1456 XXXI International Mineral Processing Congress 2024 Proceedings/Washington, DC/Sep 29–Oct 3
Pawlowsky-Glahn 2005a,b). A balance dendrogram is a tool
to represent simultaneously the relative variation between
and within groups of parts (i.e., ratios between groups of
minerals), the sample center (i.e., geometric mean), the
total variance decomposed between group of elements (i.e.,
balances), and some statistical characteristics of the bal-
ances. Each balance reflects the proportion of the total vari-
ance accounted for by the associated groups of minerals.
The longer the distance from the center of the composition,
the greater the proportion of the total variance represented
by the respective minerals.
Discriminant function analysis (DFA) of major ele-
ment data is a well-established method that has been
applied for a long time to the classification of various types
of geological samples (e.g., Klovan and Billings, 1967,
Roser, B.P., Korsch, R.J., 1988, etc). In this study, DFA
was used to classify skarn types and sulphides mineral asso-
ciations using whole rock data as a proxy for the miner-
alogical data obtain using TIMA-X. This analysis seeks to
derive a set of linear functions, based on multiple variables
(e.g., major elements in wt%) that achieve the best separa-
tion between predefined skarn types and sulphides bearing
assembles derived from TIMA-X analysis. Two linear func-
tions derived for the combined mineralogy and geochemi-
cal data described above (e.g., training dataset), have been
applied to a wider drillcore dataset (testing set) in order to
predict broadly their skarn mineralogy and dominant sul-
phides mineral assemblages. DFA has been implemented as
part of ioGAS ™ and so we used it due to its simplicity, and
because it has been demonstrated that the accuracy of linear
discriminant analysis is comparable to more sophisticated
machine learning algorithms such as random forest, and
exceed others such as classification and regression trees or
support vector machines, which have been used earlier to
predict skarn classes (Ordóñez-Calderón et al., 2017).
Principal component analysis, balances and discrimi-
nant function analysis were carried using a combination
of CoDaPack (an open source software developed by the
Research Group in Statistics and Compositional Data
Analysis at University of Girona), Microsoft ® Excel, and
ioGAS ™ (an industry standard software for interrogating
multielement geochemistry data).
Comminution
Samples were characterized by JK drop-weight (DWT),
SMC Test ®, SAG power index (SPI ®) /SAG grindability
index (SGI), BWi and Bond abrasion index (Ai).
Flotation
Flotation testing was conducted following the method-
ologies outlined in Amelunxen et al. 2018. The flotation
conditions were guided by geological and mineralogical
information (grain size, Cu species, etc.) and tailored to
optimize Cu recovery. Variables tested included the primary
grind size, collector concentration, collector types, pH,
pulp density and the effect of sulfide specific ion electrode
(SIE) potential [controlled potential sulfidization (CPS)].
RESULTS AND DISCUSSION
Geochemical and Mineralogical Results
An example of the results of geochemical and TIMA-X
analyses is presented in Table 1 and Table 2, respectively.
The analysis indicates several parameters used to define
the mineralization and their host rocks. First, it reveals the
presence of Cu sulphides including mainly chalcopyrite,
bornite and chalcocite, Cu-oxides, chrysocolla, malachite,
pitch copper wad (manganese oxides/hydroxides), and Cu
silicates such as chlorite. The presence of quartz, feldspars
and muscovite/clays indicates the presence of sedimentary
rocks and porphyry type granitic rocks, and the pyroxenes,
garnets, epidote, amphibole, and wollastonite the presence
of various skarn hosts.
The Cu concentration is strochiometric in the sul-
phides. However, copper varies in the copper secondary
minerals. For example, Cu in the goethite can vary from
0.5% to 5%, Cu in chlorite from 0.5% to 10%. The
data have been used to calculate the distribution of Cu
among the minerals and samples (Table 3). They reveal
that the Cu sulphides host most of the copper. However,
in certain samples secondary minerals such as chrysocolla,
malachite, Cu-goethite and pitch copper wad can account
for significant Cu.
The liberation, association and exposure of the indi-
vidual Cu minerals or combined (e.g., Cu sulphides includ-
ing chalcopyrite, bornite, chalcocite) are also calculated.
