Open Archive Toulouse Archive Ouverte (OATAO)

OATAO is an open access repository that collects the work of Toulouse researchers and
makes it freely available over the web where possible.

This is an author-deposited version published in: http://oatao.univ-toulouse.fr/

Eprints ID: 9054

To link to this article: DOI:10.1016/j.minpro.2013.02.009

http://dx.doi.org/10.1016/j.minpro.2013.02.009

To cite this version : Kroll-Rabotin, Jean-Sébastien and Bourgeois, Florent

and Climent, Eric Physical analysis and modeling of the Falcon
concentrator for beneficiation of ultrafine particles. (2013) International
Journal of Mineral Processing, vol. 121 . pp. 39-50. ISSN 0301-7516

Any correspondence concerning this service should be sent to the repository

administrator: [email protected]
 Physical analysis and modeling of the Falcon concentrator for
beneficiation of ultrafine particles

Jean-Sébastien Kroll-Rabotina,c , Florent Bourgeoisa,c,∗, Éric Climentb,c

a University of Toulouse, INPT-UPS-CNRS: Laboratoire de Génie Chimique, Toulouse, France
b University of Toulouse, INPT-UPS-CNRS: Institut de Mécanique des Fluides, Toulouse, France
c Fédération de recherche FERMaT, CNRS FR 3089 - Toulouse, France

Abstract
A predictive model of the Falcon enhanced gravity separator has been derived from a physical analysis of its separation principle,
and validated against experimental data. After summarizing the previous works that led to this model and the hypotheses on
which they rely, the model is extended to cover a wide range of operating conditions and particle properties. The most significant
development presented here is the extension of the analytical law to concentrated suspensions, which makes it applicable to actual
plant operating conditions. Two examples of industrial use cases are described and studied by interrogation of the model: dredged
sediment waste reduction and coal recovery from fine tailings. Comparisons with empirical studies available in the literature show a
good agreement between model predictions and industrial data. The model is then used to identify separation efficiency limitations
as well as possible solutions to overcome them. These two examples serve to show how this predictive model can be used to
obtain valuable information to improve physical separation processes using a Falcon concentrator, or to evaluate Falcon separator’s
abilities for new applications.
Keywords: Gravity concentration, Physical separation, Modeling, Fine particle beneficiation

Introduction et al., 2005). In this case, the film flows over a retention zone
that has no fluidization counter-pressure. Both series are es-
Falcon concentrators are enhanced gravity separators (EGS) sentially semi-batch: “heavy” particles are recovered by inter-
consisting of a fast spinning bowl. The bowl is fed from its rupting operation and emptying the retention zone before a new
bottom and uses centrifugal force to drain the slurry in a thin operating cycle starts. The third design — C series — operates
flowing film at its wall. During operation, part of the trans- similarly to the UF series, but adds a slot in the retention zone
ported particles is retained inside the bowl, while the other part that is equipped with discharge valves with variable size aper-
flows out with the fluid. Due to high rotation rate, the centrifu- tures. In this way, the discharge rate in the retention zone can
gal force in the flowing film can be several orders of magni- be adjusted, which makes it possible to operate the bowl con-
tude greater than Earth’ attraction. Different mechanisms have tinuously (Abela, 1997; Honaker et al., 1996; McAlister and
been identified as playing significant roles in the separation tak- Armstrong, 1998).
ing place inside the bowl (Abela, 1997; Deveau, 2006; Laplante Commercial brochures published by Falcon indicate recov-
et al., 1994; Laplante and Nickoletopoulos, 1997; Laplante and ery abilities for C and UF series down to 10 and 3 µm respec-
Shu, 1993; McAlister and Armstrong, 1998), such as particle tively for targeted applications to heavy materials (tin, tantalum,
differential settling in the bottom region of the bowl, near the tungsten, chrome, cobalt and iron). The UF series are more lim-
film inlet (Zhao et al., 2006). ited in terms of capacity (up to 20 m3 /h for the bigger bowls)
Three bowl series differ by the way they trap particles once due to their design oriented towards ultrafine particle recovery.
particles have been classified by differential settling in the flow- This study focuses on these concentrator series because of their
ing film. Falcon SB series uses fluidized annular grooves up- potential application to fine dredged sediments. Nevertheless,
stream of the bowl outlet. The retention capacity of the bowl a number of conclusions drawn from physical analysis of these
can thus be set by adjusting the counter-pressure flow rate. Fal- concentrators remain valid for other series.
con UF series uses smooth bowls with a slight reduction in
diameter at the outlet. This lip creates a non-flowing region
whose volume varies with the bowl’s opening angle (Holtham 1. Separation modeling

1.1. Physical analysis and hypotheses

∗ Corresponding author
Email addresses: [email protected]
The assumptions on which our modeling of the smooth-wall
(Jean-Sébastien Kroll-Rabotin), [email protected] Falcon device relies, have already been described in previous
(Florent Bourgeois), [email protected] (Éric Climent) publications (Kroll-Rabotin, 2010; Kroll-Rabotin et al., 2010,
2011a,b). The fundamental hypothesis on which our model- • rotation as a solid body (which may contain up to almost
ing is based is that once particles enter the bowl retention zone, 10% error as stated in section 1.2);
they never leave it. Moreover, any classification upstream of • other neglected terms detailed in the model derivation.
the flowing film is neglected: it is assumed that the impeller
at the bottom of the bowl plays no active role in the separa- Among those three points, the first one seems to be the most
tion — ot is considered in fact that it homogenizes the feed significant, and is the reason why the calibration should be eval-
that enters the flowing film. Suspension is then considered ho- uated for each bowl shape (which differs slightly between bowl
mogeneous at the film inlet which implies that the prevalent sizes and Falcon series). The fact that the calibration constant’s
separation mechanism is particle transport in the flowing film order of magnitude is around unity confirms that it only con-
before particles reach the retention zone or the bowl outlet. Our tains corrective terms and does not hide any unaccounted sig-
predictive model was then built by solving a simplified particle nificant physical phenomenon.
transport equation analytically. It is worth mentioning that due to the balance between drag
The evolution of the flowing film thickness along the bowl and centrifugal forces, theoretical particulate Reynolds num-
wall is neglected, so that the flow is modeled by the combina- ber and Archimedes number are related. For particles whose
tion of a semi-parabolic profile in the streamwise direction and settling follows Stokes’ drag law, Re p = (3π)−1 Ar. Because
a solid body rotation. This simplification is discussed in section of that, it is commonly said that particulate Reynolds number
§1.2. governs separation in gravity separators. This is not effectively
Once the flow has been modeled, a transport model is added true in this case as it does not accurately account for the effects
to it in order to predict separation. To achieve this, numeri- of flow rate and particle size, as shown is this other form of
cal and analytical solutions of Lagrangian (Kroll-Rabotin et al., equation (1a):
2010), and Eulerian tracking of particles in the film were ob- !
tained (Kroll-Rabotin et al., 2011a). 1
C p = min Ar λ Q−1 r−1p ν R H bowl , 1 (1b)
The analytical solution is obtained by neglecting particle in- 3
teractions. Therefore, it is only reliable to predict separation
in dilute suspensions. Particle transport is then governed by the In this expression, a pseudo Reynolds number appears. It is
balance between drag and centrifugal forces acting on particles. based on particle size and on velocity Q/(RH) which has no di-
For this balance to be sufficient to account for the real physics, rect physical meaning, since Q is the feed flow rate and R × H
particle inertia must be neglected. Also, in order to get an an- is half the area of the bowl azimuthal section. This simple
alytical solution, the drag law must remain linear. These two overview of the model already shows that although such gravity
assumptions limit analytical predictions to low Stokes and low concentrators are used to perform separation according to par-
particulate Reynolds numbers. However, it is shown in section ticle densities, particle size is also playing a significant role in
2.2 that it does not affect the accuracy of the model. their performances (Coulter and Subasinghe, 2005).
Finally, the recovery of a given particle type — characterized
by its radius (r p ) and density (ρ p ) — in a smooth wall Falcon 1.2. Modeling of the flow profile
bowl is given by:
A major difficulty in the flow field computation is the free
4π surface of the film which yields a boundary condition whose po-
C p = min λ Q−1 ω2 (ρ p − ρ f ) r2p µ−1 sition is unknown until the problem is fully solved. It could be
9
! solved numerically by interface tracking or with the “Volume of
Fluid” method (Dijk et al., 2001) that solves the physics contin-
Rmin Rmax Hbowl , 1 (1a)
uously between the liquid and gas phases by weighing them ac-
cording to their respective local volume fractions. This method
where C p is the equation of the partition surface. In this equa- would make it possible to compute the flowing film thinning
tion Q and ω are the operating conditions (feed and bowl rota- along the bowl wall. However, such an approach is only re-
tion rates), ρ f and µ are the carrier fluid properties (density and quired when the film thickness undergoes significant variations.
dynamic viscosity). Rmin , Rmax and Hbowl define the bowl geom- Another approach that has been described thoroughly in the
etry (base radius, radius at the outlet and height) and λ is a cal- literature gives analytical solutions of the simplified Navier-
ibration constant. Experiments yielded a value of λ = 0.68 for Stokes equations in a rotating referential (Bruin, 1969; Janse
a laboratory scale Falcon L40 equipped with a UF bowl (Kroll- et al., 2000; Langrish et al., 2003; Makarytchev et al., 1998,
Rabotin et al., 2011b). The need of a calibration constant has al- 1997).
ready been detailed extensively in Kroll-Rabotin et al. (2011b): In a Falcon concentrator, the centrifugal force due to the bowl
it actually only reflects the simplifications we have included in spinning reaches several hundreds of times the Earth’s gravita-
the model derivation such as: tion (from 100 to 600 G depending on the series). For high
• bowl geometry (as it is simplified to the 3 parameters Froude numbers Makarytchev et al. (1997) give:
Rmin , Rmax and Hbowl , while actual bowls are made of a  1/3
few parts with different opening angles, include trapping 3νQ
h = 
 
