Você está na página 1de 18

A CFD based combustion model of an entrained ow biomass

gasier
D.F. Fletcher
a,
*
, B.S. Haynes
a
, F.C. Christo
b,1
, S.D. Joseph
b
a
Department of Chemical Engineering, University of Sydney, Sydney, NSW 2006, Australia
b
Biomass Energy Services and Technology Pty. Ltd., 1 Davids Close, Somersby, NSW 2250, Australia
Received 24 August 1998; received in revised form 11 May 1999; accepted 7 June 1999
Abstract
This paper contains the description of a detailed Computational Fluid Dynamics (CFD) model developed to sim-
ulate the ow and reaction in an entrained ow biomass gasier. The model is based on the CFX package and rep-
resents a powerful tool which can be used in gasier design and analysis. Biomass particulate is modelled via a
Lagrangian approach as it enters the gasier, releases its volatiles and nally undergoes gasication. Transport
equations are solved for the concentration of CH
4
, H
2
, CO, CO
2
, H
2
O and O
2
and heterogeneous reactions between
xed carbon and O
2
, CO
2
and H
2
O are modelled. The model provides detailed information on the gas composition and
temperature at the outlet and allows dierent operating scenarios to be examined in an ecient manner. 2000
Elsevier Science Inc. All rights reserved.
Keywords: CFD modelling; Biomass energy; Gasication; Swirl ow; Combustion; Gasier
1. Introduction
The drive to reduce net greenhouse emissions has produced considerable interest in the com-
bustion of biomass. Many processes produce biomass waste that can be used for energy pro-
duction. This paper is concerned with the design and optimization of one possible system, namely
an entrained ow biomass gasier. Our aim is to use Computational Fluid Dynamics (CFD) to
prove the design of a pilot plant which is being developed, and then to use the validated model to
design a 1 MW plant.
In an entrained ow gasier the fuel (sawdust or cotton gin trash in this case) is blown into a
reaction chamber with air. The quantity of air is limited, so that there is sucient reaction to
maintain the temperature of the system at a high enough level to cause pyrolysis of the fuel, but so
that not all of the volatiles are consumed. The gas phase reactions take place very quickly, and
leave behind a mixture of CH
4
, H
2
, CO, CO
2
, H
2
O and N
2
. Once the volatiles have been released
from the particles a mixture of char (unburnt carbon) and ash remains. The char then undergoes
reaction with CO
2
and H
2
O to produce CO and H
2
.
www.elsevier.nl/locate/apm
Applied Mathematical Modelling 24 (2000) 165182
*
Corresponding author. Tel.: +61-2-9351-4147; fax: +61-2-9351-2854.
E-mail address: davidf@chem.eng.usyd.edu.au (D.F. Fletcher).
1
Current address: DSTO Aeronautical and Maritime Research Laboratories, Weapons Systems Division, Salisbury, SA 5106,
Australia.
0307-904X/00/$ - see front matter 2000 Elsevier Science Inc. All rights reserved.
PII: S 0 3 0 7 - 9 0 4 X ( 9 9 ) 0 0 0 2 5 - 6
The gas exiting the gasier is cleaned and then used to power an engine to generate electricity.
Details of the gasier design and additional background can be found in [1]. The full-scale design
examined in this paper is shown in Fig. 1. The gasier diameter is 0.8 m and the height is 7.5 m.
Fuel is introduced via the lower pair of tangential inlets and superheated steam is introduced via
the upper pair.
2. The present model
This is a very demanding system to model, involving intensely swirling ows, complex
chemistry for both homogeneous and heterogeneous reactions, and complex geometry. There are
clearly many levels of modelling which could be tried. Simple mass and energy balances can be
used to determine what would happen under equilibrium conditions but these take no account of
nite rate processes, such as gasication, or of spatial variation. At a more sophisticated level
separate thermodynamic models can be used for the dierent regions in the gasier, e.g., for the
pyrolysis zone or the gasication chamber. These provide more information than the global
model but are still very limited. Such an approach has been performed by Zhen [2] for a corncob
gasier. Here we have decided to use a CFD model which allows us to determine local infor-
mation and to include nite rate chemistry eects. These complexities provide a very signicant
computational challenge and we have therefore developed the model in a step by step manner. All
modelling reported in this paper was performed using CFX4, a multi-purpose CFD coded
marketed by AEA Technology, Harwell, UK [3].
Fig. 1. The geometry of the gasier. The lower inlets are used to inject the biomass mixed with air, and the upper inlets
are used to inject steam.
166 D.F. Fletcher et al. / Appl. Math. Modelling 24 (2000) 165182
Initially we made an extensive study of the oweld and the behaviour of particles within this
oweld. The need to use a Dierential Stress Model (DSM) for turbulence modelling and high
order dierencing on a complex body-tted grid was established for a cold system [6]. In this
paper we address combustion modelling via the addition of a simple model for devolatilization,
global reactions to model the gas phase reaction and heterogeneous reactions which cause the
xed carbon (char) to be gasied.
2.1. Conservation equations
The governing equations are given here in coordinate-free tensor notation. Favre averaging is
used throughout. An upper case symbol is used to denote a mean quantity and a lower case
symbol is used to denote a uctuating quantity where appropriate. The ow is considered to be
weakly compressible, so that the density is a function of the composition and the temperature.
Thus the density is given by
1
q
=
RT
P
0
X
N
i=1
m
i
W
i
; (1)
where ideal gas behaviour is assumed. There are N components, with component i having a
molecular weight W
i
and mass fraction m
i
. P
0
is the reference thermodynamic pressure. Conser-
vation of mass and momentum are governed by
\ (qU) = S
p
(2)
and
\ (qU U) = \ r S
pu
; (3)
where the stress tensor is given by
r = Pd
2
3
l\ Ud l(\U (\U)
T
) q(u u): (4)
The terms S
p
and S
pu
are the source terms due to the presence of the particles.
For the ke model [4] the Boussinesq hypothesis is used to relate the turbulent stress tensor to
the shear via
q(u u) =
2
3
qkd
2
3
l
T
\ Ud l
T
(\U (\U)
T
); (5)
where the turbulent viscosity is given by
l
T
= C
l
q
k
2
e
(6)
and an eective viscosity l
eff
is dened via
l
eff
= l l
T
: (7)
The usual equations for k and e are used, where
\ (qUk) \ l


