Flow Split and Water Split

in Hydrocyclone Classifiers

NFTDC Internal Report No. NFTDC/ TR - 20/ 1999-2000

March, 2000
(For limited circulation only)

This Report may not be circulated except with the permission of NFTDC. While
parts of this report may be freely quoted for Research / Academic purposes, rights
for commercial exploitation of any matter contained in this report are reserved.
 FLOW SPLIT AND WATER SPLIT IN INDUSTRIAL HYDROCYCLONES

K. NAGESWARARAO
NFTDC, Hyderabad 500 058, India. E-mail: [email protected]

ABSTRACT

The distribution of feed water into the products is an important characteristic to describe the
performance of hydrocyclones. Significant work on this topic is reported in literature. In this
paper, the features including the assumptions and approximations in developing the
equations for flow split and water split are reviewed and consequential implications are
discussed. The methodology used for the development of the generalised equations currently
in use in JKSimMet is outlined.

It is shown that the ‘feed material characteristics’ is a significant variable, which affects the
performance the flow split and water split. It is also shown that equations for water recovery
to underflow, Rf, are preferable to flow split, S. Finally, the usefulness of the generalised
equations for scale-up is illustrated..

Key words
Hydrocyclones, classification, modelling, simulation

INTRODUCTION

For a complete mass balance of the hydrocyclone feed slurry, it is necessary to account for
the distribution of water too into the products. Conventionally, this is characterised by Rf,
the ratio of mass flow rate of water in underflow to that in the feed. Other performance
measures used in literature include the flow split S and volumetric flow ratio, Rv, which are
given below -

Mass flow rate of water in underflow

R (%) = 100
f Mass flow rate of water in feed

Vol. rate in underflow

S (%) = 100
Vol. rate in overflow

S
Rv (%) = 100
S + 100
 2

Rv − Rs β
Clearly, R f = , where β is the volumetric fraction of feed solids and Rs is the solids
1− β
recovery to underflow in percent. Specifically, for dilute feed suspensions, Rv is
approximately equal to Rf.

For industrial hydrocyclones, the two comprehensive models, which have been discussed
extensively in literature, are those due to Lynch and Rao (1975) and Plitt (Plitt, 1976 and
Flintoff et al, 1987). However, Heiskanen (1996) notes that neither the Plitt model nor the
Lynch-Rao model can predict the flow split accurately. More recently, Chen et al (2000) too
opined that the lack of capability to predict the flow split is the major deficiency for all
models analysed by them, which included those of Plitt and Svarovsky (1984). However, they
had not analysed the models developed at the Julius Kruttschnitt Mineral Research Centre
(JKMRC).

It is also relevant to mention that a generalised model for cyclones (Nageswararao, 1978) has
been used with considerable success (Lynch, 1984, Lynch and Morrell, 1992, Napier-Munn
and Lynch, 1992, Napier-Munn et al, 1996). Although, the model equations are now
available in the literature (Nageswararao, 1995 and Napier-Munn et al, 1996), the details
regarding model development are not published as yet.

In this paper, the fourth in a series (Nageswararao, 1999a, 1999b and 2000) on the modelling
of industrial hydrocyclones, a comparative analysis of the equations available for flow split
and water split is presented. The basis for and the methodology of the development of the
generalised equations incorporated in JKSimMet are briefly discussed. It is shown that
equations for Rf are preferable to equations for S (or Rv) for simulation of hydrocyclone
performance. The usefulness of the generalised equations for design and scale-up is also
illustrated.

FLOW SPLIT AND WATER SPLIT - THE EQUATIONS

The relationships available in literature include equations for water rate in overflow, Wo, in
terms of feed water, Wf and spigot diameter, Du (Rao, 1966 and Lynch and Rao, 1968) and
for water recovery to underflow, Rf, using the same methodology (Lynch and Rao, 1975).
Plitt (1976) preferred flow split, S as an independent characteristic for his comprehensive
cyclone model. The equations for Rf and Rv developed by this author (Nageswararao, 1978,
1995), Asomah’s equation for Rf (Asomah and Napier-Munn, 1997) and Medronho’s
relation for Rv (Medronho and Svarovsky, 1984, Svarovsky, 1987) are all based on
dimensional reasoning.

The salient features of the more important equations, specifically, the basis, assumptions
(explicit and implicit) and approximations are outlined below.
 3

Lynch-Rao equation

For industrial hydrocyclones, the first correlation for water rate in overflow and from it for Rf
is attributable to Rao (Rao, 1966 and Lynch and Rao, 1968). The observations of Peachey
(1960) and De KoK (1962) that a linear relationship exists between the flow rate of water in
overflow, Wo, and the corresponding rate in feed, Wf, formed the basis for his equation. The
structure for Lynch–Rao equation (1968) took the form -

Wo = 1.08 Wf - 10.0 Du + K1 (1)

where the water rates are in tons/hr, Du is in inches and K1 is a constant dependent on the
feed material.

Rao and Deb Kanungo (1973) later generalised equation (1) by including De Kok’s data as

Wo = 0.97 Wf - 4.30 Du - 0.53 (2)

Later, based on extensive experimental work, Lynch and Rao (1975) retained the same
structure for the relationship between Wo and Wf. Significantly, they reported that the
coefficients in the above equation are dependent on characteristics of feed solids including
the size distribution.

