Insights into the early compressive strength of UHPC using experimental, connection weight, and ANN methods

This study aims to explore the first-day compressive strength of ultra-high-performance concrete (UHPC), a crucial characteristic for its potentially diverse applications, by employing artificial neural networks (ANNs). In this research, a range of mineral additives to formulate UHPC, such as silica fume (SF), ground granulated blast furnace slag (GGBFS), metakaolin (MK), fly ash (FA), calcium carbonate (CaC), rice husk ash (RHA), fluid catalytic cracking catalyst residue (FC3R), quartz powder (QP), and glass waste powder (GWP) are investigated. Drawing from robust literature-based data containing 604 sets, we developed a precise ANN predictive model. The model's efficacy was appraised by comparing its regression outcomes with experimental findings from 90 UHPC samples, which served as benchmarks for validation. Here, we gauged the ANN regression's precision using various statistical metrics, notably the coefficient of determination (R2). The results underscore the efficacy of the ANN methodology in estimating UHPC's 1-day compressive strength, showing R2 of 0.912 and 0.898 for the test and validation groups, respectively, outperforming previous ANN models with a unique hidden layer. Furthermore, the Connection-Weight-Approach (CWA) technique was employed to explore the interaction between UHPC ingredients and early-age compressive strength. It was concluded that specific mineral additives, notably MK, SF, FC3R, and CaC, and features like particle packing density positively influenced UHPC's initial strength. However, the introduction of FA, GP, and RHA adversely affected its strength in the early stages.