Application of neural networks to predict the quality of iron ore concentrate based on flotation data

¹ Bauman Moscow State Technical University, 105005 Moscow, Russia
² Reshetnev Siberian State University of Science and Technology, 660037 Krasnoyarsk, Russia
³ Russian State Agrarian University - Timiryazev Moscow Agricultural Academy (RSAU-MAA Named after K.A. Timiryazev), 127434 Moscow, Russia

^* Corresponding author: sofaglu2000@mail.ru

Abstract

This paper presents a study aimed at developing and testing a neural network model for predicting the percentage of silica in iron ore concentrate obtained during flotation. The problem of precise control of the silica content is critical for the mining industry, since the quality of the final product and, accordingly, its market value depend on it. During the study, data was collected from the flotation plant, their preliminary processing was carried out, including standardization and elimination of missing values. The developed neural network model included two hidden layers and was trained on real data. The evaluation of the model quality showed high results, which was confirmed by the metrics of mean square error (MSE), mean absolute error (MAE) and coefficient of determination (R²). Additionally, an analysis of the visualizations of the residuals and predicted values confirmed the accuracy and stability of the model. The results of the study demonstrate that the proposed model can be effectively used in production conditions to improve process control and improve product quality in the mining industry.