Powder Technology 326 (2018) 190–207

Contents lists available at ScienceDirect

Powder Technology

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

Review

Population balance modelling to describe the particle aggregation

process: A review
Ricardo I. Jeldres a,⁎, Phillip D. Fawell b, Brendan J. Florio b,c
a
Department of Chemical Engineering and Mineral Process, Universidad de Antofagasta, PO Box 170, Antofagasta, Chile
b
CSIRO Mineral Resources, Waterford, Western Australia, Australia
c
School of Mathematics and Statistics, University of Western Australia, Crawley, Western Australia, Australia

a r t i c l e i n f o a b s t r a c t

Article history: Unit operations used to achieve solid-liquid separation for fine particle suspensions rely upon efficient aggregate
Received 4 August 2017 formation. There is considerable potential for predictions from population balance models describing particle
Received in revised form 17 November 2017 aggregation to help optimise full-scale processes. The vast majority of studies in this area make use of the classical
Accepted 4 December 2017
coagulation equation of Smoluchowski, and while developed primarily for coalescence phenomena, it has been
Available online 8 December 2017
adapted and modified extensively to describe particle aggregation for many different substrates and procedures
Keywords:
for inducing aggregate formation. This has resulted in a wide variety of mathematical expressions, some of
Flocculation which are highly sophisticated but can only be applied successfully to a limited range of conditions. For this
Polymer bridging reason, it is necessary for researchers to understand the main aggregation mechanisms involved in the processes
Aggregate (coagulation, bridging flocculation) and how the system conditions (flow regime, particle size, solids concentra-
Fractal tion) can then influence aggregate growth, breakage and the resulting structures. Such understanding is essential
Population balance modelling for the appropriate selection of mathematical equations to then obtain a successful model that can be solved at
Review low computational cost. The main mathematical expressions developed for the different phenomena that
occur during particle aggregation (i.e. collision frequency and efficiency, aggregate breakage rate and distribu-
tion, the structure formed and their potential for restructuring over time) by different mechanisms are reviewed.
The main published studies are critically assessed, indicating their scope and the conditions under which the
models can be usefully applied. Important challenges remain towards achieving wider practical applications,
particularly in reducing reliance on empiricism. Particular emphasis is placed on research oportunities, focusing
mainly on i) the incorporation of interaction forces for colloidal systems at submicron particle; ii) importance
of achieving more reliable representations of aggregation behaviour at high solid concentration; and
iii) incorporation of PBEs within computational fluid dynamics (CFD) models that describe industrial aggregation
processes as a powerful tool for full-scale unit design and process optimisation.
© 2017 Elsevier B.V. All rights reserved.

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
2. Mathematical models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
2.1. Smoluchowski coagulation equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
2.2. Modification Smoluchowski coagulation equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
2.3. Collision frequency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
2.3.1. Rectilinear approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
2.3.2. Curvilinear approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
2.4. Collision efficiency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
2.4.1. Equations based on Adler's trajectory analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
2.4.2. Interaction forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
2.4.3. Polymer adsorption . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
2.4.4. Polymer depletion on collision efficiency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197

⁎ Corresponding author.
E-mail address: jeldresresearch@gmail.com (R.I. Jeldres).

https://doi.org/10.1016/j.powtec.2017.12.033
0032-5910/© 2017 Elsevier B.V. All rights reserved.
 R.I. Jeldres et al. / Powder Technology 326 (2018) 190–207 191

2.5. Breakage rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197

2.5.1. Exponential breakage model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
2.5.2. Power law breakage model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
2.5.3. Polymer depletion on aggregate breakage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
2.6. Breakage distribution function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
2.7. Aggregate restructuring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
2.8. Model solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
3. Analysis of selected studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
4. Research opportunities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
4.1. Reduced reliance on empiricism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
4.2. Interaction forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
4.3. High solids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
4.4. CFD for aggregation prediction in 3D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
5. Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206

1. Introduction
Solid-liquid separation processes are complex and challenging to
control. However, these are fundamental operations for numerous
industrial applications, such as in mineral recovery, tailings manage-
ment, oil extraction, paper production and wastewater treatment
[1–8].
In the case of mining and mineral processing, efficient water recov-
ery stages are needed to ensure the sustainability of the industry, given
the extreme scarcity of water in many regions of countries like Chile,
Peru, Australia and South Africa. ‘Gravity thickeners’ are the key equip-
ment towards this purpose, since they enable rapid water recovery for
use in upstream stages in high volumetric throughput applications.
Aqueous solutions of flocculants (long chain water-soluble polymers)
are added to bridge the fine particles into large fast-settling aggregates
with the aim of producing clarified liquor discharged at the thickener's
periphery (overflow) and a thickened sediment (underflow) at its
base. Similar aggregation processes are employed in settling vessels
used in other industries, although rarely seeking the same high settling
velocities.
Separation efficiency as determined through settling velocity,
quality of recovered water, and rheological properties of the resulting
sediment can be directly related to the size, structure and strength of
these aggregates [9]. For this reason, much emphasis has been placed
on studying aggregation from a microscopic point-of-view, i.e. directly
examining the physical properties of aggregates, made possible through
modern characterization techniques [10].
There have also been many attempts to describe aggregate
properties and the aggregation process through analytical models.
Such modelling on a simple one-dimensional level alone can provide
valuable insights into the factors that influence aggregation perfor-
mance [11]. When able to be incorporated into computational
fluid dynamics (CFD) modelling, the combined predictions in three-
dimensions of flow patterns with aggregate growth and breakage can
become a powerful tool for full-scale unit design and optimisation of
operation, such as for flocculation processes in thickener feedwells
[12–14]. Such outputs are now being used to as accurately as possible
generate surrogate model over the complete design space of interest
using as few simulation points as possible - once constructed,
the surrogate can be quickly reused for sensitivity analysis or in control
[15].
These models are based on population balance equations that make
use of the classical Smoluchowski coagulation equation (1917), which
describes the rate of irreversible aggregation. This equation is useful
to describe aggregates that are spherical and maintain their form
throughout the process but such aggregates are rarely observed in
practice, and the equation does not describe aggregates that have an
open and irregular structure and also change their shape over time.
The latter is what happens during polymer bridging flocculation, with
the aggregates generated by high molecular weight polymer being
readily broken or densified. The physicochemical response is complex
because many phenomena act simultaneously, being very difficult
to describe them mathematically. Because of this, the models have
a high level of empiricism that can limit their applicability to real
situations. However, many researchers have proposed modifications
that have led to the variety of mathematical expressions currently
available, some of which are highly sophisticated but may only be
applicable over a limited range of conditions. For this reason, it is
necessary for researchers to be aware of the principal phenomena
and aggregation mechanisms involved in the processes (coagulation,
bridging flocculation, aggregate structure permeability and breakage,
etc.), as well as the critical system conditions (flow regime, particle
size, solids concentration). Such understanding is fundamental towards
making an appropriate selection of the mathematical equations and
then obtaining a successful model that can be solved in reasonable
times.
The main mathematical expressions that have been developed to
describe the different phenomena that occur during the particle aggre-
gation are discussed below.
2. Mathematical models
2.1. Smoluchowski coagulation equation
Many phenomena act simultaneously when particle aggregation by
polymer bridging occurs:
• collisions between particles, polymers and any existing aggregates;
• adsorption of polymer on the solid surface;
• reorientation of adsorbed polymer chains;
• aggregation, whether that be particle-particle, particle-aggregate or
aggregate-aggregate;
• aggregate fragmentation;
• in some applications (depending on the polymer and adsorption
mechanism), re-aggregation of broken aggregates;
• aggregate restructuring and consolidation.
These processes do not occur in isolation, but rather they overlap and
potentially interact with each other, making it difficult to incorporate all
aspects of aggregation into a single model. However, there are many
mathematical models to describe specific processes, such as colloidal
aggregation/breakage, polymer adsorption or aggregate rearrangement,
which have then been fitted to experimental data [16–19].
192 R.I. Jeldres et al. / Powder Technology 326 (2018) 190–207
The classical equation proposed by Smoluchowski describes the rate
at which aggregates that contain i and j particles, respectively, come
together to form aggregates with m particles, where m = i + j. This
equation is given by:
X
dNm 1 m−1 X∞
¼ α i;m−i βi;m−i Ni Nm−i −Nm α i;m βi;m N i ;
dt 2 i¼1 i¼1
where αi, j corresponds to the collision efficiency between particles of
size i and j, βij is the collision frequency, and Ni is the number concentra-
tion of the aggregates containing i particles. For the development of
this expression, Smoluchowski made a number of basic assumptions,
such as:
i) all collisions are effective (αi, j = 1),
ii) particles are of the same size, and like the aggregates formed,
have a spherical shape;
iii) collisions are between two particles, and this is due to the
laminar movement of the fluid; and
iv) aggregate fragmentation is not considered.
Although this model is useful to describe aggregation by coales-
cence, attempts have been made to apply it to cases of particle aggrega-
tion under different flow conditions, and therefore several researchers
have proposed modifications giving greater relevance to practical
systems.
2.2. Modification Smoluchowski coagulation equation
Population balance equations can be written in continuous [20] or
discrete forms [21]. The former assumes that the aggregate size distri-
bution is continuous, generating an integral whose analytic solution is
very complicated. For the latter, aggregates are grouped depending on
their size to then conduct the equations. This generates a system of
ordinary differential equations (ODE) where the number of equations
required must be chosen to represent the full range of aggregate sizes.
The difficulty is that the ODE system becomes highly stiff, since the
dynamic evolution of the number concentration of small aggregates is
significantly different from that of larger clusters. For this reason,
discretisation of the size distribution is first performed, which allows a
Fig. 1. Dynamics of aggregation and breakage for discretised population balance, adapted from [23].
considerable reduction in the number of equations and the computa-
tional time. The most classical discretisation approach is the geometric
distribution (i.e. Vi + 1 = 2Vi), from which a model widely used
to describe the change in particle concentration was proposed by
Hounslow et al. [17] and modified for breakage by Spicer and Pratsinis
[22], given by:
ð1Þ
dNi X i−2
¼ 2 j−iþ1 α i−1; j βi−1; j N i−1 N j
dt j¼1
1 X
i−1 X
m ax1
þ α i−1;i−1 βi−1;i−1 N2i−1 −Ni 2 j−1 α i; j βi; j N j −Ni α i; j βi; j N j
2 j¼1 j¼i
X
max2
þ Γ i; j S j N j −Si N i ;
j¼i
ð2Þ
where Ni is the number concentration of aggregates containing 2i − 1
particles. Fig. 1 shows a scheme to identify the meaning of each term
of the previous expression.
The first two terms on the right of Eq. (2) describe aggregate forma-
tion in interval i, generated by collisions of aggregates belonging to
smaller size intervals. The third and fourth terms represent the loss of
aggregates in interval i because they are adhered, generating an aggre-
gate belonging to a larger size range. The fifth term represents the
gain of aggregates in interval i, resulting from the rupture of larger
aggregates, while the last term is for loss from the interval i due to
breakage, generating smaller aggregates. The superscript max1 is the
maximum number of intervals used to represent the complete aggre-
gate size spectrum, but max2 corresponds to the largest interval from
which aggregates in the current interval are produced. It should be
noted that Hounslow's expression (Eq. (2)) contains two additional
parameters to the Smoluchowski equation. Гi, j is the breakage distribu-
tion function for aggregates of the interval j which form species in the
interval i, while Si represents the breakage rate of aggregates in the
interval i.
Numerous studies have sought to understand and achieve a better
representation of the fundamental expressions within Eq. (2) (αij, βij,
Sij, and Γi, j) that describe the main physical phenomena occurring
during aggregation. For this, analytical and semi-empirical models are
 R.I. Jeldres et al. / Powder Technology 326 (2018) 190–207
used which have parameters that are fitted to experimental data and
then introduced into the balance equation. However, there are still
opportunities to improve the description of the flocculation process,
and these are discussed in later sections.
2.3. Collision frequency
The collision frequency represents the number of collisions that
occur between particles and aggregates, regardless of whether they
result in aggregation or not. There are different ways it can be described,
with three distinct mechanisms currently recognised by which colli-
sions may occur: perikinetic (Brownian motion), orthokinetic (shear
condition), and differential sedimentation (aggregate density-size). It
is important to note that the dominant mechanism depends fundamen-
tally on the size of the particles and aggregates. Submicron particles
will be influenced mainly by temperature effects (Eq. (3)), whereas
collisions for larger species will depend more on the shear conditions
(laminar or turbulent regime) and aggregate density [24] (Eqs. (4)
and (7)). Recently, Oyegbile et al. [25] summarised the main contribu-
tions in this field towards understanding the phenomenon of turbu-
lence, and also proposed a possible direction for future research. In
the case of hydrodynamic interactions, the induced velocity gradient
promotes the aggregation process, but may also be responsible for
aggregate breakage as a result of increased viscous shear stress. Signifi-
cant progress has been made in understanding physicochemical aspects
of aggregation, but turbulence remains as a misunderstood phenome-
non despite its profound influence on many engineered processes.
Researchers have used two approaches to estimate collision
frequencies from the trajectories of particles and aggregates. The classic
rectilinear trajectory approach proposed by Smoluchowski assumes that
the particles are spherical and follow rectilinear paths, i.e. the fluid has
no influence on the collisions. However, as particles approach each
other, the fluid is squeezed out from the gap between particle surfaces.
This causes the particles to rotate about each other and induces
curvilinear motion, which reduces the chances of collision.
2.3.1. Rectilinear approach
Smoluchowski developed the following analytical expression to
determine the collision frequency for Brownian motion:
 2
2kT Ri þ R j
βperikinetic ¼ ; ð3Þ
3μ Ri ∙ R j
where k is the Boltzman constant, T is the absolute temperature, μ is
the dynamic viscosity of the fluid, and Ri corresponds to the radius of
an aggregate belonging to the interval i.
The collision frequency for orthokinetic aggregation in isotropic
turbulence is estimated by Saffman and Turner [26]:
 3
βorthokinetic ¼ 1:294G Ri þ R j ; ð4Þ
where G is the mean shear rate. Although this expression is widely
accepted for the turbulent regime, it must be used with caution.
Pedocchi and Piedra [27] questioned its validity, given that the
utilisation of the root-mean-square velocity gradient may produce a
poor estimate of the coagulation rate.
A similar form of the collision frequency is known for laminar shear
flow, where the numerical constant (1.294) in Eq. (4) is replaced with
4/3 and G with γ,_ the linear shear rate. Additional expressions for both
laminar and turbulent regimes can be found in Meyer and Deglon [28].
Saffman and Turner [26] linked the mean shear rate in turbulent
flow, G, with the local turbulent energy dissipation rate per unit mass,
ε, and with the kinematic viscosity, ν:
G ¼ ðε=ν Þ1=2 ; ð5Þ
193
however, it also may be treated as the power input per unit of volume
(Φ = P/V), where G = (Φ/μ)1/2. For stirred tanks, the power input P
can be estimated by:
P ¼ N p ρ f N3I N5I ; ð6Þ
where ρf is the fluid density, NI and DI are the rotational speed and
diameter of the impeller, and Np is the power number for the particular
tank configuration used [29].
Camp and Stein [30] also incorporated the contribution of differen-
tial sedimentation towards the collision frequency:
 2  
βDS ¼ π Ri þ R j ∙ ui −u j ; ð7Þ
where ui corresponds to the settling velocity of the particle i, which in
diluted systems can be determined by Stokes law, given by:
 
