AI-Driven Composite Bridge Design Using Machine Learning and Genetic Algorithms

The design of steel-concrete composite bridges requires iterative structural analysis, code-based verification, and optimization processes that can become computationally demanding during the early design stages. This study presents an AI-assisted workflow for preliminary design and optimization of composite bridges using a parametric framework developed to automate structural verification according to Eurocode provisions called EFESTO. It combines a spreadsheet-based user interface with a Python computational backend that performs automated verification of ultimate limit states (ULS), serviceability limit states (SLS), and fatigue, while accounting for time-dependent effects such as creep and shrinkage. In the first phase, the framework incorporates a Genetic Algorithm to identify optimal structural configurations for given bridge parameters. In the second phase, it was used to systematically vary key geometric parameters to generate a database of approximately 2500 bridge configurations, including optimal section dimensions, verification demand-capacity ratios, and structural steel consumption. To accelerate the preliminary design phase, machine learning surrogate models were trained on the generated dataset to predict key design outputs directly from geometric parameters. Three approaches of Gradient Boosting, Random Forest, and Polynomial Regression were evaluated in terms of predictive accuracy for parameters such as girder height and steel consumption. The proposed workflow is demonstrated through a case study where machine learning predictions are used to estimate the design of a new bridge configuration. The predicted solution is subsequently verified and validated through finite element analysis, highlighting the potential of integrating machine learning with parametric design tools to accelerate bridge design.