Industrial-scale prediction of cement clinker phases using machine learning

Abstract

Cement production exceeds 4.1 billion tonnes annually, emitting 2.4 billion tonnes of CO2 annually, necessitating improved process control. Traditional models, limited to steady-state conditions, lack predictive accuracy for clinker mineralogical phases. Here, using a comprehensive two-year industrial dataset, we develop machine learning models that outperform conventional Bogue equations with mean absolute percentage errors of 1.24%, 6.77%, and 2.53% for alite, belite, and ferrite prediction respectively, compared to 7.79%, 22.68%, and 24.54% for Bogue calculations. Our models remain robust under varying operations and are evaluated for uncertainty and rare-event scenarios. Through post hoc explainable algorithms, we interpret the hierarchical relationships between clinker oxides and phase formation, providing insights into the functioning of an otherwise black-box model. The framework can potentially enable real-time optimization of cement production, thereby providing a route toward reducing material waste and ensuring quality while reducing the associated emissions under real-world conditions.