Some Modelling Aspects of Aggregation Kernels

and the Aggregation Population Balance

Equations

Nageswara Rao Narni1 , Gerald Warnecke2, Jitendra Kumar3 , Mirko Peglow4,

and Stefan Heinrich5
1
Department of Mathematics, Rajiv Gandhi University of Knowledge Technologies,
Hyderabad-500 032, India
narninrao@gmail.com
2
Institute for Analysis and Numerics, Otto-von-Guericke-University Magdeburg,
Universitätsplatz 2, D-39106 Magdeburg, Germany
3
Department of Mathematics, IIT Kharagpur, India
4
Institute for Process Engineering, Otto-von-Guericke-University Magdeburg,
Universitätsplatz 2, D-39106 Magdeburg, Germany
5
Institute of Solids Process Engineering and Particle Technology, Hamburg
University of Technology, Hamburg, Germany

Abstract. Aggregation is one of the important processes in chemical

and process engineering. To model this process aggregation population
balance equation is widely used. The aggregation kernel is one of the
most effective parameter of this aggregation equation. This is usually
defined as the product of the aggregation efficiency and collision fre-
quency functions. Several authors attempted to model the aggregation
kernel using different approaches like theoretical, experimental and em-
pirical approaches. In the present paper we are giving some important
modelling aspects of these kernels. One can use these modelling aspects
in other variants of the population balance equations with breakage, nu-
cleation, growth, ..., etc.

Keywords: Aggregation equation, aggregation kernel, collision frequency

function, dimensional analysis.

1 Introduction
Population balance equations are widely used in the literature to model the
granulation process in the fluidized beds on a macroscopic level. Kernels are
the major parameters in these integro-partial differential equations. Aggregation
equation is one of the important class of these equations. It describes the particle
growth through aggregation of particles. The main parameter in this aggregation
equation is the aggregation kernel.
Modelling of the aggregation kernel was done through theoretical, empirical
and experiments approaches. Smoluchowski [13] derived the shear kernel and
brownian kernel using theoretical considerations. Sastry [11], Kapur [7] obtained

J. Mathew et al. (Eds.): ICECCS 2012, CCIS 305, pp. 319–326, 2012.
c Springer-Verlag Berlin Heidelberg 2012
320 N.R. Narni et al.

the aggregation kernels by empirical approach. Aggregation kernels can be mod-

elled by fitting the experimental data to the empirical kernels. Peglow [10] ob-
tained an empirical kernel by fitting the experimental data to the empirical kernel
proposed by Kapur [7].
Due to improvement in computational power, some authors try to extract
the kernels by using computer simulations. Recently Gant et al. [6] obtained a
multi dimensional aggregation kernel by simulating a granulator. Tan et al. [14]
derived an aggregation kernel based on the principle of kinetic theory of granular
flow to the fluidized bed melt granulation.
In the present paper we wish to present some modelling aspects of these ker-
nels for population balance equations and a correction to the population balance
equation for the fluidized bed. It is shown that when the number of particles in-
creases the correction is negligible. The present modelling aspects can be applied
to population balance equations with growth, breakage and nucleation.

2 Different forms of Aggregation Equation

The aggregation equation was first derived by Smoluchowski [13]. He derived
an infinite set of differential equations describing the particle aggregation. Later
Mueller [8] derived a continuous form for these equations and is an integro-partial
differential equation of nonlinear type. The analytic solutions of these equations
exists only for a few types of kernels and can be found in [2].

2.1 Discrete Population Balance Equation

Disperse systems describes the solid or liquid particles suspended in a medium
(usually in a gas). The disperse systems are usually unstable and changes with
time due to the movement of the particles in the medium. The particles gets
the movement by external fields like gas flow, stirrer, ...etc. As a result of this
movement, the particles collides. The collision and coagulation in this system
leads to the change in the particle size (Here we took particle volume as size)
distribution. Population balance equations are very frequently used to study the
particle size distribution in the disperse systems. Smoluchowski [13] derived an
infinite set of nonlinear differential equations for his theory of rapid coagulation
process with the following assumptions:
– The number of particles per unit volume of the fluid are sufficiently small
(i.e. dilute system) with constant porosity
– Binary collisions are assumed to occur simultaneously
– Incompressible spherical particles are assumed to collide
Based on the above assumption he derived the following equations:
∞
1 
i−1
dni (t)
= Ki−j,j ni−j (t)nj (t) − ni Ki,j nj (t) (1)
dt 2 j=1 j=1
 Some Modelling Aspects of Aggregation Kernels Modelling 321

