A Study of Structured/Semantic Data Extraction from Mechanical Engineering Drawings
Arnaud Ridard, Omar Lebhar, Nikita Letov, Yacine Mahdid, Yaoyao Fiona Zhao · HAL (Le Centre pour la Communication Scientifique Directe) · 2025
Mechanical engineering drawings are essential documents for designing, manufacturing, and maintaining industrial systems. However, their complex structure-combining graphical elements, annotations, tables, and text-makes manual interpretation costly, time-consuming, and error-prone. Usual document management approaches store engineering drawings as digital files but lack efficient methods for retrieving and utilizing their semantic data. This paper explores recent computer vision and artificial intelligence advances to automate information extraction from engineering drawings. State-of-theart methods for processing symbolic, tabular, and geometrical data are reviewed, highlighting the challenges of low-level (e. g., OCR, vectorization) and high-level (e. g., 3D reconstruction, semantic annotation) interpretation. A deep learning-based case study is presented for extracting title blocks, bills of materials, and general notes. Finally, a conceptual framework is established for understanding engineering drawing and integrating multimodal data to enable applications in supply chain management and manufacturing optimization. Our findings underscore the need for open datasets, robust benchmarks, and novel AI-driven methodologies to bridge the gap between raw engineering drawings and structured, actionable information.