Plitt Equation

As summarised by Bradley (1965), early researchers (for example, Stas, 1951) working with
dilute feed slurries developed correlations for S assuming it to be dependent on Du/Do only.
However, the dependence of S on throughput was incorporated later as below -

S ∝ (Du/Do)p Q-q (3)

The coefficients p and q are 4.4 and 0.44 according to Moder and Dahlstrom (1952) and 1.75
and 0.75 according to Bradley. Plitt (1976) too considered the flow split S to be the
performance factor that could be correlated to the design and operating conditions. He
generalised equation (3) above (Plitt1) and assumed it to include all the factors, which were
identified to influence cyclone performance. A calibration factor F4 was later incorporated
by Flintoff et al (1987). The current form of the equation (Plitt2) is -

0.54φ
18.62 ρ 0.24 ( D / D )3.31 ( D 2 + D 2 ) 0.36 e
p u 0 u o (4)
S=F
4
D1.11P 0.24
c
 4

Yet, it is Rf, which is ultimately required for subsequent calculations of the cyclone
performance. For dilute suspensions we could take Rv = S/(1+S) ≅ Rf. However, estimation
of Rf for dense feed slurries, usually encountered in industrial practice, using Plitt’s equation
is a complex procedure as given below (Hinde, 1977, Plitt et al, 1990). The additional data
required for this method includes mass fraction in particle size class i in feed, fi, the Kelsall
classification size, d50c and the sharpness of classification, m .

n − 0.6931(d / d )m
S /(1 + S ) − (φ / 100) 1 − f e 50c
i
1
R = (4a)
f
n − 0.6931(d / d )m
1 − (φ / 100) 1− f e 50c
i
1

Further, it may be noted that to use Plitt’s equations (4) and (4a), both feed pressure, P and
throughput, Q are required. This is because the equation for S includes P as an independent
variable and those for d50c and m include Q. This means that for a better estimate of Rf, both
P and Q need to be measured. Alternately, when one of them is estimated from model, error
propagation is inevitable.

It is also relevant that Cilliers and Hinde (1991) noted that Plitt2 (equation 4) does not fully
take into account the feed solids concentration even after ‘calibration’. They proposed a
provisional revision with coefficients of 1.80 for Du/Do instead of 3.31 and 7.87 instead of
0.544 for the solids concentration term, φ. The obvious conclusion is that Plitt2 overestimates
the effect of Du/Do and underestimates that of φ, at least as far as their data is concerned.

The generalised equations

A generalised model for industrial hydrocyclones, which includes equations for Rf and Rv was
attempted (Nageswararao, 1978) to evaluate the dependence of the major design and
operating variables on Rf. These included the

• reduced vortex finder, Do/Dc

• reduced spigot, Du/Dc
• reduced inlet Di/Dc
• reduced length of the cylindrical section Lc/Dc and
• cone angle, θ

The diameter of the cyclone, Dc was the obvious choice as characteristic dimension of length.
 5

The operating variables considered were

• the hindered settling factor, λ , where λ = β /(1 − β ) 3 , β being the volumetric

fraction of solids in the cyclone feed, and

• {P/ (ρp g Dc)} to account for the force field generated in the cyclone.

When only the above variables are considered, we can write

Du Do Di P L
Rf = f ( , , , , c ,θ , λ ) (5)
Dc Dc Dc ρ p gD c Dc

Under the explicit assumption that the variables are separable and could be expressed as
monomial power functions and that the independent effect of unquantifiable feed
characteristics is separable1 from machine characteristics, we get

a b c d f
Du Do Di Lc P
R f = K w1 θ e
λg (6)
Dc Dc Dc Dc ρ gDc

where Kw1 incorporates the effect of all those variables, which were not considered in the
formulation of the above, in addition to the feed characteristics.

For a system where the variables, are only Do, Du, feed pressure and pulp density of the feed
slurry only, the above could be reduced to

a b f
Du Do P
R f = K w2 λg (7)
Dc Dc ρ gDc
where

c d
Di Lc
K w1= K w 2 θe (8)
Dc Dc

1
As Napier-Munn et al (1996) emphasise, the JKMRC simulation models aspire for a clear goal that of
separating ore characteristics from those of the processing machine. The generalised cyclone model
(Nageswararao, 1978) is the first of those developed at JKMRC, which explicitly decoupled the machine and
material dependence. Later, the work by W J. Whiten and his students Awachie (1983) and Narayanan (1985)
followed, leading to the development of material specific breakage functions.
 6

Calibration

The model equations as given by equations (7) and (8) can be meaningful, only if the
numerical values of a, b, . . . g can be determined. As the hydrodynamics within the cyclone
is not developed enough to evaluate these from theoretical considerations, experimental data
could be used to calibrate2 the model.

The parameters a, b, f and g in equation (7) were evaluated from a dataset (39 tests) treating
limestone of fixed size distribution ( 65% passing –53 µm, referred to as FINE by Lynch and
Rao, 1975) in a 38.1 cm hydrocyclone.

Also, the database of Nageswararao (1978) for 38.1, 25.4, 15.2 and 10.2 cm cyclones is
complementary to that of Lynch and Rao (1975) in that only inlets were different. With each
cyclone, tests were carried out with variations in vortex finder, spigot feed pressure and solids
concentration. The feed material was MEDIUM limestone (containing 50% -53µm). Kw2
values for different inlets for each cyclone were then calculated from which the coefficient c
could be estimated.

