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

net/publication/240407969

Estimating the recovery of size-liberation classes in industrial

flotation cells: A simple technique for minimizing the propagation of
the experimental error

Article  in  International Journal of Mineral Processing · January 2006

DOI: 10.1016/j.minpro.2005.09.002

CITATIONS READS
13 60

1 author:

Orivaldo Normando Savassi

SGS GEOSOL
16 PUBLICATIONS   265 CITATIONS   

SEE PROFILE

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

COMPREHENSIVE MODELLING OF FLOTATION View project

All content following this page was uploaded by Orivaldo Normando Savassi on 28 October 2020.

The user has requested enhancement of the downloaded file.

 Int. J. Miner. Process. 78 (2006) 85 – 92
www.elsevier.com/locate/ijminpro

Estimating the recovery of size-liberation classes in industrial

flotation cells: A simple technique for minimizing the
propagation of the experimental error
O.N. Savassi *
Departamento de Engenharia de Minas, Universidade Federal de Minas Gerais R. Espirito Santo 35, Belo Horizonte, CEP 30160-030, Brasil
Received 12 August 2005; received in revised form 23 August 2005; accepted 3 September 2005
Available online 4 October 2005

Abstract

A simple technique, based upon interpolation and mass balancing, is proposed for minimizing the error propagation in the
estimation of the flotation recovery of size-liberation particle classes. Liberation data obtained by a down-the-bank survey is used
to illustrate application to industrial cells.
D 2005 Elsevier B.V. All rights reserved.

Keywords: flotation; mineral liberation; interpolation; beta distribution; mass balancing

1. Background and objective sharply outside an intermediate particle size range

whose boundaries vary according to the mineral being
As the costs of industrial comminution disallow com- floated (Lynch et al., 1981). The common observation
plete liberation of the ore, a significant amount of valu- that composite particles are recovered more slowly than
able mineral entering a flotation circuit is often locked fully liberated ones was reported by Gaudin as early as
with gangue. The economics of comminution–flotation 1939. Nevertheless, the measurement of particle libera-
plants is thus strongly linked to the recovery of compos- tion for optimisation of industrial flotation circuits has
ite particles and their impact upon concentrate grade. In only been attained in recent years, with the advent of
this sense, it has been stated that bmany of the complex- scanning electron microscopy and image analysis of
ities of industrial flotation circuits are due to attempts to polished sections of ore particles (Sutherland and Got-
find the most efficient way to treat composite particles so tlieb, 1991; Barbery, 1991; King and Schneider, 1993).
that the recovery of the valuable mineral is maintained at It must be emphasized, however, that size-liberation
a maximum while the dilution of the concentrate by datasets are strongly affected by the propagation of the
gangue is minimizedQ (Lynch et al., 1981). experimental error, even when the most modern tech-
The effect of particle size and particle liberation upon niques are applied for the measurement. This is a
metallurgical performance is well established. Data from consequence of the limited number of particles that
several plants show that the true flotation recovery falls can be analysed by scanning electron microscopy at
practical cost and execution time (around 2000 particles
* Current address: MinnovEX Technologies, 1140 Sheppard Ave W.
per sample, at the time of this writing). The objective of
# 6, Toronto, Canada M3K2A2. Fax: +55 416 633 2695. this paper is therefore to propose a simple technique for
E-mail address: [email protected]. minimizing the propagation of the experimental error in
0301-7516/$ - see front matter D 2005 Elsevier B.V. All rights reserved.
doi:10.1016/j.minpro.2005.09.002
86 O.N. Savassi / Int. J. Miner. Process. 78 (2006) 85–92

