Machine Learning Prediction of High‐Strain‐Rate Compressive Properties of Rock‐Filled and Conventional Vibrated Concrete

Abstract

Rock‐filled concrete (RFC) offers material and environmental benefits for mass‐concrete structures, but its heterogeneous composition complicates prediction of its high‐strain‐rate compressive response. This study developed a machine learning framework to predict the strain rate, dynamic compressive strength, and dynamic increase factor (DIF) of RFC and conventional vibrated concrete subjected to Split Hopkinson Pressure Bar loading. The database contained 288 records representing 48 experimental conditions involving different material categories, strength grades, specimen‐preparation methods, air pressures, and striker velocities. Five regression algorithms, Elastic Net, Decision Tree, Random Forest, Support Vector Regression, and Gradient‐Boosted Regression Trees, were evaluated using the coefficient of determination (R2) and root‐mean‐square error (RMSE). Among the evaluated models, Random Forest achieved the highest accuracy for strain rate (R2 = 0.961; RMSE = 8.211 s−1) and DIF (R2 = 0.836; RMSE = 0.331), whereas Support Vector Regression performed best for dynamic compressive strength (R2 = 0.831; RMSE = 8.436 MPa). Correlation analysis indicated that strain rate was governed mainly by impact‐loading parameters, while dynamic strength and DIF reflected combined material and loading effects. The framework provides an efficient preliminary assessment tool within the investigated experimental domain.