2 ρp −ρl g ∙ R2i
ui ¼ ; ð8Þ
9μ
where ρp and ρl correspond to the particle and medium densities,
respectively. Johnson et al. [31] showed that this equation is useful for
solid spheres, but a modified form should be used for fractal aggregates
(see Eq. (20)). Dukhin et al. [32] described the role of gravity on
the particle pair interactions, including analytical simulations and
numerical stability diagrams for colloids.
The global collision frequency is, therefore, the sum of the contribu-
tions from the three mechanisms that cause movement of particles and
aggregates: Brownian movement, shear, and differential sedimentation:
 2
2kT Ri þ R j  3  2  
βij ¼ þ 1:294G Ri þ R j þ π Ri þ R j ∙ ui −u j  ð9Þ
3μ Ri ∙ R j
2.3.1.1. Aggregate with irregular and impermeable structure. So far,
aggregates have been considered as solid spheres, which undoubtedly
has serious implications for predicting their size distributions, since the
aggregates actually have irregular structures and therefore their collision
radius will be different from that of a sphere. An expression to describe
aggregate irregularity and self-similarity proposed by Mandelbrot [33]
relates the mass fractal dimension (df) to the collision radius, giving:
 1=d f
ni
Rci ¼ R0 ; ð10Þ
kc
where R0 is the radius of the primary particle, ni is the number of primary
particles contained in the aggregate i, and kc is a constant relative to the
packing density which is generally assumed to be kc = 1 [34,35]. The
mass of the aggregate (M) can be related to its characteristic length (l)
and df as M ∝ ldf, where df varies from 1 (aggregate is a line of particles)
to 3 (a solid sphere). The majority of aggregation processes will lead to
df values in the range 1.7 to 2.3, and even at the higher end of this
range, they will be low-density structures for which the majority of the
volume will be liquor.
Fig. 2 shows a computer simulation of a typical structure from diffu-
sion limited aggregation, generated by a two-dimensional off-lattice
random walk of 5000 circular primary particles. The simulation was
“killing-free” whereby a primary particle which does not meet the
aggregate and exits the domain will be reset at its birth position. This
avoids a strong bias towards collisions on long extrusions [36]. The
main value of this image is to highlight the inherent fragility of such
low-density, highly branched structures.
2.3.2. Curvilinear approach
The expressions referred to above are based on a simplistic view of
collisions between two particles, in which the effect of changes in
194 R.I. Jeldres et al. / Powder Technology 326 (2018) 190–207

Fig. 2. Simulation of diffusion-limited aggregation of 5000 particles.

fluid movement and short range forces when the particles approach are
ignored. The problem with the rectilinear approach, where particle
motion follows straight lines (Fig. 3, left), is that the collision frequency
is overestimated. The particle trajectories actually deviate due to both
hydrodynamic interactions and short range forces, thereby reducing
their collisions.
Han and Lawler [24] considered a curvilinear trajectory approach
(Fig. 3, right) and presented a set of corrections to the collision frequency
functions, given by:
βcur ði; jÞ ¼ eBr βBr ði; jÞ þ eSh βSh ði; jÞ þ eDS βDS ði; jÞ; ð11Þ

Fig. 3. Possible particle trajectories (left: rectilinear model, right: curvilinear model), adapted from [24].
 R.I. Jeldres et al. / Powder Technology 326 (2018) 190–207
where eBr, eSh and eDS are correction factors for each mechanism (Br:
Brownian motion, Sh: Shear, DS: Differential Sedimentation) and can
be determined with the following correlations:
eBr ¼ a þ bδ þ cδ2 þ dδ3 ;
8 reduced. However, as the particle size increases, the shear collision
eSh ¼ a þ bδ þ cδ2 þ dδ3 ; ð12Þ
ð1 þ δÞ3
2
þdδ3
eDS ¼ 10aþbδþcδ
where δ is the size ratio (smaller to larger) of the two colliding particles
and the coefficients a, b, c, and d can be found in Han and Lawler [24]. As
can be seen in Fig. 4a (rectilinear approach), the collision frequency
function for Brownian motion is a relatively weak function of particle
size. Brownian motion is the dominant mode of collision only when
the variable particle size is quite small (b0.1 μm). Differential sedimen-
tation is a very strong function of both particle size and the collision
frequency function for fluid shear where collisions dominate the broad
mid-region of the particle size distribution. The situation changes
Fig. 4. Collision frequency functions in a) rectilinear model and b) curvilinear model
(DS: differential sedimentation; Sh: Shear; Br: Brownian motion) [24]. The second
particle, i, diameter is fixed at 2 μm.
195
when the curvilinear collision frequency functions are considered
(Fig. 4b). For Brownian motion, the results are quite similar to those of
the rectilinear model, because the correction is relatively small. The
general shape of the curves is also similar to the rectilinear case for dif-
ferential sedimentation, although the collision frequency is drastically
frequency shows a quite different behaviour, reaching a maximum and
then reducing. When analysing the curvilinear model, it is found that
the role of the spatially averaged turbulent shear rate (G) is not as
important as previously thought from the rectilinear approach.
2.3.2.1. Aggregates with irregular and permeable structure. It has been
shown in numerous studies that fluid flow can penetrate through
an aggregate structure [18,37]. However, work based on path analysis
considers that aggregates are impermeable, which results in a consider-
able overestimation of the collision frequency. Somasundaran and
Runkana [38] showed that a permeable aggregate when experiencing
viscous drag presents a collision frequency considerably lower than
that predicted by the rectilinear model (Fig. 5). Jeldres et al. [39] subse-
quently showed that porous aggregates lose permeability over time as a
result of restructuring. They become more compact and smaller so that
the rate of decrease in collision frequency over time is higher than in the
case of impermeable aggregates (Fig. 6).
The degree of fluid penetration in an aggregate can be quantified
through a parameter called the ‘fluid collection efficiency’, η, which
corresponds to the ratio between flow passing through an aggregate
and that approaching it [40]. The magnitude of η depends on the aggre-
gate permeability and therefore its structure. Veerapaneni and Wiesner
[37] proposed that functions to calculate the collision frequency should
consider both the aggregate's permeability and its fractal dimension
(df). They presented modifications for the three mechanisms that
determine the collision frequency:
 
2kT 1 1  
βperikinetic ¼ þ Rci þ Rcj ; ð13Þ
3μ Ωi ∙ Rci Ω j ∙ Rcj
1 pffiffiffiffi pffiffiffiffiffi 3
βorthokinetic ¼ ηi Rci þ η j Rcj G; ð14Þ
6
pffiffiffiffi pffiffiffiffiffi 2  

βDS ¼ π ηi Rci þ η j Rcj  ui −uj : ð15Þ
where Ω is the drag force correction which corresponds to the ratio
between the force exerted by the fluid in a permeable aggregate and
that on an impermeable sphere [41], and u∗ is the settling velocity of a
permeable aggregate.
There are various expressions for η and Ω derived from Darcy's law
with the variation of boundary conditions [18,40,41–44]. However, for
high porosity aggregates, these expressions have been shown to be
less accurate than those derived from Brinkman's extension of Darcy's
law, due to their inability to describe the velocity distribution and the
viscous force in the boundary layer [18]. The Brinkman extension is
able to describe the appearance of stress due to the velocity distribution,
which is crucial in calculating the drag force exerted in a porous
material.
Based on the Brinkman's extension of Darcy's law, the expression
of Ω as a function of the dimensionless permeability (ξ) is derived as
follows:
 
2tanhξ
2ξ
1−
ξ
Ω¼  : ð16Þ
2 tanhξ
2ξ þ 3 1−
ξ
196 R.I. Jeldres et al. / Powder Technology 326 (2018) 190–207

Fig. 5. Collision frequency for permeable and impervious aggregates. Average particle radius: 150 nm, shear rate: 50 s−1 (filled symbols correspond to impervious aggregates) [38].

