Surrogate-assisted design optimisation method for steel-concrete joints in large-span hybrid girder bridges using machine learning
Steel-concrete joints play a critical role in force transfer in large-span hybrid girder bridges, where multiple structural parameters interact in a highly nonlinear manner. Conventional parametric analysis is often insufficient for efficient performance evaluation and design optimisation. In this article, an application-oriented surrogate-assisted design optimisation framework was established for steel-concrete joints by integrating finite element (FE) analysis, machine learning (ML) and multi-objective decision-making. A database was established based on detailed FE models covering various design parameter combinations of common hybrid girder bridges. Six ML algorithms were trained and evaluated, and the best-performing models were selected as surrogate models for predicting force transfer performance indicators. The results indicate that LightGBM and XGBoost achieve high predictive accuracy for the uniformity indices of the back and front cells, respectively. Based on the surrogate models, Pareto-based multi-objective optimisation was conducted considering force transfer uniformity and equivalent economic performance related to material consumption, construction complexity and carbon-related effects. Finally, a preferred design parameter combination can be selected by the comprehensive indicator with different weight coefficients. The proposed framework provides a systematic data-driven procedure for evaluating, interpreting, and selecting design parameter combinations of steel-concrete joints, while reducing the computational cost of repeated FE analyses.
Authors
- Xiangyong Duanmu (ORCID: https://orcid.org/0009-0003-8895-422X)
- Yufan Sun (ORCID: https://orcid.org/0009-0000-4036-8115)
- Chenxi Fu
- Dong Xu
- Yakun Wu
Institutions
- Tongji University (CN)
- China Design Group (China) (CN)
- McGill University (CA)
Publication Details
- Journal
- Structure and Infrastructure Engineering
- Published
- 2026-11-21
- DOI
- https://doi.org/10.1080/15732479.2026.2765923
- Primary Topic
- Structural Engineering and Materials Analysis
- Type
- article
- Field-Weighted Citation Impact
- 0.00