estimating the recovery of size-liberation classes in opposite, since the metallurgical data comes from sam-
industrial flotation cells. ples of much larger masses. The metallurgical data will
thus be balanced independently, with the results being
2. Methodology used to constrain the mass balancing of the liberation
data. In addition, an interpolation model will be applied
Scanning electron microscopy is usually conducted for minimizing the scatter of the liberation distributions
on particles of a narrow size interval. A portion of the measured experimentally, at same time as reconciling
sample is embedded in an epoxy resin block that, after each distribution with the grade and density of the
hardening, is cut and polished to reveal internal sections sample it originates from. These steps are discussed in
of the particles. Special care is taken to avoid segrega- more detail in the upcoming sections.
tion of denser particles during hardening. What is left of
the original sample is reserved for assaying and some- 2.1. Mass balancing the metallurgical data
times density measurement. The block is then placed
under the microscope to generate a BSE image of the A flotation cell operating at steady state can be seen
particles being scanned, while the approximated chem- as a mass conservation node with one inlet and two
ical constitution corresponding to each pixel in the outlet streams. Surveying all three streams provides
image is determined by dispersive X-ray spectrometry. experimental data redundancy, which is essential for
Image analysis allows the liberation of each individual mass balancing purposes. Benchmarking of industrial
particle to be quantified in terms of the number of pixels flotation circuits, however, is rarely conducted on indi-
of valuable mineral upon the total number of pixels vidual cells. Rather, due to cost considerations, a typical
forming the image of the particle. At the same time, mass balance flowsheet has a limited number of nodes,
the grade of the sample is estimated iteratively, as a each one encompassing a selected group of cells. Not
growing number of particles are being scanned. This all streams in the flowsheet have to be necessarily
process continues until the grade of the sample estimat- surveyed. However, to ensure that there is sufficient
ed by image analysis approaches the experimental assay. data redundancy for mass balancing, the total number
Alternatively, linear (1D) intercepts can be used instead of streams being surveyed must exceed the total number
of areal (2D) measurements. The end result is a libera- of nodes by two streams at least.
tion distribution, ranging from barren to fully liberated Most mass balancing packages currently available
particles, with several composite classes in between. can handle size-by-size datasets. The data from each
In strict terms, what is being measured on the stream is usually structured in the following hierarchy:
polished sections is the areal grade of the particles (or
sometimes the linear grade), which would justify a ! The total stream flowrate;
stereological correction for transformation to volumetric ! The total assay of the stream;
basis (Weibel, 1979; Gay, 1994; King and Schneider, ! The weight percentage of each size interval;
1997). This issue will be revisited in the experimental ! The size-by-size assays.
section below. The approach adopted here, however, is
to consider the image analysis as a practical method for After mass balancing the metallurgical data, the flow-
the measurement of volumetric grade, which, in addition rate of the valuable mineral in any target size interval
to the stere1ological error, is impaired by several other can be calculated in all streams of the flowsheet. This
sources of inaccuracy, including: a) the resolution of the information will be used to constrain the mass balancing
image is limited by the diameter of the electron beam; b) of the liberation data, as demonstrated later in the paper.
each pixel can only be assigned to a single mineral
species; c) apical sections have too few pixels for proper 2.2. Interpolation of volumetric grade distributions
estimation of particle grade and thus are routinely dis-
carded; and d) it is difficult to measure liberation in an A liberation distribution is usually discretized into
industrial flotation circuit accurately based on sets of twelve classes of volumetric grade, g V, j, each class con-
just 2000 particles, that is, less than 5 mg per epoxy taining a fraction, f V, j, of the total volume of the particles
block. being analysed per block. Class number 1 corresponds to
Therefore the technique proposed here is based upon the barren particles ( g V,1 = 0%), while class 12 corre-
the premise that the liberation data must comply with the sponds to the fully liberated ones ( g V,12 = 100%). The
metallurgical data (stream flowrate, head assays, size composite particles are assigned to ten intermediate clas-
distribution and size-by-size assays), rather than the ses (0% b g V,2 b 10%; 10% Vg V,3 b 20%; . . . ; 90% V
 O.N. Savassi / Int. J. Miner. Process. 78 (2006) 85–92 87

