310
phenomenon, in a scale of mm/day to cm/day, caused by the gradual failure of the rock mass
when exposed to high magnitude stresses, usually deriving in the total closure of the affected
galleries.
The research studies that analyzed collapse events between 2008 and 2010 in the Esmeralda
sector showed that two distinct collapse processes exist in terms of scale. The first one being
behind of the caving front and the second being ahead of it, the latter being associated to the
presence of major geological structures and their sub-parallel orientation with respect to the
caving front.
During the development of the studies and research of this phenomenon under CODELCO’s Investigation Project (IP) T13E200 known as: “Desarrollo de Modelo de Diseño Minero para la Mitigación de Colapsos en Mina El Teniente”, building a model that allowed
the conceptualization of the phenomena was deemed necessary, after studying a series of
variables which were determined to be related to it in statistical terms, considering their interdependence, using the information in Esmeralda, Reservas Norte and Dacita (the latter two
known collectively as “RENO-Dacita”) sectors as a baseline.
Regarding the above and the interdependence of the features previously mentioned, a
Gradient-Boosting based Machine Learning algorithm was built in order to obtain an optimal generalization, given the representative elements involved based on previous diagnosis of
the collapse processes and using the model’s inherent “learning ability” as an opportunity to
improve future estimations of damage risks, allowing future and justified decision making.
3 REALIZATION OF COLLAPSES AS AN UNDERLYING PHYSICAL PROCESS
The use of a machine learning model for the estimation of a vulnerability indicator applied
to the pillars in a determined production level, is due to the fact that there is no formal analytical expression that allows the definition of the phenomenon as a physical process. Thus,
it’s occurrence has to be considered as a realization of an underlying process of such characteristics, whose results can be explained by different sets of information.
Considering the above, it’s important to point out that, previous to the IP’s closure, there
is no rigorous definition of the physical process that ends up manifesting as a collapse. On
the contrary, only a definition built on the basis of the visible progressive effects produced by
the collapse currently exists.
It is then crucial to conceive these collapses as the end-stage of a process that involves
a significant interaction between tensional and deformation states which, given the LaméHoek equations for elastic solids, include the stiffness parameters of the rockmass where the
support pillar is emplaced.
It is known that the stress solicitations caused by the mining process are fundamentally dependent on the geometry of the caving front and the variant of panel caving method implemented in
a determined production sector. On the other hand, mining causes perceivable effects in terms of
seismic waves that can cause cumulative damage, thus deriving in the necessity to apply mitigation strategies such as Hydraulic Fracturing (HF). Finally, the mining operation is not a uniform
process, given the fact that at any given time different draw points receive a continuous flow of
ore priorizing productivity, resulting in the production rate being one of the essential elements in
an intent of explaining the load distributions derived from the caving process.
In order to establish the base structure of the model, an experimental design was developed in order to support the information used, which is shown in Fig. 1. It can be noted that
the information which feeds the model is subdivided into five information sets, following the
previously mentioned criteria.
4 INFORMATION SETS
Given the experimental design shown in Fig. 1, it was determined that the variables serving as
the input of the vulnerability model were to be divided into five different sets of information,
phenomenon, in a scale of mm/day to cm/day, caused by the gradual failure of the rock mass
when exposed to high magnitude stresses, usually deriving in the total closure of the affected
galleries.
The research studies that analyzed collapse events between 2008 and 2010 in the Esmeralda
sector showed that two distinct collapse processes exist in terms of scale. The first one being
behind of the caving front and the second being ahead of it, the latter being associated to the
presence of major geological structures and their sub-parallel orientation with respect to the
caving front.
During the development of the studies and research of this phenomenon under CODELCO’s Investigation Project (IP) T13E200 known as: “Desarrollo de Modelo de Diseño Minero para la Mitigación de Colapsos en Mina El Teniente”, building a model that allowed
the conceptualization of the phenomena was deemed necessary, after studying a series of
variables which were determined to be related to it in statistical terms, considering their interdependence, using the information in Esmeralda, Reservas Norte and Dacita (the latter two
known collectively as “RENO-Dacita”) sectors as a baseline.
Regarding the above and the interdependence of the features previously mentioned, a
Gradient-Boosting based Machine Learning algorithm was built in order to obtain an optimal generalization, given the representative elements involved based on previous diagnosis of
the collapse processes and using the model’s inherent “learning ability” as an opportunity to
improve future estimations of damage risks, allowing future and justified decision making.
3 REALIZATION OF COLLAPSES AS AN UNDERLYING PHYSICAL PROCESS
The use of a machine learning model for the estimation of a vulnerability indicator applied
to the pillars in a determined production level, is due to the fact that there is no formal analytical expression that allows the definition of the phenomenon as a physical process. Thus,
it’s occurrence has to be considered as a realization of an underlying process of such characteristics, whose results can be explained by different sets of information.
Considering the above, it’s important to point out that, previous to the IP’s closure, there
is no rigorous definition of the physical process that ends up manifesting as a collapse. On
the contrary, only a definition built on the basis of the visible progressive effects produced by
the collapse currently exists.
It is then crucial to conceive these collapses as the end-stage of a process that involves
a significant interaction between tensional and deformation states which, given the LaméHoek equations for elastic solids, include the stiffness parameters of the rockmass where the
support pillar is emplaced.
It is known that the stress solicitations caused by the mining process are fundamentally dependent on the geometry of the caving front and the variant of panel caving method implemented in
a determined production sector. On the other hand, mining causes perceivable effects in terms of
seismic waves that can cause cumulative damage, thus deriving in the necessity to apply mitigation strategies such as Hydraulic Fracturing (HF). Finally, the mining operation is not a uniform
process, given the fact that at any given time different draw points receive a continuous flow of
ore priorizing productivity, resulting in the production rate being one of the essential elements in
an intent of explaining the load distributions derived from the caving process.
In order to establish the base structure of the model, an experimental design was developed in order to support the information used, which is shown in Fig. 1. It can be noted that
the information which feeds the model is subdivided into five information sets, following the
previously mentioned criteria.
4 INFORMATION SETS
Given the experimental design shown in Fig. 1, it was determined that the variables serving as
the input of the vulnerability model were to be divided into five different sets of information,