l
T
r
k

\k

= P qe; (8)
\ (qUe) \ l


l
T
r
e

\e

= C
e1
e
k
P C
e2
q
e
2
k
; (9)
D.F. Fletcher et al. / Appl. Math. Modelling 24 (2000) 165182 167
and the production term is given by
P = l
eff
\U \U

(\U)
T

2
3
\ U(l
eff
\ U qk): (10)
The standard constants are used:
C
l
= 0:09; C
e1
= 1:44; C
e2
= 1:92; r
k
= 1:0; r
e
= 1:3: (11)
For the Reynolds stress calculations the model based on that of Launder et al. [5] was used. In
this model, the following equation was used to determine the components of the Reynolds stress
tensor
\ (qu u U) \ q
C
S
r
DS
k
e
u u(\u u)
T

= P U
2
3
qeI; (12)
where the shear production term is given by
P = q u u(\U)
T

(\U)u u

: (13)
The pressurestrain correlation U is split into two components, which model return to isotropy
and return to isotropy of production. They are given by
U
1
= qC
1S
e
k
u u


2
3
kI

(14)
and
U
2
= C
2S
P


2
3
PI

; (15)
where
P =
1
2
trace(P) = qu u \U: (16)
The turbulence dissipation is calculated via
\ (qUe) \ q
C
S
r
e
k
e
u u\e

= C
e1
e
k
P C
e2
q
e
2
k
: (17)
The additional constants are given by
C
S
= 0:22; C
1S
= 1:8; C
2S
= 0:6; r
e
= 1:375; r
DS
= 1:0: (18)
Note that in the present simulations no account is taken of the eect of the particles on the
turbulence eld.
Conservation of energy is given by
\ (qUH quh k\T) = S
ph
(19)
and mass conservation for a species mass fraction Y is given by
\ (qUY quy qD\Y ) = S
pY
R
f
; (20)
where k is the thermal conductivity, D the laminar diusion coecient, S
ph
and S
pY
are particle
source terms, and R
f
is the source term due to chemical reaction. The temperature was related to
168 D.F. Fletcher et al. / Appl. Math. Modelling 24 (2000) 165182
the enthalpy, H, using a cubic polynomial in order to allow for the variation of specic heat with
temperature correctly.
For the ke and Reynolds stress modelling, the Reynolds uxes are given by
qu/ =
l
T
r
T
\U and C
S
k

