Development of low carbon concrete and a machine learning integrated model tool for predicting strength and embodied carbon for lightweight concrete production

Nukah, Promise

Abstract

The use of Ground Granulated Blast-furnace Slag (GGBS) as an alternative cement replacement material has shown good results in producing near green concrete. While GGBS exhibits good cementitious properties, concerns regarding slow strength development and workability due to its non-pozzolanic nature and porosity persist. Therefore, this study presents the inclusion of light weight aggregate, GGBS, alkaline solution, and silica fume to achieve a near zero carbon concrete in the design and construction of concrete beams.

Experimental study carried out entailed the use of Lytag aggregate after pre-soaked for 24 hours. Alkaline precursors were prepared using sodium hydroxide and sodium silicate solution and used after 24hours. Mix proportions of samples were based on concrete mix design from grade 20 to 50 with silica fume and superplasticizer added as an additive and an to enhance workability and cement replaced between 60-80% with GGBS for alkaline binder ratio between 0.3 to 0.6. Sample mixes were adjusted for different values of alkaline binder ratio while sodium silicate to sodium hydroxide ratio were kept at 2.5 for all mixes. Mechanical testing carried on sample include compressive strength, Flexural testing, Creep, and slump test. Embodied carbon of concrete samples evaluated using the global warming potentials. All sample constituents were considered as input features while compressive strength and embodied carbon of samples were obtained as output features for the development of tool to predict concrete performance on compressive strength and carbon emissions. Analysis of sample performance in terms of application in structural design were carried out with derivation of beam design equation.

Mathematical modelling of compressive strength of the sample for prediction was also derived. Results indicated a 42% increase in compressive strength and a 22% increase in ultimate compressive strain for the geopolymer lightweight concrete, suggesting improved structural resistance. Additionally, embodied carbon was reduced by 46–61% for non-geopolymer lytag-based concrete and 69–77% for geopolymer concrete. Creep strain increased by 0.55% for geopolymer concrete, compared to 2.81% for non-geopolymer lytag-based concrete. The modulus of elasticity decreased with loading age for geopolymer concrete at 0.39%, contrasting with a 1.93% reduction for non-geopolymer lytag-based concrete. This study employs machine learning regression models to predict compressive strength and embodied carbon, with the XGBoost model achieving a low Mean Squared Error of 50.15 and an R² of 0.96. The developed tool effectively predicts compressive strength and embodied carbon, supporting sustainable concrete design to aligned with net-zero carbon targets.

• Thesis Type: Thesis
• Deposit Date: Oct 28, 2024
• Publicly Available Date: Jul 4, 2025
• Keywords: Green concrete, Embodied carbon, Flexural strength, Lightweight concrete, Geopolymer concrete, Creep, Compressive strength, Machine learning, Foundation structure, Artificial intelligence, Dimensional analysis, Oven dried density, Water to binder ratio, Alkaline binder ratio
• Public URL: https://uwe-repository.worktribe.com/output/13321095
• Award Date: Jul 4, 2025