Logo Minería
login

Inicie sesión aquí

MACHINE LEARNING FOR NATURAL RESOURCES

Current advances in computing are revolutionizing the way we treat data and make decisions. The development of artificial intelligence, blockchain and quantum computing technologies are at the forefront of these advances. With more than 3,000 researchers in 12 labs, IBM Research has been a pioneer in the development of these technologies across the world.

In this context, it has developed most of its research on natural resources in its research lab in Brazil, which focuses on the development of new artificial intelligence and visual analytics techniques for applications in oil and gas, agriculture and mining.

One recent advance is the use of state-of-the-art machine learning techniques that take into account the characteristics of subsurface data and expert knowledge to aid geoscientists in the discovery of natural resources. In particular, we have recently developed a methodology using machine learning for identifying new exploration targets to test potential zones of gold mineralization using drill hole data which we describe in this paper.

In order to identify new exploration targets to test potential zones of gold mineralization, geologists gather different sources of data such as drill hole observations and measurements, maps of visible geological structure and assay lab results to justify their decisions. Such activity is complex and requires expert tacit knowledge acquired through experience, which is specific to each mining project, meaning that the location's underlying geology is unique.

The activity for establishing new exploration targets can be subjective and requires many hours from highly trained individuals. A considerable amount of time is spent on interpreting geological information acquired from different sources (Häggquist and Söderholm, 2015). Most of the current tools are designed for the geologist to manually interpret the data. While current tools can handle smaller volumes, they all have issues dealing with larger mine projects which makes identifying patterns across multiple data sets difficult. Additionally, ore and waste classification is still done manually by analyzing collected geological information, which makes target identification prone to human error.

A way to tackle this problem is to use data-driven predictive modeling to represent the relationships among data, which can then be used in new regions to more readily identify exploration targets. We developed one such method based on machine learning (Zadrozny et al., 2018). The method involves a data transformation step and a predictive modeling step using convolutional neural networks. We will then summarize the results achieved using this methodology on real data from a gold mining company and discuss the conclusions of this paper.

Artículos relacionados

Se analizó situación y conflictividad entorno a la minería

Seguir leyendo
X

Ingrese sus datos aquí

X

Recuperar Contraseña

X

Recuperar Contraseña

Si tiene problemas para recuperar su contraseña contáctese con el Área de Servicio al Asociado al teléfono 313-4160 anexo 218 o al correo asociados@iimp.org.pe

X

Ha ocurrido un error al iniciar sesión

Si tiene problemas para recuperar su contraseña contáctese con el Área de Servicio al Asociado al teléfono 313-4160 anexo 218 o al correo asociados@iimp.org.pe

X

Ingrese sus datos y nos pondremos en
contacto para poder completar su compra

X

Ingrese sus datos y nos pondremos en
contacto para poder completar su compra