The fluid accumulation efficiency, η, is then: where C is a packing coefficient, which is generally assumed to be 1,
while da and dp are the aggregate and primary particle diameters,
d c respectively.
η ¼ 1− − 3 ; ð17Þ
ξ ξ One of the most used models to determine the rate of sedimentation
of permeable aggregates corresponds to the modified Stokes law, given
3 5 3 5 3 by:
where: ξ ¼ prffiffiKffi , d ¼ 3J ξ ð1− tanhξ
ξ
Þ; c ¼ − 1J ½ξ þ 6ξ − tanhξ
ξ
ð3ξ þ 6ξ Þ;
J ¼ 2ξ þ 3−3 tanhξ
2
. The permeability as represented by K can be  
ξ 2
ρp −ρl g ∙ ∅ ∙ di
calculated through different models. Li and Logan [45] showed that ui ¼ ð20Þ
18 ∙ μ ∙ Ω
the Brinkman model given by:

2
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi!
dp 3 3 8 Like Eq. (8), it can only be applied to dilute systems. However, there
K¼ 3þ − −3 ; ð18Þ are more complex models available which can be extended to other
72 1−ϕ 1−ϕ
conditions (see for instance [47–50]). Johnson et al. [31] and Tang and
Raper [51] reviewed the strengths and weaknesses of these models.
provides a realistic prediction of the permeability of aggregates within a
homogeneous porosity. More permeability models can be found in
2.4. Collision efficiency
Veerapaneni and Wiesner [36] and Thill et al. [18].
Vainshtein et al. [46] related the porosity, ϕ, and fractal dimension,
Collision efficiency, α, represents the probability that two aggregates
df, by:
remain attached when they collide, i.e. it corresponds to the ratio
 df −3 between the number of collisions ending in aggregation and the total
da number of collisions. Calculating its value is complex and it is instead
ϕ ¼ 1−C ; ð19Þ
dp often incorporated as a fitted parameter within models [23] or simply
assigned as a scalar value [3,52], but in these cases all the variables—that
102 determine the collision efficiency are then ignored. Alternatively,
101 2 min permeable empirically deduced mathematical models could be used because they
100 2 min impermeable incorporate some physical characteristic, such as size, interaction forces,
Collision frequency (cm /s), β

10-1 5 min permeable or polymer adsorption.

5 min impermeable Below are four different cases of addressing this parameter. The first
10-2
3

8 mi permeable
10-3 (Section 2.4.1) corresponds to an empirical equation proposed by
8 min impermeable
10-4 Kuster [53], useful for modelling porous aggregates. In Section 2.4.2
10-5 the interaction forces between surfaces are considered, useful when
10-6 the particles are submicron. Also, in Sections 2.4.3 and 2.4.4, the effect
10-7 of polymers is introduced, which in this case is very important towards
10-8 describing aggregation phenomena by polymeric bridging.
10-9
10-10 2.4.1. Equations based on Adler's trajectory analysis
10-11 As an example of models that use trajectory analysis, Kusters [53]
used the Adler approach [40,54] to model the path of a porous aggre-
10-12
10 15 20 25 30 gate, represented as an impermeable core surrounded by a permeable
Size interval, i shell (core-shell model). The semi-empirical function resulting from
this analysis estimated that collision efficiency is substantially lower
Fig. 6. Collision frequency factor as a function of aggregate size at three different when the aggregate size ratio is Rci/Rcj ≤ 0.1. In addition, the core-shell
flocculation times for impermeable and permeable aggregates [39]. model shows that α is markedly reduced for large and less permeable
 R.I. Jeldres et al. / Powder Technology 326 (2018) 190–207 197

structures. Later, Selomulya et al. [55] developed an analogous empirical Similar approaches were developed by Ahmad et al. [58], Atmuri et al.
model: [59] and Liang et al. [60].

2   2 3 2.4.3. Polymer adsorption

exp −x 1− ij
6 7 According to early interparticle bridging theory [61], the formation
α ij ¼ 6
4
7α max ;
5 ð21Þ of a bridge requires that an adsorbed polymer molecule on one particle
ði ∙ jÞy
interacts with bare surface on a second particle. The collision efficiency
is then proportional to the product of the surface area covered with
where x and y are fitted parameters and αmax represents the upper limit polymer adsorbed on the first particle and the uncovered fraction of
of αi, j, which is between 0 and 1. Fig. 7 shows the evolution of the the second:
collision efficiency according to Eq. (21), estimated using x = 0.1, y =
0.1, and αmax = 1. It can be seen that a higher value of α is achieved α ¼ θð1−θÞ; ð23Þ
for small aggregates of comparable size, consistent with the trend
proposed by the shell-core model. where θ is the degree of the surface covered. Swerin et al. [62] proposed
three models to determine collision efficiency, depending on the
2.4.2. Interaction forces flocculant used and the flocculation mechanism. For example, when
Surface chemistry and how this influences polymer adsorption is a flocculation is carried out by polymer bridging, the thickness of
subject of great importance, especially in relation to the behaviour of the adsorbed polymer layer affects the collision efficiency. Thus, the
sub-micron particles. Therefore, incorporating interaction force effects resulting expression includes an extension that improves the descrip-
on α implies a more accurate description of the process and potentially tion of collision efficiency, given by:
allows direct analysis of the impact of variables such as solution pH and
ionic strength. α ¼ 2θð1−θÞ þ að1− expð−bθÞÞ; ð24Þ
The relationship between total interaction energy and collision
efficiency, αi, j, can be obtained through the Fuchs stability ratio [56], where a and b are constants that can be determined experimentally.
which corresponds to the inverse of αi, j and is defined by:
2.4.4. Polymer depletion on collision efficiency
Z
  ∞
expðV T =kT Þ When the rates of aggregation and breakage are equalised, a steady-
W i; j ¼ Rci þ Rcj dS ð22Þ
Rci þRcj s2 state is established in which the aggregate size does not change
(Fig. 8a). This phenomenon describes very well coagulation processes
using soluble salts; however, in the case of polymer bridging floccula-
where VT corresponds to the interaction energy of two particles and S is
tion with high molecular weight polymers, the aggregate size reaches
the separation distance of their centres. The total energy of interaction is
a maximum from which it begins to decrease gradually until it
a function of the forces acting between surfaces, and includes van der
stabilizes, often at a much lower value (Fig. 8b). Heath and Koh [63]
Waals force, electrostatics, short range repulsion by hydration, long
indicated that this phenomenon might be produced by what they
range hydrophobic attraction, steric forces or bridging due to adsorbed
termed the ‘degradation’ of the polymer, although this included all pro-
polymer, etc. The nature and magnitude of the total energy of interac-
cesses which in effect lead to deactivation of surface-adsorbed polymer
tion depends on the characteristics of the medium (pH, ionic content,
for bridging, e.g. reconformation of polymer tails back onto the same
etc.), the surface chemistry of the solid-solvent-polymer system, and
particle's surface. Such degradation would have two effects: i) the
the type of flocculant used. A detailed description of surface forces can
collision efficiency would decrease in time, and ii) impact on the rate
be found in Israelachvili's work [57].
of breakage of aggregates (Eq. (31), discussed in more detail in
Somasundaran and Runkana [38] applied a modified DLVO theory,
Section 2.5.3). The decrease of collision frequency can be modelled by
including the effects of adsorbed polymer layers on attractive van der
exponential decay:
Waals force, bridging attraction and steric repulsion, to then analyse
hematite flocculation at different polyacrylic acid (PAA) concentrations.
α ¼ Ce−t=D ; ð25Þ

where C is the initial collision efficiency and D is a parameter that

describes the decay rate of efficiency over time. Eq. (25) should ideally
be written in terms of many other variables, such as shear rate, solids
fraction, particle size, flocculant characteristics, etc., however, it is
considered useful for its mathematical simplicity.
Selomulya et al. [55] showed that aggregates are compacted,
changing their structure during the aggregation process. They included
the evolution of the fractal dimension with time into the model, which
is discussed in Section 2.7.

2.5. Breakage rate

Breakage models are usually fitted to size distribution data to obtain

parameters for the breakage rate, since there is no theory to predict this
through first principles. Therefore the accepted way to describe the
breakage rate is through an exponential (Eq. (26)) or power law func-
tion (Eq. (28)) of the aggregate mass. However, for bridging floccula-
tion, the weakening of the aggregates resulting from the degradation
Fig. 7. The collision frequency, αi, j, was estimated using x = 0.1, y = 0.1 and αmax = 1 or deactivation of the polymer would have a significant impact on
[55]. their breakage rate (Eq. (30)).
198 R.I. Jeldres et al. / Powder Technology 326 (2018) 190–207

Fig. 8. Flocculation kinetic by a) coagulation (B is the parameter of Eq. (27)), and b) bridging flocculation (as function of shear rate) [55].

2.5.1. Exponential breakage model is the effective volumetric fraction that includes the porosity of the
The aggregate breakage rate (Si) can be determined with the semi- aggregate.
empirical function used by Kusters [53]: The degradation (or deactivation) of polymer flocculant is incorpo-
rated into the model with a decreasing collision efficiency (previously
 1=2  1=2  ε 
4 ε described), or by a higher breakage rate through a weaker aggregate.
Si ¼ exp − bi : ð26Þ
15π ν ε This last approach is shown here, with the breakage rate a function of
the solid surfaces covered with flocculant:
where εbi is the critical rate of energy dissipation that causes the break-
age, and its value can be estimated through [64]: a3 εa2 μ ∙ dagg;i
Si ¼ ; ð30Þ
θf
B
εbi ¼ ð27Þ
Rci where a3 is a fitted parameter, dagg is the aggregate diameter, and θf
is the solids surfaces covered with flocculant (mass/area) (Eq. (31)).
where B is a fitting parameter that represents the critical force needed
The value of θf is derived by measuring the flocculant dosage and the
for the breakage of aggregates. The magnitude of parameter B can be
particle surface area, the latter through techniques such as laser diffrac-
manipulated to allow a greater or lesser number of breakages in a
tion, to give:
range of size i, for a particular shear rate, and thus determine the
interval at which fragmentation begins. The breakage rate is a function mf
θf ¼ M ð1−ΘÞ; ð31Þ
that depends strongly on the shear rate, wherein G ¼ ðνε Þ1=2. In addition, As
the energy required to break an aggregate reduces when its size is
larger. where mf is the flocculant mass, As is the surface area of the solid, and
M is the mixing index, which for turbulent pipe flow can be defined as:
2.5.2. Power law breakage model pffiffifL
The power law empirical model was proposed by Pandya and M ¼ 1−e−a4 D ; ð32Þ
Spielman [65]:

Si ¼ P 1 ∙ G ∙ RPc;i2 ð28Þ

where P1 and P2 are fitted parameters.

The breakage rate (Si) increases with the shear rate and aggregate
size. Since it is known that fractal dimension can influence the breakage
rate (Fig. 9 [38]), with a higher fractal dimension leading to lower break-
age rates, this dependence should be included in future models.

