Modern analytical techniques used in the minerals processing industry can provide detailed characterization data at the particle level. However, process models that make full use of this information are currently not available, limiting the usefulness of these extensive datasets. This contribution addresses this issue. It presents a novel particle-based approach for process modelling capable of dealing with complete particle datasets and operating without human input. The method provides a probabilistic description for the behavior of individual particles in a given mineral separation unit, based on all measurable particle properties. It is applicable to any separation process that does not modify the physical dimensions of particles, i.e. it does not cover comminution. The method comprises a regularized logistic regression model with a probability adjustment step to accommodate geological variability. Even though this method supports any type of particle-level characterization data, its potential is illustrated here using data obtained by scanning electron microscope based image analysis. Constructed cases demonstrate the efficiency of the method in recreating characteristic recovery trends for magnetic separation, hydrocyclone, and flotation units. In addition, the method was used successfully to reconstruct a real processing plant with three flotation and one magnetic separation circuits. Predicted results of compositions for all the intermediate and product streams correspond well with the results reported from the plant itself. The predicted masses of the products are much affected by the quality of sampling and still require improvement. The case study illustrates that the method proposed here provides a powerful tool to understand and optimize mineral separation processes e and thus increase the resource and energy efficiency of mining operations. (c) 2020 Elsevier Ltd. All rights reserved.
原文链接
分享
引用
