An LLM-based cross-domain knowledge retrieval augmented generation method for bio-inspired solution design

Abstract

Innovative engineering design requires systematic retrieval and adaptation of cross-domain insights to foster novel solutions. While bio-inspired strategies offer potential for sustainable innovation, designers face challenges in bridging biological analogies with engineering applications. This research introduces a large language model (LLM)-based methodology integrating cross-domain knowledge retrieval-augmented generation for bio-inspired solution design. A unified knowledge graph aligns engineering and biological domains through structured entity-relationship modeling, enabling semantic retrieval of interdisciplinary patterns. The approach employs sampling algorithms to navigate cross-domain knowledge reasoning, identifying transferable biological principles relevant to engineering problems. Three LLM-powered phases are implemented: (1) Context-aware problem decomposition, (2) Retrieval-augmented scheme generation through dynamic knowledge fusion, and (3) Iterative refinement via human feedback. The system enables continuous optimization through bidirectional feedback loops, where designers guide LLM outputs while the model proposes biologically-informed design variations. Validation through wastewater treatment system development demonstrates enhanced creativity metrics and functional feasibility compared to conventional engineering design.