See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/271614258

Analysis of swirling flow in hydrocyclones operating under dense regime

Article in Minerals Engineering · May 2012

DOI: 10.1016/j.mineng.2012.01.012

CITATIONS READS

53 2,000

4 authors, including:

Eric Climent Florent Bourgeois

Institut de Mécanique des Fluides de Toulouse Ecole Nationale Supérieure des Ingénieurs en Arts Chimiques et Technologiques
130 PUBLICATIONS 2,667 CITATIONS 59 PUBLICATIONS 1,076 CITATIONS

SEE PROFILE SEE PROFILE

Arun K Majumder
Indian Institute of Technology Kharagpur
61 PUBLICATIONS 1,090 CITATIONS

SEE PROFILE

All content following this page was uploaded by Florent Bourgeois on 16 October 2019.

The user has requested enhancement of the downloaded file.

This article appeared in a journal published by Elsevier. The attached
copy is furnished to the author for internal non-commercial research
and education use, including for instruction at the authors institution
and sharing with colleagues.
Other uses, including reproduction and distribution, or selling or
licensing copies, or posting to personal, institutional or third party
websites are prohibited.
In most cases authors are permitted to post their version of the
article (e.g. in Word or Tex form) to their personal website or
institutional repository. Authors requiring further information
regarding Elsevier’s archiving and manuscript policies are
encouraged to visit:
http://www.elsevier.com/copyright
 Author's personal copy

Minerals Engineering 31 (2012) 32–41

Contents lists available at SciVerse ScienceDirect

Minerals Engineering
journal homepage: www.elsevier.com/locate/mineng

Analysis of swirling flow in hydrocyclones operating under dense regime

Aurélien Davailles a,b, Eric Climent a,b,⇑, Florent Bourgeois b,c, Arun Kumar Majumder d
a
Université de Toulouse, INPT, UPS, IMFT (Institut de Mécanique des Fluides de Toulouse), Allée Camille Soula, F-31400 Toulouse, France
b
CNRS, Fédération de recherche FERMAT, Toulouse, France
c
Université de Toulouse, INPT, UPS, LGC (Laboratoire de Génie Chimique), 4 Allée Emile Monso, BP 44362, 31432 Toulouse Cedex 4, France
d
Department of Mining Engineering, Indian Institute of Technology, Kharagpur 721 302, India

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

Article history: There are many circumstances where hydrocyclone performance and dense flow are intertwined, such as
Available online 22 February 2012 for example when feed solids flow exceeds hydrocyclone capacity during continuous operations. The work
reported here, which is part of an ongoing research effort to develop a robust CFD model for prediction of
Keywords: hydrocyclone performance, focuses on hydrocyclone operation under high solids concentration. The paper
Hydrocyclone presents the basic physics framework that accounts for solid–liquid and solid–solid interactions under
CFD hydrocyclone’s swirling flow. Operating conditions that are past the transition from spray to rope regime
Solids concentration
are deliberately chosen for this purpose. Model predictions are validated by comparison with solids split
Roping
and separation curves measured on a 100 mm diameter hydrocyclone. CFD model predictions permit tak-
ing an insightful look at the inside of a hydrocyclone under extreme operating conditions, which would be
difficult to achieve experimentally. Velocity profiles, G-force distribution and distribution of solids pre-
dicted by CFD are bound to lead to a better understanding of the separation that takes place inside a hydro-
cyclone, which may eventually help improve hydrocyclone design and performance.
Ó 2012 Elsevier Ltd. All rights reserved.

1. Introduction operating conditions. The physics of particle separation behavior

inside a hydrocyclone at dilute conditions and the CFD simulation
There are many circumstances where hydrocyclone perfor- have already been described elsewhere (Davailles et al., in press).
mance and dense flow are intertwined, such as for example when The present article, however, deals with the CFD modeling of a
feed solids flow exceeds hydrocyclone capacity during continuous hydrocyclone at high solids concentrations.
operations. Moreover, under normal operating conditions, it is gen-
erally perceived that some regions inside a hydrocyclone might be
very densely populated with solids while other regions might only 1.1. Experimental studies
experience dilute conditions. Dilute conditions are expected in the
central part of the hydrocyclone and in the cylindrical section of The hydrodynamic behavior of a hydrocyclone depends on inlet
the hydrocyclone, whereas dense conditions, as dense as packed solid concentration, feed flow rate and geometry of the device.
beds in some instances, can be expected along the hydrocyclone Fig. 1 presents two distinct behaviors, below and above the transi-
walls as well as in the conical region. With the above in mind, tion to roping (Neesse et al., 2004a). Under those two operating re-
solid–solid separation inside a hydrocyclone is bound to vary sig- gimes, visualization of solid–liquid flow patterns is difficult as the
nificantly throughout the body of an operating hydrocyclone. This accumulation of particles along the wall highly complicates the
reasoning emphasizes the need to better understand particle and optical access to the hydrocyclone’s core. Usual techniques which
water flow behavior inside a hydrocyclone under high solids have been applied to study single phase flow, such as Laser Doppler
concentration, whether dense regions occur locally or spread Velocimetry (Hsieh, 1988), are not applicable for concentrated sys-
throughout the hydrocyclone body. In an ongoing research effort, tems. Much more complex measurement tools, such as X-ray
therefore, an attempt has been made to develop a robust CFD mod- tomography (Galvin and Smitham, 1994) or radioactive particle
el for the prediction of hydrocyclone performances at different tracking (Chang et al., 2011), have, therefore, been tried. However,
it is still very challenging to gain new experimental insights into
local hydrodynamics of dense solid–liquid flows. At low feed solids
⇑ Corresponding author at: Université de Toulouse, INPT, UPS, IMFT (Institut de
concentration, the underflow discharge of a hydrocyclone has a
Mécanique des Fluides de Toulouse), Allée Camille Soula, F-31400 Toulouse, France.
E-mail addresses: [email protected] (A. Davailles), [email protected]
spray profile and a stable air core is established due to the suction
(E. Climent), fl[email protected] (F. Bourgeois), majumder1965@gmail. of air through the underflow (Fig. 1a). However, with increasing
com (A.K. Majumder). feed solids concentration the quantity of solids exiting through

