Proceedings of the

The 33rd European Safety and Reliability Conference (ESREL 2023)
3 – 8 September 2023, Southampton, UK

Approach to Generate a Simple Semantic Data Model from 2D Bridge Plans using AI-based Text Recognition

Peter Gölzhäuser1,a, Mengyan Peng2,c, Jan-Iwo Jäkel1,b, Katharina Klemt-Albert1 and Steffen Marx1

1ICoM, RWTH Aachen University, Germany.

2IMB, Technische Universität Dresden, Germany.


The digital twin is intended to serve as the basis for an improved maintenance management. However, in the case of existing bridges, a digital model of the physical structure rarely exists. Various research approaches are currently addressing this problem using advanced technologies (laser scan, AI, photogrammetry). An essential part of these efforts is the transfer of relevant semantic information from an analogue source into the digital model. This paper deals with the question of how textual information from 2D drawings of bridges can be recognised and translated into a semantic data model. For this purpose, an OCR algorithm was utilized to translate printed and handwritten textual information into machine-readable text. The information pertaining to the material properties of the examined component was subsequently assigned to its respective component and stored in a structured data table.

The choice of the OCR algorithm, the post-processing of the text recognition results, the identification of relevant information and the translation into a semantic data model are the key findings presented in this paper. It was shown that while the approach is operational, the reliable identification of information is highly dependent on the nature and form of its representation in the drawings. While text recognition has been shown to be reliable, further research is needed to process and interpret the extracted semantic information to enable a broader approach to semantic enrichment.

Keywords: Building information modeling, Optical character recognition, Bridge maintenance, Construction / technical drawings, 2D-plans, Machine learning.

Download PDF