Ideational Design Support through Patent Specification Generation Using Large Language Models

Abstract

Ideational design begins with vague and abstract ideas that must be transformed into concrete design concepts. Patent specifications are useful not only for protecting inventions as intellectual property but also for documenting design intent, technical problems, solutions, and expected effects. However, preparing patent specifications generally requires collaboration among inventors, engineers, and patent professionals, making the process time-consuming and costly. This study investigates the use of large language models (LLMs) to support conceptual design through patent specification generation and design knowledge extraction. For method inventions, we constructed an LLM-based workflow in which ideas discussed with an LLM were expanded into patent claims and embodiments and automatically converted into patent specifications. For product inventions, we proposed a three-layer representation consisting of a Feature Ontology, a Meta Process Model, and a Parametric Shape Space. Using two patent specifications of axial fans as case studies, the LLM extracted design intent and organized it into a meta-process model for representing design rationale, enabling visualization of innovation relationships between patents. The results demonstrate that current LLMs are effective for expanding design ideas, generating patent specifications for method inventions, and structuring design knowledge from existing patent documents. Although evaluating hardware geometries and physical phenomena remains difficult for current LLMs, the proposed approach enables knowledge sharing and supports ideational design by making implicit design intent explicit.