An EulereEuler approach to modeling biomass fast pyrolysis

in uidized-bed reactors e Focusing on the gas phase

Pelle Mellin
*
, Qinglin Zhang, Efthymios Kantarelis, Weihong Yang
KTH Royal Institute of Technology, Division of Energy and Furnace Technology, Brinellvgen 23, SE 100 44 Stockholm, Sweden
h i g h l i g h t s
A 3-D CFD model for fast pyrolysis in uidized beds has been developed.
Focus is on the vapor phase which permits a simplied rendering of the solids.
Predicted results largely agree with measured temperature, pressure, and bed height.
a r t i c l e i n f o
Article history:
Received 16 January 2013
Accepted 25 April 2013
Available online 7 May 2013
Keywords:
CFD
Fluidized bed
Fast pyrolysis
Pyrolysis oil
Pyrolysis
Bio oil
a b s t r a c t
A developed 3D EulereEuler CFD model, with an integrated pyrolysis model, is proposed as a way of
predicting vapor phase dynamics and product distributions in the uidized bed process for biomass fast
pyrolysis. The main interest in this work is the gases resulting from the pyrolysis mixed with the
uidizing gas. We propose therefore a simple rendering of the solid material while directing attention to
the vapor phase. At the same time the required computational resources for reaching stabilized condi-
tions in the reactor are reduced. Temperature prole, velocity prole and pyrolysis products are pre-
dicted and globally veried by a series of parallel cases, which are compared to experimental
measurements and known trends of liquid, solid and gas yields. The comparison of experimental
measurements and model predictions satisfy the accuracy of the model and on a quantitative basis, the
product yields agree with commonly known trends of bio oil versus temperature and residence time.
2013 Elsevier Ltd. All rights reserved.
1. Introduction
Kinetic models of biomass pyrolysis have evolved as a part of
understanding devolatilization during combustion processes.
However, as research progressed, pyrolysis models have been
specialized for use in gasication as well as pure pyrolysis pro-
cesses. The primary objective for gasication is to nd out the
reactivity of the resulting coal after devolatilization. In the pyrolysis
process, the models have successfully explained and predicted
product distributions as function of temperature, as well as resi-
dence time and other pyrolysis conditions.
Computational Fluid Dynamics (CFD) modeling applied to
chemical processes is an increasingly important tool in plant design
and process development, as well as an increasingly useful tool in
research work. A CFD model coupled with a kinetic model, applied
to fast pyrolysis are increasingly used to investigate such parame-
ters as heat distribution, product distributions, temperature of gas
and solid phase as well as vapor and particle residence time. Recent
work of implementing fast pyrolysis into CFD codes includes Xue
and co-workers [1,2] (uidized bed reactor), Boateng and Mtui [3]
(uidized bed), Papadikis and co-workers [4e10] (uidized bed
reactor), Papadikis and co-workers [5,11] (entrained ow reactor).
In the work by Papadikis and co-workers for uidized bed re-
actors; two Eulerian phases and one Lagrangian is the preferred
approach by which the actual path lines of the reacting biomass
particles could be calculated in a 3-D domain. Boateng and Mtui [3]
used a similar approach in a 2-D domain. In the work by Xue and
co-workers [1,2], three Eulerian phases are computed in a 2-D
domain and for one case in a 3-D domain.
The objective of this work is to develop a model of the uidized
bed based on the Eulerian two-phase framework. This approach
could be useful to time-effectively study the interaction between
the sand and biomass mix with the vapor phase at the same time as
determining the product distributions (most importantly the liquid
yield) and performance of the system.
Initially the experimental rig is described which is used for
supporting the model and providing measured values for the
different physical quantities. The numerical basis of the model is
* Corresponding author. Tel.: 46 87909022; fax: 46 8207681.
E-mail address: pmellin@kth.se (P. Mellin).
Contents lists available at SciVerse ScienceDirect
Applied Thermal Engineering
j ournal homepage: www. el sevi er. com/ l ocat e/ apt hermeng
1359-4311/$ e see front matter 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.applthermaleng.2013.04.054
Applied Thermal Engineering 58 (2013) 344e353
then presented and a base case is used to compare with measured
results. The predicted results are further examined and discussed
with reference to accuracy and predictability. Finally the inuence
of temperature and uidizing gas velocity is explored through
computing a series of cases and summarizing the results.
2. Pilot reactor
A pilot fast pyrolysis setup with has been assembled at KTH,
Sweden. Fig. 1 shows a photo of the setup. The setup includes a
preheater for the uidizing-gas, the uidized bed reactor, a cyclone
and a scrubber. The biomass is introduced by a screwfeeder directly
into the uidized bed at a height of 5.75 cm above the distributor
plate. The biomass (a mix of pine and spruce) are usually fed at rate
of 2 kg/h, which is the designed capacity of the rig.
The total height of the cylindrical stainless-steel vessel, contain-
ing the uidized bed and the freeboard, is 95 cm. Fig. 2 shows a cross
section of the uidized bed reactor. The uidizing gas carries the
pyrolysis vapors and reacted biomass (char) through a small circular
outlet (1 cm in diameter) at the top. The pyrolysis vapors and char
are then channeled through an insulated pipe and later separated by
the cyclone. When running uidized bed reactor witha uidizing gas
velocity of 0.25 m/s, a bed temperature of 500

