An Innovative Retrieval-Augmented Generation Framework for Stage-Specific Knowledge Translation in Biomimicry Design.

Area of Science:

• Biomimicry Design
• Artificial Intelligence in Design
• Human-Computer Interaction

Background:

• Translating biological strategies into design principles is challenging for designers without biological expertise.
• Existing methods like BioTRIZ and SAPPhIRE require specialized knowledge.
• Bridging the knowledge gap in biomimicry design is crucial for innovation.

Purpose of the Study:

• To introduce a Retrieval-Augmented Generation (RAG) framework integrated with the Biomimicry Design Spiral (BDS).
• To evaluate the effectiveness of a stage-specific RAG-LLM framework in assisting industrial design students.
• To provide a reproducible AI assistance model for biomimicry design.

Main Methods:

• A quasi-experimental study involving 30 industrial design students.
• Assessment of three setups: LLM-only, RAG-Small, and RAG-Large.
• Evaluation of text quality, design concept quality, and retrieval diversity across six design stages.

Main Results:

• The RAG-Large framework demonstrated superior text quality in cognitively demanding stages.
• RAG-Large retrieved a more diverse range of high-specificity biological ideas.
• The framework facilitated more coherent integration of functional, aesthetic, and semantic design aspects.

Conclusions:

• The RAG-LLM framework effectively bridges the knowledge translation gap in biomimicry design.
• This approach diminishes cognitive burden and enhances the relevance and originality of design inspirations.
• The study provides a stage-specific AI assistance model for biomimicry, though further validation is needed.