Generating model-based construction details for existing bridges using Informed Machine Learning.

数字孪生的使用为土木工程结构的操作和维护管理提供了新的可能性。这样做的先决条件是一个很少用于现有结构的已建成模型。建成模型至少应该与结构设计的细节相对应。研究项目的重点是生成桥梁的三维结构细节，并将其集成到现有的主要部件的数字模型中。这尤其包括钢筋和肌腱的细节。关于加固和安装的肌腱的必要信息不能从点云或图像等表面信息中获得。常规的执行计划和文本文档（如结构分析）是唯一的信息来源。目前，还没有一种方法可以自动评估桥梁结构的二维平面图和相关报告，从而在三维生成相关的执行细节，并将其集成到建成模型中。生成基于机器学习方法和基于知识的系统（知情机器学习）。为特定的检测和识别步骤开发知识来源、数据集和方法。将解决以下科学子目标：基于常规二维结构或加固方案的技术细节识别和分类方法。语义信息的提取方法，如单个结构细节的尺寸和标签。基于知识的概念，用于从不同文档中提取结构细节和语义信息的匹配。基于传统文本文档的附加信息的提取和集成方法。使用参数化技术生成结构细节的概念及其在相应组件中的定位。

The use of digital twins offers new possibilities for the operation and maintenance management of civil engineering structures. The prerequisite for this is an as-built model which is rarely available for existing structures. The as-built models should at least correspond to the detailing of the structural design. The focus of the research project is the generation of three-dimensional structural details for bridges and their integration into an existing digital model of the main components. This includes in particular details of reinforcement and tendons. The necessary information about reinforcement and installed tendons cannot be obtained from surface information such as point clouds or images. Conventional execution plans and text documents such as the structural analysis are the only source of information. Currently, no methods exist for automated evaluation of 2D plans and related reports of bridge structures with the purpose of three-dimensional generation of relevant execution details for integration into an as-built model. The generation is based on machine learning approaches and knowledge-based systems (Informed Machine Learning). Knowledge sources, data sets and methods are developed for specific detection and recognition steps. The following scientific sub-goals will be addressed: Methods for the recognition and classification of technical details based on conventional two-dimensional construction or reinforcement plans. Methods for the extraction of semantic information, such as dimensions and labels for individual structural details. Knowledge-based concepts for the matching of extracted structural details and semantic information from the different documents. Methods for extracting and integrating additional information based on conventional text documents. Concepts for generating structural details using parametric techniques and their positioning in the corresponding components.