Exploiting minerals more efficiently by ore sorting with artificial intelligence

Mining News Pro - The Fraunhofer Institute for Integrated Circuits IIS, together with three other partners, intends to use artificial intelligence and sensor fusion in order to analyse the concentration of valuable minerals as early as possible in the process and thereby conserve resources.
The overall objective of the research project ‘REWO-SORT’ is the evaluation of the technical feasibility and development of improved sorting technology for raw materials by means of a multimodal sensor data fusion of optical and X-ray technologies.
The early separation of low-value material in the process chain should not only increase the treatment efficiency, but also reduce the water and energy consumption in the following process steps. The sensor data fusion method is based on deep neural networks (DNNs). The project will examine the robustness of the methodology under variable geological conditions, for example different rock compositions.
The combination of laser induced plasma spectroscopy (LIBS) and multi energy X-ray imaging (ME-XRT) is particularly promising, as the technologies complement each other very well in terms of their analytical performance: LIBS is able to provide an analysis of the chemical composition of the surface, whereas ME-XRT determines elementary information of the total object volume. “The technological convergence of these two sensor technologies will enable the extrapolation of precise surface information to the entire volume. This allows us to determine representative values for the entire ore. Adaptation to varying ore types and geological parameters will be done using artificial intelligence.” explains Markus Firsching, Project Manager at the Fraunhofer Development Center for X-ray Technology, a division of Fraunhofer IIS.
The fusion of sensor technologies to be developed should provide constant and accurate monitoring of the mineralogy of the mined rock. Special feature: the geological, mineralogical, rock mechanical and metallurgical properties of the ore are determined directly while the rock material moves over a conveyor belt. Additionally, these properties will be automatically fed into geological 3D models in order to facilitate mine planning.
The project results will be used mainly in the fields of sorting primary raw materials. An application in the field of recycling is also conceivable. Both areas pose major challenges for manufacturers of sorting machines due to increasingly complex tasks. The results regarding the use of deep neural networks should be used to enable a flexible reaction to changing requirements, as well as to facilitate the teaching of the system and the configuration for sorting material flows. For the industrial partners, these advantages represent great unique selling propositions compared to their competitors and are thus highly interesting in economic terms.