Por: Luis Vargas Jeri, Ardiles Puma Qquenta y Malmco Camborda Morocho, Minsur.

Abstract

This study describes the development, implementation, and evaluation of LingoSmelter, a statistical model based on Machine Learning (ML) techniques aimed at optimizing crude tin metal recovery in the Ausmelt furnace at the Minsur smelting plant.

Traditionally, the prediction of crude metal recovery has been based on a theoretical model supported by mass and energy balances, which, although functional, presents limitations when facing operational variability and the complexity of process data.

LingoSmelter was developed using an advanced analytics workflow that integrates supervised learning algorithms (XGBoost) with evolutionary optimization methods (Differential Evolution Solver). The model was trained using historical operational data from the Ausmelt furnace, including variables such as feed composition, coal and natural gas consumption, temperature, among others, as well as the actual crude metal recovery per batch. Subsequently, an optimization process was applied to identify combinations of operational parameters that maximize the predicted performance within the technical and safety limits of the process.

Model validation was conducted using standard regression evaluation metrics (MAE, MAPE, and RMSE), comparing its performance against the theoretical model. The results show a significant improvement in the predictive accuracy of the statistical model, enabling better estimation of crude metal recovery and more informed decision-making in furnace operation. This improvement translates into greater efficiency in the use of inputs, reduced operating costs, and a potential increase in crude metal recovery of 1.78%, representing an estimated benefit of USD 1.3 million (Net Present Value) evaluated over a six-year horizon.

LingoSmelter represents a clear example of the application of Mining 4.0 technologies, integrating data science, metallurgical expertise, and a culture of continuous improvement to generate sustainable value in complex metallurgical operations.