You are on page 1of 8

Computers and Chemical Engineering 35 (2011) 11921199

Contents lists available at ScienceDirect


Computers and Chemical Engineering
j our nal homepage: www. el sevi er . com/ l ocat e/ compchemeng
Comparison of EulerianLagrangian and EulerianEulerian method for dilute
gassolid ow with side inlet
S.N.P. Vegendla, G.J. Heynderickx

, G.B. Marin
Universiteit Gent, Laboratorium voor Chemische Technologie, Krijgslaan 281, S5, Gent 9000, Belgium
a r t i c l e i n f o
Article history:
Received 23 December 2009
Received in revised form 1 September 2010
Accepted 5 September 2010
Available online 21 September 2010
Keywords:
GasSolid ow
EulerianLagrangian
EulerianEulerian
Stochastic differential equations (SDEs)
Finite Volume
Reynolds Average
a b s t r a c t
A EulerianLagrangian method is implemented to simulate turbulent two-phase gassolid riser ow,
using a mean-eld/probability density function (PDF) method. The mean-eld method is applied to the
gas phase while the PDF method is applied to the solid phase. Using the PDF method for the solid phase is
advantageous as there is no need for closure models for the convection termin the momentumequation
contrary to the situation where the mean-eld method is used. The present method is implemented to
investigate the inuence of a side inlet on the owpattern in a dilute gassolid riser conguration. When
applying the EulerianLagrangian method, a perfect correspondence with the experimental observations
is obtained. On the contrary, when using the EulerianEulerian method signicant differences with the
experimental data are observed.
2010 Elsevier Ltd. All rights reserved.
1. Introduction
Simulation of gassolid ow is vital in the chemical indus-
try to describe processes such as Fluid Catalytic Cracking (FCC)
(Das, Baudrez, Marin, & Heynderickx, 2003), adsorption of stack
gases (Das, De Wilde, Heynderickx, Marin, Vierendeels, et al., 2004;
Vegendla, Heynderickx, & Marin, 2009), and coal combustion (Van
de Velden, Baeyens, Dougan, & McMurdo, 2007). In circulating u-
idized beds, the gas phase is the carrier phase while the solid
particles behave as a dispersed phase. De Wilde, Van engelandt,
Heynderickx, and Marin (2005) and Van engelandt, De Wilde,
Heynderickx, and Marin (2007) provide experimental observations
in a dilute gassolid cold ow riser with side inlets and abrupt T-
outlets. To simulate the riser ow and to design riser geometries,
Computational Fluid Dynamics (CFD) is one of the major tools.
To model gassolid riser ow, two types of approaches are
frequently found in the literature. The rst approach is the
EulerianEulerian (two-uid model) method, the second approach
is the EulerianLagrangian method. In both methods, the gas
phase or carrier phase is considered to be a continuous phase.
The EulerianEulerian method is implemented in the work of
Gidaspow (1994), De Wilde, Heynderickx, Vierendeels, Dick, and
Marin (2002), De Wilde, Marin, and Heynderickx (2003), De Wilde,
Vierendeels, et al. (2005), DeWilde, Vanengelandt, et al. (2005), Das

Corresponding author. Tel.: +32 092644532; fax: +32 092644999.


E-mail address: Geraldine.Heynderickx@UGent.be (G.J. Heynderickx).
et al. (2003), Das, De Wilde, Heynderickx, Marin, Vierendeels, et al.
(2004), Das, De Wilde, Heynderickx, and Marin (2004), Benyahia,
Syamlal, and OBrien (2005), Jiradilok, Gidaspow, Damronglerd,
Kobes, and Mosto (2005) and Chalermsinsuwan, Piumsomboon,
and Gidaspow (2009). In this method, the solid phase is consid-
ered to be a continuous phase as well. The EulerianLagrangian
method is implemented in the work of Dhanunjay (2003) and
Deen, Van Sint Annaland, Van der Hoef, and Kuipers (2007). In this
method, the solid phase is simulated particle by particle. The latter
is also referred to as the Discrete Particle model (DPM) (Deen et
al., 2007; Dhanunjay, 2003). Thus, the solid phase is considered to
be a dispersed phase. To reduce the computational cost linked to
this Lagrangian method, the transported velocity probability den-
sity function (PDF) equation is introduced for the solid phase in the
work of Minier andPeirano (2001) andChibbaro andMinier (2008).
The PDF equation is solved using the notional/computational par-
ticles technique. The use of the PDF method can then be considered
a Lagrangian method where the physical particles are replaced by
notional particles to solve a set of equations. The main advantage of
the EulerianLagrangian method is the fact that it does not require
any modeling of the convection terms and the source terms in the
PDF equation.
When comparing the simulation results obtained using the
EulerianEulerian method with available experimental data,
remarkable differences are found (De Wilde, Van engelandt, et
al. 2005) for a riser with a solids side inlet. These differences are
seriously reduced when using the EulerianLagrangian method as
presented in this work.
0098-1354/$ see front matter 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compchemeng.2010.09.001
S.N.P. Vegendla et al. / Computers and Chemical Engineering 35 (2011) 11921199 1193
Nomenclature
A acceleration term (ms
2
)
b time scale ratio
C random uctuation velocity vector (ms
1
)
C
0
constant in the SDE model of value 2.1
f
v
velocity probability density function
g acceleration due to gravity (ms
2
)
g
0
radial distribution function
I unit matrix
k turbulent kinetic energy for gas phase (m
2
s
2
)
m mass of solid (or) notional particle (kg)
N number of notional particles
s shear stresses (kgm
1
s
2
)
t time (s)
T
L
Lagrangian time scale (s)
T
E
Eulerian time scale (s)
V sample of solids velocity vector (ms
1
)
v solids velocity vector (ms
1
)
V
r
relative velocity vector (ms
1
)
u gas velocity vector (ms
1
)
U gas velocity vector (ms
1
)
y position vector (m)
dW Wiener diffusion vector (s
1/2
)
Z state vector
ensemble average
Greek notations
density (kgm
3
)
viscosity of the gas phase (kgm
1
s
1
)
ratio of Lagrangian to Eulerian time scale

