Modeling The Size-Density Partition Surface of Dense-Medium Separators

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 12

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

net/publication/257127683

Modeling the size–density partition surface of dense-medium


separators

Article  in  International Journal of Mineral Processing · September 2003


DOI: 10.1016/S0301-7516(03)00118-2

CITATIONS READS
35 489

3 authors, including:

Venkoba Rao P. C. Kapur


Tenova
136 PUBLICATIONS   2,464 CITATIONS   
37 PUBLICATIONS   124 CITATIONS   
SEE PROFILE
SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Segregation of Particles in a Jig Bed with regard to size and density View project

Colloidal Processing of Alumina-Zirconia Ceramics View project

All content following this page was uploaded by Venkoba Rao on 14 November 2017.

The user has requested enhancement of the downloaded file.


Int. J. Miner. Process. 72 (2003) 443 – 453
www.elsevier.com/locate/ijminpro

Modeling the size–density partition surface


of dense-medium separators
B. Venkoba Rao, P.C. Kapur, Rahul Konnur *
Tata Research Development and Design Centre (TRDDC), 54B, Hadapsar Industrial Estate, Pune 411 013, India

Received 5 July 2002; received in revised form 6 June 2003; accepted 1 July 2003

Abstract

A stochastic model is proposed for describing the size – density partition surface in dense-medium (DSM) gravity/centrifugal
separators. The model is obtained by super-imposing a random walk on a simplified particle flow behavior inside the separator.
The novel feature of the model is that it permits a description of the partition surface without incorporating the pivot
phenomenon, i.e., without considering the partition number at the pivot point. The model also yields analytical expressions for
the cut size, cut density and the ecart probable. The computed separation indices are in agreement with existing empirical
relationships. Data from the published literature have been used to test the validity of the model in representing the partition
surface.
D 2003 Elsevier B.V. All rights reserved.

Keywords: modeling; size – density separation; partition surface; dense-medium separators

1. Introduction medium suspension. Particles of different sizes,


shapes and densities are separated from each other
Dense-medium (DSM) gravity/centrifugal separa- due to the differential settling rates in the dense-
tors are widely used in the mineral processing industry medium fluid whose density can be controlled. The
for classifying particles by density and size. Vast presence of centrifugal force in centrifugal separators
quantities of coal and mineral fines in a rather broad enables separation of fine particles at a rate faster than
size range of 0.5– 100 mm are treated in a host of that achievable in gravity-only concentrating units.
dynamic separators, such as dense-medium (DSM) The performance of a dense-medium separator is
cyclone, Vorsyl separator, Larcodems, Chance Cone conveniently reported in the form of a partition
Separator, Dyna Whirlpool, Tri-Flo separator and surface. Any point on the surface is a partition
others. In these systems, the feed to the separator number, which corresponds to the fraction of feed of
consists of de-slimed coal or ore particles in a dense- given size and density reporting to the underflow or
sink. Ferrara and Schena (1987) provided a schematic
representation of the general partition surface, i.e., a
* Corresponding author. Tel.: +91-20-4042310; fax: +91-20-
three-dimensional plot of partition number as a func-
4042399. tion of size and density, under ideal conditions of a
E-mail address: rahulk@pune.tcs.co.in (R. Konnur). homogeneous Newtonian fluid and free settling of

0301-7516/$ - see front matter D 2003 Elsevier B.V. All rights reserved.
doi:10.1016/S0301-7516(03)00118-2
444 B. Venkoba Rao et al. / Int. J. Miner. Process. 72 (2003) 443–453