(2)
2πω2 r2 sin β2

mechanisms, etc);
 h = 150

110
Y (µm)

90

60

30

0
0 1 2 −0.1 0 0.1 0.2 1 2 3 4 5
flow profile uX (m/s) wall normal uY (m/s) tangential uθ (m/s)
h = 100

80
Y (µm)

60

40

20

0
0 2 4 −0.1 0 0.1 0.2 2 4 6 8 10
flow profile uX (m/s) wall normal uY (m/s) tangential uθ (m/s)
– – modeled profile, — real solution
X = 5.95 cm, Q = 2 l/min, ω = {1000, 2000} rpm

Figure 1: Comparison of modeled flow profiles with DNS solutions

This simplified law expresses how film thickness changes as The only shear source in the azimuthal direction is Coriolis
a function of the operating and geometrical parameters. In par- acceleration. Indeed, fast rotation speed may induce significant
ticular, expression (2) yields the thickness ratio between the Coriolis effects that make the flow fully three-dimensional. The
bottom of the bowl (hi ) and the outlet (h f ): azimuthal velocity profile is also given by Makarytchev et al.
(1997) as:
!−2/3
hf Hbowl β
= 1+ 1  Y !
tan (3)
hi R0 2 uθ = ωr 1 + Ek−2 Pθ (6a)
4 h
 Y   Y 3 1  Y 4 Y
For a Falcon L40 with a UF bowl, this ratio is approximately 0.8
Pθ = − − 2 (6b)
which confirms the validity of the constant thickness assump- h h 4 h h
tion when compared to the shape ratio of the film azimuthal ν
Ek = (6c)
section. It only depends on geometrical properties and is ap- 2ωh sin β2
2
proximately the same for all Falcon series, even for industrial
scale bowls. As a result, variations of the film thickness are The non-dimensional parameter Ek is not explicitly identified
neglected in our modeling. in the expression given by Makarytchev et al. (1997). However,
In the streamwise direction (parallel to the bowl wall), for expressing it explicitly makes physical interpretation easier. Ek
high Froude numbers and high rotation rates, the analytical so- is the Ekman number which is the ratio of viscous effects rel-
lution given by Makarytchev et al. (1997) is a semi-parabolic ative to Coriolis inertial force. Its expression shows that when
profile: viscous effects prevail, wall rotation propagates throughout the
whole film and azimuthal velocity tends to rotation as a solid
Q Y 
body. When Coriolis effects become more and more important,
uX ≈ P (4a)
2πrh h they perturb the base flow so that it looks more and more like
 Y  3 Y Y2 ! profile Pθ . As this profile strongly depends on film thickness,
P = 2 − (4b) using constant thickness assumption in equations (6) gives in-
h 2 h h2
consistent profiles. It cannot be used when film thickness vari-
In the azimuthal plane, the only simplification that was added ation is not explicitly taken into account. However, direct nu-
to the analytical solution is the constant film thickness, so wall- merical simulation (DNS) results show that the error is smaller
normal fluid velocity is neglected: than 10% (cf. figure 1) between rotation as a solid body and real
azimuthal velocity for an opening angle of 20◦ . Consequently,
uY = 0 (5) azimuthal velocity is considered as the rotation of a solid body
 2. Coarse and dense particles
10
For a more precise consideration of the physics compared to
Fraction of total volume
in each size class (%)

what has been used in the model derivation, particle trajectory

8
may be calculated based on a more comprehensive force bal-
6
ance at each point along the trajectory. To account for drag
force, buoyancy, added mass force and the effect of pressure
4
gradient, force balance is expressed as follows (Kroll-Rabotin
et al., 2010) for the modeled velocity field described in equation
2 (4) that satisfies D~u/Dt = 0:

  d~v
0
1 10 ρp + ρ f CM =
Size (µm) dt
model: — upstream, - - overflow, · · · concentrate ~− 3
(ρ p − ρ f ) G ρ f C D k~v − ~uk ~v − ~u (7)