C and a feeding rate
of 2 kg/h, the predicted gas residence time to the outlet is about 2 s.
Pre-heated uidizing gas (N
2
) is introduced from the bottom
through a distributor plate. Heating elements inside the reactor
wall conveys additional heat to the uidized bed and the freeboard.
The sand in the uidized bed provides rapid heat transfer on the
scale of 1000

C/s to the biomass particles. The intermediate tem-
perature of 500

C is maintained which prevents extensive tar
cracking. The liquid yield of the operating setup is in the range of
60e70% by mass.
Temperature is measured at ve points along the reactor height,
at 5 cm (inside the bed), 25 cm, 45 cm, 65 cm and 85 cm above the
distributor plate. The thermocouples are positioned as to measure
the local temperature along the centerline. Pressure is measured
under the distributor plate and above the bed inside the steel
vessel; the difference provides information about the behavior of
the bed. The permanent gases are analyzed by a GCeMS with six
columns, able to detect up to C6 hydrocarbons.
3. Multiuid model
The approach in this work assumes that the main impact on the
uid phase owcomes fromthe sand particles in the system and to
a limited degree from the biomass particles. Together with a fast
evaporation and a fast kinetically controlled pyrolysis, the
assumption permits a simplied treatment of the solids using only
one Eulerian phase for the biomassesand mixture. This approach
has some limitations, such as the inability to show the actual
pathways of biomass particles and thus the char entrainment.
However the approach still introduce a lowfeed temperature to the
feeding location as well as evaporating and devolatilizing biomass
particles which gives rise to a lowtemperature zone and signicant
gas expansion in the lower part of the reactor.
The EulereEuler multiphase framework with two phases is
applied in the computation; one uid phase and one granular phase
which is a mixture of sand and biomass. The multi-uid CFD
calculation is based on continuity equations that are solved for each
phase, cell, iteration and for each time step. The commercial soft-
ware ANSYS Fluent 14.5 is used.
3.1. Mass conservation
Continuity equations account for mass, momentum, energy and
species; for the uid phase the following mass equation (1) is
solved, where a denotes volume fraction, r the density, v the ve-
locity, _ m the mass exchange between phases and S the sources
terms dened by the user.
d
dt
_
a
f
r
f
_
V$
_
a
f
r
f
v
!
f
_
_ m
f
S
f
(1)
Fig. 1. Photo of the bio oil pilot plant, in view is the insulated uidized bed reactor, the
uidizing gas preheater, the hopper and the screw feeder.
Biomass
Distributor
plate
Windbox
Fluidized bed
Freeboard
T-201
T-202
T-203
T-204
T-205
T-206
Thermocouples
H
e
a
t
i
n
g
e
l
e
m
e
n
t
s
Fig. 2. Cross section of the bubbling uidized bed reactor, which indicates the position
of the thermocouples and the heating elements.
P. Mellin et al. / Applied Thermal Engineering 58 (2013) 344e353 345
3.2. Species conservation, properties and phase exchange
For all the species in a phase, such as the compounds in the uid,
the following analogous equation (2) is solved, where Y denotes the
mass fraction of any specie i.
d
dt
_
a
f
r
f
Y
i
_
V$
_
a
f
r
f
Y
f
v
!
f
_
_ m
f
S
f
(2)
When running the experimental setup, particle size of the
irregular shaped biomass is about 700e1000 mm which is well
within the suggested range of 100e6000 mm for producing liquid
products; according to Ref. [12]. The ideal conditions for producing
liquids implies fast and uniform heating rate which means suf-
ciently hot conditions and a large surface to volume ratio of the
biomass. Both conditions are met by the current setup.
As the uidizing gases readily ush away the vapors from the
particle surface, diffusion controlled decomposition is not likely. The
pyrolysis reactions applied in this model are instead considered rst-
order kinetically controlled while being a function of particle tem-
perature. The particle temperature is determined by the balance of
the temperature surroundings (as the particle is considered ther-
mally thin) and the reactions. Evaporation of moisture is considered
as heat transfer controlled which is reasonable considering the fast
heating of biomass particles initially injected at room temperature.
Volatiles in the context of this model are activated and
decomposed by primary pyrolysis to either tars or gas and char. The
dened pyrolysis reactions correspond to the scheme developed by
Miller and Bellan [13] for three sub-components of biomass. The
rate of the reactions associated to primary pyrolysis are averaged
based on the weight fractions Y
i
of cellulose, hemicellulose and
lignin reported by Demirbas [14], for Pine (i.e. 43%, 29.4% and 27.6%
respectively). The expression for rate of decomposition of activated
volatiles is written as (3).
dC
Volatiles
dt

i C;H;L
a
s
r
s
Y
i
Y
Volatiles
M
Volatiles
A
i
exp
_
E
i
RT
s
_
(3)
The kinetic parameter A
i
and E
i
, denoted A
C
, A
H
, A
L
and E
C
, E
H
, E
L
,
are taken from Ref. [18]. Table 1 shows parameters for the
mentioned reactions associated to the primary pyrolysis. Tar is
considered as the same product from cellulose, hemicellulose, and
lignin which is subjected to secondary pyrolysis according to ki-
netics collected from Di Blasi [15], i.e. A 4.28 10
6
1/s and
E 108.0 10
3
J/mol.
The more general expression for determining the reaction rate
for pyrolyzing specie i, in phase p which is used for the secondary
pyrolysis is given by (4).
dC
i
dt

a
s
r
s
Y
i
M
i
Aexp
_
E
RT
s
_
(4)
Heat of reaction for reactions 2 and 3 (see Table 1) is taken from
Refs. [16], while the activation (reaction 1) is considered athermic.
Heat of evaporation and heat of secondary pyrolysis are calculated
by balance. The biomass feeding is dened as a mass and mo-
mentum source from six cells, at the corresponding position of the
screw feeder inlet.
The biomass consists of 92.4% volatiles, 0.6% ash and 7% mois-
ture. Table 2 shows the material properties of the solid species in
the simulation. Note that for Case 1 (see Table 3), the char density is
higher since a low density would cause unrealistic bed expansion
during the long run.
7 Nl/min of nitrogen and 0.56 g/s of biomass are continuously
added to the respective phases. A uidizing gas inow rate of
0.25 m/s is set, which is equivalent to 3.125 u
mf
. The biomass is a
mix of pine and spruce which is milled to ne particle size of about
700e1000 mm. The particle size of the sand is about 300 mm which
is the value used for the whole granular phase. 500 g of SiO
2
-sand
as bed material is placed at the bottom of the reactor at the
beginning of the computations.
3.3. Momentum conservation and phase exchange
The momentum equation for the uid is expressed in (5), where
p is pressure, s
f
equals the stress tensor, g is the gravitational force,
K
sf
is the phase momentum exchange coefcient.
d
dt
_
a
f
r
f
v
!
f
_
V$
_
a
f
r
f
Y
f
v
!
f
v
!
f
_
a
f
Vp Vs
f
a
f
r
f
g
!
K
fs
_
v
!
f
v
!
s
_
_ m
fs
v
!
f
(5)
Table 1
Kinetics and heat of reaction for the three reactions associated to primary pyrolysis.
Reaction A
C
, 1/s A
H
, 1/s A
L
, 1/s E
C
, j/mol E
H
, j/mol E
L
, j/mol Dh
R
, j/g
1. Vol. /active vol. 2.8 10
19
2.1 10
16
9.6 10
8
242.4 186.7 107.6 0
2. Active vol. /char gas 1.3 10
10
2.6 10
11
7.7 10
6
150.5 145.7 111.4 3827
3. Active vol. /tar 3.3 10
14
8.8 10
15
1.5 10
9
196.5 202.4 143.8 1277
Table 2
Materials properties used in the computation.
Materials properties Sand (SiO
2
) Char (C) Volatiles Moisture (H
2
O)
Apparent density, kg/m
3
2650 200
a
600 1000
Maximum packing factor 0.59 0.59 0.59 0.59
Bulk density, kg/m
3
1563.5 118 354 590
Heat capacity, J/kg/K 830 1583 1538 4200
a
Set to 1000 in Case 1, since bed expansion otherwise becomes unrealistic with
incorrect pressure prole as a result.
Table 3
Cases computed in the work with corresponding boundary conditions.
Case, no Biomass feeding
temperature,

C
Inlet temperature,

C Initial temperature,

C Wall temperature,

C Fluidizing gas
velocity, m/s
Computed real
time, s
1
a
25 397.7 477 510 0.25 200
2
b
e 25 25 25 0.25 25
3e8 25 450 450 510 0.15e0.40 25
9e15 25 350e550 350e550 350e550 0.25 25
a
Base case used for comparison with experimental results at hot reacting conditions.
b
Cold case without reactions which is used for comparison of bed height in experiment.
P. Mellin et al. / Applied Thermal Engineering 58 (2013) 344e353 346
For the granular phase, the analogous equations are solved
which is given in (6), where Vp
s
the solids pressure is added.
d
dt
a
s
r
s
v
!
s
V$a
s
r
s
Y
s
v
!
s
v
!
s
a
s
Vp Vp
s
Vs
s
a
s
r
s
g
!
K
fs
_
v
!
s
v
!
f
_
_ m
sf
v
!
s
(6)
The stress tensor which appears in equations (5) and (6) is
related to the internal viscous stress. For the uid phase, s
f
is
dened as (7), for the solid phase s
s
it is dened as (8).
s
f
m
f
_
_
V v
!
f
V v
!
f
_

2
3
V$ v
!
f
I
_
(7)
s
s
m
s
_
V v
!
f
V v
!
f
_

_
l
s

2
3
m
s
_
V$ v
!
f
I (8)
The symbol m denotes viscosity in both stress tensor equations;
m
f
being the molecular viscosity in the case of uid phase and m
s
the
solid shear viscosity in the case of solid phase, given by (9). l
s
denotes the bulk viscosity and is calculated as (10) from Ref. [17].
m
s
m
s;col
m
s;kin
m
s;fr
(9)
l
s

4
3
a
s
r
s
d
s
g
0;ss
1 e
ss

_
q
s
p
_
1=2
(10)
The solid shear viscosity m
s
is generally assumed to be a sum of
three components, the collisional viscosity m
s,col
(11) fromRefs. [18],
the kinetic viscosity m
s,kin
(12) from Refs. [19], the frictional vis-
cosity m
s,fr
(assumed negligible due solids volume fraction being far
from the maximum packing limit). q
s
is the granular temperature
which is estimated based on kinetic theory, from Ref. [19].
m
s;col

4
5
ard
s
g
0;ss
1 e
ss

_
q
s
p
_
1=2
a
s
(11)
m
s;kin

adr

q
s
p
_
63 e
ss

_
1
2
5
1 e
ss
3e
ss
1a
s
g
0;ss
_
(12)
The pressure based solver is used with an implicit rst order
time formulation. The standard ke per-phase turbulence model
are applied to the gas phase, but neglected for the granular phase.
Turbulence is considered since the gas accelerates close to the top
of the reactor and reaches high velocities (up to 30 m/s as seen in
the results in Section 4.2). Such high velocities correspond to a
Reynoldss number upwards of 8000, which means an emerging
fully turbulent state.
The momentum phase interaction coefcient K
fs
is estimated by
(13), following the work by Syamlal and OBrien [19]. C
D
is the drag
coefcient, Re
s
is the particle Reynolds number and v
r,s
is the par-
ticle terminal velocity (derived from (14) with coefcients dened
in (15)).
K
fs

3a
s
alr
f
4v
2
r;s
d
s
C
D
_
Re
s
v
r;s
_

v
!
s
v
!
f

(13)
Fig. 3. The gure shows pressure drop as function of uidizing gas velocity which
indicates the minimum uidization velocity, u
mf
. The difference in pressure is
measured below the distributor plate and above the bed.
0
0.2
0.4
0.6
0.8
200 300 400 500 600
R
e
a
c
t
o
r

h
e
i
g
h
t

(
m
)
Temperature (C)
Model Measured
Fig. 4. The thermal prole along the centerline of the reactor; predicted results are
shown as a black solid line. The measured temperature is indicated by black dots with
error bars showing the uctuating interval.
Table 4
Comparison of total pressure, for measured and predicted results (Case 1).
Position Measured, Pa Model, Pa
Under distributor plate (windbox) 102,700 56 e
Pressure above distributor plate 102,427 56
a
102,923
Above the bed 101,348 51 101,687
Pressure drop over bed 1079 107 1236
a
Calculated.
0.0
0.2
0.4
0.6
0.8
0.0 0.1 0.2 0.3 0.4
R
e
a
c
t
o
r

h
e
i
g
h
t
/
Y
-
d
i
r
e
c
t
i
o
n

(
m
)
Volume fraction of solids
Predicted
Expected
Fig. 5. The expected bed height compared to predicted results from Case 2. The pre-
dicted results are shown as the volume fraction of solids along the reactor centerline
and the bed height is indicated by the sharp decline in solids volume fraction at a
height of about 14 cm.
P. Mellin et al. / Applied Thermal Engineering 58 (2013) 344e353 347
v
r;s
0:5
_
A 0:06Re
s

0:06Re
s

2
0:12Re
s
2B A A
2
_ _
(14)
A a
4:14
f
B
_
a
C1
f
; a > 0:85
C
2
a
1:28
f
; a 0:85
(15)
We have tuned the values C
1
and C
2
according to the procedure
described in Ref. [20] in order to predict the right minimum uid-
ization velocity u
mf
. By monitoring the pressure drop as a function
of intrinsic uidizing gas velocity, u
mf
is estimated to 0.08 m/s. See
Fig. 3 for the pressure drop as function of intrinsic uidizing gas
velocity. By also assuming a volume fraction of 0.5, the values C
1
and
C
2
are approximated to 9.19 and 0.28 respectively; as opposed to the
default values of 2.65 and 0.8 (which predicts a higher u
mf
).
3.4. Energy conservation and phase exchange
The energy equation for the uid phase is solved as (16), where
h
f
is the internal heat transfer coefcient and h
fs
is the heat transfer
coefcient at the phase interface.
d
dt
_
a
f
r
f
h
f
_
V$
_
a
f
r
f
Y
f
v
!
f
h
f
_
a
f
vp
vt
s
f
: V v
!
f
V q
!
f
S Q _ m
fs
h
fs
(16)
The energy exchange between the uid and the solid phase is
determined by temperature difference multiplied by an exchange
coefcient h
sf
, given by (17). k
f
denotes the thermal conductivity
of the uid, d
s
particle diameter of the solid and Nu
s
the solids
Nusselt number which is approximated by (18) following the work
by Gunn [21].
h
sf