The dependence of cyclone length and cone angle, that is the coefficients d and e, were
evaluated from data obtained on a 15.2 cm hydrocyclone, where these two variables were
changed. Feed material was MEDIUM limestone as above.

The resulting equation is –

−1.19 2.40 −0.50 0.22 −0.53

Du Do Di Lc P
R f = K w1 θ − 0.24
λ0.27 (9)
Dc Dc Dc Dc ρ p gDc

Following an identical methodology, an equation for Rv was also developed. As we can see,
the effect of λ is not significant on Rv.

−0.96 1.83 −0.25 0.22 −0.31

Du Do Di Lc P
R v = K v1 θ − 0.24
(10)
Dc Dc Dc Dc ρ p gDc

We should also mention that equations (9) and (10) as such are suitable for predicting relative
changes in Rf (or Rv) with relative changes in the independent variables. However, we still
need to tune the equations for predicting the actual values for any specific feed material. In

2
The term calibration has a different meaning here than that used by Flintoff et al (1987) and Napier-Munn et al
(1996). For a detailed discussion, Nageswararao (1999b) may be referred.
 7

principle, data from a single test is sufficient to determine Kw1 and Kv1 so that the equations
are ready for prediction of Rf and Rv for the specific feed material.

For scale up applications, it is necessary to evaluate the independent effect of cyclone

diameter also, as it is not adequately accounted for. During testing of the equations on
experimental data, mean values of Kw1 and Kv1 for cyclones of different sizes treating same
feed solids did not show any significant diameter dependence. As such, tentatively, it was
proposed that they are independent of cyclone diameter (Nageswararao, 1978).
Mathematically, we can write

Kw0 ∝ Kw1 {Dc0.00} and

Kv0 ∝ Kv1 {Dc0.00}

where Kw0 and Kv0 are material dependent constants. Strictly speaking, they are independent
of the variables, which have been considered, but depend on those which have not been
included.

Other equations

Other equations in literature are due to Medronho (Svarovsky, 1987), Brookes et al (1984)
and Vallebuona et al (1995a and 1995b). Medronho’s equation takes the form

4.75
D
Rv = 1218 u EU −0.30 (11)
Duc

with a separate equation for Euler’s Number in terms of Reynolds’s number, feed solids
concentration and dimensionless inlet size, Di/Dc. It is relevant to note that in the
experimental data used for estimating the exponents in Equation (11), the feed materials
consisted of chalk and alumina hydrate. The implicit assumption is that Rv is independent of
feed material characteristics.

Brookes et al assume that

x
Du
Rf = K (12)
Du + Do

Rouse et al (1987) too confirm the above and note that the constant K and the coefficient x
are machine and material specific, thereby limiting the range of applicability. The equations
 8

proposed by Vallebuona et al are also subject to the same limitation. There are other
equations too whose range of applicability is limited (for example, Arterburn, 1982).

DISCUSSION

In developing an equation be it empirical or theoretical, assumptions (explicit or implicit)

regarding the behaviour of the physical system are made. Clearly, it is these, which bind
together the dependent and independent variables in the equation. Obviously, the closer the
assumptions are to the physical reality the greater is the predicting power and range of
applicability.

The assumptions include the variables/factors to be considered and the way they are
correlated (equation structure). It is implied that not including a certain variable in the
equation, by itself, forms a critical assumption.

Specifically with regard to the theme of this paper, there are two distinct equation structures,
the linear type (Lynch and Rao) and the power functions. A comparative analysis of the
different equations available and their predictive power is best carried out by analysing the
mathematical structure of the equations and the assumptions and approximations.

Mathematical representation

The factors identified by Lynch and Rao are Wo, Wf and Du. Their obvious assumption is that
changes in Wf and Du proportionately change the water rate to overflow.

A major limitation to the Lynch and Rao structure is the dimensional inhomogeneity. This
factor could be confusing and misleading too in attributing physical significance. The
problem is compounded in view of the inherent inability of collecting highly accurate data in
industrial circuits.

For example, consider equation (2) proposed by Rao and Deb Kanungo (1973) which is
claimed to be valid for the data of Lynch and Rao (1968) as well as de Kok (1956). From a
comparison with the original equation (1), we can note that with every 1 inch (2.54 cm)
increase in spigot, a decrease in 10 tonnes/hr and 4.30 tonnes/hr in Wo, are predicted from the
two different equations. Obviously both cannot be true for the same data. This difference is
clearly attributable to the range of data from which the two equations were developed.

Additionally, a new set of equations is proposed later (Lynch and Rao, 1975) indicating the
limited applicability of the generalisation of Rao and Deb Kanungo. Clearly, should we
choose Lynch-Rao structure, we could expect changes in Wf and Du proportionately
changing the water rate to overflow in general agreement with the experimental observations
(Lynch, 1977, Gault and Nageswararao, 1981). However, the proportionality constants shall
 9

be data specific, thereby making Lynch-Rao procedure a mere curve fitting exercise for any
new system.

This could possibly be the reason, why Cilliers and Hinde (1991) consider that Plitt model,
despite its limitations, is better than Lynch-Rao. There are other reports too of unsuccessful
validity (Hinde, 1999, Coelho and Medranaho, 1993) when an attempt is made to extend the
range of applicability.

On re-arrangement of the expression proposed by Rao and Deb Kanungo (1973), it was noted
(Nageswararao, 1978) that water rate in underflow, Wu, can be expressed as

Wu = 0.03 Wf + 1.70 Du + 0.52 (13)

where Du is in cms and the flows in tonnes/hr.

