Data-augmented probabilistic design model for shear capacity of cementitious composite strengthened reinforced concrete beams

Accurate and design-consistent prediction of the shear capacity of Fabric-Reinforced Cementitious Matrix (FRCM)-strengthened reinforced-concrete (RC) beams remains challenging because of limited experimental evidence, coupled load-transfer mechanisms, and inherent model uncertainty. This study develops an interpretable and uncertainty-aware hybrid framework that integrates machine-learning prediction, TVAE-assisted input-space enrichment, and Bayesian–Markov Chain Monte Carlo (MCMC) calibration to derive a mechanics-informed shear formulation with quantified uncertainty. The 207-specimen experimental database was used for machine-learning benchmarking and surrogate-model development. Of these specimens, 149 complete non-anchored beams were retained for TVAE-assisted enrichment and Bayesian–MCMC calibration, consistent with the applicability domain of the proposed mechanics-based formulation. Among the investigated regressors, the super-learner stacking model achieved the best test-set performance, with a coefficient of determination (R^2 ) of 0.967, a mean experimental-to-predicted ratio (μ_p ) of 100.74%, and a coefficient of variation (CoV) of 11.60%, and exhibited stable behaviour under repeated random splits. The TVAE-generated samples were introduced as down-weighted pseudo-labelled auxiliary information to improve coverage of the non-anchored input domain, rather than as independent experimental evidence. The calibrated formulation explicitly separates the concrete, internal steel, and external FRCM contributions while accounting for interaction effects. Validation against measured non-anchored experimental data showed improved agreement compared with two conventional analytical models, with R^2=0.862, μ_p=106.47%, and CoV = 26.08%. Residual and posterior uncertainty analyses further enabled preliminary reliability-oriented model-uncertainty formatting. The proposed framework provides a transparent route from data-driven prediction to mechanics-informed probabilistic assessment.

Authors

Publication Details

Journal
Engineering Applications of Artificial Intelligence
Published
2026-11-19
DOI
https://doi.org/10.1016/j.engappai.2026.116794
Primary Topic
Structural Behavior of Reinforced Concrete
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Data-augmented probabilistic design model for shear capacity of cementitious composite strengthened reinforced concrete beams

Xiangsheng Liu, XIANGSHENG LIU, XIA QIN
6 links
Engineering Applications of Artificial Intelligence
Structural Behavior of Reinforced Concrete
article

Data-augmented probabilistic design model for shear capacity of cementitious composite strengthened reinforced concrete beams

Xiangsheng Liu, XIANGSHENG LIU, XIA QIN
article en

Abstract

Accurate and design-consistent prediction of the shear capacity of Fabric-Reinforced Cementitious Matrix (FRCM)-strengthened reinforced-concrete (RC) beams remains challenging because of limited experimental evidence, coupled load-transfer mechanisms, and inherent model uncertainty. This study develops an interpretable and uncertainty-aware hybrid framework that integrates machine-learning prediction, TVAE-assisted input-space enrichment, and Bayesian–Markov Chain Monte Carlo (MCMC) calibration to derive a mechanics-informed shear formulation with quantified uncertainty. The 207-specimen experimental database was used for machine-learning benchmarking and surrogate-model development. Of these specimens, 149 complete non-anchored beams were retained for TVAE-assisted enrichment and Bayesian–MCMC calibration, consistent with the applicability domain of the proposed mechanics-based formulation. Among the investigated regressors, the super-learner stacking model achieved the best test-set performance, with a coefficient of determination (R^2 ) of 0.967, a mean experimental-to-predicted ratio (μ_p ) of 100.74%, and a coefficient of variation (CoV) of 11.60%, and exhibited stable behaviour under repeated random splits. The TVAE-generated samples were introduced as down-weighted pseudo-labelled auxiliary information to improve coverage of the non-anchored input domain, rather than as independent experimental evidence. The calibrated formulation explicitly separates the concrete, internal steel, and external FRCM contributions while accounting for interaction effects. Validation against measured non-anchored experimental data showed improved agreement compared with two conventional analytical models, with R^2=0.862, μ_p=106.47%, and CoV = 26.08%. Residual and posterior uncertainty analyses further enabled preliminary reliability-oriented model-uncertainty formatting. The proposed framework provides a transparent route from data-driven prediction to mechanics-informed probabilistic assessment.

Engineering Applications of Artificial IntelligenceVol. 188
Industry, innovation and infrastructure
Structural Behavior of Reinforced Concrete
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon