Experimental and AI-based modeling of C-25 concrete with partial cement replacement using fly ash and marble powder

This study explores the combined use of fly ash (FA) and marble powder (MP) as a binary supplementary cementitious blend for C-25 concrete, aiming to reduce clinker content without sacrificing performance. The mixes used FA–MP (1:1 by mass) to replace 0–35% of OPC, with a constant water-to-binder ratio of 0.50. Material characterization using X-ray diffraction (XRD) and scanning electron microscopy (SEM) confirmed the presence of reactive quartz-mullite phases and spherical cenospheres in FA, and predominantly calcite with angular, fine particles in MP, supporting a pozzolanic-filler synergy. The fresh and hardened properties, including slump, setting time, compressive strength (7, 14, and 28 days), strength activity index (SAI), water absorption (WA), and dry density (DD), were measured and complemented with AI/ML predictions using DNN, ELM, RF, XGB, and SVR. The results indicate an optimal replacement at 25% (SUB25), achieving a 110 mm slump, 34 MPa 28-day strength, SAI = 109.9%, WA = 1.12% (≈ 66% lower than that of the control), and DD = 2420 kg/m³. SEM of hardened mixes revealed a dense, well-interlocked C–S–H matrix at SUB25, whereas 35% replacement reintroduced micro-voids. Among the models, Random Forest yielded the most accurate predictions (R² = 0.99, MAE = 0.20 MPa, RMSE = 0.28 MPa), followed by DNN, with SVR performing the poorest. The findings demonstrate that 25% FA–MP substitution delivers balanced workability, strength, and durability while reducing cement use. Coupling microstructural evidence with robust AI/ML modeling offers a practical, data-driven pathway for optimizing sustainable concrete mixes and reducing the environmental footprint of construction.