Chemical Engineering Science

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 7

ARTICLE IN PRESS

Chemical Engineering Science 65 (2010) 21832189

Contents lists available at ScienceDirect

Chemical Engineering Science


journal homepage: www.elsevier.com/locate/ces

Interfacial energy estimation in a precipitation reaction using the atness


based control of the moment trajectories
b

Vassil Vassilev a, Michael Groschel


, Hans-Joachim Schmid c, Wolfgang Peukert a,, Gunter
Leugering b
a

Institute of Particle Technology, Cauerstrasse 4, Germany


Institute of Applied Mathematics, Martensstrasse 3, Friedrich-Alexander-University, Erlangen-Nuremberg, Germany
c
Mechanical Process Engineering and Environmental Technology, Pohlweg 55 University Paderborn, Paderborn, Germany
b

a r t i c l e in fo

abstract

Article history:
Received 25 March 2009
Received in revised form
9 December 2009
Accepted 11 December 2009
Available online 24 December 2009

In this paper, we consider a model for precipitation experiments based on the population balance
equation. The study revealed a high sensitivity of the system with respect to the modeling of intrinsic
parameters, motivating a comprehensive validation of the estimates. In the forward simulation the
impact of the inuencing parameters including surface energy, nucleus size and distribution is
investigated. Subsequently we construct a simplied model of the precipitation process in such a way
that it is orbitally at in terms of control theory, which enables the inverse calculation of the
parameters. The numerical results of the inverse simulation for the interfacial energy have been
compared to a physical model. The possibility of solving the inverse problem provides a promising way
of estimating hardly measurable quantities for more complex molecules.
& 2009 Elsevier Ltd. All rights reserved.

Keywords:
Precipitation
Population balance
Parameter identication
Process control
Flat systems
Interfacial energy

1. Introduction
Precipitation is a widespread process in industry for the
production of particles of varying size distributions, morphologies
or other determining factors. In order to improve and predict the
product properties, as well as to apply precipitation processes in
the production of nanoparticles, a further and deeper understanding of the involved steps is required. Precipitation is a
promising method for the economical production of huge
amounts of nanoparticles as it is fast and operable at ambient
temperatures. The evolution of the particle size distributions
(PSD) during the precipitation process as well as the inuence of
mixing and supersaturation have been studied experimentally
and numerically, for instance by Marchisio et al. (2002), Judat and
Kind (2004), Schwarzer and Peukert (2004a), using barium sulfate
as a test system.
Population balance equations (PBE) represent a state-of-theart model description of polydisperse particulate processes. Most
of the models need difcult-to-access data on the material
properties of the particles, e.g. surface charge and surface energy.
Depending on the suspension composition an exact estimation of
their contribution to the PSD becomes very complicated. The
interfacial energy gPF occurring in the nucleation rate is such a
 Correspondence to: Lehrstuhl fur
Feststoff und Grenzachenverfahrens
Erlangen-Nurnberg,

technik, Friedrich-Alexander-Universitat
Cauerstrasse4,
91058 Erlangen, Germany, Fax: + 49 9131 85 29402.
E-mail address: W.Peukert@lfg.uni-erlangen.de (W. Peukert).

0009-2509/$ - see front matter & 2009 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ces.2009.12.014

crucial variable, as the nucleation rate and thus the resulting PSD
are very sensitive to this parameter.
For the computation of the surface potential a frequently used
approach is a calculation based on measurements of the
zpotential. Another approach rests upon describing the interfacial
energy as a sum of the chemical and electrostatical contributions
calculated using the Grahame equation (Israelachvili, 1991).
According to the outlined sensitivity of the system, an accurate
estimation of the interfacial energy is absolutely indispensable.
Methods have to be developed to validate the different representations of this important parameter, depending on the
material properties of the particles, the composition of the
solution and the prevailing growth mechanism.
The challenging measurement of these determining factors
raises the question how to obtain estimates or validate the
parameter modeling. For the batch crystallization process an
inverse problem approach has been suggested by Vollmer and
Raisch (2003), yielding a temperature prole needed in order to
achieve a certain given crystal size distribution (CSD). They show
that the batch crystallizer model fulls the conditions of an
orbitally at system after an appropriate time scaling. Introduced
by Fliess et al. (1992, 1995) this approach was later used in
trajectory planning and feed-forward control design (for a
comprehensive review see Martin et al., 2003). In this contribution the notion of orbital atness will be used to determine the
interfacial energy resulting in a certain given PSD. This way the
interfacial energy ascertained through specic models can be
veried.

