AIDABI
Artificial Intelligence for the automated creation of a digital archive of bridge infrastructure
Principal Investigators
Prof. Dr. Pierluigi D’Acunto
Professorship of Structural Design, Department of Architecture, TUM School of Engineering and Design
Prof. Dr. Angela Dai
Professorship of Machine Learning for 3D Scene Geometry, Department of Informatics, TUM School of Computation, Information and Technology
Motivation and Goals
The maintenance of existing bridges and the design of new ones can be significantly improved by creating a comprehensive global archive of 3D and structural models of current bridges. Our project explores machine learning (ML)-based 3D reconstruction methods combined with computational structural design techniques to automate the archiving process using an image-to-3D/structure pipeline. In the current proof-of-concept phase, we focus on synthetic bridge data to validate our approach. In the long term, we aim to collect real-world bridge data to build a training set comprising structural graphs, 3D models, and bridge photographs.
Recent Results
Our recently developed synthetic bridge dataset generator can randomly produce an unlimited number of valid 3D bridge models along with corresponding renderings. The current dataset used for ML training comprises 1,000 bridges, each enriched with structural graph data linked to 3D structural models generated using the Combinatorial Equilibrium Modeling (CEM) form-finding method, structural statistical features, solid geometry models, input parameters for environmental and contextual conditions, and 20 rendered images per bridge.
Selected documentation
Conference Paper | Gunes, M. C., et al. (2024). "Synthetic dataset for ML-based image-to-3D reconstruction of bridge infrastructures" Georg Nemetschek Institute Symposium & Expo on Artificial Intelligence for the Built World 2024 |
Journal Paper (in preparation) | Gunes, M. C., et al. (2025). "ML-driven image-to-3D reconstruction for bridge infrastructure" Advanced Engineering Informatics, Elsevier |