+ reconcilied data measured by laser sizing 8 rp
Q = 4.26 l/min, ω = 1170 tr/min
By neglecting the slip between the particles and the fluid in
Figure 2: Comparison between measurements with a Malvern Mastersizer 2000
the azimuthal direction, equation (7) can be solved in the az-
and model predictions imuthal plane only:

 d~v2D
~
  
uθ = ωr throughout the rest of this article. ρp + ρ f CM = (ρ p − ρ f ) ω2 r~er + G
dt
Finally, the fluid flow is modeled by combining the rotation 3
ρ f C D k~v2D − ~u2D k ~v2D − ~u2D (8)


as a solid body for uθ with a semi-parabolic profile for uX . This −
8 rp
flow gives good approximation of the real flow in a fast spinning
bowl with a small opening angle. 2.1. Numerical resolution of the Lagrangian tracking
To extend the scope of this modeling to dense and coarse par-
1.3. Validation of the separation model ticles, equation (8) must be solved for all Stokes and particulate
The analytical model (1a) has been validated against exper- Reynolds numbers. This means that particle inertia cannot be
imental data (Kroll-Rabotin et al., 2011b) using a Falcon L40 neglected, which turns the force balance used in the analytical
equipped with a UF smooth bowl. In order to relate recovery model into a partial differential equation. It also means that drag
to particle physical properties, particles with the same density force becomes a non linear function of the slip velocity of par-
were used in the experiments. Particle size distributions in the ticles in the fluid. The expression from Schiller and Naumann
feed, concentrate and tailing streams were measured by laser (1935) is used for the drag coefficient:
diffraction sizing. This technique allowed us to determine the
value of the calibration constant λ for the device and to ver- 24  
CD = 1 + 0.15Re0.687
p (9a)
ify the model accuracy, thus validating the aforementioned hy- Re p
potheses. k~u2D − ~v2D kr p
with Re p = 2 (9b)
Figure 2 shows all the laser sizing measurements and model ν
predictions for a given experimental run. The data are plotted in
the same manner as the sizing device provides them: one point As there is no analytical solution for equation (8), numerical
for each measured size class (size ranges vary exponentially so methods are required to solve this new model. The following re-
that data points are regularly spaced on the semi-logarithmic sults have been computed using an explicit time scheme solver
scaled curve). Vertical axis directly gives the probability asso- in which fluid velocity is expressed analytically at every point
ciated with a finite size range instead of the more usual prob- using equations (4) and (5).
ability density associated with a unique size. Predictions are
plotted with continuous lines and expressed in the same unit. 2.2. Physical limitations of the separation
The good agreement between experimental results and model According to the previous observations, separation based on
predictions validates the model. The only noticeable difference a density criterion with a Falcon concentrator would be more
is that measured size distributions show smoother variations efficient for high particulate Reynolds numbers (Kroll-Rabotin
than predictions in regions where predictions vary very sharply. et al., 2010) and high Stokes numbers (Kroll-Rabotin, 2010).
Actually, laser sizers are known to have limited ability to track The analytical model is not able to take these effects into ac-
sharp variations in measured distributions. Smoothing of mea- count, but in cases of low values for these numbers, for which
surements is a laser sizing effect which by itself is enough to it is accurate, it shows that the écart probable (E p ) is propor-
explain the difference between predictions and size measure- tional to the difference between the cut-point density and the
ments, while the whole set of other data points confirms the density of the fluid carrying the suspension. This comes from
accuracy of the physics accounted for in equation (1a). model (1a) which gives a linear relation between the recovery
 1

Écart probable, E p (g/cm3 )

0.1
1100 rpm, 1 l/min
1100 rpm, 5 l/min
1100 rpm, 10 l/min
1100 rpm, 30 l/min
1100 rpm, 50 l/min
1900 rpm, 1 l/min
1900 rpm, 5 l/min
1900 rpm, 10 l/min
1900 rpm, 30 l/min
1900 rpm, 50 l/min
2100 rpm, 1 l/min
2100 rpm, 5 l/min
2100 rpm, 10 l/min
2100 rpm, 30 l/min
2100 rpm, 50 l/min
0.01
0.1 1
Density difference between cut point and carrying fluid, ∆ρ (g/cm3 )

Figure 3: Relationship between the écart probable and the cut-point density for dilute suspensions

ratio and the density difference between the particles and the the forces balanced in the scaling law (Stokes drag law and cen-
carrying fluid. trifugal force), which yields an approximated theoretical value
The analytical solution of the model relies on a few hy- (Re p th ) that is a good estimation of the real particulate Reynolds
potheses which are only valid for low Stokes and particulate number:
Reynolds numbers. In order to quantify the differences between ω rpR
4 ρp
! 2 3
the numerical solution and the analytical model in terms of pre- Re p th = −1 (10)
dicted separation efficiency, figure 3 shows the relationship be- 9 ρf ν2
tween the écart probable (E p ) and the cut-point density (ρ50 ) Combining equations (1a) and (10), the rate of change of par-
obtained by numerical resolution. The analytical model always ticulate Reynolds number with bowl size, at iso-recovery, are:
gives a value of E p /ρ50 = 0.5 for all operating conditions. Fig-
ure 3 shows that for all operating conditions, even quite away Re p ω2 R
≈ R Hbowl −1


from the actual range of operating conditions achievable by ∝ 2 (11)
Falcon bowls, points remain aligned on the same straight line Cp ω Rmin Rmax Hbowl
E p /ρ50 ≈ 0.5. This demonstrates that for all particle distribu- Equation (11) shows that in order to achieve constant separa-
tions that can be treated with a Falcon concentrator, there is no tion, increasing bowl size (radius and/or length) makes the par-
subset of operating conditions for which the effects of particle ticulate Reynolds number decrease. So, as the effect of this
inertia and of non-linear drag, which have been neglected in number is negligible on separation with a laboratory scale Fal-
the analytical law (1a) play a significant role. The conclusion con L40, it is also negligible for all bigger Falcon concentra-
is that in the operation conditions we consider in this paper, tors. In the expression of the particulate Reynolds number (10),
there is no significant error when using correlations only valid the physical quantity whose exponent prevails is particle size,
for low Stokes and particulate Reynolds numbers, so the Stokes with a cubic law. The main input parameter that could make it
and particulate Reynolds numbers can always be considered as possible to leverage particulate Reynolds effect on density sep-
low, in the sense that asymptotic behaviour for low values gives aration efficiency is by changing the washability of the treated
good results for all operating conditions at which Falcon con- suspension. Unfortunately, in the ultrafine size range, the oper-
centrators operate. ating conditions achievable with a Falcon concentrator cannot
This conclusion, confirmed by experimental results using a benefit from this favourable effect.
small scale Falcon device, can a priori be extended to all Fal-
con concentrator series since, for a given separation, particu- 3. Concentrated suspensions
late Reynolds number decreases with increasing bowl radius.
Indeed, relatively low particulate Reynolds numbers (that is The fluid flow assumptions on which the modeling relies
Re p / 1) can be estimated using the analytical expression of have been validated. The model limitations, in terms of par-
ticulate Reynolds number and Stokes number, do not affect its
validity to predict the separation of ultrafine particles. Extend-
ing the scope of the model to dense suspensions is the last lim- h
organic (1.3)
itation that needs to be overcome in order to obtain a general φ = 0.1%
model of separation with a smooth bowl Falcon concentrator.
This has been done using Eulerian tracking of the concentra- 10
tion of each particle class (Kroll-Rabotin et al., 2011a). This 0%
70
method is briefly described hereafter. %

40
3.1. Particle Eulerian tracking model

10%
As the effect of the Stokes number is negligible, particles are
always sedimenting at their terminal settling velocity. To ac-
count for the presence of other particles, hindered terminal set- 0 Lbowl
tling velocities are calculated according to the semi-empirical h
law by Concha and Almendra (1979); Concha et al. (1992). 10% organic (1.3)
The application of a hindered settling law in numerically solved
100% 70% φ = 15%
40%
model for tracking of particle concentration already proved to
be satisfactory in the study by Kim and Klima (2004a,b) on hin-
dered settling columns and by Wang et al. (2006) on centrifuge
bowls with fluidization at their wall (such as Knelson separa-

