Stress-strain behaviour of axially loaded FRP-confined natural and recycled aggregate concrete using design-oriented and machine learning approaches

Dada, Temitope E, Gong, Guobin, Xia, Jun and Di Sarno, Luigi ORCID: 0000-0001-6244-3251 (2024) Stress-strain behaviour of axially loaded FRP-confined natural and recycled aggregate concrete using design-oriented and machine learning approaches Journal of Building Engineering, 95. p. 110256. ISSN 2352-7102, 2352-7102

	Text Temitope Dada Paper Draft_rev.docx - Submitted version Download (6MB)
	Text Dada et al 2024 Final Version.pdf - Author Accepted Manuscript Available under License Creative Commons Attribution. Download (9MB) | Preview

Abstract

The reuse of recycled aggregates (RA) in fresh concrete helps to address construction waste management and reduces carbon footprints; however, the resulting recycled aggregate concrete (RAC) possesses some undesirable properties for structural applications. While strengthening RAC with fibre-reinforced polymers (FRP) helps to mitigate these issues, predicting the stress-strain behaviours of FRP-confined RAC (FRCRAC) and natural aggregate concrete (FRCNAC) can be complex. Therefore, high-performing models are required to quantify the behaviours of FRCNAC and FRCRAC as more experimental data emerge. This work employs design-oriented empirical modelling (DOEM) and machine learning (ML) approaches to predict the stress-strain behaviours of FRCNAC and FRCRAC. Data from 868 FRCNAC and 244 FRCRAC experimental specimens were assembled, and the kernel density estimator (KDE) was used to augment the limited FRCRAC training data for the ML approach. New high-performing DOEMs were developed and extensively tested against existing DOEMs. A new ML model named CATO was also implemented using the categorical boosting algorithm optimised with Optuna and trained via 10-fold cross-validation. The model performances were evaluated using the R², MAE, RMSE, MAPE, and GEC metrics. The results showed that the newly proposed DOEMs outperformed existing empirical models across all metrics. The CATO also outperformed all existing empirical and ML models, attaining high testing accuracies, with R² of 97.78 % and 91.92 % for the ultimate strength and strain respectively for FRCNAC, and with R² of 98.69 % and 94.94 % attained for the ultimate strength and strain respectively for FRCRAC. Furthermore, Zhou and Wu's (ZW) model was modified and integrated with CATO to form a novel hybrid CATO-MZW model, accurately predicting the axial stress-strain curves of FRCNAC and FRCRAC. The SHapley Additive exPlanation (SHAP) was used to identify the overall thickness of the FRP, unconfined concrete strength, and confining stress as crucial parameters in the model. Finally, the CATO-MZW was deployed online for practical use, and this tool developed in this study gives reliable, faster and more accurate predictions than existing experimental, numerical, and traditional analytical approaches.

Item Type:	Article
Uncontrolled Keywords:	Kernel density estimator, FRP-Confined concrete, Recycled aggregate concrete, Machine learning, Categorical boosting, Optuna
Depositing User:	Symplectic Admin
Date Deposited:	27 Aug 2024 07:53
Last Modified:	23 May 2026 09:11
DOI:	10.1016/j.jobe.2024.110256
Related Websites:	Author Publisher
URI:	https://livrepository.liverpool.ac.uk/id/eprint/3184054
Disclaimer:	The University of Liverpool is not responsible for content contained on other websites from links within repository metadata. Please contact us if you notice anything that appears incorrect or inappropriate.