XXXI International Mineral Processing Congress 2024 Proceedings/Washington, DC/Sep 29–Oct 3 1047
All XRF data collected is used, having Euclidean distance
as the measure of sample similarity. The dendrogram indi-
cating the relation between samples is shown in Figure 2A.
The TSF is split into 4 families (namely clusters). As dis-
played in Figure 2B, family 1 corresponds to the very north
of the TSF, families 2 and 3 to the center and partially to
the south of the body, while family 4 is mostly present on
the east and west borders of the area.
Composite samples of each family are put together by
mixing 500 g of six randomly selected samples within the
family. After homogenization, each composite is split back
into 500 g aliquots. Two of these aliquots are used for the
flotation tests of the family while the remaining four ali-
quots of each family are combined into the global compos-
ite. The 8 kg global composite sample is split into 500 g
aliquots for flotation tests.
Flotation Test Work and Product Characterization
Flotation is used here to produce a sulfide-rich concen-
trate and sulfide-poor tailings, but process optimization is
beyond the scope of this study. A Magotteaux ® bench scale
flotation device with a 2L cell is used for the experiments.
The flotation tests for all families and the global composite
are conveyed with the same parameters:
• 500 g of feed material,
• 450 rpm impeller speed,
• 5 l/min air flow rate,
• 50 g/t of technical grade Potassium Amyl Xanthate
as collector,
• 20 g/t of technical grade Methyl Isobutyl Carbinol
as frother,
• and five concentrates respectively collected over
0–30 s, 30–90 s, 90–210 s, 210–390 s, and 390–630
s time intervals.
All flotation products are dried overnight at 60 ºC and split
into 1–2 g samples with a small rotary splitter, to be pre-
pared into grain mounts for MLA analysis. Grain mounts
are prepared as B-sections to avoid issues with sedimen-
tation, following the routine proposed by Heinig et al.
(2015). After carbon coating, MLA measurements are con-
ducted at HIF in a FEI Quanta 650F scanning electron
microscope equipped with two Bruker Quantax X-Flash
5030 energy dispersive X-ray spectrometry detectors and
MLA Suite 3.1.4 for automated data acquisition. Grain-
based X-ray mapping measurements are conducted with 1
µm pixel size, 6 µm X-ray spacing, 25 kV operating voltage,
and a minimum particle area of 2 pixels.
Particle-Based Models
A particle-based model (Pereira et al., 2021) is trained with
the processing products collected from the flotation test
of the global composite to be used as a general recovery
model for the TSF. This strategy follows the assumption
that PSMs can be accurate as long as it is presented with the
different types of particles that can be found in a deposit.
It is important to bear in mind that PSMs assume that
recovery of a particle only depends on its microstructural
properties—i.e., particle-particle interactions are not con-
sidered. While this limitation might hinder the accuracy
of the PSM in this study, Pereira et al. (2022) have dem-
onstrated that PSMs can be accurate despite variations in
feed ore composition in the case of magnetic separation.
Besides, this strategy is sufficient to reach the objective of
Figure 2. Hierarchical clustering results as a dendrogram (A) and represented in the TSF map (B)
All XRF data collected is used, having Euclidean distance
as the measure of sample similarity. The dendrogram indi-
cating the relation between samples is shown in Figure 2A.
The TSF is split into 4 families (namely clusters). As dis-
played in Figure 2B, family 1 corresponds to the very north
of the TSF, families 2 and 3 to the center and partially to
the south of the body, while family 4 is mostly present on
the east and west borders of the area.
Composite samples of each family are put together by
mixing 500 g of six randomly selected samples within the
family. After homogenization, each composite is split back
into 500 g aliquots. Two of these aliquots are used for the
flotation tests of the family while the remaining four ali-
quots of each family are combined into the global compos-
ite. The 8 kg global composite sample is split into 500 g
aliquots for flotation tests.
Flotation Test Work and Product Characterization
Flotation is used here to produce a sulfide-rich concen-
trate and sulfide-poor tailings, but process optimization is
beyond the scope of this study. A Magotteaux ® bench scale
flotation device with a 2L cell is used for the experiments.
The flotation tests for all families and the global composite
are conveyed with the same parameters:
• 500 g of feed material,
• 450 rpm impeller speed,
• 5 l/min air flow rate,
• 50 g/t of technical grade Potassium Amyl Xanthate
as collector,
• 20 g/t of technical grade Methyl Isobutyl Carbinol
as frother,
• and five concentrates respectively collected over
0–30 s, 30–90 s, 90–210 s, 210–390 s, and 390–630
s time intervals.
All flotation products are dried overnight at 60 ºC and split
into 1–2 g samples with a small rotary splitter, to be pre-
pared into grain mounts for MLA analysis. Grain mounts
are prepared as B-sections to avoid issues with sedimen-
tation, following the routine proposed by Heinig et al.
(2015). After carbon coating, MLA measurements are con-
ducted at HIF in a FEI Quanta 650F scanning electron
microscope equipped with two Bruker Quantax X-Flash
5030 energy dispersive X-ray spectrometry detectors and
MLA Suite 3.1.4 for automated data acquisition. Grain-
based X-ray mapping measurements are conducted with 1
µm pixel size, 6 µm X-ray spacing, 25 kV operating voltage,
and a minimum particle area of 2 pixels.
Particle-Based Models
A particle-based model (Pereira et al., 2021) is trained with
the processing products collected from the flotation test
of the global composite to be used as a general recovery
model for the TSF. This strategy follows the assumption
that PSMs can be accurate as long as it is presented with the
different types of particles that can be found in a deposit.
It is important to bear in mind that PSMs assume that
recovery of a particle only depends on its microstructural
properties—i.e., particle-particle interactions are not con-
sidered. While this limitation might hinder the accuracy
of the PSM in this study, Pereira et al. (2022) have dem-
onstrated that PSMs can be accurate despite variations in
feed ore composition in the case of magnetic separation.
Besides, this strategy is sufficient to reach the objective of
Figure 2. Hierarchical clustering results as a dendrogram (A) and represented in the TSF map (B)