XXXI International Mineral Processing Congress 2024 Proceedings/Washington, DC/Sep 29–Oct 3 2289
APPLICATIONS
Kaolinite
Kaolinite is a common clay mineral formed from altered
rocks and a source of alumina impurities (Johnston et al.
2022). Kaolinite and its composing oxides (Al2O3 and
SiO2) are also known for being deleterious to metallur-
gical processes, reducing both efficiency and productiv-
ity. Hence, the impurities contained in minerals such as
kaolinite together with other silicates have to be removed
from mineral ores (Rodrigues 2013). As mentioned in
the previous section, the modeling of the kaolinite/water
interface by AIMD calculations is inherently restricted to
relatively small system sizes. In contrast, the trained ALFF
model may extend to nanoseconds the simulation time and
increase the size of the system up to a few thousand atoms.
Figure 5 offers a straightforward example to enlighten
this difference. It shows for the kaolinite/water model the
evolution of the enthalpy at high water coverages. As we
can see, AIMD was able to investigate the system until 2
mmol/g in terms of adsorbed amount. Although CMD was
able to extend the coverage, the simulation turns out to be
less accurate. MLFF model (onset of Figure 5) extends the
latter up to 100 mmol/g maintaining first principle accu-
racy, allowing to see well the plateau, thanks to its ability
to include much more water molecules in the model. In
particular, the MLFF model reaches the plateau at approxi-
matively –39 kJ·mol–1, nearby the enthalpy of water
condensation (–40.7 kJ·mol–1). At low equilibrium pres-
sures, the enthalpy of adsorption calculated from the MLFF
model approaches –55 kJ·mol–1, which can be compared
to the adsorption/desorption energies commonly observed
on metals and minerals surfaces: around –90 kJ·mol–1
on scheelite (Foucaud et al. 2021), –89 kJ·mol–1 on cal-
cite (Rahaman et al. 2008), and –55 kJ·mol–1 on fluorite
(Foucaud et al. 2018).
Quartz
A crucial point for a better understanding of the flotation
mechanism deals with how the surface may affect the layer-
ing of the surfactant water molecules. Several contributions
may perturb the structure of water molecules at the interfa-
cial region, such as the cleavage direction and its chemical
nature, or the pH value of the solution and the consequent
number of silanol groups on surface. By means of vibra-
tional sum frequency generation (VSFG) experiments, Du
et al. (Du et al. 1994) have shown the presence of two addi-
tional peaks at the silica/water interface with respect to liq-
uid water. These peaks have been interpreted as liquid-like
(3400 cm–1) and ice-like (3200 cm–1), since those bands
correspond to those in the vibrational spectra of pure liq-
uid water and ice, respectively. Gaigeot et al. (Gaigeot et al.
2012) investigated the correlation between pH and silanol
groups on surface. At low pH, the quartz is mostly cov-
ered by silanol groups. By contrast, at high pH the silanol
Figure 4. HDNNP scheme. (a) General NN architecture consisting of layers and nodes (i.e., the activation function). The latter
are connected by weights which need to be parametrized over the feed/back forward training. (b) Expression for the active
function and depiction of the design matrix which represents the computational demanding bottleneck over the training
APPLICATIONS
Kaolinite
Kaolinite is a common clay mineral formed from altered
rocks and a source of alumina impurities (Johnston et al.
2022). Kaolinite and its composing oxides (Al2O3 and
SiO2) are also known for being deleterious to metallur-
gical processes, reducing both efficiency and productiv-
ity. Hence, the impurities contained in minerals such as
kaolinite together with other silicates have to be removed
from mineral ores (Rodrigues 2013). As mentioned in
the previous section, the modeling of the kaolinite/water
interface by AIMD calculations is inherently restricted to
relatively small system sizes. In contrast, the trained ALFF
model may extend to nanoseconds the simulation time and
increase the size of the system up to a few thousand atoms.
Figure 5 offers a straightforward example to enlighten
this difference. It shows for the kaolinite/water model the
evolution of the enthalpy at high water coverages. As we
can see, AIMD was able to investigate the system until 2
mmol/g in terms of adsorbed amount. Although CMD was
able to extend the coverage, the simulation turns out to be
less accurate. MLFF model (onset of Figure 5) extends the
latter up to 100 mmol/g maintaining first principle accu-
racy, allowing to see well the plateau, thanks to its ability
to include much more water molecules in the model. In
particular, the MLFF model reaches the plateau at approxi-
matively –39 kJ·mol–1, nearby the enthalpy of water
condensation (–40.7 kJ·mol–1). At low equilibrium pres-
sures, the enthalpy of adsorption calculated from the MLFF
model approaches –55 kJ·mol–1, which can be compared
to the adsorption/desorption energies commonly observed
on metals and minerals surfaces: around –90 kJ·mol–1
on scheelite (Foucaud et al. 2021), –89 kJ·mol–1 on cal-
cite (Rahaman et al. 2008), and –55 kJ·mol–1 on fluorite
(Foucaud et al. 2018).
Quartz
A crucial point for a better understanding of the flotation
mechanism deals with how the surface may affect the layer-
ing of the surfactant water molecules. Several contributions
may perturb the structure of water molecules at the interfa-
cial region, such as the cleavage direction and its chemical
nature, or the pH value of the solution and the consequent
number of silanol groups on surface. By means of vibra-
tional sum frequency generation (VSFG) experiments, Du
et al. (Du et al. 1994) have shown the presence of two addi-
tional peaks at the silica/water interface with respect to liq-
uid water. These peaks have been interpreted as liquid-like
(3400 cm–1) and ice-like (3200 cm–1), since those bands
correspond to those in the vibrational spectra of pure liq-
uid water and ice, respectively. Gaigeot et al. (Gaigeot et al.
2012) investigated the correlation between pH and silanol
groups on surface. At low pH, the quartz is mostly cov-
ered by silanol groups. By contrast, at high pH the silanol
Figure 4. HDNNP scheme. (a) General NN architecture consisting of layers and nodes (i.e., the activation function). The latter
are connected by weights which need to be parametrized over the feed/back forward training. (b) Expression for the active
function and depiction of the design matrix which represents the computational demanding bottleneck over the training