CFD Modelling of Mixing Effects on the Course of Parallel

Chemical Reactions Carried out in a Stirred Tank

By Jerzy Badyga and ukasz Makowski*
Effects of turbulent mixing on the course of two fast parallel chemical reactions (neutralization of sodium hydroxide and
hydrolysis of ethyl chloroacetate) carriedout ina semibatchstirredtankreactor are experimentally investigatedandnumerically
simulated. The flow pattern in the stirred tank is predicted using CFD and experimentally validated using Laser Doppler
Anemometry. Mixing effects are modelled using three CFDbased models. In the first and the second model the Beta probability
distribution and the spiked distribution are used respectively; in the third model concentration fluctuations are neglected.
1 Introduction
Mixing affects the course of homogeneous chemical
reactions when they are at least as fast as often attained
mixing rates. In this work, the authors are particularly
interested in the influence of mixing on the course of parallel
chemical reactions (test system based on neutralization of
sodium hydroxide (A) with hydrochloric acid (B) and
hydrolysis of ethyl chloroacetate (C).
NaOHHCl
k
1
NaCl H
2
O
NaOHCH
2
ClCOOC
2
H
5

k
2
CH
2
ClCOONa
C
2
H
5
OH
carried out in the semibatch stirred tank rector. Semibatch
reactors are often applied for production of fine chemicals,
and the problemof selectivity is then of particular importance
as the product purity determines its quality and price.
Experimental results obtained in the CSTR [1] will be used
as well for investigation of the effect of operation mode and
model validation. However, comparing to modelling of the
steady-state operation carried out in the CSTR, application of
CFDtomodel the semibatch process is more difficult, because
one needs to follow numerically the process during long
feeding times.
In this paper we consider three CFD based mixing models:
the first model simulates concentration fluctuations of
reactants using Beta distribution of the mixture fraction
the second model simulates fluctuations of reactant con-
centrations by using the spiked distribution related to
multizone modelling and applies the coalescence-redisper-
sion mechanism to simulate micromixing
the third model neglects concentration fluctuations of
reactants.
Hence, micromixing will be either neglected or simulated
using two micromixing models.
2 Hydrodynamics
The reactor studied in this work was a flat-bottom
cylindrical vessel made of glass, equipped with four baffles
and the Rushton type flat blade turbine. The height of the
liquidHwas equal tothe vessel diameter T = 0.3 m. Details of
reactor geometry are shown in Fig. 1. The vessel was placed
inside a square tank filled with water in order to minimize
refraction at the cylindrical surface during LDA measure-
ments, and to secure isothermal conditions in the system
(T = 293 K). Simulations of hydrodynamics were performed
using the CFD code FLUENT, version 6.0, using standard k-e
model and the multiple reference frame approach. The
numerical gridconsisted of 500,000 hexahedral computational
cells. The grid was created with Gambit 2.0. Convergence of
computations was regarded as satisfactory when the total
normalized residual was smaller than 10
6
. The grid density
was considered as acceptable when further increase of the
number of hexahedron volume elements close to the impeller,
baffles and walls did not increase the computed value of the
Chem. Eng. Technol. 2004, 27, No. 3 DOI: 10.1002/ceat.200401992 2004 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim 225
w
H
C
D
sh
T
B+C
D
/T=1/3
/D=0.1
=0.3m
sh
/T=1
/T=1/3
C
D
T
H
D
w/T=0.2
A
z
Figure 1. Schematic presentation of the stirred tank reactor.

