基于大语言模型和多智能体协同的齿轮传动智能设计方法研究

doi:10.3901/JME.260490

摘要/Abstract

摘要： 齿轮传动是航空、风电、新能源等领域中广泛应用且不可替代的核心部件，其设计过程涉及材料、摩擦、几何等多学科耦合分析和多约束优化决策。以大语言模型(Large language models, LLM)为代表的人工智能(Artificial intelligence, AI)正在驱动新一轮科技变革，广泛应用于生物医疗与新材料等领域。然而，由于齿轮传动对设计规范和稳健性要求极高，LLM在该领域尚无成熟应用。基于大语言模型和多智能体协同的齿轮传动智能设计方法，采用分层分布式多智能体架构(Hierarchical multi-agent system, HMAS)结合状态有向图和ReAct推理行为方法，实现对复杂齿轮传动设计任务的全流程控制与参数交互，显著提升输出可靠性和可解释性。该方法可实现需求分析、构型推荐、任务推理规划、元件计算校核、系统集成设计、多目标优化、报告生成、CAD生成等全流程设计任务的自主完成，实现涵盖齿轮、轴承、轴等零部件材料、工艺、结构设计与性能评估等800余项参数的整体方案设计周期缩短至10 min以内。通过航空、风电、新能源汽车等领域中多种复杂构型齿轮传动系统案例测试与评估，本文方法设计性能和鲁棒性优于人工设计，且有效改善了目前大模型普遍存在的参数计算不准、设计依据性与探索性不足等问题，综合能力比主流大模型提升23.91%～53.60%，为AI赋能工业设计树立了新范例。

关键词: 齿轮传动, 大语言模型, 多智能体系统, 人工智能

Abstract: Gear transmission is a fundamental component in industries such as aerospace, wind power, and new energy. Its design process requires multidisciplinary coupled analyses and multi-constraint optimization involving materials, friction, geometry, and other factors. Artificial intelligence (AI), particularly large language models (LLMs), is driving a new wave of technological innovation and has been broadly applied in domains such as biomedicine and advanced materials. However, due to the stringent demands for specification adherence and robustness, mature LLM applications in gear transmission design remain undeveloped. This study introduces an intelligent gear transmission design methodology leveraging LLMs and collaborative multi-agent systems, employing a hierarchical multi-agent system (HMAS) architecture integrated with state-oriented graphs and ReAct reasoning methods. This approach enables comprehensive process management and parameter coordination in complex gear transmission design workflows, significantly enhancing output reliability and interpretability. The proposed methodology autonomously executes the entire design workflow—including requirements analysis, configuration recommendation, task reasoning and planning, component calculation and verification, system integration, multi-objective optimization, report generation, and CAD modeling—completing the overall scheme design, which involves over 800 parameters related to component materials, processes, structural design, and performance evaluation, in less than 10 minutes. Evaluation using various complex gear transmission cases in aerospace, wind power, and new energy vehicles demonstrates that the proposed method outperforms traditional manual design in terms of both performance and robustness. Additionally, it effectively resolves prevalent issues observed in mainstream LLMs, such as inaccurate parameter calculations and insufficient innovation and justification in design. Overall, the capability of this approach exceeds that of leading LLMs by 23.91% to 53.60%, establishing a new paradigm for AI-driven industrial design.

Key words: gear transmission, large language models, multi-agent systems, artificial intelligence

中图分类号:

TH132

李扬, 卢泽华, 刘桂源, 朱才朝, 刘怀举. 基于大语言模型和多智能体协同的齿轮传动智能设计方法研究[J]. 机械工程学报, 2026, 62(15): 111-125.

LI Yang, LU Zehua, LIU Guiyuan, ZHU Caichao, LIU Huaiju. Research on Intelligent Design Method for Gear Transmission Based on Large Language Models and Multi-agent Collaboration[J]. Journal of Mechanical Engineering, 2026, 62(15): 111-125.

