Revolutionizing 3D concrete printing: Leveraging RF model for precise printability and rheological prediction

In this paper, a general theoretical framework based on the random forest (RF) algorithm used for predicting the 3D printing concrete rheological properties and printability (3DPCRP) is proposed for the first time, which can avoid the subjective empirical dependence of earlier methods to control the stability of concrete printing. Specifically, the developed prediction models are categorized into two major types, namely rheological properties and printability prediction models. For the rheological properties prediction models, the input parameters include ordinary portland cement (OPC), sulfate aluminate cement (SAC), silica fume (SF), fly ash (FA), sand (S), maximum sand particle size (MAXSS), thixotropic agent (TA), early strength agent (ESA), superplasticizer/binder (SP/B), and water/binder (W/B). The printability prediction models take input parameters such as resting time (RT), DYS, SYS, PV, printing nozzle (PN), extrusion speed (ES), printing speed (PS), printing layer height (LH), and printing layer width (LW). The results of the statistical check index evaluation and shapley additive explanations (SHAP) analysis show that they all have high R2 (0.84–0.99) and low remaining statistical errors. This proves that the models developed in the study can successfully predict 3DPCRP. They can assist researchers in reliably and efficiently predicting the printability of concrete, thereby improving the likelihood of successful printing, print quality, and printing process stability.