[*] Prof. Dr. Jerzy Badyga (author to whom correspondence should be

addressed, e-mail: baldyga@ichip.pw.edu.pl), Dr. ukasz Makowski,
Faculty of Chemical and Process Engineering, Warsaw University of
Technology, ul. Warynskiego 1, 00645 Warszawa, Poland.
Full Paper
volume-weighted average rate of energy dissipation and the
related power number. In the range of parameters applied in
computations, the calculatedvalues of the power number were
not smaller than 90% of the measured value equal to 5.4. The
next step of constructing the numerical grid consisted in
introducing into the grid about 50,000 computational cells
necessary to simulate the flow outside and inside the feeding
pipes. To perform LDA measurements, the vessel was seeded
with the polyamide seeding particles of diameter equal to
5 lm, supplied by Dantec. Fig. 2 presents details of the
velocity distribution close to the feed pipe in the case of the
semibatch single feed stirred tank reactor. The agreement
between model predictions and measured values is satisfac-
tory. Results illustrating the flow pattern in the discharge
stream and the bulk are presented in Ref. [1].
3 Presentation and experimental validation of
micromixing models using CSTR experimental
data
The first method is based on using the k-e model to calculate
the flow field and the multiple-time-scale turbulent mixer
model [2,3] to simulate distributions of concentration var-
iances of the non-reacting tracer. To this end the mixture
fraction being the normalized concentration of the passive
tracer is used
1)
f =
c
0
A
c
A0
(1)
The turbulent mixer model enables calculation of the
distribution of the concentration variance r
2
S
, as well as
its inertial-convective r
2
1
, viscous-convective r
2
2
, and viscous-
diffusive r
2
3
components.
The distributions of f

and the concentration variance

components r
2
i
are modelled using the gradient diffusion
approximation:
@f
@t
u
j
@f
@x
j
=
@
@x
j
D
m
D
T
_ _
@f
@x
j
_ _
(2)
and
@r
2
i
@t
u
j
@r
2
i
@x
j
=
@
@x
j
D
m
D
T
_ _ @r
2
i
@x
j
_ _
R
Pi
R
Di
; i = 1; 2; 3 (3)
where R
Pi
and R
Di
represent the production and dissipation
terms:
r
2
1
is produced from gradients of f

:
R
P1
= 2u
/
j
f
/
@f
@x
j
= 2D
T
@f
@x
j
_ _
2
(4)
and dissipated due to the inertial-convective reduction of the
scale of inhomogeneity, producing r
2
2
R
D1
= R
P2
=
r
2
1
s
S
; s
S
=
1
R
k
e
(5)
r
2
2
, decays due to viscous-convective shrinking of compo-
nent slabs, producing r
2
3
R
D2
= R
P3
= Er
2
2
; E = 0:058
e
m
_ _1=2
(6)
r
2
3
, is dissipated due to molecular diffusion in deforming,
shrinking slabs
R
D3
= Gr
2
3
; G = E 0:303
17050
Sc
_ _
(7)
Once the spatial distributions of f

and r
2
S
are known, one
expresses the local value of the probability density of f using
the Beta probability distribution
U(f; x
*
; t) =
f
v1
(1f)
w1
_
1
0
y
v1
(1y)
w1
dy
(8)
with
v = f
f 1f
_ _
r
2
S
1
_ _
; w =
1f
_ _
f
_ _
v (9)
andcalculates the local values of the average reactionrate. For
example, the average reaction rate between the ester C and
the base A is given by
r
2
= k
2
c
A
c
B
= k
2
_
1
0
c
A
(f)c
C
(f)U(f)df (10)
226 2004 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim http://www.cet-journal.de Chem. Eng. Technol. 2004, 27, No. 3
0.8 1 1.2 1.4 1.6
-1.6
-1.2
-0.8
-0.4
0
0.4
z
tip
0.8 1 1.2 1.4 1.6
-0.08
-0.04
0
0.04
0.08
2r/D
2r/D
z
tip
2r/D
0.8 1 1.2 1.4 1.6
0
0.04
0.08
0.12
0.16
2r/D
z/T = 0.300
z/T = 0.366 z/T = 0.333
z/T = 0.266
0.8 1 1.2 1.4 1.6
-0.15
-0.1
-0.05
0
0.05
0.1
experimental
predicted (model k - )
Figure 2. Variation of the mean axial velocity close to the feed pipe;
N = 106 rpm, t
f
= 20 min.

