|dc.description.abstract||The present study applies the concept of geometallurgy to an industrial mineral case study. Geometallurgy is a holistic discipline conveying different aspects of geology, mineral processing and mining together to improve decision making, predict performance, ease the transfer of information, reduce risk and optimize mine and project performance in the whole value chain. The application of the geometallurgical concept is mainly associated to the platinum-group elements (PGE), gold and copper industries, with recent examples in the copper and iron mining industry. However, it is rarely applied in the industrial minerals mining industry, despite the known benefits of implementing geometallurgy. In this regard, the present study shows a concrete example of the application of the geometallurgical concept to the industrial mineral mining industry as a predictive tool for material characteristics and process performance.
The present thesis is focused on applying the geometallurgical concept to a nepheline syenite deposit exploited on the island of Stjernøy, northern Norway. Currently, the company base its operations on raw material and concentrate bulk chemistry analyses and concentrate yield measurements from laboratory tests done on site. It has been proven that the use of mineralogical information and predictive models in operations bring better understanding of the process behaviour and control over the resources available. In this regard, the present work aims to establish methodologies for the development of geometallurgical models for the prediction of mineralogical information such as modal mineralogy and processing behaviour such as laboratory test concentrate yield. Additionally, geometallurgical domains are defined and described to the open pit dataset to discretize in a statistically meaningful way the deposit to optimize the exploitation of the resources available.
The work was divided into three papers describing methodologies at different scales and ways to apply geometallurgical program for the creation of geometallurgical models. The first paper shows the development in MS Excel of two methodologies to predict modal mineralogy. The methodologies took into consideration the influence of mineral chemistry into estimations. The results show that a regression model has better results than a least-square model and indicated that variations in mineral chemistry do not impact greatly the prediction of the methods. The second paper proposes a methodology to predict modal mineralogy and laboratory concentrate yield based on a machine learning approach, specifically a neural network approach. The methodology was implemented in Matlab and took into consideration different input datasets to build the different neural networks. The second paper upscaled the applicability of the models to the open pit database of the company. The models successfully predicted laboratory concentrate yield and modal mineralogy throughout the open pit, which could be applied into operations. Moreover, modal mineralogy predictions followed current geological descriptions of the Nabbaren deposit and concentrate yield predictions showed an accuracy above 90% with a neural network based solely on a bulk chemistry input dataset. Finally, the third paper applies cluster analysis to the whole database of the open pit mine at Stjernøy to create different domains in the open pit mine. The domains showed a spatially logical distribution in the mine with meaningful features in accordance with the geological descriptions and processability variation. The combination of domains depicts the deposit in a geometallurgical framework for its possible implementation into resource strategy planning.
Additionally, the geometallurgical models developed proved to be readily applicable in operations. The geometallurgical models and domains developed and defined in the present study added value to the current data available to the company, with the possibility of predicting mineral characteristics and estimating material processability. Nonetheless, it is essential to further characterize, evaluate and validate the presented models and domains.||nb_NO