Loading...
Genetic algorithm-optimized and explainable machine learning for bond strength prediction between steel-FRP composite bars and concrete
Zhang, Z. Ge, W. Han, S. Cao, D. Wu, J.
Zhang, Z.
Ge, W.
Han, S.
Cao, D.
Wu, J.
Publication Date
2026-07-15
End of Embargo
Supervisor
Rights
© 2026 The Author(s). This is the Author Accepted Manuscript of the article distributed under the Creative Commons CC-BY license (https://creativecommons.org/licenses/by/4.0) in accordance with the University of Bradford Rights Retention Policy.
Peer-Reviewed
Yes
Open Access status
openAccess
Accepted for publication
2026-04-11
Institution
Department
Awarded
Embargo end date
Files
Loading...
zhang_et_al_2026.pdf
Adobe PDF, 1.9 MB
Additional title
Abstract
The interfacial bond strength between steel-FRP composite bars (SFCBs) and concrete is crucial for structural design. Existing empirical formulas often fail to capture the complex nonlinear interactions arising from the unique secondary-stiffness characteristics of SFCBs. To address this, we propose an explainable machine learning (ML) framework. A comprehensive database comprising 241 pullout test specimens was constructed, incorporating 12 key geometrical and material features. Eight supervised ML algorithms were developed to model the bond behavior, with hyperparameters automatically optimized using a Genetic Algorithm (GA) to enhance accuracy and generalization. Comparative analysis demonstrates that the GA-optimized Support Vector Regression (GA-SVR) model is superior, yielding a coefficient of determination (R2) of 0.963 on the test set, far outperforming five existing empirical formulas and other ensemble algorithms. Gaussian noise injection tests further demonstrated that the GA-SVR model possesses exceptional robustness against data perturbations due to its structural risk minimization principle. Crucially, the "black box" nature of the current model was deciphered using Shapley Additive Explanations (SHAP) and three-dimensional Partial Dependence Plots (3D-PDPs), verifying that the model successfully learned the intrinsic physical mechanisms of the composite structure. The interpretability analysis confirmed the synergistic "dual-stiffness" mechanism, where both the steel core and FRP layer positively contribute to bond capacity, while accurately capturing the shear lag effect exhibited by the negative correlation with the bond length-to-diameter ratio. Finally, a user-friendly GUI integrating the optimized model was developed for quick, physically consistent bond strength estimation in practical engineering applications.
Version
Accepted manuscript
Citation
Zhang Z, Ge W, Han S et al (2026) Genetic algorithm-optimized and explainable machine learning for bond strength prediction between steel-FRP composite bars and concrete. Engineering Structures. 359: 122773.
Link to publisher’s version
Link to published version
Link to Version of Record
Type
Article