AI-driven critical parameter optimization of sustainable self-compacting geopolymer concrete

Abstract

This study aims to optimize the critical parameters of self-compacting geopolymer concrete (SCGC) with a primary focus on enhancing compressive strength. Optimization of SCGC holds significant importance within the context of sustainable development. To implement the research, a comprehensive database was meticulously compiled by integrating data from published literature, with Fly ash and GGBFS serving as essential binders. Prior to the development of a robust machine learning model, the database underwent careful preprocessing to ensure data quality. In pursuit of optimal performance, three machine learning model's (base model) hyperparameters were diligently fine-tuned employing effective spotted hyena metaheuristic optimization technique. A hybrid model was developed using three base models and a meta-learner. The hybrid model outperformed other models owing to better accuracy and generalizability. Leveraging quasi-Monte Carlo sensitivity analysis, this study effectively identified temperature, Na2SiO3/NaOH, NaOH molarity, and percentage of superplasticizer as critical parameters with the highest impact on compressive strength. Optimization involved adjusting the temperature (25–90 °C), Na2SiO3/NaOH concentration (8–16 M), molarity (1.5–4), and superplasticizer amount (2–10%). The optimization of these crucial parameters was effectively executed using Response Surface Methodology (RSM). This study introduces a novel integrated approach within RSM, deviating from conventional quadratic or polynomial models for interaction characterization. The optimized parameters were consistent across cases involving fly ash, GGBFS, and a blend of both binders, except for temperature, which varied based on the binder type. The optimized mix parameters are used for experimental validation of results obtained through AI-driven analysis and observed to provide an expected result of compressive strength in the range of 39–43 MPa.