212 Wills’ Mineral Processing Technology
space of the thickener, oxidation of liberated parti-
cles may occur in the thickener, which may affect
subsequent processes, especially froth flotation.
The hydrocyclone
This is a continuously operating classifying device
that utilises centrifugal force to accelerate the
settling rate of particles. Its one of the most impor-
tant devices in the minerals industry, its main use in
‘mineral processing being as a classifier, which has
proved extremely efficient at fine separation sizes.
It is widely used in closed-circuit grinding oper-
ations (Napier-Munn et al., 1996) but has found
‘many other uses, such as de-sliming, de-gritting,
and thickening.
It has replaced mechanical classifiers in many
applications, its advantages being simplicity and
high capacity relative to its size. A variant, the
“water-only-cyclone”, has been used forthe cleaning
‘of fine coal (Osbome, 1985) and other minerals
A typical hydrocyclone (Figure 9.13) consists
cof a conically shaped vessel, open at its apex, or
underflow, joined to a cylindrical section, which
hhas a tangential feed inlet. The top of the cyli
drical section is closed with a plate through which
passes an axially mounted overflow pipe. The pipe
is extended into the body of the cyclone by a short,
removable section known as the vortex finder,
roane 5. ]
Figure 9.13 Hydrocyclone (trom Napior-Munn et al,
1996; Courtesy JKMRC, The University of
Queensland)
which prevents short-circuiting of feed directly into
the overflow.
‘The feed is introduced under pressure through the
tangential entry which imparts a swirling motion
to the pulp. This generates a vortex in the eyclone,
with a low-pressure zone along the vertical axis.
An air core develops along the axis, normally
‘connected to the atmosphere through the apex
opening, but in part created by dissolved air coming
‘out of solution in the zone of low pressure.
‘The classical theory of hydrocyclone action is that
particles within the flow pattern are subjected to two
opposing forces an outward centrifugal force andan
inwardly acting drag (Figure 9.14). The centrifugal
force developed accelerates the settling rate of the
particles thereby separating particles according t0
size, shape, and specific gravity. Faster settling
particles move to the wall of the cyclone, where
the velocity is lowest, and migrate to the apex
opening. Due to the action of the drag force, the
slower-settling particles move towards the zone of
low pressure along the axis and are carried upward
through the vortex-finder to the overflow.
Radius of orbting parle
cer
Contrgel oa tore
Moto of pantie
Figure 9.14 Forces acting on an orbiting particle in
the hydrocyclone
‘The existence of an outer region of downward
flow and an inner region of upward flow implies a
position at which there is no vertical velocity. This
applies throughout the greater part of the cyclone
body, and an envelope of zero vertical velocity
should exist throughout the body of the cyclone
(Figure 9.15). Particles thrown outside the envelope
of zero vertical velocity by the greater centrifugal
force exit via the underflow, while particles swept
to the centre by the greater drag force leave in the
overflow. Particles lying on the envelope of zero
velocity are acted upon by equal centrifugal and
drag forces and have an equal chance of reporting
either to the underflow or overflow.
Jorma tetera mereSEpMEENE LGR
Figure 9.15. Distribution of the vertical and radial
‘components of velocity in a hydrocyclone
Experimental work reported by Renner and
Cohen (1978) has shown that classification does
not take-place throughout the whole body of the
cyclone as the classical model postulates. Using
a high-speed probe, samples were taken from
several selected positions within a 150-mm diam-
eter cyclone, and were subjected to size anal-
ysis. The results showed that the interior of the
cyclone may be divided into four regions that
contain distinctively different size distributions
(Figure 9.16)
Essentially unclassified feed exists in a narrow
region A adjacent to the cylinder wall and roof of
the cyclone, Region B occupies a very large part
of the cone of the cyclone and contains fully clas-
sified coarse material, ie. the size distribution is
practically uniform and resembles that ofthe coarse
underflow product. Similarly, fully classified fine
material is contained in region C, a narrow region
surrounding the vortex finder and extending below
the latter along the cyclone axis. Only in the toroid-
shaped region D does classification appear to be
taking place. Across this region, size fractions are
radially distributed, so that decreasing sizes show
Classification 213
Figure 9.16 Regions of similar size distribution
within eycione
maxima at decreasing radial distances from the
axis. The cyclone was nun at low pressure, $0 the
region D may be larger in production units.
