You are on page 1of 9

Chem. Eng. Technol. 2009, 32, No.

6, 939947

Stefan Iglauer1
Hans-Joachim Warnecke2
1

Earth Science and Engineering


Department, Imperial College
London, London, UK.
Universitt Paderborn,
Technische Chemie und
Verfahrenstechnik, Paderborn,
Germany.

939

Research Article

Simulation and Experimental Validation of


Two-Phase Flow in an Aerosol-Counter Flow
Reactor using Computational Fluid Dynamics
A simulation of the hydrodynamic behavior of an aerosol-counter flow reactor
was conducted using an Euler-Lagrange method. The simulation results were
then verified with experiments. The process simulated was a separation process
required during the production of biodiesel (fatty acid methyl ester). In this process, the liquid ester/glycerol phases are continuously injected through a hollow
cone nozzle with an overpressure of 106 Pa into the reactor, operated at 15000 Pa.
The liquid is atomized because of the pressure drop and a liquid particle spray is
generated with an inlet velocity of 44.72 m/s. Water vapor of temperature 333 K
is injected tangentially through two side, gas inlets with an inlet velocity of
1.2 m/s. Excess methanol is subjected to a mass transfer from the liquid phase
into the gas phase, which is withdrawn through the head of the reactor and condensed in an external condenser unit. The stripping of the methanol off the liquid
leads to a sharp interface between the glycerol and the ester phase, which can then
be easily separated by gravity or pumping. The gas velocity field, pressure field
and the liquid particle trajectories were calculated successfully. Simulated dwell
time distribution curves were derived and analyzed with the open-open vessel
dispersion model. Experimental dwell time distribution curves were measured,
analyzed with the open-open vessel dispersion model, and compared with the
simulated curves. A good consistency between simulated and measured Bodenstein numbers was achieved, but 25 % of the simulated particles exited at the
reactors head, contrary to experimental observations. The difference between
simulated and measured dwell times was within one order of magnitude.
Keywords: Aerosol-counter flow reactor, Biodiesel, Computational fluid dynamics,
Euler-Lagrange method, Two-phase flow
Received: January 20, 2009; revised: February 11, 2009; accepted: February 27, 2009
DOI: 10.1002/ceat.200900034

Introduction

Aerosol-counter flow reactors are commonly used reactors in


the chemical industry for continuously stripping contaminating volatile organic or inorganic substances off product
streams. This process is called desodorization and results in
higher product purity. Example industrial processes include
removal of styrol out of polymer dispersions [1], methanol

Correspondence: Dr. S. Iglauer (s.iglauer@imperial.ac.uk), Earth


Science and Engineering Department, Imperial College London, Prince
Consort Road, SW7 2AZ London, UK.

2009 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim

out of fatty acid methyl ester [2, 3], or ammonia out of contaminant liquids or percolate water [4, 5, 6].
This work was part of a larger industrial biodiesel production project and the work was carried out to improve understanding of the fluid dynamic processes in the reactor, so that
the reactor design can be further optimized. Biodiesel is a renewable fuel resource which can be manufactured out of agricultural raw materials.
During the production of biodiesel (fatty acid methyl ester),
a two-phase liquid product stream of 1,2,3-propanetriol (glycerol) and fatty acid methyl ester results, mixed with methanol
and rapeseed oil, which was not chemically converted. Because
of the reversibility of the reaction, an excess of methanol is
added in order to shift the thermodynamic equilibrium to the
right side (Fig. 1) [2].

http://www.cet-journal.com

940

Chem. Eng. Technol. 2009, 32, No. 6, 939947

S. Iglauer et al.

The particles do not deform.