particles. An illustrative plot of a partition surface


generated by a stochastic model—discussed latter—is
shown in Fig. 1. The partition surface is a function of
operating conditions such as solid volume concentra-
tion and medium viscosity (Davis and Napier-Munn,
1987; Ferrara and Schena, 1987; Scott and Napier-
Munn, 1992). This surface may become steeper or
flatter depending on changes in media and ore char-
acteristics and separator design parameters. Sectioning
of the surface by planes normal to the density axis
yields a family of density-based partition curves,
shown in Fig. 2. Similarly, sectioning of the surface
by planes normal to the size axis yields a family of
size-based partition curves, shown in Fig. 3. In the
former case, particles of density less than the separa-
tion density show reverse classification. In other Fig. 2. Density based classification curves for different particle
words, the particle recovery to underflow decreases densities generated from Fig. 1.
with an increase in particle size. Regardless of density,
as particle size decreases, the density-based partition
curves flatten, and as particle size approaches zero, tion process/equipment. The pivot phenomenon
the partition curve becomes completely flat with the could arise in solid –liquid density separation sys-
partition number equal to the fluid recovery in the tems owing to the presence of particles with neutral
underflow. buoyancy which, irrespective of their size, will
As shown in Fig. 3, under the above mentioned separate in the same proportion as the medium pulp.
idealized conditions at least, the size-based partition This implies that the partition curves pass through a
curves pass through a unique pivot point whose single point, the so-called pivot point. The coordi-
coordinate is a characteristic of the specific separa-

Fig. 1. A typical partition surface generated using the stochastic


model (Eq. (18)). The parameter values are A =  23.373  10 3, Fig. 3. Size-based classification curves for different particle sizes
B =  0.63, c = 1.611 and qp = 1497. generated from Fig. 1.
B. Venkoba Rao et al. / Int. J. Miner. Process. 72 (2003) 443–453 445

nates of the pivot points, namely, the pivot density


qp and the pivot partition number Yp are referred to
as the pivot parameters. In the past, while many
authors have observed the pivot phenomena in
experimental data (Collins et al., 1983; King and
Juckes, 1984; Restarick and Krnic, 1991), its exis-
tence has also been utilized for the purpose of
process modeling by Scott and Napier-Munn (1992).
In practice, ferrosilicon or magnetite solids that
make up the media in a dense-medium suspension
themselves settle under the influence of gravity/cen-
trifugal forces. This settling causes segregation of the
media particles according to their size and density (in
a manner similar to the partitioning of ore particles)
and results in density gradients within the separating
unit. As a consequence, the density of the underflow
medium is greater than the density of the overflow
medium and, as seen in Fig. 4 (see discussion in
Section 5), the experimental partition curves for
different particle sizes may not cut at a single point
but over a region.

2. Background and current separator models

The efficiency of separation represented by the


partition surface can be improved or controlled at an
optimum level provided a mathematical parametric
representation of the partition surface and the corre-
lations between partition surface parameters and op-
erating and design variables of the separating unit are
established.
A survey of the status of modeling and simulation
of dense-medium separation processes at the begin-
ning of the 1990s was provided by Napier-Munn
(1991). Two of the better-known mathematical models
for the performance of classifiers in a single particle
attribute are due to Lynch (1965) and Plitt (1971).
Using data obtained from several experimental tests,
Lynch applied the following Whiten’s regression
equation to describe the classification curve of a
hydrocyclone classifier by size

expðaxÞ  1
Y¼ ð1Þ
expðaxÞ þ expðaÞ  2
Fig. 4. (a, b and c) Comparison of size based separation curves
generated by the stochastic model for low, medium and high values Plitt derived a simple model for the classification
of SSE. curve by assuming that mixing due to the turbulence
446 B. Venkoba Rao et al. / Int. J. Miner. Process. 72 (2003) 443–453

within the classifier is responsible for imperfect sep- Substitution of a from Eq. (5) into Eq. (3), fol-
aration in the classifier. Plitt’s model equation is given lowed by some simplification yields
by
1
m Yj ¼ ð6Þ
Y ¼ 1  exp½x lnð2Þ ð2Þ 1 þ exp½1:099ðq50  qj Þ=Ep