13
tors). In order to account for the polydispersity of the treated

0%
suspension, the hindered settling model for monodisperse sus-
pensions is applied to each particle class, while concentration
effects on the medium sum the effects of all density and size
classes (Bürger et al., 2000; Concha and Bürger, 2002). 0 Lbowl
h
mineral (2.6)
 P 
~v p − ~u s = ~vC&A r p , ρ p , ρ s , µ, ω2~r, φ p (12a)
φ = 15%
~vC&A,p φ p
P
~u s = ~u − (12b)
1 − φp
P
 P 
ρs = 1 − φ p ρ f + φ p ρ p
P
(12c)
In this way, terminal settling velocity of each particle can be
computed in every point of the flowing film. By discretizing the
flowing film in the azimuthal plane, particle concentration in the
100%
70%

10%
40%

whole film was computed for each particle class by solving its
convection by the particle velocity fields using a finite volume 0 Lbowl
method. h
total volume concentration
3.2. Concentration fields and particle interactions φ = 15%
Figure 4 shows the concentration fields for different particle
70 60%
classes at different solid concentrations. The feed suspension %
washability used to plot these graphs is a sample dredged sed-
80%

iment consisting of two phases. The first one is organic with

a relative density of 1.292 and the other one is mineral with a
90%

relative density of 2.676. Their mass fractions are 11.5% and

88.5%. Their size distributions are modeled by fitted Rosin-
100%

Rammler distributions with d63.2 = 12.94 µm, k = 0.6369 for

the organic phase and d63.2 = 23.35 µm, k = 0.3285 for the 0 Lbowl
mineral phase respectively. Computations have been run for a volume fraction: 0 max
Falcon L40 equipped with a smooth UF bowl at a flow rate of — local concentration value compared to inlet (iso-lines)
5 l/min, a bowl rotation speed of 1200 rpm and solid concentra-
tion in the feed stream of 30 wt.%.
Figure 4: Volume fraction of a 23.5 µm particle from a sample sediment in
Particle interactions induce several phenomena:
the flowing film of a Falcon L40 with UF bowl operating at Q = 5 l/min and
• settling being hindered, particles settle slower and reach ω = 1200 rpm
the wall bowl later, which leads to a decrease of the recov-
ery to concentrate for any given particle class, provided
 100

41.6 µm

µm
31.3 µ

23.5 µ

µm
17.7

13.3

µm
Fraction of particles having reached the wall (%)

97.8 µm
130 µm

55.3 µm

10
80

µm
7.5
m
55.3 µ

µm
41.6

m
60

3µ
31.

6µm
5.
µm
.5
40 23

µm µm
.7 4.2
17

20
µm
3.2
.3 µ
m
13
µm
2.4

10 µ
m
µm
7.5
0
0 0.5 1 1.5 2

Distance from the inlet (X/Lbowl )

–– ρ p = 1.292 g/cm3 , — ρ p = 2.676 g/cm3

Figure 5: Cumulative distribution onto the bowl wall for each particle class

that all other operating parameters remain constant. This aries of the inlet, outlet and retention zone. Figure 5 shows
phenomenon appears clearly when comparing the two first how the flux of particles that reach the wall is distributed along
graphs in figure 4; the bowl wall, for the same suspension and operating condi-
• low density particles are concentrated inside the flowing tions that were used previously. It is easy to visualize the
film before they can reach the wall. As shown in the third rate of recovery of each particle class by reading the value of
graph of figure 4, this is due to the suspension density in the relative cumulative flux at the abscissa corresponding to
the film and to the back flow of dense and coarse particles the bowl length. Some particle classes completely settle be-
that settle faster, whose settling creates a fluid flow in the fore the bowl outlet and are totally recovered (case of particles
opposite direction for volume conservation. This opposes ρ p = 2.676 g/cm3 and r p = 13.3 µm), while, due to the local
the settling of lighter and smaller particles. Eventually, increase of concentration in the film, some particles do not set-
for the least dense particles in the suspension, the most tle at all before the outlet (case of particles ρ p = 1.292 g/cm3
concentrated region in a given particle class is not at the and r p = 10 µm). For sizes and densities in the vicinity of the
inlet but inside the flowing film. Once heavier particles cut-point values, the recovery has a value that is neither 0 nor
have all settled, lighter particles are able to settle and reach 100% at the abscissa corresponding to the bowl length. Thus,
the wall. partition function (C p ) can be read as the value of the curves
plotted in figure 5 at abscissa 1, for each particle class. In con-
For broad size distributions, these effects are noticeable on ditions such as the ones used for that plot, density separation is
every particle class, while from a general point of view, the almost perfect for 10 µm particles (bold lines on figure 5).
amount of solid in the film always decreases while particles set-
tle and get trapped (cf. last graph in figure 4). 3.3. Corrected analytical model
The recovery to concentrate of each particle class is calcu- In order to completely account for both previously identified
lated by considering mass fluxes that flow through the bound- particle interaction mechanisms, it is necessary to compute the
volume fractions of all particle classes anywhere inside the film. ond effect can be modeled by a coefficient that varies with the
However, for the sake of practical use of the model, it was found concentration in the freed stream. This coefficient appears to be
that their effect could be modeled in a simpler way by adding a linear function of the volume fraction of the solids, as shown
two parameters in the analytical model: one of these parameters in figure 7 which shows calculated values of this coefficient ob-
accounts for the effects of the local density of the suspension, tained by minimizing the error of the model predictions, as well
while the other one accounts for the global elongation of the as their linear regression. This hindered settling coefficient is
particle trajectories due to hindered settling. then approximately 1 − 1.6φ, where φ is the volume fraction of
As the resolution of the concentrated problem is done in a the solids in the feed stream.
completely different way than the previously described dilute Equation (1a) becomes:
one, the application of this model to dilute suspensions does
not give the exact same result as the analytical model (1a). The  4π
C p = min λ (1 − 1.6φ)
numerical resolution has not been fine tuned so that it converges 9
to the analytical model under dilute conditions. Only the trends

Q−1 ω2 (ρ p − ρ s ) r2p µ−1 Rmin Rmax Hbowl , 1 (13a)
from the numerical resolution are needed in order to extend
the analytical law. A multiplicative coefficient (of value 0.8) where density in the tailings (ρ s ) is calculated this way:
has been included in the analytical expression in order to make
the two approaches give the same results in dilute conditions,
ZZ
ρs = ρ f + φ ρ p − ρ f 1 − C p ffeed dr p dρ p