0892-6875/$ - see front matter Ó 2012 Elsevier Ltd. All rights reserved.
doi:10.1016/j.mineng.2012.01.012
 Author's personal copy

A. Davailles et al. / Minerals Engineering 31 (2012) 32–41 33

clones and dense suspensions. This implies a degree of compromise

between the accuracy of the physics and the computational cost.
Processes at both the macroscopic and local scales have some
influences on hydrocyclone’s behavior; hence both scales must
be embedded in models for making prediction of hydrocyclone
performance. Quite a few experimental and numerical studies
helped build a strong understanding of single-phase flow inside
hydrocyclones (Kelsall, 1952; Hsieh and Rajamani, 1991; Brennan,
2006).
In contrast, the sensitivity of separation performance to feed
solids concentration possibly has not been studied so far and,
therefore, remains unexplained. One reason for this situation is
that the measurement of slurry velocity profiles or solids spatial
distribution anywhere inside the hydrocyclone is an experimen-
tally challenging task.
It is comparatively simple to obtain such information from CFD
simulations based on the understanding of swirling flow patterns
inside a hydrocyclone. The physics and the CFD modeling dealing
with dilute suspensions was validated with our experimental data
as well as with a set of published data (Davailles et al., in press).
Fig. 1. Hydrocyclone: (a) dilute and (b) dense slurries. The present study focuses on the analysis of swirling flow in hydro-
cyclones operating under dense flow regimes having the following
the underflow discharge also increases and above a critical point objectives:
(Neesse et al., 2004b) the rate of accumulation of solids in the con-
ical portion exceeds the rate of discharge. As a result the spray pro-  Development of a modeling scheme based on the strong foun-
file collapses and the air suction is also hindered. This condition is dation of physics of fluid flow and particle–fluid interactions.
generally termed as roping (Fig. 1b). Neesse and Dueck (2007)  Simulation of the roping conditions in a 100 mm diameter
opined that the air core cannot totally disappear while the exit is hydrocyclone using an Eulerian multi fluid model with RANS
plugged by particles. An air core remains within the hydrocyclone turbulence modeling.
and vanishes on the particles bed. However, the small size of the  Validation of the model predicted data with experimental
air core and the absence of connection with the atmosphere se- data.
verely limit its influence on the hydrodynamics for this operating
regime. It is not within the scope of this paper to investigate the transi-
tion between spray and rope discharge despite the occurrence of
1.2. Numerical simulations and the objectives this transition is an important issue from an operational stand-
point. Indeed, the issue at stake here is that of the effect of dense
CFD studies on hydrocyclones generally aim at giving new in- suspension on swirling patterns and particle separation inside
sights on local flow patterns, such as velocity profiles, pressure dis- the hydrocyclone. This yields the choice of operating conditions
tribution and spatial distribution of solids. These information are used throughout this work.
difficult to access through experimentation and are critical for bet-
ter understanding of separation performances.
2. Description of CFD model
However, numerical modeling of all physical aspects still re-
mains a challenging research topic due to the complexities associ-
The three dimensional numerical simulations were performed
ated with the swirling flow behavior of concentrated slurries
with NEPTUNE_CFD v1.08@Tlse (Laviéville and Simonin, 1999,
within a hydrocyclone. The geometry is not an obvious one and
2006; Ozel et al., 2010; Galassi et al., 2009). NEPTUNE_CFD is a
contributes significantly to flow streams repartition between the
multiphase flow simulation software developed within the frame-
two exits. The level of turbulence is high corresponding to a hier-
work of a French industrial consortium.
archy of spatial flow structures interacting selectively with the par-
The behavior of dispersed multiphase flow is modeled using the
ticles. Due to strong swirling, the anisotropy of the flow couples
general Eulerian multi-fields balance equations which are obtained
particle centrifugation and triggers air focusing in the core of the
from the fundamental conservation laws of physics (mass conser-
device. The intrinsic three phase (gas–liquid–solid) flow with a
vation and momentum balance), restricted to Newtonian mechan-
large amount of particles make hydrocyclones a real challenge
ics. Conservation laws are written under differential form that are
for CFD codes.
valid at any time and location within the continuum, except across
Large Eddy Simulation (LES) appears to be the most promising
interfaces between two physical phases. At interfaces, jump condi-
approach for modeling flow turbulence under dilute feed flow rate
tions derived from continuous equations are written and inte-
(Slack et al., 2000; Delgadillo and Rajamani, 2005) as it preserves
grated through source and sink terms in the equations. This
turbulence spatial structure. For particle laden flows, LES can be
results in a fully coupled set of equations.
coupled with Discrete Particle Simulation. Without momentum
coupling between the solid phase and the liquid phase, this permits
to evaluate separation curves under dilute regime based on trajec- 2.1. Transport equations
tory analysis. Unfortunately, this technique is very time consuming
and the relevance of the model for concentrated slurries needs fur- The multiphase Eulerian model is built on a kinetic approach
ther theoretical developments and validation tests on academic based on a joint fluid-particle Probability Density Function (Simo-
configurations. This approach is still reserved for small scale study nin, 1996). The set of PDF’s moment transport equations (mass,
and only thousands of particles, while the work that is presented momentum, particle agitation) is derived from a Boltzmann-like
here is concerned with the simulation of industrial size hydrocy- kinetic equation of the PDF.
 Author's personal copy

