Prediction of multi-stage froth flotation efficiency of complex lead–zinc sulfide ore using an integrated ensemble neural network–random forest model
Kwanghui Jo,Jinyoung Je,Donwoo Lee,Heechan Cho,Kwanho Kim,Kwangsuk You
DOI: https://doi.org/10.1016/j.mineng.2024.108669
IF: 5.479
2024-03-29
Minerals Engineering
Abstract:Despite the long history of flotation in the mineral processing industry, its prediction and understanding remains a great challenge owing to its many variables acting in a complex manner. In this study, we introduced machine learning (ML) models to predict the grade and yield of a multi-stage flotation process of a complex lead–zinc sulfide ore. Over 100 batch flotation tests were conducted in a stepwise manner to characterize different rougher-cleaner-scavenger configurations. Performing a pre-flotation of talc prior to sulfide flotation remarkably improved the grade of lead concentrate. The experimental data were divided into four subsets for the ML models: rougher without pre-flotation, rougher with pre-flotation, cleaner, and cleaner-scavenger datasets. Different ML models were evaluated to determine whether they could predict the lead grade, zinc grade, and yield related to key flotation parameters, including particle size, reagent dosage, and pulp pH. The integrated ensemble neural network and random forest model yielded the best prediction results with R 2 values of 0.924, 0.902, 0.973, and 0.894 for the rougher without pre-flotation, rougher with pre-flotation, cleaner, and cleaner-scavenger subsets, respectively. The developed ML model, with the connection of subset models, effectively predicted the flotation outcome of the rougher-cleaner-scavenger circuit, demonstrating a better prediction performance than previous methods. This indicates that the developed ML model can potentially predict flotation process performance and evaluate the efficiency of newly designed multi-stage froth flotation processes.
engineering, chemical,mining & mineral processing,mineralogy