u u\U
T
; (21)
respectively. The turbulent Prandtl number (r
T
) is set to 0.9.
2.2. Particle transport model
It is assumed that the biomass, which consists of a mixture of seeds, cotton and pieces of stalk,
can be represented as spherical particles with an appropriate average diameter. There is consid-
erable inhomogeneity in the biomass and this is represented by using a distribution of particle
sizes. Constitutive physics laws are required to model the transfer of momentum, heat and mass
between the particles and the gas phase. The laws used for each of these transfer processes are
described below.
2.2.1. Momentum transfer
The path of a particle is tracked by solving the following equation
m
dU
p
dt
= F
D
; (22)
where m is the particle mass, U
p
the particle velocity and F
D
is the drag force, given by
F
D
=
1
8
pd
2
q
g
C
D
[U
R
[U
R
; (23)
where d is the particle diameter, q
g
the gas density, C
D
the drag coecient and U
R
is the relative
velocity between the gas and the particle. The drag coecient is calculated from [7]
C
D
= 24(1 0:15Re
0:687
)=Re; (24)
where Re is the slip Reynolds number given by
Re =
q
g
[U
R
[d
l
g
(25)
and l
g
is the viscosity of the gas.
The eect of turbulence on the particles is modelled using an eddy interaction model [8].
Particles are assumed to interact with a succession of randomly directed eddies. The character-
istics of the eddy are determined from the calculated turbulence quantities. It is assumed that the
variance of the speed of the ow in the eddies is 2k, that an individual eddy exists for a time t
E
given by
t
E
= 1:5
1=2
C
3=4
l
k

(26)
and that the eddy lengthscale is
L
E
= C
3=4
l
k
3=2

: (27)
D.F. Fletcher et al. / Appl. Math. Modelling 24 (2000) 165182 169
2.2.2. Heat transfer
Heat transfer to the particles is modelled as the sum of contributions from convective, latent
heat and heat of gasication components. Convective heat transfer is modelled using the well-
known RanzMarshal correlation [9], which gives
Nu = 2 0:6Re
1=2
Pr
1=3
; (28)
where Nu is the Nusselt number and Pr the Prandtl number of the gas. The convective heat
transfer component is then given by
_
Q
c
= pdk
g
Nu(T
g
T
p
); (29)
where k
g
is the thermal conductivity of the gas.
The component due to phase change for each species i is given by
_
Q
Mi
=
_
M
i
L


Z
T
T
ref
(c
pgi
c
ppi
) dT

; (30)
where L is the latent heat at reference temperature T
ref
. In the simulations the heat of pyrolysis
was assumed to be zero for all species. Therefore the only contribution from this term was from
cases in which water remained in the trash.
Finally, for the char species the heat of gasication was added to the particle, i.e.,
_
Q
G
=
_
M
C;CO
2
H
CO
2

_
M
C;H
2
O
H
H
2
O

_
M
C;O
2
H
O
2
; (31)
where the heats of gasication for the CO
2
, H
2
O and O
2
reactions are given by 14:37 MJ/kg of C
reacted, 10:94 MJ/kg of C reacted and 32.8 MJ/kg of C reacted, respectively.
The total heat transfer from the particles was assumed to be the sum of the above components.
2.2.3. Mass transfer
Release of the volatile components was assumed to occur via ash pyrolysis (see later) so that
_
M
vol
= H(T T
rel
)M
vol
=s; (32)
where H(T) is the Heaviside function, s a time constant (set to 0.1 s) and T
rel
is a critical tem-
perature for pyrolysis to commence (set to 500 K). In this model the volatile release rate is
proportional to the mass of volatiles remaining in the particle and is switched on or o depending
whether the temperature is above or below a critical value.
For the char the reaction rate was assumed to be the minimum of the kinetic and diusion
rates. For the kinetic rate, the following rst order rate was used
_
M
Ci
= k
/
i
X
i
exp
E
Ai
RT