34 A. Davailles et al. / Minerals Engineering 31 (2012) 32–41

The multi-field mass balance equation for field k is written According to Gourdel et al. (1999), the momentum exchange
(when k = l, we refer to the liquid and k = p to the class p of due to collisions between particle q and particle p can be written
particles): as:
@ @ mq 1 þ ec H1 ðZ pq Þ
ðak qk Þ þ ðak qk U k;i Þ ¼ 0 ð1Þ Iq!p;i ¼ ap qp ðU p;i  U q;i Þ ð9Þ
@t @xi mp þ mq 2 scpq
with ak, qk, Uk, the volumetric fraction, the density and the mean mp and mq are respectively the mass of particle p and particle q,
velocity of phase k. ec is the inelastic restitution coefficient and scpq the collision char-
The multi-fluid momentum balance equation for phase k is de- acteristic time.
fined as follows: The momentum exchange term presented in Eq. (9) comple-
@ @ ments the monodisperse model. When mp = mq(monodisperse),
ðak qk U k;i Þ þ ðak qk U k;j U k;i Þ Up = Uq yielding Iq?p,i equals to zero.
@t @xj
H1(Zpq) is a theoretical function that permits adapting the ki-
@P X @
¼ ak þ Iðq!kÞ;i þ ak qk g i þ T k;ij ð2Þ netic theory of gases to that of granular flow.
@xi q¼p;g @xj

with P the mean pressure, gi the acceleration due to gravity and Tk,ij 2.2. Turbulence modeling
the effective stress tensor.
I(l?p),i accounts for momentum transfer rate from the liquid to Ensemble averaging is usually applied to instantaneous Navier–
solid phase (friction), while I(q?p),i represents the momentum ex- Stokes equations in order to study industrial flows. These new
change by collision between particles q and p. In order to account equations are called Reynolds Averaged Navier–Stokes (RANS)
for collisions between different particle sizes, the mean momen- equations.
tum transfer between phases I(q?p),i has been generalized in this The effective stress tensor, Tk,ij, in Eq. (2) can be written as
study compared to the validation on dilute regime (Davailles follows:
et al. (in press)). T k;ij ¼ ak qk Rk;ij þ Hk;ij ð10Þ
The mean liquid to particle momentum transfer can be modeled
using an estimate of the drag acting on particles. It contains two contributions: a turbulent kinetic stress tensor
Rk,ij due to turbulence (or fluctuations in phase k) and a collisional
ap qp
Iðl!pÞ;i ¼ Iðp!lÞ;i ¼ V ð3Þ (or molecular viscosity) term Hk,ijfor the solid (alt. liquid) phase.
sFlp r;i The constitutive relations for viscosity and diffusivity are derived
1 3 ql hC D ip in the framework of the kinetic theory for dry granular flows.
with ¼ hjV r ji ð4Þ
sFlp 4 qp dp According to Batrak et al. (2005), the solid stress tensor Tp,ij can
be written as:
Kinetic stress
sFlp is the particle relaxation time scale, and h ip is the ensemble zfflfflfflfflfflfflfflfflfflfflfflfflfflffl
 ffl}|fflfflfflfflfflfflfflfflfflfflfflfflfflffl
 ffl{  
average operator over the particulate phase (Simonin, 1996). Vr,i is @U p;m @U p;i @U p;j 2 @U p;m
T p;ij ¼ Pp  kp dij  lp þ  dij
the average of the local relative velocity which can be expressed in @xm @xj @xi 3 @xm
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
terms of the averaged velocity between phases and the drift veloc- Collisional stress
ity. The mean drag coefficient related to an isolated particle, hCDip
can be written as a function of particulate Reynolds number. Under with lp = lp,col + lp,cin + lp,fric.
moderately concentrated suspension to dense regime, the multi- The terms lp,col and lp,cin represent the kinetic viscosity and the
fluid approach accounts for local particle volume fraction ap collisional viscosity respectively (Batrak et al. (2005)).
through a correction to the drag law. Under these conditions, we When the particulate volume concentration is high (typically
adopt the approach published by Gobin et al. (2003), who proposed ap P 0.5), the assumption of instantaneous binary collision is often
to use a combination of two relations: Wen and Yu (1965) for di- inaccurate as enduring contacts are prone to occur. Frictional ef-
lute cases and Ergun (1952) for concentrated regime. fects are accounted for by adding a frictional viscosity lp,fric as de-
8 scribed by Srivastava and Sundaresan (2003).
>
> if ap > 0:3 Because of the strong anisotropy of the flow, the Reynolds Stress
>
> D E D E 
>
> Model (RSM) due to Launder et al. (1975) seems naturally better
>
< Min C Wen Yu
; C Ergun
D D
hC D ip ¼
p p
ð5Þ suited than models based on an assumption of isotropy of the Rey-
>
> if ap 6 0:3 nolds stress tensor, such as the familiar k   turbulence model.
>
> D E
>
> With RSM, the turbulent viscosity approach has been comple-
>
: C Wen Yu
D
p mented and the Reynolds stresses hqfuf,iuf,jif are computed directly
to close RANS equations. This is done by solving the differential
with transport equations, leading to heavier computational costs. The
D E ap 7 exact Reynolds stress transport equation accounts for the direc-
C Ergun
D ¼ 200 þ ð6Þ tional effects of the Reynolds stress fields. The Reynolds stress
p Rep 3
model has been extended to multiphase flows.
8 For the particles, particle fluctuating kinetic energy, q2p and fluid-
> if Rep < 1000
>
>   particle fluctuating velocity covariance, qfp are solved through a set
D E >
< a1:7 24
1 þ 0:15Re0:687 of coupled transport equations. For polydisperse solid phase, the
C Wen
D
Yu
¼ l Rep p
ð7Þ
p >
> particle agitation transport equation contains some specific terms
>
> if Rep P 1000
: (Gourdel et al., 1999; Batrak et al., 2005) that account for three phe-
0:44a1:7
l nomena: particle agitation dissipation by non-elastic polydisperse
collisions, production of particle agitation by the mean solid rela-
al ql dp hjV r ji tive velocity and finally, transfer of particle agitation between dif-
Rep ¼ ð8Þ
ll ferent classes of the solid phase.
 Author's personal copy