(0)
for i = 1, 2, 3, ..., with the initial conditions ni (0) = ni ≥ 0.
The above equation (1) is known as the discrete coagulation equation. The
complexity of the above system is dependent on the form of the kernel Ki,j .
The function Ki,j is known as the coagulation kernel. It describes the intensity
of the particle interactions between particle classes i and j. The coagulation
kernel is non-negative and symmetric, i.e. Ki,j ≥ 0 and Ki,j = Kj,i . The un-
known non-negative function ni (t) is the concentration of particles with size
i−1
i, i ≥ 1. The term 12 j=1 Ki−j,j ni−j (t)nj (t) is known as the birth term and
∞
ni j=1 Ki,j nj (t) is knows as the death term.

2.2 Continuous Population Balance Equation

Mueller [8] derived the continuous form for the above equation as an integro
partial differential equation defined as:

 u
∂n(t, u) 1
= K(t, u − v, v)n(t, u − v)n(t, v)dv
∂t 2 0
 ∞
−n(t, u) K(t, u, v)n(t, v)dv, (2)
0

with the initial condition n(0, u) = n0 (u) ≥ 0. The Kernel K(t, u−v, v) describes
the coagulation of particles of size u − v with particles of size v. The first term
of the right hand side of equation(2) describes birth of the particles. The second
term in the right hand side of the equation(2) describes the death of the particles.

3 Dimensional Analysis of the Equation

Dimensional analysis is one of the primitive modelling tool for Physicist, Engi-
neers and Applied mathematicians for modelling a physical phenomena. It allows
us to calculate the scale factors when using physical models. We are going to
use it in calculating the scaling factor K(u, v), which is known as aggregation
kernel. To discuss the concepts of dimensional analysis, we need the following
definitions:

– We defines the mass M as the mass of the particle with units Kilogram.
– We defines the volume for two physical quantities:
1. Volume of fluid as Vf luid and has the dimension length L3 for notational
purpose we writes it as Lf luid , which is an external parameter corre-
sponds to space coordinate with units Meter.
2. Volume of particles as Vparticles and has the dimension length L3 for no-
tational purpose we writes it as Lparticles , which is an internal parameter
corresponds to particle property with units Meter.
– We defines time as T and has units Seconds.
322 N.R. Narni et al.

3.1 Dimensional Analysis for Discrete Equation

– Particle concentration ni (t) is defined as the number of particles per unit
volume of the fluid and had the dimension M 0 L−3f luid T
0

– The coagulation kernel Ki,j is defined as the number of particle interactions

in unit volume of fluid per unit time and had the dimension M 0 L3f luid T −1

3.2 Dimensional Analysis of the Continuous Equation

Case 1:
– Particle concentration n(t, u) is defined as the number of particles per unit
volume of fluid per unit particle volume and had the dimension
M 0 L−3 −3
f luid Lparticle T
0

– Dimension of kernel K(t, u, v) is M 0 L3f luid T −1

Case 2:
– Suppose for the continuous equation we have dimension of n(t, u) as Number
of particles per Unit volume of particles. That is we are considering the
number density as function of time and material volume with the dimension
M 0 L−3 0
particle T .
– Dimension of the kernel K(t, u, v) is M 0 L0particle T −1 . This is usually called
as collision frequency kernel in the literature [4].

3.3 Number Density Selection and Dimensional Analysis of the

Kernel in Applications
When we are trying to use this aggregation equation for a particulate process,
the dimensions of the kernels can be calculated as:
In the study of atmosphere sciences the particle volume is negligible compared
to the volume of the air. So in such models discrete kernel is widely used by
considering the number density as function of fluid and time. In this case the
discrete kernel K(t, u, v) has the dimension L3air T −1 .
In a fluidized bed the volume of the bed is not constant. Therefore the porosity
of the bed changes with the time. So in such situations it is necessary to consider
both, the volume of fluid and volume of the particles in the calculation of particle
concentration. Since the porosity is not constant, we consider the case 1 and the
kernel has the dimension L3f luid T −1 .
As we stated earlier, the number density can be a function of any material co-
ordinate. In certain applications the mass of the particles is important compared
to volume of the particles. In crystallization processes, particles are formed as a
result of change in the thermodynamics of the material. So in such processes the
number density solely depends on the amount of material present in the vessel.
Therefore we can replace the number density as a function of material and time,
and having the dimension M −1 L−3 . In this case the aggregation kernel has the
dimension of M L3 T −1 . Here L3 represents the volume of the vessel.
 Some Modelling Aspects of Aggregation Kernels Modelling 323

