3160 XXXI International Mineral Processing Congress 2024 Proceedings/Washington, DC/Sep 29–Oct 3
over the flotation process and to replicate conditions that
could be scaled up for industrial applications. For the
flotation tests, two reagents were used with fixed dosage.
EscaidTM 110 from ExxonMobil (Hydrocarbon fluid,
Product No. 20171206) was used at a dosage of 350 g/t as
promoter (collector), to enhance the natural hydrophobic-
ity of graphite surfaces. Methyl isobutyl carbinol (MIBC)
supplied by Alfa Aesar (99% C6H14O, Product No.
A13435) was used as a frother with a dosage of 150 g/t.
Testing workflow consisted of mixing the pulp for 120 s for
stabilising the pulp chemical conditions, collector condi-
tioning for 180 s, frother conditioning for 120 s, and col-
lection of four concentrates over 0–30 s, 30–60 s, 60–120
s, and 120–540 s time intervals. Rotor speed and air flow
rate were kept constant for both tests, respectively at 1000
rpm and 5 L/min. Samples were pretreated with attrition
before froth flotation experiments at 40% a solid content
water dispersion with a Ultra Turrax high shear mixer
(IKA, dispersing unit S25N-25F, Königswinter, Germany)
operating at 16000 rpm for a duration of 10 min.
After each flotation experiment, the mass of the four
overflow products was measured before and after drying
to determine the mass and water pulls. Drying was per-
formed in an oven under natural convection for 12 h at
45°C. Finally, representative dry samples were prepared for
characterization. Flotation recoveries were calculated with
Eq. (1), where Ri means recovery of component i, C the
overflow product mass, ci the grade of phase i in the over-
flow product, F the feed mass and fi the grade of phase i in
the feed.
R R f
C c
i
i
i #
#
=(1)
The enrichment factor of phase i in product p =(EFi p )were
calculated with Eq. (2), where ci p and ci f are the grades of
phase i in the product and in the feed, respectively.
EF
c
c
1
i
p
i
f
i
p
=-(2)
Scanning Electron Microscopy (SEM)-Based
Automated Image Analysis
SEM-based automated image analysis, exemplified by
techniques such as Mineral Liberation Analysis (MLA)
streamlines mineralogical studies by automating sample
characterization. MLA employs SEM and EDX (Energy-
Dispersive X-ray Spectroscopy) technologies to characterize
particle microstructures. For MLA measurements, sample
aliquots (1.8g) were obtained using a rotary splitter, manu-
ally deagglomerated, and embedded in epoxy resin to create
grain mounts (25 mm diameter). The mounts underwent
slicing, a 90-degree rotation, and re-embedding into
B-sections to minimize particle settling bias (Heinig et al.,
2015). Analysis was carried out using a FEI Quanta 650F
SEM equipped with two EDS detectors and MLA-Suite
software for automated data acquisition at the Helmholtz
Institute for Resource Technology (HIF). Grain-based
X-ray mapping (GXMAP) measurements were carried out,
using operational conditions of 15kV acceleration voltage,
10 nA probe current, and a resolution of 0.5 µm/px com-
bined with a X-ray stepsize of 10 px. These comprehensive
measurements, averaging 140,000 particles per sample,
provided crucial insights into phase composition and facili-
tated the assessment of recovery processes (Bachmann et al.,
2017), particularly in the context of LIBs (Vanderbruggen
et al., 2021a). Furthermore, offer additional potential for
utilization in particle-based separation models (Pereira et
al., 2021b).
Particle-Based Separation Models
The PSM method of Pereira et al. (2021b) consists of a
least absolute shrinkage and selection operator (LASSO)-
regularized logistic regression. Particle descriptors are the
data quantified with automated mineralogy—size (equiva-
lent circle diameter, ECD), shape (aspect ratio and solid-
ity), modal composition, and surface composition—for
individual particles. Data transformation consists of:
• Log transformation of size and shape variables,
• The square of the log transformed ECD is added as
a new variable,
• Modal and surface composition vectors are closed
to sum to 100% (van den Boogaart and Tolosana-
Delgado, 2013) individually,
• A categorical variable indicating the main phase
composing the particle (in mass) is added to the
dataset—namely mainphase.
Particle data, compiled in a training dataset that contains
the same number of particles for each process stream, is
used to train the PSM. From all particle data, 70% is used
for model training and the remaining 30% for validation.
The response variable of the multinomial logistic regression
is the particle class (e.g., concentrate 1, tailings, etc.). An
ancova framework based on an interaction between size and
shape variables and mainphase is used in the model to cap-
ture a singular relation between particle geometric descrip-
tors and recovery for each phase. After training, the model
intercepts are adjusted to reflect the actual particle balance
(analogous to mass balance) of the process. An individual
model is trained for each of the process conditions tested
(low and high percentage of solids).
over the flotation process and to replicate conditions that
could be scaled up for industrial applications. For the
flotation tests, two reagents were used with fixed dosage.