6k
f
a
s
a
f
Nu
s
d
2
s
(17)
Nu
s

_
7 10a
f
5a
2
f
__
1 0:7Re
0:2
s
Pr
1=3
_

_
1:33 2:4a
f
1:2a
2
f
_
Re
0:7
s
Pr
1=3
(18)
3.5. Boundary conditions, mesh and time step size
The wall boundary condition is set to no-slip with a xed tem-
perature, as shown in Table 3. For Cases 3e8 the uidizing gas
velocity is changed in increments of 0.05 m/s and for Cases 9e15
the inlet-, initial- and wall temperature is changed in increments of
50

C. The heating elements only reach up to 65 cmof the bed; thus

a decreased wall temperature of 200

C is set 65 cm above the
distributor plate in the model. Plug ow is assumed at the
distributor plate with a specied velocity-inlet condition.
The time step size for all cases started from 2 10
4
s. As gra-
dients stabilize a gradual increase of the step size is possible. For all
cases, step size could be increased to 5 10
3
s and in the long
calculation Case 1, the time step size could be increased to a nal
value of 1 10
2
s. The residuals for all time steps, is allowed to
decline below 1 10
3
before progressing to the next step.
The mesh used in the model consists of 69,294 hexagonal cells.
The grid size in Y-direction is uniformly 9.5 mm. In the X and
Fig. 6. Shows a 3-D image of modeling results for Case 2, after 25 s. A cross section
along the Z 0 plane indicates the volume fraction along the height and depth, while
an iso-surface shows the surface of the uidized bed. The iso-surface of the volume
fraction has a value of 0.1.
101325
101725
102125
102525
102925
103325
0
0.8
1.6
2.4
3.2
4
0 0.15 0.3 0.45
P
r
e
s
s
u
r
e

[
P
a
]
R
e
s
i
d
e
n
c
e

t
i
m
e
s

[
s
]
Velocity [m/s]
Residence time Pressure
u
mf
Fig. 7. Shows the decreasing gas residence time for higher uidizing gas velocities, for
Cases 3e8. Also the pressure above the bed increases slightly as the uidizing gas
velocity increases which is shown on the right axis.
P. Mellin et al. / Applied Thermal Engineering 58 (2013) 344e353 348
Z-direction, grid size varies as the mesh is rened toward the
centerline. However, the largest cells are 4.11 mm in both X and Z-
direction; with a maximum volume of 1.6$10
7
m.
The sand considered in the model may be categorized as Geldart
B type particles. According to simulations made by Parmentier et al.
[22] for Geldart B particles; parameters such a bed height is
insensitive to quadratic grid sizes varying from 4.6 cm to 9.86 mm.
In developing the model, the same observation was made. A grid
size with maximum side length of 9.5 mm was deemed sufcient,
as the gas phase is primarily the focus of this work. The resulting
cell number which can be considered low permits a short
computation time.
4. Results and discussion
This section shows results provided by the model for the Cases
shown in Table 3. In Section 4.1, Case 1 is compared to measured
values of Pressure and Temperature and Case 2 is compared to
expected bed height. In the next Section 4.2, the inuence of
uidizing gas velocity is studied in Cases 3e8, and in the section
after, 4.3, the inuence of temperature is covered by Cases 9e15.
For Case 1 the actual time for computing the nal results took
8260 min, or 137.7 h, using 4 cores on a HP Compaq 81100 Elite CMT
PC. The other cases are computed in parallel with aid of the Na-
tional Infrastructure for Computing (SNIC 001-11-26) via PDC. The
Fig. 8. Cross sections along the Z 0 plane which displays the temperature proles at different uidizing gas velocities, for Cases 3e8.
P. Mellin et al. / Applied Thermal Engineering 58 (2013) 344e353 349
required time for computing Case 1 is comparably low for a 3-D
domain. In the work by Xue et al. [1], one week of computation is
required for 100 s of physical time with a 2-D grid of only 940 cells.
The yields calculated throughout the paper are based on mass
fraction. Gas and Liquid yield are the outlet ow rate of corre-
sponding compounds divided by the biomass feeding rate. Char is
summed up over the whole domain and divided by the total
computation time and the feeding rate per second. This results in
the sum of the three product yields slightly deviating from 1.
4.1. Comparison of predicted and experimental results of the base
case
For comparison with model results, we used an experimental
base case (Case 1) with an inlet ow rate of 0.25 m/s, an inlet
temperature of 397.7

C, and a maintained wall temperature of
510

C. The gas residence in the predicted base case is 2.1 s. The
temperature prole compared to model results is shown in Fig. 4
and the pressure is compared to modeling results in Table 4. For
the thermocouple located in the bed: measured temperature is
397.7