The particles are spherical and free
of contaminations.
O
O
R
OH

The mass balance of the particles is


R
O
+ 3 MeOH
HO
+ 3
ignored.
OH
MeO
R
O
R
Compressible fluids up to a Mach
number, Ma1), of 0.3 are treated as
O
incompressible fluids; for Ma larger
(c)
(d)
(a)
(b)
than 0.3, the resistance coefficient,
cw, is a function of the Reynolds
Figure 1. Reaction scheme of re-esterification. (a) rapeseed oil, (b) methanol, (c) 1,2,3-pronumber, Re, and Ma; cw = f(Re, Ma).
pantriol, (d) fatty acid methyl ester = biodiesel.
In the discrete Lagrange projection,
each cell represents the same fluid eleIn order to strip off excess methanol, the liquid two-phase
ment at every time step. External and surface forces, which act
product stream is sprayed into an aerosol-counter flow reactor
on the cell, are easier to determine and it is so easier to dethrough a hollow cone nozzle [7] with a 0.5-mm bore and an
scribe the response of the elements. After calculation of the
opening angle of 70. The volume flow into the reactor
new velocities, the fluid elements move to their new positions;
amounts to 0.002777 m3/s, which corresponds to an inlet flow
this is achieved by shifting the coordinate system according to
velocity of 44.72 m/s. The operating pressure of the reactor is
the new velocities.
15000 Pa and the dynamic flow pressure at the inlet nozzle is
The movement of the disperse phase is computed via ordin106 Pa. Water vapor is tangentially injected with a velocity of
ary differential equations. The equations are given below.
1.2 m/s through two side inlets at a temperature of 333 K. The
Eq. (1) calculates the spatial position of the particle.
methanol is subjected to a mass transfer from the liquid phase
into the gas phase [8]. The breaking of the liquid injection
d~
n ~
C
(1)
stream by the pressure drop increases the interfacial area so
dt
that the mass transfer of methanol is accelerated [9, 10]. The
~
n is the position of the particle in the mathematical space, t is
methanol desaturated liquid phase is collected at the bottom
time, and ~
C is the mathematical velocity. The trajectories are
of the reactor, and the methanol-water vapor phase is led
calculated in the mathematical space in order to simplify the
through the reactors head into a condenser unit where it is licalculation of the trajectories and the transfer at control volquified and removed. The stripping of methanol off the bioume interfaces [16].
diesel product phases results in a sharp phase separation beAfter converting back into physical space, the change of the
tween the glycerol phase and the ester phase as methanol acts
velocity vector of the particle is calculated by applying Newas a solubility enhancer between these phases. This basically
tons second law:
enables a simple two-phase gravity separation, and the biodiesel can be easily separated by pumping or simple drainage.
d~
v
The 3D gas velocity and pressure fields, and the trajectories
m ~
Ftotal
(2)
dt
of the liquid particles were calculated with the Euler-Lagrange
method [1113]; the computational results were verified exwhere ~
Ftotal is the total force acting on the particle and m is the
perimentally by dwell time measurements.
mass of the particle. ~
Ftotal consists of two components, namely:
The gravity force, ~
Fg ; considering the gravitational constant,
~
~
g,
the
particle
diameter,
2 Theoretical Background
 d, and the particle density, q, Fg
1
3
~
amounts to Fg pd qb qa ~
g.
O

2.1 The Euler-Lagrange Method


The Euler-Lagrange method sets one phase as continuous and
homogeneous, and the second phase as disperse [14]. The
homogeneous phase is treated as a continuum based on the
Euler approach by solving the energy, mass, and momentum
balances. The movement of the disperse phase is solved with
the Lagrange method by following the trajectories of a large
number of particles through the mathematical space.
The Euler-Lagrange model makes the following assumptions:
The disperse phase can exchange momentum, mass, and energy with the continuous phase. These exchanges are described via interaction terms.
The particle volumes are significantly smaller than the volume of the continuous phase.