so that the effect of the physical parameters can be quantified (13b)
compared to this reference and then be included directly in the
handling of concentration effects into the analytical law. Again, Here, C p and ρ s are interdependent and an iterative resolution
it does not hide any physics, as confirmed by the model’s fit- is needed to solve this set of equations. The expression of ρ s
ting to all the trends that have already been described (particle depends on the feed suspension washability ( ffeed ), which meets
density and size, bowl rotation and flow rates). This correction common understanding that the partition function depends on
factor only accounts for the error due to the numerical solution the washability of the suspension being processed, when the
of the model and to the accuracy of its underlying models, and volume fraction (φ) becomes non-negligible.
ensures that the asymptotical trend of the numerical simulation Figure 8 compares separation predictions obtained by solv-
matches with the analytical law in dilute conditions. As a result, ing for the concentrations of all particle classes in the whole
it should not be included in the extended analytical law, since film against predictions from the corrected analytical law (13).
equation (1a) already fits experimental data. However, for com- The results of both approaches are in a very good agreement,
parisons shown in figures 6, 7 and 8, this factor has been applied which confirms that the two corrective terms that have been
to the analytical law as there is no way to add it in the numerical added to the analytical law (1a) are sufficient to capture the
model with certainty that it does not impact the physics. mechanisms induced by particle interactions that impact the
The effect of suspension density can be modeled using an ap- separation. This also means that the physical mechanisms af-
parent density in model (1a) that differs from that of the carrier fecting separation are generally the same for both dilute and
fluid. However, this apparent density is not the suspension den- concentrated conditions, since concentration effect does not
sity in the feed stream. Indeed, concentration decreases grad- change the form of equation (1b) but only adds new factors into
ually in the film while particles reach the bowl wall and get it.
trapped in the retention zone. Consequently, the density in the Finally, equation set (13) is a very satisfactory separation
feed stream is the maximum value of the apparent density in the model for smooth bowl Falcon concentrators. Its predictions
whole film, as shown in the last graph in figure 4). Conversely, can be used to evaluate Falcon’s efficiency for processing ultra-
according to the second graph in figure 4, it appears that low fine suspensions.
density particles are concentrated before they reach the wall.
Thus, they all quickly reach the bowl wall once denser parti-
cles have settled and the local suspension density has decreased. 4. Applications to industrial cases
Density in the overflow is the last suspension density in the film
Separation of ultrafine particles is an important challenge in
before the outlet, which means that this is also the last medium
a number of industrial and environmental applications. In this
density that particles will experience before leaving the bowl.
article, we turn our attention to the beneficiation of dredged sed-
As a result, particles that can settle in these conditions have
iments and fine coal tailings. The former application was in fact
already settled, while the others are still carried by the film.
the practical problem that supported the work that is presented
This critical density is thus the one that must be included in the
in the paper.
model. Figure 6 confirms this observation by providing com-
parisons between density in the tailings and optimal apparent
density yielded by fitting analytical model predictions against 4.1. Beneficiation of dredged sediments
concentrated medium simulation results. Every year, millions of tons of sediments are dredged from
When increasing the concentration, the distances before im- harbors and various water ways around the globe, hence benefi-
pact at the wall of all particle classes increase due to their set- ciation of dredged sediments has become a significant industrial
tling being hindered by particle-particle interactions. This sec- business and a major environmental issue. Dredged sediments
 1.2

)
Probability density
g/cm3
function ( µmwt.%
40 wt.%
Suspension density (g/cm3 )

mineral

organic
1.1

2.68 100
Den
sity 10
(g/ 1.29 1
cm 3 µm)
1 ) Size (

Figure 9: Washability of a sample of dredged sediment

0 5 10 15 20
Feed solid fraction (vol.%)
— density in overflow, - - - model fitting
Q = 2 l/min, ω = 1000 tr/min
can be considered as a two-phase material of fine sand and or-
ganic particles. In many instances, the sand fraction coarser
Figure 6: Comparison between density in overflow and apparent suspension than 10 µm is free of pollutants, which implies that the +10 µm
density obtained with least squares method sand fraction can potentially be beneficiated. Unfortunately,
current treatment plants are only able to concentrate sand parti-
cle larger than 60 to 80 µm with physical separation processes,
1
which may represent only half of the potentially valuable sand
particles. Given the tonnages involved, there is a very strong
incentive to identify robust technologies for concentrating sand
0.9
particles from 10 µm upwards. In this size range however, the
Correction factor

y=
1−
1.5 competing effects of particle size and density are significant,
97
3x despite the 1.3 s.g. difference between sand and organic parti-
0.8
cles.
Previous studies (Detzner, 1995) attempted to concentrate
dredged sediments using small diameter hydrocyclones. How-
0.7
ever, this solution suffered from important wear issues and still
could not achieve efficient separation below 63 µm. Achieving
0 5 10 15 20
the desired result implies both a size-based separation above
Feed solid fraction (vol.%)
10 µm and a density-based separation between 2.6 and 1.3 s.g.
— value of hindered settling coefficient, ··· linear fit
Q = 2 l/min, ω = 1000 tr/min Since our work has shown that the UF Falcon concentrator op-
erates a separation based on both particle size and particle den-
Figure 7: Least squares calculated correction factor to be included in analytical sity in this range, it seemed interesting to assess the potential of
law to account for separation dependency on feed solid fraction
the UF Falcon for this industrial application by interrogation of
our model (13). The washability of the used dredged sediments
is given in figure 9.
100
50 3020 15 µm 10 µm 7.5 µm As shown in equation (1) and detailed in section 2.2, the par-
5 µm ticle transport regime remains the same for all the conditions at
80
3 µm which a Falcon concentrator may operate. Although they weigh
differently on separation, the flow rate and the rotation rate can
60 be grouped into a single parameter Q−1 ω2 . Equation (13a) con-
C p (%)

firms that this remains valid under dilute and concentrated con-
40 ditions. Adjusting both the flow rate and the rotation rate makes
2 µm
it possible to reach a ratio of approximately 250 between the
20 maximum and minimum values of this unique parameter with a
1 µm Falcon L40. This permits processing a broad range of slurries
0.5 µm
0
1 10
with the same device.
Particle density (g/cm3 ) To simplify the use of this parameter, a reference operating
condition set is chosen with Q0 = 5 l/min and ω0 = 1200 rpm.
Suspension made of particles ranging in 1.05–8 g/cm3 and 0.5–50 µm
φ = 10%, Q = 3 l/min, ω = 1200 tr/min The behavior at other operating conditions is then studied rela-
— analytical law, + simulation result tive to this reference, as per the horizontal axis in figure 10.
Different set points were applied to a typical sediment washa-
Figure 8: Recovery according to particle density: comparison between simula-
bility whose valuable content, defined here as the sand fraction
tion results and extended analytical law (13)
in the 10 to 150 µm range, is approximately 40% (same as in
 section 3.2, also plotted in figure 9). The performance of the
100
UF Falcon can be investigated by varying the feed solid fraction
and the parameter that couples flow and rotation rates which
Mass recovered to concentrate
and concentrate grade (%)

80
φ was defined earlier. The results obtained for a Falcon L40 are
plotted in figure 10. It shows that even though concentration
60 effects improve separation efficiency, their contribution may be
overruled by the flow and rotation rates. This figure shows two
40 quantities that vary with the operating conditions: the mass re-
covery to concentrate and the amount of non-valuable material
20 in the concentrate. As expected from grade-recovery analysis,
φ
both of them increase with Q−1 ω2 , which means that the more
0 valuable material is recovered, the more non-valuable material
0.1 1 10
Q−1 ω2 is retained too, since all particles undergo the same physics,
0 ω0
Q−1 2 though their quantitative response is different.
— total mass ratio to concentrate As expected, figure 10 shows that the lower the recovery, the
––– non-valuable content in concentrate (organics and −10 µm minerals) cleaner the concentrate. The best result is obtained with the
−→ increasing feed concentration (φ = {0, 21, 38} wt.%)
combination of high flow rate and slow bowl rotation speed.
Figure 10: Impact of operating parameters on separation result for a typical Even in this case, the concentrate is still around 3.7 % of pre-
dredged sediment
sumably polluted material, that is minus 10 µm sand and or-
ganic particles. Depending on the nature of the pollution asso-
100 ciated with the sediments, such a grade may be acceptable for
beneficiation, however it comes at the cost of recovering only
Recovery of valuable content (%)

