Solver-free artificial intelligence framework for stress-strain prediction and rapid field visualisation of fibre-reinforced polymer-confined recycled aggregate concrete

Dada, Temitope E, Gong, Guobin, Xia, Jun and Di Sarno, Luigi ORCID: 0000-0001-6244-3251 (2026) Solver-free artificial intelligence framework for stress-strain prediction and rapid field visualisation of fibre-reinforced polymer-confined recycled aggregate concrete Engineering Applications of Artificial Intelligence, 163. p. 113130. ISSN 0952-1976

Text Dada et al._Final.docx (1).pdf - Author Accepted Manuscript Available under License Creative Commons Attribution. Download (19MB) | Preview

Abstract

Accurately predicting the stress-strain behaviour of fibre-reinforced polymer (FRP)-confined recycled aggregate concrete (FRCRAC) remains challenging due to the complex mechanics introduced by recycled aggregate. This study presents a novel two-stage machine-learning (ML) framework with a mechanics-inspired visualisation module (MIVM) to predict and visualise the stress-strain behaviour of FRCRAC. First, Optuna-optimised Categorical Boosting (CATO) models are used to predict ultimate axial strength, axial strain, and hoop strain. These predictions are then integrated into Long Short-Term Memory (LSTMO) models to construct full axial and hoop stress-strain curves, forming a CATO-LSTMO framework. Trained via ten-fold cross-validation on a combination of 194 experimental and 600 synthetic datasets, CATO-LSTMO significantly outperforms conventional analytical and hybrid models with a coefficient of determination, R2, above 98 %. Secondly, a solver-free MIVM that replicates finite element (FE) behaviour is proposed to enhance the physical interpretation of ML models. Three dedicated Categorical Boosting regressors are trained on over 350,000 nodal field outputs from Abaqus simulations to predict the three-dimensional stress, strain, and displacement distributions across 21 loading frames. These predictions are then scaled using the previously ML-generated stress-strain curves to reconstruct the framewise three-dimensional contour plots. The MIVM visualisations achieve results comparable to Abaqus outputs, while being about 500 times faster. This study contributes to artificial intelligence through novel hybrid ML frameworks and solver-free surrogate visualisation. The integrated CATO-LSTMO-MIVM framework is deployed as an interactive web application, and it offers practical use in engineering applications for rapidly estimating and visualising the stress-strain behaviour of FRCRAC.

Item Type:	Article
Divisions:	Faculty of Science & Engineering Faculty of Science & Engineering > School of Engineering Faculty of Science & Engineering > School of Engineering > Civil and Environmental Engineering
Depositing User:	Symplectic Admin
Date Deposited:	20 Nov 2025 08:07
Last Modified:	20 Nov 2025 08:07
DOI:	10.1016/j.engappai.2025.113130
Related Websites:	Publisher
URI:	https://livrepository.liverpool.ac.uk/id/eprint/3195482
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.