2009 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim

The friction force, ~


Fr , which constitutes the major part of
the total force, and is calculated as follows:
1
~
vr j~
vr
Fr pd2 qa cw j~
8

~
vr is the relative velocity between the continuous and the
disperse phase, and cw is the drag coefficient which is a function of the Reynolds number [17].
Other forces were not considered in this simulation.
In this work, the gas phase, a, was considered to be the continuous phase and was, therefore, calculated via the Euler
method. The liquid phase, b, was considered to be disperse,
and was calculated using the Lagrange method. Gas velocities

1)

List of symbols at the end of the paper.

http://www.cet-journal.com

Chem. Eng. Technol. 2009, 32, No. 6, 939947

Computational fluid dynamics

in three space dimensions, ~


v, and the pressure, pa, of the gas
phase were calculated; moreover, the trajectories of the disperse liquid particles were computed.

941

2.2 Turbulence Model


The operating pressure and temperature of the reactor combined with the gas and liquid injection velocities resulted in
turbulent fluid flow in the reactor. This turbulence was modeled with the standard k-e model [1821]. The k-e model assumes that the flow is fully turbulent and the effects of molecular viscosity are negligible. We selected the k-e model because
we observed that the water vapor and the aerosol flowed turbulently through the reactor space. Moreover, the k-e model is
robust and computationally economic.
The k-e model utilized in this work solved the following
equations:
The continuity equation:
 ~
va 0

(3)
a

and the momentum balance:




va
qa~
 leff ~
va
~
 qa~
va
va
t

T 
px  leff ~
va
~
Ftotal

Figure 2. Geometry of the aerosol-counter flow reactor.

(4)

where ~
va is the average velocity of the continuous phase, ~
Ftotal
is the total force acting on the particle, leff is the effective viscosity, and px is a modified pressure.
The following algorithm was used to compute the particle
transport. At first, the partial differential equations for the
continuous phase were solved and, subsequently, the ordinary
differential equations for the Lagrange model were solved. The
Lagrange model therefore calculated an additional iteration for
the gas phase. If such a computational step is conducted once,
then no further influence of the particle on the continuous
phase is considered. If the iteration is repeated more than
once, the influence of the particle on the continuous phase is
considered. In this simulation, 50 such iteration cycles were
performed in order to account for the particle-gas interaction.

2.3

tained 7800 cells (see Figs. 3 and 4); the cells in the gas inlets
were somewhat distorted (see Fig. 4), which could not be
avoided because of the geometry of the inlets and the use of a
structured mesh.
The adapted block-structured H-mesh contained more cells,
specifically around the liquid injection nozzle and, therefore,
required more computational power than the non-adapted
mesh. It was shown that the adapted mesh does not signifi-

Reactor Geometry and Mesh Generation

The geometry of the aerosol counter-flow reactor is shown in


Fig. 2.
The tangential gas inlets are located at points b and c, respectively; the vapor outlet is at the reactor head, d, and the
liquid outlet is at the reactor bottom, a. The x-axis is the reactors vertical axis, the y-axis is perpendicular to the tangential
gas inlets, and the z-axis is parallel to the gas inlets. The u-velocity is the velocity component in x-direction, the v-velocity
in the y-direction, and the w-velocity in the z-direction.
Three different mathematical meshes were tested, namely
a block-structured H-mesh, an adapted block-structured
H-mesh, and an O-mesh. The block-structured H-mesh con-

2009 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim

Figure 3. Frontal (xy) slice through the block-structured H-mesh,


showing the whole reactor.

http://www.cet-journal.com

942

Chem. Eng. Technol. 2009, 32, No. 6, 939947

S. Iglauer et al.

Results were visualized with the post-processing modules of


this program.
Two simulations were performed; the first computation included 20 particles while the second computation included
100 particles. This was done in order to investigate the effect
of the number of particles (and, therefore, the momentum interaction between the gas and the liquid phases) on the solution. The 100-particle calculation was performed with the gas
velocity field computed by the 20-particle solution as the starting condition in order to save computation time.