A. Davailles et al. / Minerals Engineering 31 (2012) 32–41 35

2.3. Mesh and boundary conditions

The whole volume of the 100 mm diameter hydrocyclone has

been meshed, as shown in Fig. 2. Simail software, which is associ-
ated to NEPTUNE_CFD v1.08@Tlse, was used to build the mesh. The
3D mesh was first generated by a full rotating extrusion of a cross-
section of the hydrocyclone. Then, in order to prevent the existence
of a singular axis at the center of the hydrocyclone, the cylindrical
volume located around the central axis has been meshed sepa-
rately. Overall, the total number of cells comprises 450,000 hexa-
hedral and pentahedral structured control volumes.
Boundary conditions at the feed inlet are prescribed assuming a
uniform distribution of solids across the section. Atmospheric pres-
sure is imposed at the underflow and overflow exits. The suspen-
sion can leave the domain freely through those exits. Along the
hydrocyclone walls, standard turbulent friction functions are used
for the liquid phase (law of the wall). A zero flux
 condition is ap-
dU
plied on the velocity field for particles dnp ¼ 0 .

3. Experimental work

Fig. 3. Geometric features of Neyrtec’s HC100 hydrocyclone.

As justified earlier, operating conditions were chosen past the
transition from spray to rope, such that discharge of the 100 mm
hydrocyclone pilot test rig was a steady rope. Table 1
Nominal features of Neyrtec’s HC100 hydrocyclone.
3.1. Experimental set-up Model HC 100
Diameter (mm) 100
The pilot test rig dimensions are reported in Fig. 3. Standard glo- Feed rate (m3/h) 7–13.5
bal information, including operating parameters and performance Feed pressure (kPa) 60–250
characteristics (partition function) can be accessed with this set- Cut size (lm) 7–18
Material Polyurethane
up.
Overflow (OF) diameter (mm) 33
The Neyrtec’s HC100 hydrocyclone (100 mm diameter) is made Vortex finder length (mm) 50
of interchangeable polyurethane parts, allowing various geometri- Underflow (UF) diameter (mm) 18
cal configurations to be tested (see technical specifications in Table
1). The length of the cylindrical body can be adjusted using
100 mm extensions, and several diameter spigots (from 6 mm to
18 mm) are available. One 100 mm cylindrical extension was used cyclone. Slurry mass flow rate is monitored using an FSM4000
for the experiments and simulations reported here, with either electromagnetic flowmeter from ABB Automation Products GmbH.
6 mm or 18 mm diameter spigots. One single 50 mm vortex finder The mass flowmeter readings were calibrated from pure water
was use in the experiments. flow up to 50 wt% solids (28 vol.%).
The slurry is pumped around a closed circuit using a C40 Silica particles were used in the experiments. The particle size
SCHABAVER 2.2kW centrifugal pump. Both underflow and over- distribution is plotted in Fig. 4. The material itself is 98.8% SiO2
flow are recirculated to a 1 m3 agitated sump that feeds the hydro- (according to the manufacturer), hence it is assumed to consist of
a single constituent with specific gravity 2.65.
It was found during the early phases of our work that the geom-
etry of the simulated hydrocyclone should closely match the
geometry of the real hydrocyclone, particularly in the inlet region.
As seen in Fig. 5, the inlet pipe section is connected to the cylindri-
cal body of the hydrocyclone through a rectangular and tangential
channel through a 10 mm  30 mm section. The sudden reduction
in cross-sectional area forces a sharp increase of the inlet flow
velocity, which the simulation must model accurately. Indeed,
the initial velocity of water and particles as they enter the hydro-
cyclone are utterly critical.

3.2. Experimental results

This section presents the experimental data that were obtained

with feed solids concentration in the range 10–50 wt% (4 vol.% and
27 vol.%).

3.2.1. Effect of feed solids concentration on hydrocyclone performance

The feed solids concentration was varied in 10 wt% increments
between 10 wt% and 50 wt%. Feed volumetric flowrate was set to
Fig. 2. Axial and cross-section view of the mesh. 2.6  103 m3 s1 (156 L min1). Overflow and underflow streams
 Author's personal copy

36 A. Davailles et al. / Minerals Engineering 31 (2012) 32–41

Fig. 4. Size distribution of silica particles: measured (symbols) and discretized for
numerical simulation (solid line).