In these models, Y is the corrected efficiency of This is a logistic function, which was also proposed
separation, i.e., the corrected fraction of feed passing independently by Klima and Luckie (1989). Imposing
to the underflow, x = d/d50 is the ratio of the particle the conditions that (i) all size-based partition curves
size d and the particle size corresponding to 50% pass through the pivot point Yp at particle density qp,
separation d50. The parameters a and m reflect the and (ii) all these partition curves at different particle
sharpness of separation. Note that similar expressions sizes are linked to the partition surface through an
with different values of parameters can be written for empirical power law of ecart probable expressed in
classification of particles by density q. particle size (Scott and Napier-Munn, 1992)
While the Lynch and Plitt models are reason-
ably adequate for describing the classification Ep ¼ kd n ð7Þ
curve for size-only (or density-only) separation, where the proportionality constant k incorporates the
these are not convenient for representing the par- effect of viscosity and the strength of forces acting
tition surface of size – density separation in dense- within the separator, while the parameter n is related to
medium type equipment. This is because a direct the flow conditions prevailing within the separator. Eq.
extension from one attribute (say, density) to two (6) transforms to the particle size –density partition
attributes (density and size) would require a surface in percent partition number coordinate as
knowledge of the d50 value for each density or follows
density class. Lynch and Napier-Munn (1986) and
Scott and Napier-Munn (1992) got around this diffi- 100
Y ¼
culty by utilizing the pivot point phenomenon as a 1þ exp½lnðYp1  1Þ þ 1:099ðqp  qÞ=ðkd n Þ
constraint for extending the Lynch model for density
ð8Þ
partition curve to a size – density partition surface.
These authors begin with the Lynch reduced partition This model consists of four parameters YP, qP, k
equation in discrete density intervals and n that can be estimated from the experimental
expðaqj =q50 Þ  1 data. Although not derived by the authors, expressions
Yj ¼ ; j ¼ 1; 2 . . . ð3Þ for cut size and cut density can be extracted from the
expðaqj =q50 Þ þ expðaÞ  2 model Eq. (8). For the sake of completeness, the
derived relations for the two separation parameters
Here, Yj is the partition number defined as the
are given below
weight fraction of density species qj in the feed that
reports to sink after separation, q50 is the cut density !ð 1n Þ
and a is a shape parameter. The ecart probable, Ep, lnð3Þ ðq  qp Þ
d50 ¼ ð9Þ
which reflects the separation inefficiency at the given k lnðYp1  1Þ
particle size, is given by
  It can be seen from Fig. 2 that the above expression
q75  q25 q50 3ð3ea  2Þ
Ep ¼ ¼ ln ð4Þ for cut size is defined for
2 2a ea þ 2
1. q>qP, when YP V 0.50 (i.e., for YP V 50%)
For typical values of a which tend to be greater
2. q < qP, when YP z 0.50 (i.e., for YP z 50%)
than 4

lnð3Þq50 kd n
Ep ¼ ð5Þ q50 ¼ qp þ lnðYp1  1Þ ð10Þ
a lnð3Þ
B. Venkoba Rao et al. / Int. J. Miner. Process. 72 (2003) 443–453 447

