The Heuristic Design Innovation Approach for Data-Integrated Large Language Model

In an era characterized by the relentless emergence of big data and the continuous evolution of artificial intelligence, the traditional design chain is undergoing a significant reconstruction. The performance of existing general-purpose large language models (LLMs) in specific domains falls considerably short of expectations. This study aims to invigorate the utilization of a vast amount of high-quality design data, integrating multimodal methods to deeply embed AI into the design process. Taking industrial design as a case study, this research selected databases of entries from three internationally recognized design awards, encompassing key data fields such as names of works, award statuses, and design descriptions, totaling 84,773 design data entries. Through extracting and analyzing design data, this research fully exploits the data resources of the design and intelligent manufacturing industry, constructing a heuristic design innovation method that incorporates information from award-winning works. By fusing design data with LLM, this study developed DIABot, a heuristic design innovation tool based on LLMs, inspired by the ReAct method. Combining extensive design data, DIABot offers design guidance through data retrieval integration. Furthermore, an evaluation test involving nine participants was organized to compare DIABot with human designers and ChatGPT in terms of design assistance, with the results affirming DIABot’s effectiveness in supporting design tasks. This research provides valuable insights and explorations into the application of AI in the design field, offering designers more comprehensive, refined, and precise support, thereby promoting the deeper development and application of artificial intelligence in industrial design.