Development and evaluation of a soft sensor with AI and machine learning to predict P80 in milling

Palma-Villanueva R.M. · Boletin de Ciencias de la Tierra (2026) · DOI: 10.15446/rbct.n59.121702

Resumen

This study developed a soft sensor model to predict the P80 value in the semi-autogenous grinding (SAG) circuit of the copper concentrator plant at Minera Las Bambas S.A., in Peru. P80 is a critical parameter for process efficiency, but its direct measurement faces instrumental limitations. Thirteen machine learning algorithms were trained using operational data from four cyclone batteries. The optimized CatBoost model showed the best performance, with R² values above 0.80 and low dispersion between predicted and actual values. The most relevant variables for prediction were specific energy consumption, throughput, and solids percentage. This Softsensor model can replace the particle size analyzer in case of maintenance or other events, ensuring operational continuity. © The author; licensee Universidad Nacional de Colombia.

Instituciones: Universidad Nacional de San Agustín de Arequipa