2.5.3. Polymer depletion on aggregate breakage

As mentioned earlier, Heath et al. [11] considered that polymer
depletion from the fluid is one of the main causes for aggregates to
decrease in size over time. It is captured as part of their overall descrip-
tion of breakage irreversibility for polymer bridging flocculation in
mineral systems, given by:
 1 !
dΘ ϕeff 3
¼ a1 εa2 μ ∙ ϕ ∙ M 1−Θ ; ð29Þ
dt ϕ

where Θ is termed the degree of flocculant degradation, a1 and a2 are Fig. 9. Variation of the fragmentation rate constant (Si) with respect to the size of the
fitted parameters, ε is the rate of energy dissipation per unit mass, μ is aggregate and its fractal dimension (average particle radius: 150 nm, shear rate:
the suspension viscosity, ϕ is the volumetric fraction of solids, and ϕeff 100 s−1, fragmentation rate parameter, B = 5 μm) [38].
 R.I. Jeldres et al. / Powder Technology 326 (2018) 190–207 199

where a4 is a fitted parameter, f is the pipe friction factor ( f = 0.0791/ appears as a direct indicator of the structure of the aggregate [71].
Re1/4), L is the pipe length [m], and D is the pipe diameter [m]. For this reason, Thomas et al. [72] stated that knowledge of the
fractal dimension makes flocculation models more applicable to
2.6. Breakage distribution function real systems. Bushell et al. [73] described the main experimental
techniques used to determine the fractal dimension of particle
Apart from knowing the rate at which the aggregates are broken, it aggregates, with microscopy and light scattering the most frequently
is necessary to establish the size distribution acquired by aggregate applied.
fragments. For this, a breakage distribution function is used (Γi, j), Many researchers have therefore included a fractal dimension in
which describes the distribution of fragments in interval i arising from flocculation models of porous aggregates, where the general assump-
aggregate breakage n the interval j. There are many mathematical tion was that the structure remains constant throughout the process.
forms to represent this function [66], which can utilise: However, it has been experimentally observed that aggregate structure
is not constant and compaction occurs as flocculation proceeds. This
• binary distributions, where aggregates divide into two fragments of
phenomenon has been observed both in laminar [74] and turbulent
equal size;
regimes [22,55,75].
• ternary distributions, in which they are divided into three fragments;
The evolution of df can be expressed as a function of the following
or
parameters:
• normal breakage distributions, where the fragments are normally
distributed in smaller size ranges. dd f
¼ functionðaggregate size; particle size; aggregation rate
dt
Spicer and Pratsinis [22] claimed that the application of any distribu- =fragmentation; shear; interaction forces; additional factorsÞ
tion preserves the shape of the final aggregate size distribution,
although experiments by Flesch et al. [67] showed that the use of a Selomulya et al. [55] used the following empirical expression:
normal breakage distribution (Eq. (34)) could improve the prediction

 c2 
of steady-state size distributions. However, the most used breakage dd f d  
¼ c1 þ c3 AB d f ; max −d f ; ð35Þ
function corresponds to the binary distribution, where an aggregate dt d0
breaks into two pieces of equal mass. Although it is not the most realistic
description, it is the easiest to implement and does not include adjust- where d and d0 correspond to the aggregate and particle diameters,
able parameters. Erosion of small fragments has also been suggested respectively, c1 and c3 are fitted parameters, and dmax is the maximum
as a mechanism by which aggregate size is reduced [68], but this value of df obtained for a particular process. The coefficient AB is
mechanism is less popular and is not well supported by experimental used to represent the combined effect of formation and breakage
data. of aggregates, highlighting the impact of both mechanisms on
compaction.
• Binary breakage distribution function [69]:
iX
max iX
max
The binary breakage function is easy to implement, requires no addi- AB ¼ jterm i of aggregationjx jterm i of fragmentationj;
tional adjustment parameters and yet adequately predicts aggregate 1 1
sizes. It is simply defined by:
where:
Vj 2 3
Γi; j ¼ for j ¼ i þ 1 and Γi; j ¼ 0 for j ≠ i þ 1; ð33Þ X
i−2
Vi 1
6 2 j−iþ1 α i−1; j βi−1; j N i−1 N j þ α i−1; j βi−1; j N 2i−1 7
6 2 7
6 j¼1 7
jterm i of aggregationj ¼ abs6 7;
where max2 = i + 1 (from Eq. (2)), which is applied to the equation 6 X
i−1 X
max1 7
4 −N 2 αj−i
β N −N α β N 5
Vi = 2i − 1V0, Vj = 2iV0, and V0 is the volume of the primary particle. i i−1; j i−1; j j i i; j i; j j
j¼1 j¼1

• Normal breakage distribution function [70]: 2 3

X
max2
When normally distributed fragments are considered, the breakage jterm i of fragmentationj ¼ abs4−Si Ni þ Γ i; j S j N j 5;
j¼i
distribution function is given by:

Z  !
νj bi
1 ν−ν fa Bonanomi et al. [76] proposed a simplistic alternative model to
Γi; j ¼ pffiffiffiffiffiffiffiffiffiffiffiffi exp − dv; max2 ¼ max1; ð34Þ describe the restructuring of aggregates during flocculation through
νi bi−1 2πσ f 2σ 2f
the following empirical relationship:
where νfa is the mean volume of the breakage distribution, σf is the  
ν
dd f
standard deviation σ f ¼ λf , νf is the volume of the aggregate subjected ¼ γ d f ; max −d f ; ð36Þ
dt
to breakage, and λ is a variable that when it decreases, the breakage
distribution is broadened. where γ is a time constant and df, max is the maximum value of the fractal
Aggregate breakage has received little attention in comparison to dimension.
aggregation, although it is important towards obtaining fundamental
knowledge about: i) the size and number of daughter fragments, 2.8. Model solution
ii) whether breakage is due to fluid shear or collisional effects, iii) the
relationship between shear and breakage, and iv) breakage as a function The modelling of aggregation using population balance equations
of aggregate size. involves two steps. The first is to identify the main phenomena and
aggregation mechanisms, according to the physical conditions of the
2.7. Aggregate restructuring system (solid concentration, size distribution, type of flocculant, hydro-
dynamics), and therefore, make an adequate selection of mathematical
As previously mentioned, the formation of clusters in aggregation models that describe the process. The second step is the estimation
processes leads to self-similar structures, so that the fractal dimension of the model parameters, which is achieved with a conventional sum-
200 R.I. Jeldres et al. / Powder Technology 326 (2018) 190–207
of-squares minimisation, in this case between the modelled and exper-
imental mean aggregate diameters:
X
t¼t max  2
min ψ ¼ d½4; 3exp −d½4; 3model ;
t¼0
where d[4, 3] corresponds to the mean volumetric diameter, and is suit-
able for use when representing gravitational solid-liquid separation
processes:
 P
P adsorption densities with increasing solids concentrations in aggregat-
ni di di P n d4
3
mi di πρp =6
d½4; 3 ¼ ¼  P ¼ P i i3
M πρp =6 n i di
3
n i di
Ni corresponds to the number concentration of aggregates containing
2i−1 particles (#/cm3) in the range i, obtained from the PBE (Eq. (2)) and
di can be calculated by:
 
di ¼ 2ði−1Þ=d f d0 :
Usually, the maximum number of size intervals has been 30,
covering aggregates containing 1 to 229 particles [39].
The integration of Eq. (2) requires different inputs (e.g. primary
particle size, shear rate of the system, the number concentration of
primary particles, the aggregate fractal dimension at the beginning
and end of the process, among others), but these will depend on the
equations selected to describe the flocculation process.
3. Analysis of selected studies
Table 1 summarises the principal studies in this area, with the
“Implications” column drawing out some of the key messages of
broader significance. While there are features in common across some
of the studies, the table serves to not only highlight the wide range of
applications involving aggregation but also how the diversity of scales
(particle size, solids volume fraction, shear) leads to very different
modelling approaches. Certainly, there has been no single model that
can be readily applied across all scales.
If narrowed down to those studies involving predominantly turbu-
lent flow regimes as being of the greatest industrial relevance, it still
remains apparent that some level of specific customisation is required
in almost every case. Perhaps the single largest factor in this is the
impact of particle size, and Fig. 10 seeks to represent the distinct
responses to applied aggregation conditions of sub- and supra-micron
solids, the former dominated by colloids and the latter by high through-
put industrial feeds.
Variation in surface charge is one way in which interaction forces
manifest for particle behaviour, with a reduction in Zeta potential
through double layer compression (increasing ionic strength) or modi-
fication (pH adjustment, addition of multivalent cations or charged
polymers able to promote closer approach of particles and the onset of
aggregation). Such forces are weak and therefore only small aggregates
are observed under moderate or high shear, while their effect dimin-
ishes when particle sizes increase above a few micrometres.
As already discussed in Section 2.4.3, the fraction of surface coverage
can influence the collision efficiency according to Eq. (23) or refine-
ments of this relationship. Olsen et al. [89] took this further to also pro-
pose a generalized geometric model describing the collision efficiency
factor for hetero-aggregation, i.e. the probability of collisions resulting
in aggregation for systems comprised of two oppositely charged species.
Such models all revolve around 50% available surface (or values close
to this after taking into account geometrically excluded surface)
representing an optimum for aggregation. This is certainly the case for
sub-micron particles, with which the possibility of over-dosing of poly-
mer being detrimental to the extent and rate of aggregation is well
known. It is very rare for such high surface coverage to be approached
for larger particles, with very large aggregate sizes generally achieved
at quite low coverages [90]. As a consequence, increases in dosage will
normally favour aggregation.
ð37Þ
The possible exception to this is when polymers are dosed in a
staged or step-wise manner. Required polymer dosages can be greatly
reduced in this manner, with the formation of small aggregates after a
first dosing stage greatly reducing the effective surface area seen by
polymer in the next stage. The concept of excluded surface area was
used by Chaplain et al. [91] to explain observed reduction in plateau
ð38Þ ing systems, and the contribution of aggregation to effective surface
coverage has been theoretically modelled [92].
The very different time-scales observed for aggregation of sub- and
supra-micron particles is largely a reflection of the typical solids volume
fractions and applied shear regimes. However, there is also a contribu-
tion from the mechanism by which aggregation is promoted. For
processes that solely involve surface charge modification and are there-
fore reversible, there are usually no detrimental effects from growth
ð39Þ
requiring longer reaction times. However, when polymer-bridging
between particles is the mechanism, the adsorbed polymer tails are
likely to reconform with time, such that they are no longer available
for bridging. Similarly, aggregate breakage will proceed through bridg-
ing polymer chain rupture, and the chain fragment has greatly reduced
activity. The consequence of both processes is a short timeframe in
which aggregation can take place, often only a matter of seconds, before
irreversible breakage becomes dominant. Some hybrid aggregation
mechanisms are proposed, i.e. bridging by polyelectrolytes onto oppo-
sitely charged particles, but aggregation with any long chain polymer
will have some degree of time-based irreversibility. The benefits of
staged polymer dosing discussed above can therefore be greatly dimin-
ished if the time between doses is too long.
Almost all studies of colloidal particle aggregation will conclude that
higher solid volume fractions give higher particle collision rates and,
given the long time-scales discussed above, thereby favour aggregation.
This can extend to larger particles at low solid volume fractions, but on
increasing the volume fraction a critical value will be reached at which
aggregation achieves an effective total volume that the suspension
viscosity is then influenced [11]. Beyond this value, the aggregate size
achieved will be reduced, even with increased flocculant dosages. The
response to solid volume fraction can only be captured in modelling
with knowledge (or a reliable estimate) of the fractal properties of the
aggregates.
4. Research opportunities
Since Smoluchowski proposed the first model to describe the
particle aggregation process through population balances, there has
been notable progress in mathematical, computational and experimen-
tal techniques of aggregate characterization. However, there are still
important challenges that impact upon the wider use of PBEs in
industrial aggregation applications. The following are some of the
main knowledge gaps.
4.1. Reduced reliance on empiricism
Many of the expressions used in models based on population
balances correspond to heuristic models that present some degree of
intuition or empiricism. These expressions need to be refined in such a
way as to capture the impact of variables relevant to the process, but
based on fundamental principles. As an example, the aggregate break-
age rate is usually described with an exponential law with respect to
aggregate size. However, properties that are critical to determining
this behaviour, such as the structure or resistance of aggregates, have
not been incorporated. Flesch et al. [67] showed that the selection of
different models to describe the breakage distribution function has a
 R.I. Jeldres et al. / Powder Technology 326 (2018) 190–207
Table 1
Summary of selected studies on flocculation modelling using population balance equations.
Authors Substrate system
Kusters [53] Aqueous suspensions of mono-sized spherical
polystyrene (~1 μm). NaCl used to induce
coagulation. Solid concentration: 10% v/v.
Han and Two particle sizes were presented; one set at a
Lawler [24] constant diameter of 2 μm, the other varied over
three orders of magnitude (dp = 0.1–100 μm).
Particle density 2.65 g/cm3; G 32 s−1.
Flesch et al. Monodisperse, spherical, polystyrene particles (d0
[67] 0.87 μm) was studied in a baffled, stirred tank (2.8 L).
Solid volume fraction: 1.4 × 10−5.
Aluminium sulphate hydrate (10 mg/L) as
coagulant.
Thill et al. [18] Latex particles (0.8 μm) in 0.25 L suspensions
(volume fraction 6 × 10−5).
Selomulya Experiments carried out in a 1 L baffled mixing
et al. [55] tank using an axial flow impeller. Mean diameter of
monodisperse latex particles was 380 nm.
Runkana et al. Hematite suspensions in the presence of salts (KCl)
[77] and two PAA flocculants with different molecular
weight (1.36 × 106 and 3.69 × 104 g/mol) at
different concentrations.
Bonanomi Experiments carried out in a 1 L baffled mixing
et al. [76] tank using an axial flow impeller. Mean diameter of
monodisperse latex particles was 380 nm.
Somasundaran Two experimental systems:
and a) Hematite (72 nm) flocculated at different PAA
Runkana concentrations; solid concentration: 2.4 × 1016
[78] particles/m3;
b) Polystyrene latex (330 nm) flocculated with
cationic polyacrylamide; solid concentration:
0.05% w/w.
Ding et al. [79] Activated sludge (~1 μm). Aggregation through
presence of polymeric material in the sludge
sample, with no external agent added to the
system to promote aggregation.
Heath et al. Calcium carbonate (calcite) of various grades,
[11] particle size (2.36–24.3 μm) and solid fractions
(0.0123–0.0615 v/v), flocculated with a
commercial 30% anionic high-molecular weight
polymer in turbulent pipe flow.
Runkana et al. Polystyrene latex particles (165 nm) flocculated by
[80] high molecular weight (16 × 106 g/mol) cationic
quaternary ammonium polyacrylamide; solid
concentration 0.05% w/w.
Approach
Semi-empirical model to determine collision
efficiency of permeable aggregates, using Adler's
[40,54] description of aggregate trajectories,
modelling permeable shells-impermeable cores.
A curvilinear approach is presented as a set of
corrections to the rectilinear collision frequency
function for each mechanism. The effects of
hydrodynamic interactions and van der Waals
attraction were included, leading to a more realistic
curvilinear view of collisions.
Effect of aggregate structure on shear-induced
coagulation and fragmentation modelled by a
rectilinear approach to the orthokinetic formation of
impervious and fractal aggregates in laminar and
turbulent regimes. Collision efficiency accounted for
by Kusters et al. model [35]. Two fragment size
distribution functions (binary and normal) examined,
latter giving best agreement between theory and
experiment.
Numerical model considering df variation. The time
in which experiments were simulated was divided
into several time intervals in which aggregates
have a different df. A curvilinear approach considers
perikinetic, laminar orthokinetic and differential
settling formation of permeable aggregates.
Collision efficiency an adjustable parameter.
PBM demonstrates aggregate structure and possible
rearrangement influence on aggregate size
development towards steady-state. A rectilinear
approach used to consider both perikinetic and
turbulent orthokinetic formation of impervious
aggregates. An analogous empirical form to Kusters'
model of collision efficiency was proposed (Eq. (21)).
A rectilinear approach used in the absence of
applied shear, considers Brownian motion and
differential sedimentation of impervious and
fractal aggregates. Collision efficiency factor for
aggregates computed as reciprocal of Fuchs'
stability ratio between particles, incorporating
DLVO theory, where Hamaker expression describes
adsorbed polymer layers.
Model contains main features of coagulation in
shear flows, including aggregation, breakage, and
time evolution of cluster df, tuned to reproduce
literature experiments. A rectilinear approach used,
considering Brownian motion and turbulent
formation of impervious and fractal aggregates.
Evolution of df from simplified Selomulya et al. [55]
expression for collision efficiency.
Examined how adsorbed structure and polymer
conformation control stability and efficiency of
aggregation. Accounts for stirred suspension
aggregation/fragmentation, irregular structures
and surface forces. Collision rate from a rectilinear
approach in absence of applied shear, considering
Brownian motion and differential sedimentation of
impervious aggregates.
Modelled effect of different mixing conditions,
considering orthokinetic formation of impervious
spherical aggregates. Did not accurately capture full
experimental data, especially at long times.
PBM of flocculation in turbulent pipe flow,
accounting for fluid shear, flocculant dosage,
particle size and solid fraction. Rectilinear approach
through turbulent collision kernel of Saffman and
Turner [26] for impervious aggregates. Irreversible
aggregate size decrease from polymer chain
scission or deactivation. Aggregate porosity
through df estimated from size and settling velocity
measurements.
Extension of earlier work [77] incorporating a
collision efficiency model taking into account
bridging attraction or steric repulsion, van de
Waals attraction and electrical double layer
repulsion between particles.
201
Implications
Permeable shells-impermeable cores (shell-core)
model implies that the attachment probability
would be notably reduced for larger and less
permeable aggregates.
Concludes importance of velocity gradient (G)
apparently over-emphasized in the traditional view
of aggregation based on rectilinear collision
frequency functions. G is far less significant for a
curvilinear, heterodisperse view of flocculation.
The effect of bridging flocculation was not
considered, so it is expected the usefulness of this
model will be restricted to describing particle
aggregation by coagulation in dilute systems, both
in laminar and turbulent regimes.
Aggregate restructuring was found to be an important
mechanism in explaining aggregation kinetics.
It is expected that the model may be useful to
describe the flocculation processes with
sub-micron and larger solids at low solids fractions,
in the laminar regime.
Aggregate structure variation incorporated within
PBM by changes of df. Modelling showed the extent
of restructuring determined transition of average
aggregate size to steady-state. Therefore, this model
can describe both coagulation and flocculation
processes in dilute systems, within the isotropic
turbulent regime.
Model used to describe the flocculation process, but
does not include any mechanism to describe
aggregate size decreasing. Therefore, it is expected the
model can describe aggregation only by coagulation
in dilute system, without orthokinetic movement.
The model may be useful to describe bridging
flocculation processes in dilute systems under
isotropic turbulent conditions.
The main contribution was the incorporation of the
modified DLVO theory to estimate collision
efficiency. The model was validated with
experimental data where aggregate size increases
until reaching a steady-state. Therefore, it does not
describe bridging flocculation.
Aggregation rate constant (β0) proportional to shear
rate and ‘selection rate constant’ (S0) size-
independent, but size-dependent collision efficiency.
Models dilute system coagulation in laminar regime.
Increasing suspension viscosity with solid volume
fraction captured. Inclusion of these additional
terms lets PBM describe aggregation and breakage
kinetics for higher solid fractions in mineral
processing thickeners. Model usefully describes
flocculation at high solid concentration, under
isotropic turbulence.
(continued on next page)
202 R.I. Jeldres et al. / Powder Technology 326 (2018) 190–207

