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

Project summary

One of the most effective approaches to capitalize on available resources in the context of existing bridge infrastructure is to extend the lifespan of the global stock of aging bridges through monitoring, maintenance, and rehabilitation. On the other hand, material resources can be used effectively in new bridge designs by generating efficient structural forms inspired by relevant examples of built bridges. The possibility of easily acquiring geometric and structural data on existing bridges worldwide would contribute significantly to achieving these two challenging goals. To this end, this project aims to establish an artificial intelligence (AI) framework for the automated generation of a digital archive of 3D mesh and structural models of existing bridges using 2D images of bridges available online. A fully supervised machine learning (ML) model will be first developed to learn the association of bridge images to 3D models based on a manually generated training dataset of representative existing bridges. Computational methods like combinatorial equilibrium modeling (CEM) and finite element method (FEM) will be used to produce the 3D structural models in the training dataset. By leveraging novel semi-supervised 3D reconstruction methods, the trained ML model will then be able to automatically reconstruct 3D models of any existing bridge for which a set of pictures is available online. The applicability of the proposed AI framework will be demonstrated in two real-world applications related to the maintenance of existing bridges and the conceptual design of new bridges.