Sustainable design of self-compacting concrete integrating explainable machine learning, physics-based maturity models, and multi-objective optimisation

Aldawish, Abdulaziz 2026. Sustainable design of self-compacting concrete integrating explainable machine learning, physics-based maturity models, and multi-objective optimisation. PhD Thesis, Cardiff University. Item availability restricted.

Preview	PDF (Thesis) - Accepted Post-Print Version Download (5MB) | Preview
	PDF (Cardiff University Electronic Publication Form) - Supplemental Material Restricted to Repository staff only Download (301kB)

Abstract

Self-compacting concrete (SCC) flows and consolidates under its own weight, yet its mix design remains largely empirical, rarely treating workability, strength, cost, and sustainability as simultaneous objectives. This thesis develops an integrated, interpretable framework unifying explainable machine learning for fresh-state prediction, a physics-based temperature-adaptive maturity model, and life-cycle driven multi-objective optimisation. A database of 2504 SCC mixtures from 176 published studies was assembled, and ten algorithms were benchmarked for predicting slump flow, T₅₀, V-funnel time, and L-box ratio. Extra Trees performed best, achieving R² = 0.61 for slump flow and 0.63 for T₅₀ under five-fold cross-validation, and distribution-free conformal prediction intervals delivered 87–93% coverage, with SHAP analysis identifying the aggregate frame, the paste–water balance, and superplasticiser dosage as the dominant variables. Against fifty Kuwait production batches, mean slump flow was reproduced to within 9 mm with every batch inside its 90% conformal interval, confirming transfer to field practice. Strength development is captured by a maturity model with a linear temperature adaptive activation energy, calibrated on 196 mean-strength observations from 588 cube tests spanning four curing temperatures (10–50 °C), seven ages (1–90 days), and seven high-strength SCC mixtures. The model achieved R² = 0.929 (RMSE = 4.74 MPa), reduced systematic bias at the temperature extremes relative to the ASTM C1074 constant-activation-energy baseline, is favoured by the Akaike information criterion (the Bayesian criterion marginally favours the simpler form), and generalised to 120 independent literature observations with R² = 0.881 (RMSE = 6.26 MPa). Finally, cradle-to-gate life cycle assessment was integrated with NSGA-II optimisation, minimising embodied CO₂ and material cost while maximising predicted slump flow under workability and compositional constraints; the resulting 100 Pareto-optimal designs deliver up to 40% lower embodied CO₂ at or below dataset-average cost while meeting the EFNARC envelope with 90% confidence.

Item Type:	Thesis (PhD)
Date Type:	Completion
Status:	Unpublished
Schools:	Schools > Engineering
Uncontrolled Keywords:	1. Self-compacting concrete 2. Supplementary cementitious materials 3. Machine learning 4. Explainable artificial intelligence 5. Maturity method 6. Multi-objective optimisation
Date of First Compliant Deposit:	1 September 2026
Last Modified:	01 Sep 2026 13:08
URI:	https://orca.cardiff.ac.uk/id/eprint/189239

Actions (repository staff only)

Edit Item