Infocz與首爾大學合作開發工業圖紙AI模型
Infocz, Seoul National University to Develop AI Model for Industrial Drawings
Infocz與首爾大學計算機科學與工程系展開聯合研究,開發以整張圖紙為單一輸入的 Image-to-Graph foundation model,識別圖紙中的符號及其連接關係,並轉換為圖表數據。研究將使用約20,000張真實工業圖紙訓練模型,並把成果逐步整合至 DELTAflow。技術計劃由管道及儀表流程圖擴展至單線圖、控制邏輯圖等工程圖紙。

Ontology-based enterprise AI company Infocz said on Nov. 12 that it has launched joint research with Seoul National University’s Department of Computer Science and Engineering, led by Professor Kim Gun-hee’s laboratory, to develop AI technology capable of recognizing entire industrial drawings.
Infocz and the SNU research team plan to develop an Image-to-Graph foundation model using a whole-scale approach that processes an entire drawing as a single input. The technology is designed to preserve details in high-resolution drawings while analyzing their overall structure and context, identifying symbols and their interconnections, and converting the information into graph data.
“Conventional methods that recognize drawings by dividing them into sections have limitations in identifying connections between equipment across an entire drawing,” Infocz said. “Industrial drawing recognition requires not only identifying numerous pieces of equipment, pipes and instruments as individual objects, but also interpreting their connections and understanding the context of the entire drawing.”
The research will use DELTAflow, Infocz’s proprietary drawing recognition solution. According to the company, DELTAflow incorporates technology that extracts equipment and piping connections and process flows from unstructured engineering drawings and converts them into structured data. Infocz said its customers use DELTAflow to analyze the impact of new equipment installations on production processes and to review designs.
The two parties will also develop training datasets specialized for industrial drawings. Infocz said it is using approximately 20,000 real-world industrial drawings to train AI models on various drawing structures and connections between equipment. The goal is to develop an AI model with recognition accuracy and processing efficiency suitable for practical industrial applications.
Through the joint research, Infocz plans to expand the technology’s capabilities beyond piping and instrumentation diagrams (P&IDs) to include single-line diagrams (SLDs), control logic diagrams and other engineering drawings. The research results will be gradually incorporated into DELTAflow to enhance automated analysis and search of large-scale drawings, exploration of equipment interconnections, and comparison of drawing revision histories.
“In the long term, we plan to connect graph data extracted from drawings with enterprise design, equipment and operational data, expanding its applications to AI agent-based workflow automation,” Infocz said.
來源:The Elec · thelec.net