ARTICLE IN PRESS
V. Vassilev et al. / Chemical Engineering Science 65 (2010) 21832189

In the following sections the forward simulation is presented.


Additionally sensitivity tests for the decisive parameters are
performed. Next the method of moments and the orbital atness
are presented and veried. At last the numerical results from the
forward and inverse simulation are evaluated.

2. Theoretical background
The general population balance equation (1) as applied to
engineering systems is given below in its differential form,
possibly depending on the spatial position ~
r:


@nx; ~
r
@x
r~
 nx; ~
r Bagg nx; ~
u r  nx; ~
r r
r Dagg nx; ~
r :
@t
@t
1
Using this formulation the evolution of the number density
distribution undergoing a convection ~
u r is described. In the ideal
plug-ow reactor model a complete radial mixing can be
assumed. Therefore it is sufcient to observe only the 1D spatial
resolution rz along the z-direction which represents the mixer
axis.
In the following the residence time t is introduced in order to
transform the one dimensional spatial evolution of the PSD into a
system evolving with respect to the time. In this case the space
convective term ~
u vanishes and a reformulation of the general PBE
reduces for an ideally mixed batch process to
@nx; t
@GS; x; t  nx; t
Bhom S; gPF ; t  f x
Bagg nx; tDagg nx; t;
@t
@x

2
assuming that the total volume of the system remains constant.
For a detailed justication of the model derivation in the case of
precipitation experiments see Schwarzer and Peukert (2004b).
Here, nx; t represents the number density distribution of
particles at time t, f x the density distribution of nuclei sizes,
Bhom the homogeneous nucleation rate, Gx the particle growth
rate and Bagg n; x resp. Gagg n; x the source and sink terms due to
agglomeration.
Usually, the critical cluster size for stable nuclei is approximated by
xc

2  gPF  Vm
:
kB T  nlnS

Due to numerical reasons in simulations a narrow lognormal size


distribution around this cluster size is assumed. Various test runs
showed that the width of this initial distribution does not
inuence the nal PSD.
The homogeneous nucleation rate Bhom can be calculated based
on the classical nucleation theory. Details on the derivation can be
found for example in Mersmann et al. (2000).
r
p
gPF
 Vm
Bhom S; gPF ; t 1:5  D KSP  S  NA 7=3
kB T
!


16p
gPF 3 Vm2
:
4

exp 

3
kB T
nlnS2
The interfacial energy gPF has a strong inuence on the nucleation
rate Bhom which results in a high sensitivity of the precipitation
process on the values of the interfacial energy (Fig. 1). Therefore a
change of the interfacial energy leads to a signicant change of
the resulting PSD.
In general, particle growth is a complex process including
several steps. Predominantly the transport of ions to and the
subsequent integration into the surface determines the speed of
the growth process. Based on several assumptions like equal

1020
nucleation rate Bhom [1/m3s]

2184

1016
1012
108

PF = 0.10 J/m

PF = 0.12 J/m

104

PF = 0.14 J/m

1
10

100
supersaturation [-]

1000

Fig. 1. Homogeneous nucleation rate as a function of supersaturation and


interfacial energy.

transport rates and an instantaneous change of the equilibrium


surface charge (see Schwarzer and Peukert, 2004b) the growth
rate G for spherical particles at high supersaturations can be
expressed by
p
2ShDM KSP S1

