Machine learning applied to predicting phase assemblages of hardened cementitious systems

Recently, the use of supplementary cementitious materials (SCMs) as a partial replacement of Portland cement has become necessary to minimize CO₂ emissions associated with concrete, while the wide variations in their types and properties make predictions of the final products difficult, requiring a significant experimental effort. In this research, machine learning (ML) algorithms were explored to predict the major phases of Portland cement blended with SCMs. An artificial neural network ML algorithm was used for prediction in order to provide adaptable input that can be utilized for various SCMs. The application of the model was validated by forming and analyzing the relationship between the phases, water-to-cement ratio, and main oxide compositions of SCMs. The results demonstrated that ML models can be used effectively to predict phases of SCM-blended cements. The outcome of this study is expected to provide valuable information for concrete mix designs involving the use of SCMs.