g V,11 b100%). The average volumetric grade, g V, of the The interpolated grade of each class, g*V,j , is then
distribution is given by the following sum (all equations calculated by integration of the incomplete beta distri-
in this paper apply to a target size interval, all grades are bution within the limits of that class.
expressed as a fraction rather than a percentage): Rather than being adjusted directly, the parameters a
X
12   and b are calculated from the first and second moments,
gV ¼ gV; j dfV; j ð1Þ n 1C and n 2C, of the incomplete beta distribution (Barb-
j¼1
ery, 1991):
The average volumetric grade can also be estimated
n1C  n2C
from the assay, g M, and the density of the original a ¼ n1C d ð10Þ
sample along with the density of the valuable mineral, n2C  n21C
as in: n1C  n2C
qsample b ¼ ð1  n1C Þd ð11Þ
gV ¼ gM d ð2Þ n2C  n21C
qvaluable
The first moment of the distribution is equivalent to
Note, however, that the grades calculated by Eqs. (1)
the average volumetric grade of the composite particles,
and (2) may differ significantly depending on the error
while the second moment is a function of the spread, r,
propagation in measuring the liberation distribution of
around the average, that is:
the original sample. The objective of the interpolation
method proposed in this section is thus to reconcile the *
n1C ¼ gVC ð12Þ
liberation distribution with the assay of the sample, so
that the interpolated values (superscript b*Q) always n2C ¼ r2 þ n21C ð13Þ
satisfy the conditions:
The spread r is estimated directly from the experi-
X12  
q mental data by:
gM ¼ valuable d * j df V;
gV; *j ð3Þ " #2
qsample j¼1 11  
X X
11  
X
12 r2 ¼ 2
gV; j dfVC; j  gV; j dfVC; j ð14Þ
*j ¼1
f V; ð4Þ j¼2 j¼2
j¼1
Finally, the volumetric fractions of the composite par-
Prior to the interpolation, the volumetric fractions of
ticles are re-normalized to satisfy Eq. (4) while the para-
the composite particles (2 V j V 11) are normalized by
meters c and d are adjusted to minimize an objective
excluding the barren and fully liberated particles, that is:
function in terms of the cumulative volumetric fractions,
fV; j
fVC; j ¼ ð5Þ raw and interpolated. Therefore, an entire liberation
1  fV;1  fV;12 distribution can be interpolated by adjusting those two
The core of the technique proposed here consists in parameters only. The results are not affected by the initial
recalculating the volumetric fractions of the barren and guesses. More importantly, the interpolated distribution
the fully liberated particles by adjusting two interpola- is reconciled with the assay of the original sample.
tion parameters:
* ¼ fV;1 d ð1 þ cÞ
f V;1 ð6Þ 2.3. Transformation of interpolated distributions from
volumetric to mass basis
* ¼ fV;12 d ð1 þ dÞ
f V;12 ð7Þ
The interpolation parameters c and d also determine In order to prepare the interpolated distributions for
the average volumetric grade of the composite particles, mass balancing, it is necessary to transform the data
g VC, as per the mass conservation of the valuable from volumetric to mass basis. Knowledge of the den-
mineral: sity of each liberation class, q j , allows the transforma-
* tion to be conducted according to:
g V  f V;12
* ¼
gVC ð8Þ qj
* *
1  f V;1  f V;12 f M; * jd
* j ¼ f V; ð15Þ
qsample
Note that g V in Eq. (8) is given by Eq. (2). The
normalized fractions of the composite particles are then qvaluable
gM; *j
* j ¼ gV; ð16Þ
interpolated by the incomplete beta distribution (upj is qj
the upper limit of the jth composite class):
Obviously, q j is not measured directly. Rather it is
* ¼ Betaða; b; upj Þ
f VC;j ð9Þ calculated from image analysis, given the density of each
88 O.N. Savassi / Int. J. Miner. Process. 78 (2006) 85–92

mineral species in the sample. Alternatively, a linear classes of that size interval to be calculated by (recovery
model can be used, which is exact for binary systems: expressed as a fraction rather than a percentage):
  con
qj*¼ qvaluable  qgangue dgV;
* j þ qgangue ð17Þ fM;j Csize
Rnode
j ¼ d
feed F
ð19Þ
In multi-mineral systems, Eq. (17) represents an fM;j size