gs
drag co-efcient (kgm
3
s
1
)
z
p
solids particle response time (s)
turbulent dissipation energy for gas phase (m
2
s
3
)

g/s
gas (or) solid volume fraction of the riser
(m
3
gas/solid
m
3
reactor
)
standardized joint normal random vector
bulk viscosity (kgm
1
s
1
)
0 granular temperature (m
2
s
2
)
granular dissipation of energy (m
2
s
3
)
o
k
, o

constants in the gas turbulence model of value 1.0


and 1.3 respectively
C
1
, C
2
, C

constants in the gas turbulence model of value


1.44, 1.92 and 0.09 respectively
^ time rate of change vector of the gas phase velocity
seen along particle trajectories (ms
2
)
Subscripts
E Eulerian
g gas
i indices of x, y, z-direction
l grid number
L Lagrangian
max maximum
o inlet
p particle
s gas seen by solid particle (or) solid particle
sim simulation
Superscripts
c collisional
k kinetic
n particle number
t turbulent
- time average value
In multiphase ows, the PDF equation of the dispersed phase
(i.e. the solid phase in FCC units and the liquid phase in internal
combustion engines) is solved using the notional-particle based
Monte-Carlo technique, while the carrier phase conservation equa-
tions are solved using the Finite Volume (FV) technique (i.e. the gas
phase in both FCC units and internal combustion engines). Refer-
ence is made to the work of Minier and Peirano (2001), Chibbaro
and Minier (2008) and Subramaniam and Haworth (2000).
The EulerianLagrangian and the EulerianEulerian method are
both implemented to simulate a gassolid riser ow. The Reynolds
Averaged Navier-Stokes (RANS) equations for the gas phase and
the transported velocity PDF equation for the solid phase are
considered. Tosolve the RANS equations, the FVtechnique is imple-
mented. To solve the transported velocity PDF equation, using
conventional techniques like the FV and the nite difference (FD)
technique is computationally prohibitive due to the high dimen-
sionality of the problem (i.e. the large number of independent
variables). To overcome this problem, Monte-Carlo simulations
are proposed. By implementing a solution technique based on a
Monte-Carlo approach, the simulation results are free from false
numerical diffusion (Fox, 2003; Pope, 1987). Remark that the use
of the transported velocity PDF equation is not limited to the solid
phase hydrodynamic properties (velocity and volume fraction), but
can be applied for reactive species in a multiphase owproblemas
well. The PDF equations for the reactive species are discussed in
Vegendla et al. (2009).
The outline of the paper is as follows. InSection2the mathemat-
ical modelingof thegassolidowis presented. Section3covers the
notional-particle solution technique. The hybrid algorithm used to
determine the solidgas ow, is presented in Section 4. The results
are discussed in Section 5. Finally, conclusions are proposed.
2. Mathematical modeling
2.1. EulerianEulerian model
As the gas and solid phase are modeled using the
EulerianEulerian method, the mass and momentum conser-
vation equations are Reynolds averaged (Ranade, 2002). For the
solid phase, the mass and momentum conservation equations
are derived based on the Kinetic Theory of Granular Flow (KTGF)
(Gidaspow, 1994).
2.1.1. Gas phase
The time smoothedconservationequations for the turbulent gas
ow are obtained from the instantaneous equations by Reynolds
averaging of the mass and momentum equations (see Eqs. (1) and
(2) in Table 1). The instantaneous gas velocity u is expressed as
u = u +u

