3226 XXXI International Mineral Processing Congress 2024 Proceedings/Washington, DC/Sep 29–Oct 3
recover only 11.2% of NMC input mass on average. Still,
the long tails of the outcome distributions indicated that,
for some process parameters (See table 3), NMC recovery
and grade could be quite high. The remaining question was
whether it possible to identify the process that maximizes
mass recovery for both NMC and graphite, and what com-
bination of process parameters would produce this.
A data-driven approach to process analysis facilitates
multi-objective optimization, allowing us to find numerical
solutions that simultaneously satisfy several objectives. The
focus was on the fractional mass recovery data for NMC
and graphite. Figure 5a) illustrates the selection of simu-
lated processes where recovery for both NMC and graphite
is over 70%. These 130 data points constitute only 1.3%
of all process variations. The data distribution illustrates
that large mass recovery is easier for graphite than NMC.
This is in line with observations in Figure 4, where graphite
recovery tends to favour large values, but large NMC recov-
ery features only in the tail of the distribution. It was also
observed that no parameter combination produces recover-
ies over 88% for graphite or NMC. Considering the vast
number of simulations performed, it can be confidently
said that this is a limitation of the process design, which
could be addressed by modifying processing stages, fixed
parameters, or ranges of variable parameters.
Figure 5b) illustrates the 13 best process outcomes,
which achieve mass recovery over 85% for both processes.
Here, we selected 3 optimal process outcomes for detailed
analysis. Process O1 balances best recovery for both mate-
rials, while processes O2 and O3 slightly favor recovery
of NMC and graphite respectively. Table 3 presents the
process parameters behind the optimal results. For each of
the parameters, we compared the optimal values against
the variation range. Some shared parameter settings were
observed: high values of the Fixed SGE (Pre-milling) and
very low Magnetic Field Strength clearly lead to optimal
recovery. But elsewhere, parameter values could be very
different, indicating that not all parameters affect the out-
comes to the same extent. Further feature analysis work
with data science models would clarify this aspect of the
simulations.
Table 2. Summary of NMC and graphite mass flow and grade distributions for 10,000 process outcomes: range (minimum -
maximum) and average result. Mass recovery is computed as percentage fraction of the input feed.
Parameter Range Mean Recovery range, %Recovery mean, %Feed
Graphite in
Concentrate 2
Mass, t/h 0.0–23.3 20.9 0.0–86.3 77.4 27.0
Grade, %95.0–99.8 99.4 — — 18.0
NMC in
Tailings 2
Mass, t/h 0.0–31.7 4.0 0.0–89.6 11.2 35.6
Grade, %2.5–62.5 8.8 — — 23.7
Figure 5. Process data with optimal mass recovery for NMC in tailings 2 and graphite in concentrate 2: a) data where at least
70% of both materials was recovered b) data with simultaneous recovery of over 85%
recover only 11.2% of NMC input mass on average. Still,
the long tails of the outcome distributions indicated that,
for some process parameters (See table 3), NMC recovery
and grade could be quite high. The remaining question was
whether it possible to identify the process that maximizes
mass recovery for both NMC and graphite, and what com-
bination of process parameters would produce this.
A data-driven approach to process analysis facilitates
multi-objective optimization, allowing us to find numerical
solutions that simultaneously satisfy several objectives. The
focus was on the fractional mass recovery data for NMC
and graphite. Figure 5a) illustrates the selection of simu-
lated processes where recovery for both NMC and graphite
is over 70%. These 130 data points constitute only 1.3%
of all process variations. The data distribution illustrates
that large mass recovery is easier for graphite than NMC.
This is in line with observations in Figure 4, where graphite
recovery tends to favour large values, but large NMC recov-
ery features only in the tail of the distribution. It was also
observed that no parameter combination produces recover-
ies over 88% for graphite or NMC. Considering the vast
number of simulations performed, it can be confidently
said that this is a limitation of the process design, which
could be addressed by modifying processing stages, fixed
parameters, or ranges of variable parameters.
Figure 5b) illustrates the 13 best process outcomes,
which achieve mass recovery over 85% for both processes.
Here, we selected 3 optimal process outcomes for detailed
analysis. Process O1 balances best recovery for both mate-
rials, while processes O2 and O3 slightly favor recovery
of NMC and graphite respectively. Table 3 presents the
process parameters behind the optimal results. For each of
the parameters, we compared the optimal values against
the variation range. Some shared parameter settings were
observed: high values of the Fixed SGE (Pre-milling) and
very low Magnetic Field Strength clearly lead to optimal
recovery. But elsewhere, parameter values could be very
different, indicating that not all parameters affect the out-
comes to the same extent. Further feature analysis work
with data science models would clarify this aspect of the
simulations.
Table 2. Summary of NMC and graphite mass flow and grade distributions for 10,000 process outcomes: range (minimum -
maximum) and average result. Mass recovery is computed as percentage fraction of the input feed.
Parameter Range Mean Recovery range, %Recovery mean, %Feed
Graphite in
Concentrate 2
Mass, t/h 0.0–23.3 20.9 0.0–86.3 77.4 27.0
Grade, %95.0–99.8 99.4 — — 18.0
NMC in
Tailings 2
Mass, t/h 0.0–31.7 4.0 0.0–89.6 11.2 35.6
Grade, %2.5–62.5 8.8 — — 23.7
Figure 5. Process data with optimal mass recovery for NMC in tailings 2 and graphite in concentrate 2: a) data where at least
70% of both materials was recovered b) data with simultaneous recovery of over 85%