AI-Based Engineering and Production Drawing Information Extraction
Christoph von der Haar, Hangbeom Kim, Lukas Koberg
Abstract
Abstract The production of small batches to single parts has been increasing for many years and it burdens manufacturers with higher cost pressure. A significant proportion of the costs and processing time arise from indirect efforts such as understanding the manufacturing features of engineering drawings and the process planning based on the features. For this reason, the goal is to automate these indirect efforts. The basis for the process planning is information defined in the design department. The state of the art for information transfer between design and work preparation is the use of digital models enriched with additional information (e.g. STEP AP242). Until today, however, the use of 2D manufacturing drawings is widespread. In addition, a lot of knowledge is stored in old, already manufactured components that are only documented in 2D drawings. This paper provides an AI(Artificial Intelligence)-based methodology for extracting information from the 2D engineering and manufacturing drawings. Hereby, it combines and compiles object detection and text recognition methods to interpret the document systematically. Recognition rates for 2D drawings up to 70% are realized.