Fig. 6. Measured partition function for 10 wt% to 50 wt% feed solids concentration
with the 18 mm diameter spigot.

Table 2
Feed solids concentration influence on cut-sizes and solids recovery to the underflow.

d50 (lm) d50c (lm) j^ Rf (%) SI

10 wt% 19.25 19.5 0.62 2.0 0.55

20 wt% 30.55 31.5 0.53 5.0 0.53
30 wt% 40.99 46.0 0.47 16.9 0.48
40 wt% 64.84 75.3 0.32 13.4 0.40
50 wt% 68.05 95.3 0.36 21.4 0.36

tern. It is noted that the hydrocyclone did not choke under the con-
ditions of Table 2.
As the objective of this study is the understanding of the hydro-
cyclone hydrodynamics with roping discharge, this paper focuses
on feed solids content higher than 30 wt% (14 vol.%).

Fig. 5. HC100 injection. 3.2.2. Effect of spigot diameter

In order to provide different conditions against which to test
CFD predictions, experiments were carried out using two spigot
were sampled manually by short successive cuts of the full diameters, 6 mm and 18 mm. Clearly, reducing the spigot diameter
streams. A sample of feed material was taken for every experiment with a roping hydrocyclone is a practical nonsense, however it
at the hydrocyclone underflow, using a low flowrate for which all makes sense in the context of creating additional dense regime
the slurry would report to the underflow. Particle size distributions conditions against which to test the CFD simulation model and
of the feed, the underflow and the overflow were measured by la- study the swirling flow inside the hydrocyclone. Partition functions
ser diffractometry using a Malvern 2000S, using a silica refractive that were obtained with 30 wt% feed solids concentration are plot-
index of 1.49. Every data set was reconciled using standard mass ted in Fig. 8 for both spigot diameters. The important point here is
balance techniques, from which partition curves were calculated. that hydrocyclone performance changes significantly between
Results obtained with the 18 mm diameter spigot are presented these conditions, which reflect significant changes in water and
in Fig. 6 and reported in Table 2. The measurements follow ex- particle flow inside the hydrocyclone. It is noted that some classi-
pected trends for feed solids concentration and cut size, such as re- fication is still taking place despite the roping discharge of the
ported by Nageswararao et al. (2004). Indeed, the cut size (d50 or hydrocyclone.
d50c) and the concentration of solids in the underflow both increase It was found that a threefold decrease in spigot diameter gave
with feed solids content. In addition, the steepness of the partition an increase in d50 from 37 lm to 91 lm under these roping condi-
function slope (sharpness index SI) decreases as solids concentra- tions. Separation sharpness was similar with both spigots; how-
tion is increased. Water recovery was not measured in this study. ever water recovery to underflow increased (from about 10% to
The values of water recovery to the underflow Rf, which are re- 20%) with the larger spigot diameter.
ported in Table 2 were read directly from the partition functions.
These are found to increase from 2% to about 20% with feed solids
concentration in the range 10–50 wt%. As expected, the perfor- 4. Numerical simulation at high solids concentration
mance of the hydrocyclone worsens monotonically with increasing
feed solids concentration. According to earlier statements from Section 2.3 about the ab-
From 30 wt% solids concentration onwards, the underflow dis- sence of an air core at high solids concentration, which was also
charges as a rope. For sake of clarity, Fig. 7 shows the discharge illustrated in Fig. 1, hydrocyclone operation in dense regime was
of the hydrocyclone at 10 wt% (4 vol.%) and 30 wt% (14 vol.%) so computed by simulating water and solids phase only. The compu-
the reader can readily appreciate the drastic change in flow pat- tational ease that is gained from having to simulate two phases
 Author's personal copy

A. Davailles et al. / Minerals Engineering 31 (2012) 32–41 37

Fig. 7. Photographs of the underflow discharge with feed solid content of 10 wt% and 30 wt%.

As with the measurements, the simulations predict that the

recovery of water to underflow increases with feed solids concen-
tration, with a significant increase between 40 wt% and 50 wt%.
These changes in hydrocyclone performance areassociated with
a changing distribution of solids inside the hydrocyclone body. Ta-
ble 3 gives the mass distribution of the water and the 4 size classes
inside the hydrocyclone as a function of feed solids concentration.
For consistency of the results, it can easily be verified from the data
in Table 3 that the volume of water and solids inside the hydrocy-
clone is constant and equal to 1.82 L.
Results at 30 wt% and 40 wt% feed solids concentration are
nearly same. Partition functions in Fig. 9 share the same shape
and numerical results are close to each other. Table 3 shows a
strong gradient between the mass of small and large particles
inside the hydrocyclone, which reflects that a strong size classifica-
tion effect is still taking place, despite the hydrocyclone operating
in the roping regime. Solids concentration inside the hydrocyclone
Fig. 8. Measured influence of spigot diameter on separation curves.
is also higher than the feed (36 wt% and 46 wt% against 30 wt% and
40 wt%), which indicates that the hydrocyclone is thickening the
particles at the underflow. As it will be shown later, these results
instead of three (with gas) is partly offset by the added complexity indicate that the hydrocyclone body length is such that even under
of simulating dense particle flow in the hydrocyclone. roping conditions, there is sufficient body length inside which the
The particle size distribution used in the experiments was dis- flow is swirling for some classification to take place, giving parti-
cretized into four size classes, each class accounting for 25 wt% of cles sufficient residence time for separation to occur.
the experimental size distribution. In the end, the simulation uses
four types of particles with sizes equal to the arithmetic mean size
of the four size classes. These four sizes are 6 lm, 25 lm, 50 lm
and 96 lm respectively. Thus, the size distribution used in the sim-
ulations is a step function that follows the experimental size distri-
bution. Both size distributions were plotted in Fig. 4.
The feed flowrate used in the simulation is set equal to the
experimental flowrate (2.6 L s1). Water and silica mass flow rates,
which depend on the solid to liquid ratio in the feed, are calculated
using 1.0 s g. for water and 2.65 s g. for silica particles.