;
5
GS; x; t
rP
x
where Sh is the Sherwood number and D the constant diffusion
coefcient.
Applying the denition of the Sherwood number Sh kD x=D,
the mass transfer coefcient kD is the last remaining parameter
depending on the particle size.
In order to meet further needed requirements, the diffusive
transfer coefcient is henceforth assumed to be constant. Thus the
particle growth rate G turns into
p
2kD M KSP
 S1:
6
GS; t

rP

Concerning aggregation and agglomeration, the impact due to the


Brownian as well as shear-induced mechanisms was found to
increase with supersaturation as a result of increasing particle
numbers. Under the simulated conditions presented in this paper
these effects were found to have negligible inuence due to
successful electrostatic stabilization (Schwarzer et al., 2006).
Both nucleation and growth term depend on the current
composition of the system through interfacial energy and supersaturation values. Based on the concept of a Gibbs dividing
surface and assuming constant temperature as well as equilibrium throughout the three-phase system (solid, interface,
liquid), the Gibbs adsorption isotherm can be derived analogously
to the GibbsDuhem equation (Hunter, 1986; Sugimoto, 1996,
1999). Using the Grahame equation (Eq. 5.28, taken from
Israelachvili, 1991) the following expression for the interfacial
energy can be derived):
Z cs
X
gPF g0 R  T  Hi  ai 
sc dc:
7
i

This equation permits to calculate the interfacial energy gPF of a


particle-uid interface at equilibrium. The rst term on the righthand side is a material-specic contribution to the interfacial
energy, the second term accounts for the chemical contribution
due to sorption on the interface and the last term stands for the
electrostatic contribution arising for charged species. It is
important to emphasize that the summation in the Grahame
equation includes all ion species in the electrolyte solution. Using

ARTICLE IN PRESS
V. Vassilev et al. / Chemical Engineering Science 65 (2010) 21832189

this equation the value for the interfacial energy can be calculated
for every time step using the corresponding concentrations.
For the investigation of the precipitation reaction barium
sulfate BaSO4 was chosen as the product is hardly soluble and
thus the equilibrium is shifted to the right side:
Ba2 SO2
4 -BaSO4 k

Mersmann et al. (1994) showed that in a stirred tank reactor


nearly perfect mixing can be assumed for the case when the
discontinuous phase reactant is added near to the stirrer at
sufcient stirring speed. For the case of steep mixing gradients the
supersaturation almost reaches the value of the supersaturation
of the complete mixing case and results in a very similar PSD.
With decreasing steepness of the mixing gradient the maximally
reached supersaturation does not attain that high values and
results in particles with larger diameters and wider PSD.
The interfacial energy is a quantity which is difcult to
measure but has a decisive impact on the nucleation term. In
the considered situation transport-limited growth is predominant
such that the growth rate is independent of the surface energy.
The change of the relation between the nucleation and growth
terms inuences the amount of formed particles. The value of the
interfacial energy is assumed to be constant throughout the
simulation. Fig. 3 shows the inuence of the interfacial energy for
an educt composition for which the physical model of the
interfacial energy predicts the value gPF 0:126 J=m2 . The shown
variation using a constant mixing gradient inuences strongly the
process as can be seen on the evolution of the supersaturation.
The simulation parameters are shown above in Fig. 2 with a
mixing time tmix 47 ms. As the value of the interfacial energy
decreases the nucleation terms gain inuence. This leads to
increased nucleation and thus a fast reduction of the
supersaturation leading to a relatively narrow PSD. The 720%
change of the value of the interfacial energy shifts the mean
particle size from 35 to 220 nm and inuences also the width of
the PSD.
Precisely the interfacial energy is considered to be a function of
the chemical composition of the system. The increase of the
interfacial energy between the start and end value is, however,
only about 1% of the initial value. The comparison between two
simulations, with constant and varying values (see Fig. 4), shows
only a very slight difference of the resulting PSD. Therefore the
use of a constant value is sufcient for simulations without
loosing accuracy.
As shown in the Theoretical Section the size of the nuclei is
modeled by (3) depending on the interfacial energy as well as on
the supersaturation value. Calculations with different nuclei sizes
are conducted to investigate the possibility of eliminating this
dependency. In Fig. 5 the results for four different calculations are
presented. The rst curve represents the calculated nucleus size
from the Kelvin equation (3) at every time step. The constant
values used are the minimal and average value for the nuclei size,
and the so-called crystallographic density which is the molecule
size when the BaSO4 is considered as ideal solid matter.