Figure 4. Axial (yz) slice through the block-structured H-mesh,


showing the reactor at the gas inlet height.

cantly improve resolution and simulation results as compared


to the non-adapted mesh, and it was therefore not used in subsequent calculations.
The O-grid contained 196100 cells. This mesh was not suitable for simulations as the gas inlets could not be incorporated. Moreover, the nozzle was located exactly on the singularity and shifting the mathematical space would of course
introduce an error. The large number of cells costing more
computational time was another disadvantage of this O-grid.
However, one advantage of this O-grid was the higher average
orthogonality of the cells so that the coordinate transformation would be easier to calculate.

2.4

Boundary Conditions

The normal velocities at the gas inlet faces were set to 1.2 m/s
and standard logarithmic wall laws were applied. The mass
flow boundary conditions were Neumann conditions and the
mass flux was set in a way that all gas exited through the reactors head as observed in the experiments.

Simulated Continuous Gas Phase and Disperse


Liquid Phase

Calculated particle trajectories and the gas velocity field for the
20-particle solution are presented in Figs. 59.

Physical and Mathematical Parameter Selection

The hollow cone nozzle was simulated by point ~


x = (0.9, 0, 0),
in which all particle trajectories started. This introduced a
small error as the bore of the nozzle was 0.5 mm. However, this
error is small as the diameter of the reactor was 0.4 m. The
gravity vector was set to (9.81 m/s2, 0 m/s2, 0 m/s2), the gas
density was set to 0.2 kg/m3, and gas viscosity, la, was set to
1.72 105 Pa s. The density of the disperse phase was set to
1000 kg/m3 and the radius of the particles was set to
0.00005 m. The numerical CFD solver did not permit transient
calculations for the Lagrange model. Therefore, stationary
conditions were selected.

2.5

3.1

Results and Discussion

The computations were conducted with the commercial software program, CFX 4.2 [15], on an IRIX SGI workstation.

2009 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim

Figure 5. Trajectory of one selected particle.

Some particles left through the reactors head; these were


classified as outliers because experimentally, all liquid exited
through the reactors bottom. The outliers probably resulted
from the character of the k-e turbulence model. The k-e model
has been developed for description of single-phase flow and is
therefore not arbitrarily applicable to two-phase flow. In this
work, the disperse phase has influence on the flow field of the
continuous phase; this is particularly evident at the injection
nozzles location, but also occurs in the rest of the reactors
volume. Because of this effect due to the interaction between
the two phases, the k-e model produces inaccuracies. If all outliers are eliminated, the particle trajectory paths result as
shown in Fig. 6.
The hollow cone character of the liquid nozzle spray was experimentally visually observable and was reproduced by the

http://www.cet-journal.com

Chem. Eng. Technol. 2009, 32, No. 6, 939947

Computational fluid dynamics

943

Figure 8. w-velocity of the gas phase [m/s].

Figure 6. Trajectories of all particles exiting the reactor at the


bottom.

Figure 7. v-velocity of the gas phase [m/s].

calculation (cp. cone formed by particle trajectories downstream of the nozzle; Fig. 6). This is evidence that the simulation is more or less correct. The w-velocity amounted to
1.2 m/s and 1.2 m/s, respectively, at the inlet faces as set by
the boundary condition. The gas stream at these faces was
mixed tangentially due to the tangential geometry. This mixing
was intensified within the reactor. The v-velocity showed a radial mixing in the yz-plane at the height of the gas inlets. The
absolute velocities were lower than the w-velocities because, in
the tangentially inflowing gas, momentum transfer already
happened.
The residuals for the calculations are displayed in Tab. 1 for
the 20-particle computation and for the 100-particle computation.

2009 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim

Figure 9. u-velocity of the gas phase [m/s].

The residual depends on the value of the state variable and


the number of cells as the residual is an average over all errors
of all cells. The residual of the mass the most important variable was comparatively high, but acceptable. Because of this,
the solution can be considered to be an approximation. The
convergence of the u-, v- and w-velocity was of the same order
so that this is also an approximate solution. The modest convergence was possibly due to the Gaussian probability function, which is implemented in the turbulent particle dispersion
model. This model was, however, required in order to characterize the statistical behavior of the particle trajectories and to
gain a physically meaningful solution.

http://www.cet-journal.com

944

Chem. Eng. Technol. 2009, 32, No. 6, 939947

S. Iglauer et al.

Table 1. Parameter residuals.