1) List of symbols at the end of the paper

Full Paper
The local concentrations of reactants of parallel chemical
reactions are expressed as functions of the mixture fraction f
using the relation
f =
c
A
(c
B
c
C
)c
B0
c
C0
c
A0
c
B0
c
C0
(11)
together withthe methodof interpolationof c
C
(f) betweenthe
concentrations of reagents of the instantaneous and infinitely
slowreaction. Moreover, the neutralization reaction is treated
as instantaneous, soreactants AandBdonot co-exist. Inother
words a single reaction progress variable and a conditional
moment closure are used. This means that it is enough to
integrate the balance of C to obtain concentration distribu-
tions of all reactants
@c
C
@t
u
j
@c
C
@x
j
=
@
@x
j
D
m
D
T
_ _
@c
C
@x
j
_ _
r
2
(12)
The model was used to interpret experimental data in Ref.
[1], where a premixture of hydrochloric acid (B) and ethyl
chloroacetate (C) was fed through one feeding port, and
solution of sodium hydroxide (A) was fed through the second
feeding port to the CSTR. The final selectivity was expressed
by
X
S
=
c
C0
c
C
c
A0
(13)
where c
i0
are average feed concentrations. Experiments are
carried out for c
A0
= c
B0
= c
C0
. Predictions of Model I are
compared with experimental data in Fig. 3.
Model II is a five-zone finite-mode PDF model that can be
represented by five spikes, or more exactly it corresponds to a
presumed joint scalar PDF of the form:
U w; x; t ( ) =

N
n=1
p
n
x
*
; t
_ _

m
a=1
d w
a
c
n ( )
a
x
*
; t
_ _ _ _
(14)
where the product is over all chemical species and the sum is
over all environments. Such a general approach to the multi-
environment micromixing models was presented by Fox [4]
and applied afterwards to reactive mixing and precipitation
using three or four environments. To apply the micromixing
model for interpretation of reactive mixing, one needs to
define a mechanism for simulation of mass exchange between
the environments, and micromixing can then be interpreted as
a probability flux [4]. In the following, we use for
simulations the coalescenceredispersion mechanism in the
Eulerian frame. Due to this mechanism, at any position in the
system each environment i can interact with all other
environments present at the same point of the system.
When environment i interacts with environment j than each of
them loses cp
i
p
j
of probability p
i
and p
j
per second with the
rate dependent on the micromixing parameter c that
represents frequency of coalescence-redispersion events.
The fluids are mixed and either transported to the environ-
ment (i + j)/2 if (i + j)/2 is an integer number, or splitted into
two equal parts and transferred into the two closest
environments. These two closest environments include
environments i and j themselves, but then there is no variation
of p
i
and p
j
but pure exchange of species. Environments
number 1 and N only lose their probability and can be
thus regarded as completely segregated. Similar balances for
the volume weighted concentrations p
n
c
n ( )
a
of each chemical
species with source terms r
a
_
c
*
n ( )
_
p
n
are solved together with
balances for the probability p
n
. In aqueous solutions
molecular diffusivity of reacting species does not significantly
affect the reaction progress (Sc < 3000 is here a condition [3]
for neglecting effects of molecular diffusivity), so only the
inertial-convective and viscous-convective mechanisms of
micromixing are considered. In the case of Model I relaxation
of the concentration spectrum is included in the model; in the
case of Model II we apply two limiting solutions for the
micromixing rate presented by Heinz and Roekaerts [5]:
case a) is for homogeneous decaying turbulences and yields
c = E
1
s
S
_ _1
, case b) is for local equilibrium turbulence
and results in c = min E; s
1
S
_ _
= E E s
1
S

_
+
E s
1
S
_ _
)=2. Results of computation of effects of mixing on
the course of parallel chemical reactions are compared with
predictions of both models in Fig. 3 and 4. Experimental
results are taken from Ref. [1]. Fig. 3 shows that both models
well predict trends of influenceof the meanresidence timeand
mixing rate on the reaction selectivity defined by equation
(13). Model IIa overestimates experimental data. There can
be several reasons for this effect. The first reason can be a too
Chem. Eng. Technol. 2004, 27, No. 3 http://www.cet-journal.de 2004 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim 227
0
0.04
0.08
0.12
0.16
0.2
experimental
106 rpm
214 rpm
Model I
106 rpm
214 rpm
Model II, case a
106 rpm
214 rpm

