萃智理論在加速產品開發與解決複雜工程問題上具備強大潛力，然而其高度的抽象性、高度的知識發想與陡峭的學習曲線，使得多數缺乏專業背景之工程師與管理者在實務導入時面臨極大困難。傳統的試錯法缺乏方向，而單純應用大型語言模型則易產生缺乏工程依據的「幻覺」或虛幻解。為解決上述困境，本研究旨在研究如何整合大型語言模型與萃智理論之創新流程輔助系統。此系統基於 OpenAI ChatGPT GPTs 平台開發，採用無程式碼設計，將其定位為「虛擬創新顧問」。
在研究方法上，系統於流程前端導入六何法與魚骨圖等系統性提問框架，輔助使用者釐清模糊情境；隨後透過「思維鏈」提示詞與內建知識庫（包含 39x39 矛盾矩陣與 40 項發明原則），引導使用者將具體問題轉化為標準技術或物理矛盾，並推導出具學理依據之解法方向。為驗證系統成效，本研究選取「不完美防水拉鍊之產品設計」與「射出成型良率之製程改善」雙案例進行實測。研究創新性地將「類似問題類似解」原則與「專利發想」機制納入系統後端，強制 AI 搜尋並對接具備實體機構的真實設備專利。結果顯示，本系統不僅能有效收斂問題，更成功將 TRIZ 產出的抽象概念轉化為工程上可直接評估與落地的「真實解」。
總結而言，本研究所建置之萃智理論人工智慧代理系統，成功將繁複的創新方法論轉化為對話式的智能引導流程。不僅大幅降低了萃智理論的學習門檻，實現了「創新民主化」，更為企業提供了一套低成本、高效率且具防呆機制的實務研發輔助工具，替未來人機協同創新與工程管理實踐開創了新的應用典範。

TRIZ (Theory of Inventive Problem Solving) demonstrates strong potential in accelerating product development and addressing complex engineering problems. However, its high level of abstraction, heavy reliance on knowledge generation, and steep learning curve make it difficult for engineers and managers without specialized backgrounds to adopt in practice. Traditional trial-and-error approaches lack clear direction, while the direct use of large language models (LLMs) often leads to “hallucinations” or unrealistic solutions that lack engineering validity. To address these challenges, this study aims to study an innovation-support system that integrates LLMs with TRIZ methodology.
The system is developed on the OpenAI ChatGPT GPTs platform using a no-code approach and is positioned as a “virtual innovation consultant.” In terms of methodology, the system introduces structured questioning frameworks—such as the 5W1H method and fishbone diagrams—at the front end to help users clarify ambiguous situations. Subsequently, through chain-of-thought prompting and an embedded knowledge base (including the 39×39 contradiction matrix and the 40 inventive principles), the system guides users in transforming specific problems into standard technical or physical contradictions, thereby deriving theoretically grounded solution directions.
To validate the effectiveness of the system, two case studies were conducted: the product design of an imperfect waterproof zipper and process improvement for injection molding yield. This study innovatively integrates the principle of “similar problems, similar solutions” with a patent ideation mechanism in the backend, forcing the AI to retrieve and align with real-world equipment patents that contain concrete physical implementations. The results demonstrate that the system not only effectively converges problem definitions but also successfully transforms abstract TRIZ concepts into practical, engineering-feasible “real solutions."
In conclusion, the TRIZ-based AI agent developed in this study successfully converts complex innovation methodologies into an interactive, dialogue-driven guidance process. It significantly lowers the learning barrier of TRIZ and realizes the concept of “democratizing innovation.” Moreover, it provides enterprises with a low-cost, high-efficiency, and error-proof R&D support tool, opening new possibilities for human-AI collaborative innovation and engineering management practices.

[1]World Intellectual Property Organization (2024). World Intellectual Property Report 2024: Making innovation policy work for development. World Intellectual Property Organization. 取自https://www.wipo.int/publications/en/details.jsp?id=4746. 最後瀏覽日：2026/04/21。
[2]Altshuller, G. (1984). Creativity as an Exact Science: The Theory of the Solution of Inventive Problems. Gordon and Breach, London, UK.
[3]OpenAI. (2023). GPTs: Custom Versions of ChatGPT. GPT-4 Technical Report.
[4]Shin, W.-S., Lee, S.-H., & Sue, H.-J. (2024). Fostering Tech Innovation: Exploring TRIZ and ChatGPT Integration for Developer and Entrepreneur Challenges. Journal of Innovation & Knowledge, 18(4), 588–597.
[5]Szczepanik, K., & Chudziak, J. A. (2025). TRIZ Agents: A Multi-Agent LLM Approach for TRIZ-Based Innovation. Proceedings of the 17th International Conference on Agents and Artificial Intelligence (ICAART), 1, 196-207.
[6]Jiang, S., & Luo, J. (2025). AutoTRIZ: Artificial Ideation with TRIZ and Large Language Models. Advanced Engineering Informatics, 65, 103312.
[7]Mohammadi, A., & Zeng, Y. (2025). Enhancing TRIZ through Environment-based Design Methodology Supported by a Large Language Model. AI EDAM: Artificial Intelligence for Engineering Design, Analysis and Manufacturing, 39, e12, pp. 1–18.
[8]WIPO (2024). World Intellectual Property Report 2024: Making innovation policy work for development. World Intellectual Property Organization. 取自 https://www.wipo.int/publications/en/details.jsp?id=4746 最後瀏覽日：2026/04/21.
[9]Designorate, TRIZ Method for Problem Solving. 取自 https://www.designorate.com/triz-method-problem-solving/。最後瀏覽日：2026/04/20。
[10]Ishikawa, K. (1986). Guide to Quality Control. Asian Productivity Organization, Tokyo, Japan..
[11]Weng, L. (2023, June 23). LLM Powered Autonomous Agents. Lil'Log. https://lilianweng.github.io/posts/2023-06-23-agent。最後瀏覽日:2026/04/20。
[12]Schick, T., et al. (2023). Toolformer: Language models can teach themselves to use tools. Advances in Neural Information Processing Systems, 36, 6853-6871.
[13]Shinn, N., Labash, B., & Gopinath, A. (2023). Reflexion: Language agents with iterative self-reflection and feedback. Advances in Neural Information Processing Systems, 36, 9321-9343.
[14]Chen, L., et al. (2024). TRIZ-GPT: An LLM-Augmented Method for Problem-Solving. Proceedings of the ASME IDETC-CIE 2024, Washington, DC, USA.
[15]黃文正（2018）。運用萃智理論提升顯示器製造廠材料管理之彈性-以凌暉科技股份有限公司為例。國立成功大學工程管理碩士在職專班
[16]陳俊銘（2025）。應用萃智方法提升射出成形良率之改善。國立成功大學工程管理碩士在職專班
[17]Lee, J. H. (2021). Waterproof zipper with improved sealing structure. Korean Patent No. 10-2261012. 取自 https://patents.google.com/patent/KR102261012B1/en. 最後瀏覽日：2026/04/21.
[18]潘沂泓（2018）。運用 TRIZ 理論提升廢硫酸回收效率。國立成功大學工程管理碩士在職專班
[19]楊志輝（2018）。運用萃智理論探討鐵道列車熱軸箱感測器功能改善。國立成功大學工程管理碩士在職專班
[20]Wang, L., et al. (2024). A survey on large language model based autonomous agents. Frontiers of Computer Science, 18(6), 186345.