approximation by which all minerals apart from the If the node encompasses N equivalent flotation cells
valuable are lumped together as to constitute a com- in series, the recovery per cell is given by:
bined gangue species, whose density is given by:  1=N
qsample  qvaluable dgV Rcell
j ¼ 1  1  Rnode
j ð20Þ
qgangue ¼ ð18Þ
1  qV Eq. (20) is useful for analysing down-the-bank sur-
Note that Eqs. (17) and (18) reconcile the liberation veys, as demonstrated in the following section.
distribution with the density of the original sample.
3. Experimental
2.4. Mass balancing of size-liberation data
A down-the-bank survey was conducted at the prima-
The same calculation platform used for mass balanc- ry rougher of a Pb–Zn industrial flotation circuit. As
ing the size-by-size metallurgical data can be used for shown in Fig. 1, the feed was conditioned ahead of the
mass balancing the liberation data of a target size bank, with extra collector and frother being added in both
interval. To that end, the data in each stream being cells number 1 and 5. Air was supplied to the impellers
balanced must be re-organized in the following manner: via a manifold whose intake was at the tailings end of the
bank. Cell number 1 was operated as a conditioning tank
! Replace the total stream flowrate by the mass bal- with the air valve completely closed so that there was no
anced flowrate of the target size interval; concentrate being produced. All other cells were operat-
! Replace the head assay by the mass balanced assay ed with the air valves fully opened, resulting in increas-
of the valuable mineral in the target interval, g M; ing air flowrate in going from cell number 2 to 12. Froth
! Replace the weight percentage of the size intervals depth in all cells was 35 cm approximately.
by that of the liberation classes, f M,j ; Control room information provided a ballpark estimate
! Replace the size-by-size assays of the valuable min- for the feed flowrate, while the individual concentrate
eral by the grade of the liberation classes, g M,j . flowrates were measured by timed lip samples. All sam-
ples were sized and assayed to obtain size-by-size metal-
In order to force the liberation data comply with the lurgical data. A total of thirty two samples, encompassing
metallurgical data, both the flowrate of the target size four size intervals ( 75 + 53 Am and cyclosizer CS2, CS3
interval and the grade of the valuable mineral must be and CS4) in eight streams were analysed by QEM * SEM
assigned a very small standard deviation. This keeps (Sutherland and Gottlieb, 1991) to obtain liberation data.
those values fixed upon mass balancing the liberation
data. Using zero standard deviations should be avoided, 4. Results
though, as this may render the mass balancing algo-
rithm unstable. 4.1. Metallurgical data
Given the flowrate of the target size interval in the
feed and concentrate streams of any mass conservation Galena, sphalerite and pyrite assays were calculated
node, F size and C size, allows the recovery of the liberation from Pb, Zn and Fe assays according to mineral stoi-

EtX EtX
MIBC MIBC
air

feed

1 2 3+4 5+6 7+8+9 10+11 12 tail

Fig. 1. Industrial rougher bank from which the liberation was obtained.
 O.N. Savassi / Int. J. Miner. Process. 78 (2006) 85–92 89

Table 1
Stream characteristics before and after mass balancing
Stream Flow (tph) Solids (%) Galena (%) Sphalerite (%) Pyrite (%) NSG (%)
Raw Bal Raw Bal Raw Bal Raw Bal Raw Bal Raw Bal
Feed 154 182 35 34 8.71 8.35 14.1 14.0 39.7 39.4 37.5 38.3
Con 2 2.5 2.7 54 56 46.3 46.3 11.0 11.0 33.4 33.5 9.25 9.13
Con 3 + 4 6.4 6.9 55 56 43.1 43.2 11.6 11.6 34.9 34.8 10.4 10.3
Con 5 + 6 6.6 6.0 55 52 38.8 38.8 12.5 12.5 37.5 37.5 11.2 11.2
Con 7 + 8 + 9 5.6 5.4 56 55 31.9 32.0 12.9 12.8 42.4 41.5 12.9 13.6
Con 10 + 11 2.4 2.3 56 55 27.9 28.0 13.1 13.1 44.2 44.2 14.7 14.7
Con 12 0.9 0.9 54 54 24.0 24.0 12.7 12.7 46.9 46.8 16.3 16.4
Tailing 129 158 32 33 3.88 3.83 14.2 14.2 39.4 39.6 42.5 42.4