(1)
The Reynolds stresses appearing in the time averaged gas momen-
tumequation are modeled using the standard k model (see Eqs.
(4) and (5) in Table 1) (Launder and Spalding, 1974).
2.1.2. Solid phase
The solid phase conservation equations are derived from the
KTGF (Gidaspow, 1994). The complete set of turbulent solid phase
equations are reported in Table 2. An extensive discussion of these
equations is found in De Wilde, Vierendeels, et al. (2005) and De
Wilde, Van engelandt, et al. (2005). The constitutive equations for
the calculation of the gas and solid phase properties are summa-
rized in Table 3. An extensive discussion of these equations is found
in De Wilde et al. (2002, 2003), De Wilde, Vierendeels, et al. (2005),
De Wilde, Van engelandt, et al. (2005) and Benyahia et al. (2003,
2005).
1194 S.N.P. Vegendla et al. / Computers and Chemical Engineering 35 (2011) 11921199
Table 1
Flow model: gas phase conservation equations.
Continuity equation
(T 1.1)

t
(g g ) +

y
.(g g u) = 0
Momentum equations (x, y, z)
(T 1.2)

t
(g g u) +

y
.(g g uu) =
(P+(2]3)g k)
ij
y


y
.(g sg ) + g g g gs(u v)
(T 1.3) where sg = s
m
g
+ s
t
ij
=
_
_
g
2
3
g
__
.u
y
_
+
_
g +
t
g
_
_
_
u
y
_
+
_
u
y
_
1
__
Turbulent kinetic energy equation
(T 1.4)

t
(g g k) +

y
(g g uk) =

y

_
g
g +
t
g
o
k
k
y
_
+
_
g
t
g
_
_
u
y
_
+
_
u
y
_
1
__
:
_
u
y
_
g g
Turbulence dissipation equation
(T 1.5)

t
(g g ) +

y
(g g u) =

y

_
g
g +
t
g
o

y
_
+ C
1

k
_
g
t
g
_
_

y
u
_
+
_

y
u
_
1
_
:
_

y
u
_
_
C
2
g g

2
k
Table 2
Flow model: solid phase conservation equations.
Continuity equation for solid phase
(T2.1)
(ssp)
t
+
(sspv)
y
= 0
Momentum equations (x, y, z)
(T 2.2)
(sspv)
t
+
(sspvv)
y
=
Ps
y

(sss)
y
+ sspg + gs(u v)
(T 2.3) where ss =
_
_
s
2
3
s
__

y
.v
_
+ s
_
_
v
y
_
+
_
v
y
_
1
__
Granular temperature equation
(T 2.4)
3
2
_

t
(ssp0) + (ssp0 v)

= (p
k
+ ^c ) : v (q
k
+q
c
) +c
_
1
2
mC
2
_
+ gsu

C 3gs0 = (Psl + sss) : v +


_
sks
0
y
_
3gs0
2.2. EulerianLagrangian model
As the gas phase is modeled using the Eulerian method, the
mass and momentum conservation equations are Reynolds aver-
aged (Ranade, 2002). For the solid phase, the model equations are
derived from the transported velocity PDF equation (Chibbaro &
Minier, 2008).
2.2.1. Gas phase
The time smoothed conservation equations for the turbu-
lent gas ow are obtained from the instantaneous equations
by Reynolds averaging of the mass and momentum conser-
vation equations. The turbulence of the gas phase is mod-
eled using the standard k model as described in Section
2.1.1.
Table 3
Flow model: constitutive equations.
Dissipation of granular temperature by inelastic particle collision
(T 3.1) = c
_
1
2
mC
2
_
= 3(c
2
1)s
2
sg
0
0
_
4
dp
_
0

_1
2
.v
_
Radial distribution function
(T 3.2) g
0
= 1 +4s
_
1+2.5s+4.5904
2
s
+4.515439
3
s
_
1((s]s,max))
3

.67802
_
Solid phase pressure
(T 3.3) Ps =[1+2(1+e)sg
0
]ss0
Bulk viscosity solid phase
(T 3.4) s =
4
3
ssdpg
0
(1 + c)
_
0

Shear viscosity solid phase


(T 3.5) s =
c
s
+

k
s
(1+(8]5)((1+c)]2)sg
0
)(1+(8]5)sg
0
)
sg
0
(T 3.6)
c
s
=
4
5
ssdpg
0
(1 + c)
_
0

