You are on page 1of 10

The SAG grindability index test

Peter Amelunxen
a,
, Patricio Berrios
b
, Esteban Rodriguez
b
a
AME Ltda, Cerros de Camacho 440, Dep F-17, Santiago de Surco, Lima, Peru
b
Aminpro Chile SpA, Cerro San Cristobal 95111, Quilicura, Santiago, Chile
a r t i c l e i n f o
Article history:
Received 13 June 2013
Accepted 31 August 2013
Available online 5 October 2013
Keywords:
Semi-autogenous grinding
SAG milling
SGI test
Grinding
a b s t r a c t
In this paper, the authors undertake a critical review of the Starkey test and the publicly available infor-
mation related to the test equipment, procedures, and scale-up methodology. The following recommen-
dations are proposed to improve the test method:
1. The test should be conducted for a xed grinding time of 120 min, regardless of the time required to
reach 80% passing Tyler #10 mesh.
2. The test should be conducted with constant time intervals of 15, 30, 60, and 120 min (cumulative) in
order to facilitate the application of geostatistics to the resulting index values. This would also allow
for multiple tests to be conducted in parallel (through the use of multiple mill rollers).
3. The feed size should be prepared using a more rigorous procedure to ensure constant mass in each of
the course screen fractions.
4. The curve of nished product versus time should be modeled and the resulting index calculated from
the model for a standard feed size distribution, so that errors attributable to the sample preparation
step are minimized.
The improved feed preparation steps and the use of constant grinding intervals enables the develop-
ment of a faster alternative to the standard test that is more cost effective for high volume geometallur-
gical programs.
In addition to the updated procedures, a new calibration equation is proposed, with calibration factors
for pebble crushing, ne feed and autogenous grinding, based on information in the public literature.
Detailed descriptions of the test equipment, procedures, and calibration are provided, and it is proposed
that this become an open standard procedure for SAG mill hardness testing, particularly for soft to med-
ium-hard ores, over which range the test is most effective.
2013 Elsevier Ltd. All rights reserved.
1. Background
Since the Starkey mill was introduced more than 15 years ago,
the world has seen many changes in the approach used for SAG
circuit design (in this manuscript, the term SAG will be used to
refer to both SAG and FAG milling). While it is no longer the only
available point hardness test method, it is the only one available
based on a purely empirical tumbling mill test. Originally devel-
oped, calibrated, and marketed by Minnovex Technologies under
the name SAG Power Index

(SPI

), the technology was acquired


by SGS in 2005. The technology and scale-up calculations are
believed by many to be condential and proprietary, but during
the mid-1990s the mill and test equipment specications were
published openly and sold to various mining corporations (Baeza
and Villanueva, 1994; Gonzalez, 2000), and the primary scale-up
equation has also found its way into the public domain through
various publications and presentations. Today, any individual or
organization may freely build and market a test system, and there
are now at least six laboratories offering the test, not including SGS
and the various mining companies who purchased the equipment
in the mid-1990s. The authors believe that one of the main imped-
iments to the widespread adoption of the test has been a general
lack of awareness of the information available that is required for
constructing the equipment. This knowledge has been dissemi-
nated in various languages throughout conference proceedings,
journal papers, websites, and university research theses; indeed,
one of the aims of this paper is to rationalize these disparate tech-
nical sources into one coherent document.
A second objective of this research program is to provide a crit-
ical review of the test method, equipment specications, data
interpretation, and calibration equation, insofar as permitted by
the information available in the public domain.
0892-6875/$ - see front matter 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.mineng.2013.08.012

Corresponding author.
E-mail address: peter@aminpro.com (P. Amelunxen).
Minerals Engineering 55 (2014) 4251
Contents lists available at ScienceDirect
Minerals Engineering
j our nal homepage: www. el sevi er. com/ l ocat e/ mi neng
1.1. Nomenclature
The Starkey laboratory mill has been used since 1994 to predict
the autogenous mill specic energy (Starkey and Dobby, 1996). The
test was originally termed the Starkey SAGtest (Starkey et al., 1994)
and the laboratory SAG mill was called the Starkey laboratory SAG
mill (Starkey and Dobby, 1996), or just the Starkey SAG Mill
(Starkey et al., 1994). The test and equipment were commercialized
by Minnovex Technologies, a Canadian company that is nowpart of
SGS Lakeeld Minerals, under the name SAG Power Index

, or SPI

test (Starkey et al., 1994), terms which were later trademarked in


Canada (CIPO, 2012). During the 1990s the test equipment,
procedures and calibration information were purchased by several
institutions (Baeza and Villanueva, 1994), and the test continues to
be used in its original public form(Asmin Industrial, 2013; Aminpro
Chile, 2013), which is essentially the same as that of the SPI

