Mineral Processing
Meysam Nikfarjam; Ardeshir Hezarkhani; Farhad Azizafshari; Hamidreza Golchin
Abstract
Geometallurgical modeling (GM) plays a crucial role in the mining industry, enabling a comprehensive understanding of the complex relationship between geological and metallurgical factors. This study focuses on evaluating metallurgical varibles at the Sungun Copper mine in Iran by measuring and predicting ...
Read More
Geometallurgical modeling (GM) plays a crucial role in the mining industry, enabling a comprehensive understanding of the complex relationship between geological and metallurgical factors. This study focuses on evaluating metallurgical varibles at the Sungun Copper mine in Iran by measuring and predicting process properties, including semi-autogenous power index (SPI), recovery (Re), and concentration grade. To overcome the additivity limitations of geostatistical methods, we utilized machine learning algorithms for enhanced predictive modeling, aiming to improve decision-making and optimize mining operations in geometallurgy. The research incorporates crucial data inputs such as sample coordinates, grades, lithology, mineralization zones, and alteration to assess the accuracy and reliability of different machine learning regression methods. The Relative Standard Deviation (RSD) is highlighted as a significant metric for comparing the accuracy of predicted process properties. Evaluation metrics such as Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and coefficient of determination (R2) further confirm the superiority of specific modeling methods in certain scenarios. The K-Nearest Neighbors (KNN) method exhibits superior accuracy, lower error metrics (RMSE and MAE), and a higher R2 for modeling the SPI test. For modeling Cu grade in concentrate, Support Vector Regression (SVR) proves to be effective and reliable, outperforming the Multilayer Perceptron (MLP) method. Despite MLP's high R2, its higher RSD suggests increased uncertainty and variability in the predictions. Therefore, SVR is considered more suitable for modeling Cu grade in concentrate. Findings optimize operations at Sungun Copper mine, improving decision-making, efficiency, and profitability.
Exploration
Ahmadreza Erfan; Saeed Soltani Mohammad; Maliheh Abbaszadeh
Abstract
Machine learning (ML) has significantly transformed multiple disciplines, including mineral resource evaluation in mining engineering, by facilitating more accurate and efficient estimation methods. Ensemble methods, as a fundamental component of modern machine learning, have emerged as powerful ...
Read More
Machine learning (ML) has significantly transformed multiple disciplines, including mineral resource evaluation in mining engineering, by facilitating more accurate and efficient estimation methods. Ensemble methods, as a fundamental component of modern machine learning, have emerged as powerful tools that robust techniques that integrate multiple predictive models to improve performance beyond that of any individual learner. This study proposes a novel ensemble method for estimating ore grades by localizing the base learner weights in ensemble method. Ordinary kriging, inverse distance weighting, k-nearest neighbors, support vector regression, and artificial neural networks have been used as the base learners of the algorithm. In ML base learners, coordinates (easting, northing and elevation) of samples have been defined as input nodes and grade has been defined as target. The proposed method has been validated for predicting the copper grade (Cu%) in Darehzar porphyry deposit. The performance of proposed method has been by individual base learners and famous ensemble methods. This comparison shows that performance of proposed method is better than other ones. The findings highlight the necessity of adapting ensemble methods to address spatial variability in geological data, thereby establishing a robust framework for ore grade estimation.