C 5 (1.3%), 25 cm above the distributor plate:
458.5

C 3.8, 45 cm above: 486.2

C 2.7, 65 cm above:
478.3

C 2.9 and 85 cm above: 436.7

C 3.85. Variations of the
temperature are small and the response time of the thermocouples
sufcient for capturing the uctuations.
Note that thepredictedtemperaturedeviates inthemiddle-part of
the reactor; while exact reasons may be hard to distinguish, keep in
mind that the thermocouples inserted in the gas ow not only mea-
sures the uids physical temperature. Radiation from the walls and
contact with particles also contributes to heating the thermocouples.
The expected bed height, which is compared with results from
Case 2 in Fig. 5, is about 14 cmat cold conditions (with a uidization
velocity of 0.25 m/s). The expected bed height is determined by
using the Ergun equation [23] to predict the pressure drop as
function of gas velocity below u
mf
. Since the values agree with the
observed pressure drop in the xed bed region in Fig. 3, the Ergun
equation may be used to calculate bed expansion at u
mf
. The bed
expansion is then presumed constant as the uidizing gas velocity
is increased to 0.25 m/s. As visible in Fig. 3, the predicted bed height
ts the expected closely. Fig. 6 shows a 3-D view of the reactor for
Case 2, after 25 s. The color map indicate volume fraction of solids.
A cross-section surface and an iso-surface shows the surface of the
bed and the prole along the Z 0 plane.
4.2. Inuence of uidizing gas velocity in a series of cases
The uidizing gas velocity affects both the bed behavior and the
time the particles and gases reside in the reactor. The effect on the
liquid yield is not self-evident as residence time and heat transfer is
important for the tar cracking and the fast pyrolysis respectively.
In this work, the residence time is calculated based on volume of
the reactor divided by the volumetric ow rate at the outlet. In
Fig. 7, the gas residence time is shown as a function of inlet ow
rate, alongside the slight pressure increase due to the increased
ow rate. The increased pressure arises from gas trying to escape
through the small orice at the top of the reactor. A slightly pres-
surized process induces ow of vapors throughout the system;
however leakage must carefully be prevented since noxious fumes
might escape the reactor.
The temperature will differ slightly in each case as the increased
ow rate of gas will supply more heat to the system. See Fig. 8 for
the temperature proles in Cases 3e8 with increasing uidizing gas
velocity. The temperature proles reveal a rather homogenous
temperature which increases slightly as the ow rate of uidizing
gas adds more heat to the system.
Fig. 9 shows the tar mass concentration and upwards-velocity
along the reactor height, for three cases. The peak in tar mass
concentration, in each individual case, represents the pyrolysis
which releases the tar. The peak is delayed at higher ow rates
which is reasonable since the bed expands and the biomass is
carried upwards. Rapid declines above the peaks represent the tar
cracking which occurs throughout the whole length of the reactor.
The acceleration of gases is primarily due to heating of gases
since expansion is assumed in accordance with the ideal gas law.
However, a contribution of the pyrolysis is visible in the graphs, as
the gas accelerates after the peak of tar concentration. The accel-
eration at the top of the rector (right hand side of Fig. 9) in all cases
occurs because of the small outlet. The velocity can reach as high as
30 m/s (in Case 8) but such large numbers are not displayed in
Fig. 9.
Fig. 10 displays the pyrolysis product yields as function of inlet
gas velocity. The tendency is a high char yield at high owrates. The
decreased residence time at higher ow rates seems to have a less
pronounced effect than anticipated. Instead the char yield increases
somewhat as the biomass is lifted at a high angel to the top of the
bed. Here, less heat is supplied by the sand which causes char
formation and some heat develops from this reaction which is
visible in Fig. 10. Since the actual path lines of the biomass particles
-1.25
-1.00
-0.75
-0.50
-0.25
0.00
0.25
0.50
0.75
1.00
1.25
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
1.00
0 0.2 0.4 0.6 0.8
Y
-
v
e
l
o
c
i
t
y

(
m
/
s
)
T
a
r

m
a
s
s

c
o
n
c
e
n
t
r
a
t
i
o
n

(
k
g
t
a
r
/
m
rr
3
)
Reactor Height/Y-direction (m)
Tar concentration Tar concentration Tar concentration
Y-velocity Y-velocity Y-velocity
a) 0.15 m/s b) 0.25 m/s c) c 0.35 m/s
Fig. 9. Shows tar mass concentration and gas Y-velocity along reactor height, for a set
of curves at different uidizing gas velocities; (a) Case 3, 0.15 m/s, (b) Case 5, 0.25 m/s,
(c) Case 7, 0.35 m/s.
0
0.2
0.4
0.6
0.8
0 0. 15 0. 3 0. 45
Y
i
e
l
d
,

m
a
s
s

f
r
a
c
t
i
o
n
Velocity (m/s)
Liquid Char Gas
u
mf
Fig. 10. Shows pyrolysis product distributions as function of uidizing gas velocity, the
predicted values correspond to Cases 3e8.
P. Mellin et al. / Applied Thermal Engineering 58 (2013) 344e353 350
are not considered, the results should be conrmed by a more
intricate model in order support this conclusion.
4.3. Inuence of temperature in a series of cases
The temperature is of signicant importance to the product
yields of the pyrolysis process. A number of cases with increasing
wall, initial and inlet temperature are computed to see the tem-
perature proles and product yields. Fig. 11 shows the temperature
proles with increasing temperature boundary conditions. As seen
in the proles the temperature evens out before passing half the
height of the reactor. The biomass feeding from the right side is
visible as a trailing lowtemperature zone in the gas phase, which is
due to the low feeding temperature, evaporation of moisture and
the slightly endothermic tar forming reaction. At the highest tem-
perature (550