(Verret
et al., 2011; Baeza and Villanueva, 1994).
The test should not be confused with the SAGDesign test devel-
oped by Starkey & Associates, which is of similar genesis but em-
ploys a different analytical approach (Starkey et al., 2006).
Because Starkey no longer bears any afliation with the test
(Starkey, 2013), the authors suggest that a new name be adopted
to reect the public nature of the test procedures. With this in
mind, we propose the terms SAG Grindability Index (SGI) and
SAG Grindability Index Test (SGI test) for the index value and
test, respectively. These terms are used for the remainder of this
manuscript when referring to the public test and methods. Usage
of the terms SPI

and SAG Power Index

will continue in the context


of the proprietary test system of the current trademark owners.
1.2. Test equipment
The Starkey laboratory mill dimensions and operating condi-
tions are described in great detail in the references listed at the
end of this paper. The important characteristics are described be-
low, and scaled drawings are included in Appendix B. The authors
test system, which was used for the work described in this paper,
was constructed to the specications described below.
The mill has a 304.8 mm diameter by 101.6 mm length (Verret
et al., 2011). The mill is equipped with six one inch by one inch
charge lifters located at 60 intervals, as can be observed in the
published photographs shown in Fig. 1a and b (Baeza and Villanu-
eva, 1994; Baeza and Domingo, 2000). A 5 kg ball charge of 1
1
=
4
in.
balls is used (Verret et al., 2011; Gonzalez, 2000).
The mill reportedly rotates at 70% of critical velocity (Verret
et al., 2011), or 54 rpm (Gonzalez, 2000). There is controversy sur-
rounding the mill rotational speed. The calculation of mill critical
velocity (V
c
) is derived from the balance of gravitational force
and the centrifugal force acting on a ball, and is given by Eq. (1),
after Austin and Concha (1994)
V
c

42:2

D d
p 1
where D is the weighted average inside liner diameter and d is the
ball diameter. Using this equation, the resulting mill rotational
speed should be 57 rpm, i.e. higher than the published value of
54 rpm.
For large mills the ball diameter is negligible relative to the mill
diameter, and the equation is often simplied to (Napier-Munn,
1996)
V
c

42:2

D
p 2
This equation yields the published mill rotational speed of
54 rpm; hence, it is assumed that the test designers used the sim-
plied equation for computing the rotational velocity.
Back-calculating the fraction of critical speed using the correct
equation indicates that the test actually operates at a rotational
speed of approximately 66% of critical, contrary to the published
value of 70%.
1.2.1. Mill rotational speed
To investigate the impact energy imparted to the ore particles
as a function of the mill rotational speed, the mill was outtted
with a transparent acrylic lid and rotated with a ball charge at var-
ious fractions of critical speed (no ore charge was used to avoid vis-
ibility loss because of dust). High denition video was used to
measure the release angle (a) and toe angle (b), where 12 oclock
is 0 (Fig. 2). Two distinct zones were observed in the video, an
abrasion zone and an impact zone where the cascading ball
impacts the toe of the charge. These can be seen in Fig. 2 and in
Video 1.
The impact energy (E
i
), in joules, of a single ball-on-liner colli-
sion was calculated using equation (3):
E
i

1
2
mv
2
3
where m is the mass of a ball (approximately 130 g) and v is the
velocity of the ball at impact, calculated from the horizontal (v
i
)
and vertical (v
j
) components of the velocity vector. Letting r equal
the mill radius (m) and v
s
be the circumferential speed of the mill
shell (m/s), the velocity components are given by
v
i
v
s
cos a 4a
Fig. 1. (a and b) The Starkey laboratory mill at El Teniente metallurgical lab (Baeza
and Domingo, 2000).
P. Amelunxen et al. / Minerals Engineering 55 (2014) 4251 43
v
j

v
s
sina
2
2gh
q
4b
where
h rcos a cos2p b 4c
Knowing the impact energy per collision, the energy per unit
time (E
t
in joules per second) is calculated assuming each ball
undergoes a single cycle per mill revolution. Since there are 38
balls in the standard SGI charge, of which approximately seven
pertain to the abrasion zone and are, therefore, not used to gener-
ate impact breakage, E
t
is given by
E
t

