Geometallurgical Domaining by Cluster Analysis: Iron Ore Deposit Case Study

Abstract

Geometallurgy integrates many aspects of geology, mineralogy, resource modeling, mine planning, metallurgy, and process control to optimize mining operations. Small-scale metallurgical samples that determine the natural variability of the processing response in the deposit are the building blocks of a geometallurgical model. In order to take representative samples for ore characterization and metallurgical testing, it is necessary to partition a deposit into homogeneous regions in terms of processing properties, called geometallurgical domains. Quantitative rock characteristics such as chemical assay, petrophysical properties, mineralogy, and texture are used to form similar groups with regard to processing properties. This study explores a body of multivariate data to detect classes with similar inherent multivariate characteristics. As a reliable and fast method, cluster analysis identifies geometallurgical classes within a multivariate framework in the deposit, which helps in choosing and characterizing samples and performing small-scale test. The eastern part of the Dardvey iron ore deposit was selected as the case study. Three model-free clustering approaches including hierarchical clustering, k-means clustering, and self-organizing maps (SOMs) were investigated. The results show that k-means and SOM performed similarly and outperformed hierarchical clustering. The resulting domains were confirmed by geological logging and were separated well both in attribute and geographical spaces.