φ
90 18 % of the total mass, which corresponds to 44 % recovery of
the valuable material. This is not sufficient to justify processing
80 such a material in industrial cases, especially because these val-
ues rely on a high concentration which is hard to keep constant
70 in industrial processes.
Figure 11 summarizes the data presented in figure 10 (con-
60
tinuous lines) and more in the form of a usual grade-recovery
curve. The results show that for a sample dredged sediment, it
50
is hardly possible to obtain a concentrate with less than 5 % of
70 75 80 85 90 95 100 presumably polluted content. This is too high to consider this
Concentrate grade (%) product as a clean product for any realistic situation, especially
as this result is obtained at the best separation conditions and
— sample sediment washability (see figure 9)
––– sample sediment washability with finer organic particles
comes at the cost of recovering only half the non-polluted ma-
−→ increasing feed concentration (φ = {0, 21, 38} wt.%) terial which would only decrease the final amount of produced
Figure 11: Grade-recovery curves of different sediment washabilities at differ- waste by 30 %. Figure 12 shows that the separation curves are
ent operation conditions quite steep. Therefore, Falcon separators produce a really sharp
separation. The issue here is that the cut point dependency on
particle size does not match what is needed for beneficiating
100
dredged sediments. As the separation limitations mostly arise
Recovery to concentrate, C p (%)

from the particle size distribution, they cannot be solved by us-

80
ing multiple passes through Falcon concentrators.
Following the results which showed that the UF Falcon is
60
unable to meet the +10 µm objective with the natural dredged
sediment washability, the next course of action was to inves-
40 tigate whether changing the washability itself might give new
scope for beneficiation. This possibility was considered since
20 organic particles were found through experiments to be signif-
icantly more sensitive to attrition than sand particles, so that
0
1 10 100
intensive mixing for instance could reduce the size of organic
Particle size (µm)
particles selectively. This behavior, which is specific to the
material of interest, will increase the particle size differential
– – ρ = 1.29 g/cm3 , — ρ = 2.65 g/cm3 between sand and organic particles, thereby providing the UF
Falcon with more favorable conditions for separation.
Figure 12: Predicted partition curves of a sample sediment with a Falcon L40 The new size distribution of organic particles was modeled
operating at 0.18 Q−1 ω2 /(Q−1
0 ω0 ), feed solid fraction being 21 wt.%.
2
using a Rosin-Rammler distribution with d63.2 = 10 µm and ex-
 Table 1: Sample mass distribution of coal fines
1.2
density range (g/cm3 )
mass %
1.2 – 1.40 1.40 – 1.45 1.45 – 1.50 1.50 – 8.18
1.19
100 – 200 4.04 3.06 3.01 4.25
size range (µm)

80 – 100 4.29 3.23 3.03 5.09

Cost function
63 – 80 0.87 0.58 0.51 0.95 1.18
40 – 63 5.45 3.41 1.88 3.25
20 – 40 7.33 5.39 2.08 4.52
5 – 20 5.99 6.35 7.97 12.39 1.17
0–5 0.41 0.00 0.00 0.67
1.16

ponent k = 1 (cf. 3.2). Analyzing Falcon’s ability to treat this 1.15

0 5 10 15 20 25
new washability using the same method as with plain sediment Feed solid fraction (vol.%)
washability yields significantly better separation efficiency (cf.
Q = 15 l/min, ω = 2400 rpm
figure 11). In fact, running at 0.13Q−1 0 ω0 gives a concen-
2

trate grade of approximately 5% and a recovery between 25

and 30 wt.%. It is observed that the grade does not depend Figure 14: Search of optimal feed solid concentration for ultrafine coal recovery
strongly on the concentration of solids in the feed stream any- by minimizing cost function (14)
more. These conditions are far more interesting for industrial
applications than the previous ones. The actual benefits that can
be expected from such a process will depend on the distribu- 4
tion of the contamination throughout the sediment washability,
which will control the sand fraction that can be beneficiated. It 3.5 25%
is noted that minus 150 µm particles were used in these simula-
Density (g/cm3 )

tions. Better results can be expected with the Falcon concentra- 3

tor if it is fed with a narrower size distribution, such as minus
63 and 80 µm, which are achievable cut sizes with industrial 2.5 50%
dredged sediment treatment plants.
2

4.2. Coal fines

1.5 75%
Recovery of coal fines from plant tailings is another in- 3 10 20 30 100
1 2 5 50
dustrial process that faces ultrafine particle beneficiation chal- Size (µm)
lenges. Falcon concentrator’s potential application for this pro-
cess was evaluated by applying the analytical model (13) to
a sample washability of coal fines (cf. table 1 and figure 13) Figure 15: Partition function: recovery to overflow (1 − C p ) of sample ultrafine
coal at 15 l/min, 2400 rpm and 16 vol.%.
whose grade is around 70%.
To quantify the separation efficiency and to determine con-
ditions at which the Falcon concentrator should be operated in
order to concentrate this coal, the following function was mini- 4

3.5
50%
Density (g/cm3 )

3
25%
2.5 75%
3)
Probability density
function ( µm g/cm
%

1.5
ρcoal
1
2 10 20 40 60 80 100 120 160 200
1.8 1.6 1.4 100 m) Size (µm)
e (µ
1.2 1
Density (g
/cm 3) Siz

Figure 16: Partition function: recovery to overflow (1 − C p ) of coal fines over