Figure 3. Influence of mean residence time s and stirrer speed N on the

selectivity X
S
observed in the continuous-flow stirred tank reactor;
c
B0
= c
C0
= 20 mol m
3
; c
A0
= 1000 mol m
3
.
Full Paper
small micromixing frequency c characteristic for decaying
turbulence. Fig. 4 shows that this effect is small and thus does
not explain this discrepancy. Moreover, the difference
between cases IIa and IIb decreases with increasing rate of
energy dissipation, when effects of viscosity become negli-
gible. The second reason can be an effect of the spurious scalar
dissipationresulting fromrepresenting the joint scalar PDFby
a finite number of environments [4]. Numerical results for a
tubular reactor [4] suggest that this effect is negligible already
for four environments. However, this conclusion is most
probably only valid for a reactant concentration ratio close to
unity. Hence, the third reason can be related to a too small
number of environments in the case when the concentration
andvolume ratios of reactants are very high. Numerical results
for the tubular reactor [6] show that for reactant volume and
concentration ratios equal to 10, the number of environments
that yield the same results as the Beta distributions should be
equal to 30 whereas for the volume and concentration ratios
equal to 100 this number increases to roughly 300. A proper
choice of number of environments requires further investiga-
tions. First results showthat application of a too small number
of environments can significantly affect the results of
computations. Inthe following parts of this paper micromixing
is therefore interpreted using Beta distribution.
4 Experimental
The parallel chemical reactions employed to investigate
the mixing effects in the semibatch process were the
neutralization of sodium hydroxide (A) by hydrochloric acid
(B) and the alkaline hydrolysis of ethyl chloroacetate (C).
The experimental procedure was similar to that described by
Bourne and Yu [7]. Experiments were carried out at
T = 293 K with k
2
= 23 dm
3
mol s
1
. The stirred tank con-
taining initially V
BC,0
= 0.02079 m
3
of the premixture of HCL
(B) and CH
2
ClCOOC
2
H
5
(C), each at a concentration of
40 mol m
3
, was operated in the semibatch mode, whilst
the relative volume of NaOHsolution added (c
A0
= 2000 mol
m
3
) was 2% of the tank capacity (V
A0
= 4.16 10
4
m
3
). The
feed point was located in the impeller discharge stream at the
impeller disc level, at a distance of 0.09 m from the axis
(Fig. 1). The feed tube (diameter equal to 0.001 m) was used
to exclude backmixing. Experiments were carried out for the
feed time t
f
equal to 8, 10, 15 and 20 min, respectively.
Measurements werecarriedout for therotational stirrer speed
in the range of 106 rpm N 214 rpm. In order to calculate
the final selectivity of parallel reactions, eq. (13), the
concentrationof ethyl chloroacetate was chromatographically
(HPLC) measuredbefore andafter experiments. c
C
ineq. (13)
represents the final ester concentration in the tank after the
process, whilst c
A0
= V
A0
c
A0
= V
A0
V
BC;0
_ _
and c
C0
=V
BC;0
C
C0
= V
A0
V
BC;0
_ _
. It is important to note that X
S
represents a relative amount of the substrate Aconverted into
the byproduct S (ethanol).
5 Modelling of micromixing effects in a semibatch
stirred tank reactor
Application of CFD to model the semibatch process is
difficult because it is necessary to follow the process during
long feeding times. This problem is solved by discretizing the
feed into r portions of equal volume. For each portion the
differential balance equations for mass, momentum and
species are solved. Integration is carried out until the quasi-
steady-state solution is obtained; then the next portion of the
fresh feed is added and the equations are again integrated till
the new quasi-steady-state in the system is reached. For each
computation step the process is thus interpreted as the steady-
state continuous flow system, however, the outflowing
mixture is collected and added back to the system before
introducing the next portion. This method of treating the
semibatch process seems to be similar to the one used by
Patterson and Randick [8].
For each addition the procedure consist of the following
steps:
the k-e model is first appliedtocalculate distributions of the
average velocity u
j
, the average kinetic energy k and the
average rate of the energy dissipation e;
the mixture fraction distribution

f
/
is calculated;
The mixture fraction, is interpreted in the case of the
semibatch reactor as a mass fraction of the fresh feed that is
diluted by the older bulk fluid. To determine distribution of

f
/
two methods are used. The first one is based on CFD
simulations of the conversion of chemical reactions of order
zero for various initial reactant concentrations c
Y
. The
reactant of the zero-order reaction is fed into the system by
the A-inlet. As the rateof consumptionof the reactant of the
zero-order chemical reaction is independent of the feed
concentration, the dilution effect determines