Being the thermodynamical driving force of the reaction the


supersaturation is dened as
s
aBa2 cBa2  aSO2 cSO2
4
4
9
S
KSP
The activity coefcients were obtained from the semi empirical
method proposed by Bromley (1973).
In this context it has to be considered that the incomplete
dissociation of the sulfuric acid in its second stage
2

(HSO
4 $H SO4 ) is described by the dissociation constant
Kdiss as dened below:
Kdiss

aH cH  aSO2 cSO2
4

10

aHSO4 cHSO4

3. Forward simulations
For the numerical calculations we used PARSIVAL. PARSIVAL is
a commercial software package from CiT-Computing in Technology GmbH and represents a common tool, which solves efciently
the entire integro-differential equation. The implemented numerical algorithm is described in detail in Wulkow et al. (2001).
PARSIVAL has already been used for the forward simulation
(Schwarzer and Peukert, 2004a; Schwarzer et al., 2006) in terms
of calculating the PSD and its evolution. The reaction is
implemented on the basis of the equations and dependencies
shown in Section 2.
Mixing represents a key issue in chemical reactions. As the
procedure of mixing is of low interest in our current investigation
the mixing is reduced to a mixing gradient which instantaneously
mixes the components perfectly to a certain specied extent.
Prole 1 in Fig. 2corresponds to instantaneous complete mixing.
The other prole represent mixing with a linear gradient which
starts at time 0 and no mixing and reaches complete mixing at a
dened time between 0.1 and 47 ms. The simulations were
conducted with barium concentration cBa2 0:25 mol=l and
sulfate concentration cSO2 0:165 mol=l. The SO24 ions are in
4
the continuous phase, the volume used in the simulation was 1 m3
and the interfacial energy was calculated for this composition to be
gPF 0:126 J=m2

600
400

density distr. [1/nm]

supersaturation [-]

0.05

profile 1
profile 2
profile 3
profile 4
profile 5
profile 6

800

200
0
10-4

2185

profile 1
profile 2
profile 3
profile 4
profile 5
profile 6

0.04
0.03
0.02
0.01
0

10-3

10-2
time [s]

10-1

50

100
150
200
particle size [nm]

250

Fig. 2. Variation of the mixing gradient. (a) Different mixing proles; (b) resulting PSD (normalized number density distribution).

ARTICLE IN PRESS
2186

V. Vassilev et al. / Chemical Engineering Science 65 (2010) 21832189

600
400

density distr. [1/nm]

800
supersaturation [-]

0.04

PF = 0.10
PF = 0.11
PF = 0.12
PF = 0.125
PF = 0.13
PF = 0.135
PF = 0.14

200
0
10-3

PF = 0.10
PF = 0.11

0.03

PF = 0.12

PF = 0.125
PF = 0.13

0.02

PF = 0.135
PF = 0.14

0.01
0

10-2

10-1

100

time [s]

200 300 400


particle size [nm]

500

600

Fig. 3. Inuence of varying of interfacial energy. (a) Supersaturation prole; (b) resulting PSD (normalized number density distribution).

0.04
PF

density distr. [1/nm]

interfacial energy [J/m2]

0.124

PF = const

0.1235

0.123

PF
PF = const

0.03
0.02
0.01
0

0.1225
0

0.01 0.02 0.03 0.04 0.05 0.06


time [s]

100
200
particle size [nm]

300

Fig. 4. Small variation of the interfacial energy. (a) Interfacial energy prole; (b) resulting PSD (normalized number density distribution).

xc
xc = 0.4 nm
xc = 0.7 nm
xc = 1.2 nm

xc [nm]

0.004
0.003
0.002
0.001
0

0.04
density distr. [1/nm]

0.005

xc
xc = 0.4 nm
xc = 0.7 nm
xc = 1.2 nm

0.03
0.02
0.01
0

0.01 0.02 0.03 0.04 0.05 0.06


time [s]

100

200

300

time [s]

Fig. 5. Variation of the nuclei size. (a) Size of the nuclei; (b) resulting PSD (normalized number density distribution).

