Prediction of the R3 Test-Based Reactivity of Supplementary Cementitious Materials: A Machine Learning Approach Utilizing Physical and Chemical Properties

Abstract

This study utilized machine learning (ML) models to investigate the effect of physical and chemical properties on the reactivity of various supplementary cementitious materials (SCMs). Six SCMs, including ground granulated blast furnace slag (GGBFS), pulverized coal fly ash (FA), and ground bottom ash (BA), underwent thorough material characterization and reactivity tests, incorporating the modified strength activity index (ASTM C311) and the R3 (ASTM C1897) tests. A data set comprising 46 entries, derived from both experimental results and literature sources, was employed to train ML models, specifically artificial neural network (ANN), support vector machine (SVM), and random forest (RF). The results demonstrated the robustness of the ANN model, achieving superior prediction accuracy with a testing mean absolute error (MAE) of 9.6%, outperforming SVM and RF models. The study classified SCMs into reactivity classes based on correlation analysis, establishes a comprehensive database linking material properties to reactivity, and identifies key input parameters for predictive modeling. While most SCMs exhibited consistent predictions across types, GGBFS displayed significant variations, prompting a recommendation for the inclusion of additional input parameters, such as fineness, to enhance predictive accuracy. This research provided valuable insights into predicting SCM reactivity, emphasizing the potential of ML models for informed material selection and optimization in concrete applications.