M
C
; (33)
where M
C
is the mass of carbon and X
i
is the mole fraction of the reacting gas.
For diusion
_
M
Ci
= 2pdDY
i
q
g
W
C
W
i
(34)
where d is the particle diameter, D the diusion constant, Y
i
the mass fraction of species i in the
gas, q
g
the gas phase density and W
i
is the molecular weight of species i. The value of the diusion
coecient was evaluated at the lm temperature.
170 D.F. Fletcher et al. / Appl. Math. Modelling 24 (2000) 165182
3. Chemical reaction modelling
Rates must be specied for the volatile release, as well as the homogeneous and heterogeneous
reactions.
3.1. Devolatilization
Firstly, a model of the devolatilization process is required. This process is very uncertain, with
the composition of the volatiles released being very dependent on the heating rate and the tem-
perature as discussed by Bingyan et al. [10]. For various woods, they found that at mid tem-
peratures (400C < T < 700C) the pyrolysis process produced char, tar, CO
2
, CO, H
2
O, H
2
, CH
4
and C
n
H
m
, and that for temperatures above 700C and long residence times (68 s) all of the tar
was cracked to H
2
, CH
4
and C
n
H
m
.
The gasier operates in the high temperature, long residence time regime. Therefore it was
decided to model pyrolysis via the use of a prescribed time constant of release and to prescribe the
mass fractions of CH
4
, H
2
, CO
2
, CO and H
2
O released. The devolatilization process was assumed
to be energetically neutral because the heat of devolatilization is generally small and of uncertain
sign [11].
The composition of the volatiles is not arbitrary, as it depends on the composition of the
original biomass. Here we used experimental data for the elemental composition (C,H,O) of the
biomass and the xed carbon content to provide a constraint on the product composition. Also
we assumed that the heat of combustion of the volatiles released would be the same as that for the
gas composition coming from the biomass, where we obtained the heat of combustion of
the biomass from a general formula for the heat of combustion as a function of the composition.
The higher heating value, as determined by the Institute of Gas and Technology [12] was used,
which gives
HHV (kJ=kg) = 2:32(146:58C 568:78H51:53(ON) 6:58Ash 29:45); (35)
where C, H, O, N and Ash are expressed in weight percentage on a dry basis. With these as-
sumptions a volatile composition was always found for any given temperature. (In all cases of
practical interest we found a unique composition but this is not guaranteed by the above con-
ditions.) The temperature at which the volatiles were assumed to form had an inuence on the
nal composition but was generally weak over the temperature range 300600C. Table 1 gives
the results of such a computation for two types of biomass, namely sawdust and cotton gin trash.
Except for the rst row, which gives the moisture content in the original biomass, all other
Table 1
Composition of the biomass for various fuels. All compositions are given in terms of mass fraction
Fuel Sawdust Cotton trash
Moisture 9.1% 10.4%
Composition (dry basis) C
0:514
H
0:072
O
0:414
C
0:502
H
0:057
O
0:441
CH
4
0.22 0.17
H
2
0.02 0.00
CO 0.24 0.02
CO
2
0.38 0.34
H
2
O 0.00 0.05
O
2
0.00 0.09
Char 0.14 0.22
Ash 0.001 0.11
D.F. Fletcher et al. / Appl. Math. Modelling 24 (2000) 165182 171
quantities are presented on a dry basis. There are some signicant dierences between the two
fuels. In the case of sawdust the volatiles contain signicant quantities of H
2
and CO, which are
not present in the cotton trash. Also there is almost twice the mass of char and a very signicant
mass of ash in the cotton trash case. The latter is due to the large amount of soil in the roots of the
plants.
3.2. Gaseous phase reactions
We have used simple global reactions to describe the gas phase chemistry. The reactions used
are:
CH
4
2H
2
O
k
1
CO
2
4H
2
206 MJ=kmol; (36)
CH
4
2O
2