chiometry. The non-sulphide gangue assay was estimat- around 63 Am and above. Note also that even at 11 Am
ed as the remainder. The mass balancing results in there is still more than 30% of the galena locked with
Table 1 show minor alteration in relation to the raw gangue. This highlights the impact of the composite
metallurgical data. The only significant exception oc- particles upon grade-recovery in an industrial flotation
curred for the feed flowrate, as a consequence of the circuit, since the ultimate separation is determined by
poor initial estimate. Equally good results are shown in the locking of the valuable mineral.
Fig. 2 for the size-by-size data. The strong correlation As shown in Appendix A (AVAILABLE ON-LINE)
between raw and mass balanced results in this case is the average particle size of each cyclosizer interval in
due to some extent to the low data redundancy, since Table 2 takes into account the effect of particle density
there are only eight streams for a total of six mass throughout the twelve liberation classes. Note the in-
conservation nodes. Collecting samples of the interme- verse linear relationship between the average particle
diate tailings, by syphoning pulp from the outlet of size and the proportion of liberated galena.
individual cells, would improve data redundancy in Prior to mass balancing the liberation data, various
similar cases. techniques were tentatively applied for the stereological
correction. Unfortunately, this resulted in recoveries
4.2. Liberation data estimates well above 100% for many of the liberation
classes, which precludes the use of the stereological
The liberation distributions obtained by QEM * SEM correction in the present case.
are summarized in Appendix A (AVAILABLE ON- Comparison between raw and interpolated libera-
LINE). The proportion of liberated particles with volu- tion distributions is shown in Fig. 3. As can be seen
metric grade above 90% in each size interval is shown there is little bias in the fine size intervals, as indicated
in Table 2 for the rougher feed. As can be seen, by both the correlation coefficient and the slope of
liberation in this case becomes very poor for particles the correlation. Most significant alteration is in the
 75 + 63 Am interval, suggesting that the coarse par-
80 ticles were more strongly affected by the propagation
of the experimental error in measuring the liberation
data.
60 Results of mass balancing the interpolated distri-
after mass balancing

butions are shown in Fig. 4. Two important aspects

must be emphasised: a) the flowrate of the valuable
40

Table 2
size wt% slope = 1.00 r2 = 0.99
Galena liberation in the rougher feed
Galena slope = 1.00 r2 = 1.00
20
Sphalerite slope = 1.00 r2 = 0.99
Size interval Weighted average Proportion
particle size (Am) of liberated
Pyrite slope = 1.00 r2 = 1.00
galena (%)
NSG slope = 1.00 r2 = 1.00
0 75 + 53 Am 63 13
0 20 40 60 80
Cyclosizer CS2 24 55
size-by-size assays and weight percent
Cyclosizer CS3 16 63
Fig. 2. Comparison between raw and mass balanced size-by-size data. Cyclosizer CS4 11 69
90 O.N. Savassi / Int. J. Miner. Process. 78 (2006) 85–92

100 100
combined rougher concentrate
80 80

recovery of galena
after interpolation

60 60

40 40

-75+53 slope = 0.92 r 2 = 0.93

20 CS2 slope = 0.97 r 2 = 0.97
20
CS3 slope = 1.02 r2 = 0.98
CS4 slope = 1.03 r 2 = 0.98
0
0
0 20 40 60 80 100 0 10 20 30 40 50 60 70
volumetric fraction by image analysis average particle size (microns)
Fig. 3. Comparison between raw and interpolated volumetric fractions Fig. 5. Effect of particle size upon the recovery to the combined
for liberation classes of four size intervals. concentrate of an industrial rougher bank.

mineral in each stream was calculated from the mass not smooth since no interpolation was conducted on the
balanced metallurgical data and kept fixed upon mass direction of the particle size. Note the sharp decrease in
balancing the liberation data; and b) the mass bal- recovery outside the 10 to 30 Am size range, a trend
ancing of the liberation data was conducted for the observed in many other industrial flotation systems
same class throughout different streams, while the (Lynch et al., 1981). The drop in ultrafines range is
interpolation was conducted for different classes in due to low efficiency of bubble-particle collision, as
the same stream (a cross-check between interpolation those particles tend to follow the pulp streamlines
and mass balancing). Therefore, the fact that the strictly and thus slip around the bubbles (Dobby and
distributions balance well demonstrates that the inter- Finch, 1987). The behaviour of the coarse particles, on
polation renders stronger physical meaning to the raw the other hand, corresponds to the lack of liberation (see
liberation data. Table 2).
Fig. 6 illustrates the effect of particle liberation
4.3. Effect of size and liberation upon particle recovery upon the recovery of a target size interval. As can
be seen, recovery increases with increasing particle
The effect of particle size upon galena recovery is liberation, the gradient being much sharper at both
shown in Fig. 5 for mass balanced results. The curve is ends of the distribution. Note the scatter of the esti-

100 20
from raw liberation data
after interpolation and
80
15 mass balancing
recovery of class
after mass balancing

