Development and optimization of geopolymer concrete with compressive strength prediction using particle swarm-optimized extreme gradient boosting

Abstract

This study presents an experimental investigation of the workability and strength behavior of eco-friendly binary geopolymer concrete (GPC) containing ground granulated blast furnace slag (GGBS) and sugarcane bagasse ash (SBA). Initially, this study optimizes binder content (GGBS) in GGBS-based GPC by utilizing the Taguchi method by examining various variables, including alkaline activator-to-binder ratio (AAS/B), sodium hydroxide molarity (M), sodium silicate-to-sodium hydroxide ratio (SS/SH). After optimizing the binder (GGBS) content, SBA is incorporated to formulate a binary GPC by partially replacing GGBS with SBA at substitution levels of 0 %, 5 %, 10 %, 15 %, and 20 %. The effects of varying the AAS/B, M, and SS/SH ratios at different SBA additions on the workability and compressive strength of binary GPC were analyzed. Split and flexural strength were tested on optimized binary GPC samples with varying SBA replacement levels. Moreover, machine learning prediction models using the XGBoost and hyperparameter-optimized XGBoost using particle swarm optimization (PSO) were developed to predict the 28th day compressive strength of binary GPC. Finally, the cost efficiency of binary GPC at different SBA replacement levels was determined. The experimental findings demonstrate that an AAS/B ratio of 0.6, an SS/SH ratio of 2.5, and a sodium hydroxide molarity of 12 M provide an optimal balance between workability and compressive strength. Furthermore, the binary GPC incorporating GGBS and SBA demonstrated compressive strengths ranging from 57 to 79 MPa after curing for 28 days at ambient temperature. This study suggests that 15 % SBA was the optimal replacement level in GGBS-based GPC without significantly compromising its mechanical properties. The prediction outcomes demonstrate that the PSO-XGBoost model is highly effective in predicting binary GPC compressive strength, with an R² value of 0.97. According to the SHAP (Shapley Additive exPlanations) study, the compressive strength of binary GPC was substantially impacted by the quantities of GGBS and SBA.