The resulting PSDs for the different nuclei size coincide to a


great extend showing that sophisticated modeling is dispensable.

4. Method of moments for the precipitation reaction


The method of moments is a method which reduces the
description of the PSD to the calculation of a number of moments.
The k th moment of a number density distribution is dened as
follows:
Z 1
mk t
xk nx; t dx;
11
0

where n is the unnormalized number density distribution.


Integrating over the entire population density function, i.e. the
zeroth moment m0 t, gives per denition the number of particles.

Furthermore, the third moment m3 t is proportional to the total


volume of particles per unit suspension volume.
To apply the method of moments to the model derived in
Section 2, we use the fact that the number of particles increases
only through nucleation and that there is a maximum particle size
limiting the distribution density function. Thus, under the present
experimental conditions the PBE model (2) can be written by the
following coupled moment equations:
dm0 t
Bhom S; gPF ; t;
dt

12

dmk t
kGS; tmk1 t:
dt

13

Eq. (12) reects the assumption that the number of particles only
changes through nucleation. The other moments depend on each
other through relation (13), which can be checked by partial

ARTICLE IN PRESS
V. Vassilev et al. / Chemical Engineering Science 65 (2010) 21832189

integration. Collecting now all facts and premises the full model
turns out to be
@nx; t
@GS; tnx; t

;
@t
@x
n0; t

Bhom S; gPF ; t
;
GS; t

nx; 0 0:

14

15
16

Using the moment technique the characteristic properties of the


PSD like mean particle size, deviation and skewness can be
calculated. The method is connected with a low computational
effort, but on the other hand the continuous development of the
PSD is not completely obtained.

The ability of simulating the progress of precipitation reactions


has led to new insights into the prevailing system dynamics.
Naturally, the next step will consist in the attempt to inuence the
process in a favorable manner. In this context control theory
provides a methodical approach to invert the system dynamics to
compute the inputs required to perform a specic task. This
inversion may involve nding appropriate inputs or parameters to
trigger a control system from one state to another or may involve
nding inputs to follow a desired trajectory for some or all of the
state variables of the system. Desired inputs could be for instance,
temperature proles, feed concentrations or the intensity of mixing.
The most common structure to exploit is a linear dependence
of one output of the system with respect to a variable input
parameter. As the systems that we seek to control become more
complex, the use of a linear structure alone is often not sufcient
to solve the control problems that are arising in applications. This
is especially true in the case when nonlinear dynamics dominate.
In this paper we focus on a specic class of systems, called
differentially at systems, for which the structure of the
trajectories of the nonlinear dynamics can be completely
characterized, i.e. two distinct trajectories may not intersect.
Applying this in the present context means that if two particles
are formed at different points in time in the process, their
resulting nal size must stay distinct.
A starting point for introducing the notion of atness is the
differential equation
y0 y0 A Rn ;

17

with the right-hand side depending on a parameter u from a set


U  Rm . The set U is called the set of control parameters and
contains all admissible inputs resp. strategies affecting the system
states yt coupled through the dynamic system.
Eq. (17) is called differentially at, if there exists an observation
or output wt A Rm being uniquely dened by a function h of the
system states in combination with the input and its rst r
derivatives:
n

m r 1

-R
h : R  R
s:t:
00
w hy; u; u0 ; u ; . . . ; ur :

18

On the other hand the system states and inputs have to be


explicitly characterized by the output and nitely many of its
time derivatives:
00

y fw; w0 ; w ; . . . ; wr1 ;
00

u cw; w0 ; w ; . . . ; wr :

In this case w is called a at output and the complete system can


be retrieved exploiting only the observation function (18).
If a sufciently smooth trajectory of the at output is given, the
evolution of the correlated moments can be computed without
solving a differential equation (Vollmer and Raisch, 2003). As
outlined in Moya et al. (2001) or Guay (1999) the notion of
differential atness can be applied to a larger class of ordinary
differential equations after an appropriate time-scaling. If the
transformed system offers the structure of atness, the original
system is called orbitally at. The used scaling function is
therefore subject to certain conditions to ensure invertibility, in
order to subsequently recover the desired input in real time.
A new time variable t is introduced by
dt
jyt; ut;
dt

tt0 t0

21

given a positive and bounded scaling function j jyt; ut.


