The Mill Family Model: A decision-support framework for tailored grinding media selection in semi-autogenous grinding and overflow ball mills

• Mill Family Model was developed using K Means clustering on industrial mill data. • Mills were grouped by ore, design and operation – factors influencing media wear. • Five clusters were identified for SAG and overflow ball mills, respectively. • Model supports grinding media selection based on wear-driven environment clusters. • Model acts as a decision-support tool to optimize grinding media selection process. Grinding media selection plays a critical role in the performance, cost-efficiency, and sustainability of semi-autogenous grinding (SAG) and overflow ball mill operations. However, variability in ore characteristics, mill design, and operating conditions complicates the identification of optimal media. This study introduces a data-driven framework, termed the Mill Family Model, to classify SAG and overflow ball mills into operational clusters that share similar grinding environments. Using K Means clustering on industrial datasets, mills were grouped based on ore type, abrasiveness, mill size, speed, feed size, mill rate, charge levels, and slurry pH – parameters known to influence grinding media wear mechanisms. Principal Component Analysis (PCA) was applied to visualize cluster separations and highlight contributing variables. The analysis identified five distinct clusters for each mill type, each exhibiting different combinations of mechanical and chemical conditions. These clusters reflect unique wear environments, suggesting that different grinding media, defined by chemistry, microstructure, size, hardness and toughness, may be better suited to specific clusters. The Mill Family Model is not intended to replace wear testing but rather to complement it by standardizing early-stage decision-making and guiding grinding media selection based on inferred dominant wear mechanisms, namely impact, abrasion, and corrosion. It provides a scalable, repeatable framework for aligning media strategies with specific mill environments, supporting both operational decision-makers and consumable suppliers. Future work will extend the dataset scope and incorporate grinding media properties and wear metrics to improve the model’s predictive capabilities and broader applicability in optimizing consumable development and selection.