Automated machine learning modeling of synergistic effects in tailings–SCM mixtures for strength prediction and multi-objective design optimization

Researchers have blended mine tailings with supplementary cementitious materials (SCMs) to partially replace ordinary Portland cement (OPC), leveraging synergistic interactions to overcome the strength limitations of using tailings alone. However, excessive reliance on commercial SCMs can undermine the original goal of utilizing tailings as a waste resource to reduce cost and carbon emissions, making careful proportioning essential. Experimentally exploring all possible combinations to determine the best trade-offs among these multiple objectives is impractical. To address this, present study integrates an automated ML framework, AutoGluon tabular, with the non-dominated sorting genetic algorithm (NSGA) for efficient multi-objective optimization. AutoGluon selected the tabular prior data fitted network version 2 (TabPFNv2) as the best-performing model for UCS prediction. Evaluation on testing data further demonstrated its superior accuracy, producing the lowest prediction error of 2.022 MPa. Combining TabPFNv2 with NSGA, alongside other objective functions for cost and emissions, generated multiple mixture designs that outperformed experimental formulations. Finally, a practical software, SynergyCem, was introduced to bring sophisticated ML modeling and optimization techniques into real-world applications. Moreover, SynergyCem allows customization of design problem for different materials, and regions, making it a versatile tool for real-world use. Overall, this study establishes a practical, efficient pathway for designing tailings–SCM mixtures, promoting their adoption as economical and environmentally-friendly cementitious materials worldwide.