j=1
j=i
F
n
j
)z
i
dx =
_

2a
g
T
4

z
i
dx +
_

2a
s
T
4
s

z
i
dx
where
C
1
= 1 +
1
w

w
_
3

j=1
|n
j
|
_
C
2
=
1
w

w
C
3
=
3

j=1
j=i
|n
j
|
3.2.3 Finite Differences
We suppose that = [0, l
x
] [0, l
y
] [0, l
z
]. Then the boundary condition above reads

w
F
i
+
F
i
x
i
n
i
=
2

w
T
4
w
on .
The proposed algorithm is the following:
Step 0. Input F
0
i
, 0 i 3
Step n+1. Given F
n
i
, we obtain F
n+1
1
as the solution to the problems dened on lines
parallel to the x axis
_

x
_

F
n+1
1
x
_
+ KF
n+1
1
= b
1n
in
x
= [0, l
x
] {y
j
} {z
l
},

F
n+1
1
x
n
1
+
w
F
n+1
1
=
2

w
(T
n
w
)
4
on
x
= {(0, y
j
, z
l
)} {(l
x
, y
j
, z
l
)},
with
b
1n
= 2
s
s
3

j=1
j=1
F
n
j
+
2a
g
(T
n
)
4

+
2a
s
(T
s
)
4

.
This leads to the simple linear system
A{F
1
} = S
44
where
A =
_
_
_
_
_
_
_
d
1
p
d
1
s
d
1
s
.
.
.
.
.
.
.
.
.
.
.
. d
nx1
s
d
nx1
s
d
nx
p
_
_
_
_
_
_
_
S =
_
_
_
_
_
_
_
_
_
_
_
_
1
2
h
1
b
1
1
+
2

w
(T
w
(x
1
, y
j
, z
l
))
4
.
.
.
1
2
(h
i1
+ h
i
)b
1
i
.
.
.
1
2
h
nx1
b
1
nx
+
2

w
(T
w
(x
nx
, y
j
, z
l
))
4
_
_
_
_
_
_
_
_
_
_
_
_
d
p
=
_
_
_
_
_
_
_
_
_
_
_
_
_
1
2

1
+
2
h
1
+
w
+
1
2
h
1
K
1
.
.
.
1
2

i1
+
i
h
i1
+
1
2

i
+
i+1
h
i
+
1
2
(h
i1
+ h
i
)K
i
.
.
.
1
2

nx1
+
nx
h
nx1
+
w
+
1
2
h
nx1
K
nx
_
_
_
_
_
_
_
_
_
_
_
_
_
d
s
=
_
_
_
_
_
_

1
2

1
+
2
h
1
.
.
.

1
2

nx1
+
nx
h
nx1
_
_
_
_
_
_
F
1
= (F
11
, . . . , F
1nx
)
T
and F
1i
= F
n+1
1
(x
i
, y
j
, x
l
), i = 1, . . . , n
x
.
In a similar way, we proceed with lines parallel to the axis y and z, giving values of
F
n+1
2
and F
n+1
3
, respectively.
3.3 Advection-diusion equations
The advection-diusion-reaction equations of the gas-phase model are of the general
form,

Du(x, t)
Dt
((x, t)D(x)u(x, t)) + c(x, t)u(x, t) = f(x, t).
More specically, the energy and mass conservation equations are of this type. In order
to solve them we use a second order semi-Lagrange-Galerkin scheme. We discretize the
convective term using the formula
Du
Dt
=
u
t
+v u
1
2t
_
3u
n+1
4u
n