The dynamic system (17) turns into

5. Notion of orbital atness

y0 t f yt; ut;

2187

19
20

y0 t f yt; utjyt; ut:

22

6. Model verication for orbital atness


In Sections 2 and 4 we have derived a population balance
equation determining the evolution of the number density
distribution:
@nx; t
@GS; tnx; t

;
@t
@x
n0; t

Bhom S; gPF ; t
;
GS; t

nx; 0 0:

23

24
25

Due to the conservation of mass, an additional equation can be


derived in order to describe the concentration at time t by the
third moment m3 t:
ct c0 

rP kv
M

m3 t:

26

This equation is only valid under the assumption that the volume
of the uid remains constant during the precipitation. The present
process characteristics legitimate the assumption, as an inorganic
substance with ionic bonds is precipitated and the particle
concentration was under 3.0 wt%.
Thus, the supersaturation S which depends on the concentration at time t is also related to the third moment m3 t.
Conclusively the birth and growth rates are solely characterized through m3 t and gPF t. Hence the differential equations for
the rst four moments characterizing the formed particles can be
written as
dm0 t
Bm3 t; gPF t;
dt

27a

dm1 t
Gm3 t  m0 t;
dt

27b

dm2 t
2Gm3 t  m1 t;
dt

27c

dm3 t
3Gm3 t  m2 t:
dt

27d

In the above stated evolution of moments all dependencies are


unambiguously accessible. Using the notion of differential
atness, we are going to show that the system states m0 ; . . . ; m3
together with the control, realized by gPF , form a nite-dimensional non-linear at system.

ARTICLE IN PRESS
V. Vassilev et al. / Chemical Engineering Science 65 (2010) 21832189

jt

28

Gm3 t

introduces a new time variable t by means of


dt Gm3 tdt;

t0 0:

29

Invertibility is provided through the requested permanent supersaturation of the solution. Otherwise particles may be dissolved
and the corresponding moment trajectories fail to be unique. The
time-scaling commutes the system (27a)(27d) to
dm0 t
Bm3 t; gPF t
;

dt
Gm3 t

dm2 t
2m1 t;
dt

30c

30d

Moreover, as a conclusive point for orbital atness, properties (19)


and (20) remain to be validated for our approach. Analyzing the
structure of the underlying system of equations,
31

is supposed to be an appropriate choice for the observation


function. Being a state variable itself, m3 obviously fulls Eq. (18).
Furthermore, the latter two Eqs. (19) and (20) are checked by
differentiating the output with respect to t:
dwt
3m2 t;
dt

32a

d wt
6m1 t;
dt2

32b

d wt
6m0 t;
dt3

32c

d wt
Bm3 t; gPF t
:
6
Gm3 t
dt4

0.1265

0.126

0.1255

0.01

0.02

0.03

time [s]
30b

wt m3 t

f(t)

0.125

30a

dm1 t
m0 t;
dt

dm3 t
3m2 t:
dt

0.127

interfacial energy [J/m ]

In order to proof that the system (27a)(27d) of ordinary


differential equations is orbitally at, we need to apply an
appropriate time-scaling. Requiring that the solution is kept
supersaturated, the time scaling function

32d

Using the rst three Eqs. (32a)(32c), plus the observation


variable w itself to construct a proper function jw; w0 ; . . . ; wiv ,
the required correlation between the system states and the
output is shown. The last Eq. (32d) implies the remaining
condition (20), as the parameter gPF can be obtained by solving
the nonlinear equation. Recapitulating the fact that wt denes a
at output, the system (30a)(30d) is differentially at and
consecutively the original system (27a)(27d) yields the property
of orbital atness.
The shown equivalence constitutes the basis to identify the
interfacial energy from a given evolution of moments. We use this
fact to calculate gPF t and subsequently validate or adapt the
modeling of the homogeneous nucleation rate. However, the
relationship is still expressed in the new notion of time t. In order
to retrieve the original time prole, Eq. (29) has to be inverted
analytically or numerically.