(T 3.7)
k
s
= 1.016
5
16
m
dp
2
_
0

Turbulent viscosity of gas phase


(T 3.8)
t
g
= Cg
k
2

Interphase momentum transfer coefcient


(T 3.9) If g - 0.80, gs = 150

2
s
g
g
(dps)
2
+1.75
sg
dps

u v

(T 3.10) If g : 0.80, gs =
3
4
C
d
sg
dps
g

u v

g
2.65
(T 3.11) Rcp =
g g dp

uv

g
(T 3.12)
C
d
= 0.44 for Rcp : 1000
C
d
=
24
Rcp
(1 +0.15Rcp
0.687
) for Rcp - 1000
S.N.P. Vegendla et al. / Computers and Chemical Engineering 35 (2011) 11921199 1195
2.2.2. Solid phase
The hydrodynamic properties of the solid phase are obtained
from the transported velocity PDF equation as discussed in Minier
and Peirano (2001).
The state vector Z for the solid phase is dened as:
Z = [V, y] (2)
The PDF j
v
(V; y, t) is dened as the probability of observing the
event that the values of the particle velocities v at a time t and
a position y are found in the differential neighborhood of given
values of the sample space V:
j
v
(V; y, t)dV = P[V - v V + dV] (3)
The one-point transported velocity PDF equation is dened as
(Chibbaro & Minier, 2008; Minier & Peirano, 2001):
(j
V
)
t
+ V
(j
V
)
y
+
(A
p
j
V
)
V
P
=

V
s
[^
s
|(y, V
p
, V
s
)j
V
] (4)
where represents the ensemble average, based onspecied num-
ber of samples, and where A
p
is an acceleration term given by:
A
p,i
=
V
s,i
V
p,i
z
p
+ g
i
(5)
In Eq. (4), the rst termon the Left Hand Side (LHS) corresponds
to the time rate of change of the velocity probability, the second
term corresponds to convection due to the mean particle velocity
in physical space and the third term represents the acceleration of
the solidparticles inthe particle velocityspace. The Right HandSide
(RHS) corresponds to the rate of change of the gas phase velocity
seen along the particle trajectories.
2.2.2.1. Model equations for solid phase. The stochastic differential
equation (SDE) to determine the position y
p,i
of a particle is dened
as:
dy
p,i
= V
p,i
dt (6)
The SDE to determine the velocity V
p,i
of a particle is dened as:
dV
p,i
=
V
s,i
V
p,i
z
p
dt (7)
The SDE to determine the gas velocity V
s,i
seen by the solid
particles is dened as:
dV
s,i
=
1

g
P
y
i
dt+(V
p,j
U
g,j
)
U
g,i

y
j
dt
1
1
L,i
(V
s,i
U
g,i
)dt
+

(C
0
b
i

k
k
+
2
3
_
b
i

k
k
1
_
dW
i
x (8)
dW
i
=
i
_
dt (9)
where dW
i
is the Wiener-diffusion term characterized by a Gaus-
sian process with a zero mean and a variance dt (Fox, 2003; Pope,
1985).
The time scales in the longitudinal and transverse directions are
given by:
1
L,i
=
1
L
_
1 +
2
(V
r,i

2
](2k]3))
(10)
where is the ratio of the Lagrangian to Eulerian time scale (T
L
/T
E
),
and V
r,i
is the relative velocity of the gas velocity seen by the solid
particles and the velocity of the solid particles.
The Lagrangian time scale is given by:
1
L
=
1
(1]2 +3]4C
0
)
k

(11)
b
i
=
1
L
1
L,i
(12)
k =
3
2

3
i=1
b
i
u
2
g,i

3
i=1
b
i
(13)
The transported velocity PDF equation (Eq. (4)) needs to be
solvedfor a large number of notional particles that covers the entire
computational domain of the cold ow riser, in order to obtain an
accurate estimate of the mean velocity eld, V
V
l
=