Note that the functional form of Eq. (10) is similar tively. Given particle size and density and liquid
to the one proposed independently by Collins et al. viscosity, this equation can be used to calculate the
(1983). transient particle position h(t), which denotes the
A different approach for modeling separation in position of a particle relative to the entry point h = 0
dense-medium separators has been recently devel- at time t
oped by Hu et al. (2001). They employ a modified
suspension model for predicting the partition curves hðtÞ ¼ ðvs  vl Þt þ Vr t ð12Þ
of smaller size fractions using experimental data
The velocity Vr a random variable which deter-
obtained from density tracer tests carried out using
mines the probability of a particle reporting to float or
large size particles in dense-medium cyclones. This
sink stream, can be written as
approach is based on the work of Schubert and Neese
(1973) and Clarkson and Wood (1993), who have hðtÞ
developed a general model for turbulent cross-flow Vr ¼ þ vl  vs ð13Þ
t
wet classification.
In this work, a stochastic model is proposed for Next, the central limit theorem is invoked to justify
describing the size – density partition surface in that the random velocity Vr is normally distributed
gravity/centrifugal dense-medium separators without with zero mean and variance r2. The cumulative
invoking the pivot phenomena. The model is a distribution function form of the random velocity is
phenomenological model obtained by considering a   
simplified particle flow behavior inside the separator. 1 Vr
FðVr ; 0; rÞ ¼ 1 þ erf 0:5 ð14Þ
It is seemingly much more general and versatile than 2 2 r
the existing models in the sense that it yields analyt-
ical expressions for various performance indices of Substitution of Eq. (13) into Eq. (14) yields the
dense-medium separators. Moreover, as shown in following distribution in h
Appendix A, it is possible to derive the Scott and   
1 h þ vl t  vs t
Napier-Munn model from the proposed model. Y ðhÞ ¼ 1 þ erf  lVhVl
2 20:5 tr
ð15Þ
3. Stochastic model
This equation represents the transient partition
surface at a height h and at time t. It can be seen that
Following Klima and Luckie (1989), it is assumed
Y(-l) = 0 and Y(l) = 1, thus fulfilling the boundary
that the separator is fed at an arbitrary height h = 0.
conditions for a distribution function. At height h = 0
Inside the separator, the movement of the particles is
(where the feed enters the separator), or as time t ! l
governed by: (i) gravity force, (ii) hydrodynamic force
(at any height in the vessel), Eq. (15) is reduced to
due to fluid flow and (iii) random forces resulting
from interparticle interactions coupled with circular 1h  v  v i
s l
motion. Under the influence of these forces, particles Y¼ 1  erf ð16Þ
2 20:5 r
either move to the sink stream at h = l, or the float
stream at h =  l. It is assumed that fluid flow This implies that under steady-state conditions, the
opposes particle settling. The instantaneous velocity destination (overflow or underflow stream) of a particle
of a particle is given by of given size and density is completely determined by a
normally distributed random process associated with
dh the relative velocity of the particle given by (vs  vl). A
¼ vs  vl þ Vr ð11Þ convenient expression for settling velocity of solid
dt
particles is given by Johnson et al. (1996)
where vs, vl and Vr are the settling velocity of the
particle, drift velocity due to flow of the fluid or gd c ðq  qp Þ
vs ¼ 0 < cV2 ð17Þ
medium and random velocity in the separator, respec- 18l
448 B. Venkoba Rao et al. / Int. J. Miner. Process. 72 (2003) 443–453

where g is acceleration due to gravity, qp is the using the stochastic model given by Eq. (18) is shown
medium density and l is medium viscosity. Substi- in Fig. 1.
tuting Eq. (17) in Eq. (16) and lumping variance and
the other constants into the terms A and B yields the
following equation for partition surface expressed as 4. Model validation
a percentage
The validity of the model was tested on 16 sets of
Y ¼ 50½1  erf ðAd c ðq  qp Þ  BÞ ð18Þ published experimental data by Palowitch and Deur-
brouck (1979), Collins et al. (1983), King and Juckes
where (1984), Ferrara and Schena (1987), Restarick and
Krnic (1991) and Scott and Napier-Munn (1992).
g
A¼ pffiffiffi ð19Þ Table 1 gives best-fit parameters for each of the
18 2lr experimental data sets. It should be noted that these
data were obtained from different types of gravity/
and
centrifugal separators, which are indicated in the table.
vl The data sets 14 – 16 in Table 1, taken from the work
B ¼ pffiffiffi ð20Þ of Ferrara and Schena (1987), are for the separation of
r 2
media particles only, separated into magnetic and non-
A, B, c and qp are the parameters of the size – density magnetic fractions. The Levenberg– Marquardt least-
partition surface. The parameters A and c together squares minimization technique was used to obtain
account for the following effects: (i) viscosity of the the best fits of the parameters from the partition data
medium; (ii) turbulence in the system; (iii) hindered and the sum of squared errors (SSE). The flow
settling; and (iv) particle shape. A three-dimensional parameter c, which is indicative of the flow condi-
plot of the size – density partition surface obtained tions prevailing within the separation zone of the