Parameter

Residuals
20 particles

Residuals
100 particles

10 kgm /s

10 kgm/s

102 kgm/s2

101 kgm/s2

102 kgm/s2

101 kgm/s2

102 kgm/s2

10 kgm /s

10 kgm /s
1

10 kgm/s

[23], Fritz und Fritzer [24]). The analytical solution of Eq. 6


for this open-open dispersion model is given by Eq. 7.

10 kgm /s
2

s
Bo
Bo1 h2
Eh

exp
4h
4ph

4
3

where E is the dwell time distribution function.


The average normalized dwell time results as:
s
2

h 1
sh
Bo

10 kgm/s

The pressure drop within the reactor amounted to ca. 4 Pa


over the total reactor height and it decreased continually. The
isobars were almost parallel and parallel to the yz-axis. An
irregularly low pressure was located around the injection
nozzle as expected.

3.2 Verification of Simulated Results with


Quantitative Experimental Data
The simulation results for the pressure and the u-, v-, and
w-velocities are physically reasonable; in order to further verify
these results, simulated average dwell times and Bodenstein
numbers were compared to experimentally measured ones.

Calculation of Average Dwell Time and


Bodenstein Number

The average dwell time of the tracer in the experiments was


calculated with the mass balance (Eq. 5); Eq. 5 is also called
the dispersion model.
c

c
2 c
Dax 2
x
x

(5)

where c is the tracer concentration, t is elapsed time from tracer injection start, u is the average flow velocity in the axial direction, x is the reactor length, and Dax is the axial dispersion
coefficient.
The dispersion model describes an ideal piston-like flow
with ideal mixing in the radial direction (Baerns et al. [22],
Levenspiel [23], Fritz und Fritzer [24]). The dimensionless
form of Eq. 5 is given by Eq. 6.
c
1 2 c

h Bo z2

c
z

(6)

where the Bodenstein number, Bo = uL/Dax, the normalized


dwell time, H = t/tn = tu/L, and the normalized reactor length,
z = x/L. tn is the hydrodynamic dwell time.
We solved Eq. 6 with the open-open dispersion model,
which assumes that there is no discontinuity in tracer concentration at reactor inlet and outlet (Baerns et al. [22], Levenspiel

2009 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim

(8)

and the standard deviation, rh, of the average normalized


dwell time amounts to:
rh

r
2
8

Bo Bo2

(9)

As can be seen in Eq. 9, the Bodenstein number, Bo, is an indicator for the standard deviation of the average normalized
dwell time, i.e., Bo describes the narrowness of the dwell
time curve.
The average simulated dwell time and Bodenstein number
of the simulated particles were calculated in an analogous way
to quantify reactor flow parameters and make comparisons
with the experiments in a straightforward manner.

3.4
3.3

(7)

Simulated Dwell Times

The code computed a trajectory for each particle, including


position vector, velocity vector, elapsed time, and corresponding residuals of each state variable. The calculated data included dwell times so that the cumulative dwell time curve
could be derived. Five out of 20 particles were considered outliers as they exited the reactors head.
In a least-square statistical approach, the simulated dwell
time points were fitted with a Boltzmann-shaped curve. A
Boltzmann curve was selected because it represents a good description of the theoretical, ideal cumulative dwell time.
The fitting function is given in Eq. 10.
y

100

 100
x 11:48689
1 exp

(10)

4:33133

Eq. 10 was differentiated so that the dwell time distribution


could be derived. This differentiation was performed in order
to be able to compare the simulated results with the experimental data, i.e., to further analyze the data with the openopen vessel dispersion model [3, 2224].
The open-open vessel dispersion model was utilized to extract the Bodenstein number (Bo) and the average dwell time.
For decreasing Bo, the error associated with using the openopen vessel dispersion model increases. For Bo = 20, the error
is ca. 10 %. As the experimental error is estimated to be of the
same magnitude, the open-open vessel dispersion model was
used for data evaluation.