N
l
n=1
m
n
l
v
n
l

N
l
n=1
m
n
l
(14)
where m
n
l
is the unit mass of a notional particle and V
n
l
is the
velocity vector for the nth particle in a given mesh cell l.
3. Notional-particle approach
When performing a 3D simulation of a riser ow, the trans-
ported velocity PDF equation to be solved is a function of 3
independent variables (x, y, z), and of the velocity vector for the
solid phase V. As a result, it is computationally prohibitive to solve
the transported velocity PDF equation accounting for all indepen-
dent variables (position and velocity vectors), using conventional
numerical techniques such as a FV and/or a FD technique, as dis-
cussed in the introduction. On the other hand, the Monte-Carlo
particle mesh technique provides an efcient way to solve such
a large number of equations. In these methods, the transported
velocity PDF equationfor all independent variables are represented
by a large number of notional particles, which can be randomly
or uniformly distributed over the ow domain (Fox, 2003). These
notional particles represent the particle properties such as posi-
tion and velocity. The position of these so-called notional particles
in physical space (Eq. (6)) and their velocity vector (Eqs. (7) and (8))
are determined by solving a set of SDEs. The SDEs for the notional
particles are the replicas of the transported velocity PDF equation
(Eq. (4)). From the solution of the SDEs for the notional particles,
the solution of the transported velocity PDF equation (Eq. (4)) is
obtained.
Lagrangian tracking of the notional particles consists of the fol-
lowing three steps: (1) determine the cell a notional particle is
locatedin, (2) interpolate the eldvariables at the notional-particle
location, and (3) update the position and velocity vector of the
notional-particle according to Eqs. (7 and 8), using a time integra-
tion method (Kolhapure, Fox, Daiss, & Maehling, 2005).
During time integration, the notional particles will freely move
in the physical riser space, during each time step. To determine the
position of a grid cell a notional particle is located in, the modied
brute force search algorithm is implemented as described by Apte,
Mahesh, Moin, and Oefelein (2003) and Vegendla et al. (2009). The
Eulerian properties of the gas phase need to be updated prior to
the interpolation of the position of the notional particles using a
tri-linear inverse distance weighting method (Sadarjoen, Walsum,
Hin, & Post, 1997). Next, to update the position and the velocity of
a notional particle, a rst-order explicit forward Euler scheme is
implemented (Vegendla et al., 2009).
4. Hybrid algorithm
As seen in Fig. 1, the hybrid algorithm applies the FV technique
to solve the gas phase equations and the Monte-Carlo technique to
follow the solid phase notional particles through the grid. Before
starting the simulation, the gas and the solid phase hydrodynamic
properties are initialized. Gas phase simulations are performed
until the required convergence is achieved. Next, the properties of
1196 S.N.P. Vegendla et al. / Computers and Chemical Engineering 35 (2011) 11921199
Iterate PDF code
Start
Initialize FV and PDF code
Iterate FV code until
convergence
Interpolate gas flow properties
No
t>t
sim
Yes
Stop
V
Fig. 1. Simulation procedure for cold ow riser.
the gas phase are interpolated to the position of the solid notional
particles as discussed above. The simulation continues, until the
specied number of time steps is achieved. The solid phase proper-
ties are averaged over a grid cell (Eq. (14)). The obtained averaged
solid velocity is used to calculate the solid phase contribution to
the gas phase conservation equations, such as the drag term in the
gas phase momentum equation. The overall combination of itera-
tions is alternated using the FV method and the PDF method until
the specied convergence is achieved. The time-average method is
implemented, using 2000 time steps, once a statistically stationary
state is reached.
5. Results and discussion
The riser geometry and the operating conditions for the pre-
sented simulation results are collected in Table 4. A schematic
diagram of the pilot plant riser is given in Van engelandt et al.
(2007). The riser has one solids side inlet having a 35