4.1. Effect of feed solids concentration

Partition functions predicted by numerical simulation, defined

as the ratio of the mass flux of each particle size in the underflow
to that in the feed, are presented in Fig. 9. Results obtained in very
dilute condition (1 wt%) are used to illustrate the best separation
that can be achieved with the hydrocyclone. With increasing feed
solids content, separation efficiency is predicted to deteriorate as Fig. 9. CFD predictions of separation curve as a function of feed solids concentration
it is observed experimentally. (symbols: simulation, solid lines: experimental measurements).
 Author's personal copy

38 A. Davailles et al. / Minerals Engineering 31 (2012) 32–41

Table 3
Mass distribution of water and solids inside the hydrocyclone body.

Inlet content Water (kg) Silica (kg)

6 lm 25 lm 50 lm 96 lm Total
30 wt% 1.509 0.163 0.182 0.247 0.241 0.832
14 vol.% 64.4% 7.0% 7.8% 10.5% 10.3% 35.6%
40 wt% 1.376 0.231 0.249 0.322 0.383 1.185
20 vol.% 53.7% 9.0% 9.7% 12.6% 15.0% 46.3%
50 wt% 1.311 0.328 0.326 0.332 0.370 1.355
28 vol.% 49.2% 12.3% 12.2% 12.4% 13.9% 50.8%

As feed solids concentration is increased further, the gradient

between the mass of small and large particles decreases, to the
point where it is almost nil with the 50 wt% case. In this case,
the solids distribution inside the hydrocyclone is almost equal to
that of the feed. The solids concentration inside the hydrocyclone Fig. 11. G-force profiles in the cylindrical section of the hydrocyclone, taken
is 51%, which is also almost identical to the feed. There is hardly 100 mm under the roof.
any classification taking place in this case, which indicates that
there is hardly any zone inside the hydrocyclone where flow is
separation. This change in behavior is entirely due to the concen-
swirling at a sufficiently high rate. This will be confirmed later
tration of solids inside the hydrocyclone, as shown hereafter.
through observation of solids concentration distribution inside
Fig. 12 shows the distribution of liquid volume fraction (alt. the
the device.
distribution of solids concentration) inside the hydrocyclone. With
The tangential velocity, which drives the separation inside the
30 wt% and 40 wt% feed solids content, it is apparent that particles
hydrocyclone, is strongly affected by the feed solid concentration.
undergo a separation inside the hydrocyclone, with a strong gradi-
As shown in Fig. 10, tangential velocity predicted at 1 wt% is signif-
ent of water (alt. solids) concentration from bottom to top of the
icantly higher than tangential velocity obtained at the high solids
hydrocyclone. In the 50 wt% case, water (alt. solids) concentration
concentrations used in this study. Profiles at 30 wt% and 40 wt%
is homogeneously distributed throughout the body of the
are close to each other, which agrees with the closeness of the par-
hydrocyclone, which implies that little separation is taking place
tition functions predicted at these concentrations (cf. Fig. 9). The
inside the hydrocyclone. Due to the solid concentration, local vis-
same profile with 50 wt% solids content in the feed is very differ-
cosity of the suspension increases, which dissipates the energy that
ent, as was the partition function, with values of tangential velocity
would otherwise be available to operate the separation inside the
half those obtained at 30 wt% and 40 wt% solids concentrations. 2
U hydrocyclone.
Keeping in mind that the separation force is proportional to rghp
According to Figs. 13 and 14, the tangential velocity and the
(Uhp is the particles tangential velocity, r the radius and g the accel-
associated G-force decrease steadily from the top to the bottom
eration due to the gravity), the maximum value of this acceleration
of the hydrocyclone. The bottom of the conical section is filled with
factor, which is a key separation driver decreases from about 900
particles and no longer contributes to the classification perfor-
for a feed solids content of 1 wt% to 280 for 30 wt%, 230 for
40 wt% and only 40 for 50 wt% (Fig. 11).
The reduction in tangential velocity is perhaps best understood
when it is recognized that should the inside of the hydrocyclone
rotate as a solid rotation, the tangential velocity would be linear.
What this means is that increasing feed solids concentration yields
velocity distribution in the hydrocyclone that tends to reduce the
differential motion of particles, which then worsens particle

Fig. 10. Water tangential velocity profile in the cylindrical section of the hydro- Fig. 12. Distribution of liquid volume fraction (=1 – solid volume fraction) inside
cyclone, taken 100 mm under the roof. the hydrocyclone.
 Author's personal copy

A. Davailles et al. / Minerals Engineering 31 (2012) 32–41 39

Fig. 13. Evolution of the tangential velocity along the height of the hydrocyclone
Fig. 15. Influence of spigot diameter on separation curves.
with 30 wt% feed solids content.

Table 4
Influence of spigot diameter on the mass of liquid and solids inside the hydrocyclone
body.