http://www.cet-journal.com

Chem. Eng. Technol. 2009, 32, No. 6, 939947

Computational fluid dynamics

For the 20-particle calculation, Bo amounted to 36 and the


average dwell time was 8.6 s.
In terms of the 100-particle computation, the number of
particles exiting the reactors head was 26 and showed a relative match with the solution of the 20-particle calculation. The
cumulative dwell time points were again fitted with a least
square-Boltzmann curve; Eq. 11:
100

 100
y
x 11:7241
1 exp

(11)

3:35367

Using the open-open vessel dispersion model, this calculation resulted in a Bodenstein number of 25 and an average
dwell time of 8.6 s.

Experimental Measurements

In order to verify the simulated results, dwell times were measured in the aerosol-counter flow reactor [3]. The experimental conditions are summarized in Tab. 2; six experiments were
conducted in total.

amounted to 6 s. A correction time (tcorrection = 10.294 s) was


subtracted from the measured dwell time curve to account for
these effects.

4.2

Experimental Results

The electrodes recorded 300 readings per second; these data


were averaged in such a way that two data points per second
resulted. The correction time, tcorrection, was considered in the
analysis and the measured data was analyzed with the openopen vessel dispersion model. The resulting Bodenstein numbers and average dwell times are presented in Tab. 3. The data
was highly reproducible and the results of experiment 2 are
exemplarily shown in Fig. 10.

Table 3. Experimental Bodenstein numbers and average dwell


times.
run

Bodenstein number

Average dwell time [s]

14.9

38.4

14.5

39.3

Table 2. Experimental conditions given during dwell time measurements.

15.4

38.5

14.4

37.3

Temperature

333 K

14.7

38.1

Pressure

15000 Pa

14.2

38.0

Liquid flow rate

2.7778 10 m /s

Vapour mass flow


per inlet

55.5556 105 kg/s

4.1

945

Using the open-open vessel dispersion model, the resulting


Bodenstein number amounted to 15 and the average dwell
time to 38 s. The good reproducibility of the experimental
results is confirmed by the low standard deviation (< 0.1) for
all observed variables with a confidence interval of 95 %.

Experimental Procedure

Several millilitres of an aqueous 0.356-molar KCl tracer solution were injected as a Dirac impulse into the reactor
through an external pipe. The tracer solution was pressurized to a pressure above 106 Pa in order to overcompensate
the nozzle overpressure. The solution had to flow through
the external pipe to reach the nozzle, this time, tinject, was
calculated using Eq. 12. It had been assumed that the pipe
was completely filled and behaved ideally. This was experimentally confirmed as the liquid hollow cone did not
break up.
tinject

A r lr
4:294 s
V_

(12)

with a pipe cross-sectional area, Ar = 3.848105 m2, a


pipe length, lr = 0.31 m, and a volume flow rate,
V_ = 2.7778106 m3/s.
The measurement electrode was not directly installed at
the reactors exit. Because of this, the liquid had to flow an
additional path from the exit to the electrode. This additional time, tout, was experimentally determined and

2009 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim

Figure 10. Experimentally-observed dwell time distribution curve of one


selected experiment.

http://www.cet-journal.com

946

Chem. Eng. Technol. 2009, 32, No. 6, 939947

S. Iglauer et al.

Comparison of Experimental and


Simulated Results

The computed cumulative dwell time curves were consistent


with common sigmoidal cumulative dwell time curves. The
computation therefore captured the physical character of this
process.
The computed results are compared with the experimental
results in Tab. 4.
Table 4. Bodenstein numbers and average dwell times for simulations and experiments.
System

Bodenstein number

average dwell time [s]

