Generating Execution Details for Existing Bridges with ML

Conventional 2D plans are the most common source of information about structural details of existing structures. Complex 2D technical plans (e.g. reinforcement plans) contain a large number of symbols, text, dimensions and connection lines that represent both physical and logical relationships between the information. For a reliable and efficient analysis and extraction of this information, appropriate expert knowledge must be taken into account. In this project the goal is to develop new methods for automating the generation of three-dimensional execution details in bridge construction and integrating them into existing digital building models. The project aims to analyze conventional 2D plans and text documents of bridge constructions to extract execution details like reinforcement and tendons. By leveraging machine learning, deep learning and knowledge-based systems, the project will create automated evaluation methods for bridge construction plans and related text documents, enabling the generation of subject models with execution details. The project will also address the challenge of integrating execution details into digital building models that already contain information about the main components of the bridge. The methods developed will involve recognizing and classifying technical details, extracting semantic information, assigning and locating execution details, extracting additional information from text documents, and employing parametric approaches to generate execution details within the digital building model.