An open source platform for predictive geometallurgy, R. Tolosana-Delgado, E. Schach, N. Kupka, M. Buchmann, K. Bachmann, M. Frenzel, L. Pereira, and M. Rudolph, and K.G. van den Boogaart

Geometallurgy - the science of predicting the behaviour of ores along the value chain based on their geological and mineralogical characteristics - has barely moved beyond process minerology studies. These studies typically link process behaviour to the mineralogical and microstructural characteristics of pristine or processed ores in a qualitative or semi-quantitative manner. Eventually, a trial-and-error optimisation of the operation is achieved. However, to deliver on its promise, geometallurgy should progress beyond this current approach to a fully quantitative understanding of process behaviour, including the various uncertainties involved, namely predictive geometallurgy. One of the barriers currently hindering this development is the neglect of most ore characterisation data obtained in geometallurgical studies. This is mostly due to the large extent of this information exceeding the capacity of conventional software. To address this challenge, we developed a data mining platform consisting of an SQL database, an R front-end in-house package, and back-ends to interpret data from several analytical instruments. This platform enables the use of the statistical and data mining power of the R open-source environment for any geometallurgical study. Potential applications range from ore characterisation to process understanding and forecasting, and from geostatistical prediction to operational control and optimisation. The scripting abilities of R allow for the: 1) processing of many streams in loops, thus freeing the user from repeating tedious click-and-drop tasks; 2) computation of virtually any derivate or aggregate quantity from the data; 3) distribution of the work among several processor clusters; and 4) keeping of a log file for the calculations done. This contribution presents the building blocks of this platform, and illustrates with several examples its potential to enable predictive geometallurgy. Keywords: Data mining, entropy, kernel methods, LASSO regression, predictive geometallurgy