31v
rpm
60
E
i
5
Table 1 shows the results of the energy calculations for various
mill speeds. Also shown is the estimated specic energy input per
tonne of ore (E
s
in kWh/t), calculated for an assumed SGI of 60 min.
The data indicate that the highest impact energy per collision
occurs at a rotational speed of 50 rpm, but the highest energy input
per unit time occurs near 56 rpm. Because there is only minimal
difference between the total energy input at 54 rpm and 56 rpm,
the publicly reported speed of 54 rpm is likely appropriate despite
the fact that it is not 70% of critical. Ongoing work is aimed at test-
ing different ores with different speeds to validate this conclusion.
1.3. Original test procedures
The original procedures for the test consist of stage crushing the
sample to two control points100% minus
3
=
4
in. and 80% passing
in. (Verret et al., 2011). The sample is screened and then placed
in the mill with the ball charge. The test is a dry batch grinding test
in which the mill is rotated and the sample ground until it reaches
80% passing Tyler #10 mesh, or 1.7 mm (Starkey et al., 1994). Be-
cause the time required to achieve 80% passing 1.7 mm is not
known before starting the test, an initial grinding time is selected,
the sample is ground then removed from the mill and screened on
a large Tyler #10 mesh screen. The mass of minus 10 mesh
material produced per revolution is then used to approximate
the time required for the subsequent grinding interval. The entire
charge is then returned to the mill and ground for the next time
period. The sequence is repeated until the test achieves completion
(Amelunxen, 2003, pp. 1921). Fig. 3 shows the results of a typical
test.
1.4. Revised test procedures
The above procedures were reviewed in a previous publication
by one of the authors (Amelunxen, 2003, pp. 1921, 5557, 6265)
and several recommendations were made to improve the test
(Amelunxen, 2003 p. 108). These are summarized as follows:
The curve of mass retained on Tyler #10 mesh (1.7 mm) versus
cumulative grinding time should be mathematically modeled
and the resulting index be interpolated or extrapolated. Several
alternative model forms were suggested, depending on the ore
hardness.
The practice of performing several iterations near the end of the
test (the three points above 120 min on the x-axis in Fig. 3) is
not necessary given that the curve can be modeled and the
end result interpolated or extrapolated.
The test should be conducted for constant grinding time inter-
vals, as this facilitates the mathematical or geostatistical han-
dling of the data (it generates additive parameters, rather
than a non-additive index values).
These recommendations have been adopted in the present
work. The times selected for the xed grinding periods are 15,
30, 60 and 120 min (cumulative). Beyond 120 min, the slope of
the curve starts to approach zero. As a result, for ores with SGI val-
ues over approximately 150 min, relatively small errors in the feed
size or competency can lead to very large differences in the result-
ing SGIin other words, the test begins to break down. This is
mostly likely a result of the fact that the impact energies required
to fracture the coarsest particles in the feed are higher than those
attainable with such a small mill.
This problem has been mitigated, in part, by introducing a small
change to the feed preparation procedures. The new procedures
require the feed to be prepared to a controlled size distribution.
The new size distribution was selected to reect the average
size distribution from the authors database of SGI tests conducted
for commercial projects before the new procedures were
implemented. It is shown in Table 2.
The proposed standard test procedures are appended.
Fig. 2. Image capture from SGI mill at 54 rpm showing two breakage zones.
Table 1
SGI mill energy vs. rotational speed.
Speed (rpm) Release
angle ()
Toe angle () Ei (j/ball) Et (j/s) Es (kWh/t)
48 28 137 0.35 8.84 4.42
50 21 133 0.36 9.30 4.65
52 20 128 0.35 9.45 4.72
54 13 126 0.35 9.95 4.97
56 10 120 0.34 9.98 4.99
58 2 105 0.30 9.12 4.56
60 0 90 0.26 7.98 3.99
SPI Test Result
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150
Time (minutes)
R
e
l
a
t
i
v
e

P
e
r
c
e
n
t

P
l
u
s

T
y
l
e
r

s
t
a
n
d
a
r
d

#
1
0

m
e
s
h

(
1
.
7

m
m
)
SPI Definition
Test Iterations
Model
Fig. 3. SPI

test grinding curve showing completion point, from (Amelunxen, 2003,


p. 21).
44 P. Amelunxen et al. / Minerals Engineering 55 (2014) 4251
1.5. Revised test interpretation
Several equation forms were investigated for modeling the SGI
test, including those described previously by Amelunxen (2003, p.
55). It was found that the best performance is given by a variant of
the Swebrec

equation (Ouchterlony, 2005) often used for model-


ing cumulative size distributions. The form of the equation is
Pt
f x
1 f x
6
where P(t) is the percent retained on a Tyler #10 mesh screen at a
grinding time of t (minutes) and
f x m
ln0:25
lnm
20


