DUCTILE: Agentic LLM Orchestration of Engineering Analysis in Product Development Practice
A new LLM agent system handles messy, changing data formats that break traditional automation in industrial design.
A team from Chalmers University of Technology and GKN Aerospace has introduced DUCTILE (Delegated, User-supervised Coordination of Tool- and document-Integrated LLM-Enabled), a novel framework for automating complex engineering analysis. Traditional automation in product development relies on rigid scripts that fail when interfaces, data formats, or methodologies change. DUCTILE addresses this brittleness by using an LLM agent to perform adaptive orchestration—interpreting documented practices and inspecting input data to dynamically plan a processing path—while offloading the deterministic execution to verified, trusted engineering tools. This creates a flexible, human-in-the-loop system where the engineer supervises and provides final judgment.
The system was demonstrated on a real-world industrial structural analysis task at an aerospace manufacturer. The DUCTILE agent successfully navigated input deviations in format, units, naming conventions, and methodology that would have caused traditional scripted pipelines to break. Evaluation against expert-defined acceptance criteria and deployment with practicing engineers confirmed it produces correct, methodologically compliant results consistently across independent runs. The paper also delves into the practical consequences of such agentic automation, exploring the tension between removing mundane tasks and the potential for creating a new, potentially exhausting supervisory role for engineers, highlighting a shift in the nature of engineering work itself.
- DUCTILE uses an LLM agent for adaptive planning while trusted tools handle deterministic execution, creating a resilient human-in-the-loop system.
- Successfully tested on aerospace structural analysis, handling deviations in data format, units, and naming that break traditional scripts.
- The research discusses the real-world impact of agentic automation, including changes to engineering roles and supervisory burdens.
Why It Matters
This moves AI beyond simple chatbots into robust, real-world engineering workflows, automating complex tasks while keeping experts in control.