Particles are classified in groups based on mineral-of-inter-
est area percent: pure (100% of the total particle area by
volume), free (≥95%), and liberated (≥80%). The non-
liberated grains have been classified according to associa-
tion characteristics, where binary association groups refer
to particle area percent greater than or equal to 95% of
the two minerals or mineral groups. The complex groups
refer to particles with ternary, quaternary, and greater min-
eral associations including the mineral of interest. Table 4
shows a summary of the liberation and association of the
Pawlowsky-Glahn 2005a,b). A balance dendrogram is a tool
to represent simultaneously the relative variation between
and within groups of parts (i.e., ratios between groups of
minerals), the sample center (i.e., geometric mean), the
total variance decomposed between group of elements (i.e.,
balances), and some statistical characteristics of the bal-
ances. Each balance reflects the proportion of the total vari-
ance accounted for by the associated groups of minerals.
The longer the distance from the center of the composition,
the greater the proportion of the total variance represented
by the respective minerals.
Discriminant function analysis (DFA) of major ele-
ment data is a well-established method that has been
applied for a long time to the classification of various types
of geological samples (e.g., Klovan and Billings, 1967,
Roser, B.P., Korsch, R.J., 1988, etc). In this study, DFA
was used to classify skarn types and sulphides mineral asso-
ciations using whole rock data as a proxy for the miner-
alogical data obtain using TIMA-X. This analysis seeks to
derive a set of linear functions, based on multiple variables
(e.g., major elements in wt%) that achieve the best separa-
tion between predefined skarn types and sulphides bearing
assembles derived from TIMA-X analysis. Two linear func-
tions derived for the combined mineralogy and geochemi-
cal data described above (e.g., training dataset), have been
applied to a wider drillcore dataset (testing set) in order to
predict broadly their skarn mineralogy and dominant sul-
phides mineral assemblages. DFA has been implemented as
part of ioGAS ™ and so we used it due to its simplicity, and
because it has been demonstrated that the accuracy of linear
discriminant analysis is comparable to more sophisticated
machine learning algorithms such as random forest, and
exceed others such as classification and regression trees or
support vector machines, which have been used earlier to
predict skarn classes (Ordóñez-Calderón et al., 2017).
Principal component analysis, balances and discrimi-
nant function analysis were carried using a combination
of CoDaPack (an open source software developed by the
Research Group in Statistics and Compositional Data
Analysis at University of Girona), Microsoft ® Excel, and
ioGAS ™ (an industry standard software for interrogating
multielement geochemistry data).
Comminution
Samples were characterized by JK drop-weight (DWT),
SMC Test ®, SAG power index (SPI ®) /SAG grindability
index (SGI), BWi and Bond abrasion index (Ai).
Flotation
Flotation testing was conducted following the method-
ologies outlined in Amelunxen et al. 2018. The flotation
conditions were guided by geological and mineralogical
information (grain size, Cu species, etc.) and tailored to
optimize Cu recovery. Variables tested included the primary
grind size, collector concentration, collector types, pH,
pulp density and the effect of sulfide specific ion electrode
(SIE) potential [controlled potential sulfidization (CPS)].
RESULTS AND DISCUSSION
Geochemical and Mineralogical Results
An example of the results of geochemical and TIMA-X
analyses is presented in Table 1 and Table 2, respectively.
The analysis indicates several parameters used to define
the mineralization and their host rocks. First, it reveals the
presence of Cu sulphides including mainly chalcopyrite,
bornite and chalcocite, Cu-oxides, chrysocolla, malachite,
pitch copper wad (manganese oxides/hydroxides), and Cu
silicates such as chlorite. The presence of quartz, feldspars
and muscovite/clays indicates the presence of sedimentary
rocks and porphyry type granitic rocks, and the pyroxenes,
garnets, epidote, amphibole, and wollastonite the presence
of various skarn hosts.
The Cu concentration is strochiometric in the sul-
phides. However, copper varies in the copper secondary
minerals. For example, Cu in the goethite can vary from
0.5% to 5%, Cu in chlorite from 0.5% to 10%. The
data have been used to calculate the distribution of Cu
among the minerals and samples (Table 3). They reveal
that the Cu sulphides host most of the copper. However,
in certain samples secondary minerals such as chrysocolla,
malachite, Cu-goethite and pitch copper wad can account
for significant Cu.
The liberation, association and exposure of the indi-
vidual Cu minerals or combined (e.g., Cu sulphides includ-
ing chalcopyrite, bornite, chalcocite) are also calculated.
Particles are classified in groups based on mineral-of-inter-
est area percent: pure (100% of the total particle area by
volume), free (≥95%), and liberated (≥80%). The non-
liberated grains have been classified according to associa-
tion characteristics, where binary association groups refer
to particle area percent greater than or equal to 95% of
the two minerals or mineral groups. The complex groups
refer to particles with ternary, quaternary, and greater min-
eral associations including the mineral of interest. Table 4
shows a summary of the liberation and association of the