Integrated experimental, machine learning, and life-cycle assessment of fly ash-silica fume based self-compacting geopolymer concrete.

This study investigates the performance of self-compacting geopolymer concrete (SCGC) incorporating fly ash (FA) and silica fume (SF) as aluminosilicate precursors, with emphasis on mechanical behaviour, durability, microstructural characteristics, environmental impact, and machine-learning-based performance prediction. Four geopolymer SCC mixes were prepared using SF to substitute FA at 0, 5, 10, and 15% by weight and an OPC-based SCC of the same class of strength was prepared to serve as a benchmark. Fresh properties were assessed as per EFNARC standards after which compressive, split tensile and flexure strength were done at intervals of 28, 90, and 180 days. Sorptivity, rapid chloride permeability (RCPT) and ultrasonic pulse velocity (UPV) were used to determine durability, whereas scanning electron microscopy (SEM) was used to measure the microstructural evolution. Findings reveal that SF is a considerable improvement to fresh and hardened geopolymer SCC, the best performance being found at a 10% replacement. The G10 mix recorded the best compressive strengths of 65.3 MPa at 180 days, which was 21% higher than the FA only geopolymer mix, and tensile and flexural strengths were 14-18% higher compared to FA only geopolymer mix. The performance of durability increased significantly, as sorptivity was reduced by about 21% and RCPT was lower than 1000 Coulombs (extremely low permeability) and the largest values of UPV were the highest, which shows the presence of a dense and uniform internal structure. Refined pore structure and well-developed the formation of aluminosilicate gels in the G10 mix had been confirmed by SEM observations. Parameters of mix, fresh properties, and curing age were used as inputs to develop machine learning models (KNN, SVM, Decision Tree, and Random Forest). Among them the predictive accuracy of the Random Forest model (R