20 particle model

36

8.6

100 particle model

25

8.6

Experiment

15

38

The 100-particle model gave a better prediction of the experimentally observed Bodenstein number than the 20-particle
model, because the momentum transfer between the dispersed
and the continuous phase was described more accurately.
Overall, the simulations showed a reasonable match with
the experiments in terms of the Bodenstein number, while
average dwell times were estimated with an accuracy of one
magnitude (different by a factor of 4.4).
The difference in average dwell time was due to wall effects
as observed during the experiments a significant number of
particles hitting the reactors wall adhered to the wall and flowed down along the wall. This significantly extended the experimental dwell time and explains the difference. As we had
insufficient experimental data about this wall effect, i.e., we
had only qualitative visual observations, we did not further
attempt to model it with CFD.

Summary and Conclusions

A simulation of the hydrodynamic behavior of an aerosolcounter flow reactor has been performed using an Euler-Lagrange method. The simulated results were compared with
experimental results.
The mass balance of the gas velocity field computation
showed a moderate convergence with a residual of 101 kg/s.
The residuals of the u-, v-, and w-velocities were 101 kgm/s2
and the inaccuracies resulted from the particle trajectory integration. The solution for the gas velocity field is an approximate solution which is physically realistic.
The simulated cumulative dwell time points were fitted with
a Boltzmann curve and derived after time. These simulated
dwell time distribution curves and the experimentally-measured dwell time distributions were analyzed with the openopen vessel dispersion model. A good consistency between
simulated and measured Bodenstein numbers was achieved,
but 25 % of the simulated particles exited at the reactors head,
which was contrary to experimental observations.

2009 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim

The simulated and measured dwell times were within the


same order of magnitude. The difference mainly arose due to
wall effects as a significant percentage of the aerosol hitting the
reactors wall adhered to the wall and flowed down on it, significantly extending the dwell time. This wall phenomenon
was not included in the presented simulations, because only
qualitative data, i.e., visual observations, were available. In order to implement these wall effects as boundary conditions,
the elastic behavior of the particles at the given thermodynamic conditions should be thoroughly investigated experimentally. A further improvement would be to measure the
particle volume distribution and feeding this information into
the model, as this has an influence on the momentum transfer
between the continuous and the disperse phases.
An increase in particle numbers was tested in order to
achieve better convergence and a better description of the
phase interactions. This indeed resulted in a smoother cumulative dwell time curve and a better prediction of the Bodenstein number. However, the computational costs massively increased with larger particle numbers. If a suitable gas velocity
field is given as a starting condition in this coupled process,
the computational effort is minimized.
The Gaussian probability function in the turbulent particle
dispersion model deteriorated convergence because of its statistical character. An alternative would be to resolve the turbulences with a very fine mathematical grid. However, this would
require enormous computer capacities.

Acknowledgements
We would like to thank Dr. Heggemann and Dr. Bremer for
conducting the experimental measurements.

Symbols used
~
xp
m
t
~
v
p
q
l
~
Fg
~
Fw
~
Ftotal
~
Fr
k
e
~
g
Ma
Re
Bo
~
n
*

C
d
cw

[m]
[kg]
[s]
[m/s]
[Pa]
[kg/m3]
[Pas]
[N]
[N]
[N]
[N]
[m2/s2]
[m2/s3]
[m/s2]
[]
[]
[]
[m]
[m/s]
[m]
[]

position vector of particle


mass
time
velocity
pressure
density
dynamic viscosity
gravity force
drag force
total force
friction force
turbulent kinetic energy
dissipation rate
gravity constant
Mach number
Reynolds number
Bodenstein number
position of the particle in the
mathematical space
mathematical velocity
particle diameter
drag coefficient

http://www.cet-journal.com

Chem. Eng. Technol. 2009, 32, No. 6, 939947

A
u
v
w
V_
c
u

[m2]
[m/s]
[m/s]
[m/s]
[m3/s]
[mol/m3]
[m/s]

