Flotation is an important process used to separate valuable minerals selectively from unwanted gangue. The Julius Kruttschnitt Mineral Research Centre (JKMRC) has developed a number of flotation methodologies for assessing and predicting flotation. One of these methodologies is the property-based approach, which allows a discrete description of the flotation performance as a function of the physical properties (both size and liberation) of the mineral particles.
One aspect restricting the application of this methodology is the considerable amount of data required to mass balance a flotation circuit. In particular, the gathering of mineral liberation data is costly, in terms of time, money and resources. The error that propagates in each measurement and calculation associated with the parameters used in the methodology can be greater compared to other approaches, which could have serious implications for the interpretation of the results.
The aim of this thesis is to develop a framework for laboratory and industrial flotation data analysis enabling a robust prediction of size-by-liberation flotation recovery that will improve predictive capabilities in flotation performance.
The uncertainties of the overall flotation recovery and overall kinetic constant values as a function of particle size and mineral liberation classes were assessed through propagation of error analysis.
A factor analytic model called Positive Matrix Factorisation (PMF) has been used to condense the information of recovery values obtained from physical-property-based size-by-liberation distribution for the minerals of interest in the feed stream of continuous flotation tests. PMF uses the uncertainties associated with the flotation recovery values to calculate the weights of residuals, setting a more optimal balanced scaling that also leads to optimal use of the available data.
Three different data-sets have been used to develop a methodology to predict the size-by-liberation recovery distribution. The data-sets analysed are:
i) Flotation performance of high-grade lead (galena) ore from BHP Billiton’s Cannigton Mine in a 40L continuous flotation cell, reported in Welsby’s PhD (2010);
ii) Flotation performance of high-grade lead (galena) ore from BHP Billiton’s Cannigton Mine in a 3L continuous flotation cell, reported in Vianna’s PhD (2004) at five different collector
iii) Flotation performance of low-grade copper (chalcopyrite) ore from Newcrest’s Telfer Mine in a 150 m3 continuous flotation cell, collected during this PhD project.
The overall flotation recovery values by size and liberation class and associated uncertainties were used to identify any consistent relationship between these physical properties. As these values are obtained from a mass balance of the system, they reported smaller propagated errors than the modelled flotation rate constant values, with the coefficient of variation for recovery ranging between 12.8 % and 17.4 %.
The application of PMF with a rank 1 model fitted the matrix of size-by-liberation recoveries for data-sets i and ii, while rank 2 fitted for data-set iii, so that residuals of the fit were comparable to the used data uncertainties. The two vectors resulting from PMF analysis followed clear trends as functions of particle size and galena liberation that provided the tools for developing new models for predicting flotation performance.
This thesis proposes a model that integrates mineral recovery, mass recovery and mass fraction results on a size-by-size level, with the mineral liberation analysis of a specific particle size fraction to predict flotation recovery values. A comparison of such a model with the experimental data-sets indicates that the predicted size-by-liberation recovery values reported good agreement, presenting on average differences smaller than 11% in all the six flotation case studies (data-sets i and ii) examined.
An extension of the model has been developed to enable the prediction of changes in chemistry. This was performed by relating the flotation recovery by mineral liberation classes with the surface coverage of collector onto the galena particles using data-set ii. Again, good agreement was found between the predicted and the experimental data and, more importantly, a significant reduction in the number of physical measurements required compared to the current status of the approach.
In the case of the low-grade plant survey data (data-set iii), the higher rank needed to describe the data requires that full mineralogical data is collected and provides a practical limitation of the extent to which the approach can be applied to predict size-by liberation recovery values. Despite this limitation, the methodology indicated that there were two clear components in the data, a highly liberated component above 90% chalcopyrite content, and composites below this value.
This work has demonstrated that by using the PMF methodology, the propagated uncertainties can be incorporated to an appropriate framework for predicting size-by-liberation flotation recovery. The development of new models and associated methodologies and the limitations of their application are shown. For the first time, chemistry has been included in a property-based flotation model for a real ore, offering the potential to improve the ability to predict flotation performance.