Through the Fire and Flames

Abstract

Industrial copper smelting is a popular pyrometallurgical process that enriches the concentration of copper in concentrated mineral ores. The composition of these ores vary depending on their geographical origin and form the feedstock for a flash furnace. Currently, only limited technologies are available to determine the compositional information of feed materials, and process control is impossible. A new method of compositional analysis for process control in industrial copper smelting is presented in this thesis. The development of a flame emission system (patent-pending) using a high-temperature acetylene burner to measure the elemental composition and mineralogy of a powdered feed material without sample preparation is described. A spectrometer records the flame emission spectrum several times a second. An analysis of the spectral features revealed atomic and molecular emission features related to the composition of the material. The analysis is complicated by the blackbody background, the many short-lived species in the flame, and inner-filter effects that prevented the extraction of quantitative information from conventional spectroscopic analysis. A comparison of supervised and unsupervised models and spectral preprocessing techniques was conducted to predict the composition of powdered materials from complex emission data. An Artificial Neural Network (ANN) algorithm was appropriately trained on over 8700 spectra collected from 47 industrially relevant powdered samples with well-characterized chemical and physical properties. The ANN provides a robust prediction of the elemental and mineralogical composition, in addition to the particle size distribution of concentrate composition, with an approximate accuracy of less than 1% (m/mtotal), 2 % (m/mtotal), and 1 um, respectively. It was 3-times more accurate than the PLS models. Furthermore, an ANN parameter analysis revealed that predictors of greatest influence can be related back to known spectral features. An additional study on the combustion of copper concentrates using a lab-scale drop tower reactor is described. A complete elemental and mineralogical analysis of the concentrates before and after combustion allowed for the development of a preliminary PLS model to describe the desulphurization characteristics of concentrated sulphide ores based on their fractional mineralogy and, with continued training, could be used to predict the combustion of new blends.