That is, Wu = 0.03 (Wf - meanWf) + {1.7 Du + 0.52 + 0.03 meanWf}

where meanWf is the mean water flow rate for the data from which equation (2) was derived.

An equation for Rf was then rewritten as

Rf (in %) = 100 Ks/ Wf + 3 (1 - meanWf/ Wf )

where Ks = 1.7 Du + 0.52 + 0.03 meanWf

For the data of Rao, where Wf varies from 70 –110 tonnes/hr for a majority of tests, the above
reduces to

Rf (in %) = 100 Ks/ Wf ± 1 (approx.) (14a)

As a simplification the above could be approximated as

Rf = 100 Ks/ Wf (14b)

We can note that the absolute error in prediction of Rf using equations (14a) and (14b) is of
the order of ± 3 even at the extreme values where Wf differs from the mean value by a factor
of 2.

However, if Ks is assumed to be a function of spigot only and writing Wf as Q (1-fs), we

could assume a structure for Rf similar to Q. It is apt to remark that such an analysis led to
the generalised equation structure later (Nageswararao, 1978).
 10

Interestingly, Rao and colleagues (Bandyopadhyay et al, 1985) continued using the original
structure. For example, they reported that, equation (1) with a site specific K could be used
for the cyclone at the Rakha Concentrator of Hindustan Copper Ltd. It should be noted that
there was apparently no change in spigot.

All other equations, namely Plitt, Nageswararao, Asomah, Brookes et al etc essentially have
the same structure. The factors (variables) which have been taken into consideration and the
way they are bound together in the equation structure however vary.

The distinguishing feature in the methodology proposed by Brookes et al and Vallebuona et

al is the assumption that the model parameters are application specific. That is, again a
curve fitting exercise similar to Lynch-Rao, the only difference being the structure of the
equations.

However, in the equations proposed by Plitt, Nageswararao and Asomah, the model
parameters are not application dependent. These however include material specific constants,
which must be determined from experimental data. Some salient differences are discussed
below.

Interaction of variables

A significant difference between the generalised equation (s) (Nageswararao) and Plitt’s
equation concerns the interactive nature of the effects of Do and Du. In quantitative terms,
Plitt’s model predicts different S values although, we change the spigot by the same amount
for different vortex finders. For example, let us consider a 10% increase in Du. The
generalised equation predicts an increase of 26% in Rf, irrespective of what the other
variables are. However, the increase in S predicted by Plitt’s equation is dependent on the
vortex finder too.

The distinctly different perceptions in modelling the physical reality, hence the assumptions
between Plitt and the author may be noted.

Effect of feed inlet

Plitt explicitly ignored the independent effect of inlet on flow split, while this is identified as
an independent variable by Nageswararao and Asomah. So far, the geometry of the inlet
shape is approximated and the effect of inlet is assumed to be the same as that of a circle of
equal area. There are indications that the flow regime could be affected by the inlet shape
and geometry (Rogers, 1998).
 11

Effect of cyclone length

Plitt (1976) considered the free vortex height, h as an independent factor in his equation, thus
simplifying the effect to be of the same magnitude whether due to change in cone angle or the
cylinder length. However, a clear distinction of the effects of Lc and θ on Rf is made in the
later models developed at JKMRC (Nageswararao, Asomah and Napier-Munn).

Effect of angle of inclination of the cyclone

Asomah is to be credited for quantifying the effect of angle of inclination of the cyclone as a
variable (Asomah and Napier-Munn, 1997). We can observe that their data are in conformity
with their conclusions that the larger diameter cyclones are more effected than the smaller
cyclones (Asomah and Napier-Munn, 1996). However, the major limitation is that the model
structure does not represent the observations adequately. For fixed geometry cyclones, their
predictive equation takes the form –

Rf ∝ exp (- k α/180)

where, k is a constant.

Clearly, the mathematical function they assumed, predicts that changes in (∆Rf/Rf) with
changes in angle of inclination (∆α/α ) to be independent of the cyclone diameter, in
contradiction to the experimental observations.

Effect of feed material characteristics

Understandably, the original equation proposed by Lynch and Rao (1968) did not include the
effect of size distribution of feed as a variable, although, a material specific constant was
incorporated. A bolder conjecture later followed when a universal equation independent of
feed characteristics (including the material and its size distribution) for water in overflow was
proposed by Rao and Deb Kanungo (1973).

Significantly, the equations proposed by Lynch and Rao (1975) implicitly take into account
feed characteristics.

A distinctive feature of Plitt’s original model (1976) is the implicit assumption that cyclone
performance is independent of feed material, for example copper ore or silica. Additionally,
Plitt’s equations explicitly ignores the effect of feed size distribution even when the same
feed material (for example, silica or limestone) is used. Implicitly this amounts to a further
approximation that cyclone performance is independent of the feed size distribution as well.
It is not only logical to expect cyclone performance to depend on the feed size distribution
 12

but that it is actually so is clearly shown by Lynch and Rao (1975)and Hinde (1985). The
model parameters, that is the regression coefficients accordingly correspond to these
approximations.

However, when it was recognised that the ‘feed material characteristics’ is a significant
variable, Flintoff et al (1987) suggested incorporation of what they termed as the ‘calibration
parameters’ a posteriori in the model equations (Plitt2).