C), a slight increase of temperature is visible above
the feeding location. This is due to biomass particles being
entrained by the gas ow beneath and the heat supply rate thus
declining. A declining heat supply rate favors the char forming
Fig. 11. Cross sections along the Z 0 plane displays gas temperature for Cases 9e15 with increasing initial, inlet and wall temperature.
-1.25
-1.00
-0.75
-0.50
-0.25
0.00
0.25
0.50
0.75
1.00
1.25
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
1.00
0 0. 2 0. 4 0. 6 0. 8
Y
-
v
e
l
o
c
i
t
y

(
m
/
s
)
T
a
r

m
a
s
s

c
o
n
c
e
n
t
r
a
t
i
o
n

(
k
g
t
a
r
/
m
rr
3
)
Reactor Height/Y-direction (m)
Tar concentration Tar concentration Tar concentration
Y-velocity Y-velocity Y-velocity
a) 400C b) 475C c) 550C
Fig. 12. Shows mass concentration of tar and Y-velocity of the gas along reactor height,
for a set of temperature cases (a) Case 9, 400

C, (b) Case 12, 475

C, (c) Case 15, 550

C.
P. Mellin et al. / Applied Thermal Engineering 58 (2013) 344e353 351
reaction (see Table 1) which is exothermic and leads to increasing
temperature. Fig. 12 shows the tar mass concentration and
upwards-velocity along the reactor height, for three cases. The tar
cracking is clearly visible as the lowest tar mass concentration for
the highest temperature. In the velocity prole, the case with the
highest temperature shows the highest acceleration due to thermal
expansion of the gases and the signicant volume of released
permanent gases due to the secondary pyrolysis.
Fig. 13 shows the corresponding total yields of pyrolysis prod-
ucts, as predicted by the model. The drop in liquid yield at high
temperature is signicant and is offset by a sharp increase in the gas
yield. This shows the secondary pyrolysis in full effect and is
consistent with earlier ndings by Mesa-Prez et al. [24] where a
high gas yield is caused by raising the temperature from 470

C to
550

C. The predicted liquid yield in this case is as high as 73.3% at
450

C; however the actual process yield will be lower since the tar
cracking continues outside the reactor.
5. Discussion and conclusion
The results show that as long as product yields and uid phase
dynamics is main interest, this approach is adequate. In general, the
model shows good agreement with experimental measurements
and is a good foundation for further development. The computa-
tional time for calculating one case until stabilized conditions is
promising as other approaches take much longer time.
The effect of uidization gas velocity shows that decreased
residence time by increased owrate has little impact on the liquid
yield (in the range of 450

