Data Set of Concrete Beam–Slab Bridge Superstructures for Machine Learning–Based Estimation of Structural Parameters

Abstract The load rating of existing road bridges often requires construction drawings to analytically determine internal forces and member capacities. When such documentation is unavailable, engineers must rely on costly load tests, advanced field surveys, or expert judgment. To support the development of estimation models for unknown structural parameters, this paper presents a curated data set of beam–slab bridges, assembled from actual construction drawings obtained primarily from the archives of the Colombian National Institute of Roads (INVIAS). The data set includes 420 reinforced concrete (RC) decks and 892 RC and post-tensioned prestressed concrete girders, with over 60 parameters covering geometry, materials, and reinforcement details. Records were reviewed, cataloged, and filtered to ensure completeness and consistency. The data set is publicly available for download from a data repository. An exploratory analysis revealed clear relationships between material specifications and construction periods, as well as between span lengths and reinforcement or prestressing amounts. To demonstrate the data set’s utility for analytical bridge assessment, regression models were developed to estimate key reinforcement parameters, primarily using geometric predictors. For example, the best-performing model for total prestressing steel area ( A ps ) achieved a validation root-mean-square error (RMSE) of 10% of the training set mean, and predictions for bottom steel ( A s , bot ) in RC girders yielded a validation RMSE of 21% of the training set mean. These results confirm the capability of the data set to support data-driven estimation of structural details in bridges lacking documentation. The potential for further applications is also discussed.