XXXI International Mineral Processing Congress 2024 Proceedings/Washington, DC/Sep 29–Oct 3 41
deposit, although additional samples from proximal depos-
its were also included. The mineral grades of the sample
set were determined with QEMSCAN, with 10% of the
sample grades validated with quantitative X-ray diffraction
(QXRD). Accompanying 35-element chemical analyses
were also used to validate the QEMSCAN data. EPMA
analyses were used to accurately determine the composition
of sphalerite, from which the Mn distribution in the sample
could be calculated.
This complete dataset was then submitted for data
analysis and machine learning with a project partner. Once
the representativity of the sample set was established by
comparison with the Gamsberg drill hole database, and
exploratory data analysis was performed to identify impor-
tant features for predictive modelling, regression model-
ling was carried out. These models were used to generate
algorithms based on the chemistry and a single categori-
cal variable to predict key mineral grades (e.g., sphalerite,
pyrrhotite, pyrite), as well as sphalerite composition (Price
et al., 2023). Once these models were developed at the
required accuracy, they were deployed into the entire drill
hole database generating a geometallurgical block model,
with Figure 5 illustrating the range of Zn content in sphal-
erite (low Zn in sphalerite indicates impure sphalerite with
higher Fe+Mn). Such a model offers numerous opportuni-
ties for mine planning including improvements in ROM
ore blending to manage concentrate quality, stockpiling
and mine scheduling. In such a way, a strategy was imple-
mented to overcome the Mn problem, of which process
mineralogy was a key component in understanding the Mn
distribution. Using geometallurgical techniques, that are
seeing rapid and increasing application of machine learning
MPO PEO_PO PEO_PY
0
20
40
60
80
100
Sphalerite
Pyrrhotite
Pyrite
Galena
Othersulphides
Pyrolusite
Magnetite
Quartz
Feldspar
Mica+Chlorite
Amphibole
Pyroxenoid
Pyroxene
Garnet
Others
a). b).
MPO PEO_PO PEO_PY
0
20
40
60
80
100
Sphalerite
Pyrrhotite
Alabandite
Pyrolusite
Amphibole
Pyroxenoid
Pyroxene
Garnet
Figure 4. An example of (a) the mineral grades and (b) Mn deportment of the three main mineralised horizons at Gamsberg
contributing to the ROM. Mineral grades were determined through QEMSCAN analysis and Mn distribution through a
combination of QEMSCAN and EPMA analysis. Data from Molifie and Becker (2022)
Mineralgrade(wt.%) Mndistribution(wt.%)
deposit, although additional samples from proximal depos-
its were also included. The mineral grades of the sample
set were determined with QEMSCAN, with 10% of the
sample grades validated with quantitative X-ray diffraction
(QXRD). Accompanying 35-element chemical analyses
were also used to validate the QEMSCAN data. EPMA
analyses were used to accurately determine the composition
of sphalerite, from which the Mn distribution in the sample
could be calculated.
This complete dataset was then submitted for data
analysis and machine learning with a project partner. Once
the representativity of the sample set was established by
comparison with the Gamsberg drill hole database, and
exploratory data analysis was performed to identify impor-
tant features for predictive modelling, regression model-
ling was carried out. These models were used to generate
algorithms based on the chemistry and a single categori-
cal variable to predict key mineral grades (e.g., sphalerite,
pyrrhotite, pyrite), as well as sphalerite composition (Price
et al., 2023). Once these models were developed at the
required accuracy, they were deployed into the entire drill
hole database generating a geometallurgical block model,
with Figure 5 illustrating the range of Zn content in sphal-
erite (low Zn in sphalerite indicates impure sphalerite with
higher Fe+Mn). Such a model offers numerous opportuni-
ties for mine planning including improvements in ROM
ore blending to manage concentrate quality, stockpiling
and mine scheduling. In such a way, a strategy was imple-
mented to overcome the Mn problem, of which process
mineralogy was a key component in understanding the Mn
distribution. Using geometallurgical techniques, that are
seeing rapid and increasing application of machine learning
MPO PEO_PO PEO_PY
0
20
40
60
80
100
Sphalerite
Pyrrhotite
Pyrite
Galena
Othersulphides
Pyrolusite
Magnetite
Quartz
Feldspar
Mica+Chlorite
Amphibole
Pyroxenoid
Pyroxene
Garnet
Others
a). b).
MPO PEO_PO PEO_PY
0
20
40
60
80
100
Sphalerite
Pyrrhotite
Alabandite
Pyrolusite
Amphibole
Pyroxenoid
Pyroxene
Garnet
Figure 4. An example of (a) the mineral grades and (b) Mn deportment of the three main mineralised horizons at Gamsberg
contributing to the ROM. Mineral grades were determined through QEMSCAN analysis and Mn distribution through a
combination of QEMSCAN and EPMA analysis. Data from Molifie and Becker (2022)
Mineralgrade(wt.%) Mndistribution(wt.%)