Machine Learning and Experimental Identification of Parameters for Discrete Element Method Simulations: Binary Mixtures of Pellets and Lump Ores

Increasing the proportion of lump ore in ironmaking reactors can contribute to enhancing economic efficiency and reducing carbon emissions. Reliable discrete element method (DEM) simulations of such systems require accurate particle‐shape representation and efficient calibration of input parameters. In this study, an integrated framework for DEM parameter calibration that combines experimental repose angle measurements, 3D scanning with multisphere modeling, and a genetic algorithm–optimized random forest (GARF) model is proposed. GARF is embedded in the calibration workflow to efficiently explore the high‐dimensional parameter space and predict contact parameters with reduced computational demand. Experimental results are used as benchmarks to validate the calibrated parameters, and strong agreement is observed between GARF‐assisted DEM simulations and measurements. The calibrated contact parameters provide a robust reference for simulating irregular particle mixtures, while the proposed GARF‐based workflow enhances the efficiency and applicability of DEM parameter identification in ironmaking research.