Hydrocyclones are almost universally used in
grinding circuits (Figure 9.17) because oftheir high
capacity and relative efficiency. They can also clas-
sify over a very wide range of sizes (typically
5-500m), smaller diameter units being used for
finer classification,
Cyclone efficiency
‘The commonest method of representing cyclone
efficiency is by a performance or partition curve
(Figure 9.18), which relates the weight fraction,
for percentage, of each particle size in the feed
Which reports to the apex, or underflow, to the
particle size. It is analogous to the partition curve
for density separation (Chapter 11). The cut point,
‘or separation size, of the cyclone is defined as the
size for which 50% of the particles in the feed
report to the underflow, i.e. particles of this size
have an equal chance of going either with the over-
flow or underflow (Svarovsky, 1984). This point214 Wills’ Minaral Processing Technology
Figure 9.17 Ball mill n closed cirout with
rocyciones, Batu Hijau mine, Indonesia (Courtesy
JKMAG and JiTech Ply Ltd)
3
Feed appearing in undertow, %
a
Size (um)
Figure 9.18 Partition curve for hydrocyclone
is usually referred to as the dsp size. The sharp-
ness of the cut depends on the slope of the central
section of the partition curye; the closer to vertical
the slope, the higher is the efficiency. The slope
of the curve can be expressed by taking the points
at which 75 and 25% of the feed particles report
to the underflow. These are the dys and dos sizes,
respectively. The efficiency of separation, or the
so-called imperfection J, is then given by
dys— ds
2d
Many mathematical models of hydrocyclones
include the term “corrected dy." taken from the
rected” classification curve. Kelsall (1953)
suggested that solids of all sizes are entrained inthe
coarse product liquid by short-cireuiting in direct
proportion to the fraction of feed water reporting
to the underflow.
For example, if the feed contains 16th"' of
material of a certain size, and 12th reports tothe
underflow, then the percentage of tis size reporting
to the underflow, and plotted on the normal parti-
tion curve, is 75%.
However, if, say, 25% of the feed water reports
to the underflow, then 25% of the feed material will
short-circuit with it; therefore, 4th” of the size
fraction will short-circuit to the underflow, and only
8th leave in the underflow due to classification,
‘The corrected recovery of the size fraction is thus
(0.14)
‘The uncorrected partition curve can therefore be
‘corrected by utilising the equation
ya2ik
I-k
where y’ is the corrected mass fraction of a pa
ular size reporting to underflow, yi the actual mass
fraction of a particular size reporting to the under-
flow, and R is the fraction of the feed liquid which
is recovered in the coarse product stream, The
corrected curve thus describes particles recovered
to the underflow by true classification. It should
be noted that Kelsall’ assumption has been ques-
tioned, and Flintoff et al. (1987) reviewed some of
the arguments. However, the Kelsall correction has
the advantages of simplicity, utility, and familiarity
through long use. Figure 9.19 shows uncorrected
and corrected classification curves.
‘The method of construction of the partition
curve can be illustrated by means of an example.
It is easily performed in a spreadsheet. Suppose
(9.15)09)
1% ood to undertow
eo Ge (
Sizo——
Figure 9.19 Uncorrected and corrected
classification curves
a cyclone is being fed with quartz. (density
2700kgm-*) in the form of a slurry of pulp density
1670kgm_®. The cyclone underflow density is
1890kgm-*, and the overflow 1460kgm™~*.
Using Equation 3.6, the percentage solids by
weight in the cyclone feed is 63.7%. Therefore, the
lution ratio (water-solids ratio) of feed is,
36.3
7
Similarly, the underflow and overflow dilution
ratios can be calculated to be 0:34 and 1,00 respec-
tively.
If the cyclone is fed at the rate of Fth™ of dry
solids and the underflow and overflow mass flow-
rates are U and Vth! respectively, then, since the
total amount of water entering the cyclone must
equal the amount leaving in unit time
O.STF=034U+V
=0.57
\34U + (F-U)
652
The underflow is thus 65.2% of the total feed
weight and the overflow is 34.8% of the feed.
