Thesis Recomendaciones de geometrías para revestimientos de molinos en gran minería basadas en algoritmos de inteligencia artificial
Loading...
Date
2025-02
Journal Title
Journal ISSN
Volume Title
Program
Ingeniería Civil Informática
Campus
Campus Casa Central Valparaíso
Abstract
Esta memoria tuvo como propósito crear un modelo con técnicas de machine learning que permita generar geometrías de revestimientos de molinos SAG. Se realizó una manipulación de datos y características relevantes donde se consideró las proyecciones lineales y no lineales del desgaste esperado. El algoritmo propuesto esta basado en XGBoost y logró una puntuación en la métrica R2 de 0.828 y un error del 15 % en el RMSE en comparación con los datos reales de desgaste. En consecuencia, la geometría dibujada del perfil se baso en estos resultados.
El modelo permitirá a la empresa contar con una herramienta confiable al momento considerar una duración objetivo de la vida útil de los revestimientos, evitando así campañas de supervisión desfasadas o innecesarias. Ademas, esto llevará a ahorrar en costes operacionales relacionados con la detención de los molinos.
The purpose of this work was to create a model using machine learning techniques to generate SAG mill liner geometries. A manipulation of relevant data and features was performed where linear and nonlinear projections of expected wear were considered. The proposed algorithm is based on XGBoost and achieved an R2 metric score of 0.828 and an error of 15 % in the RMSE compared to the actual wear data. Consequently, the drawn profile geometry was based on these results. The model will allow the company to have a reliable tool when considering a target life of the liners, thus avoiding outdated or unnecessary monitoring campaigns. In addition, this will save on operational costs related to mill shutdowns.
The purpose of this work was to create a model using machine learning techniques to generate SAG mill liner geometries. A manipulation of relevant data and features was performed where linear and nonlinear projections of expected wear were considered. The proposed algorithm is based on XGBoost and achieved an R2 metric score of 0.828 and an error of 15 % in the RMSE compared to the actual wear data. Consequently, the drawn profile geometry was based on these results. The model will allow the company to have a reliable tool when considering a target life of the liners, thus avoiding outdated or unnecessary monitoring campaigns. In addition, this will save on operational costs related to mill shutdowns.
Description
Keywords
Industria minera, Aprendizaje automático, Molinos de bola, Procesamiento de datos