iDesignGPT enhances conceptual design via large language model agentic workflows.

Area of Science:

• Engineering Design
• Artificial Intelligence
• Human-Computer Interaction

Background:

• Traditional engineering design methods struggle with complexity and creativity.
• Emerging AI tools offer potential but require integration with established practices.

Purpose of the Study:

• To introduce iDesignGPT, a framework combining large language models (LLMs) with engineering design methodologies.
• To enable dynamic multi-agent collaboration for enhanced conceptual design.
• To provide quantitative insights using design metrics like novelty and diversity.

Main Methods:

• iDesignGPT framework integrates LLMs with multi-agent collaboration for design tasks.
• Incorporation of design metrics (coverage, diversity, novelty) for quantitative evaluation.
• Performance evaluation on six public design challenges and two user studies.

Main Results:

• iDesignGPT demonstrated competitive novelty and superior originality and modularity compared to GPT-4o and Deepseek-r1.
• Expert assessments and metric evaluations validated iDesignGPT's performance.
• User studies indicated positive reception and reduced mental demand for novice designers.

Conclusions:

• iDesignGPT is a practical framework for integrating LLM agents with engineering design methods.
• It offers metrics-driven support for conceptual design, benefiting both experts and novices.
• The framework facilitates dynamic multi-agent collaboration for complex design challenges.