x
; 7a
m
x

ln
t
0
tx

ln
t
0
t
50
7b
and
m
20

ln
t
0
t
20

ln
t
0
t
50
7c
where t
x
is the indeterminate time and t
0
, t
20
, and t
50
are tted
parameters that represent the cumulative grinding times at which
P(t) is equal to 0%, 20%, and 50%, respectively. Hence, by denition
the SGI is equal to t
20
.
2. Lower cost, lower precision test procedures
Analysis of a data set of approximately 150 tests conducted
using the SGI mill indicate that it is possible to produce a lower
cost, lower precision version of the SGI test by performing a single
grinding iteration with the standard feed charge for a constant
time interval of 60 min. The resulting product is then screened
on a Tyler #10 mesh screen (1.7 mm) and an empirical equation
is used to determine the SGI from the screen analysis results. The
new test only requires a single grinding iteration of 60 min, so
one technician can operate three or maybe four individual SGI
mills in parallel. This allows for a team of four technicians to pro-
duce approximately three tests per hour, or 24 tests in an eight
hour shift (two technicians for sample preparation, one to operate
the SGI mills and sieve shakers, and one for miscellaneous tasks
including sample logging, data logging, quality control, and clean-
up). We can estimate the unit sales cost under such a scenario by
assuming a gross labor cost of US$60/h and generous allowances
for xed overheads and other costs. The estimates, shown in
Table 3, yield a unit price of US$150 per test, compared to approx-
imately US$450 for the standard test procedures detailed above.
Because anybody can build and operate the test equipment, no
additional fees are assumed.
By adopting the short test procedures, only minor loss of delity
is incurred, as the empirical correlation shows a coefcient of
determination of almost 0.98 (Fig. 4). This is equivalent to approx-
imately 0.16 kWh/t at one sigma level of condencei.e. a stan-
dard error of only 2% for the data set used in this study.
Notwithstanding the marginally lower precision, the modied
test procedures have a superior value proposition. For example
(see Table 4), with a hypothetical budget of US$25,000, one could
elect to perform either 56 standard tests or 167 modied tests.
The modied tests would yield mean mill power requirements or
mean throughput estimates that have approximately half the
test-related error of those derived from the standard tests. In addi-
tion to the cost benets, the modied test delivers an index value
(the mass retained after 60 min) that is naturally additive and
therefore is more conducive to geostatistical handling (for a discus-
sion on the mathematical additivity of the SGI, see Amelunxen,
2003, pp. 5465).
Table 2
Standard SGI feed size distribution.
Tyler screen Opening (lm) Mass (g)
3/4
00
19,050 0
1/2
00
12,700 400
3/8
00
9525 400
4-Jan 6350 400
#6 3350 400
#10 1700 150
pan 0 250
Table 3
Cost estimate for short and standard SGI tests.
Calculation Short SGI
test
Standard SGI
test
Units
Tests per hour 3 1 tests/h
Hours per shift 8 8 h
No. of technicians 4 4 techs
Productivity factor 1 1
Cost of labor, gross $60 $60 $/h
tech.
Cost per test, labor $80 $240 $/test
Overheads and other costs
Administration (20%) $16 $48 $/test
Engineering and supervision
(15%)
$12 $36 $/test
Quality assurance and control
(10%)
$8 $24 $/test
Financing and depreciation (7%) $6 $17 $/test
Maintenance (6%) $5 $14 $/test
Operating supplies (5%) $4 $12 $/test
Contingency (5%) $4 $12 $/test
Prot (20%) $16 $48 $/test
Total overheads $70 $211 $/test
Total cost per test $150 $451 $/test
Fig. 4. Comparison of modied (short) SGI test and standard SGI test.
Table 4
Error estimates for a hypothetical characterization program, showing superior results
from the modied test.
Test type: Modied Standard
Budget available $25,000 $25,000
Cost $150 $450
No. of tests 167 56
Assumed mean kWh/t (metric) 7.33 7.33
Standard error of mean kWh/t (metric) 0.012 0.021
Relative standard error of mean 0.16% 0.28%
P. Amelunxen et al. / Minerals Engineering 55 (2014) 4251 45
3. SAG mill specic energy calibration
3.1. Closed circuit without pebble crushing
The original calibration equation was developed during the
mid-1990s under sponsorship by the Mining Industry Technology
Council of Canada (MITEC) and the results were published in the
proceedings of the SAG 96 conference held in Vancouver, Canada.
The relevant equation expressed the mill power draw per unit
throughput as a function of the SPI and T
80
, as follows (Starkey
and Dobby, 1996):
SAG
kWh
t

2:2 0:1SPI
T
0:33
80
8
where the T
80
is the 80% passing size of the SAG screen undersize. In
1999 additional data points were published, then totaling 13 differ-
ent plants (Kosick and Bennett, 1999) together with a new, nonlin-
ear, calibration equation. In that publication, however, the value
scale of the abscissa was not shown (Fig. 5a). Nevertheless, by
assuming that the four valid concentrators from the 1996 publica-
tion (Starkey and Dobby, 1996) are also included in the 1999 graph
(why would they not be?), the value scale can be back-calculated by
superposing the 1996 points on the 1999 plot and tting the value
of n to give the best match. The results (Fig. 5b) yield the scale off
the abscissa, which can then be used to derive the values of the
remaining points and the constants of the new nonlinear calibration
equation.
The equation, shown in the formgiven by Dobby et al., (2001), is
SAG
kWh
t
5:9
SGI

T
80
p

0:55
9
The calibration equation shown above is the primary calibration
equation and is only valid for the reference circuit consisting of a
SAG mill operating in closed circuit without pebble crushing and
with nominal 6 in. SAG mill feed (Kosick and Bennett, 1999). For
mills operating with ner feed and/or in-circuit pebble crushing,
an adjustment factor is used to account for the reduced SAG mill
specic energy (Dobby et al., 2001). The adjustment factor, termed
f
SAG
, is then used as follows (Dobby et al., 2001):
SAG
kWh
t
5:9
SGI

T
80
p

0:55
f
SAG
10
3.2. Fine feed and pebble crushing
Fig. 6, after Amelunxen (2003, p. 24), shows the calibration scat-
terplots for grinding mills operating with ne feed (left) and with
in-circuit pebble crushers (right). The graphs indicate that the val-
ues of f
SAG
are:
With ne feed, f
SAG
= 0.9.
With pebble crushing, f
SAG
= 0.85.
Presumably, f
SAG
values are additive for optimized circuits. SAG
mills operating in closed circuit with both ner feed and in-circuit
pebblecrushing would, therefore, show f
SAG
values approximately
equal to product of the ne feed and pebble crushing correction
factors, i.e. approximately 0.77. Audit data from such a circuit
(Candelaria) is available (Amelunxen et al., 2011) and corroborates
this notion (Fig. 7).
3.3. Open circuit SAG milling
More recently, data from Los Bronces SAG mill circuit has be-
come available (Becerra and Amelunxen, 2012), in which the SAG
mill was operating in both SABC-B conguration, where the
crushed pebbles report to the ball mill feed chute, and SABC-AB
conguration, in which the pebble crusher product is split between
both SAG and ball mill sections. In this conguration, the apparent
transfer size is calculated as the weighted average of the SAG
classier undersize and the pebble crusher product. The data are
shown in Fig. 8. Some interesting observations can be made:
The mill efciency (if f
SAG
can be taken as a measure of ef-
ciency) was very similar to that of the Candelaria mill, in spite
y = 5.9x
0.55
0
2
4
6
8
10
12
14
16
18
20
22
0 1 2 3 4 5 6 7 8 9
P
l
a
n
t

