XXXI International Mineral Processing Congress 2024 Proceedings/Washington, DC/Sep 29–Oct 3 1407
Future work will refine the model by incorporating the
lengths of the conveyors, updating the failure assumptions
to reflect the change in ore attributes if a quantifiable and
causally plausible correlation is found. In this example, the
likelihood of an unplanned down in the crusher considered
the ore hardness as determined in the ore control model.
Similarly, any correlation between the ore characteristics as
determined by the ore control model and the downstream
processing performance can be used to further constrain
the model behavior. Lastly, time series studies will be used
to validate that the modeled arrival times for the ore is con-
sistent with the observed processing outcomes as a way of
ground-truthing the model.
Of these improvements, the most challenging will
likely be identifying a consistent correlation between the
ore attributes and downstream downtimes. One reason for
this may be that the datasets are noisy and may have gaps
or include metrics that are not suitable for this purpose. For
example, some of the ore control attributes are not based
on metallurgical testing but rather are calculated based on
proxy data. Maintenance data may rely on process control
operators manually assigning a reason for the down, and
over half the recorded downs do not have a specified rea-
son. Additionally, it is necessary to use weighted average
and summed downtimes when calculating correlations.
Not doing so results in a noisy data set correlated with
another noisy dataset, thus any potential correlations will
be missed. However, using an overly broad window dur-
ing the weighted average and summation calculations will
overly smooth the data, which may also disguise any poten-
tial correlations. Therefore, a systematic approach is nec-
essary to determine the optimal combination of weighted
average window, summation window, and time lag between
the two data sets.
Currently, the model replicates mineral processing
from 2022, however once calibrated the model inputs
can be replaced to use planned production instead for a
forward-looking model. In that case, the timing of ore
deliveries would be based upon a Poisson or exponential
distribution, depending on whether one wanted to model
the time between truck arrivals or model the number of
trucks per unit of time. In either case, the existing DES
model, once calibrated, could easily be adjusted to run on
predicted ore deliveries instead.
While the goals of the current study were modest, they
lay the groundwork for additional processing modules.
Passing the foreign keys to the original mine data source
as attributes adds the ability to connect the location of the
in-situ ore to downstream processing outcomes. Using a
series of smaller model capsules while tracking the foreign
key elements allows each component of the model to han-
dle the minimum amount of data necessary, as additional
geoattributes can be added or dropped as needed for each
mini-model. Thus, if downstream processing components
are sensitive to any geological attributes not modeled here,
these can be retrieved and incorporated into the compo-
nents of the processing system for which they are relevant.
CONCLUSION
A successful geometallurgical DES model could be used to
drive concrete operational improvements at an active min-
ing operation based on the same geostatistical inputs used
in long range planning and actual mine production mea-
surements. Such a model would be more directly related
to an operation than geometallurgical test work completed
during the feasibility stage of project development, while
incorporating more fundamental geologic information
than the typical process control model. The goal is to truly
integrate orebody knowledge with production in an explicit
manner, thus driving significant operational improvements
and offering new ways of monitoring production. Given
the high variability of geologic and operational data, how-
ever, such techniques will in no way replace conventional
lab analyses for metallurgical testing/ore characterization
work and this novel technique will require much develop-
ment, including more extensive ground-truthing, before
the full potential of such an approach can be realized.
In future research, the outputs of the DES model will
be used to compare the specific process outcomes, both
modeled and observed, to the in situ geologic characteris-
tics of the orebody, with the goal of identifying new stra-
tegic insights into processing behavior and providing more
visibility into the performance of the ore during processing.
The ultimate goal is to achieve improved operational per-
formance through more precise and geologically-informed
ore routing criteria or blending requirements, especially as
they relate to acid consumption and clay-related processing
issues. This will be combined with additional DES models,
for a more comprehensive assessment of operational perfor-
mance as it relates to the geologic characteristics of the ore.
Once validated, such models can be used for comparisons
of processing outcomes to the original mining data using
common data mining techniques. Between the original
mining and processing data and the modeled production
outcomes, it will be possible to develop predictive and pre-
scriptive analytics tools that can offer insights into past pro-
duction and well as predict future returns.