Water Silica (kg)

(kg)
6 lm 25 lm 50 lm 96 lm Total
UF 6 mm 1.339 0.146 0.172 0.276 0.391 0.985 42 wt%
UF 18 mm 1.509 0.163 0.182 0.247 0.241 0.832
36 wt%
Variation 11% 10% 5% 12% 62% 18%

Fig. 14. Evolution of the G-force profiles along the height of the hydrocyclone with
30 wt% feed solids content.

mance of the hydrocyclone. The solid that fills this region of the
hydrocyclone reduces the volume available for separation, as it
gives an effective geometry for classification that is different from
the physical geometry of the hydrocyclone. In the end, the simula-
tion results show that the internal geometry of the hydrocyclone
where swirling flow is occurring gradually evolves from that of a
classifying cyclone to that of a stub cyclone. The notion of effective
versus physical geometry of the hydrocyclone, as modified by the
solids concentration inside the hydrocyclone, is deemed an
important concept that underlies separation efficiency of the
hydrocyclone.

4.2. Spigot diameter influence

As discussed in the previous Section 4.1, an increase of the total Fig. 16. Influence of spigot diameter on liquid volume fraction.
solids mass inside the hydrocyclone reduces the separation perfor-
mance by slowing down the tangential velocity (hence the G-force)
inside the hydrocyclone, which is the real separation driver. This separation performance (see Fig. 15). As per the measurement,
also yields an increase in both water and fines recovery to the the fines ratio decreases with decreasing spigot diameter, which
underflow. A further decrease of the spigot diameter from 18 mm is correlated to the water split.
to 6 mm was found in Section 3.2.2 to give a lower recovery to The change from the 18 mm to the 6 mm spigot affects particles
underflow with both fines and water under the same feed condi- in a different manner, which depends on their diameter. This can
tions, recalling that the hydrocyclone was already roping with be appreciated through the evolution of the mass of individual size
the 18 mm spigot. classes inside the hydrocyclone are reported in Table 4.
The same trend is found by numerical simulations, however With a small spigot diameter, the largest particles are more
simulation tends to overestimate the effect of spigot diameter on concentrated inside the hydrocyclone, and more of the smaller
 Author's personal copy

40 A. Davailles et al. / Minerals Engineering 31 (2012) 32–41

particles report to the overflow. Despite the reduced separation 99% of the feed mass flow rate report to the overflow, which
performance (see Fig. 15), the hydrocyclone is able to separate par- would correspond to the hydrocyclone choking in practice. De-
ticles under the conditions of the test. spite the fact that the simulations give interesting trends about
Fig. 16 reveals that there is a low solids concentration area in the effect of dense regime on hydrocyclone separation behavior,
the cylindrical part of the hydrocyclone with both spigots. The con- it also shows that the current simulation CFD code is not yet able
sequence is that there is still sufficient tangential velocity (see to predict all that happens inside the hydrocyclone as the hydro-
Fig. 17) to yield some classification. The increase in cut size is cyclone did not choke under these conditions. The discrepancies
not correlated with a decrease in tangential velocity 100 mm from can also been seen in the predicted and measured partition func-
the roof, as seen with the larger spigot at the highest solids concen- tions of Fig. 15.
tration (Fig. 10).
The change in spigot diameter is responsible for this evolution
5. Conclusion and perspectives
of the behavior. According to Table 4 and Fig. 18, there is almost
twice as much of the largest particles inside the hydrocyclone
CFD simulation offers significant potential for studying and
with the 6 mm diameter spigot than with the 18 mm spigot.
understanding the separation mechanisms that take place inside
Moreover, due to the small underflow diameter and by the com-
a hydrocyclone. This can potentially give practitioners and
paction of particles in the conical part, the flow rate passing
researchers interesting guidelines for improving the performance
through the vortex finder increases and drives out more particles,
of hydrocyclones and possibly improve their design.
irrespective of their diameter. With the 6 mm diameter spigot,
There are many circumstances where hydrocyclone perfor-
mance and dense flow are intertwined, however the effect of dense
regimes inside a hydrocyclone are not understood. This work has
established that CFD modeling can easily provide meaningful
information about the properties of the swirling flow inside a
hydrocyclone where experimental investigations would be utterly
complex.
Comparison between CFD simulation results and measured val-
ues of solids split and separation functions measured on a 100 mm
diameter hydrocyclone permitted conclusion that CFD simulations
are able to predict performance trends, even under extreme roping
conditions.
Through varying feed solids concentration and spigot diame-
ters, a number of distinct operating regimes were produced inside
the hydrocyclone. It was found that solids concentration is more or
less evenly distributed throughout the hydrocyclone. Dense re-
gions can form inside the hydrocyclone, mainly towards the bot-
tom part of the conical section, which dissipate energy through
viscous effects. The swirling flow, with the high tangential veloci-
Fig. 17. Water tangential velocity profile in the cylindrical section of the hydro- ties (G-force) that are the classification driver cannot develop
cyclone, taken 100 mm under the roof.
throughout the entire hydrocyclone body. This can be described
as the hydrocyclone having an apparent, or effective geometry that
differs from the physical one. At the highest solids concentration
tested in this study, the conical section of the hydrocyclone was al-
most entirely filled with a packed bed of particles, so that the
apparent geometry resembled that of a stub cyclone, yielding a
thickening behavior of the hydrocyclone with nearly all the process
water reporting to the overflow.
Overall, this study has established that it is possible to simulate
the behavior of the swirling flow inside hydrocyclone under dense
regimes. Even though it is necessary to continuously improve on
the quality of the physics embedded into the CFD model, it was
found that analysis of simulation results can help understand the
separation trends observed in experiments. A great deal of infor-
mation that is directly relevant to practical operation of hydrocy-
clones can be made from analysis of the flow characteristics
inside the hydrocyclone. It is believed that many important practi-
cal issues such as short-circuiting, the Rf factor or the fish-hook ef-
fect will soon find interesting explanations from analysis of CFD
simulations.

Acknowledgements

The authors would like to acknowledge TOTAL S.A. for funding

this study through a PhD CIFRE program and the scientific support
of C. Leroi and C. Yacono. The simulations were performed at the
High Performance Computing centre CALMIP (Calcul en Midi-
Fig. 18. Influence of spigot diameter on 96 lm diameter particles. Pyrénées) under the project P0406.
 Author's personal copy

A. Davailles et al. / Minerals Engineering 31 (2012) 32–41 41

References Hsieh, K., Rajamani, R., 1991. Mathematical model of the hydrocyclone based on
physics of fluid flow. AIChE Journal 37.
Kelsall, D., 1952. A study of the motion of solid particles in a hydraulic cyclone.
Batrak, O., Patino, G., Simonin, O., Flour, I., Le Guevel, T., Perez, E., 2005. Unlike
Chemical Engineering Research and Design 30, 87–108.
particles size collision model in 3d unsteady polydispersed simulation of
Launder, B., Reece, G., Rodi, W., 1975. Progress in the development of a reynolds-
circulating fluidised bed. In: 8th International Conference on Circulating
stress turbulence closure. Journal of fluid mechanics 68, 537–566.
Fluidized Bed, Hangzhou, China.
Laviéville, J., Boucker, M., Quemerais, M., Mimouni, S., Mechitoua, N., 2006.
Brennan, M., 2006. CFD simulations of hydrocyclones with an air core comparison
NEPTUNE_CFD V1.0 – Theory Manual. NEPTUNE Report H-I81-2006-04377-EN
between large eddy simulations and a second moment closure. Chemical
– Nept_2004_L1.2/3, EDF R& D.
Engineering Research and Design 84, 495–505.
Laviéville, J., Simonin, O., 1999. Equations et modèles diphasiques du Code Astrid
Chang, Y., Ilea, C., Aasen, Ø., Hoffmann, A., 2011. Particle flow in a hydrocyclone
3.4 et du Code saturne polyphasique. Technical Report HE-44/99/041/A, EDF R&
investigated by positron emission particle tracking. Chemical Engineering
D.
Science.
Nageswararao, K., Wiseman, D., Napier-Munn, T., 2004. Two empirical hydrocyclone
Davailles, A., Climent, E., Bourgeois, F., 2012. Fundamental understanding of
models revisited. Minerals Engineering 17, 671–687.
swirling flow pattern in hydrocyclones. Separation and Purification
Neesse, T., Dueck, J., 2007. Air core formation in the hydrocyclone. Minerals
Technology. doi:10.1016/j.seppur.2011.12.011.
Engineering 20, 349–354.
Delgadillo, J., Rajamani, R., 2005. A comparative study of three turbulence-closure
Neesse, T., Schneider, M., Golyk, V., Tiefel, H., 2004a. Measuring the operating state
models for the hydrocyclone problem. International journal of mineral
of the hydrocyclone. Minerals Engineering 17, 697–703.
processing 77, 217–230.
Neesse, T., Schneider, M., Golyk, V., Tiefel, H., Dueck, J., Buntenbach, S., 2004b.
Ergun, S., 1952. Fluid flow through packed columns. Chemical Engineering Progress
Hydrocyclone operation at the transition point rope/spray discharge. Minerals
48, 89–94.
Engineering 17, 733–737.
Galassi, M., Coste, P., Morel, C., Moretti, F., 2009. Two-phase flow simulations for PTS
Ozel, A., Fede, P., Simonin, O., 2010. 3D numerical prediction of gas-solid flow
investigation by means of NEPTUNE CFD code. Science and Technology of
behavior in CFB risers for Geldart A and B particles. In: Proceedings of the 20th
Nuclear Installations 2009, 12.
International Conference on Fluidized Bed Combustion.
Galvin, K., Smitham, J., 1994. Use of X-rays to determine the distribution of particles
Simonin, O., 1996. Continuum Modelling of Dispersed Two-phase Flows
in an operating cyclone. Minerals Engineering 7, 1269–1280.
Combustion and Turbulence in Two-phase Flows (Lecture Series 1996–02).
Gobin, A., Neau, H., Simonin, O., Llinas, J., Reiling, V., Sélo, J., 2003. Fluid dynamic
Von Karman Institute for Fluid Dynamics, Rhode Saint Genese.
numerical simulation of a gas phase polymerization reactor. International
Slack, M., Prasad, R., Bakker, A., Boysan, F., 2000. Advances in cyclone modelling
Journal for Numerical Methods in Fluids 43, 1199–1220.
using unstructured grids. Transactions of IChemE 78, 1098–1104.
Gourdel, C., Simonin, O., Brunier, E., 1999. Two-maxwellian equilibrium distribution
Srivastava, A., Sundaresan, S., 2003. Analysis of a frictional-kinetic model for gas-
function for the modelling of a binary mixture of particles. Circulating Fluidized
particle flow. Powder Technology 129, 72–85.
Bed Technology, vol. VI. Frankfurt am Main, Germany.
Wen, C., Yu, Y., 1965. Mechanics of Fluidization. In: Chem. Eng. Prog. Symp. Ser., p.
Hsieh, K., 1988. Phenomenological Model of the Hydrocyclone. Ph.D. Thesis. Dept. of
100.
Metallurgical Engineering, University of Utah.

View publication stats