L
Dax
T
z
tn
E

[m/s]
[m2/s]
[]
[]
[s]
[]

projected frontal area


velocity in x-direction
velocity in y-direction
velocity in z-direction
volume flow rate
concentration
average flow velocity in axial
direction
reactor length
axial disperion coefficient
normalized dwell time
normalized reactor length
hydrodynamic dwell time
dwell time distribution function

Subscripts
a
b
t
p
eff

gas phase
liquid (particle) disperse phase
turbulent
particle
effective

Superscripts
x

modified
average

References
[1] D. Meier, Ph.D. Thesis, University of Paderborn (Germany)
1998.
[2] K. Scharmer, G. Golbs, I. Muschalek, K. Zimalla, Abschlussbericht; Vergleich der Verfahren zur Umesterung und zur Aufreinigung des Hauptproduktes, GET Gesellschaft fr Entwicklungstechnologie mbH, Aldenhoven, Germany, 1993.
[3] M. H. Heggemann, Ph.D Thesis, University of Paderborn
(Germany) 2001.
[4] S. A. Beg, S. Obaid-ur-Rehman, M. M. Hassan, Chem. Ing.
Tech. 1989, 61 (4), 338.
[5] M. H. Heggemann, M.Sc. Thesis, University of Paderborn
(Germany) 1997.

2009 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim

Computational fluid dynamics

947

[6] A. K. Bin, P. Machniewski, Inz. Chem. I. Proces. 1998, 19 (2),


371.
[7] Schlick GmbH & Co., Katalog: Schlick-Dsen, Filter, Apparatebau, Untersiemau bei Coburg, Germany 1997.
[8] H. Brauer, Stoffaustausch einschlielich chemischer Reaktionen, Verlag Sauerlnder, Frankfurt am Main 1971.
[9] E. Moreira et al., Chem. Eng. Comm. 2005, 192, 1017.
[10] M. A. Moris, F. V. Diez, J. Coca, Sep. Purification. Technol.
1997, 11, 79.
[11] A. Kayode Coker, in Modeling of chemical kinetics and reactor
design (Ed.: A. Kayode Coker), Gulf Publication, Texas 2001.
[12] C. Gentric, D. Mignon, J. Bousquet, P. A. Tanguy, Chem.
Eng. Sci. 2005, 60, 2253.
[13] V. V. Ranade, in Computational flow modeling for chemical
engineering (Ed.: V. V. Ranade), Academic Press, California,
2002.
[14] C. Drumm, H. J. Bart, Chem. Eng. Technol. 2006, 29, 1297.
[15] AEA-TECHNOLOGY, CFX-4.2 Solver Manual, AEA-Technology plc, Harwell, UK 1997.
[16] H. R. Schwarz, Numerische Mathematik, B. G. Teubner, Wiesbaden 1997.
[17] L. Schiller, A. Naumann, Verfahrenstechnik, Verein Deutscher
Ingenieure, Berlin, 1935, 77, 318.
[18] B. E. Launder, D. B. Spalding, The numerical computation
of turbulent flows, Computer methods in applied mechanics
and engineering 1974, 3, 269.
[19] D. C. Wilcox, Turbulence modeling for CFD, DCW Industries
Inc., La Canda, California, USA 1993.
[20] S. Ghaniyari-Benis et al., Chem. Eng. Technol. 2009, 32 (1),
93.
[21] L. D. Landau, E. M. Lifshitz, Fluid Mechanics, 6 (Course of
Theoretical Physics), 2nd ed., Butterworth-Heinemann, Oxford 2000.
[22] M. Baerns, H. Hofmann, A. Renken, Chemische Reaktionstechnik, Thieme Verlag, Stuttgart 1999.
[23] O. Levenspiel, Chemical Reaction Engineering, Wiley & Sons,
New York 1972.
[24] E. Fitzer, W. Fritz, Technische Chemie, Springer Verlag, Berlin
1982.

http://www.cet-journal.com

You might also like