Multidimensional particle-based process characterization – A case study of dry magnetic separation of a skarn ore, M. Buchmann, E. Schach, M. Rudolph, U. Peuker, G. van den Boogaart, and R. Tolosana-Delgado

The multidimensional characterisation of processes based on various particle properties is a key requirement for a detailed prediction on certain process streams. To access the relevant particle properties (e.g. particle size, particle density and mineral content) automated mineralogy (i.e. Mineral Liberation Analyser) is applied. The kernel density estimation is used to transfer the received discrete particle data into continuous distributions without the necessity of binning the information and the related loss of information. From these continuous distributions, multidimensional separation curves are estimated. The uncertainty of these representations is tracked via bootstrap resampling. For the evaluation of the separation efficiency of the multidimensional separation curves, the statistical entropy is calculated. The statistical entropy represents a local (for each property combination) as well as a global (for the investigated property range) measure of separation efficiency and can be weighted with various property distributions (e.g. mass or mineral mass distribution) to allow for a comparison of different parameter settings and processes. With the help of a case study of a dry magnetic separation of a cassiterite bearing skarn ore from the Erzgebirge (Germany), the methodology is presented and the results are discussed in detail. Keywords: Multidimensional, magnetic separation, kernel density estimation, mineral liberation, statistical entropy