A significant assumption of Asomah (Asomah and Napier-Munn, 1997) concerns the feed
characteristics. In combining the data from different materials with widely varying
characteristics (specific gravity and feed size distribution), for evaluating their model
parameters by regression, they implicitly made an ad hoc assumption that P40, the 40%
passing size of feed solids, solids density, ρs, and the viscosity of slurry, µsl, represent
completely the dependence on cyclone performance. This could be considered more rigorous
than Plitt (1976).

However, while Flintoff et al (1987) have introduced the so-called ‘calibration parameters’
only after the model predictions were found to be unsatisfactory, Asomah (Asomah and
Napier-Munn, 1997) introduced the so-called ‘system constants’, at the same time of
introducing his model. Presumably, the feed material characteristics could not be totally
taken care of by P40, ρs and µsl, in view of the extensive validation on databases as reported.

To summarise Plitt, Asomah and Svarovsky essentially approximate the model parameters
obtained from data treating multiple feeds to be the same, as what would be obtained with a
single feed. The consequential errors in model parameters with this approximation are
discussed in detail elsewhere (Nageswararao, 1999b).

COMPARISON OF MODEL PREDICTIONS

In view of the contentious assumptions and approximations in developing, Plitt’s model

discussed earlier, we have investigated quantitatively the efficacy of Plitt’s equation.

We may recall that the industrial database for evaluation of model parameters (regression
coefficients) of Plitt model is exclusively that of Rao (1966). As such, we should expect that
Plitt model predictions for Rao’s data should be more accurate than any other model.
Accordingly, calculations of the so-called calibration parameters for equation for S are done
for the Rao data of 20-inch cyclone treating silica. Although at least one data set is necessary
and sufficient too (in principle) for further predictions, the mean value of this parameter for
all the data is calculated. This mean value is then used for estimating S for the model fitting
data. The results are shown in Figures 1.
 13

The generalised model includes an equation for the volumetric flow split Rv as well in
addition to the water recovery to underflow Rf. Since S = Rv / (1- Rv), this characteristic too
could be predicted. Mean Kv1 for the data is calculated and the results of computation are
shown in the same figure for comparison. The flexibility of the generalised equation
(Nageswararao, 1978, 1995) can be seen in the fact that this characteristic also could be
directly estimated with reasonable degree of accuracy.

For the sake of completeness, we show in Figure 2, the predicted values of Rf against the
experimentally determined values for the data of Rao.

For a broader comparison, it is possible to use the relation of Shi and Napier-Munn (1996), to
estimate viscosity of the feed slurry and hence Rf using Asomah’s equation. However, with
such a procedure, we would not be able to distinguish errors in prediction due to propagation
of errors in estimated viscosity and those due to Asomah’s equation itself. As such, no
comparison of Asomah’s equation is attempted, since Rao’s data does not include measured
viscosity.

In this context, we may mention that the data of Hinde (1985) too corroborates that the
generalised equations for Rv and Rf are satisfactory. The results of computation shall be
communicated in due course (Nageswararao and Hinde, 2000). Significantly, the data
demonstrate clearly that size distribution of feed solids is a significant variable.

Scale-up and design applications

Lynch-Rao (1975) database includes tests on 10.2, 15.2, 25.4 and 38.1 cm hydrocyclones
treating limestone with a feed size distribution of 40% passing –53 µm, described as
COARSE. As these data have not been used in evaluating the model parameters, the
efficacy of the generalised equations for scale-up could be validated with these data.

We have calculated Kw0 (=Kw1) and Kvo (= Kv1) from the data of 38.1cm hydrocyclone.
These values are then used for predicting Rf and Rv at different design and operating
conditions for the above database. The results of computation are shown in Figures 3a and
3b. We may note that the predicted Rv values are about 15% higher for 25.4 cm cyclones
when Kvo obtained from 38.1 cm unit is used for calculations. Similarly, Rf Values are about
30% less for 15.2 cm cyclones and 5% less for 10.2 cm cyclones, if the predictions are done
either with 38.1 or 25.4 cm cyclones.
 14

70

60
Predicted flow split ( %)

50

40

30
SPlitt
20
SNageswararao
10

0
0 10 20 30 40 50 60 70
Observed flow split (%)

Figure 1. Prediction of flow split for model fitting data (Rao,

1966) of Plitt
 15

40
35
30
Predicted Rf (%)

25
20
15
10
5
0
0 5 10 15 20 25 30 35 40
Observed Rf (%)

Figure 2 Prediction of water recovery to underflow using the

generalised model (Nageswararao, 1978) for the data
Rao (1966)
 16

60

50
Cyc. Dia
Predicted Rf

40 38.1 cm
25.4 cm
15.2 cm
30 10.2 cm

20

10
10 20 30 40 50 60
Observed Rf

Figure 3a. Model predictions from 38.1 cm cyclone for

COARSE limestone
 17

60

50
Predicted Rv

40 Cyc.Dia
38.1 cm
25.4 cm
30 15.2 cm
10.2 cm

20
20 30 40 50 60
Observed Rv

Figure 3b. Model predictions from 25.4 cm for COARSE

limestone
 18

For the data with MEDIUM limestone of Lynch and Rao, such clear anomalies are not
apparent as can be seen from Figures 4a and 4b. Here Kw0 is determined from 38.1 cm
cyclone data, while Kvo is determined from 10.2 cm unit.