7. Numerical results
Using the notion of the previously proven orbital atness, the
calculation of the optima value of the interfacial energy becomes

Fig. 6. Comparison of the temporal evolution of the surface energy (line: value
taken for forward simulation, cross: data from inverse simulation).

PF calculated by inverse simulation [J/m2]

2188

0.128
0.126
0.124
0.122
0.12

+0.8%
0.8%

0.118
0.116
0.114
0.112
0.112 0.114 0.116 0.118

0.12

0.122 0.124 0.126 0.128

PF by Grahame equation [J/m2]


Fig. 7. Comparison of the interfacial energy by the Grahame equation and the
inverse simulation.

feasible. A forward simulation in PARSIVAL solving the full PBE


with predened values for the educt concentration was conducted. The calculated evolution of the third moment acted as an
input for the inverse simulation. In Fig. 6 the obtained values for
the interfacial energy are presented. At later time steps the
inuence of the nucleation rate and thus of the control decreases
to such an extent that no reliable calculation of considered
parameter is possible. At this point the particle number in the
system has reached 99.99% of its end value. This simulation is
conducted for a certain number of test cases with different educt
concentrations. In Fig. 7 the predicted values of the interfacial
energy by the inverse simulation is compared to the physical
model. For the considered precipitation reaction the Grahame
equation yields consistent results to validate the presented
approach. In contrast, for the case of more complex molecules
or structures no proper model of this decisive parameter is
available. The stated comparison encourages that this technique
may provide at least a reasonable range for the interfacial energy
allowing for further investigation. All the values deviate 7 0:8%
from the values for the interfacial energy calculated by means of
the Grahame equation. Thus a quantity like the interfacial energy,
which is difcult to measure, can be calculated from the third
moment being in direct correlation to the amount of precipitated
substance and therefore accessible by measurements.

ARTICLE IN PRESS
V. Vassilev et al. / Chemical Engineering Science 65 (2010) 21832189

Since the results are in good agreement with the theoretical


prediction (conf. Fig. 7) the underlying model has to be consistent
with the prevailing reaction mechanisms. The steps interconnecting
the calculated values with the measured output, i.e. the third
moment m3 , are thus veried and must correspond to the considered
precipitation process on the main lines. This way a validation of the
nucleation rate given by Eq. (4) has been achieved in an indirect way
by conrming the inverse calculation of the interfacial energy.

t
T
~
u
Vm
x

In this work, the modeling of the particle synthesis process in


precipitation reactions was discussed. Starting from the population balance equation, the sensitivity to the involved parameters
motivated a further investigation of their modeled values.
Therefore, a new method for validating the model of the
interfacial energy by means of a transformed moment model was
investigated. Using the orbital atness property of the resulting
system of ordinary differential equations, accessible measurements of the third moment representing the volume fraction
provide information on the evolution of the interfacial energy
acting as a control variable. Calculating the optimal control yields
therefore an estimate of this decisive parameter.
For more complex molecules the obtained parameter estimates computed by the simplied moment technique can
subsequently be used in a more elaborated simulation procedure.
The resulting evolution of the particle size distribution provides
then a continuous description of the system states based on the
improved parameter estimates.

9. Notation
ai
B
Bagg
Bhom
c
D
D
Dagg
f x
G
Hi
kB
kD
kv
KSP
M

activity coefcient, dimensionless


general birth rate density function, m4 s1
birth rat6e density function of agglomeration, m4 s1
nucleation rate, m3 s1
molar concentration, kmol m3
diffusion coefcient, m2 s1
general death rate density function, m4 s1
death rate density function of aggregation, m4 s1
density distribution of nuclei sizes, m1
growth rate, m s1
Henry-type adsorption coefcient of species i, m3 m2
Boltzmann constant (1:381  1023 ), J K1
mass transfer coefcient, m s1
volume shape factor, dimensionless
2

solubility product, kmol m6


1

molecular weight, kg kmol


nx; t particle number density of size x, m4
1
NA
Avogadros number (6:023  1026 ), kmol
~
spatial position in the main duct of the mixer, m
r
R
ideal gas constant (8314), J kmol K1
S
Sh
t

supersaturation, dimensionless
Sherwood number, dimensionless
normal time, s

residence time, s
temperature, K
velocity eld in the main duct of the mixer, m s1
molecular volume, m3
particle size, m

