Testing the validity of ore texture descriptors used in mineral liberation modeling

Abstract

Determining the extent of liberation of the valuable mineral from its gangue matrix achieved at the comminution stage is of great importance in optimizing the downstream recovery processes in most mineral processing operations. Reasonably accurate comminution models are available for the prediction of the product size distributions, but models that describe the degree of liberation achieved is scarce in the published literature. In this regard, the liberation model based on linear intercept distributions of mineral and gangue proposed by King (1979), its application to predict the flotation performance of an industrial plant demonstrated by Schaap (1979) and liberation models based on geometric probability theory by Davy (1984) and Barbery (1991) may be considered useful contributions. However, models proposed by Schaap and King are based on ore texture information obtained from the parent rock, i.e. intercept length distributions of mineral and gangue measured on polished sections. While such information on the distribution of mineral in particles can be obtained by scanning electron microscope (SEM) based techniques, such as, Mineral Liberation Analyzer (MLA) and Quantitative Evaluation of Minerals by Scanning Electron Microscopy (QEMScan), more complex functions such as the covariance functions and proximity functions used in Barbery’s models are more difficult to determine. In the absence of measured data Barbery derived functions for the above descriptors based on assumed ore textures, such as Poisson polyhedra and Boolean textures. Such models are applicable at best for low grade ores but not for real ores at higher grades.This paper describes a method of measuring the textural descriptors using image analysis techniques using SEM images of the parent rock and comminuted particles. The conventional texture descriptors, such as linear mineral and gangue intercept distributions, covariance function and the proximity function that describe particle properties, have been obtained from 2-D images and compared with those used by Barbery in the prediction of liberation characteristics of comminuted particles. It was found that the functional forms for the proximity function of the particles proposed by King and the covariance function suggested by Barbery were not suitable for the high grade ore type used in this work. The method described in this work adopts numerical methods to determine the above functions without any assumption of ore texture or particle shape. The techniques outlined in this work could be applied to extract more useful ore texture information from MLA type images that are familiar to most mineral processing practitioners.