We may also mention that the installation of 15” cyclones based on a priori model predictions
at Newcrest Mining Ltd’s Celebration plant is a typical example of selecting optimum design
geometry for a specified duty. As reported by Lynch and Morrell (1992), post installation
sampling surveys confirmed that observed solids in overflow of 34.9% compare favourably
with 34.5% from model prediction

Sources of error in model predictions

Apart from the unavoidable sampling and analysis errors in the collection of base data, the
major source of error in estimating Kw0 and Kv0 is the error in measurement of the spigot size.
This would become significant particularly when we are scaling up from small diameter
cyclones (say, 10.2-15.2 cm). For example, let us consider a case when the true spigot size
is 26 mm. Even when a value of 25-mm (assumed from the manufacturer’s specifications) is
used in calculations, the error in Kwo and hence in predicted Rf values would be of the order
of 10%. The systematic error would be 20%, should the true value be 27mm.

It is thus not surprising that Napier-Munn et al (1996) caution specifically in this regard,
“all internal dimensions of the cyclone must be measured and not assumed. Normal wear,
and maintenance errors can lead to significant discrepancies in cyclone dimensions”.

However, errors in estimation of Rf do not contribute to errors in the prediction of product

(overflow) size distribution (Nageswararao, 1999b), although they are propagated to the mass
flow rates of solids in the products and hence to pulp densities.

General remarks

No discussion on cyclone models for industrial applications is complete without a mention of

the unique contribution of Lynch and Rao. For the sake of completeness, we put on record
the following, which are specifically relevant to the theme of this paper.

The most significant contribution of Lynch and Rao is undoubtedly their original model
published in 1968, which paved the way for modern simulation methods and their application
to industrial circuits (Lynch, 1977). Their exhaustive industrial database (Rao, 1966) also
forms the ‘hardcore’ for the original Plitt model (Plitt, 1976).
 19

50

40
Rf Predicted

Cyc. Dia
30
38.1 cm
25.4 cm
15.2 cm
20 10.2 cm

10
10 20 30 40 50
Rf Observed

Figure 4a. Model predictions from 38.1 cm cyclone for

MEDIUM limestone
 20

50

40
Rv Calculated

Cyc. Dia
30 38.1cm
25.4 cm
15.2 cm
10.2 cm
20

10
10 20 30 40 50
Rv Observed

Figure 4b. Model predictions from 10.2 cm cyclone for

MEDIUM limestone
 21

Further, Lynch and Rao (1975) explicitly identified that ‘feed characteristics’ is a significant
variable, which affects the cyclone performance. This formed an important assumption for
the later model by this author (Nageswararao, 1978)3. Their cumulative database provided
the basis for testing the hypothesis that the effect of feed size distribution on cyclone
performance is separable.

In this context, we may emphasise that so far as applicability for industrial units is concerned,
the accuracy of the model parameters for Plitt’s equation (s) is almost wholly dependent on
the precision of their earlier database (Rao, 1966) while for the generalised model
(Nageswararao) their more extensive data (Lynch and Rao, 1975).

Finally, it is apt to remark that we can not describe any model (or equation) as true or false,
although we may attribute theoretical, empirical, phenomenological or any other qualifier
depending on the methodology adopted in arriving it. All models are subject to the limitation
that they are merely approximations of the physical reality, based on simplifying assumptions
(hypotheses). Moreover, errors in data if used, for evaluating the model parameters, are
carried forward. Consequently, model predictions cannot be relied upon totally. The only
yardstick for comparison is how useful the model is for our objective.

To that end, it appears that the generalised equations (Nageswararao) have proved their
usefulness in control, and design applications (Napier-Munn et al, 1996). However, they are
still inadequate to the extent that an extremely useful control variable, namely the inclination
of cyclone is not accounted for. Fortunately, the equation structure permits further
refinement to incorporate this variable as well as any other factors which may be discovered
in future.

SUMMARY

1. Dimensional inhomogeneity and the choice of structure restrict the generalised

applicability of Lynch-Rao equation beyond site-specific situations. Similarly, the
equations proposed by Brookes et al and Valebuona et al too are site-specific.

2. ‘Feed material characteristics’ is a significant variable, which affects the performance of

industrial hydrocyclones including the flow split and water split. It is necessary to
consider this factor in prediction equations for better accuracy.

3
The author is privileged to have the guidance of Professors A. J. Lynch and T. C. Rao, from whom he learnt
the ABC-s of hydrocyclones during his Graduate Studies in 70-s. Their exhaustive database made all the
difference to evaluate the validity of various assumptions and approximations in the development of a
generalised model (Nageswararao, 1978 and 1995).
 22

3. The equations proposed by Plitt and Svarovsky are based on an implicit assumption that
the effect of feed material characteristics could be ignored. However, Plitt’s equation is
later modified by ad hoc inclusion of a material-specific calibration parameter.
Additionally, using Plitt’s equation to calculate Rf from S involves a convolute procedure
resulting in possible error propagation. Consequently, for applications in closed grinding
circuits, equations for direct estimation of Rf are preferable to equations for S from which
it is to be calculated.

4. The inclusion of material-specific ‘system constant’, a posteriori in the equation for Rf by

Asomah indicates that the effect of feed material characteristics could not be totally
accounted by P40, ρs and µsl. Furthermore, the equation structure assumed to quantify the
effect of cyclone inclination is inconsistent with the experimental observations.