Table 1
Summary of partition surface parameters of the stochastic model and sum of squared errors (SSE) for gravity/centrifugal separator data taken
from the literature
S. No. Separator typea Stochastic model Reference Remark
3
B qp A  10 c SSE
1 DSMC  0.630 1497  23.373 1.611 0.285 Scott and Napier-Munn (1992) Survey 1.1
2 DSMC  0.200 1454  8.774 1.159 0.393 Scott and Napier-Munn (1992) Survey 1.2
3 DSMC 0.183 1431  7.432 1.022 0.332 Scott and Napier-Munn (1992) Survey 1.3
4 DSMC  0.540 1563  4.188 1.117 0.282 Scott and Napier-Munn (1992) Survey 2.1
5 DSMC  0.312 1554  4.694 0.970 0.245 Scott and Napier-Munn (1992) Survey 2.2
6 DSMC 0.081 1551  3.817 1.062 0.168 Scott and Napier-Munn (1992) Survey 2.3
7 DSMC  0.325 1336  1.098 1.474 0.282 Scott and Napier-Munn (1992) Survey 3.1
8 DSMC  0.367 1403  20.908 0.596 0.055 King and Juckes (1984)
9 VORSYL  0.652 2758  2.663 0.459 0.116 Collins et al. (1983) Fig. 9
10 DSMC  0.057 1219  32.087 0.773 0.043 Restarick and Krnic (1991)
11 DSMC  0.887 3009  2.596 1.067 0.171 Lynch and Napier-Munn (1986)
12 CCS  0.897 1329  1.899 0.781 0.029 Palowitch and Deurbrouck (1979)
13 DSMC  0.605 1311  2.049 0.688 0.061 Palowitch and Deurbrouck (1979)
14 TRI FLO 0.114 2462  14.370 1.631 0.052 Ferrara and Schena (1987) Fig. 8
15 TRI FLO 0.117 2572  4.587 1.117 0.064 Ferrara and Schena (1987) Fig. 9
16 MDC  0.595 3817  8.520 0.796 0.020 Ferrara and Schena (1987) Fig. 10
Pivot partition number Yp is expressed as a fraction, particle size d is in millimeters and density qp is in kg/m3.
a
DSMC, dense-medium cyclone; VORSYL, Vorsyl separator; TRI FLO, tri-flow separator; MDC, media densifying cyclone; CCS, chance
cone separator.
B. Venkoba Rao et al. / Int. J. Miner. Process. 72 (2003) 443–453 449