45 µm at 30 l/min, 700 rpm and 20 vol.%.
Figure 13: Sample washability of coal fines
mized: Falcon L40 (Q = 30 l/min, ω = 700 rpm and 30 wt.% feed
p concentration), the predicted recovery ratio is around 30% for a
cost(Gout , Cm ) = (K (1 − Gout ))2 + (1 − Cm )2 (14)
grade quality of approximately 98%, which confirms that the
This cost function involves the concentrate grade (Gout ) and the limitations on concentrating ultrafine coal do not appear for
mass fraction of recovered solids (Cm ) which both are parame- coal that is coarser than 45 µm. This is consistent with observa-
ters whose values must be as high as possible in order to achieve tions detailed in the literature. Moreover, the authors who stud-
good separation from an industrial point of view. Initially, the ied coal beneficiation with a Falcon concentrator did not neces-
product grade is given a far bigger weight (with K = 10), be- sarily need to operate it under such extreme conditions. Con-
cause it is a mandatory condition for industrial markets, while trary to UF bowls, they were using SB bowl in which they could
recovery rate is a process optimization. As detailed in sec- adjust the recovery rate by varying the fluidization counter-
tion 4.1, there is no such thing as a unique set of optimum pressure. It is thus acceptable that their results are quantita-
operating conditions. Instead, there is a continuous range of tively different from our results. The important point is the fact
bowl rotation rates and corresponding flowrates that perform that our model is able to predict the patently opposite trends
the same best achievable separation. Operating at such opti- between coal fines and ultrafines.
mal conditions according to this cost function (Q = 15 l/min, Figure 16 shows the partition function obtained at such ex-
ω = 2400 rpm) results in a concentrate grade of 92% and a re- treme operating conditions with a Falcon L40. The density
covered mass fraction of around 15%. Over the whole range of of clean coal is indicated. From this figure, it appears that,
operating conditions, it does not seem possible to get a product with such conditions, coal is concentrated below 100 µm: 25%-
with a significantly higher grade. recovery curve is equal to clean coal density in the vicinity of
Figure 14 shows the effect of the concentration of solids in 100 µm (coal being less dense than the impurities, the product
the feed stream on function (14) while other operating param- is the overflowing fraction). This corresponds to the size range
eters remain constant. It reaches its minimum value for a feed targeted by such Falcon concentrators. On the contrary, sepa-
volume fraction φ ≈ 0.16 (that is around 25 wt.%). At this feed ration concentrates more and more denser particles that contain
solid concentration, suspension density in the overflow is still more gangue when size decreases. Since these are extreme op-
much lower than the density of coal (1.01 g/cm3 ), so the ef- erating conditions for this device and for the analytical model,
fect of concentration is not expected to significantly improve it does not make sense to use these results to deduce the criti-
the separation according to particle densities. What is actually cal size below which separation becomes inefficient, especially
observed here is a shift in the size cut point, which is consistent since this size depends on the washability of the processed sus-
with the observed grade quality limitation due to the inability to pension. Figure 16 still gives an interesting clue about separa-
efficiently separate the different phases based on their densities. tion trends.
This is illustrated by figure 15 that shows the partition function It might still be possible to overcome this limitation by lever-
at optimal operating conditions. The cut-point line is almost aging fluidization in SB bowls. But, mechanistic models cur-
vertical, meaning that the separation is mostly size dependent. rently lack the ability to account for it. Studies that used flu-
This observation is in good agreement with the analysis of idization were purely empirical. Although they provide valu-
Abela (1997), Bradley et al. (2000) and Honaker et al. (1994– able quantitative results, they are not able to draw conclusive
2004). Indeed, they conclude that Falcon concentrators are not interpretation about separation trends that would help define po-
able to beneficiate coal below 45 µm. To overcome this limita- tential improvements to the coal enrichment process. To extend
tion, some studies focused on improving the density cut-point the presented analytical model to other Falcon series, deeper
using a dense medium (Honaker and Patil, 2002; Honaker et al., understanding of the fluid flow in the retention zone of Falcon
2000). Model (13) is able to predict the impact of such a solu- SB and C bowls would be needed.
tion through its effect on concentration in the overflow (which is
the product in the case of coal while denser tailings are retained 5. Conclusions
in the bowl).
In order to compare model results against conclusions by This paper presents a predictive analytical model for ultra-
these authors, size fraction below 45 µm was removed from fine particle separation with a smooth-bowl Falcon concentra-
the previously used washability. Separation predictions for this tor. This model was developed through a detailed analysis of
new washability show a different behavior. In fact, many op- the physics of this concentrator, on the assumption that the pre-
erating condition sets lead to product grade as high as 100% vailing separation mechanism is particle transport in a flow-
(some of them come at the cost of a very low recovery rate). ing film. The model has been successfully validated using a
The previously defined cost function is here almost indepen- range of experimental and numerical investigations. While the
dent from the value of parameter K (for 0 ≤ K ≤ 10), which model described here enables more comprehensive interpreta-
shows that it almost only accounts for the impact on the recov- tion of experimental data, it in fact is an ideal tool with which
ered fraction. With the coarser new washability, this function a more general analysis of the concentrator’s intrinsic strengths
has no minimum within the operating condition ranges that are and limitations could be investigated. This is in sharp contrast
reachable with a Falcon L40. It improves monotonically when to experimental studies (Duan et al., 2009; Oruç et al., 2010),
increasing flow rate, decreasing bowl rotation speed or increas- which provide only information on the concentrator’s ability to
ing feed solid fraction. For extreme operating conditions of the process specific suspensions. This analytical model can also
be used for process investigation or optimization, and two case Appendix B. Nomenclature
studies, for dredged sediments and coal fines, were described.
This work represents a significant step forward in the simula- Modeling of the particle dynamics
tion of separation processes involving a Falcon concentrator, or
any device whose principles of operation are similar. λ : calibration constant
In the future, it is expected that this model will be extended to P : flow profile
the study of continuous discharge concentrators. These are very u : fluid velocity
sensitive to fluctuations in their feed stream due to the low am- v : particle velocity
plitude of concentration variations they can withstand in their G : gravity acceleration
discharge valves. Our model is able to relate this concentration C D : drag coefficient
to the feed properties and the discharge rate, which would make
it possible to relate the feed fluctuations to their impact on the
separation. Operating conditions

Q : flow rate
Appendix A. Time scales ω : rotation rate of the bowl
C p : partition (or cut) function
Cm : mass ratio to product
Though physical separators’ typical usage is for processing Cv : volume ratio to product
non-reacting slurries, and as such, time scales are of limited in-
f : washability (size and density distribution of particles, by
terest in these units, two specific time scales can be evaluated:
volume)
the residence time of particles in the concentrator, and the op-
erating cycle duration.
As particles entering the retention zone are definitely Physical quantities of the fluid and the solids
trapped, their residence time is infinite. For the fraction flow-
ing out with the carrying fluid, their slip velocity is negligible r p : radius of the particle class ”p”
compared to the fluid velocity in the streamwise direction. Let φ p : volume fraction of particle class ”p”
∆t be the average residence time of the overflowing particles, ρ p : density of the particles
then: ρ f : density of the carrying fluid
ρ s : density of the mixture
Lbowl Lbowl A f ilm
∆t = = µ : viscosity of the carrying fluid
umean Q ν : kinematic viscosity (= µ/ρ f )
= V f ilm Q = π R2 h Q−1
−1
(A.1)

For a Falcon L40, operating at 15 l/min and spinning at Geometry and coordinates
1500 rpm, ∆t is approximately 0.2 s.
h : film thickness
The duration of an operating cycle is constrained by the vol-
r : distance to the bowl’s axis of rotation
ume content of the retention zone (Vtrap ≈ 16 millilitres). The
R : radius of the bowl
cycle must be shorter than the time required to fill the reten-
tion zone. Let ∆T be the operating cycle duration, it can be H : height of the bowl
estimated by: β : opening angle of the bowl
X : streamwise direction
Vtrap k pack Y : wall normal direction
∆T = (A.2)
Q Cv θ : azimuthal direction

where k pack is the packed volume fraction of the solids and Cv is

the volume fraction of particles retained in the bed. Cv depends Dimensionless parameters
on the partition function and on the volume of the particles in Re p : particulate Reynolds number
the slurry:
Ar : Archimedes number
X Ek : Ekman number
Cv = C p Vp (A.3)
p