k
W
h
/
t
SPI * T
80
-0.5
10
Fig. 5. SPI

Calibration equation, left (Kosick and Bennett, 1999), with superposi-


tion of 1996 data points shown by blue squares (right). (For interpretation of the
references to color in this gure legend, the reader is referred to the web version of
this article.)
Fig. 6. SAG specic energy calibration plots for ne feed (left) and pebble crushing (right).
46 P. Amelunxen et al. / Minerals Engineering 55 (2014) 4251
of operating in open circuit or semi-open circuit conguration.
Counterintuitively, the mill efciency was higher (lower f
SAG
)
during the three surveys in which some of the crushed pebbles
were returned to the ball mill.
As expected, the apparent transfer size of the SABC-AB congu-
ration was ner than that of the SABC-B conguration, reect-
ing the reduced ow of coarse pebble crusher product to the
ball mills. This observation may partly explain the lower f
SAG
values for the SABC-B conguration. Because these particles
tend to have sharper, less rounded edges than their same-sized
counterparts in the SAG mill discharge, they may lead to greater
production of nes or nished product as these edges get
rounded off upon their return to the SAG mill. Therefore, as
long as they are already ner than the SAG mill classier slots
and the SAG mill throughput is limited by the size reduction
of the coarse, plus-grate size material, then passing some of
them through the SAG mill one more time before sending them
to the ball mills may not be such a bad idea.
3.4. Fully autogenous milling
The Bagdad mill consists of ve autogenous grinding lines fol-
lowed by ball milling. Using the available SPI and plant survey data
(Amelunxen et al., 2011), we can compare the f
SAG
values from Bag-
dad to those of the other circuit congurations discussed above. It
can be seen that the Bagdad ABC-A conguration averages f
SAG
val-
ues of approximately 0.62. This is an improvement over that of the
SABC-B and SABC-AB conguration of Los Bronces discussed above,
but it should be noted that this does not include the power of the
pebble crushers, and Bagdad operated with a high circulating load
(routinely above 50%), as can be seen from the appended data set.
Fig. 9 shows the Bagdad calibration data relative to the previously
discussed points.
While the relationship between predicted and actual specic
energy is weaker with the Bagdad data (this is likely a result of dif-
ferences in pebble crusher loads), the average f
SAG
value of 0.62 can
nevertheless be used as a guide for those considering fully autoge-
nous grinding circuits.
3.5. Discussion of error
At this point, it should be noted that there is some scatter in
Fig. 9. This is expected due to the difculties of obtaining represen-
tative samples and mass balance estimates for large grinding cir-
cuits, particularly those involving high circulating loads and
pebble crushing. Amelunxen (2003, p. 107) estimates the standard
error of the specic energy at between 20% and 26% due to error of
the primary calibration equation and sub-models for T
80
and f
SAG
.
Furthermore, studies performed at Chino Mines and elsewhere
indicate that normal drill-hole collar intervals may be outside the
range of typical SGI variograms (Amelunxen, 2001; Amelunxen,
2003, p. 107), and therefore, error reduction due to geostatistical
correlation may not be signicant (at least not in the horizontal
directions). As such, the central limit theorem provides a reason-
able basis for estimating error for different forecast periods; i.e.
S
r

