Abstract

Ore pre-concentration by sensor-based sorting, brings a huge potential for significant energy savings, more efficient mineral processing, reduced waste generation and significantly reduced environmental impact. Currently, the main problem in pre-concentration is related to too high content of valuable materials in waste fractions, caused by inefficient sorting. Therefore, a new sorting system has been developed to achieve accurate sorting especially at low mineral content, being typical for CRM ores.    The system employs high selectivity X-ray detectors operating together with other sensors like RGB and SWIR (Short-Wave-Infra-Red) cameras, installed in the same sorting unit. This combination significantly increases number of parameters describing processed materials thus allowing accurate identification of ores representing metals like nickel, copper, magnesium, tungsten, REE and many others. By detailed analysis of the internal particle structure by XRT sensors and the particle surface by additional optical cameras, it is also possible to detect waste rocks and minerals which do not contain any valuable material. Therefore, the sorting system, provides a dual identification and verification of sorted material where CRM ores and waste rocks are identified parallelly, to finally provide an accurate separation of waste rocks with minimal content of any valuable material.    It is done by special correlation models, which compare data from different sensors. Advanced digital filtration tools with smart data fusion are used, to compare information available in different formats and resolutions. The pixel resolution of the X-ray sensors is from 50 µm to 0.8 mm, the RGB cameras provide 0.2-0.5 mm resolution and the SWIR sensors bring 2-3 mm pixel resolution. In addition, X-ray systems operate with multiple energy analysis, RGB cameras provide three color channels and the SWIR system can bring up to 256 band channels. It is therefore very challenging to analyze this huge data stream in an optimal way, to provide still efficient data processing in short time intervals, which is very critical to achieve a high sorting capacity. Final decision is done by AI based models employing convolutional neural networks, which in addition can provide self-learning possibilities during sorter operation to achieve an optimal operation.    The paper describes sorting application examples for copper and REE ores. For copper ore preconcentration, the feed material had 1.2-2% of Cu, which was concentrated up to 1.5-2.4% of Cu after sorting and where the sorted waste fraction had Cu content below 0.15% (being below the cutoff grade). This Cu content of 0.15% was fully acceptable by the mine where the current flotation tailings had 0.2-0.25 % of Cu from regular processing. Similar pre-concentration effects were achieved in REE-bearing iron-dolomite carbonatites (FDC), where 25-35% of waste rocks could be removed resulting in 33-50% pre-concentration in the product fraction. This brings significant savings and environmental benefits as the REE ores require high cost and complicated further processing. The presented sorting system has already been applied in the industrial scale in pre-concentration of different metal ores, providing huge economic and environmental savings. The paper presents the complete development from laboratory, pilot to industrial scale applications.