n
+ u
n1

n1
_
,
where

n
(x) = (x, t
n+1
; t
n
)
is the position at time t = t
n
of the uid particle that is at point x at time t
n+1
and that
moves with velocity v. The characteristic curves (x, s; t) are the unique solution of the
ODE
d(x, s; t)
dt
= v((x, s; t), t), s, t [0, T]
(x, s; s) = x.
45
To solve this ODE we can use a fourth-order Runge-Kutta method or a third-order
xed point method with the operator T : dened by
T = t
_
1.5v
n
0.5v
n1
_
that satises Ty = y with
y = x (x, t
n+1
; t
n
).
The latter makes possible to adapt the time step.
3.4 Algorithm for the solid phase model
In order to solve the combustion model introduced in the previous sections, we have
designed an algorithm which distinguishes, at each time step, the combustion stage in
which the particle is. In Figure 3.4 we can see a scheme of this algorithm that we apply
to each single particle.
Figure 8. Algorithm for solving the combustion model
46
4 Numerical Results
4.1 Analysis of numerical diusion
Results of the study of the numerical diusion when solving a convection-diusion
equation with the nite element method introduced in the previous section can be seen
in Figure 4.1. They could be compared, for instance, with those of [5].
Figure 9. Comparison of results for a benchmark 3D test
4.2 A coal ame simulation
A detailed study of the structure of a turbulent pulverized coal ame has been done
in Hwang et al [3, 4].
Our purpose will be to do a full numerical simulation of the turbulent pulverized coal
burner, used in these articles (see Figure 10).
Figure 10. Scheme of the experiment
47
In this study, methane is supplied to ignite the two-phase jet because the coal injection
rate is very small and ame stabilization is impossible for a pure pulverized coal ame.
The methane ow rate is the minimum amount needed to form a stable ame. The
experiment is designed as follows: rst the air is supplied to the main burner port and the
methane to the annular slit burner. Next the gas ame due to the air and the methane
is formed, and nally, when the methane diusion ame becomes stable, the pulverized
coal particles are dropped.
In order to make a numerical simulation of this experiment we proceed to the following
steps:
1. We make a simulation with Fluent of the rst stage of the experiment, until the
methane ame is stable. Figure 11 shows the geometrical domain for the CFD
simulation. The data corresponding to ows are: through the coal-air inlet 1.810
4
kg/s of air and 1.49 10
4
kg/s of coal are introduced whereas 1.67 10
5
kg/s of
gas enters through the gas inlet.
Figure 11. Geometry of the burner
2. With the results of the gas phase obtained from the Fluent simulation we solve our
combustion model.
3. We solve the combustion model with Fluent (namely Discrete Phase Model) starting
at the same initial data.
4. We compare results provided by both methods.
Figures 12 and 13 show the homogenized sources of total mass and O
2
, respectively,
obtained with both models.
48
Figure 12. Source of mass (f
m
)
Figure 13. Source of O
2
(f
m
O
2
)
Furthermore, with respect to unburned coal we have obtained the following results
from the two dierent models:
Fluent:
Volume integral of mass source: 6.34664 10
5
kg/s
Mass of released volatiles: 94.54 %
Mass of gasied char: 50.50 %
BFL:
Volume integral of mass source: 7.55143 10
5
kg/s
Mass of released volatiles: 98.31 %
Mass of gasied char: 44.53 %
49
5 Conclusions
We have proposed a mathematical model for a pulverized coal furnace and subse-
quently numerical methods for solving it. For validation purposes, we have simulated an
experiment involving the combustion of pulverized coal particles in a cylindrical domain.
Results obtained have been compared with those given by Fluent.
Acknowledgment
Part of this work was supported by Ministerio de Educaci on y Ciencia through project
ENE2005-09190-C04-01: Investigacion Aplicada en Conversi on Limpia de Combustibles
Fosiles (CLCF)
References
[1] K. Annamalai and W. Ryan. Interactive processes in gasication and combustion. II Isolated
carbon, coal and porous char particles. Progr. Energy Combust. Sci., 19:383446 (1993).
[2] A. Berm udez and J.L. Ferrn and A. Li n an. The modelling of the generation of volatiles, H
2
and CO, and their simultaneous diusion controlled oxidation, in pulverised coal furnaces.
Combustion Theory and Modelling, 11:949976 (2007).
[3] S-M. Hwang, R. Kurose, F. Akamatsu, H. Tsuji, H. Makino and M. Katsuki. Observation of
detailed structure of turbulent pulverized-coal ame by optical measurement (Part 1, time-
averaged measurement of behavior of pulverised-coal particles and ame structure). JSME
International Journal, 49(4):13161327, 2006.
[4] S-M. Hwang, R. Kurose, F. Akamatsu, H. Tsuji, H. Makino and M. Katsuki. Observation
of detailed structure of turbulent pulverized-coal ame by optical measurement (Part 2, in-
stantaneous two-dimensional measurement of combustion reaction zone and pulverised coal
particles). JSME International Journal, 49(4):13281335, 2006.
[5] M.R. Kaazempur-M. and C.R. Ethier. An ecient characteristic Galerkin scheme for the
advection equation in 3-D. Comput. Methods Appl. Mech. Engrg., 191:5345-5363, 2002.
[6] J.M. Macinnes and F. V. Bracco. Stochastic particle dispersion modeling and the tracer-
particle limit Phys. Fluids, 12:28092824 (1992).
[7] F.A. Williams. Combustion Theory. The Benjamin/Cummings Publ. Co., Inc, Menlo Park,
1985.
[8] A.D. Gosman and E. Ioannides. Aspects of computer simulation of liquid-fueled combustors.
J. Energy, 7(6):482490, 1983.
[9] Q. Zhou and M.A. Leschziner. A time-correlated stochastic model for particle dispersion in
anisotropic turbulence. 8th Turbulent Shear Flows Symposium., Munich, 1991.
50