Agrupamento de domínios e modelagem geometalúrgica da Mina Chapadão, Ouvidor-GO

Abstract

Mineral deposit modeling has been improved over the years with the incorporation of geological, chemical, and metallurgical information. However, the diversity of this information makes this activity expensive since the data are not unified in procedures, sample support, or quantities. The non-additivity and non-linearity of certain attributes impose further difficulties. Recently, new approaches in computational geometallurgy have shown remarkable progress in the analysis of spatial data to overcome the challenges that would restrict the creation of these models. This study is driven by these new methodologies and employs machine learning techniques to enable the Chapadão Mine model, facing difficulties such as the geological complexity of transitional contacts and overlapping structures, in addition to the lack of uniformity in the database, that is, the metallurgical information is significantly smaller compared to the others. This situation is not exclusive to Chapadão, but is common to most mineral deposits, which often limit mineral modeling to the primary attributes of the rock. The generalization of metallurgical models is a valid option when the relationships between geological characteristics and processing responses are well defined. From this perspective, ore typologies were defined, whose geological and chemical characteristics are compatible with the metallurgical results. The fuzzy algorithm was tested to manage the uncertainty of the studied phenomenon and to classify the drill samples into typological domains modeled by kriging of the indicators. Three machine learning techniques were analyzed to predict the process attributes. The satisfactory results of the predictive models allowed the geostatistical simulation of the primary variables for each domain of the geometallurgical model that were reconciled with the production data.