C), the effect of uidization gas velocity is

likely larger at higher temperatures.
The model shows a signicant drop in liquid yield at higher
temperature which is consistent with previous work. The path lines
of the biomass particles are inexactly predicted by the model,
therefore such phenomena as biomass entrainment which causes
char formation, must be veried by a more rened model.
Given that Geldart B particles are used and sufciently small
biomass particles are fed to the uidized bed, the model can be
applied to an industrial-scale plant. Other approaches discussed in
the introduction are in some instances needlessly complex and
when modeling a larger integrated system, other methods might be
to be too elaborate to be practically applicable.
In closing remark, the products of fast-pyrolysis encompass a
large number of species. Tar compounds, in the context of
gasication, are generally considered as molecules exceeding the
molecular weight of benzene i.e. 78 g/mol [25]. In the case of py-
rolysis processes, the tarry compounds constitute up to 40 percent
of the initial biomass. For the model to be useful beyond deter-
mining product yields as a result of structural aspects, the chem-
istry of the bio oil has to be modeled more accurately. Secondary
reactions occurring to specic components would be one inter-
esting aspect that could be explored by CFD models. The compu-
tational requirements of the model proposed in this work offers
room for an extended chemical model with more reactions and
species.
Acknowledgements
This work was supported by the Swedish National Infrastructure
for Computing (SNIC 001-11-26) via PDC. KIC Innoenergy and The
Swedish Energy Agency are acknowledged for funding of the
project. Preem, Boson Energy, Sveaskog and The Division of
Chemical Technology at KTH are acknowledged for analysis, gen-
eral- and technical support. We are grateful to all above named
organization for the opportunity to carry out this research.
References
[1] Q. Xue, T. Heindel, R. Fox, CFD model for biomass fast pyrolysis in uidized-
bed reactors, Chem. Eng. Sci. 66 (2011) 2440e2452.
[2] Q. Xue, D. Dalluge, T. Heindel, R. Fox, R. Brown, Experimental validation and
CFD modeling study of biomass fast pyrolysis in uidized-bed reactors, Fuel
97 (2012) 757e769.
[3] A. Boateng, P. Mtui, CFD modeling of space-time evolution of fast pyrolysis
products in a bench-scale uidized-bed reactor, Appl. Therm. Eng. 33e34
(2012) 190e198.
[4] K. Papadikis, S. Gu, A. Bridgwater, CFD modelling of the fast pyrolysis of
biomass in uidised bed reactors: modelling the impact of biomass shrinkage,
Chem. Eng. J. 149 (2009) 417e427.
[5] K. Papadikis, H. Gerhauser, A. Bridgwater, S. Gu, CFD modelling of the fast
pyrolysis of an in-ight cellulosic particle subjected to convective heat
transfer, Biomass and Bioenergy 33 (2009) 97e107.
[6] K. Papadikis, S. Gu, A. Bridgwater, 3D simulation of the effects of sphericity on
char entrainment in uidised beds, Fuel Process. Technol. 91 (2010) 749e758.
[7] K. Papadikis, S. Gu, A. Bridgwater, A CFD approach on the effect of particle size
on char entrainment in bubbling uidised bed reactors, Biomass and Bio-
energy 34 (2010) 21e29.
[8] K. Papadikis, S. Gu, A. Bridgwater, CFD modelling of the fast pyrolysis of
biomass in uidised bed reactors. Part B: heat, momentum and mass transport
in bubbling uidised beds, Chem. Eng. Sci. 64 (2008) 1036e1045.
[9] K. Papadikis, S. Gu, A. Bridgwater, Computational modelling of the impact of
particle size to the heat transfer coefcient between biomass particles and a
uidised bed, Fuel Process. Technol. 91 (2010) 68e79.
[10] K. Papadikis, S. Gu, A. Bridgwater, Geometrical optimization of a fast pyrolysis
bubbling uidized bed reactor using computational uid dynamics, Energy
Fuels 24 (2010) 5634e5651.
[11] K. Papadikis, S. Gu, A. Bridgwater, H. Gerhauser, Application of CFD to model
fast pyrolysis of biomass, Fuel Process. Technol. 90 (2009) 504e512.
[12] A. Bridgwater, Principles and practice of biomass fast pyrolysis processes for
liquids, J. Anal. Appl. Pyrol. 51 (1999) 3e22.
[13] R. Miller, J. Bellan, A generalized biomass pyrolysis model based on super-
imposed cellulose, hemicellulose and lignin kinetics, Combust. Sci. Technol.
126 (1996) 97e137.
[14] A. Demirbas , Thermochemical conversion of biomass to liquid products in the
aqueous medium, Energy Sources 27 (2005) 1235e1243.
[15] C. Di Blasi, Modeling chemical and physical processes of wood and biomass
pyrolysis, Prog. Energy Combust. 34 (2008) 47e90.
[16] J. Rath, M.G. Wolnger, G. Steiner, G. Krammer, F. Barontini, V. Cozzani, Heat
of wood pyrolysis, Fuel 82 (2003) 81e91.
[17] C.K. Lun, S.B. Savage, D. Jeffrey, N. Chepurnity, Kinetic theories for granular
ow: inelastic particles in Couette ow and slightly inelastic particles in a
general ow eld, J. Fluid Mech. 140 (1984) 223e256.
[18] D. Gidaspow, R. Bezburuah, J. Ding, Hydrodynamics of circulating uidized
beds, kinetic theory approach, in: Fluidization VII, Proceedings of the 7th
Engineering Foundation Conference on Fluidization, 1992, pp. 75e82.
[19] M. Syamlal, W. Rogers, T.J. OBrien, MFIX Documentation: Theory Guide, U.S.
Department of Energy, Morgantown, 1993.
[20] F. Vejahati, N. Mahinpey, N. Ellis, M.B. Nikoo, CFD simulation of gasesolid
bubbling uidized bed: a new method for adjusting drag law, Can. J. Chem.
Eng. 87 (2009) 19e31.
0
0.2
0.4
0.6
0.8
375 425 475 525 575
Y
i
e
l
d
,

f
r
a
c
t
i
o
n

b
y

m
a
s
s
Temperature (C)
Liquid Char Gas
Fig. 13. Shows pyrolysis product distributions as function of reactor temperature, the
predicted values correspond to Cases 9e15.
P. Mellin et al. / Applied Thermal Engineering 58 (2013) 344e353 352
[21] D. Gunn, Transfer of heat or mass to particles in xed and uidized beds, Int. J.
Heat Mass Transfer 21 (1978) 467e476.
[22] J.F. Parmentier, O. Simonin, O. Delsart, A numerical study of uidization
behaviour of Geldart B, A/B and A particles using an Eulerian multiuid
modeling approach, in: Circulating Fluidized Bed Technology IX, Proceedings
of the 9th International Conference on Circulating Fluidized Beds, 2008, pp.
331e336.
[23] S. Ergun, Fluidowthroughpackedcolumns, Chem. Eng. Prog. 48(1952) 89e94.
[24] J.M. Mesa-Prez, J.D. Rocha, L.A. Barbosa-Cortez, M. Penedo-Medina,
C.A. Luengo, E. Cascarosa, Fast oxidative pyrolysis of sugar cane straw in a
uidized bed reactor, Appl. Therm. Eng. 56 (1e2) (2013) 167e175.
[25] L. Chunshan, S. Kenzi, Tar property, analysis, reforming mechanism and model
for biomass gasication e an overview, Renew. Sustain. Energy Rev. 13 (2009)
594e604.
P. Mellin et al. / Applied Thermal Engineering 58 (2013) 344e353 353