n
p 11
where S is the standard error of the mean throughput estimate for a
given production period, r is the standard error of the throughput
estimate for a point hardness sample (including calibration errors),
and n is the number of samples that represent the ore in that pro-
duction period.
Note that when modeling for design purposes, one must also
consider the error inherent in the estimated mill power draws;
Morrell (1993) estimates this at 5.4% for one sigma.
For a complete analysis and discussion of the error, including
the error as it relates to the forecasting time period, refer to Amel-
unxen (2003, pp. 66104).
3.6. Estimating the transfer size
One of the limitations of the scale-up methodology described
above is that the transfer size must be known in order to estimate
the specic energy required for a given SAG mill circuit. One can
either select a transfer size, and then determine the mill power re-
Fig. 7. SAG specic energy calibration plot for both ne feed and pebble crushing,
after (Amelunxen et al., 2011).
Fig. 8. Los Bronces data set showing open circuit and partially open circuit SAG
milling.
Fig. 9. Bagdad survey data.
P. Amelunxen et al. / Minerals Engineering 55 (2014) 4251 47
quired to achieve it (given a throughput target), or one can select
the installed power and tonnage and then use the resulting trans-
fer size to size the ball mills (presumably on the basis that the mill
operators would then select the grates, pebble ports, crushing con-
guration, and classier opening such that the calculated transfer
size can actually be achieved in practice).
The transfer size conundrum results from fact that in practice
the transfer size is not exogenousit is itself a function of the mill
throughput and ore hardness (for a xed system). Over long time
periods it may be reasonable to assume that the grates, pebble
ports, and other system parameters will be optimized to achieve
the desired mean transfer size, but for simulations targeting spe-
cic process congurations or shorter timeframes (such as those
performed on point samples representing ore reserve blocks or
snapshot process audits), the energy based scale-up methodol-
ogy described above should be used in conjunction with either
phenomenological or semi-empirical models that incorporate
breakage and selection functions in some form or another. For this
purpose the authors favor an energy-based phenomenological
breakage model that will be presented as part of a future
publication.
4. Conclusions
This paper has provided a comprehensive review of the publicly
available information related to the Starkey test and calibration
methodology. It has also provided an analysis of test data gener-
ated using the SGI test mill constructed by the authors as shown
in the appended schematics. A case has been made for making
the following changes to the terminology, procedures, and scale-
up methods of the test:
The terms Starkey mill, Starkey test, and Starkey index are
obsolete, confusing, and should be dropped from the lexicon.
The authors have proposed the terms SAG grindability index
or SGI mill and test.
To reduce error, the feed size distribution for the test should
be xed for the coarse fractions above the Tyler #10 mesh
screen.
The test should be conducted for xed grinding time intervals of
15, 30, 60, and 120 min, after which the test is completed.
The grinding curve should be modeled using a variant of the
Swebrec equation and the SGI extrapolated or interpolated
from the tted equation.
A shorter, lower cost/lower precision version of the test has
been recommended for large geometallurgical projects that
require very large datasets and can tolerate relatively minor
reductions in test delity.
An updated calibration equation has been provided, together
with calibration factors for ner feed, pebble crushing, and fully
autogenous milling. The appendices contain detailed drawings of
the laboratory mill, a summary of published calibration data, and
the test procedures for the standard and short versions of the SGI
test. The authors suggest that these should constitute a freely
available, open standard SGI test.
Appendix A. Previously published calibration data
See Figs. A1.a, A1.b, A2.a and A2.b.
No Source Plant Circuit Feed
size F80
SPI TPH
dry
Pebble
circ load
Power Prod
size T80
S.E. SPI
shell
S.E. Plant
shell
Fsag
name (in) (min) (t/hr) (%) (kW) (lm) (kWh/t) (kWh/t)
1 18 Bagdad ABC-A 5.6 100 624 58 4501 1857 8.62 7.22 0.84
2 18 Bagdad ABC-A 5.5 91 854 57 3974 4725 6.89 4.65 0.67
3 18 Bagdad ABC-A 3.8 92 844 56 4474 4843 6.89 5.30 0.77
4 18 Bagdad ABC-A 3.2 120 655 59 4477 7914 6.97 6.83 0.98
5 18 Bagdad ABC-A 4.3 118 647 55 4490 6678 7.21 6.93 0.96