4 Derivation of Collision Frequency Functions

Aggregation processes lead to a reduction in the total number of particles and an

increase in the average size. Therefore agglomeration leads a major change in the
system as the time changes. As a result, the aggregation rate will be effected,
which is a function of collision frequency function and aggregation efficiency.
Here we are deriving an expression for collision frequency function.
Let Ni,j be the number of collisions occuring per unit time per unit volume
between the two classes of particles of volumes vi and vj . All particles are as-
sumed to be spherical in shape, and collisions are treated as binary hard sphere
collisions. The collision frequency function Ki,j can be written in terms of con-
centrations of particles of volumes vi and vj as:

Ni,j = Ki,j ni nj , 1 ≤ i, j ≤ I

where ni , nj are the number concentrations of classes i, j. The number I denotes

the total number of classes. The parameter Ki,j is the frequency function among
different classes i and j. The collision frequency function Ki,j is a function of
flow properties, i.e. particle size, gas velocity, temperature, etc.
The number of collisions occuring during time tsim in the fluidized bed of
volume Vbed is given as

tot Ni Nj
Ni,j = Ki,j tsim Vbed , (3)
Vbed Vbed

where Ni , Nj are the number of particles of class i, j present in the fluidized bed
Nj
of volume Vbed , i.e. ni = VNbed
i
, nj = Vbed .

From the Equation (3), the collision rate function is obtained as

tot
Ni,j Vbed
Ki,j = . . (4)
Ni Nj tsim

In case of collisions within a size class, i.e. i = j, the collision rate function is
obtained as
tot
Ni,i Vbed
Ki,i = Ni (Ni −1)
. (5)
tsim
2

which differs from the previous one due to the correction for self collisions of the
particles, which was observed by Hu and Mei [5]. The authors observed a devia-
tion of about 2 percent in their simulations. This correction is negligible in the
case of continuous spectrum, because of the assumption of large concentrations,
in which, we take the approximation Ni (Ni − 1)/2 ≈ Ni2 /2. The details of the
derivation of (5) by using a combinatorial approach can be found in Sastry [12].
324 N.R. Narni et al.

5 Correction to Aggregation Equation

Let us consider the combinatorial approach of the particle collisions among dif-
ferent classes and within the same class of particles. Here classes are based on the
volume of the particle, i.e, V0 , 2V0 , 3V0 , ... and V0 is the initial particle volume
class.
The Figure 1 describes the number of possible collisions among two classes
A and B which contain 2 and 3 particles respectively. The number of possible
collisions between A and B are 6. The number of possible collisions among the
particles of class B are 3.

Fig. 1. Combinatorial calculation of the particle collisions

We can generalize the above approach as follows. Suppose there are Ni parti-
cles of ith class and Nj particles of jth class at a given instant. Then the number
of possible collisions among ith class and jth class are Ni Nj . When we have col-
lisions among the particles of class i, then the number of possible collisions are
Ni (Ni − 1)/2.

Table 1. Simulations for collisions among the same size particles

Simulation N0 Same size collisions (5) Same size collisions (4) Difference
1 5000 6.7155e-05 6.7088e-05 6.7155e-08
2 10000 1.6566e-05 1.6557e-05 8.2828e-09
3 15000 1.4514e-05 1.4509e-05 4.8379e-09
4 20000 1.1776e-05 1.1773e-05 2.9440e-09

We can apply the above combinatorial approach to the fluidized bed as follows.
Let Ki,j be the collision frequency function of the fluidized bed. Then the number
of collisions among the classes i and j per unit time and per unit volume are
Ni,j = Ki,j Ni Nj . In case of collisions among the particles in the ith class are
Ni,j = Ki,j Ni (Ni − 1)/2. The case of collisions among the same class are not
considered during the derivation of the aggregation equation. So here we are
showing the effect of this in modelling the aggregation kernel in Table 1. In
this simulation we considered 10 classes of particles with an equal number of
particles in each class. For details of initial conditions of the simulations and
 Some Modelling Aspects of Aggregation Kernels Modelling 325

Discrete Particle Model (DPM) see Narni [9]. We made simulations with an
increasing the number of particles.
From the above simulation Table 1, we can observe that when the number of
particles is high, i.e. N0 is large enough, this new correction term has little effect
on the aggregation equation. Therefore in industrial level fluidized beds the new
correction is negligible, since the number of particles is very high.

