AI-Driven Feature Extraction from Engineering Drawings
Customer Challenge
The Air Force required artificial intelligence (AI)-driven analysis of 2D engineering drawings to support digital sustainment activities for their weapon systems. Presently, engineers manually extract data from 2D drawings, but the volume of Air Force data makes large-scale extraction impractical.
Innovative Solution
Illumination Works enhanced its existing Linnea™ 2D Feature Extraction and Analysis Pipeline to more accurately analyze low quality Air Force engineering drawings. ILW implemented improved optical character recognition (OCR) and incorporated large language models (LLM), vision language models (VLM), and deep learning image segmentation algorithms to significantly increase the accuracy and range of information extracted from Air Force drawings.
Benefits/Outcomes
- Automatically extract 25+ key entities (e.g., part number, CAGE, materials) with high accuracy (80%+)
- Enhanced OCR and segmentation, achieving 85%+ accuracy on complex, information-dense drawings
- Containerized and deployed the improved
Linnea capability into an AWS GovCloud environment leveraging continuous integration/continuous deployment (CI/CD) methods
Domain Expertise
- Engineering
- Advanced Manufacturing
Business Value
- Eases engineers’ manual burden, saving time and costs by automating the data extraction from engineering data inputs
- Accelerates sustainment and advanced manufacturing workflows through AI-driven automation
- Extends potential use of extracted engineering data to a variety of sustainment activities
Toolbox
- ILW Linnea™ AM Decision Support Tool
- AI, NLP, Computer Vision, OCR, VLM, LLM
- AWS GovCloud, CI/CD Methods