Table 1 (continued)

Authors Substrate system Approach Implications

Antunes et al. Precipitated calcium carbonate (0.5 μm)
[81] flocculated with cationic polyacrylamides (very
high molecular weight, medium charge density);
solids volume fraction 1 × 10−2.
Sang and Precipitated calcium carbonate (PCC) flocculated
Englezos [82] with two cationic flocculants of different charge
density (0.57 and 1.08 mequiv/g). The mean particle restructuring to account for decreasing sizes.
size was 2.71 μm. Solid concentration: 0.05 wt%.
Li [83] Activated sludge collected from a full-scale
wastewater treatment plant, diluted with filtered
effluent (0.45 μm).
Au and Leong Colloidal Silica 5020 (0.50 μm) aggregated with
[84] cationic polymer R-8140; solid concentration 20 wt%. fragmentation. Experiments and PBM showed pure
Jeldres et al. Two particle/reagent systems:
[39] i) Polystyrene (0.87 μm) and hydrated aluminium restructuring. A modified rectilinear approach
sulphate; volume fraction: 1.4 × 10−5.
ii) Precipitated calcium carbonate (0.5 μm) and
cationic polyacrylamide (high molecular weight,
medium charge); volume fraction: 1 × 10−2.
Nassar et al. Asphaltenes at fixed concentration of 1000 mg/L
[85] and with different Heptol mixtures. Commercial
silica, alumina, and magnetite nanoparticles used
as inhibitors of asphaltene growth; average size:
1.5 nm.
Vlieghe et al. Monodispersed latex (2.15 μm) destabilized with
[86] sodium chloride at a concentration of 1.3 mol/L;
solid volume fraction: 3.5 × 10−5.
Liang et al. Submicron-sized Pr-ZrSiO4 in the presence of
[60] polymer surfactant SDBS and potassium chloride at motion and differential sedimentation. Collision
different concentrations; particle size: 370 nm;
experiments were carried out at various solid
fractions.
Wang [87] Precipitated calcium carbonate (2.53 μm)
aggregated by cationic starches of low (0.26 meq/g) perikinetic and orthokinetic formation of
and high (1.01 meq/g) charge densities; solid
concentration 0.05 wt%.
Li et al. [88] Activated sludge aggregated with extracellular
polymeric substances (EPS). Particle size: 2.64 μm; collision efficiency) used to evaluate EPS/zeta
3
solid concentration: 0.1 kg/m .
Rectilinear approach to perikinetic and orthokinetic
formation of impervious and fractal aggregates.
Selomulya equation (Eq. (21)) used for collision
efficiency, modelling aggregation, breakage and
restructuring. Parameters related to flocculant
concentration/branching, aggregate structure.
Rectilinear approach to perikinetic and turbulent
orthokinetic formation of impervious aggregates, with
Selomulya equation (Eq. (21)) used for collision
efficiency. Related fitted parameters to experimental
conditions.
A rectilinear approach, considering orthokinetic
formation of impervious and fractal aggregates. An
analogous empirical form of the Kusters model for
collision efficiency was used, as proposed by
Selomulya et al. [55].
Investigated high shear impacts on aggregate
fragmentation was negligible and re-aggregation
occurred for high shear breakage.
PBM captures both aggregate permeability and
used, considering both perikinetic and orthokinetic
formation of fractal and permeable aggregates.
Aggregate restructuring was examined by
evolution of df using an empirical equation that
needs initial and final df.
A rectilinear approach was used, considering
perikinetic and orthokinetic aggregation, with
porosity treated as a fitted parameter and the
collision efficiency assumed to equal to 1. A term
was introduced into the PBM model that represents
aggregate inhibition.
PBM adapted to describe aggregation and breakage
of fractal aggregates under varying turbulent
conditions. A rectilinear approach was used,
considering orthokinetic formation of impervious
aggregates. The collision efficiency was described
by Selomulya equation (Eq. (21)).
A rectilinear approach applied, considering Brownian
efficiency estimated with Fuchs theory of stability as a
function of interaction forces between particles. PBM
by Runkana et al. [77] modified, introducing volume
mean aggregate size, leading to a decrease of collision
efficiency on aggregation.
A rectilinear approach employed, considering both
impervious aggregates. Collision efficiency
described by the Selomulya equation (Eq. (21)).
Model describes size decreasing by aggregate
compactness, quantified through mass df.
Balance factor (ratio breakage rate coefficient to
potential effects on aggregation and breakage.
Rectilinear approach for orthokinetic formation of
impervious aggregates, collision efficiency a fitted
parameter.
The model developed in this work can be efficiently
applied to describe flocculation processes in dilute
systems under the turbulent regime.
Higher charge density starch gave lower collision
efficiency, lower restructuring rate, and a higher
energy dissipation rate for aggregate break-up.
Model describes both coagulation and flocculation in
dilute systems under an isotropic turbulent regime.
A normal breakage distribution was used, and it
does not include any mechanism to describe the
aggregate size decreasing. The model, therefore,
may best be applied to coagulation processes in
dilute systems.
At lower breakage speed, excellent agreement
between experimental and model size density
functions obtained, but agreement less satisfactory
a higher speed.
Aggregate properties (permeability, breakage rate,
collision rate) changed considerably over time during
restructuring. Collision frequency significantly
smaller for permeable than impervious aggregates.
Model may be useful to describe bridging flocculation
in dilute systems under anisotropic turbulence.
The model used to describe the behaviour of
asphaltene aggregation, considering adsorption of
asphaltenes onto different metal oxide nanoparticles
might be useful to describe coagulation processes in
dilute systems under a laminar flow regime.
Most influential of 4 parameters describe (i) power
by which mean shear rate is raised in breakage
kernel, and (ii) colliding aggregate size in the
collision efficiency model, highest for small
aggregates. Model describes coagulation in dilute,
turbulent systems with aggregate size plateau.
Attempts to describe solids fraction effects, but
limitations include assuming constant viscosity
(not influenced by solids fraction) and ignoring
aggregate fragmentation. As a consequence, the
model was only successful at low solids fractions.
The model can describe coagulation processes in
dilute systems without orthokinetic movement.
High charge starch gives lower collision efficiency,
higher energy dissipation rate and lower
restructure rate. Collision efficiency decreased with
increasing shear. Model describes dilute system
coagulation and flocculation under turbulence.
The balance factor obtained from PBM was linearly
correlated with either EPS content or decreasing
absolute values of zeta potential.