separator, varies from a low of 0.46 to a high of At the cut size d50 for a given particle density, or
about 1.63. It should be noted that the flow parameter equivalently at the cut density q50, for a given particle
is expected to be close to 2 for laminar flow size, the partition number is 50%. This implies that the
conditions and decreases as the flow undergoes a settling velocity of particle vs is equal to the drift
transition to the turbulent flow regime (Johnson et velocity of the fluid vl (Eq. (18)). Hence
al., 1996). For a better appreciation of the fit
!ð 1c Þ
obtained, Fig. 4 shows three representative model B
simulations for low, medium and high values of d50 ¼ ð22Þ
Aðq  qp Þ
SSE.
and
5. Discussion B
q50 ¼ qp þ ð23Þ
Ad c
The partition surface represented by Eq. (18) is
the cumulative form of a standard normal distribu- Eq. (23) is of the same form as Eq. (10), which has
tion in A(q  qp)d c  B. This means that a plot of been derived using the empirical correlations pro-
A(q  qp)d c  B versus Y results in a standard posed by Scott and Napier-Munn (1992), as well as
normal distribution curve N(0, 1) with all experi- the following empirical relation first proposed by
mental points collapsing onto this curve. The mean Collins et al. (1983)
of this curve occurs at a partition number of 50%,
where the particle-settling velocity A(q  qp)d c is q50 ¼ qs þ sd w ð24Þ
balanced by fluid drift B as discussed later in this
section. This validates the assumption of taking zero where qs, s and w are constant parameters. These latter
mean in the derivation of the stochastic model. At authors determined the separation density qs, defined
steady state, owing to the properties of the normal as the separation density achieved at infinitely large
distribution, a plot of the partition number Y versus particle size, from the asymptote of curve of q50
A(q  qp)d c or (qUqp)d c will result in all the versus particle size d. Clearly, in view of Eq. (23),
experimental points (from a single experiment) qs corresponds to the pivot density qp for very large
collapsing onto a single curve. Typically in the particle sizes.
former case, the partition curve shifts by B units Comparison with hydrocyclone performance (size
either to the left or right of the standard normal attribute only) indicates that the parameter Yp in a
distribution curve N(0,1) based on the value of B dense-medium separator (size– density attributes) is
being negative or positive, respectively. the by-pass fraction of particles of neutral buoyancy
It will be seen that the model Eq. (18) for the and particles of almost zero size recovered at the
partition surface has been obtained without incorpo- underflow. In an ideal situation, by-pass fraction Yp
rating the pivot phenomenon as a constraint. In fact, should have a value equal to the media recovery in
the pivot partition number Yp can be readily obtained underflow Rm (Ferrara and Schena, 1987). However,
from Eq. (18) for an idealized homogeneous and Yp < Rm because of crowding effects (Scott and
Newtonian medium and a given feed medium density Napier-Munn, 1992). The correlation between the
qf = qp pivot partition number Yp and the percentage of
medium pulp reporting to underflow Rm is shown
Yp ¼ 50½1 þ erf ðBÞ ð21Þ in Fig. 5. Yp is expected to be close to Rm when the
viscosity of the medium is high. For low viscosity
This relation suggests that the pivot partition num- systems, in general, Yp is less than Rm due to
ber Yp is dependent on parameter B, which incorpo- crowding effects near the apex (Scott and Napier-
rates the particle drift due to fluid velocity. The drift Munn, 1992).
velocity B takes negative (positive) values when It has been observed that there exists a density
Yp < 50% (>50%). differential between overflow and underflow media
450 B. Venkoba Rao et al. / Int. J. Miner. Process. 72 (2003) 443–453

in part be attributed to the errors inherent in the


experimentation process.
Another important indicator of sharpness of sepa-
ration in dense-medium units is the steepness of the
partition surface, which is traditionally defined in
terms of the ecart probable Ep. Using the definition
given in Eq. (4), the ecart probable Ep for a given
particle size can be obtained from Eq. (18)

0:477 logð3Þ
Ep ¼  c ð25aÞ
Ad c Ad c

This equation is similar to the empirical power law


in Eq. (7) proposed by Scott and Napier-Munn (1992).
The observed and predicted relationship between Ep
and d, based on three sets of data from Collins et al.
(1983), Restarick and Krnic (1991) and Scott and
Napier-Munn (1992), is shown in Fig. 8. The param-
Fig. 5. Correlation between the computed pivot partition number Yp eter c in Eq. (25a) represents the slope of the Ep
and medium pulp to underflow Rm. The data have been taken from versus d curve. The slope of these curves varied from
the works of Collins et al. (1983), Restarick and Krnic (1991) and a low of about 0.46 to a high of about 1.63, thus
Scott and Napier-Munn (1992).
indicating the prevalence of both laminar and turbu-
lent particle flow regime in the dense-medium unit.
suspensions owing to the segregation of media par- Although Ep in Eq. (25a) appears to be indepen-
ticles (Collins et al., 1983; Restarick and Krnic, 1991; dent of the parameters qp and B (and hence of the
Scott and Napier-Munn, 1992). In spite of the density fluid drift velocity vl), this is actually not the case as
differential, Fig. 6 shows a linear relationship between can be seen by using the relationship given in Eqs.
the predicted pivot density qp and the feed medium
density qf.
A pivot density close to the desired separation
density and a low bypass of the feed material are
characteristics of good performance of a dense-medi-
um separation unit. A small by-pass (low Yp) implies a
negative value of B (Eq. (A3)). Negative B indicates
that the fluid is drifting in a direction opposite to the
direction of settling particles, i.e., the fluid offers
greater resistance to the settling of particles. In other
words, for reducing the by-pass, there should a greater
resistance to the settling of particles by the fluid
moving up.
Fig. 7 shows the relationship between the observed
and predicted cut density q50 for data taken from
Collins et al. (1983), Restarick and Krnic (1991) and
Scott and Napier-Munn (1992). It can be seen that the
predictions of the stochastic model closely match the
observed values. It is clear that for small particle sizes,
the predicted cut densities tend to fall below the Fig. 6. Correlation between pivot density qp and feed medium
values obtained experimentally. This mismatch can density qf for the data mentioned in Fig. 7.
B. Venkoba Rao et al. / Int. J. Miner. Process. 72 (2003) 443–453 451

