Abstract Despite the emerging new data capturing technologies and advanced modelling systems, the process of geometric digital twin modelling for existing buildings still lacks a systematic and completed framework to streamline. As-is Building Information Model (BIM) is one of the commonly used geometric digital twin modelling approaches. However, the process of as-is BIM construction is time-consuming and needed to improve. To address this challenge, in this paper, a semi-automatic approach is developed to establish a systematic, accurate and convenient digital twinning system based on images and CAD drawings. With this ultimate goal, this paper summarises the state-of-the-art geometric digital twinning methods and elaborates on the methodological framework of this semi-automatic geometric digital twinning approach. The framework consists of three modules. The Building Framework Construction and Geometry Information Extraction (Module 1) defines the locations of each structural component through recognising special symbols in a floor plan and then extracting data from CAD drawings using the Optical Character Recognition (OCR) technology. Meaningful text information is further filtered based on predefined rules. In order to integrate with completed building information, the Building Information Complementary (Module 2) is developed based on neuro-fuzzy system (NFS) and the image processing procedure to supplement additional building components. Finally, the Information Integration and IFC Creation (Module 3) integrates information from Module 1 and 2 and creates as-is Industry Foundation Classes (IFC) BIM based on IFC schema. A case study using part of an office building and the results of its analysis are provided and discussed from the perspectives of applicability and accuracy. Future works and limitations are also addressed.

中文翻译：

---

基于图像和CAD图纸的既有建筑半自动几何数字孪生

摘要 尽管出现了新的数据采集技术和先进的建模系统，但现有建筑的几何数字孪生建模过程仍然缺乏一个系统的、完整的框架来简化。原样建筑信息模型 (BIM) 是常用的几何数字孪生建模方法之一。但是，按原样进行 BIM 构建的过程非常耗时，需要改进。为了应对这一挑战，本文开发了一种半自动方法，以建立基于图像和 CAD 图纸的系统、准确和方便的数字孪生系统。带着这个最终目标，本文总结了最先进的几何数字孪生方法，并详细阐述了这种半自动几何数字孪生方法的方法框架。该框架由三个模块组成。建筑框架构造和几何信息提取（模块 1）通过识别平面图中的特殊符号，然后使用光学字符识别 (OCR) 技术从 CAD 图纸中提取数据来定义每个结构组件的位置。根据预定义的规则进一步过滤有意义的文本信息。为了与完整的建筑信息集成，基于神经模糊系统（NFS）和图像处理程序开发了建筑信息补充（模块 2），以补充额外的建筑组件。最后，信息集成和 IFC 创建（模块 3）集成来自模块 1 和 2 的信息，并基于 IFC 模式创建原样的行业基础类 (IFC) BIM。从适用性和准确性的角度提供并讨论了使用办公楼的一部分的案例研究及其分析结果。未来的工作和局限性也得到了解决。