f
/
.
228 2004 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim http://www.cet-journal.de Chem. Eng. Technol. 2004, 27, No. 3
experimental
Model I
Model II, case a
Model II, case b
[m
2
s
-3
]
Figure 4. Influence of the energy dissipationratee onthe selectivity X
S
observed
inthe continuous-flowstirredtankreactor; s = 20 min; c
B0
= c
C0
= 20 mol m
3
;
c
A0
= 1000 mol m
3
.
Full Paper
f
/
= c
Y2
c
Y1
_ _
= c
Y02
c
Y01
_ _
(15)
The second method is simpler but less exact. It is based on
renormalization of the mixture fraction, which in the case of
semibatch feeding reads:
f
/
= f f
b
_ _
= 1 f
b
_ _
; f
b
= Q
A
t= V
B0
Q
A
t
_ _
(16)
Both methods give almost identical results, so the second
one is recommended as easier to use.
distributions of variances r
2
1
; r
2
2
; r
2
3
are calculated from eqs
(3)(7);
a closure based on Beta distribution is used to compute
concentration distributions and the final selectivity value.
For comparison we also have performed computations
using the model that neglects concentration fluctuations of
reactants and is based on local average reactant concentra-
tions with k
2
c
A
c
C
replaced by k
2
c
A
c
C
. In the following,
this model is called Model III. Fig. 5 and 6 show comparisons
of experimental data with predictions of Models I and III. It
can be clearly seen fromexperimental data that an increase of
both the stirrer speed and the feed time decreases production
of the byproduct and thus decreases X
S
. Both models predict
these tendencies, however, neglecting effects of concentration
fluctuations (Model III) results in much smaller X
S
values
than observed in experiments. Hence, Model III is not
recommended. Results of Model I (r = 10) are in good
agreement with experimental data. It has been checked
whether sensitivity of the selectivity to the feed discretization
is small; increasing r from 10 to 30 results in increase of X
S
by
not more than 3.5%. Fig. 7 compares experimental results and
model predictions obtained in the semibatch reactor with
those for the CSTR. One cansee that the effect of the changing
operationmode is well predictedby the model. Fig. 8 shows an
example of computer visualization of reaction plumes
computed with Model I. The reaction zone presented in the
figure is equivalent to the zone of basic solution (c
A
>10
7
mol/
dm
3
). Results show a significant increase of the reaction zone
with time. The reaction zone is localized very close to the
feeding point at early stages of the process and spreads out
towards the end of the process. Hence application of the test
reaction systemto determine local values of the rate of energy
Chem. Eng. Technol. 2004, 27, No. 3 http://www.cet-journal.de 2004 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim 229
0 0.04 0.08 0.12
0
0.02
0.04
0.06
0.08
0.1
0.12
t
f
=15 min
experimental
Model I
Model III
S
0 0.04 0.08 0.12
0
0.02
0.04
0.06
0.08
0.1
0.12
t
f
=20 min
experimental
Model I
Model III
S
[m
2
s
-3
]
[m
2
s
-3
]
Figure 5. Influence of the energy dissipationratee onthe selectivity X
S
observed
in the semibatch stirred tank reactor for different feed time values t
f
;
c
B0
= c
C0
= 40 mol m
3
; c
A0
= 2000 mol m
3
.
Figure 6. Influence of the feed time t
f
on the selectivity X
S
observed in the
semibatch stirred tank reactor; N = 214 rpm; c
B0
= c
C0
= 40 mol m
3
; c
A0
=
2000 mol m
3
.
0 0.04 0.08 0.12
0
0.02
0.04
0.06
0.08
0.1
Experimental
= 20 - CSTR
t
f
= 20 - SBR
Predicted - Model I
= 20 - CSTR
t
f
= 20 - SBR
X
s
[m
2
s
-3
]
Figure 7. Influence of the energy dissipation e on the selectivity X
S
observed in
the continuous-flowstirred tank reactor and the semibatch stirred tank reactor;
s = t
f
= 20 min; c
B0
= c
C0
= 20 mol m
3
; c
A0
= 1000 mol m
3
.
Full Paper
dissipation from the reaction product distribution X
S
yields
only a crude estimate of e.
6 Discussion and conclusions
I. TheCFDbasedmodels well predict experimental data for
parallel chemical reactions carried out in stirred tank
reactors provided that modelling of micromixing and
relaxation of the concentration spectrum are included
(models I and IIb in this paper). This conclusion seems
not to agree with observations of Brucato et al. [9] that
effects of micromixing phenomena in stirred tanks are
negligible. However, there is a simple explanation for this
discrepancy: for fast injection applied by Brucato et al. [9]
(batch system) effects of micromixing are small, whereas
for slow, controlled addition they determine product
distribution. This phenomenon can be explained using
time constant analysis, see [3], pages 606608.
II. The finite-mode model requires using a large number of
environments for high values of the concentration ratio;
the performance of the model may be improved by
correcting for the effects of spurious scalar dissipation.
III. The presented approach to modelling has been validated
using experimental data from the stirred tank reactor of
diameter T = 0.3 m. However, completeness of the
model enables us to predict consequences of scale-up to
the large industrial scale. For example when using a
common scale-up criterion based on constant power per
unit mass during scale-up, one gets the same time
constants for the viscous-convective andviscous-diffusive
mixing in both systems, but the rate of turbulent diffusion
increases with increasing the scale, D
T
~ k
2
=e ~ e
1=3
L
4=3
,
whilst the rate of inertial-convective mixing decreases,
s
S
~ k=e ~ L
2=3
=e
1=3
(time constant increases). Conse-
quently, the decay of concentration fluctuations is slower
in large-scale systems and selectivity X
S
is higher.
Moreover at not too high power input the controlling
mechanism can change during scale-up from the viscous-
convective to inertial-convective. Hence, scale-up based
on e = const. is not recommended.
Received: December 1, 2003 [CET1992]
References
[1] J. Badyga, M. Henczka, . Makowski, Trans. IChemE 2001, 79(A), 895.
[2] J. Badyga, Chem. Eng. Sci. 1989, 44, 1175.
[3] J. Badyga, J. R. Bourne, Turbulent Mixing and Chemical Reactions,
Wiley, Chichester, 1999.
[4] R. O. Fox, Chem. Eng. Process. 1998, 37, 521.
[5] S. Heinz, D. Roekaerts, Chem. Eng. Sci., 2001, 56, 3197.
[6] J. Badyga, M. Jasinska, Inz. i Ap. Chem. 2002, 41, 24.
[7] J. R. Bourne, S. Yu, Ind. Eng. Chem. Res. 1989, 33, 41.
[8] G. K. Patterson, J. Randick, Proceedings of the Tenth European
Conference on Mixing, Elsevier, Delft, 1989, 53.
[9] A. Brucato, M. Ciofalo, F. Grisa, R. Tocco, Chem. Eng. Sci. 1989, 55,
291.
Symbols used
c
i
[mol dm
3
] concentration of substance i
c