Fig. 8. (a, b and c) Comparison of the variation of actual and


Fig. 7. (a, b and c) Comparison of the variation of the actual and predicted ecart probable Ep with particle size d for the data
predicted cut density q50 with particle size d. mentioned in Fig. 7.
452 B. Venkoba Rao et al. / Int. J. Miner. Process. 72 (2003) 443–453

(20) and (23) to eliminate the denominator in Eq. applied to simulate the partition surface of all types
(25a). This yields the following expression of dense-medium gravity/centrifugal separators.

pffiffiffi
0:477 2 Acknowledgements
Ep ¼ rðqp  q50 Þ ð25bÞ
vl
We acknowledge Prof. Mathai Joseph, Executive
Since, in real systems, r p 0, q50 does not coincide Director, Tata Research Development and Design
with qp. Further, there is a finite upper bound on vl. It Centre (TRDDC) for his encouragement and for
therefore follows that Ep does not tend to be zero, providing management support.
even for very large particle sizes.
Experimental evidence suggests that Ep increases
with an increase in viscosity (Ferrara and Schena, Appendix A
1987; Scott and Napier-Munn, 1992). This is espe-
cially true for finer particles and therefore implies that The Scott and Napier-Munn model is derived from
an increase in viscosity should result in a flatter the stochastic model from Eqs. (7) and (10), the ecart
partition surface. Examination of the stochastic model probable can be written as
shows that this is indeed the case. The parameters A
and c in Eq. (18) together determine the steepness of ðq50  qp Þ
the partition surface. An increase in the viscosity of Ep ¼ lnð3Þ ðA1Þ
lnðYp1  1Þ
the suspension mainly leads to a pronounced decrease
in the magnitude of A (Eq. (19)). From Eq. (18), it can
be seen that a reduction in the magnitude of A has the Also, substituting the value of Ad c from Eq. (23)
undesirable effect of flattening the partition surface. into Eq. (25a) results in

ðq50  qp Þ ðq50  qp Þ
Ep ¼ 0:477 c  logð3Þ
6. Conclusions B B
ðA2Þ
A stochastic model is proposed for representing the
particle size –density partition surface without invok- Further equating Eqs. (A1) and (A2) and simplify-
ing the existence of a pivot point. The model satis- ing yields
factorily predicts the observed relationships between
(i) the pivot partition number Yp and the medium pulp B ¼ logðYp1  1Þ ðA3Þ
reporting to underflow Rm and (ii) the pivot density qp
and feed medium density qf. In addition, predictions Hence
of the phenomenological model are consistent with 1
key empirical relationships, such as the dependence of Yp ¼ ðA4Þ
1 þ 10B
(i) ecart probable Ep on particle size d and (ii) cut
density q50 on particle size d. Moreover, the model Another expression for Yp is given by Eq. (21) of
parameters have physical relevance to the process of the stochastic model
particle separation. The model reveals ill effects of
increased viscosity on the particle separation through Yp ¼ 0:5½1 þ erf ðBÞ ð21Þ
parameter A by flattening the partition surface. The
value of parameter B, which incorporates the particle Eqs. (21) and (A4) can be used to obtain the
drift due to fluid velocity, determines the amount of following approximation to the error function
by-pass of particles in the concentrator. The value of
parameter c indicates the degree of turbulence in the 2
erf ðxÞ ¼ 1 ðA5Þ
separator. The model is more generic and can be 1 þ 10x
B. Venkoba Rao et al. / Int. J. Miner. Process. 72 (2003) 443–453 453