6 Conclusions

In the present paper we gave few modelling aspects of aggregation equation

and a correction to it. The dimensional analysis of the aggregation equation
gives a clear insight into the scaling parameter (kernel) of the equation and its
applicability to physical models. The present Discrete Particle Model simulations
shows that the new correction for large particle systems has negligible effect.
The present dimensional analysis can be extended to population balance equa-
tions with combined processes such as breakage, nucleation, growth, ... etc. The
present strategies are very useful for modelling multi particle aggregation (i.e.,
ternary,...etc.) and multidimensional population balance equations.

Acknowledgments. This work was supported by the Graduiertenkolleg-828,

”Micro-Macro-Interactions in Structured Media and Particles Systems”, Otto-
von-Guericke-University, Magdeburg, Germany. The authors gratefully acknowl-
edge for funding through this Ph. D. program.
Nomenclature

n(t, u) Number density at time t of particle property u —–

t Time sec
Ni,j Number of collisions between class i and class j m3 /sec
ni Particle concentration of class i 1/m3
nj Particle concentration of class j 1/m3
Ni Number of particles of class i dimensionless
Nj Number of particles of class j dimensionless
Ki,j Collision rate function m3 /sec
Vbed Volume of the bed m3
tstep Simulation time steps sec
u, v Volume of the particles m3

References
[1] Friedlander, S.K.: Smoke, dust and haze. John Wiley & Sons, New York (1977)
[2] Gelbard, F., Seinfeld, J.H.: Numerical solution of the dynamic equation for par-
ticulate systems. Journal of Computational Physics 28, 357–375 (1978)
[3] Hoomans, B.P.B.: Granular dynamics of Gas-Solid two-phase flow. PhD thesis,
Twente University, The Netherlands (2000)
[4] Hounslow, M.J.: The population balance as a tool for understanding particle rate
processes. Kona, 179–193 (1998)
326 N.R. Narni et al.

[5] Hu, K.C., Mei, R.: A note on particle collision rate in fluid flows. Phys. Fluids 10,
1028–1030 (1998)
[6] Gantt, J.A., Cameron, I.T., Litster, J.D., Gatzke, E.P.: Determination of coales-
cence kernels for high-shear granulation using DEM simulations. Powder Technol-
ogy 170, 53–63 (2006)
[7] Kapur, P.C.: Kinetics of granulation by non-random coalescence mechanism.
Chemical Engineering Science 27, 1863–1869 (1972)
[8] Mueller, H.: A calculation procedure for heat, mass and momentum transfer in
three dimensional parabolic flow. Int. J. Heat and Mass Transfer 15, 1787 (1922)
[9] Rao, N.N.: Simulations for modelling of population balance equations of particu-
late processes using discrete particle model. PhD thesis, Otto-von-Guericke Uni-
versity, Magdeburg, Germany (2009)
[10] Peglow, M.: Beitrag zur modellbildung eigenschaftsverteilten dispersen syste-
men am Beispiel der wirbeschicht-Spruenhagglomeration. PhD thesis, Otto-von-
Guericke University, Magdeburg, Germany (2003)
[11] Sastry, K.V.S.: Similarity size disribution of agglomerates during their growth by
coalescence in granulation or green pelletization. International Journal of Mineral
Processing 2, 187–203 (1975)
[12] Sastry, K.V.S., Fuerstenau, D.W.: Size distribution of agglomerates in coalescing
dispersed phase systems. I & EC Fundamentals 9(1), 145–149 (1970)
[13] Smoluchowski, M.V.: Mathematical theory of the kinetics of the coagulation of
colloidal solutions. Z. Phys. Chem. 92, 129–168 (1917)
[14] Tan, H.S., Goldschmidt, M.J.V., Boerefijn, R., Hounslow, M.J., Salman, A.D.,
Kuipers, J.A.M.: Building population balance model for fluidized bed melt granu-
lation: lessons from kinetic theory of granular flow. Powder Technology 142, 103–
109 (2004)