Data-driven compressive strength estimation and mix optimization for ultra-high-performance concrete

Abstract

Ultra-high-performance concrete (UHPC) has found wide application in key infrastructure projects, including skyscrapers, bridges and subsurface facilities, owing to its outstanding mechanical strength and long-term durability. Its broader adoption, however, is limited by performance variability, high production costs and substantial carbon emissions. To overcome these issues, this research develops a unified framework that integrates Bayesian model updating, life cycle assessment (LCA) and ant colony optimization (ACO) to predict UHPC's mechanical and sustainability performance and optimize mix proportions. Using 102 sets of compressive test data, a physically interpretable and highly accurate strength prediction model was developed, identifying the water-to-binder ratio, silica-to-binder ratio, sand-to-binder ratio, fiber volume fraction and superplasticizer dosage as key factors. An LCA model then quantified how these variables influence the carbon footprint and cost of strength-equivalent UHPC. The ACO algorithm enabled both bi- and tri-objective mix optimization. Key findings include: (i) Reducing water-to-binder and sand-to-binder ratios while increasing silica fume, fiber and superplasticizer contents enhances strength, with the first two factors having the greatest impact. (ii) The Bayesian model surpasses traditional linear models in accuracy, robustness and interpretability. (iii) At equal strength, increasing sand-to-binder lowers cost and emissions; altering silica fume–to–binder ratio slightly affects cost while notably reducing emissions; higher fiber or superplasticizer increases both. (iv) The framework achieves up to 50 % reductions in carbon footprint and cost, and minimizes fiber use for compressive strengths below 185 MPa.