Design automation towards sustainable self-compacting concrete via machine learning models

Cui, Tianyi 2024. Design automation towards sustainable self-compacting concrete via machine learning models. PhD Thesis, Cardiff University. Item availability restricted.

Preview	PDF (Thesis) - Accepted Post-Print Version Download (6MB) | Preview
	PDF (Cardiff University Electronic Publication Form) - Supplemental Material Restricted to Repository staff only Download (759kB)

Abstract

This thesis investigates the design, optimisation, and performance prediction of self-compacting concrete (SCC), with a focus on high strength SCC (HSSCC) and steel fibre reinforced SCC (SFRSCC). The research addresses critical challenges in SCC mix design and performance evaluation, aiming to improve efficiency, sustainability in modern construction. A pragmatic mix design methodology for HSSCC was developed, incorporating supplementary cementitious materials (SCMs) to achieve target compressive strengths of 70-100 MPa. This methodology reduces cement consumption and carbon emissions, contributing to sustainable construction practices. Design charts were created to provide practical guidance for selecting optimal mix proportions. Experimental validation confirmed that the proposed HSSCC mixes met performance requirements, with significant reductions in CO₂ emissions and improvements in fresh and hardened properties. Machine learning models, including support vector machines (SVM), artificial neural networks (ANN), decision trees (DT), and random forests (RF) were employed to predict the properties of SCC mixes containing fly ash. The results highlight the potential of machine learning to replace traditional methods by efficiently capturing the complex interactions between SCC components and their performance. Feature importance analysis provided a detailed understanding of the contributions of specific mix components, offering valuable guidance for optimising SCC formulations. For SFRSCC, advanced ensemble learning models, including RF, gradient boosting decision trees (GBDT), XGBoost, and LightGBM, were applied alongside traditional machine learning approaches. Ensemble methods consistently outperformed traditional models in predicting compressive, tensile, and flexural strengths. Feature analysis was also employed for the best- iv performing models to assess the impact of the components on SFRSCC properties. This research makes significant contributions by introducing a pragmatic mix design methodology, integrating machine learning for efficient performance prediction, and providing advanced ensemble models for analysing SFRSCC properties. Overall, this research advances the understanding of SCC and its derivatives, contributing to the development of sustainable concrete technologies.

Item Type:	Thesis (PhD)
Date Type:	Completion
Status:	Unpublished
Schools:	Schools > Engineering
Date of First Compliant Deposit:	1 May 2025
Last Modified:	01 May 2025 15:18
URI:	https://orca.cardiff.ac.uk/id/eprint/177971

Actions (repository staff only)

Edit Item