Stochastic analysis of 3D concrete printing process with curvature and inclination by explainable data-driven modelling

Abstract

The increasing adoption of 3D concrete printing (3DCP) in construction highlights the importance of understanding the stochastic behavior of the printing process to ensure quality control. This study proposes an explainable data-driven stochastic analysis framework, incorporating SHapley Additive exPlanation (SHAP), to evaluate and explain the impact of material uncertainty on the printing process for walls with curvature and inclination. Among seven machine learning algorithms examined, the Sparse Gaussian Process Regression model demonstrated superior predictive performance and was selected for data-driven modeling. SHAP-based analysis identified the degree of inclination and initial cohesion as the most critical factors influencing the printing process, surpassing other material, geometric, and printing features in importance. Stochastic analysis revealed that increasing the degree of inclination reduces both the buildability of the 3DCP process and associated uncertainty, while geometric curvature enhances buildability but increases its variation. Printing configurations, such as print speed and layer height, had negligible effects on buildability and uncertainty within small-scale geometries. Regardless of printing geometry and configurations, initial cohesion was identified as the most influential contributor to process uncertainty, making it a key focus for optimization to reduce variability and enhance reliability in 3DCP processes.