angle with
the riser axis. The solids inow is oriented downward as seen in
Fig. 2. The presented results are obtained by simulating the bot-
tom of the riser with a free top outlet, to limit the computational
time. The center of the side inlet is positioned at 0.5mriser height.
Limiting the simulations to the bottompart of the riser is validated
i ii
YZ-plane
-0.05 m
+0.05 m
Side inlet
Top-view
XZ-plane
+0.05 m
-0.05 m
35
0
YZ-plane
0.5 m
1 m
Air
Solids
0.51 m
Fig. 2. Schematic diagram of the riser (i) side view and (ii) top view.
by the fact that the inlet and outlet effects are quickly dissipated
and do not inuence one another in small diameter risers. More-
over, in the dilute owcase, no downward owhas been observed
in the riser (Chan, Seville, Parker, & Baeyens, 2010; De Wilde, Van
engelandt, et al. 2005). On the other hand, when the solids loading
fromthe side inlet increases (L-valve recycling), a downward solids
velocity component, prior to the particle pick-up by the gas stream,
was observed by Chan, Seville, Fan, and Baeyens (2009).
In the present work, the simulation results obtained using
the EulerianEulerian (two-uid model) method and the
EulerianLagrangian method, are compared with the experi-
mental results of Van engelandt et al. (2007). The experimental
observations of Van engelandt et al. (2007) show that there is
bypassing of the solids near the solids side inlet, with steep
gradients of the solids velocities. However, it was also observed
that these side inlet effects are quickly dissipated in the small
diameter riser.
As seen in Fig. 3, the simulated solids velocity radial pro-
les presented at a riser height of 0.52m (just above the top of
the side inlet) in the YZ-plane (see Fig. 2), calculated using the
EulerianLagrangian method agree very well with the obtained
experimental data. On the other hand, the EulerianEulerian simu-
lation results considerably deviate from the experimental data. In
the latter simulation, a fast diffusion of the solids over the entire
riser cross section near the side inlet is calculated as compared to
the experimental observations. This diffusion gives a more uniform
solids fraction in the YZ-plane resulting in a higher solids velocity
inthat plane. Thus, the solids velocity is overpredictedas compared
to the experimental data. An incorrect numerical diffusion of the
results due to the use of EulerianEulerian method is at the origin
of the overprediction.
As seen in Fig. 3, the solids velocity near the side inlet is low
as compared to the solids velocity near the riser wall opposite to
the solids inlet. The latter is mainly due to side inlet effects. As
seen in the Fig. 4(i), the experimental solids velocities near the riser
wall opposite to the solids inlet are somewhat higher than the inlet
supercial gas velocity (5.31ms
1
). These higher solids velocities
are mainly due to the fact that the gas pushes the solids towards
this side wall, resulting in a rise of the gas and the solids velocities.
InFig. 4, thesimulated(i) solids volumefraction, (ii) solids veloc-
ityand(iii) RMS solids velocityuctuationradial proles, are drawn
at a riser height of 0.52m in the XZ-plane (see Fig. 2). As seen
in Fig. 4, near the side inlet, bypassing of the solids is observed
and the solids velocity is high at the off-center of the riser. The
Fig. 3. Comparison of the simulated and the measured radial proles of the solid
velocities across the YZ-plane (see Fig. 2) at a riser height of 0.52m (symbols: cir-
cles, experiment; dashed line, EulerianLagrangian (see equations in Table 1 and Eq.
(7)) and solid line, EulerianEulerian (see equations in Tables 13)). Conditions: see
Table 4.
S.N.P. Vegendla et al. / Computers and Chemical Engineering 35 (2011) 11921199 1197
Table 4
Operating conditions.
Experiments Simulations
Riser diameter (m) 0.1 0.1
Side inlet position at riser height (m) 0.5 0.5
Outlet surface area (m
2
) 0.00785 0.00785
Inlet surface area for gas (m
2
) 0.00785 0.00785
Inlet solids mass ux in riser (kgm
2
s
1
) 3 (7% side valve opening of 0.005m
2
) 3 (side inlet area 0.005m
2
)
Gas ow rate (m
3
h
1
) 150 150
Density of the solid (kgm
3
) 1550 1550
Diameter of the solid particles (m) 77 77
uo (ms
1
) [0,0, 5.31] [0,0, 5.31]
vo (ms
1
) (side inlet) 35

