Machine learning-driven prediction of mechanical properties for 3D printed concrete

Mechanical performance is crucial for 3D printed concrete, as it directly influences the structural load-bearing capacity and safety. However, the inherent complexities and variability of the material pose significant challenges in achieving accurate predictions of its mechanical performance. This study introduces a novel approach to predict the compressive strength (CS) and flexural strength (FS) of 3D printed concrete using machine learning (ML) methods. A comprehensive database containing 254 CS tests and 210 FS tests was established to train four ML models: artificial neural networks, random forest, extreme gradient boosting, and categorical boosting algorithms (CatBoost). The CatBoost model demonstrated superior performance, with R2 values of 0.929 for CS and 0.967 for FS on the test set. To provide insights into the model's predictions, partial dependence plots and Shapley Additive Explanations were employed, revealing that the water-to-binder ratio (n(W/B)) and the content of ordinary Portland cement are critical factors influencing CS, while n(W/B) and the content of ground granulated blast furnace slag significantly affect FS. This innovative ML-driven approach offers a robust framework for accurately predicting the mechanical properties of 3D printed concrete, thereby enhancing its application in structural engineering.