An open source platform for predictive geometallurgy

Geometallurgy (the science of predicting the behaviour of ores along the value chain on the basis of their geoscientific characteristics) has barely moved beyond process mineralogy studies. These studies typically describe the mineralogical and microstructural characteristics of pristine or processed ores, for qualitative or semi-quantitative assessment of the processing performance and, eventually, a trial-and-error optimisation of the operation. But to deliver on its promise, geometallurgy should progress beyond this to provide fully quantitative models of such primary ore characteristics and the performance of their processing, both in average value and, most importantly, keeping track of their uncertainty. Only then it will be possible to optimise the operation in a predictive way. For this goal, we have developed a data mining platform consisting on a SQL database, an R front-end in-house package, and back-ends interpreting data from several analytical instruments into the database. A structured relational SQL database is useful to organise all available information (geometry, phase, chemistry, physical properties) of mineral grains, particles and samples. The R front-end allows to use the statistical and data mining power of this open-source environment in a transparent way for any geomet task, from ore characterisation to process fitting and forecasting, from geostatistical prediction to operational optimisation.
With regard to the back-ends, currently we import XRD, XRF and mineral liberation analysis measurements, and we will incorporate other data sources. The scripting abilities of R allow to process many streams in loop freeing the user of repeating tedious click-and-drop tasks; compute virtually any quantity; distribute the work among several clusters; or keep a log file of the calculations done, for accountability. This contribution presents the building blocks of this package, and their use with several examples in geometallurgical tasks.