k
2
CO
2
2H
2
O35:7 MJ=kmol; (37)
H
2

1
2
O
2

k
3
H
2
O242 MJ=kmol; (38)
CO
1
2
O
2

k
4
f
CO
2
283 MJ=kmol; (39)
COH
2
O
k
5f
CO
2
H
2
41:1 MJ=kmol: (40)
We used the data from Jones and Lindstedt [13] for reactions (36)(38) and (40), and that of
Westbrook and Dryer [14] for reaction (39). Initially, the use of the rates given by Westbrook and
Dryer for the oxidation of CO caused considerable numerical problems. After checking, it was
found that the rates are not consistent with the equilibrium constant available in standard texts.
Examination of the reaction rates showed that the reverse rate was unphysically high and it was
set to zero. Very recently, this and other inconsistencies in the Westbrook and Dryer scheme have
been reported in the literature by Polifke et al. [15]. Reaction (5), the watergas shift reaction, is
worthy of special mention. There are a variety of reaction rates in the published literature.
Gururajan et al. [16] discuss the fact that this reaction is catalysed by, for example, iron in the ash.
In this study we have used the forward reaction rate of Jones and Lindstedt and calculated the
reverse rate using the equilibrium constant given by Gururajan et al. [16].
The above rates are based solely on the chemical reaction rate and take no account of the
turbulence mixing time. In order to include both eects we adopted the approach of Bakke and
Hjertager [17], in which the turbulent mixing rate is calculated via an eddy breakup model and we
set the reaction rate for the fuel R
fu
via
R
fu
= min[R
chem
; C
ebu
q

k
min(m
fu
; m
ox
=s)[; (41)
where m
fu
and m
ox
are the mass fractions of the fuel and oxidant and s is the stoichiometric ratio.
As is usual in this approach, all rates were evaluated using the local values of the mean tem-
perature and mean concentrations. The constant C
ebu
was set to 4. This procedure was applied to
all reactions, including the water gas shift reaction where it was applied separately to the forward
and backward rates.
172 D.F. Fletcher et al. / Appl. Math. Modelling 24 (2000) 165182
3.3. Heterogeneous phase reactions
The following char reactions were assumed:
C H
2
O
k
6
COH
2
; (42)
C CO
2

k
7
2CO; (43)
C O
2

k
8
CO
2
: (44)
The reaction rates are known to vary signicantly depending on the composition of the char.
Therefore we used values determined experimentally from samples of cotton trash char [18]. These
data were tted using the rst order reaction model, given in (Eq. 33), to yield the coecients
given in Table 2.
4. Solution procedure
The above equations were solved using CFX4 [3] which is a general purpose CFD model. It
solves the above partial dierential equations using the nite volume method on a body tted
grid. The SIMPLEC algorithm was used together with the RhieChow algorithm to provide a
stable solution on a non-staggered mesh. All convection terms were modelled using van Leer
dierencing for accuracy [19]. A typical mesh contained ~90,000 cells. Mesh sensitivity studies
were performed as part of the model development work and this mesh density proved to be a
reasonable compromise between the competing needs for accuracy and manageable run times.
4.1. Initial conditions
In order to start the calculations the lower chamber was lled with air at 1000 K, so that
devolatilization would occur once the particles entered the chamber. The upper chamber con-
tained 30% steam at 400 K.
4.2. Boundary conditions
Calculations are performed here for the full-scale gasier. Therefore, at each of the lower inlets
a mass ow rate of 0.13 kg/s (960 kg/h) of air at 120C and 0.2 kg/s (720 kg/h) of biomass was
injected. The average air velocity was 30 m/s. Relationships for fully developed turbulence were
applied at the inlets to determine k, and the Reynolds stress components.
In order to represent the spread of particle sizes in the biomass a Rossin Rammler distribution,
given by
Prob(diameter < d) = 1 exp

