Thesis Análisis comparativo entre modelo empírico de vibraciones en campo lejano y modelo basado en técnica de machine learning
Loading...
Date
2023
Journal Title
Journal ISSN
Volume Title
Program
Ingeniería Civil de Minas
Campus
Campus Santiago San Joaquín
Abstract
El área de perforación y tronadura es una de las primeras operaciones de producción en una faena minera y tienen por finalidad el arranque del mineral o estéril del macizo rocoso. Con estas operaciones se generan excavaciones o bancos según el tipo de explotación con el que se trabaje, ya sea en minas a cielo abierto o subterráneas. Para un mejor aprovechamiento de las configuraciones y diseños de las mallas de voladura se busca aplicar la energía justa y necesaria para una correcta fragmentación del material, considerando los daños que puede producir en el macizo rocoso (ya sea en la superficie de este, como en bermas, talud del banco, sectores críticos, entre otros), junto otorgar la mayor seguridad a los trabajadores. Actualmente son muchas las fórmulas y métodos propuestos por diversos autores para el cálculo del esquema de perforación en voladuras de banco el cual genere buenas fragmentaciones, tenga buen cuidado de sectores aledaños y que exista un control de las vibraciones, el problema que surge de aquello es que no todos los parámetros se pueden determinar con igual precisión, ya que, depende de la caracterización del macizo rocoso, de la geometría del diseño, de la carga de explosivo, de la implementación de la malla, secuenciamiento, entre otros.
La presente memoria de título se basa en la definición de modelos predictivos de vibraciones en campo lejano, el cual considera una distancia de tres a cinco veces el largo de columna cargada, teniendo en cuenta que en esta escala los daños producidos se deben al comportamiento que posea la roca, en estas zonas se pueden deslizar estructuras en base a ondas vibratorias que superen un umbral limite, por lo que tener un monitoreo de la velocidad peak de la partícula puede disminuir el daño mediante predicciones.
En los estudios generados por personal de Enaex a diversas empresas mineras, los monitoreos de vibración permiten encontrar constantes de sitio junto con parámetros de atenuación de vibraciones en zonas en particular, ya sea analizando un dominio geotécnico o una expansión completa. En base a lo anterior, calcular e interpretar estos parámetros permite conocer los cambios en las condiciones del terreno.
Se describen dos modelos de regresión, uno correspondiente al mayormente utilizado en la industria elaborado por Devine y también se propone un modelo basado en una metodología de machine learning. En ambos casos se presentan sus parámetros con sus correspondientes intervalos de confianza, y sus pruebas de bondad de ajuste. El modelo con mejor ajuste se basa en la cantidad de datos muestrales, además de considerar los coeficientes de determinación más altos, junto con tener los errores más bajos.
The drilling and blasting area are one of the first production operations in mining and its purpose is to remove ore or waste from the rock mass. With these operations, excavations or banks are generated depending on the type of exploitation with which one works, whether in open-cast or underground mines. For a better use of the configurations and designs of the blasting meshes, the aim is to apply the exact and necessary energy for a correct fragmentation of the material, considering the damage that it can produce in the rock mass (whether on its surface, as in the berms and/or slope of the bank), together with granting the greatest safety to the workers. Currently there are many formulas and methods proposed by various actors for the calculation of the drilling scheme in bench blasting, the problem that arises from that is that not all parameters can be determined with equal precision, since it depends on the characterization of the rock mass, the geometry of the design, the explosive charge, the implementation of the mesh, sequencing, among others. This thesis is based on the definition of predictive models of vibrations in the far field, which considers three to five times the length of the loaded column, considering that on this scale the damage produced is due to the behavior of the rock, in these areas’ structures can slide based on vibration waves that exceed a threshold limit, so monitoring the peak velocity of the particle can reduce damage through predictions. In the studies generated by Enaex personnel for various mining companies, vibration monitoring allows site constants together with vibration attenuation parameters in particular areas to be assessed, whether analyzing a geotechnical domain or a complete expansion. Based on the above, calculating and interpreting these parameters allows knowing the changes in the ground conditions. Two regression models are described, one corresponding to the one most widely used in the industry developed by Devine and a model based on a machine learning methodology is also proposed. In both cases, their parameters are presented with their corresponding confidence intervals, and their goodness-of-fit tests. The model with the best fit is based on the amount of sample data, in addition to considering the highest coefficients of determination, along with having the lowest errors.
The drilling and blasting area are one of the first production operations in mining and its purpose is to remove ore or waste from the rock mass. With these operations, excavations or banks are generated depending on the type of exploitation with which one works, whether in open-cast or underground mines. For a better use of the configurations and designs of the blasting meshes, the aim is to apply the exact and necessary energy for a correct fragmentation of the material, considering the damage that it can produce in the rock mass (whether on its surface, as in the berms and/or slope of the bank), together with granting the greatest safety to the workers. Currently there are many formulas and methods proposed by various actors for the calculation of the drilling scheme in bench blasting, the problem that arises from that is that not all parameters can be determined with equal precision, since it depends on the characterization of the rock mass, the geometry of the design, the explosive charge, the implementation of the mesh, sequencing, among others. This thesis is based on the definition of predictive models of vibrations in the far field, which considers three to five times the length of the loaded column, considering that on this scale the damage produced is due to the behavior of the rock, in these areas’ structures can slide based on vibration waves that exceed a threshold limit, so monitoring the peak velocity of the particle can reduce damage through predictions. In the studies generated by Enaex personnel for various mining companies, vibration monitoring allows site constants together with vibration attenuation parameters in particular areas to be assessed, whether analyzing a geotechnical domain or a complete expansion. Based on the above, calculating and interpreting these parameters allows knowing the changes in the ground conditions. Two regression models are described, one corresponding to the one most widely used in the industry developed by Devine and a model based on a machine learning methodology is also proposed. In both cases, their parameters are presented with their corresponding confidence intervals, and their goodness-of-fit tests. The model with the best fit is based on the amount of sample data, in addition to considering the highest coefficients of determination, along with having the lowest errors.
Description
Keywords
Mineria a cielo abierto, Modelo predictivo de vibraciones, Aprendizaje automático