Development of printable concrete mix using ML models

Abstract

The technique of three-dimensional (3D) concrete printing (3DCP) has emerged as a highly advanced construction method, attracting considerable attention in recent times. Machine learning (ML) models are employed in this study to predict the most effective mix designs for 3D-printable concrete (3DPC). By analyzing various data inputs, these models help identify the optimal mix proportions that would ensure the best performance in 3D concrete printing. The approach also predicts the compressive strength (CS) of 3D printed materials. Five ML models are utilized in this study: Random forest (RF), Support vector machine (SVM), Extreme Gradient boosting (XGBoost), Decision Tree (DT), and Gradient boosting machine (GBM). The compressive strength data were collected for 150 research papers, and with a mix design of over 500. Hyper-parameter optimization techniques are used to optimize the parameters of the ML models. On CS, the accuracy is approximately𝑅2=0.82,0.81,0.80. The SHAP analysis detects that water/binder (W/B) ratio and ordinary Portland cement content are the most effective parameters for CS. The ML models incorporated with SHAP analysis disclose the relationship between the input variables and the mechanical performance of 3DCP. They could provide valuable information for the performance-based design of the mix proportion of 3DCP. The amount of clay and silica fumes also influences the properties, such as buildability. The VMA and SP are influential parameters in extrudability. In this paper, the compression test is performed on every direction of layers, i.e., perpendicular to printing direction, parallel to printing direction in X axis, and parallel to printing direction in Y axis. The compressive strength values for various loads provide the best loading direction plane inputs.