Ai-guided proportioning and evaluating of self-compacting concrete based on rheological approach

Abstract

Self-Compacting Concrete (SCC) has gained significant popularity due to its exceptional workability performance. However, designing SCC poses more challenges than ordinary concrete, as it must fulfill requirements for filling ability, passing ability, and segregation resistance. Regrettably, existing test techniques lack the ability to simultaneously evaluate all these properties, and relying on experiential knowledge and cognition without tangible physical meaning. Although rheology examines the flow and deformation of fluid materials and is closely related to SCC’s properties, the relationship between SCC composition, rheology, and properties remains unclear due to the complexity of the factors involved and the absence of effective tools. This study introduces a novel approach by utilizing a random forest algorithm to create multiple interpretable machine learning models for predicting the rheology, workability, and mechanical properties of SCC. Additionally, SHapley Additive exPlanation (SHAP) and Partial Dependence Plot (PDP) methods were integrated with the models to analyze how SCC composition impacts its properties by altering the rheology of the mixture. The models exhibit high accuracy in predicting both rheology and SCC properties (R2 = 0.93 ∼ 0.98, Index of Agreement = 0.92 ∼ 0.99). According to the SHAP and PDP analysis, yield stress and plastic viscosity were negatively correlated with slump flow, L-box ratio, and segregation rate, while exhibiting a positive correlation with V-funnel time and strength. Furthermore, the dependence of the segregation rate on yield stress was observed to be stronger at relatively low yield stress levels (below 40 Pa). These models provide valuable insights for designing and evaluating SCC mixtures tailored to specific requirements. Additionally, the study explores underlying mechanisms and offers guidelines for proportioning SCC in different design scenarios. The findings contribute to the advancement of SCC technology and have significant implications for the construction industry.