with riser axis [1.4, 0, 0.9] [1.4, 0, 0.9]


solids volume fraction at the core of the riser (Fig. 4(i)) obtained
using the EulerianLagrangian method compares very well with
the experimental data. Near the side wall, the predicted results
using the EulerianLagrangian method showsome deviations from
the experimental data. This is due to the implemented numerical
simulation technique which predicts less diffusive results. Over-
all, the solids volume fractions are somewhat over predicted as
compared to the experimental observations. On the other hand, it
is clear that the EulerianEulerian simulation results are far more
diffusive than the experimental data and the EulerianLagrangian
results. This results in solids volume fraction values that are largely
underpredicted at the core of the riser, while they are considerably
overpredicted near the solid wall of the riser. As expected, the sim-
ulated solids velocity proles (Fig. 4(ii)) showan opposite behavior
to the solids volume fraction proles.
In Fig. 4(iii), the solids axial Root Mean Square (RMS) velocity
uctuations in the XZ-plane are presented. The experimental RMS
velocity uctuations are found to be high at the position of the
solids side inlet. These RMS velocity uctuations are a measure of
the mixing of the solid particles in the riser ow. The higher the
uctuations, the higher the mixing rate of the solid particles. The
simulated RMS velocity uctuations across the radial direction are
found to be well matching with the experimental data.
As seen in Fig. 5, the simulated elds for the solids vol-
ume fraction, solids velocity, gas phase turbulent kinetic
energy and its dissipation using the EulerianEulerian method
and the EulerianLagrangian method clearly differ. For
EulerianLagrangian simulations, the solids are highly con-
centrated in the core of the riser (Fig. 5(i)), especially at the
height of the side inlet. For EulerianEulerian simulations the
solids volume fractions have over-all lower values as compared
to the values obtained when performing EulerianLagrangian
simulations (Fig. 5(i)). At the bottom of the riser, below the side
inlet section (that is below 0.48m), no solids are simulated using
either of the two methods. This is a consequence of the dilute
solids ow in the riser. The inowing solids are directly taken
upward by the gas ow. Using both the simulation methods, the
solids volume fractions are found to become more uniform above
the riser height of 0.54m. The latter is due to the fast diffusion of
the solids over the cross section of the riser. But, as already dis-
Fig. 4. Comparison of the simulated and measured radial proles across the XZ-plane (see Fig. 2): (i) solids volume fraction; (ii) solids velocity; (iii) solids RMS velocity, at the
riser height of 0.52m (symbols: circles, experiment; dashed line, EulerianLagrangian (see equations in Table 1 and Eq. (7)) and solid line, EulerianEulerian (see equations
in Tables 13)). Conditions: see Table 4.
1198 S.N.P. Vegendla et al. / Computers and Chemical Engineering 35 (2011) 11921199
Fig. 5. Fields of EulerianLagrangian (EL) (see equations in Table 1 and Eq. (7)) and EulerianEulerian (EE) (see equations in Tables 13) simulations: (i) solids volume
fraction; (ii) solids velocity; (iii) gas phase turbulent kinetic energy; and (iv) gas phase turbulent kinetic energy dissipation, in the XZ-plane (perpendicular to the middle of
the side inlet). Conditions: see Table 4.
cussed in Fig. 4, the EulerianEulerian simulation results are found
to be more diffusive than the EulerianLagrangian simulation
results.
Using the EulerianLagrangian method, low solids velocities
are predicted in the core of the riser as compared to the val-
ues near the riser wall. This effect is not observed for the
EulerianEulerian results. Using the latter simulation method, the
solids velocities are found to be more uniform in the riser, except
near the side inlet (Fig. 5(ii)). At the side inlet, when compar-
ing both the simulation results of the EulerianLagrangian and
the EulerianEulerian method, remarkable differences in the solids
velocities are observed. Again, these difference are due tothe differ-
ences in diffusiveness of the solids volume fraction as discussed in
theaboveparagraph, i.e. differenceinthesolids ratemixing. As seen
in Fig. 5(iii and iv), EulerianLagrangian simulation results predict
high values for the gas phase turbulent kinetic energy and its dissi-
S.N.P. Vegendla et al. / Computers and Chemical Engineering 35 (2011) 11921199 1199
pationas comparedto EulerianEuleriansimulations, starting from
the side inlet to the top of the riser.
To summarize, the simulation results obtained using the
EulerianLagrangian method compare well with the experimen-
tal observations, qualitatively and quantitatively. The simulation
results obtained using the EulerianEulerian method deviate con-
siderably from the experimental observations. These differences
are mainly due to the diffusive nature of the EulerianEulerian
simulation method as compared to the EulerianLagrangian
method and to the experimental data. The main advantage of the
EulerianLagrangian method is the fact that there is no numerical
diffusionwhensolvingthesolidphaseequations duetothetracking
of the notional particles in the riser.
6. Conclusions
The EulerianLagrangian method enables to simulate a dilute
gassolid riser cold owaccounting in an exact way for the convec-
tion in the solid phase momentum equation using the transported
velocity PDF equation for the solid phase. In this method, the gas
phase is considered as a continuous phase while the solid phase is
considered as a dispersed phase. Reynolds Averaged Navier-Stokes
(RANS) equations are formulated for the gas phase with the tur-
bulence of the gas phase modeled using the standard k model.
The Finite Volume (FV) technique is implementedtosolve these gas
phase equations. Stochastic differential equations (SDEs) are intro-
duced as replicas of the transported velocity PDF equation for the
solidphase. Anotional-particle basedMonte-Carlo meshtechnique
