Analysis of waste glass as a partial substitute for coarse aggregate in self-compacting concrete: An experimental and machine learning study

With the growing environmental burden to minimize debris and recycle, the concrete industry has replaced wasted glass with concrete composition elements in multiple ways. This study provides a comprehensive analysis of integrating waste glass aggregate (WGA) at various proportions (0 %, 5 %, 10 %, 15 %, and 20 %) as substitutes for coarse aggregates along with a consistent 20 % silica fume (SF) as cement replacement in self-compacting concrete (SCC). In addition to determining the mechanical, durability and microstructural properties of SCC, this research also uses advanced machine learning (ML) techniques to predict the mechanical properties accurately. By employing slump flow, j-ring flow and v-funnel time the rheology of fresh SCC was evaluated. Mechanical properties were assessed through the evaluation of compressive, splitting tensile, and flexural strength, whereas electrical resistivity, water permeability, rapid chloride permeability (RCPT), and accelerated mortar bar test (AMBT) were employed to evaluate durability. The microstructure was further examined using scanning electron microscopic (SEM) and energy-dispersive X-ray spectroscopy (EDS) analysis. The replacement of WGA with a constant SF in SCC improved flowability while lessening its passing ability. Furthermore, the study indicated that adding 5 % WGA along with 20 % SF to SCC resulted in a slight decrease of 4.53 % in compressive strength and 3.61 % in flexural strength compared to the reference mix containing 0 % WGA and 20 % SF after 28 days, which yields a favorable outcome. Findings from further assessments revealed that SCC's durability characteristics were enhanced by adding WGA and SF. Expansion of the Ca/Si ratio observed in EDS analysis significantly impacts the strength, while voids and cracks patterns detected in SEM analysis, result in an inferior microstructural performance with the incorporation of WGA. Notwithstanding both the ML models exhibiting excellent regression coefficients (R²), the random forest model showed superior accuracy and reliability in its results.