Classification 215
‘The performance curve for the cyclone can now
be prepared by tabulating the data as in Table 9.1.
Columns 1, 2, and 3 represent the sereen anal-
yyses of the overflow and underflow, and columns
4 and 5 relate these results in relation to the feed
‘material. Column 4, €...is prepared by multiplying
the results of column 2 by 0.652. Adding column
4 10 column 5 produces column 6, the reconst-
tuted size analysis of the feed material. Column 8
is determined by dividing each weight in column 4
by the corresponding weight in column 6, Ploting
column 8 against column 7, the arithmetic mean of
the sieve size ranges, produces the partition curve,
from which the dsp (177.5 jm) can be determined,
‘The partition curve can be corrected by utilising
Equation 9.15. The value of R in this example is
65.20.
100x057 ~
Lynch and Rao (reported in Lynch [1977
describe the application of the “reduced efficiency
curve”, which is obtained by plotting corrected
weight percentage of particles reporting to the
underflow against the actual size divided by the
comected dgy (Figure 9.20), and suggests that it
can be used to derive the actual performance
curve after any changes in operating conditions,
the curve being independent of hydrocyclone diam-
ter, outlet dimensions, or operating conditions.
‘A number of mathematical functions have been
suggested to describe the reduced efficiency curve.
‘The commonest are. reviewed by Napier-Munn
eral. (1996),
Although partition curves are extremely useful in
assessing classifier performance, the minerals engi-
neer is usually more interested in knowing fineness
of grind (ie. cyclone overflow size analysis) than
the cyclone dso, A simple fundamental relationship
between fineness of grind and the efficiency curve
of a hydrocyclone has been developed by Kawatra
and Seitz (1985),
39
Mathematical models
of hydrocyclones
Many attempts have been made to capture the key
relationships between hydrocyclone operating and
geometrical variables in models for use in design
and optimisation, with some success. Progress is
being made in using computational fluid dynamics216 Wills’ Mineral Processing Technology
Table 9.1
a @) a @ 0 6) ” 6)
Sice (um) We Wi % offeed Reconstiuted Nominal size Sof
feed (arithmetic feed
uF oF ur On mean) VE
F168 “7 rae 96 5 1000
na wae 142 sss 1000
295-589 250 59 163 24 184 422.0 886
208-295 14 90 48 34 79 2515 ws,
187-208 63 17 41 a 82 1773 500
10-147 43 2 34 39 70 128s 443
74-108 29 79 19 27 46 wo 413
-74 mm 54301289 30.1 = 372
oval 100 100281000
demonstrates some of the process trends well, but
z cannot be used directly in practical design or oper-
: ational situations.
: ‘The most widely used of the published empir
5 ical models are probably those of Pitt (1976)
2 ‘and Nageswararao (1995, though first published in
zg 1978). Plitt published a slightly modified form of
Figure 9.20 Reduced efficiency curve
to model hydrocyclones from first principles (e.g.
Brennan et al,, 2003; Nowakowski et al, 2008)
though this approach is very computationally-
intensive and not yet mature. All the models
commonly used in. practice are sill essentially
‘empirical in nature.
Bradley's seminal book (1965) listed eight equa
tions for the cutsize, and these have increased
significantly since then. Bradley's own equa-
tion based on the equilibrium orbit hypothesis
(Figures 9.14 and 9.13) was:
(525)
Qc(0, — Pr)
where D; = cyclone diameter, p= fuid viscosity,
,— feed flow rate, p,— solids density, py =
fluid density, n= hydrodynamic constant (0.5 for
particle laminar flow), k= constant incorporating
other factors, particularly cyclone geometry.
d, (9.16)
his model in 1980 (Flintoff et al, 1987). These
‘models, based on a phenomenological description
of the provess with numerical constants determined
from large databases, are described in Napier-Mum.
tal, (1996) and were reviewed and compared more
recently by Nageswararao etal, (2004),
Plit’s modified model for the corrected cut size
gg) in microns is:
39.7 D8 DP DI?" 5 exp(0.063C,)
(DAP 40% goss (a3
dae
@.17)
where D,, Dj, Dy, and D, are the inside diame-
ters of the cyclone, inlet, vortex finder, and apex
respectively (cm), 7 isthe liquid viscosity (cP), C,
is the feed solids volume concentration (%), h is
the distance between apex and end of vortex finder
(cm), kis a hydrodynamic exponent to be estimated
from data (default value for laminar flow 0.5),
is the feed flow rate (Umin), and p, is the sotids
density (g/em'). Note that for non-circular inlets,
D,= V4A/a where A (cm?) is the cross-sectional
area of the inlet.‘The equation for the volumewie flow rate of
slurry to the cyclone, Q,, is:
aia all pepe
a exp(O003C,)
B vier Piste pressure drop sro the cyclone in
KPa (1 psi =6.896 kPa). F, and F, in Equations 9.17
and 9.18 are material-specific constants that must
te determined from texts with the fod material
eceoed
Plitt also reports equations for the flow split
between underflow and overflow, and for the effi-
ciency parameter in the reduced efficiency curve.
Ncpevvacras' model tna eas for
eurecied cot tan, preeame-low ‘ale, end flow
‘split, though not efficiency. It also requires the esti-
mate of feed-specific constants from data, though
first approximations can be obtained from libraries
‘of previous case studies, This requirement for
feed-specific calibration emphasises the important
effect which feed conditions have on hydrocyclone
performance.
Asomah and Napier-Munn (1997) reported an
empirical model which incorporates the angle of
inclination of the cyclone, as well as explicitly the
slurry viscosity, but this has not yet been validated
into largestlo use which bus been enjoyed by
the Pit and Nagenwararao models
“asf general approximation forthe flow rate
i hydrocyclone
095 x 10°VP D
Flow rate and pressure drop together define the
useful work done in the cyclone:
(9.18)
(9.19)
PQ
Power =o kW (9.20)
= flow rate (m°/h), P= pressure drop
(kPa), and D= cyclone diameter (cm). The power
can be used as a first approximation to size the
‘pump motor, making allowances for head losses
and pump efficiency.
‘These models are easy to incorporate in spread-
sheets, but are particularly useful in process design
and optimisation using dedicated computer simula-
tors such as JKSimMet (Napier-Munn et al., 1996)
‘and MODSIM (King, 2001), or the flowsheet simu-
lator Limn (Hand and Wiseman, 2002). They can
also be used as a virtual instrument or “soft sensor”
(Morrison and Freeman, 1990; Smith and Swartz,
Classification 217
1999), inferring cyclone product size from geom-
etry and operating variables as an alternative t0
using an on-line sizer (Chapter 4).
Scale-up and design of
hydrocyciones
A preliminary scale-up from a known situation (eg.
a laboratory or pilot plant test) 19 the unknown
(eg. a full production installation) can be done via
the basic relationships between cut-size, eyclone
diameter, flow rate, and pressure drop. These are:
amo) (4) (3) Ge)
(21)
and Fa($
(9.22)
where P is pressuite drop, Q flow rate, D cyclone
diameter, the subscripts | and 2 indicate the known
‘and scale-up applications respectively, and n,_ are
constants which are a function of the flow condi-
tions. The theoretical values (for dilute slurries and
particle laminar flow in small cyclones) are: m=
15, m=0.5, ny=0.5, 1, 0, and
ng =4.0, The constants to be used in practice will
depend on conditions and which particular model
is favoured. In particular, high feed solids concen-
trations will substantially influence both cut size
(Gncrease) and pressure drop at a given flow rate
(reduce). There is no general consensus, but in most
applications the following values will give more
realistic predictions: ny = 1.54, n,=0.43, m=
0.72, ny=0.22, ns =2.0, and n,=3.76,
‘These relationships tell us thatthe diameter, flow
rate, and pressure must be considered together
For example, cut-size cannot be scaled purely on
cyclone diameter, as a new diameter will bring
either a new flow rate or pressure or both. For
example, if it is desired to scale to a larger eyclone
at the same cut size, then dq, = dy, and Dy =
D,(P,/P,)™!™.
Classification efficiency can sometimes be
improved by arranging several cyclones in series
to re-treat overflow, underflow or both. Svarovsky
(1984) has pointed out that if V cyclones with iden-
tical classification curves are arranged in series,
cach treating the overflow of the previous one, then