The stochastic model given by Eq. (18) can be tion. In: Wood, P. (Ed.), Proc. of the 3rd Int. Conf. On Hydro-
written in fractional form as cyclones. BHRA, Oxford, pp. 101 – 110.
Hu, S., Firth, B., Vince, A., Lees, G., 2001. Prediction of dense-
medium cyclone performance from large size density tracer test.
Y ¼ 0:5½1 þ erf ðB  Ad c ðq  qp ÞÞ ð18Þ Min. Eng. 14, 741 – 751.
Johnson, C.P., Li, X., Logan, B.E., 1996. Settling velocities of
fractal aggregates. Environ. Sci. Technol. 30, 1911 – 1918.
Converting the above equation to logistic form King, R.P., Juckes, A.H., 1984. Cleaning of fine coals by dense-
using Eq. (A5) medium cyclone. Powder Technol. 40, 147 – 160.
Klima, M.S., Luckie, P.T., 1989. Application of an unsteady-state
1 1 pulp-partition model to dense-medium separations. Coal Prep. 6,
Y ¼ ðAd c ðqqp ÞBÞ
¼ lnð10ÞðAd c ðqqp ÞBÞ 227 – 240.
1 þ 10 1þe Lynch, A.J., 1956. The characteristics of hydrocyclones and their
ðA6Þ application as control units in communition circuits, Report
Progress No. 6, Dept. of Mining and Metallurgy, University of
Equating Eqs. (7) and (25a) Queensland, Australia.
Lynch, A.J., Napier-Munn, T.J., 1986. Dense-medium separation
research at the Julius Kruttschnitt mineral research centre. Trans.
logð3Þ
Ad c ¼  ðA7Þ Inst. Min. Metall. (Section C: Min. Proc. Extr. Metall.) 95,
kd n C221 – C225.
Napier-Munn, T.J., 1991. Modelling and simulating dense-medium
Substituting Eqs. (A3) and (A7) in Eq. (A6) results separation processes—a progress report. Min. Eng. 4, 329 – 346.
in the Scott and Napier-Munn model given in Eq. (8). Palowitch, E.R., Deurbrouck, A.W., 1979. Wet concentration of
coarse coal, part 1: dense-medium separation. In: Leonard, J.W.
(Ed.), Coal Preparation, 4th ed. SME, New York, pp. 1 – 36.
Chapter 9.
References Plitt, L.R., 1971. The analysis of solid – solid separations in classi-
fiers. Can. Min. Metall. Bull., 1 – 6.
Clarkson, C.J., Wood, C.J., 1993. A model of dense-medium cy- Restarick, C.J., Krnic, Z., 1991. The effect of underflow/over-
clone performance. Coal Prep. 12, 101 – 115. flow ratio on dense-medium cyclone separation. Min. Eng. 4,
Collins, D.N., Turnbull, T., Wright, R., Ngan, W., 1983. Separation 263 – 270.
efficiency in dense media cyclones. Trans. Inst. Min. Metall. Schubert, H., Neese, T., 1973. The role of turbulence in wet clas-
(Section C: Min. Proc. Extr. Metall.) 92, C38 – C51. sification. In: Jones, M.J. (Ed.), Proc. of the 10th International
Davis, J.J., Napier-Munn, T.J., 1987. The influence of medium Mineral Processing Conference. Inst. Mining and Metallurgy,
viscosity on the performance of dense-medium cyclones in coal London, pp. 213 – 239.
preparation. In: Wood, P. (Ed.), Proc. 3rd Int. Conf. on Hydro- Scott, I.A., Napier-Munn, T.J., 1992. Dense-medium cyclone model
cyclones. BHRA, Oxford, pp. 155 – 165. based on the pivot phenomenon. Trans. Inst. Min. Metall. (Sec-
Ferrara, G., Schena, G.D., 1987. Cycloning in dense media separa- tion C: Min. Proc. Extr. Metall.) 101, C61 – C76.

View publication stats

You might also like