Publications Repository - Helmholtz-Zentrum Dresden-Rossendorf

Resource and energy efficiency are essential for the raw-materials industry to secure a sufficient and economically feasible supply of minerals and metals for society in the coming decades. This task becomes more challenging as the complexity of primary resources increases. Mineral processing plant control systems, an important tool for guaranteeing efficient plant operations, are currently based on processing models that only consider bulk chemical and physical properties. They do not incorporate particle-level data – a significant limitation when dealing with complex bulk materials. This contribution presents a novel particle-based prediction model capable of dealing with complete particle datasets (i.e. no dimensionality reduction required), of operating without human-input and able to provide the probability of each particle in a system to deport to any one of the material streams within a given operation. The method is applicable to any processing unit that does not modify the physical dimensions of particles, such as comminution.
The particle-based prediction model consists of a regularized logistic regression model with a probability adjustment step to accommodate geological variability. Even though the method supports different types of particle-level characterization data, it is built around data obtained by scanning electron microscope-based image analysis. Constructed cases demonstrate the method’s efficiency in recreating characteristic recovery trends for magnetic separation, hydrocyclone and flotation units. In addition, the system is used to reconstruct a complete processing plant with three flotation and one magnetic separation circuits. Predicted results of masses and compositions for all of the intermediates and products correspond well to the results reported from the plant itself. The provided probabilities allow for the modelling of the interaction between particle properties and machine parameters, and can later be used for process simulation and optimization.