d
d
mean

n

(45)
Table 2
Kinetics data for the char reactions
Reaction k
/
(s
1
) E
A
(kJ/kmol)
k
6
1:4 10
8
179.5
k
7
3:4 10
7
179.5
k
8
9:35 10
4
82.8
D.F. Fletcher et al. / Appl. Math. Modelling 24 (2000) 165182 173
was used, where typically we set d
mean
= 0:2 mm and n = 2. The biomass was assumed to be dry
when it entered the gasier, with the water initially present in the biomass being injected as steam
with the heated air.
At each of the upper inlets 0.02 kg/s (140 kg/h) of steam at 400C was injected. The inlet
velocity was 122 m/s.
4.3. Convergence of the simulations
Obtaining a solution to this problem is dicult because of the very dierent timescales of the
hydrodynamics and the combustion. With no combustion, a solution to the hydrodynamics
problem can be obtained straightforwardly in a few hundred iterations. However, with com-
bustion present there is a strong turbulencechemistry interaction, via the density change and
turbulent mixing. In order to obtain a solution we had to use false time-stepping for the enthalpy
and mass fractions.
The calculations are relatively robust provided that the false timestep used to under-relax the
mass fraction equations is small enough (typically 5 ms) and the particle source term under-re-
laxation factors are small enough (typically 0.05). Both of these requirements stem from the fact
that the reaction rates are high and must be under-relaxed to prevent unphysical results devel-
oping in a cell. With this arrangement, approximately 80 outer iterations around the particle
transport model are needed with 50 uid iterations per particle iteration. The calculations require
typically two days of cpu time on a 333 MHz DEC-Alpha when started from scratch.
In test cases in a simple tube without swirl it was possible to reduce the mass error to extremely
low values but in the gasier simulations it always remained at approximately 5% (this is the error
for each cell summed over the complete model). Examination of the results showed that the
oweld tends to become asymmetric, especially when the Reynolds stress model was used, which
is almost certainly a prelude to the onset of unsteady ow. This is consistent with the experimental
observations, where a time varying composition was observed.
5. Example results
Simulations were carried out by rst converging a k calculation and then using this as an
initial guess for a Reynolds stress simulation. The calculated results, in terms of temperature and
gas composition, from these dierent models were very similar. This came as a surprise, given the
very signicant oweld dierences observed in the cold ow simulations. Figs. 2 and 3 show
the swirl and axial velocity distributions across the centre of the vessel at a height of 3 m from the
base. This location is about 0.5 m downstream of the obstacle. It is clear from the gures that
there are relatively small dierences between the results of the two dierent models. The swirl
velocity shows that the k model predicts a forced vortex and the DSM model predicts a free/
forced vortex but that the swirl intensity is similar. There are dierences in the axial velocity
distribution near the centre of the gasier but there is close similarity in the near wall region.
These gures also illustrate the point made earlier about the non-symmetry of the oweld. The
dierence in the solutions of the two models was similar to that illustrated throughout the vessel
and the large dierences in the ow regimes in the gasier observed in the cold case [6] were not
observed here. In the current situation there is a large density, and hence owrate, change due to
combustion and gasication. This appears to dominate the oweld and make it much less
sensitive to the turbulence model. As the k simulations are much faster to run, this model was
used in most simulations.
174 D.F. Fletcher et al. / Appl. Math. Modelling 24 (2000) 165182
Example results are presented here for a case of a fuel of cotton gin trash. In all subsequent
gures the radius of the gasier has been scaled up by a factor of two to improve the clarity. Fig. 4
shows the vertical velocity distribution. There is a region of down ow in the lower chamber
beneath the blu body and just downstream of it, as indicated by the hashed zone in the plot.
There is also a recirculation zone near the exit, which serves to trap particles within the gasier.
The swirl component of the velocity is shown in Fig. 5. This shows that there is very strong swirl
within the device, which persists to the exit. The temperature distribution is shown in Fig. 6. The
temperature is high in the combustion zone around the tangential inlets and then falls with height
up the gasier, especially in the upper chamber where gasication is taking place. The outlet
temperature is approximately 1050 K, which is consistent with the values seen during operation.
There is a hot zone in the base of the gasier where combustion is occurring.