60

10

40

-75+53 slope = 1.00 r 2 = 0.99 5

20 CS2 slope = 1.01 r 2 = 1.00 cell number 2
CS3 slope = 1.01 r 2 = 1.00 cyclosizer interval = CS3
CS4 slope = 1.03 r 2 = 1.00
0
0
0 20 40 60 80 100
0 20 40 60 80 100
interpolated mass fraction volumetric grade of class
Fig. 4. Comparison between interpolated and mass balancedmass Fig. 6. Effect of particle liberation upon the recovery to the concen-
fractions for liberation classes of four size intervals. trate of an industrial rougher cell.
 O.N. Savassi / Int. J. Miner. Process. 78 (2006) 85–92 91

Table 3 Table 6
Down-the-bank recovery of the liberation classes in the 75 + 53 Am Down-the-bank recovery of the liberation classes in the CS4 cyclosi-
size interval zer interval
Nominal particle size = 63 Am Nominal particle size = 10 Am
Vol. grade Cell 2 Cell 4 Cell 5 Cell 7 Cell 10 Cell 12 Vol. Grade Cell 2 Cell 4 Cell 5 Cell 7 Cell 10 Cell 12
0 0.01 0.02 0.02 0.01 0.01 0.01 0 1.32 1.81 1.93 1.37 1.03 0.83
0 to10 0.02 0.07 0.11 0.08 0.05 0.03 0 to10 4.73 6.85 5.75 3.65 3.53 3.32
10 to 20 0.12 0.30 0.54 0.33 0.25 0.13 10 to 20 6.86 6.96 8.08 6.32 4.04 3.63
20 to 30 0.18 0.40 0.76 0.44 0.34 0.18 20 to 30 7.34 7.29 8.73 6.97 4.27 3.84
30 to 40 0.24 0.52 1.01 0.56 0.45 0.24 30 to 40 7.82 7.73 9.38 7.62 4.55 4.10
40 to 50 0.30 0.68 1.34 0.73 0.58 0.31 40 to 50 8.31 8.31 10.2 8.37 4.93 4.46
50 to 60 0.38 0.90 1.79 0.96 0.77 0.41 50 to 60 8.91 9.07 11.2 9.31 5.45 4.94
60 to 70 0.48 1.24 2.49 1.31 1.06 0.58 60 to 70 9.63 10.1 12.5 10.5 6.19 5.65
70 to 80 0.61 1.84 3.71 1.95 1.57 0.90 70 to 80 10.5 11.6 14.4 12.3 7.36 6.78
80 to 90 0.80 3.11 6.33 3.40 2.78 1.69 80 to 90 11.7 14.0 17.5 15.2 9.55 8.99
90 to 100 0.75 8.04 15.8 9.62 8.03 6.55 90 to 100 12.8 24.6 26.0 20.2 13.4 12.6
100 1.64 7.07 15.7 14.8 10.0 7.35 100 19.8 24.4 27.1 22.1 22.2 16.4