Both k pack and Cv vary from one slurry to another and are easier Acknowledgments
to get experimentally. When running experiments with silica
sand (D50 = 8 µm), critical operating cycle duration was of the This work is funded by the French “Agence Nationale pour la
order of a few minutes when retaining about half of the solids. Recherche” (ANR), in the framework of the PROPSED project.
References modelling of the Falcon concentrator for beneficiation of ultrafine particles.
Minerals Engineering 23 (4), 313–320, special Issue: Physical Separation.
Kroll-Rabotin, J.-S., Bourgeois, F., Climent, E., 2011a. Beneficiation of con-
Abela, R. L., 1997. Centrifugal concentrators in gold recovery and coal pro- centrated ultrafine suspensions with a Falcon UF concentrator. CIM Journal
cessing. In: Extraction Metallurgy Africa. 2 (4), 189–199.
Bradley, P., Patil, D. P., Ho, K., 2000. Development and demonstration of an Kroll-Rabotin, J.-S., Bourgeois, F., Climent, E., 2011b. Experimental valida-
enhanced gravity separator for coal cleaning. Tech. rep., Illinois Clean Coal tion of a fluid dynamics based model of the UF Falcon concentrator in the
Institute. ultrafine range. Separation and Purification Technology 92, 129–135.
Bruin, S., 1969. Velocity distributions in a liquid film flowing over a rotating Langrish, T. A. G., Makarytchev, S. V., Fletcher, D. F., Prince, R. G. H.,
conical surface. Chemical Engineering Science 24, 1647–1654. 2003. Progress in understanding the physical processes inside spinning cone
Bürger, R., Concha, F., Fjelde, K.-K., Hvistendahl Karlsen, K., 2000. Numeri- columns. Chemical Engineering Research & Design 81 (1), 122–130.
cal simulation of the settling of polydisperse suspensions of spheres. Powder Laplante, A. R., Buonvino, M., Veltmeyer, A., Robitaille, J., Naud, G., 1994. A
Technology 113, 77–318. study of the Falcon concentrator. Canadian Metallurgical Quarterly 33 (4),
Concha, F., Almendra, E. R., 1979. Settling velocities of particulate systems, 279–288.
2. settling velocities of suspensions of spherical particles. International Jour- Laplante, A. R., Nickoletopoulos, N., 1997. Validation of a Falcon model with
nal of Mineral Processing 6, 31–41. a synthetic ore. Canadian Metallurgical Quarterly 36 (1), 7–13.
Concha, F., Bürger, R., 2002. A century of research in sedimentation and thick- Laplante, A. R., Shu, Y., 1993. A comparative study of two centrifugal con-
ening. Kona: powder and particle journal 20, 38–69. centrators. In: 25th Annual Meeting of the Canadian Minerals Processors.
Concha, F., Lee, C. H., Austin, L. G., 1992. Settling velocities of particulate Paper 5. pp. 18–36.
systems, 8. batch sedimentation of polydispersed suspensions of spheres. In- Makarytchev, S. V., Langrish, T. A. G., Prince, R. G. H., 1998. Structure and
ternational Journal of Mineral Processing 35 (3-4), 159–175. regimes of liquid film flow in spinning cone columns. Chemical Engineering
Coulter, T., Subasinghe, G. K. N., 2005. A mechanistic approach to modelling Science 53 (8), 1541–1550.
Knelson concentrators. Minerals Engineering 18, 9–17. Makarytchev, S. V., Xue, E., Langrish, T. A. G., Prince, R. G. H., 1997. On
Detzner, H. D., 1995. The Hamburg Project METHA: Large scale separation, modelling fluid flow over a rotating conical surface. Chemical Engineering
dewatering and reuse of polluted sediments. European Water Pollution Con- Science 52, 1055–1057.
trol 6 (5), 38–42. McAlister, S. A., Armstrong, K. C., 1998. Development of the Falcon concen-
Deveau, C., 2006. Improving fine particle gravity recovery through equipment trator. In: Society for Mining, Metallurgy and Exploration Annual Meeting.
behavior modification. In: 38th Annual Meeting of the Canadian Mineral Oruç, F., Özgen, S., Sabah, E., 2010. An enhanced gravity method to recover
Processors. Paper 31. pp. 501–517. ultra-fine coal from tailings: Falcon concentrator. Fuel 89, 2433–2437.
Dijk, P. E., Janse, A. M. C., Kuipers, J. A. M., van Swaaij, W. P. M., 2001. Schiller, L., Naumann, Z., 1935. A drag coefficient correlation. Zeitschrift des
Hydrodynamics of liquid flow in a rotating cone. International Journal of Vereines Deutscher Ingenieure, 77–318.
Numerical Methods for Heat & Fluid Flow 11 (5), 386–412. Wang, X., Miles, N. J., Kingman, S., 2006. Numerical study of centrifugal
Duan, C., Wen, X., Shi, C., Zhao, Y., Wen, B., He, Y., 2009. Recovery of fluidized bed separation. Minerals Engineering 19, 1109–1114.
metals from waste printed circuit boards by a mechanical method using a Zhao, Y.-M., Wen, X.-F., Shi, H.-X., Jiao, H.-G., Tao, Y.-J., 2006. Study on
water medium. Journal of Hazardous Materials 166, 478–482. metals recovery from −0.074 mm printed circuit boards by enhanced gravity
Holtham, P., Gee, B., Dunne, R., Gregory, S., 2005. Recovery of fine gold separation. The Chinese Journal of Process Engineering 6 (2), 201–204.
particles using a Falcon ‘B’ separator. In: Deschenes, G. (Ed.), International
Symposium for the Treatment of Gold Ores.
Honaker, R. Q., Das, A., 2004. Ultrafine coal cleaning using a centrifugal
fluidized-bed separator. Coal Preparation 24 (1–2), 1–18.
Honaker, R. Q., Patil, D. P., 2002. Parametric evaluation of a dense-medium
process using an enhanced gravity separator. Coal Preparation 22 (1), 1–17.
Honaker, R. Q., Paul, B. C., Wang, D., Ho., K., 1995. Enhanced gravity sep-
aration: an alternative to flotation. In: Society for Mining, Metallurgy and
Exploration Annual Meeting. High Efficiency Coal Preparation: An Inter-
national Symposium.
Honaker, R. Q., Paul, B. C., Wang, D., Huang, M., 1994. Application of cen-
trifugal washing for fine coal cleaning. In: Society for Mining, Metallurgy
and Exploration Annual Meeting.
Honaker, R. Q., Reed, S., 1995. A fine coal circuitry study using column flota-
tion and gravity separation. Tech. rep., Department of Energy.
Honaker, R. Q., Singh, N., Govindarajan, B., 2000. Application of dense-
medium in an enhanced gravity separator for fine coal cleaning. Minerals
Engineering 13 (4), 415–427.
Honaker, R. Q., Wang, D., 1998. Falcon concentrators: a high capacity fine coal
cleaning technology. In: Society for Mining, Metallurgy and Exploration
Annual Meeting.
Honaker, R. Q., Wang, D., Ho, K., 1996. Application of the Falcon concentrator
for fine coal cleaning. Minerals Engineering 9 (11), 1143–1156.
Janse, A. M. C., Biesheuvel, P. M., Prins, W., van Swaaij, W. P. M., 2000. Gran-
ular flow in a rotating cone partly submerged in a fluidized bed. American
Institute of Chemical Engineers Journal 46 (3), 499–508.
Kim, B. H., Klima, M. S., 2004a. Development and application of a dynamic
model for hindered-settling column separations. Minerals Engineering 17,
403–410.
Kim, B. H., Klima, M. S., 2004b. Simulation of hindered-settling column sep-
arations when processing fine anthracite refuse. Coal Preparation 24, 261–
275.
Kroll-Rabotin, J.-S., 2010. Analyse physique et modélisation de la séparation
centrifuge de particules ultrafines en film fluant: Application au séparateur
industriel Falcon. Ph.D. thesis, Institut National Polytechnique de Toulouse.
Kroll-Rabotin, J.-S., Bourgeois, F., Climent, E., 2010. Fluid dynamics based