is used to solve these equations, thus determining the position
and the velocity of these computational particles. The advantage
of using the latter technique is free from false numerical diffusion
in the simulation results as compared to the use of the FV tech-
nique to solve the solid phase transported velocity PDF equation.
The EulerianLagrangian method is found to be adequate to sim-
ulate a side inlet dilute cold ow riser. The obtained simulation
results using the EulerianLagrangian method corresponds qual-
itatively and quantitatively with the available experimental data.
On the contrary, the simulation results obtained when using the
EulerianEulerian method are found to deviate considerably from
the experimental data.
References
Apte, S. V., Mahesh, K., Moin, P., & Oefelein, J. C. (2003). Large-eddy simulation of
swirling particle-laden ows in a coaxial-jet combustor. International Journal of
Multiphase Flow, 29, 13111331.
Benyahia, S., Syamlal, M., & OBrien, T. J. (2005). Evaluation of boundary conditions
used to model dilute, turbulent gassolid ows in a pipe. Powder Technology,
156, 6272.
Chalermsinsuwan, B., Piumsomboon, P., &Gidaspow, D. (2009). Kinetic theory based
computation of PSRI riser. I. Estimate of mass transfer coefcient. Chemical Engi-
neering Science, 64, 11951211.
Chan, C. W., Seville, J., Fan, X., &Baeyens, J. (2009). Solidparticlemotioninastandpipe
as observedbypositronemissionparticle tracking. Powder Technology, 194(12),
5866.
Chan, C. W., Seville, J. P. K., Parker, D. J., & Baeyens, J. (2010). Particle velocities
and their residence time distribution in the riser of a CFB. Powder Technology,
doi:10.1016/j.powtec.2010.05.008
Chibbaro, S., & Minier, J. P. (2008). Langevin PDF simulation of particle deposition in
a turbulent pipe ow. Aerosol Science, 39, 555571.
Das, A. K., Baudrez, E., Marin, G. B., & Heynderickx, G. J. (2003). Three-dimensional
simulation of a uid catalytic cracking riser reactor. Industrial and Engineering
Chemistry Research, 42, 26022617.
Das, A. K., De Wilde, J., Heynderickx, G. J., Marin, G. B., Vierendeels, J., & Dick, E.
(2004). CFDsimulation of dilute phase gassolid riser reactors. I. A newsolution
method and ow model validation. Chemical Engineering Science, 59, 167186.
Das, A. K., De Wilde, J., Heynderickx, G. J., & Marin, G. B. (2004). CFD simulation of
dilute phase gassolid riser reactors. II. Simultaneous adsorption of SO
2
-NOx
from ue gases. Chemical Engineering Science, 59, 187200.
Deen, N. G., Van Sint Annaland, M., Van der Hoef, M. A., & Kuipers, J. A. M. (2007).
Review of discrete particle modeling of uidized beds. Chemical Engineering
Science, 62, 2844.
De Wilde, J., Heynderickx, G. J., Vierendeels, J., Dick, E., & Marin, G. B. (2002). An
extension of the preconditioned advection upstream splitting method for 3D
two-phase ow calculations in circulating uidized beds. Computers and Chem-
ical Engineering, 26(12), 16771702.
De Wilde, J., Marin, G. B., & Heynderickx, G. J. (2003). The effects of abrupt T-outlets
in a riser: 3D simulation using the kinetic theory of granular ow. Chemical
Engineering Science, 58(36), 877885.
De Wilde, J., Vierendeels, J., Heynderickx, G. J., & Marin, G. B. (2005). Simultane-
ous solution algorithms for EulerianEulerian gassolid ow models: Stability
analysis and convergence behavior of a point and a plane solver. Journal of
Computational Physics, 207, 309353.
De Wilde, J., Van engelandt, G., Heynderickx, G. J., & Marin, G. B. (2005). Gassolids
mixing in the inlet zone of a dilute circulating uidized bed. Powder Technology,
151, 96116.
Dhanunjay, S. B. (2003). Simulation of granular and gassolid ows using Discrete
Element Method. Carnegie Mellon University, PhD thesis.
Fox, R. O. (2003). Computational models for turbulent reacting ows (1st ed.). Cam-
bridge University Press.
Gidaspow, D. (1994). Multiphase ow and uidization: Continuum and kinetic theory
descriptions. New York: Academic Press.
Jiradilok, V., Gidaspow, D., Damronglerd, S., Kobes, W. J., &Mosto, R. (2005). Kinetic
theory based CFD simulation of turbulent uidization of FCC particles in a riser.
Chemical Engineering Science, 61, 55445559.
Kolhapure, N. H., Fox, R. O., Daiss, A., & Maehling, F. O. (2005). PDF simulation of
ethylene decomposition in tubular LDPE reactors. AIChE Journal, 51, 585606.
Launder, B. E., &Spalding, D. B. (1974). The numerical computationof turbulent ow.
Computer Methods in Applied Mechanics and Engineering, 3, 269.
Minier, J. P., & Peirano, E. (2001). The PDF approach to turbulent polydispersed two-
phase ows. Physics ReportsReview Section of Physics Letters, 352, 1214.
Pope, S. B. (1985). PDF methods for turbulent reactive ows. Progressive Energy
Combustion Science, 11, 119192.
Pope, S. B. (1987). Consistency conditions for random-walk models of turbulent
dispersion. Physics of Fluids, 30(8), 23742379.
Ranade, V. V. (2002). Computational ow modeling for chemical reactor engineering.
Academic Press.
Sadarjoen, A., Walsum, T. V., Hin, A. J. S., & Post, F. H. (1997). In G. Nielson, H. Hagen,
& H. Mller (Eds.), Particle tracing algorithms for 3D curvilinear grids, scientic
visualization: Overviews, methodologies, and techniques (pp. 311335). IEEE CS
Press.
Subramaniam, S., &Haworth, D. C. (2000). Aprobability density function method for
turbulent mixing and combustion on three-dimensional unstructured deform-
ing meshes. Journal of Engine Research, 1, 171190.
Van de Velden, M., Baeyens, J., Dougan, B., & McMurdo, A. (2007). Investigation of
operational parameters for an industrial CFB combustor of coal, biomass and
sludge. China Particuology, 5(4), 247254.
Vanengelandt, G., De Wilde, J., Heynderickx, G. J., &Marin, G. B. (2007). Experimental
study of inlet phenomena of 35

inclined non-aerated and aerated Y-inlets in a


dilute cold-ow riser. Chemical Engineering Science, 62(12), 339355.
Vegendla, S. N. P., Heynderickx, G. J., & Marin, G. B. (2009). Probability density
function simulation of turbulent reactive gassolid ow in a riser. Chemical
Engineering and Technology, 32(3), 492500.

You might also like