Greek letters

g0
8. Conclusion

2189

gPF
mi
n
rP
s
c
cs

material specic contribution of interfacial energy,


J m2
interfacial energy of particle-uid interface, J m2
i-th moment of particle distribution, various
moles of ions in 1 mole of solute, dimensionless
density of particles, kg m3
surface charge density, C m2
electrical potential, V
surface potential, V

References
Bromley, L.A., 1973. Thermodynamic properties of strong electrolytes in aqueous
solutions. A.I.Ch.E. Journal 19 (2), 313320.
Fliess, M., Levine, J., Martin, P., Rouchon, P., 1995. Flatness and defect of nonlinear
systems: introductory theory and examples. International Journal of Control
61 (6), 13271361.
Fliess, M., Levine, J., Martin, P., Rouchon, P., 1992. Sur les systemes non lineaires
differentiellement plats. Competes Rendus LAcademic des Sciences Paris
I/315, 619624.
Guay, M., 1999. An algorithm for orbital feedback linearization of single-input
control afne systems. Systems Control Letters 38, 271281.
Hunter, R.J., 1986. Foundations of Colloid Science, vols. 1 and 2. Oxford University
Press, Oxford, United Kingdom.
Israelachvili, J., 1991. Intermolecular and Surface Forces, second ed. Academic
Press, London, Great Britain.
Judat, B., Kind, M., 2004. Morphology and internal structure of barium sulfate
derivation of a new growth mechanism. Journal of Colloid and Interface
Science 269 (2), 341353.
Marchisio, D., Barresi, A., Garbero, M., 2002. Nucleation, growth, and agglomeration in barium sulfate turbulent precipitation. A.I.Ch.E. Journal 48 (9),
20392050.
Martin, Ph., Murray, R.M., Rouchon, P., 2003. Flat systems, equivalence and
trajectory generation. Technical Report.

Mersmann, A., Angerhofer,


M., Franke, J., 1994. Controlled precipitation. Chemical
Engineering & Technology 17 (1), 19.

Mersmann, A., Bartosch, K., Braun, B., Eble, A., Heyer, C., 2000. Moglichkeiten
einer

vorhersagenden Abschatzung
der Kristallisationskinetik. Chemie Ingenieur
Technik 71 (1-2), 1730.
Moya, P., Ortega, R., Netto, M.S., Praly, L., Pico, J., 2001. Application of nonlinear
time-scaling for robust controller design of reaction systems. International
Journal of Robust and Nonlinear Control 12 (1), 5769.
Schwarzer, H.-C., Peukert, W., 2004a. Combined experimental/numerical study on
the precipitation of nanoparticles. A.I.Ch.E. Journal 50 (12), 32343247.
Schwarzer, H.-C., Peukert, W., 2004b. Tailoring particle size through nanoparticle
precipitation. Chemical Engineering Communications 191 (4), 580606.
Schwarzer, H.-C., Schwertrm, F., Manhart, M., Schmid, H-J., Peukert, W., 2006.
Predictive simulation of nanoparticle precipitation based on the population
balance equation. Chemical Engineering Science 61 (1), 167181.
Sugimoto, T., 1996. A new approach to interfacial energy: 1. Formulation of
interfacial energy. Journal of Colloid and Interface Science 181, 259274.
Sugimoto, T., 1999. A new approach to interfacial energy: 2. Interfacial energies of
different interfaces under the inuence of adsorption. Journal of Physical
Chemistry B 103, 273286.
Vollmer, U., Raisch, J., 2003. Control of batch cooling crystallization processes
based on orbital atness. International Journal of Control 76, 16351643.
Wulkow, M., Gerstlauer, A., Nieken, U., 2001. Modelling and simulation of
crystallization process using PARSIVAL. Chemical Engineering Science 56,
25752588.

You might also like