Mining Goes Digital – Mueller et al. (Eds)
© 2019 Taylor & Francis Group, London, ISBN 978-0-367-33604-2
551
Modelling and forecasting geometallurgical recovery
at a phosphate mine
L.B. Andrade & I.E. Cabral
Federal University of Ouro Preto, Brazil
J.F.C.L. Costa
Federal University of Rio Grande do Sul, Brazil
ABSTRACT: As human population grows, so does the pressure to improve the efficiency
on the use of mineral resources which are finite, nonrenewable and fundamental to human
existence and its development. Lately there has been a growth in the application of geometallurgy models, an instrument that allows the connection between ore characteristics and mineral processing in a predictive model that is essential for mine planning. Integrated control of
the quality of the ore from its mining to its final product can be accomplished by building a
geological-typological model, establishing the spatial relationships at the deposits to the different ore-type categories. This article presents an ore-type model built using clustering and
indicator kriging to identify and map the domains. Additionally, the metallurgical recovery
within each domain is predicted using a multivariate linear regression model. The results
were compared against real recovery proving the accuracy and precision of the predictions
derived from the model. The methodology is illustrated at a major phosphate deposit
1 INTRODUCTION
Geometallurgy is a field based on a multidisciplinary approach aimed at integrating geological, mineralogical and metallurgical information and yielding a spatial quantitative predictive
model for production management (LISHCHUK, 2015). The favorable characteristics of an
ore body that require a geometallurgical model are high variability, low mineral grade and
complex mineralogy. The phosphate deposit studied in this article is an Alkaline-carbonate
complex, situated at Serra do Salitre with an average P2O5 lower than other active phosphate
mines, where intense weathering has occurred, forming new mineral formations and creating a complex mineralogical environment. The concept of geometallurgy fits the problem
definition, since it furnishes expected plant performance depending on the feedstock, integrating the quality control of the ore and decreasing the uncertainty that surrounds recovery
performance.
The metallurgical recovery information can be achieved experimentally at a pilot plant or
by lab tests, that require time and money. One way to reduce the cost of the experiments is to
identify chemical components, currently analyzed in the ore, with possible correlation with
the metallurgical recovery. To identify patterns according to ore’s x-ray features, a machine
learning technique was applied; to predict its behavior, whereby regression equations were
established. In addition, to model the ore-type boundaries the concept of Indicator Kriging was proposed. Machine learning is a method of data analysis that automates analytical
model building. It is a branch of artificial intelligence based on the idea that systems can
learn from data, identify patterns and make decisions with minimal human intervention.
The Machine Learning technique applied in the study was the Hierarchical Cluster Analysis,
which involves grouping unsupervised data points without including targets, so we do not tell
the system where to go, the system must understand itself from the data given. In this way,
© 2019 Taylor & Francis Group, London, ISBN 978-0-367-33604-2
551
Modelling and forecasting geometallurgical recovery
at a phosphate mine
L.B. Andrade & I.E. Cabral
Federal University of Ouro Preto, Brazil
J.F.C.L. Costa
Federal University of Rio Grande do Sul, Brazil
ABSTRACT: As human population grows, so does the pressure to improve the efficiency
on the use of mineral resources which are finite, nonrenewable and fundamental to human
existence and its development. Lately there has been a growth in the application of geometallurgy models, an instrument that allows the connection between ore characteristics and mineral processing in a predictive model that is essential for mine planning. Integrated control of
the quality of the ore from its mining to its final product can be accomplished by building a
geological-typological model, establishing the spatial relationships at the deposits to the different ore-type categories. This article presents an ore-type model built using clustering and
indicator kriging to identify and map the domains. Additionally, the metallurgical recovery
within each domain is predicted using a multivariate linear regression model. The results
were compared against real recovery proving the accuracy and precision of the predictions
derived from the model. The methodology is illustrated at a major phosphate deposit
1 INTRODUCTION
Geometallurgy is a field based on a multidisciplinary approach aimed at integrating geological, mineralogical and metallurgical information and yielding a spatial quantitative predictive
model for production management (LISHCHUK, 2015). The favorable characteristics of an
ore body that require a geometallurgical model are high variability, low mineral grade and
complex mineralogy. The phosphate deposit studied in this article is an Alkaline-carbonate
complex, situated at Serra do Salitre with an average P2O5 lower than other active phosphate
mines, where intense weathering has occurred, forming new mineral formations and creating a complex mineralogical environment. The concept of geometallurgy fits the problem
definition, since it furnishes expected plant performance depending on the feedstock, integrating the quality control of the ore and decreasing the uncertainty that surrounds recovery
performance.
The metallurgical recovery information can be achieved experimentally at a pilot plant or
by lab tests, that require time and money. One way to reduce the cost of the experiments is to
identify chemical components, currently analyzed in the ore, with possible correlation with
the metallurgical recovery. To identify patterns according to ore’s x-ray features, a machine
learning technique was applied; to predict its behavior, whereby regression equations were
established. In addition, to model the ore-type boundaries the concept of Indicator Kriging was proposed. Machine learning is a method of data analysis that automates analytical
model building. It is a branch of artificial intelligence based on the idea that systems can
learn from data, identify patterns and make decisions with minimal human intervention.
The Machine Learning technique applied in the study was the Hierarchical Cluster Analysis,
which involves grouping unsupervised data points without including targets, so we do not tell
the system where to go, the system must understand itself from the data given. In this way,