321
− In Andes Norte Project, the results show there are 109 pillars labeled as vulnerable by the
PVI from a total of 691 (PVI ≥ 30%), which is roughly 16% of the total. Most of these pillars tend to be located in the vicinities of lithological contacts, such as CMET-Diorite and
CMET-Anhydrite Breccia, and in the area where the caving process will be initiated when
the Project finally begins its operation. This latter area is coincident with the one where the
vertical loads are mostly unfavorable, due to the existence of high terrain elevations over
the surface.
7 CONCLUSIONS
According to the previous discussion, the following conclusions can be drawn:
a. Regarding the developed model:
− The algorithmic process carried out using the historical data of El Teniente Mine allowed the
construction a model with high predictive capabilities regarding the collapse phenomena.
− The features used in the model development are sufficient to capture the phenomena
occurrence via the algorithmic procedure.
− The model has a 91% diagnostic power, based on PVI values and the model’s quality, estimated by its ROC curve.
− Based on a PVI cutoff value of 30%, a very high accuracy across the training set is reached,
having a 96% value for damaged pillars (sensibility). For undamaged pillars, this value
reaches up to 94% (specificity).
− The current model serves as a very good baseline for future accuracy improvements.
b. Regarding the model’s application over production sectors currently under development and
future projects:
− The results obtained by the model’s application are reasonably coherent with the current
situation presented in the mine. The diagnosed vulnerable pillars tend to be located in historically known risky areas, which supports the coherence of the whole research.
− The results suggest that there is a significant structural, geological and environmental component aspect regarding the potential occurrence of collapses in pillars.
c. Regarding the global perspective:
− From the individual features used to build the model, the most important aspects are the
shear components of stresses and strains for an end-stage cavity as result of the mining
process of a production sector and the seismic energy densities in both components, P and
S, which shows the significance of the geological discontinuities over collapses occurrences.
− The mining operation can be assumed as a collection of features, each one of them being a
necessary but not sufficient condition for a collapse process to occur. This way, only after
reaching a series of limit values for each of these features, the collapse process can be triggered, given the environmental conditions of the rockmass where pillars are emplaced.
Finally, the vulnerability model formerly developed has been proposed as a safety criterion, destined to evaluate the limit condition of the explanatory variables formerly defined
for this study as representative of the underlying collapse process. This way, for any current
or future sector, this tool provides an insight of the regarding environmental and operational
conditions that may or may not end in collapse damage for support pillars, allowing the
possibility of future decision making processes oriented to modify operational and mining
parameters in order to minimize the occurrence of collapse phenomena.
ACKNOWLEDGEMENTS
The authors wish to thank CODELCO Chile, Division El Teniente for their support and
authorization to use the required data in this research and Universidad de Santiago de Chile’s
(USACH) Mining Engineering Department for their technical and financial support as well
as providing the materials for the computational processes.
− In Andes Norte Project, the results show there are 109 pillars labeled as vulnerable by the
PVI from a total of 691 (PVI ≥ 30%), which is roughly 16% of the total. Most of these pillars tend to be located in the vicinities of lithological contacts, such as CMET-Diorite and
CMET-Anhydrite Breccia, and in the area where the caving process will be initiated when
the Project finally begins its operation. This latter area is coincident with the one where the
vertical loads are mostly unfavorable, due to the existence of high terrain elevations over
the surface.
7 CONCLUSIONS
According to the previous discussion, the following conclusions can be drawn:
a. Regarding the developed model:
− The algorithmic process carried out using the historical data of El Teniente Mine allowed the
construction a model with high predictive capabilities regarding the collapse phenomena.
− The features used in the model development are sufficient to capture the phenomena
occurrence via the algorithmic procedure.
− The model has a 91% diagnostic power, based on PVI values and the model’s quality, estimated by its ROC curve.
− Based on a PVI cutoff value of 30%, a very high accuracy across the training set is reached,
having a 96% value for damaged pillars (sensibility). For undamaged pillars, this value
reaches up to 94% (specificity).
− The current model serves as a very good baseline for future accuracy improvements.
b. Regarding the model’s application over production sectors currently under development and
future projects:
− The results obtained by the model’s application are reasonably coherent with the current
situation presented in the mine. The diagnosed vulnerable pillars tend to be located in historically known risky areas, which supports the coherence of the whole research.
− The results suggest that there is a significant structural, geological and environmental component aspect regarding the potential occurrence of collapses in pillars.
c. Regarding the global perspective:
− From the individual features used to build the model, the most important aspects are the
shear components of stresses and strains for an end-stage cavity as result of the mining
process of a production sector and the seismic energy densities in both components, P and
S, which shows the significance of the geological discontinuities over collapses occurrences.
− The mining operation can be assumed as a collection of features, each one of them being a
necessary but not sufficient condition for a collapse process to occur. This way, only after
reaching a series of limit values for each of these features, the collapse process can be triggered, given the environmental conditions of the rockmass where pillars are emplaced.
Finally, the vulnerability model formerly developed has been proposed as a safety criterion, destined to evaluate the limit condition of the explanatory variables formerly defined
for this study as representative of the underlying collapse process. This way, for any current
or future sector, this tool provides an insight of the regarding environmental and operational
conditions that may or may not end in collapse damage for support pillars, allowing the
possibility of future decision making processes oriented to modify operational and mining
parameters in order to minimize the occurrence of collapse phenomena.
ACKNOWLEDGEMENTS
The authors wish to thank CODELCO Chile, Division El Teniente for their support and
authorization to use the required data in this research and Universidad de Santiago de Chile’s
(USACH) Mining Engineering Department for their technical and financial support as well
as providing the materials for the computational processes.