Fig. 3. A comparison of the axial velocity distribution across the centre of the gasier at a height of 3 m from the base
for the k and DSM models.
Fig. 2. A comparison of the swirl velocity distribution across the centre of the gasier at a height of 3 m from the base
for the k and DSM models.
D.F. Fletcher et al. / Appl. Math. Modelling 24 (2000) 165182 175
Fig. 7 shows the methane mass fraction. It is evident that the mass fraction is highest in
the pyrolysis zone close to the inlets and then remains relatively constant as it is convected out of
the gasier. The hydrogen distribution, shown in Fig. 8, follows the opposite trend since most of
Fig. 4. The axial velocity distribution. The regions of down ow are shown hatched.
Fig. 5. The swirl velocity distribution, showing strong swirl throughout the device.
176 D.F. Fletcher et al. / Appl. Math. Modelling 24 (2000) 165182
Fig. 6. The temperature distribution, showing a hot zone at the base of the gasier and gradual cooling due to gas-
ication in the upper region.
Fig. 7. The methane mass fraction distribution, showing production close to the wall in the zone where devolatilization
takes place.
D.F. Fletcher et al. / Appl. Math. Modelling 24 (2000) 165182 177
the H
2
is produced in the upper chamber due to steam gasication and the water gas shift re-
action. The CO mass fraction distribution is shown in Fig. 9. In this case there is some CO
produced in the lower chamber but most is produced due to gasication in the top part of the
Fig. 9. The mass fraction of CO, showing signicant production near the exit where particles are trapped.
Fig. 8. The hydrogen mass fraction distribution, showing production throughout most of the gasier.
178 D.F. Fletcher et al. / Appl. Math. Modelling 24 (2000) 165182
gasier and production near the top exit is evident. Inspection of the particle tracks showed that
particles were being trapped here and undergoing gasication in this location. The CO
2
distri-
bution, shown in Fig. 10, shows the opposite behaviour, as expected. In the region where the
particles are being gasied the CO
2
concentration is reduced signicantly. The steam mass
fraction, shown in Fig. 11, displays a complex pattern. Steam is produced in the combustion
region at the gasier base and then reacts with CH
4
and CO. In the upper chamber there is a large
input of steam which is responsible for gasication of the char and production of H
2
via the water
gas shift reaction. The O
2
mass fraction, shown in Fig. 12, is zero almost everywhere, as it reacts
as soon as it enters the chamber.
The model provides details of the gas composition at exit. In this simulation the exit gas is
composed of 11% CO, 15% CO
2
, 23% H
2
, 0.06% CH
4
, 9% H
2
O and 40% N
2
, by volume. These
gures are consistent with the compositions obtained in the small-scale gasier (see below), al-
though the current calculations predict more H
2
production.
In the simulation 83% of the cotton gin trash entering the device was gasied. This is consistent
with the fact that the cotton gin trash contains 11% ash, leaving 6% of the xed carbon unburnt.
In the simulations particles were tracked for 240 s and in this time most had accumulated in the
region just below the exit. In the real device these are getting expelled from the gasier, and it is
quite likely that this is due to transient eects not modelled here.
The simulation results were found to be insensitive to the mean input particle diameter over the
range tested, 0.31.0 mm. This is because the volatiles are released very quickly in the base and the
gasication process is relatively slow, and therefore the results are not very sensitive to the particle
size.
Fig. 10. The mass fraction of CO
2
, showing signicant amounts of CO
2
present in the gasier and CO
2
consumption in
the gasication zone.
D.F. Fletcher et al. / Appl. Math. Modelling 24 (2000) 165182 179
Fig. 11. The mass fraction of steam, showing production in the lower chamber and signicant addition at the upper
inlets.
Fig. 12. The mass fraction of oxygen, showing almost immediate consumption in the lower chamber.
180 D.F. Fletcher et al. / Appl. Math. Modelling 24 (2000) 165182
A simulation was also run with sawdust as a fuel. In this case the gas composition at the exit
was 13% CO, 14% CO
2
, 24% H
2
, 5% CH
4
, 11% H
2
O and 33% N
2
. This shows that there is clearly
a dierence between the two fuels, with much more CH
4
present in this case.
With the above results looking very promising a calculation was also run for the small-scale
gasier. This proved to be very hard to converge because although the geometry was similar, the
inlets were much smaller giving very high injection velocities, e.g. 200 m/s at the steam inlet.
However, convergence to a similar level was obtained using smaller relaxation factors. This gave
an exit composition on a dry basis of 10% CO, 12% CO
2
, 20% H
2
and 1.2% CH
4
, compared with
16% CO, 14% CO
2
, 10% H
2
, 1% CH
4
measured in the experiments. Again the hydrogen gener-
