Geometallurgy is a growing area within a mineral processing industry. It brings together tasks of geologists and mineral processing engineers to do short and medium term production planning. However, it is also striving to deal with long term tasks such as changes in either production flow sheet or considering different scenarios. This paper demonstrates capabilities of geometallurgy through two case studies from perspective of Minerals and Metallurgical Engineering division Lulea University of Technology. A classification system of geometallurgical usages and approaches was developed in order to describe a working framework. A practical meaning of classification system was proved in two case studies: Mikheevskoye (Russia) and Malmberget (Sweden) projects. These case studies, where geometallurgy was applied in a rather systematic way, have shown the amount of work required for moving the project within the geometallurgical framework, which corresponds to shift of the projects location within the geometallurgical classification system.

As part of a geometallurgical program for the Leveänimei iron ore mine, the Davis tube was used as proxy to classify ore types, predict iron recoveries in wet low-intensity magnetic separation (WLIMS), and to estimate liberation of mixed particles. The study was conducted by testing 13 iron ore samples with a Davis tube and a laboratory WLIMS. Ore feed was studied for modal mineralogy and liberation distribution with Automated Scanning Electron Microscopy. Data analyses to detect the patterns and data dependencies were done with multivariate statistics: principal component analysis, and projection to latent structures regression. Results show that a simple index (X_LTU) based on mass pull (yield) in the Davis tube is capable of easy classification of magnetite ores. Using Davis tube mass pull and iron recovery, together with iron and Satmagan head grades may predict iron recovery in WLIMS. Also, the variability in Fe-oxides liberation pattern for magnetite semi-massive ores can be explained with the chemical composition of the Davis tube concentrate. It is concluded that the Davis tube test is better used only for marginal ores, since iron oxide minerals tend to be fully liberated in high-grade magnetite massive ores after grinding. The developed models may be used in populating a production block model.

The main purpose of geometallurgy is to develop a model to predict the variability in the mineralprocessing performance within the ore body. Geometallurgical tests used for developing such a model need to be fast, practical and inexpensive and include as an input data relevant and measureable geological parameters like elemental grades, mineral grades and grain size. Important in each geometallurgical program is to define the number of samples needed to be sent for geometallurgical testing to enable reliable metallurgical forecast. This is, however, a complicated question that does not have a generic answer.

To study the question on sampling a simulation environment was built including a synthetic orebody and sampling & assaying module. A synthetic Kiruna type iron oxide - apatite deposit was established based on case studies of Malmberget ore. The synthetic ore body includes alike variability in rock types, modal mineralogy, chemical composition, density and mineral textures as its real life counterpart. The synthetic ore body was virtually sampled with different sampling densities for a Davis tube testing, a geometallurgical test characterising response in magnetic separation. Based on the test results a forecast for the processing of the whole ore body was created. The forecasted parameters included concentrate tonnages, iron recovery and concentrate quality in terms of iron, phosphorous and silica contents.

The study shows that the number of samples required for forecasting different geometallurgicalparameters varies. Reliable estimates on iron recovery and concentrate mass pull can be made with about 5-10 representative samples by geometallurgical ore type. However, when the concentrate quality in terms of impurities needs to be forecasted, the sample number is more than 20 times higher. This is due to variation in mineral liberation and shows the importance of developing techniques to collect qualitative information on mineral and ore textures in geometallurgy.

Reconciliation of geological, mining and mineral processing information is a costly and time demanding procedure with high uncertainty due to incomplete information, especially during the early stages of a project, i.e., pre-feasibility, feasibility studies. Lack of information at those project stages can be overcome by applying synthetic data for investigating different scenarios. Generation of the synthetic data requires some minimum sparse knowledge already available from other parts of the mining value chain, i.e., geology, mining, mineral processing. This paper describes how to establish and construct a synthetic testing environment, or “synthetic ore body model” by integrating a synthetic deposit, mine production, constrained by a mine plan, and a simulated beneficiation process. The approach uses quantitative mineralogical data and liberation information for process simulation. The results of geological and process data integration are compared with the real case data of an apatite iron ore. The discussed approach allows for studying the implications in downstream processes caused by changes in upstream parts of the mining value chain. It also opens the possibility of optimising sampling campaigns by investigating different synthetic drilling scenarios including changes to the spacing between synthetic drill holes, composite length, drill hole orientation and assayed parameters.

A spatial model for process properties allows for improvedproduction planning in mining by considering the process variability ofthe deposit. Hitherto, machine-learning modelling methods have beenunderutilised for spatial modelling in geometallurgy. The goal of thisproject is to find an efficient way to integrate process properties (ironrecovery and mass pull of the Davis tube, iron recovery and mass pull ofthe wet low intensity magnetic separation, liberation of iron oxides, andP_80) for an iron ore case study into a spatial model using machinelearningmethods. The modelling was done in two steps. First, the processproperties were deployed into a geological database by building nonspatialprocess models. Second, the process properties estimated in thegeological database were extracted together with only their coordinates(x, y, z) and iron grades and spatial process models were built.Modelling methods were evaluated and compared in terms of relativestandard deviation (RSD). The lower RSD for decision tree methodssuggests that those methods may be preferential when modelling non-linearprocess properties.