AUTHOR=Villena Toro Javier , Wiberg Anton , Tarkian Mehdi 

TITLE=Optical character recognition on engineering drawings to achieve automation in production quality control

JOURNAL=Frontiers in Manufacturing Technology

VOLUME=Volume 3 - 2023

YEAR=2023

URL=https://www.frontiersin.org/journals/manufacturing-technology/articles/10.3389/fmtec.2023.1154132

DOI=10.3389/fmtec.2023.1154132

ISSN=2813-0359

ABSTRACT=Digitization is a necessary step to reach quality control automation for production of mechanical products. This paper focuses on autonomous text detection and recognition in engineering drawings to enable automated production quality control. Engineering drawings are one of the main and efficient information carriers for production, however, what is easily readable by humans becomes a challenge for computer vision. To reach automation in quality control, data transfer between analog drawings and CAD/CAM software must be seamless.

Most optical character recognition (OCR) algorithms have shortcomings in their application to mechanical drawings because of their inherent complexity. The stack of measurements, orientation, the inclusion of tolerances, and special symbols such as geometric dimensioning and tolerancing (GD&T) contribute to the limited data available due to intellectual property concerns.

The methodology of this paper is divided into five stages. First, image processing techniques are used to classify and identify key elements for the comprehension of the drawing. The output is three elements: information block and tables, feature control frames, and the rest of the image. For each of these elements, an OCR pipeline is proposed. With the developed tool it is possible to create a seamless integration between engineering drawings and quality control. The tool achieved a precision and recall of 90% in detection, a F1-score of 94% in recognition, and a character error rate of 8%.  The proposed tool called eDOCr is shared with the research community through Github.