significant impact on steady-state size prediction, and the best descrip-
tion is generated when using a normal breakage distribution, although
the most commonly used version is simply a binary breakage distribu-
tion. Expressions describing the breakage rate and distribution function
have been given little attention, despite their importance being demon-
strated experimentally.
The works of Heath et al. [11,63] are an excellent example of this
problem, with the PB developed for bridging flocculation in mineral sys-
tems being successfully incorporated into CFD models used to optimise
full-scale feedwells in a variety of applications. However, this still relied
upon a fractal dimension and a set of fitted parameters that can only be
obtained from extensive measurement programs to quantify floccula-
tion kinetics in turbulent pipe flow; this has not always been considered
practical or economical, with the alternative to then use generic param-
eters from previously studied systems. The incorporation of other
fundamental relationships would not necessarily eliminate the need
for some degree of experimental customisation, but may at least greatly
reduce the scale and complexity of such measures.
4.2. Interaction forces
The interaction forces between particles is a field that has been nota-
bly developed from the theoretical and experimental point-of-view [57]
and is of dominant importance in the phenomena of submicron particle
aggregation [93]. Nonetheless, there are very few works where the
 R.I. Jeldres et al. / Powder Technology 326 (2018) 190–207 203

Fig. 10. How key processes and responses that must be captured in aggregation modelling vary, depending upon the primary particle size.

effect of interaction forces on the aggregation processes of colloidal par-
ticles has been incorporated through population balances.
This has particular relevance in systems which are complex from the
colloidal point-of-view, but which appear in a significant number of
industrial cases, for example, the aggregation of fine particles in highly
saline media. While theory in this area is quite strong for a range of stan-
dard substrates (e.g. polymer latexes and sub-micron particle standards
with very narrow size distributions), there is a much more limited
understanding of the behaviour of real particle systems encountered
industrially. As discussed in Section 3, part of the problem is that these
systems typically extend to larger sizes at which colloidal behaviour
becomes problematic.
To make the problem even more complex, these feeds can possess
broad particle size distributions and complex mineralogies; for exam-
ple, tailings from hydrometallurgy can contain fine clays and other silt
phases along with coarse sand. The majority of the mass may be in the
coarse solids, but particle number will be dominated by the finer phases.
From the aggregation point-of-view, it can be acceptable to ignore the
coarser solids that may only have a “ballasting” effect on aggregation
(i.e. if coaggregated with fines, their main impact is to accelerate set-
tling). Sieliechi et al. [94] suggested such ballasting may in fact lead to
some aggregate compaction, although this may be more likely when
the fines are mainly sub-micron and the structures are of very low
density. A focus on how the fines are aggregated still remains the
most productive path to practically relevant outcomes.
4.3. High solids
Undoubtedly, a major goal regarding the description of real systems
is to improve how aggregation processes are captured at high concen-
trations of solids. However, few works can be applied under these
conditions [11] and there are many outstanding challenges in this
area. For example, particle sedimentation takes considerable impor-
tance in determining the collision frequency, but the models based on
204 R.I. Jeldres et al. / Powder Technology 326 (2018) 190–207
the Stokes law are only valid in diluted regime; therefore, appropriate
expressions must be considered to describe the sedimentation of
concentrated suspensions [95]. It should be noted that recent studies
have shown advances in sedimentation in non-Newtonian fluids,
wherein Betancourt et al. [96] presented an explicit theoretical equation
to calculate terminal settling velocities of spherical particles. The
equation was based on an extension of the theory of boundary layer
over a sphere for the flow of a non-Newtonian fluid.
The rheological behaviour of flocculated suspensions is expected to
be complex and far from that of a Newtonian fluid. This implies that
classical properties such as the viscosity will be a function of the flow
regime experienced by the system, and it becomes a major challenge
to understanding the impact of the rheological properties on key
aspects of the aggregation process, such as the orthokinetic collision
frequency.
Perhaps the most fertile area for development in relation to aggrega-
tion at high solids is the direct in-pipe addition of polymer solutions at
very high dosages prior to tailings deposition, which under optimised
conditions leads to rapid release of liquor to then achieve solids concen-
trations well above what can be achieved from gravity thickening [97].
While such effects can be readily captured experimentally, at this
stage there is no fundamental basis to adequately describe the process.
High surface coverage by the polymer is likely to lower collision effi-
ciencies (i.e. fewer sites for adsorption), and while high solids volume
fractions will increase collision rate, they will increase shear-induced
breakage of aggregates. Increased viscosity as a consequence of the
high solids should also slow aggregation rates. The closer proximity of
particles/aggregates and the high dosages may contribute to forming a
denser core aggregate structure, but there is still limited physical quan-
tification of this. An increased understanding of the process could lead
to the rapid release of liquor being achieved at lower polymer dosages,
but full modelling may be complicated by the fact that the effect only
manifests on deposition (i.e. the treated suspension is discharged from
a pipe onto a slightly sloping surface.
4.4. CFD for aggregation prediction in 3D
The majority of the studies described here make use of PBEs to
describe aggregation in either batch reactors where the applied shear
is averaged across the vessel, or in continuous plug flow through a reac-
tor. The incorporation of PBEs within computational fluid dynamics
(CFD) models that describe industrial aggregation processes represents
a potentially powerful tool for full-scale unit design and process optimi-
sation, with the ability to link predictions of aggregation with the full 3D
hydrodynamics then capturing the impacts of shear history and solids
volume fraction.
The polymer-bridging flocculation process in gravity thickener
feedwells as used in mineral processing is known as a successful exam-
ple of CFD predictions for aggregate size [12,14,98]. While application of
CFD to lower solids wastewater processes significantly predate those in
the minerals domain, there are fewer examples of the incorporation of
aggregation modelling into CFD, although PBEs have been used with
CFD to describe other wastewater processes, e.g. bubble coalescence
[99]. This may to a large extent reflect quite different priorities in
terms of optimising performance – wastewater applications are typified
by slow settling velocities and the requirement for high clarity, and as a
consequence, the hydrodynamic information provided by more basic
CFD is often sufficient to both identify short-circuiting flows or currents
that are liable to entrain solids to an overflow, and then explore designs
that address this. Lee and Molz [100] did use CFD to model turbulent
flocculation and sedimentation processes in a low solids sediment
retention pond, tracking aggregate size evolution and mass balances
in a simplified two-dimensional system, but even this work gave greater
emphasis to the mass predictions.
The mineral processing applications cited above all seek to maximise
predicted aggregate size as an indication of optimised performance.
None of these publications provide adequate detail to properly evaluate
the PBEs used or their implementation. In the case of Nguyen et al.
[12], a number of feedwell designs were considered under different
operating conditions (volumetric flow rate, solids fraction) for calcite
flocculation using the model and parameters as described by Heath
et al. [11], although it is known that they had to seek modifications
to the commercial CFD code. Their main observation was that the
predicted hydrodynamics greatly influenced the residence time under
shear and the extent of solids dilution prior to the onset of aggregation.
The impact of residence time can clearly be seen in the predicted
streamlines and aggregate sizes within two common feedwell designs
under typical operating conditions. Fig. 11a shows that in a simple
open feedwell with a single tangential feed entry, only part of the
unflocculated feed (blue spheres) has sufficient momentum to travel
around the inner feedwell wall to adequately contact with the flocculant
solution at the downstream sparge location to then achieve a significant
aggregate size (orange-red spheres, N 400 μm). Much of the feed instead
starts to sink on entry and is poorly contacted with flocculant, with the
short duration under shear inside the feedwell predicted to prevent any
appreciable aggregate growth. However, the addition of a shelf (width
~ 10% of the feedwell diameter) ensured the entire feed had a longer
residence time within the feedwell (Fig. 11b). Good mixing of particles
and flocculant is predicted in the swirling flow above the shelf, while
milder shear conditions below the shelf favoured aggregate growth,
and far more consistent aggregation is expected at the discharge plane.
Most fine particle feed streams that enter mineral processing
feedwells are solids dilution deficient, particularly in tailings applica-
tions, i.e. aggregate sizes (and thereby settling rates) will be limited
without solids dilution prior to flocculation. With solids volume fraction
having such a significnat impact on aggregate size, it can be seen that
the CFD predictions can provide valuable information, even when not
fully customised to the industrial feed in question. Tanguay et al. [14]
took the work of Nguyen et al. [12] further, focusing on just the open-
with-shelf feedwell design but considering aggregation with two
polymer flocculants for which distinct sets of PBE parameters had
been obtained [101]. One was a conventional acrylamide/acrylate
copolymer (BASF Magnafloc® 336), while the other also contained an
alternative anionic functionality (BASF Rheomax® DR 1050). The
calculation of the particle trajectories from CFD predicted flow fields
was based on mass-conserving fluxes. The polymer giving a higher
fractal dimension (Rheomax® DR 1050) was known to produce larger
aggregate sizes at a higher solids volume fraction, and the correspond-
ing CFD predictions (Table 2) indicated that this could greatly increase
mass throughput, but only at high feed concentrations and higher floc-
culant dosages. Once again, this highlighted that detailed customisation
of the model to give the most accurate quantification of aggregate sizes
may only be necessary in applications with high demands on the extent
of flocculation or when the system is very sensitive to the solids volume
fraction.
Nguyen et al. [12] and Tanguay et al. [14] made use of the classes
methods for the solution of PBEs, which involves the discretisation of
the size domain into discrete intervals or classes. With 30 or more size
classes to then solve a number of equations across a vessel mesh that
may contain N105 elements, model convergence times can be measured
in weeks, even with powerful computing resources. Ruan et al. [98]
instead applies the method of moments (MOM), which involves
converting the PBE to equations in terms of the moments of the number
density, thereby achieving better computational efficiency. While Ruan
et al. [98] provide only limited detail, the implementation of the MOM in
CFD for modelling aggregation and breakage in a Taylor-Couette vessel
has been described in detail [102,103], while the perfomance of various
forms of MOM implemenation have been compared [104]. Vlieghe et al.
[86] used the MOM to implement an aggregation-breakage model that
accounts for fractal aggregate structures in a way that changes in the
hydrodynamic conditions do not involve adjusting the model parame-
ters, the moments being raw properties of the distribution that contain
 R.I. Jeldres et al. / Powder Technology 326 (2018) 190–207 205

Fig. 11. Streamlines and aggregate size as predicted by CFD for calcite flocculation in (a) an open feedwell and (b) an open-with-shelf feedwell [12] (4 m feedwell diameter, 1000 m3 h−1
feed rate, 5% w/v calcite of median particle size ~24 μm, 20 g t−1 flocculant dosed 90° downstream of the feed entry). Images used with permission of The Southern African Institute of
Mining and Metallurgy.

more information about the aggregate population than characteristic
diameters. However, this was only applied at laboratory scale to a
simple latex system.
Falola et al. [105] noted the possibility that more than one distribu-
tion can have the same set of moments, and even when a higher number
of moments are used, the non-uniqueness of the moment inversion
cannot be ruled out without other information. They also refer to
limitations to models that only consider size independent growth, but
proposed refinements to address this. Given that the size distribution
is an essential output from modelling multi-phase flows that change
over time and space, approaches for reconstructing distribution from
moments are well known [106], but are themeselves often computa-
tionally expensive. Hutton et al. [107] have proposed an approach for
doing such recontructions more efficiently. Mazzei [108] also raised
concerns on the implementation of MOM in CFD in relation to multi-
phase flows and the failure to properly model diffusion in real space.
The relative merits of sacrificing knowledge of the finer details of the
aggregate growth and breakage processes in seeking computational
efficiency from a moments approach in CFD needs to be considered in
terms of the application in question. For low-solids coagulation-based
applications with quite simple responses to solids fraction and applied
shear, such as in the study of estuarine particulate dynamics [109], the
sacrifice is likely to be justifiable. However, this is much less likely in
polymer-bridging flocculation applications where aggregation is very
sensitive to the applied conditions, and in particular when the demands
on the aggregation process (settling rate or fines capture) are high.