Future work will refine the model by incorporating the
lengths of the conveyors, updating the failure assumptions
to reflect the change in ore attributes if a quantifiable and
causally plausible correlation is found. In this example, the
likelihood of an unplanned down in the crusher considered
the ore hardness as determined in the ore control model.
Similarly, any correlation between the ore characteristics as
determined by the ore control model and the downstream
processing performance can be used to further constrain
the model behavior. Lastly, time series studies will be used
to validate that the modeled arrival times for the ore is con-
sistent with the observed processing outcomes as a way of
ground-truthing the model.
Of these improvements, the most challenging will
likely be identifying a consistent correlation between the
ore attributes and downstream downtimes. One reason for
this may be that the datasets are noisy and may have gaps
or include metrics that are not suitable for this purpose. For
example, some of the ore control attributes are not based
on metallurgical testing but rather are calculated based on
proxy data. Maintenance data may rely on process control
operators manually assigning a reason for the down, and
over half the recorded downs do not have a specified rea-
son. Additionally, it is necessary to use weighted average
and summed downtimes when calculating correlations.
Not doing so results in a noisy data set correlated with
another noisy dataset, thus any potential correlations will
be missed. However, using an overly broad window dur-
ing the weighted average and summation calculations will
overly smooth the data, which may also disguise any poten-
tial correlations. Therefore, a systematic approach is nec-
essary to determine the optimal combination of weighted
average window, summation window, and time lag between
the two data sets.
Currently, the model replicates mineral processing
from 2022, however once calibrated the model inputs
can be replaced to use planned production instead for a
forward-looking model. In that case, the timing of ore
deliveries would be based upon a Poisson or exponential
distribution, depending on whether one wanted to model
the time between truck arrivals or model the number of
trucks per unit of time. In either case, the existing DES
model, once calibrated, could easily be adjusted to run on
predicted ore deliveries instead.
While the goals of the current study were modest, they
lay the groundwork for additional processing modules.
Passing the foreign keys to the original mine data source
as attributes adds the ability to connect the location of the
in-situ ore to downstream processing outcomes. Using a
series of smaller model capsules while tracking the foreign
key elements allows each component of the model to han-
dle the minimum amount of data necessary, as additional
geoattributes can be added or dropped as needed for each
mini-model. Thus, if downstream processing components
are sensitive to any geological attributes not modeled here,
these can be retrieved and incorporated into the compo-
nents of the processing system for which they are relevant.
CONCLUSION
A successful geometallurgical DES model could be used to
drive concrete operational improvements at an active min-
ing operation based on the same geostatistical inputs used
in long range planning and actual mine production mea-
surements. Such a model would be more directly related
to an operation than geometallurgical test work completed
during the feasibility stage of project development, while
incorporating more fundamental geologic information
than the typical process control model. The goal is to truly
integrate orebody knowledge with production in an explicit
manner, thus driving significant operational improvements
and offering new ways of monitoring production. Given
the high variability of geologic and operational data, how-
ever, such techniques will in no way replace conventional
lab analyses for metallurgical testing/ore characterization
work and this novel technique will require much develop-
ment, including more extensive ground-truthing, before
the full potential of such an approach can be realized.
In future research, the outputs of the DES model will
be used to compare the specific process outcomes, both
modeled and observed, to the in situ geologic characteris-
tics of the orebody, with the goal of identifying new stra-
tegic insights into processing behavior and providing more
visibility into the performance of the ore during processing.
The ultimate goal is to achieve improved operational per-
formance through more precise and geologically-informed
ore routing criteria or blending requirements, especially as
they relate to acid consumption and clay-related processing
issues. This will be combined with additional DES models,
for a more comprehensive assessment of operational perfor-
mance as it relates to the geologic characteristics of the ore.
Once validated, such models can be used for comparisons
of processing outcomes to the original mining data using
common data mining techniques. Between the original
mining and processing data and the modeled production
outcomes, it will be possible to develop predictive and pre-
scriptive analytics tools that can offer insights into past pro-
duction and well as predict future returns.