Machine learning-driven multi-objective optimization of sustainable engineered cementitious composites: Balancing...

This study addresses the dual challenge of enhancing the mechanical performance and environmental sustainability of Engineered Cementitious Composites (ECC) by replacing natural sand with Municipal Solid Waste Incineration Bottom Ash (MSWIBA). Fifteen ECC mixtures were experimentally tested across varying W/B ratios (0.24-0.30) and BA replacement levels (0-100%). Results showed BA inclusion decreased flowability and strength, with compressive strength dropping from 74.1 MPa (control) to 32.7 MPa (full replacement at W/B = 0.24). Optimal experimental performance, characterized by robust strain-hardening (ϵ_u>2.5%), occurred at W/B = 0.27 with 25%-50% BA. Environmentally, BA reduced CO₂ emissions per cubic meter from 0.9665 kg/m³ to 0.8788 kg/m³. To guide sustainable mix design within this experimentally investigated 15-mixture design space, a Machine Learning (ML)-driven Multi-Objective Optimization (MOO) framework was established using the experimental dataset as an in-domain database. A Support Vector Regression (SVR) model (validation R = 0.95 for fc) was used as a surrogate predictor, and the Ant Colony Optimization (ACO) algorithm maximized a composite objective derived from the Entropy Weight Method (EWM), balancing compressive strength, tensile strength, and CO₂/f_c (weights: wenv = 0.3364, w_fc = 0.3286, w_ft = 0.3350). The framework identified a recommended in-domain mix at W/B = 0.2526 and BA = 20.00%, with predicted fc = 61.09 MPa, ft = 5.56 MPa, and CO₂/f_c = 0.0154 kg/(m³⋅MPa). This optimized point should be interpreted as an interpolation-based design recommendation within the investigated range rather than as a universally applicable predictive result.