Infosys-SNU Launch Joint Research on AI for Full Recognition of Industrial Drawings... Developing a Domain-Specific Foundation Model
TECHWORLD ·
✦ AI Summary
Infosys announced on the 12th that it has launched joint research on 'AI technology for full recognition of industrial drawings' with Professor Kim Geon-hui's lab in the Department of Computer Science and Engineering at Seoul National University.
The two sides are pursuing the development of a 'Whole-scale' approach and an 'Image-to-Graph' foundation model.
The research target is P&ID, and the plan is to later expand to SLD and control logic diagrams.
BizCrushWant to follow Korean talks and live streams in real time?Try CapCrush on the web
Infosys announced on the 12th that it has launched joint research on "AI technology for full recognition of industrial drawings" with Professor Kim Geon-hui's lab in the Department of Computer Science and Engineering at Seoul National University. As the headline indicates, the development target is a drawing-specific foundation model. AI technology for full recognition of industrial drawings is considered one of the more technically challenging areas even in the global AI industry.
Infosys has its own developed drawing recognition and ontology auto-conversion technology, and it has supported AX at companies across many industries, including semiconductors, energy, refining, chemicals, and finance. It has also independently developed a drawing recognition solution, DELTAflow. The solution is equipped with technology that extracts equipment and piping connection relationships and process flows from unstructured design drawings, then converts the extracted information into structured data.
The research was launched by combining Infosys' accumulated field application experience with Seoul National University's research capabilities. Through this effort, the two sides plan to advance the technology further in a direction that will push AI for full recognition of industrial drawings, an area with a high level of technical difficulty, to the next stage.
Infosys said the prevailing drawing recognition method in industrial settings is segmented drawing recognition. However, it said segmented drawing recognition has limitations in identifying the connection relationships among all equipment in a drawing.
The core of industrial drawing recognition technology lies in the individual recognition of multiple objects such as equipment, piping, and instruments, reading the connection relationships between objects, and understanding the overall context of the drawing, it said. This challenge is regarded as difficult even for general-purpose AI and multimodal AI.
Accordingly, the Infosys and Seoul National University research team said it will use a "Whole-scale" approach. It also said it plans to develop an "Image-to-Graph" foundation model.
The "Whole-scale" approach processes the entire drawing as a single input. This model approach preserves fine details in high-resolution drawings, analyzes the overall structure and context, and identifies symbols within the drawing and the connections between them to convert the information into graph data.
The two sides are working in parallel on core AI model development and the construction of training data specialized for industrial drawings. Infosys is using about 20,000 real industrial drawing images to learn various drawing structures and equipment connection relationships. The two sides plan to focus on developing an AI model at that level, with the goal of securing recognition accuracy suitable for real product deployment and processing efficiency suitable for real product deployment.
Infosys said the research goal is to train the structural characteristics of industrial drawings. It also said the research aims to develop an AI model dedicated to industrial drawings that can recognize complex connection relationships and to secure foundational AI technologies specialized for industry.
The initial research target is P&ID. Starting with this, Infosys plans to advance the technology so it can analyze SLD and control logic diagrams, and expand the scope of analysis to various engineering drawings.
Infosys plans to apply the research results to DELTAflow in stages. Through this, it intends to strengthen DELTAflow's large-scale drawing automatic analysis and search functions, equipment connection relationship exploration, and drawing revision history comparison capabilities. In the long term, it plans to connect graph data extracted from drawings with a company's design, equipment, and operations data, and expand usage into AI agent-based workflow automation.
Tak Jeong-su, CEO of Infosys, said that AI's precise understanding of the overall structure and connection relationships in industrial drawings is an unresolved global technical challenge, and that this is why the company is pursuing joint research with Seoul National University. Tak said the joint research aims to present a new technical standard in the field of industrial drawing recognition, and that the company plans to apply the research results to its own solution, DELTAflow. He added that it expects drawing-based work innovation through the application of DELTAflow.
Source: TECHWORLD · Lee Gwang-jae
Original: https://www.epnc.co.kr/news/articleView.html?idxno=409627
References
This article was produced with the help of an automated content generation algorithm.
Source: TECHWORLD
View originalThis article was summarized and organized by BizCrush based on the original article from TECHWORLD. For exact quotations and full details, please refer to the original article.
Keep reading
BizCrush