A Soft Sensor for Inferring Energy Efficiency of Wet-Closed Ball Mills

Abstract The comminution process accounts for most of the energy consumption in ore beneficiation plants. Measuring energy consumption is crucial to ensure efficient control and meet granulometric specifications with low energy consumption. However, this measurement is not straightforward and requires specialized equipment to quantify product retention at a specific mesh size. This process is labor-intensive, expensive, and requires significant maintenance. In this work, a new approach using a soft sensor strategy is proposed to infer the energy efficiency of a wet-closed ball mill. This tool can aid the operation teams in decision-making. A case study was conducted using real data from one of the world’s largest mining companies, VALE SA, evaluating multiple modeling approaches and validation methodologies. While initial experiments with a multilayered perceptron trained with the Levenberg–Marquardt algorithm showed promising results under standard validation, we demonstrate that proper chronological validation for time-series data is essential for a more realistic performance assessment. The final model, an ensemble combining multiple predictors, achieved a correlation coefficient (R) of 68.79% in rigorous time-series validation, providing a methodologically sound solution for energy efficiency inference in industrial applications.