ation is too high and there was signicant excess steam.
6. Conclusions
A detailed CFD model of an entrained ow biomass gasier has been developed, based on the
CFX4 package. Models of nite rate chemistry in the gas phase and char reactions have been
added to the standard model. This produces a large number of coupled equations, which together
with the need to perform Lagrangian particle tracking, makes this a huge computation. However,
with current workstations solving such systems is now a practical proposition.
The current simulations show that in the reacting case the dierences between using a full
Reynolds stress model and the k model are minimal, whereas in the non-combusting case they
were pronounced. This appears to be because of the larger volume owrates associated with
combusting ows and the fact that the gasication reactions are not sensitive to the detailed
velocity eld around particles but more on the residence time. Therefore the computationally
cheaper model is quite acceptable for use in the determination of the eect of parameter varia-
tions.
It is clear from the results presented that the model is potentially a very powerful tool and, with
further validation against detailed experimental data, will aid signicantly with the design process
of such gasiers. The initial calculations suggests that simulations to examine the eect of gasier
height and the steam ux in the upper inlets could be benecial in process optimization.
Acknowledgements
This work was partially supported by an Australian Research Council Grant in collaboration
with Biomass Energy Services & Technology Pty. Ltd. We wish to thank Dr. Phil Stopford and
Dr. Nigel Wilkes from CFX International for their considerable assistance. We thank Melita
Jazbec for assistance in the determination of the volatiles composition of the biomass.
References
[1] S. Joseph, T. Denniss, R. Lipscombe, S. Errey, D.F. Fletcher, B.S. Haynes, The development and testing of an air/
steam blown entrained ow gasier fueled with cotton waste and sawdust, Proceedings of Bioenergy 96 The
Seventh National Bioenergy Conference, 1520 September 1996, Nashville, Tennessee, USA, vol. I, 1996, pp. 304
311.
[2] F. Zhen, A biomass pyrolysis gasier applicable to rural China, Fuel Sci. Technol. Int. 11 (1993) 10251035.
[3] CFX 4.1 Flow solver: User guide, CFX International, AEA Technology, Harwell Laboratory, Didcot,
Oxfordshire, UK, 1996.
[4] B.E. Launder, B.I. Sharma, Application of the energy dissipation model of turbulence to the calculation of ow
near a spinning disc, Lett. Heat and Mass Transfer 1 (1974) 131138.
D.F. Fletcher et al. / Appl. Math. Modelling 24 (2000) 165182 181
[5] B.E. Launder, G.J. Reece, W. Rodi, Progress in the development of a Reynolds-stress turbulence closure, J. Fluid
Mech. 68 (1975) 537566.
[6] D.F. Fletcher, B.S. Haynes, J. Chen, S.D. Joseph, Computational uid dynamics modelling of an entrained ow
biomass gasier, Appl. Math. Modelling 22 (1998) 747757.
[7] R. Clift, J.R. Grace, M.E. Weber, Bubbles, Drops and Particles, Academic Press, New York, 1978.
[8] J.S. Shuen, L.D. Chen, G.M. Faeth, Evaluation of a stochastic model of particle dispersion in a turbulent round
jet, AIChE J. 29 (1983) 167.
[9] W.E. Ranz, W.R. Marshall, Evaporation from drops, Part II. Chem. Eng. Prog. 48 (1952) 173.
[10] X. Bingyan, W. Chuangzhi, L. Zhengfen, Z. Xiguang, Kinetic study of biomass gasication, Solar Energy 49
(1992) 199204.
[11] C. Di Blasi, Modelling and simulation of combustion processes of charring and non-charring solid fuels, Prog.
Energy Combust. Sci. 19 (1993) 71104.
[12] American Solar Energy Society (ASES), Advances in Solar Energy, ASES, vol. 2, 1985.
[13] W.P. Jones, R.P. Lindstedt, Global reaction schemes for hydrocarbon combustion, Combust. Flame 73 (1988)
233249.
[14] C.K. Westbrook, F.L. Dryer, Simplied reaction mechanisms for the oxidation of hydrocarbon fuels in ames,
Combust. Sci. Technol. 27 (1981) 3142.
[15] W. Polifke, K. D obbeling, T. Sattelmayer, D.G. Nicol, P.C. Malte, A NO
x
prediction scheme for lean-premixed
gas turbine combustion based on detailed chemical kinetics, J. Eng. Gas Turbines and Power 118 (1996) 765772.
[16] P.K. Gururajan, P.K. Agarawal, J.B. Agnew, Mathematical modelling of uidized bed coal gasiers, Trans.
IChemE 70 (1992) 211238.
[17] J.R. Bakke, B.H. Hjertager, The eect of explosion venting on empty vessels, Int. J. Num. Meth. Eng. 24 (1987)
129140.
[18] T.G. Newbury, B.S. Haynes, Report on the characterisation of oxygen, carbon dioxide and steam gasication for
a cotton trash char, Department of Chemical Engineering, University of Sydney, 1997.
[19] N.A. Shore, B.S. Haynes, D.F. Fletcher, A.A. Sola, Numerical aspects of swirl ow computation, in R.L. May,
A.K. Easton (Eds.), Computational Techniques and Applications Conference: CTAC95, 37 July 1995,
Melbourne, Australia, 1996, pp. 693700.
182 D.F. Fletcher et al. / Appl. Math. Modelling 24 (2000) 165182

Você também pode gostar