mates arising from the experimental error in the raw

Table 4
liberation data.
Down-the-bank recovery of the liberation classes in the CS2 cyclosi- Down-the-bank recoveries calculated by Eq. (20) are
zer interval presented in Tables 3–6. For the sake of simplicity,
Nominal particle size = 22 Am results from only one of the cells in each group are
Vol. grade Cell 2 Cell 4 Cell 5 Cell 7 Cell 10 Cell 12
reported. As can be seen, there is a clear trend for
recovery to increase with particle volumetric grade, a
0 0.36 0.52 0.49 0.41 0.40 0.31
0 to10 1.22 1.65 2.13 1.28 1.20 1.11
trend that is even more pronounced for the coarser
10 to 20 2.89 3.98 4.29 2.33 2.68 1.73 particles. Also, the coarser particles are more strongly
20 to 30 3.47 4.81 5.02 2.72 3.15 1.97 affected by the extra collector addition in cell number 5.
30 to 40 4.05 5.65 5.78 3.14 3.63 2.24 The recovery of all classes increases from cells 2 to 4
40 to 50 4.71 6.61 6.66 3.65 4.18 2.58 due to increasing airflow mainly. The drop from cells 5
50 to 60 5.53 7.77 7.74 4.31 4.87 3.04
60 to 70 6.59 9.26 9.16 5.22 5.81 3.70
to 12, on the other hand, is related to the destabilization
70 to 80 8.08 11.3 11.2 6.61 7.22 4.80 of the froth structure as the pulp became depleted of
80 to 90 10.4 14.6 14.4 9.14 9.81 7.06 hydrophobic particles (Subrahmanyan and Forssberg,
90 to 100 17.5 21.6 19.1 13.3 11.9 10.3 1988). This was corroborated by the visual aspect of the
100 11.8 18.7 25.8 21.7 16.0 14.7 froth towards the tailings end of the bank, with the top
layer exhibiting much larger bubbles and reduced solids
coating as well as extensive bubble bursting and bubble
coalescence.
Table 5
Down-the-bank recovery of the liberation classes in the CS3 cyclosi-
zer interval 5. Conclusions
Nominal particle size = 15 Am
Vol. grade Cell 2 Cell 4 Cell 5 Cell 7 Cell 10 Cell 12
Using simple interpolation and mass balancing tech-
niques can minimize the propagation of the experi-
0 0.66 1.07 1.15 0.71 0.61 0.54
0 to10 2.46 3.36 2.99 2.53 2.21 1.69
mental error in the estimation of the recovery of size-
10 to 20 6.55 7.97 7.56 5.42 5.63 4.13 liberation classes. Forcing the liberation data comply
20 to 30 7.15 8.62 8.27 5.74 6.03 4.38 with the metallurgical data increases the reliability of
30 to 40 7.60 9.14 8.87 5.99 6.31 4.55 the recovery estimates, since the metallurgical data is
40 to 50 8.01 9.65 9.47 6.21 6.55 4.69
obtained from samples of much larger masses. Results
50 to 60 8.43 10.2 10.1 6.45 6.78 4.81
60 to 70 8.89 10.8 10.9 6.70 7.02 4.93 from a down-the-bank survey indicate a clear trend for
70 to 80 9.45 11.7 11.9 7.02 7.30 5.06 increasing recovery with particle volumetric grade,
80 to 90 10.5 13.3 14.2 7.93 8.25 5.65 which is even more pronounced for the coarse particles.
90 to 100 13.5 18.9 23.1 10.1 9.98 6.46 Chemical addition, air flow and froth structure also play
100 16.6 22.5 25.5 21.9 14.1 12.6 a significant role.
92 O.N. Savassi / Int. J. Miner. Process. 78 (2006) 85–92

Acknowledgements Dobby, G.S., Finch, J.A., 1987. Particle size dependence in flotation
derived from a fundamental model of the capture process. Inter-
national Journal of Mineral Processing 21, 241 – 260.
The author wishes to thank all of whom contributed Gaudin, A.M., 1939. Principles of Mineral Dressing. McGraw-Hill
for the completion of this paper, in special S. Gay and M. Book, New York.
Andrusiewicz, of the JKMRC, for early discussions on Gay, S., 1994. Liberation Modelling using Particle Sections, PhD
the stereological correction and mass balancing of size- thesis, University of Quensland.
King, R.P., Schneider, C.L., 1993. An effective SEM-based image
liberation classes, as well as G. Wilkie and P. Gottlieb for
analysis system for quantitative mineralogy. KONA 11.
assistance in obtaining and analysing the QEM*SEM King, R.P., Schneider, C.L., 1997. Comparison of stereological
data at the CSIRO back in 1996. A scholarship from the correction procedures for liberation measurements. Transactions
Brazilian Government for a PhD at the University of of the Institution of Mining and Metallurgy 104, c155 – c161.
Queensland is also acknowledged with thanks. Lynch, A.J., Johnson, N.W., Simpson, H.J., Thorne, C.G., 1981.
Mineral and Coal Flotation Circuits: Their Simulation and Con-
trol. Elsevier Scientific, Amsterdam.
Appendix A. Supplementary data Subrahmanyan, T.V., Forssberg, K.S.E., 1988. Froth stability, particle
entrainment and drainage in flotation—a review. International
Supplementary data associated with this article Journal of Mineral Processing 23, 33 – 538.
can be found, in the online version, at doi:10.1016/ Sutherland, D.N., Gottlieb, P., 1991. Application of automated quan-
j.minpro.2005.09.002. titative mineralogy in mineral processing. Minerals Engineering 4
(7-11), 753 – 770.
Weibel, E.R., 1979. Sterological Methods, vol. 1. Academic Press,
References New York.

Barbery, G., 1991. Mineral Liberation: Measurement, Simulation and

Practical use in Mineral Processing. Editions GB, Quebec.

View publication stats