6 18 Bagdad ABC-A 3.1 120 721 66 4116 1523 10.96 5.71 0.52
7 18 Bagdad ABC-A 2.7 82 696 75 4158 2173 8.04 5.97 0.74
8 18 Bagdad ABC-A 6.5 149 691 89 4434 2971 10.24 6.42 0.63
9 18 Bagdad ABC-A 4.8 101 663 42 4322 1096 10.91 6.51 0.60
10 18 Bagdad ABC-A 3.5 140 640 89 3810 990 13.41 5.95 0.44
11 18 Bagdad ABC-A 4.2 114 642 75 3915 1748 10.23 6.10 0.60
12 18 Bagdad ABC-A 4.6 118 696 78 4274 3685 8.49 6.14 0.72
13 18 Bagdad ABC-A 7.8 112 521 61 4452 963 11.92 8.54 0.72
14 18 Bagdad ABC-A 5.1 92 670 62 4382 1685 9.17 6.54 0.71
15 18 Bagdad ABC-A 6.0 121 591 33 4366 619 14.05 7.39 0.53
16 18 Bagdad ABC-A 7.7 104 670 61 4341 1092 11.08 6.47 0.58
17 18 Bagdad ABC-A 3.6 132 554 72 4300 1105 12.61 7.77 0.62
18 18 Bagdad ABC-A 5.1 127 589 38 4409 756 13.70 7.48 0.55
19 18 Bagdad ABC-A 3.1 112 498 53 4453 386 15.38 8.94 0.58
20 18 Bagdad ABC-A 3.0 88 599 77 4114 764 11.13 6.87 0.62
21 18 Candelaria SABC-B 3.8 87 1360 26 9982 1573 9.09 7.34 0.81
22 18 Candelaria SABC-B 3.7 119 1090 25 10907 786 13.07 10.01 0.77
23 18 Candelaria SABC-B 3.6 93 1457 26 10436 1888 8.97 7.16 0.80
24 18 Candelaria SABC-B 3.7 104 1282 21 10566 1764 9.71 8.24 0.85
48 P. Amelunxen et al. / Minerals Engineering 55 (2014) 4251
Appendix A (continued)
No Source Plant Circuit Feed
size F80
SPI TPH
dry
Pebble
circ load
Power Prod
size T80
S.E. SPI
shell
S.E. Plant
shell
Fsag
name (in) (min) (t/hr) (%) (kW) (lm) (kWh/t) (kWh/t)
25 18 Candelaria SABC-B 3.6 132 1500 26 10505 2899 9.66 7.00 0.72
26 18 Candelaria SABC-B 3.4 89 1596 19 10421 3689 7.26 6.53 0.90
27 18 Candelaria SABC-B 3.4 110 1358 23 10620 1905 9.81 7.82 0.80
28 18 Candelaria SABC-B 4.0 131 1964 21 9978 2242 10.33
29 18 Candelaria SABC-B 4.1 151 1297 21 10602 2178 11.25 8.18 0.73
30 18 Candelaria SABC-B 3.3 116 1693 27 10317 3171 8.78 6.09 0.69
31 18 Candelaria SABC-B 4.0 124 1693 25 10568 3311 9.00 6.24 0.69
32 18 Candelaria SABC-B 3.9 171 1151 27 10734 2674 11.39 9.33 0.82
33 18 Candelaria SABC-B 2.9 176 1072 36 10852 2100 12.37 10.12 0.82
34 19 L. Bronces SABC-B 1.8 80.8 2488 11211 5200 6.28 4.51 0.78
35 19 L. Bronces SABC-AB 2.1 72.4 3173 10755 3950 6.38 3.39 0.58
36 19 L. Bronces SABC-AB 1.2 67.4 2970 10716 3300 6.44 3.61 0.61
37 19 L. Bronces SABC-AB 1.7 75.9 2922 11327 2800 7.19 3.88 0.58
38 19 L. Bronces SABC-B 2.2 90.1 2656 13882 5750 6.49 5.23 0.87
39 19 L. Bronces SABC-B 3.1 63.0 2344 13059 4200 5.81 5.57 1.04
40 19 L. Bronces SABC-B 4.0 68.4 2553 12945 5250 5.72 5.07 0.96
41 19 L. Bronces SABC-B 3.5 73.4 2511 12932 3300 6.75 5.15 0.83
42 19 L. Bronces SABC-B 2.8 74.4 2563 12847 5300 5.97 5.01 0.91
43 3 HVC SABC-A 32.0 1659 6380 2570 4.21 3.54 0.84
44 3 HVC SABC-A 43.0 1252 6786 1925 5.37 4.99 0.93
45 3 HVC SABC-A 60.0 919 6722 1880 6.49 6.73 1.04
46 3 Selbaie SABC-A 42.0 297 2555 323 8.66 7.91 0.91
47 3 Selbaie SABC-A 54.0 306 3024 337 9.83 9.09 0.93
48 3 Selbaie SABC-A 53.0 268 3200 308 9.97 10.99 1.10
49 3 Q-Cartier SABC-A 8.0 1018 3655 515 3.06 3.30 1.08
50 3 Q-Cartier SABC-A 9.0 1164 4514 480 3.33 3.57 1.07
51 3 Q-Cartier SABC-A 12.0 1045 5005 460 3.94 4.41 1.12
52 16 unknown 18.4 18.4 1.00
53 16 unknown 14.5 14.3 0.98
54 16 unknown 14.4 14.1 0.98
55 16 unknown 11.3 11.0 0.98
56 16 unknown 9.9 9.2 0.93
57 16 unknown 8.8 7.8 0.89
58 16 unknown 6.3 6.8 1.09
59 16 unknown 4.5 3.7 0.82
Fig. A1.a. Detail SGI mill shell. Fig. A1.b. SGI mill shell Section A.
P. Amelunxen et al. / Minerals Engineering 55 (2014) 4251 49
Appendix B. SGI mill drawings
The following construction drawings depict the SAG grindabili-
ty test mill constructed to perform the research described in this
paper. The drawings were reconstructed from the publicly avail-
able sources as described in the body of this manuscript.
Appendix C. SGI test procedures
Standard SGI test procedures:
1. Using a laboratory jaw crusher, stage crush the sample to
approximately 100% minus
3
=
4
in. and approximately 80% minus
in.
2. Screen the material on the following screens, and create a con-
trolled SGI feed size distribution using standard laboratory
splitting equipment and procedures:
Screen lm Mass (g)
3/4
00
19,050 0
1/2
00
12,700 400
3/8
00
9525 400
1/4
00
6350 400
#6 3350 400
#10 1700 150
pan 0 250
1. Place the 5 kg ball charge and 2 kg mass charge in the SGI mill
and grind it for 15, 30, 60, and 120 min, measuring the size
distribution after each cycle (with the same screens as shown
above).
2. Using least squares methods, t Eqs. 6, 7a, 7b, and 7c to the
curve of% retained on a Tyler #10 mesh screen versus time,
and using the resulting equation to calculate the time required
to reach 80% passing Tyler #10 mesh (note that only the Tyler
#10 mesh screen is required after each cycle; the entire stack
