Abstract

Modeling of flotation processes is complex due to the large number of variables involved and the lack of knowledge on the impact of operational parameters on the response(s), and given this problem, machine learning algorithms emerge as an alternative interesting when modeling dynamic processes. In this work, different artificial neural network (ANN) architectures for modeling the mineral concentrate in a rougher-cleaner-scavenger (RCS) circuit based on the main process variables are generated (variables as the recovery of the rougher, cleaner and scavenger cells, along with disaggregated variables). Analysis of the global sensitivity was performed to study the importance of the individual and joint performances of the stages of the flotation circuit, reflected by sensitivity indicators that allow to infer the impact that the stages and operational parameters produce on the dependent variables (mineral concentrate in rougher, cleaner and scavenger cells, in addition to the global concentration in the RCS circuit). It should be noted that the ANN is a useful tool for modeling dynamic systems such as flotation, while sensitivity analysis shows that the operation of the three threads turns out to be crucial for the subsequent evaluation of the circuit, while the Unbundled variables that most interact with the overall recovery are gas flow rate, bubble and particle diameters, bubble velocity, particle density, and surface tension.