5. The equations for Rf and Rv proposed by this author are not subjected to tests as yet by
independent researchers, though to date, it appears that their usefulness is proved at
JKMRC and the users of JKSimMet. There is scope for further improvements in
prediction.

6. There is a need to study further the effect of inclination of the cyclone, which appears to
be an extremely useful control variable.

ACKNOWLEDGEMENTS

I am grateful to Dr. Ch. Sridhara Rao, Director II, C.C. R&D, Hyderabad, for going through
the manuscript and suggesting improvements.

My sincere appreciation goes to Dr Adrian Hinde, Specialist, Particulate Systems, MINTEK,

Randburg, South Africa. E-interaction with him continues to be immensely valuable to my
Research Programme.
 23

REFERENCES

Arterburn, R.A., The sizing and selection of hydrocyclones, In Design and Installation of Comminution
Circuits, Eds., A. L. Mular and G.V. Jergensen, AIME, New York, 1982, Ch. 32, pp 592-607.

Asomah, I. K and Napier-Munn, T. J., The Performance of inclined hydrocyclones in mineral processing, In
Proceedings -Hydrocyclones '96, Eds., D. Claxton, L. Svarovsky and M.T. Thew, Cambridge, (Mech Eng pubs,
London), 1996, pp 273 – 287.

Asomah, I. K and Napier-Munn, T. J., An empirical model of hydrocyclones, incorporating angle of cyclone
inclination, Minerals Engineering, 1997, 10(3) 339 - 347.

Awachie, S.E.A., Development of crusher models using laboratory breakage data, Ph.D. Thesis, University of
Queensland, 1983, as cited by Napier-Munn et al, op cit.

Bandyopadhyay, A., Sanyal, P., Rao, T.C and Khare, R. N., Process analysis & development of model for the
grinding & Classification circuit at Rakha concentrator of Hindustan Copper Limited, 15th International Mineral
Processing Congress, Cannes, 1985, Vol. III, 144–153.

Bradley, D., The Hydrocyclone, 1965, Pergamon Press, Oxford.

Brookes, G. F., Miles, N. J and Clayton, J.S., Hydrocyclone performance related to velocity parameters,
Proceedings of 2nd International Conference on Hydrocyclones, BHRA, Cranfield, 1984, pp 67-81.

Chen, W., Zydek, N and Parma, F., Evaluation of hydrocyclone models for practical applications, Paper
presented at the Engineering Foundation Conference on Solid – Liquid Separation Systems, Hawaii, April 18-
23, 1999, Separation and Purification Technology, 2000 (in press).

Cilliers, J. J and Hinde, A. L., An improved hydrocyclone model for backfill preparation. Minerals Engineering,
1991, 4 (7 –11) 683- 691.

Coelho, M.A.Z and Medranaho, R. A., An evaluation of the Plitt and Lynch & Rao models for the
hydrocyclones. In Hydrocyclones, Analysis and Applications, Eds., L. Svarovsky and M.T. Thew, Kluwer,
Dordrecht, 1993, pp 63-72.

De Kok, S.K., Symposium on recent developments in use of hydrocyclones in mill operation – a review,
Journal of Chemical Metallurgical and Mining Society of South Africa, 1956, 56, pp 281-294.

Flintoff, B.C., Plitt , L. R and Turak, A. A., Cyclone modelling: A review of present Technology, CIM Bulletin,
1987, 80 (905): 39-50, September.

Gault, G. A and Nageswararao, K., 1981. Hydrocyclone classification studies of a complex nickel ore.
Proceedings of Australasian Institute of Mining and Metallurgy, 1981 (285) 61-65, December.

Heiskanen, K. G. H., Developments in wet classifiers, International Journal of Mineral Processing, 1996, 44-
45: 29-42.

Hinde, A.L., Control of milling circuits using simple feed back loops, Grinding Theory and Practice, SAIMM
Vacation School, Johannesburg, 1977, pp 192-214.
 24

Hinde, A. L., Classification and concentration of heavy minerals in grinding circuits. Paper No. 85-126, 114th
AIME Annual Meeting, New York City, 1985.

Hinde, A.L., Private communication, 1999.

Lynch, A. J., Mineral Crushing and Grinding Circuits. 1977, Elsevier.

Lynch, A. J., Computers in mineral processing, XVIII APCOM, 1984, London, pp 26-30.

Lynch, A. J and Rao, T. C., Studies on the operating characteristics of hydrocyclone classifiers, Indian Journal
of Technology, 1968, 6(4) 106 - 114.

Lynch, A. J and Rao, T. C., Modelling and Scale-up of hydrocyclone classifiers. Proceedings of 11th
International Mineral Processing Congress, 1975, Cagliari, pp 245- 269.

Lynch, A. J and Morrell, S., The understanding of comminution and classification and its practical application in
plant design and operation, In Comminution - Theory and Practice, Proceedings of Annual Meeting, Ed: S. K.
Kawatra, AIME, 1992, pp 405 – 426.

Medronho, R.A and Svarovsky, L., Tests to verify hydrocyclone scale up procedure, Proceedings of 2nd
International Conference on Hydrocyclones, 1984, BHRA, Cranfield, Paper A1, pp 1-16.

Moder, J.J and Dahlstrom, D.A., Fine size, close specific gravity solid separation with the liquid-solid cyclone,
Chemical Engineering Progress, 1952, 48(2) 75-88.