Table 2 5. Summary
Theoretical maximum throughput of solids through an open-with shelf feedwell for calcite
flocculated with two different flocculants, based on CFD predicted settling velocity and
Population balance modelling of aggregate growth and breakage in
volume fraction post-feedwell. Adapted from Tanguay et al. [14]. Feedwell dimensions
and conditions as in Fig. 11b. particle suspensions draws together fundamental understanding of
the many constitutive processes to provide predictions of the aggregate
Feed solids concentration Maximum throughput of solids (tonne/h)
properties (typically size) over time for selected experimental
(% w/w) BASF Magnafloc® 336 BASF Rheomax® DR 1050 conditions. Such modelling has been successful in diverse applications
5 141 168 that include a wide range of particle systems and distinct aggregation
10 182 323 mechanisms. While common features and functions are identified,
15 157 359 there are also many applications involving functions that are distinct
20 159 348
or requiring specific customisation, and certainly there is no one
206 R.I. Jeldres et al. / Powder Technology 326 (2018) 190–207
all-encompassing model that can be applied across systems with any
effect.
The fundamental understanding of interaction forces and associated
aggregation processes is strongest for colloidal solids at low solid
volume fractions, and hence the population balance equations for
sub-models in this domain are less likely to require fitted parameters.
Industrial aggregation processes will tend to involve larger or broader
particle size distributions, often under turbulence and where the solids
volume fraction can have a complex impact on performance. The use of
polymer bridging flocculants also adds a number of ill-defined compli-
cations. Population balance modelling can then require assumptions
and simplifications that can be less-than-satisfying to the modeller,
but still provide powerful practical outcomes. The potential to identify
functions that reduce empiricism in these models is large, thereby
greatly increasing the scope for applications.
Acknowledgements
This work was supported by INNOVA CORFO Projects CSIRO Chile
[10CEII-9007].
References
[1] R. Hogg, Flocculation and dewatering, Int. J. Miner. Process. 58 (2000) 233–236.
[2] J. Zhong, X. Sun, C. Wang, Treatment of oily wastewater produced from refinery
processes using flocculation and ceramic membrane filtration, Sep. Purif. Technol.
32 (2003) 93–98.
[3] S.R. Wickramasinghe, B. Han, S. Akeprathumchai, A. Jaganjac, X. Qian, Modeling
flocculation of biological cells, Powder Technol. 156 (2005) 146–153.
[4] M.G. Rasteiro, F.A.P. Garcia, P. Ferreira, A. Blanco, C. Negro, E. Antunes, The use of
LDS as a tool to evaluate flocculation mechanisms, Chem. Eng. Process. 47 (2008)
1323–1332.
[5] R.I. Jeldres, P.G. Toledo, F. Concha, A.D. Stickland, S.P. Usher, P.J. Scales, Impact of
seawater salts on the viscoelastic behavior of flocculated mineral suspensions,
Colloids Surf. A Physicochem. Eng. Asp. 461 (2014) 295–302.
[6] C.S. Lee, J. Robinson, M.F. Chong, A review on application of flocculants in
wastewater treatment, Process. Saf. Environ. Prot. 92 (2014) 489–508.
[7] G. Quezada, R.I. Jeldres, C. Goñi, P. Toledo, A. Stickland, P. Scales, Viscoelastic
behaviour of flocculated silica sediments in concentrated monovalent chloride
salt solutions, Miner. Eng. 110 (2017) 131–138.
[8] C. Goñi, R.I. Jeldres, P.G. Toledo, A.D. Stickland, P.J. Scales, A non-linear viscoelastic
model for sediments flocculated in the presence of seawater salts, Colloids Surf. A
Physicochem. Eng. Asp. 482 (2015) 500–506.
[9] J. Gregory, Monitoring particle aggregation processes, Adv. Colloid Interf. Sci. 147-
148 (2009) 109–123.
[10] L. Liang, Y. Peng, J. Tan, G. Xie, A review of the modern characterization techniques
for flocs in mineral processing, Min. Eng. 84 (2015) 130–144.
[11] A.R. Heath, P.A. Bahri, P.D. Fawell, J.B. Farrow, Polymer flocculation of calcite:
population balance model, AICHE J. 52 (2006) 1641–1653.
[12] T.V. Nguyen, J.B. Farrow, J. Smith, P.D. Fawell, Design and development of a novel
thickener feedwell using computational fluid dynamics, J. S. Afr. I. Min. Metall.
112 (2012) 939–948.
[13] M.E. Gheshlaghi, A.S. Goharrizi, A.A. Shahrivar, Simulation of a semi-industrial pilot
plant thickener using CFD approach, Int. J. Min. Sci. Technol. 23 (2013) 63–68.
[14] M. Tanguay, P. Fawell, S. Adkins, Modelling the impact of two different flocculants on
the performance of a thickener feedwell, Appl. Math. Model. 38 (2014) 4262–4276.
[15] D.W. Stephens, P.D. Fawell, Surrogate based sensitivity analysis of process equip-
ment, Appl. Math. Model. 35 (2011) 1676–1687.
[16] M. Smoluchowski, Versuch einer mathematischen theorie der koagulations-kinetic
kolloider losungen, Z. Phys. Chem. 92 (1917) 129–142.
[17] M.J. Hounslow, R.L. Ryall, V.R. Marshall, A discretized population balance for nucle-
ation, growth, and aggregation, AICHE J. 34 (1988) 1821–1832.
[18] A. Thill, S. Moustier, J. Aziz, M.R. Wiesner, J.Y. Bottero, Flocs restructuring during ag-
gregation: experimental evidence and numerical simulation, J. Colloid Interface Sci.
243 (2001) 171–182.
[19] C. Coufort, D. Bouyer, A. Liné, B. Haut, Modelling of flocculation using a population
balance equation, Chem. Eng. Process. 46 (2007) 1264–1273.
[20] D. Ramkrishna, The status of population balances, Rev. Chem. Eng. 3 (1985) 49–95.
[21] L.A. Austin, Review: introduction to the mathematical description of grinding as a
rate process, Powder Technol. 5 (1971) 1–17.
[22] P.T. Spicer, S.E. Pratsinis, Coagulation and fragmentation: universal steady-state
particle-size distribution, AICHE J. 42 (1996) 1612–1620.
[23] C.A. Biggs, P.A. Lant, Modelling activated sludge flocculation using population bal-
ances, Powder Technol. 124 (2002) 201–211.
[24] M. Han, D.F. Lawler, The (Relative) Insignificance of G in Flocculation, vol. 84,
A.W.W.A, 1992 79–91.
[25] B. Oyegbile, P. Ay, S. Narra, Flocculation Kinetics and Hydrodynamic Interactions in
Natural and Engineered Flow Systems: A Review, vol. 21, E.E.R, 2016 1–14.
[26] P.G. Saffman, J.S. Turner, On the collision of drops in turbulent clouds, J. Fluid Mech.
1 (1956) 16–30.
[27] F. Pedocchi, I. Piedra-Cueva, Camp and Stein's velocity gradient formalization, J. En-
viron. Eng. 131 (2005) 1369–1376.
[28] C.J. Meyer, D.A. Deglon, Particle collision modeling - a review, Min. Eng. 24 (2011)
719–730.
[29] J.N. Tilton, in: R.H. Perry, D.W. Green (Eds.), Perry's chemical engineer's handbook,
7th edMcGraw-Hill, New York 1997, pp. 634–636.
[30] T. Camp, P. Stein, Velocity gradients and internal work in fluid motion, J. Boston.
Soc. Civ. Eng. 30 (1943) 219–237.
[31] C.P. Johnson, X. Li, B.E. Logan, Settling velocities of fractal aggregates, Environ. Sci.
Technol. 30 (1996) 1911–1918.
[32] A.S. Dukhin, S.S. Dukhin, P.J. Goetz, Gravity as a factor of aggregative stability and
coagulation, Adv. Colloid Interf. Sci. 134–135 (2007) 35–71.
[33] B.B. Mandelbrot, Self-affine fractals and fractal dimension, Phys. Scr. 32 (1985)
257–260.
[34] M.R. Wiesner, Kinetics of aggregate formation in rapid mix, Water Res. 26 (1992)
379–387.
[35] K.A. Kusters, J.G. Wijers, D. Thoenes, Aggregation kinetics of small particles in agi-
tated vessels, Chem. Eng. Sci. 52 (1997) 107–121.
[36] A.Y. Menshutin, L.N. Shchur, Test of multiscaling in DLA model using an off-lattice
killing-free algorithm, Phys. Rev. E 73 (2006), 011407.
[37] S. Veerapaneni, M.R. Wiesner, Hydrodynamics of fractal aggregates with radially
varying permeability, J. Colloid Interface Sci. 177 (1996) 45–57.
[38] P. Somasundaran, V. Runkana, Modeling flocculation of colloidal mineral suspen-
sions using population balances, Int. J. Miner. Process. 72 (2003) 33–55.
[39] R.I. Jeldres, F. Concha, P.G. Toledo, Population balance modelling of particle floccu-
lation with attention to aggregate restructuring and permeability, Adv. Colloid
Interf. Sci. 224 (2015) 62–71.
[40] P.M. Adler, Interaction of unequal spheres, I. Hydrodynamic interaction: colloidal
forces, J. Colloid Interface Sci. 84 (1981) 461–474.
[41] G. Neale, N. Epstein, W. Nader, Creeping flow relative to permeable spheres, Chem.
Eng. Sci. 28 (1973) 1865–1874.
[42] D.N. Sutherland, C.T. Tan, Sedimentation of a porous sphere, Chem. Eng. Sci. 25
(1970) 1948–1950.
[43] S. Haber, R. Mauri, Boundary conditions for Darcy's flow through porous media, Int.
J. Multiphase Flow 9 (1983) 561–574.
[44] J.H. Masliyah, G. Neale, K. Malysa, T.G.G. Van De Venc, Creeping flow over a com-
posite sphere: solid core with porous shell, Chem. Eng. Sci. 42 (1987) 245–253.
[45] X. Li, B.E. Logan, Permeability of fractal aggregates, Water Res. 35 (2001) 3373–3380.
[46] P. Vainshtein, M. Shapiro, C. Gutfinger, Mobility of permeable aggregates: effects of
shape and porosity, J. Aerosol Sci. 35 (2004) 383–404.
[47] Y. Magara, S. Nambu, K. Utosawa, Biochemical and physical properties of an acti-
vated sludge on settling characteristics, Water Res. 10 (1975) 71–77.
[48] F. Concha, E.R. Almendra, Settling velocities of particulate systems, 1. Settling ve-
locities of individual spherical particles, Int. J. Miner. Process. 5 (1979) 349–367.
[49] Y. Adachi, M. Kamiko, Sedimentation of a polystyrene latex floc, Powder Technol.
78 (1993) 129–135.
[50] L. Gmachowski, Hydrodynamics of aggregated media, J. Colloid Interface Sci. 178
(1996) 80–86.
[51] P. Tang, J.A. Raper, Modelling the settling behaviour of fractal aggregates–a review,
Powder Technol. 123 (2002) 114–125.
[52] J. Zhang, X. Li, Modeling particle-size distribution dynamics in a flocculation
system, AICHE J. 49 (2003) 1870–1882.
[53] K.A. Kusters, The Influence of Turbulence on Aggregation of Small Particles in
Agitated Vessels(Ph.D. thesis, Eindhoven University of Technology, Netherlands)
1991.
[54] P.M. Adler, Streamlines in and around porous particles, J. Colloid Interface Sci. 81
