Exploring the fresh and rheology properties of 3D printed concrete with fiber reinforced composites (3DP-FRC): a novel approach using machine learning techniques

Abstract This study focuses on the prediction models for four parameters related to the fresh and rheological properties of 3DP-FRC: spreading diameters (S PD ), dynamic yield stress (DYs), static yield stress (SYs) and plastic viscosity (PV), respectively. Five machine learning (ML) algorithms were employed, namely artificial neural network (ANN), random forest (RF), decision tree (DT), categorical boosting (CatBoost), and extreme gradient boosting (XGBoost). An extensive dataset was compiled including 373 (S PD ) and 219 (SYs, DYs, PV) from various literature comprising experimental results. Fifteen input parameters were identified as the most influential factors affecting the fresh and rheological properties. These parameters include OPC, W/B, W/S, FA, LP, SF, SP, VMA, W, h f , R i , AR, t sf , F t , and S time /R time . This study found strong correlations between the developed ML models and the experimental outcomes from both the training and testing datasets. The models demonstrated exceptional accuracy and provided precise predictions for S PD , SYs, DYs, and PV. The correlation coefficients (R 2 ) ranged from 0.94 to 0.99 for S PD , 0.93 to 0.99 for SYs, 0.98 to 0.99 for DYs, and 0.98 to 1.00 for PV, with consistent results observed across both the training and testing datasets. Moreover, the model’s precision was assessed using different error metrics, including root mean square error (RMSE), mean square error (MSE), coefficient of variation in root-mean-square error (CVRMSE), and mean absolute error (MAE). Sensitivity analysis was performed to identify their impact. Additionally, fiber dependent analysis was conducted to assess the effectiveness of different fiber types on the fresh and rheological properties (S PD , SYs, DYs, and PV). In conclusion, the ML models were effectively trained and optimized, resulting in accurate and highly predictive capabilities for the parameters of interest.