Nageswararao, K., Further developments in the modelling and scale-up of industrial hydrocyclones. PhD Thesis,
University of Queensland, Brisbane, 1978.

Nageswararao, K., A generalised model for hydrocyclone classifiers, Proceedings of Australasian Institute of
Mining and Metallurgy, 1995, 300 (2) 21.

Nageswararao, K., Normalisation of the efficiency curves of hydrocyclone classifiers Minerals Engineering,
1999a, 12 (1) 107-118.

Nageswararao, K., Reduced efficiency curves of industrial hydrocyclones – An analysis for plant practice,
Minerals Engineering, 1999b, 12(5) 517-544.

Nageswararao, K., A critical analysis of fish-hook effect in hydrocyclone classifiers, Paper presented at the
Engineering Foundation Conference on Solid – Liquid Separation Systems, Hawaii, April 18-23, 1999,
Separation and Purification Technology, 2000 (in press).

Nageswararao, K and Hinde, A.L., Mathematical modelling of industrial hydrocyclone classifiers, (to be
communicated), 2000.

Napier-Munn, T. J., Morrell, S., Morrison. R.D and Kojovic, T., Mineral Comminution Circuits - Their
Operation and Optimisation, 1996, JKMRC Monograph Series, Julius Kruttschnitt Mineral Research Centre,
University of Queensland.

Napier-Munn, T. J and Lynch, A. J., The modelling and computer simulation of mineral treatment processes –
Current status and future trends, Minerals Engineering, 1992, 5(2) 143-167.
 25

Narayanan, S.S., Development of a laboratory single particle breakage technique and its application to ball mill
modelling and scale-up, Ph.D. Thesis, University of Queensland, 1985.

Peachey, C.G., Distribution of water in large diameter cyclones under operating conditions, International
Mineral Processing Congress, Institute of Mining and Metallurgy, London, 1960, pp 147- 156.

Plitt, L. R., A mathematical model of the hydrocyclone classifier, CIM Bulletin, 1976, 69 (776) 114-123,
December.

Plitt, L. R., Conil, P and Broussaud, P., An improved method of calculating the water spilt in hydrocyclones
Minerals Engineering, 1990, 3(5) 533 –535.

Rao, T. C., The characteristics of hydrocyclones and their application as control units in comminution circuits.
PhD Thesis, University of Queensland, 1966.

Rao, T.C and Deb Kanungo, P., Water division by industrial hydrocyclones, Transactions of Institute of Mining
and. Metallurgy (Section C: Mineral Processing and. Extractive Metallurgy), 1973, 79: C105-C107.

Rogers, B., 1998. The effect of a new cyclone shape on wear life and separation performance, Paper presented
in Comminution-98, Brisbane, Australia.

Rouse, B.D., Clayton, J.S., Brookes, G.F., Confirmation of modelling techniques for small diameter cyclones,
Proceedings of 3rd International Conference on Hydrocyclones, BHRA, Oxford, 1987, pp 7-20.

Shi, F and Napier-Munn, T.J., A model for slurry rheology, International Journal of Mineral Processing, 1996,
47: 103-123.

Stas, M., The influence of orifices in the washing characteristics of hydrocyclones, Proceedings of 4th
International Mineral Processing Congress, Progress In Mineral Dressing, Stockholm, 1957, pp 161-186.

Svarovsky, L., Hydrocyclones, 1984, Holt, Rinehart and Winson Ltd., London.

Svarovsky, L., Selection of hydrocyclone design and operation using dimensionless groups, Proceedings of 3rd
International Conference on Hydrocyclones, Oxford, England, 1987, Paper A1.

Vallebuona, G., Casali, A and Beviliacqua, P., Small diameter cyclones: Performance prediction by empirical
models, Minerals and Metallurgical Processing, 1995a, 12(4) 184-188, November.

Vallebuona, G., Casali, A., Ferrara, G., Leal, O and Beviliacqua, P., Modelling for small diameter cyclones,
Minerals Engineering, 1995b, 8(3) 321 – 327.
 26

LIST OF IMPORTANT SYMBOLS

a, b,...g parameters in equation for Rf

d size of the particle, µm
d50c Kelsall classification size, µm
Dc, Do, Du, Di diameters of the cyclone, vortex finder, spigot and inlet, cm
EU Euler number
f functions of ...
fi size distribution of feed solids
g acceleration due to gravity
F4 calibration parameter in Plitt’s equation for flow split
h free vortex height
K1 material dependent constant in Lynch-Rao equation
Ks function of spigot (equation (13)
Kv1, Kw1, Kvo, Kw0 material dependent constants in generalised equations
Lc length of the cylindrical section of the cyclone
p, q parameters in equations for flow split (equation 2)
P cyclone feed pressure
P40 40% passing size of feed solids
µsl viscosity of feed slurry
Q throughput of the cyclone, lit/min
Rf recovery of water to underflow
Rv volumetric recovery of feed slurry to underflow
S volumetric flow split (volumetric flow in underflow/ volumetric flow
in overflow)
Wf water rate in feed, tons/hr
meanWf mean water rate in feed, tons/hr
Wo water rate in overflow, tons/hr
α inclination of cyclone with the vertical, degrees
β, φ volumetric fraction of feed solids
λ hindered settling factor, β/(1−β)3
θ full cone angle, degrees
ρp density of feed pulp
ρs density of feed solids