EscaidTM 110 from ExxonMobil (Hydrocarbon fluid,
Product No. 20171206) was used at a dosage of 350 g/t as
promoter (collector), to enhance the natural hydrophobic-
ity of graphite surfaces. Methyl isobutyl carbinol (MIBC)
supplied by Alfa Aesar (99% C6H14O, Product No.
A13435) was used as a frother with a dosage of 150 g/t.
Testing workflow consisted of mixing the pulp for 120 s for
stabilising the pulp chemical conditions, collector condi-
tioning for 180 s, frother conditioning for 120 s, and col-
lection of four concentrates over 0–30 s, 30–60 s, 60–120
s, and 120–540 s time intervals. Rotor speed and air flow
rate were kept constant for both tests, respectively at 1000
rpm and 5 L/min. Samples were pretreated with attrition
before froth flotation experiments at 40% a solid content
water dispersion with a Ultra Turrax high shear mixer
(IKA, dispersing unit S25N-25F, Königswinter, Germany)
operating at 16000 rpm for a duration of 10 min.
After each flotation experiment, the mass of the four
overflow products was measured before and after drying
to determine the mass and water pulls. Drying was per-
formed in an oven under natural convection for 12 h at
45°C. Finally, representative dry samples were prepared for
characterization. Flotation recoveries were calculated with
Eq. (1), where Ri means recovery of component i, C the
overflow product mass, ci the grade of phase i in the over-
flow product, F the feed mass and fi the grade of phase i in
the feed.
R R f
C c
i
i
i #
#
=(1)
The enrichment factor of phase i in product p =(EFi p )were
calculated with Eq. (2), where ci p and ci f are the grades of
phase i in the product and in the feed, respectively.
EF
c
c
1
i
p
i
f
i
p
=-(2)
Scanning Electron Microscopy (SEM)-Based
Automated Image Analysis
SEM-based automated image analysis, exemplified by
techniques such as Mineral Liberation Analysis (MLA)
streamlines mineralogical studies by automating sample
characterization. MLA employs SEM and EDX (Energy-
Dispersive X-ray Spectroscopy) technologies to characterize
particle microstructures. For MLA measurements, sample
aliquots (1.8g) were obtained using a rotary splitter, manu-
ally deagglomerated, and embedded in epoxy resin to create
grain mounts (25 mm diameter). The mounts underwent
slicing, a 90-degree rotation, and re-embedding into
B-sections to minimize particle settling bias (Heinig et al.,
2015). Analysis was carried out using a FEI Quanta 650F
SEM equipped with two EDS detectors and MLA-Suite
software for automated data acquisition at the Helmholtz
Institute for Resource Technology (HIF). Grain-based
X-ray mapping (GXMAP) measurements were carried out,
using operational conditions of 15kV acceleration voltage,
10 nA probe current, and a resolution of 0.5 µm/px com-
bined with a X-ray stepsize of 10 px. These comprehensive
measurements, averaging 140,000 particles per sample,
provided crucial insights into phase composition and facili-
tated the assessment of recovery processes (Bachmann et al.,
2017), particularly in the context of LIBs (Vanderbruggen
et al., 2021a). Furthermore, offer additional potential for
utilization in particle-based separation models (Pereira et
al., 2021b).
Particle-Based Separation Models
The PSM method of Pereira et al. (2021b) consists of a
least absolute shrinkage and selection operator (LASSO)-
regularized logistic regression. Particle descriptors are the
data quantified with automated mineralogy—size (equiva-
lent circle diameter, ECD), shape (aspect ratio and solid-
ity), modal composition, and surface composition—for
individual particles. Data transformation consists of:
• Log transformation of size and shape variables,
• The square of the log transformed ECD is added as
a new variable,
• Modal and surface composition vectors are closed
to sum to 100% (van den Boogaart and Tolosana-
Delgado, 2013) individually,
• A categorical variable indicating the main phase
composing the particle (in mass) is added to the
dataset—namely mainphase.
Particle data, compiled in a training dataset that contains
the same number of particles for each process stream, is
used to train the PSM. From all particle data, 70% is used
for model training and the remaining 30% for validation.
The response variable of the multinomial logistic regression
is the particle class (e.g., concentrate 1, tailings, etc.). An
ancova framework based on an interaction between size and
shape variables and mainphase is used in the model to cap-
ture a singular relation between particle geometric descrip-
tors and recovery for each phase. After training, the model
intercepts are adjusted to reflect the actual particle balance
(analogous to mass balance) of the process. An individual
model is trained for each of the process conditions tested
(low and high percentage of solids).