is included in these procedures just to avoid blinding and to col-
lect additional information in the event that it is required in the
future).
Appendix D. Supplementary material
Supplementary data associated with this article can be found, in
the online version, at http://dx.doi.org/10.1016/j.mineng.2013.
08.012.
References
Amelunxen, P., 2003. The Application of the SAG Power Index to ore Body Hardness
Characterization for the Design and Optimization of Autogenous Grinding
circuits. Department of Mining, Metals, and Materials Engineering, McGill
University, Montreal, Canada.
Amelunxen, P., Bennett, C., Garretson, P., Mertig, H., 2001. Use of geostatistics to
generate an ore body hardness data set and to quantify the relationship
between sample spacing and the precision of the throughput predictions. In:
Barratt, D., Allan, M., Mular, A. (Eds.), International Autogenous and
Semiautogenous Grinding Technology 2001 (SAG 2001), Vancouver, Canada
Amelunxen, P., Mular, M.A., Vanderbeek, J., Hill, L., Herrera, E., 2011. The effects of
ore variability on HPGR trade-off economics. In: Major, K., Flintoff, B.C., Klein,
B., McLeod, K (Eds.), International Autogenous and Semiautogenous Grinding
and High Pressure Grinding Roll Technology 2011 (SAG 2011). Vancouver,
Canada.
Aminpro Chile SpA, 2013. <http://www.aminpro.cl> (accessed 28.02.13).
Asmin Industrial, 2013. <http://www.asmin.cl> (accessed 27.02.13).
Austin, L.G., Concha, F., 1994. Diseo y Simulacin de Circuitos de Molienda y
Clasicacin. Santiago, Chile.
Baeza, R., Domingo, 2000. Caracterizacin del Consumo Especco de Energa para
Molienda SAG en Molino de Laboratorio Starkey (PowerPoint Presentation).
Codelco Division El Teniente, Chile.
Baeza, D., Villanueva, B., 1994. Mtodo Simplicado a Escala Laboratorio para la
Determinacin Del Consumo Especico De Energa SAG. Procemin, Santiago,
Chile.
Becerra, M., Amelunxen, P., 2012. A comparative analysis of grinding circuit design
methodologies. In: 9th International Mineral Process Conference (PROCEMIN),
Santiago, Chile.
Canadian Intellectual Property Ofce (CIPO), 2012. Trademark registration number
TMA567312, <http://www.cipo.ic.gc.ca> (accessed 06.01.12).
Dobby, G, Bennett, C., Kosick, G., 2001. Advances in SAG circuit design and
simulation applied to the mine block model. In: Barratt, D., Allan, M., Mular, A.
(Eds.), International Autogenous and Semiautogenous Grinding Technology
2001 (SAG 2001), Vancouver, Canada.
Gonzalez, E.A., 2000. Sensibilidad de Parmetros de Molienda SAG en Ensayo de
Laboratorio Starkey. Universidad de Santiago de Chile, Faculty of Engineering,
Department of Metallurgical Engineering, Santiago, Chile.
Kosick, G., Bennett, C., 1999. The value of ore body power requirement proles for
SAG circuit design. In: Proceedings of the 31st Annual Canadian Mineral
Processors Conference (CMP 1999), Ottawa, Canada.
Morrell, S., 1993. The Prediction of Power Draw in Wet Tumbling Mills, Ph. D Thesis,
Julius Kruttschnitt Mineral Research Center, Department of Mining and
Metallurgical Engineering, University of Queensland, Queensland, Australia,
pp. 150.
Napier-Munn, T.J., 1996. Mineral Comminution Circuits: Operation and
Optimisation. JKMRC, Brisbane, Australia.
Fig. A2.b. Detail lid rubber liner.
Fig. A2.a. Detail SGI mill.
50 P. Amelunxen et al. / Minerals Engineering 55 (2014) 4251
Ouchterlony, F., 2005. The Swebrec Function: Linking Fragmentation by Blasting
and Crushing, Mining Technology, Transactions of the Institute of Mining and
Metallurgy of Australasia, 114.
Starkey, J., Dobby, G., 1996. Application of the minnovex SAG power index at ve
Canadian SAG plants. In: Mular, A., Barratt, D., Knight, D. (Eds.), International
Autogenous and Semiautogenous Grinding Technology 1996 (SAG 1996),
Vancouver, Canada.
Starkey, J, Dobby, G., Kosick, G., 1994. A new tool for hardness testing. In:
Proceedings of the 26th Annual Meeting of the Canadian Mineral Processors
(CMP 1994), Ottawa, Canada.
Starkey, J, Hindstrom, S., Nadasdy, G., 2006. SAGDesign testing what it is and why
it works. In: Allan, M., Major, K., Flintoff, B.C., Klein, B., Mular, A. (Eds.),
International Autogenous and Semiautogenous Grinding Technology 2006 (SAG
2006), Vancouver, Canada.
Starkey, J., 2013. Private communication.
Verret, F.O., Chiasson, G., McKen, A. 2011. SAG mill testing an overview of the test
procedures available to characterize ore grindability. In: Major, K., Flintoff, B.C.,
Klein, B., McLeod, K. (Eds.), International Autogenous and Semiautogenous
Grinding and High Pressure Grinding Roll Technology 2011 (SAG 2011),
Vancouver, Canada.
P. Amelunxen et al. / Minerals Engineering 55 (2014) 4251 51

You might also like