i
[mol dm
3
] concentration of reactant of
instantaneous reaction
c
0
i
[mol dm
3
] concentration of nonreacting tracer
D
m
[m
2
s
1
] molecular diffusivity
D
T
[m
2
s
1
] turbulent diffusivity
E [s
1
] engulfment parameter
F [] dimensionless concentration of
nonreacting tracer
G [s
1
] molecular diffusion parameter
K [m
2
s
2
] kinetic energy of turbulence
k
1
, k
2
[dm
3
mol
1
s
1
] second order rate constant
L [m] integral scale of turbulence
N [rpm] stirrer speed
p
i
[] probability
t
f
[s] feeding time
u [m s
1
] local average velocity
u
tip
[m s
1
] tip velocity
X
S
[] product distribution
V
i
[m
3
] volume
e [m
2
s
3
] local average value of the rate of energy
dissipation
e [m
2
s
3
] average rate of energy dissipationinthe
tank
230 2004 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim http://www.cet-journal.de Chem. Eng. Technol. 2004, 27, No. 3
Figure 8. Visualization of the reaction plume observed in the semibatch stirred
tank reactor: N = 174 rpm, t
f
= 8 min, c
B0
= c
C0
= 40 mol/m
3
, c
A0
= 2000 mol/
m
3
; a) after 48 s, b) after 240 s, c) after 480 s.
Full Paper
c [s
1
] micromixing parameter
m [m s
2
] kinematic viscosity
r [] feed discretization
r
2
S
, r
2
i
[] dimensionless concentration variances
s [s] mean residence time
s
s
[s] time scale for the inertial-convective
mixing
Abbreviations
CFD computational fluid dynamics
CSTR continuous-flow stirred tank reactor
LDA laser Doppler anemometry
PDF probability density function
SBR semibatch stirred tank reactor
______________________
Chem. Eng. Technol. 2004, 27, No. 3 http://www.cet-journal.de 2004 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim 231
Full Paper