(1981) 531–535.
[55] C. Selomulya, G. Bushell, R. Amal, T.D. Waite, Understanding the role of
restructuring in flocculation: the application of a population balance model,
Chem. Eng. Sci. 58 (2003) 327–338.
[56] N. Fuchs, Ueber die stabilitat und aufladung der aerosole, Z. Phys. 89 (1934)
736–743.
[57] J.N. Israelachvili, Intermolecular and Surface Forces, 3rd ed. Elsevier, London, 2011.
[58] A. Ahmad, M. Chong, S. Bhatia, Population balance model (PBM) for flocculation
process: simulation and experimental studies of palm oil mill effluent (POME)
pretreatment, Chem. Eng. J. 140 (2008) 86–100.
[59] A.K. Atmuri, M.A. Henson, S.R. Bhatia, A population balance equation model to pre-
dict regimes of controlled nanoparticle aggregation, Colloids Surf. A Physicochem.
Eng. Asp. 436 (2013) 325–332.
[60] L. Liang, Y.P.Z. Wang, Prediction of aggregation behavior of submicron-sized
particles of praseodymium-doped zirconium silicate in aqueous suspension by
population balance model, Particuology 25 (2016) 83–92.
[61] T.W. Healy, V.K. La Mer, The energetics of flocculation and redispersion by
polymers, J. Colloid Sci. 19 (1964) 323–332.
[62] A. Swerin, L. Ödberg, L. Wagberg, An extended model for the estimation of
flocculation efficiency factors in multicomponent flocculant systems, Colloids
Surf. A Physicochem. Eng. Asp. 113 (1996) 25–38.
[63] A.R. Heath, P.T.L. Koh, Combined Population Balance and CFD Modelling of Particle
Aggregation by Polymeric Flocculant, Third International Conference on CFD in the
Minerals and Process Industries, Melbourne, Australia, 2003 (339-334).
[64] A.A. Potanin, On the computer simulation of the deformation and breakup of
colloidal aggregates in shear flow, J. Colloid Interface Sci. 157 (1993) 399–410.
[65] J.D. Pandya, L.A. Spielman, Floc breakage in agitated suspensions: effect of agitation
rate, Chem. Eng. Sci. 38 (1983) 1983–1992.
 R.I. Jeldres et al. / Powder Technology 326 (2018) 190–207
[66] B. Han, S. Akeprathumchai, S.R. Wickramasinghe, Flocculation of biological cells:
experiment vs theory, AICHE J. 49 (2003) 1687–1701.
[67] J.C. Flesch, P.T. Spicer, S.E. Pratsinis, Laminar and turbulent shear-induced floccula-
tion of fractal aggregates, AICHE J. 45 (1999) 1114–1124.
[68] L.A. Glasgow, X. Liu, Response of aggregate structure to hydrodynamic stress,
AICHE J. 37 (1991) 1411–1414.
[69] W. Chen, R.R. Fisher, J.C. Berg, Simulation of particle size distribution in an
aggregation-breakup process, Chem. Eng. Sci. 45 (1990) 3003–3006.
[70] C. Coulaloglou, L. Tavlarides, Description of interaction processes in agitated liquid-
liquid dispersions, Chem. Eng. Sci. 32 (1977) 1289–1297.
[71] D.W. Shaefer, J.E. Martin, P. Wiltzius, D.S. Cannell, Fractal geometry of colloidal ag-
gregates, Phys. Rev. Lett. 52 (1984) 2371–2374.
[72] D.N. Thomas, S.J. Judd, N. Fawcett, Flocculation modelling: a review, Water Res. 33
(1999) 1579–1592.
[73] G.C. Bushell, Y.D. Yan, D. Woodfield, J. Raper, R. Amal, On techniques for the mea-
surement of the mass fractal dimension of aggregates, Adv. Colloid Interf. Sci. 95
(2002) 1–50.
[74] R.C. Sonntag, W.B. Russel, Structure and breakup of flocs subjected to fluid stresses:
I. Shear experiments, J. Colloid Interface Sci. 113 (1987) 399–413.
[75] C. Selomulya, R. Amal, G. Bushell, T.D. Waite, Evidence of shear rate dependence on
restructuring and breakup of latex aggregates, J. Colloid Interface Sci. 236 (2001)
67–77.
[76] E. Bonanomi, J. Sefcik, M. Morari, M. Morbidelli, Analysis and control of a turbulent
coagulator, Ind. Eng. Chem. Res. 43 (2004) 6112–6124.
[77] V. Runkana, P. Somasundaran, P.C. Kapur, Mathematical modeling of polymer-
induced flocculation by charge neutralization, J. Colloid Interface Sci. 270 (2004)
347–358.
[78] P. Somasundaran, V. Runkana, Investigation of the flocculation of colloidal suspen-
sions by controlling adsorbed layer microstructure and population balance model-
ling, Chem. Eng. Res. Des. 83 (2005) 905–914.
[79] A. Ding, M.J. Hounslow, C.A. Biggs, Population balance modelling of activated
sludge flocculation: investigating the size dependence of aggregation, breakage,
and collision efficiency, Chem. Eng. Sci. 61 (2006) 63–74.
[80] V. Runkana, P. Somasundaran, P.C. Kapur, A population balance model for flocculation
of colloidal suspensions by polymer bridging, Chem. Eng. Sci. 61 (2006) 182–191.
[81] E. Antunes, F.A.P. Garcia, P. Ferreira, A. Blanco, C. Negro, M.G. Rasteiro, Modelling
PCC flocculation by bridging mechanism using population balances: effect of poly-
mer characteristics on flocculation, Chem. Eng. Sci. 65 (2010) 3798–3807.
[82] Y. Sang, Englezos, Flocculation of precipitated calcium carbonate (PCC) by cationic
tapioca starch with different charge densities. II: population balance modelling,
Colloids Surf. A Physicochem. Eng. Asp. 414 (2012) 520–526.
[83] Z. Li, Modelling the activated sludge flocculation process using population balance
model (PBM), Adv. Mater. Res. 610-613 (2013) 1372–1376.
[84] P. Au, Y. Leong, Fragmentation of polymer-bridged silica flocs by high shear
impact: experiment and population balance modelling, Asia Pac. J. Chem. Eng. 10
(2015) 542–555.
[85] N.N. Nassar, S. Betancur, S. Acevedo, C.A. Franco, F.B. Cortés, Development of a
population balance model to describe the influence of shear and nanoparticles
on the aggregation and fragmentation of asphaltene aggregates, Ind. Eng. Chem.
Res. 54 (2015) 8201–8211.
[86] M. Vlieghe, C. Coufort-Saudejaud, A. Liné, C. Frances, QMOM-based population bal-
ance model involving a fractal dimension for the flocculation of latex particles,
Chem. Eng. Sci. 155 (2016) 65–82.
[87] X. Wang, Population balance modeling of precipitated calcium carbonate (PCC)
flocculation induced by cationic starches, Int. J. Miner. Process. 148 (2016) 9–14.
[88] Z. Li, P. Lu, D. Zhang, G. Chen, S. Zeng, Q. He, Population balance modeling of acti-
vated sludge flocculation: investigating the influence of extracellular polymeric
substances (EPS) content and zeta potential on flocculation dynamics, Sep. Purif.
Technol. 162 (2016) 91–100.
207
[89] A.F.G. Olsen, S. Biggs, G.J. Jameson, An improved collision efficiency model for
particle aggregation, J. Chem. Phys. 125 (2006) 1849061–1849069.
[90] A.T. Owen, P.D. Fawell, J.D. Swift, D.M. Labbett, F.A. Benn, J.B. Farrow, Using
turbulent pipe flow to study the factors affecting polymer bridging flocculation
of mineral systems, Int. J. Miner. Process. 87 (2008) 90–99.
[91] V. Chaplain, M.L. Janex, F. Lafuma, C. Graillat, R. Audebert, Coupling between
polymer adsorption and colloid particle aggregation, Colloid Polym. Sci. 273
(1995) 984–993.
[92] V.N. Kislenko, Mathematical model of polymer adsorption accompanied by floccu-
lation, J. Colloid Interface Sci. 226 (2000) 246–251.
[93] S. Farrokhpay, L. Filippov, Aggregation of nickel ore particles using polyacrylamide
homo and copolymers with different charge densities, Powder Technol. 318
(2017) 206–213.
[94] J. Sieliechi, B. Lartiges, S. Skali-Lami, J. Kayem, R. Kamga, Floc compaction during
ballasted aggregation, Water Res. 105 (2016) 361–369.
[95] F. Concha, Settling velocities of particulate systems, Kona Powder Part. J. 27 (2009)
18–37.
[96] F. Betancourt, F. Concha, L. Uribe, Settling velocities of particulate systems part 17:
settling velocities of individual spherical particles in Power-Law non-Newtonian
fluids, Int. J. Miner. Process. 143 (2015) 125–130.
[97] P. Wells, L. Charlebois, J. Diep, B. Moyls, O. Omotoso, A. Revington, M. Weiss, Inline
flocculation, in: R.J. Jewell, A.B. Fourie (Eds.), Paste and Thickeden Tailings - A
Guide, 3rd edAustralian Centre for Geomechanics, Nedlands, Western Australia
2015, pp. 231–241.
[98] Z. Ruan, C. Li, C. Shi, Numerical simulation of flocculation and settling behaviour of
whole-tailings particles in deep-cone thickener, J. Cent. South Univ. Technol. 23
(2016) 740–749.
[99] A. Karpinska, J. Bridgeman, CFD-aided modelling of activated sludge systems - a
critical review, Water Res. 88 (2016) 861–879.
[100] B. Lee, F. Molz, Numerical Simulation of Turbulence-induced Flocculation and
Sedimentation in a Flocculant-aided Sediment Retention Pond, vol. 19, E.E.R,
2014 165–174.
[101] A. Grabsch, P. Fawell, S. Adkins, A. Beveridge, How achieving a higher aggregate
density through flocculant selection can impact upon polymer-bridging floccula-
tion, Int. J. Miner. Process. 124 (2013) 83–94.
[102] D.L. Marchisio, R.D. Vigil, R.O. Fox, Implementation of the quadrature method of
moments in CFD codes for aggregation–breakage problems, Chem. Eng. Sci. 58
(2003) 3337–3351.
[103] D.L. Marchisio, M. Soos, J. Sefcik, M. Morbidelli, Role of turbulent shear rate distri-
bution in aggregation and breakage processes, AICHE J. 52 (2006) 158–173.
[104] L. Silva, R. Rodriguez, J. Mitre, P. Lage, Comparison of the accuracy and performance
of quadrature-based method for population balance problems with simultaneous
breakage and aggregation, Comput. Chem. Eng. 34 (2010) 286–297.
[105] A. Falola, A. Borissova, X. Wang, Extended method of moment for general
population balance models including size dependent growth rate, aggregation
and breakage kernels, Comput. Chem. Eng. 56 (2013) 1–11.
[106] V. John, I. Angelov, A.A. Öncül, D. Thévenin, Techniques for the reconstruction of a
distribution from a finite number of its moments, Chem. Eng. Sci. 62 (2007)
2890–2904.
[107] K. Hutton, N. Mitchell, P.J. Frawley, Particle size distribution reconstruction: the
moment surface method, Powder Technol. 222 (2012) 8–14.
[108] L. Mazzei, Limitations of quadrature-based moment methods for modeling inho-
mogeneous polydisperse fluidized powders, Chem. Eng. Sci. 66 (2011) 3628–3640.
[109] X. Shen, J.P.-Y. Maa, Numerical simulations of particle size distributions:
comparison with analytical solutions and kaolinite flocculation experiments,
Mar. Geol. 379 (2016) 84–99.