Answers are grounded in this paper and cite their sources.
UHPC offers exceptionally high compressive strength but carries a heavy embodied-carbon footprint due to its elevated cement content, motivating the use of SCMs, including fly ash, silica fume, and GGBFS as partial cement replacements. However, accurate prediction remains challenging due to complex nonlinear interactions among the mixture constituents, fibre properties, curing conditions, and specimen geometry. Although existing ML studies capture these coupled effects and provide deterministic predictions, they often overlook uncertainty and heteroscedastic variability, leaving their reliability untested. This study developed an uncertainty-aware ML framework, using a harmonized database of 2,032 low-carbon UHPC mixtures incorporating one or more of FA, SF, and GGBFS, to predict compressive strength (23.6–––298 MPa) across curing ages of 1 to 90 days. Among six evaluated ML models (RF, XGBoost, AdaBoost, LightGBM, CatBoost, and ANN), LightGBM achieved the highest accuracy with R 2 = 0.945 and RMSE = 9.41 MPa. Prediction uncertainty was quantified by integrating six conformal prediction methods with each model and evaluating 36 ML–CP combinations using EC, MIW, AIS, and ES. Among all combinations, LightGBM–Jackknife achieved the best uncertainty-aware performance, providing a strong balance between accuracy and interval quality (EC = 0.959, MIW = 33.3 MPa, AIS = 38.49 MPa, and ES = 0.753). Experimental validation on five binary and ternary mixes at 14 and 28 days returned full interval coverage at both ages, with mean errors of 6.81% and 1.52% and mean interval widths of 21.96 MPa and 20.39 MPa, respectively, demonstrating the direct applicability of the proposed framework for low-carbon UHPC mix design within the validated domain of compressive strength ≥ 60 MPa.
No takes yet. Share an insight, caveat, or question.
Malik et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: