A workflow to create geometallurgical clusters without looking directly at geometallurgical variables

Abstract

Cluster analysis is frequently used to help in individualizing stationary domains. Its application in creating geometallurgical clusters can follow two approaches. The first utilizes geometallurgical test variables in cluster analysis to define domains based on the geometallurgical database. This approach, common in mining, often lacks sufficient data for accurate 3D modelling. The second approach uses the secondary information widely available in the deposit, such as the chemical, lithological and mineralogical variables, present in the core samples, and correlating them with the metallurgical variables presented in the smaller geometallurgical database. This indirect solution chooses the highly correlated variables with the geometallurgical response using only them as inputs in the cluster analysis. This approach permits defining the geometallurgical clusters using only drillhole samples, without directly looking at the geometallurgical information. Once the drillhole samples outnumber geometallurgical samples in the case study (4862 against 40), the spatial modeling of the geometallurgical clusters using secondary information is more accurate. The workflow proposed starts using recursive feature elimination to define the main explanatory variables affecting the target geometallurgical response. Next, clustering construction using k-means and other techniques is made, followed by building a decision tree to assign each drillhole sample to a geometallurgical cluster. Finally, it concludes with the 3D modeling to classify each block according to its geometallurgical domain. All analyses were made in the Viga mine, an iron ore deposit located at the iron quadrangle in Minas Gerais state, Brazil. The workflow and results proved to be adequate and promising to be implemented at industrial scale.