导出引用管理器 EndNote|Reference Manager|ProCite|BibTeX|RefWorks

链接本文: http://www.cjmenet.com.cn/CN/10.3901/JME.260490

http://www.cjmenet.com.cn/CN/Y2026/V62/I15/111

参考文献

[1] 刘怀举，张博宇，朱才朝，等. 齿轮接触疲劳理论研究进展[J]. 机械工程学报，2022，58(3)：95-120. LIU Huaiju，ZHANG Boyu，ZHU Caichao，et al. State of art of gear contact fatigue theories[J]. Journal of Mechanical Engineering，2022，58(3)：95-120.
[2] CHEN Taimin，ZHU Caichao，CHEN Jinxiao，et al. A review on gear scuffing studies：Theories，experiments and design[J]. Tribology International，2024，196：109741.
[3] LI Yang，LIU Huaiju，CHEN Yiming，et al. Probabilistic gear fatigue life prediction based on physics-informed transformer[J]. Expert Systems with Applications，2024，249：123882.
[4] LU Zehua，CHEN Yiming，LIU Huaiju，et al. A high-power-density design method for polymer gear systems via an adaptive non-dominated sorting genetic algorithm III and surrogate sub-models[J]. Materials & Design，2024，240：112875.
[5] ETESAM Y，CHEONG H，ATAEI M，et al. Deep generative model for mechanical system configuration design[C]//Proceedings of the AAAI Conference on Artificial Intelligence，2025，39(16)：16496-16504.
[6] CAN E，BOZCA M. Optimising the Geometric parameters of a gear in a tractor transmission under constraints using kisssoft[J]. Acta Mechanica et Automatica，2023，17：145-159.
[7] LI Cheng，LI Jianzhong，MU Song，et al. Simulation analysis of the tooth modification about wind power gearbox based on Romax software[J]. IOP Conf Ser：Mater Sci Eng，2020，788：012005.
[8] JAYAKIRAN R，PANDU R. Development of knowledge based parametric CAD modeling system for spur gear：An approach[J]. Alexandria Engineering Journal，2018，57：3139-3149.
[9] LIU Wanyi，CHEN Long，TANG Zhenzhou. large language model aided multi-objective evolutionary algorithm：A low-cost adaptive approach[J]. ArXiv Preprint，2024，arXiv:2410.02301.
[10] FRENKEL ME，EMARA H. ChatGPT-3.5 and -4.0 and mechanical engineering：Examining performance on the FE mechanical engineering and undergraduate exams[J]. Computer Applications in Engineering Education，2024，32：e22781.
[11] LIU Beiming，CUI Zhizhou，HU Siteng，et al. LLM evaluation based on aerospace manufacturing expertise：Automated generation and multi-model question answering[J]. ArXiv Preprint，2025，arXiv:2501.17183.
[12] YAMADA Y，BAO Y，LAMPINEN AK，et al. Evaluating spatial understanding of large language models[J]. ArXiv Preprint，2023，arXiv：2310.14540.
[13] ZHANG Ran，LI Hongwei，QIAN Xinyuan，et al. On large language models safety，security，and privacy：A survey[J]. Journal of Electronic Science and Technology，2025，23：100301.
[14] FERRAG MA，TIHANYI N，DEBBAH M. From LLM Reasoning to autonomous AI agents：A comprehensive review[J]. ArXiv Preprint，2025，arXiv：2504.19678.
[15] PANDEY S，XU R，WANG W，et al. OpenFOAMGPT：A RAG-augmented LLM agent for OpenFOAM-based computational fluid dynamics[J]. Physics of Fluids，2025，37：035120.
[16] 陈庆，柳雨生，段练达，等. 大语言模型融合知识图谱的风电运维问答系统研究[J]. 综合智慧能源，2024，46(9)：61-68. CHEN Qing，LIU Yusheng，DUAN Lianda，et al. Research on a wind power operation and maintenance Q&A system based on large language models and knowledge graphs[J]. Integrated Intelligent Energy，2024，46(9)：61-68.
[17] TAO Laifa，LIU Haifei，NING Guoao，et al. LLM-based framework for bearing fault diagnosis[J]. Mechanical Systems and Signal Processing，2025，224：112127.
[18] ZHANG Hua，ZHANG Meihang，YAN Wei，et al. Energy consumption prediction method for variable conditions machining based on hybrid fine-tuning large model[J]. Computer Integrated Manufacturing Systems，202531(12)：4441.
[19] LIANG Guannan，TONG Qianqian. LLM-Powered AI agent systems and their applications in industry[J]. ArXiv Preprint，2025，arXiv：2505.16120.
[20] YAO Shunyu，ZHAO J，YU Dian，et al. ReAct：Synergizing reasoning and acting in language models[J]. ArXiv Preprint，2023，arXiv：2210.03629.
[21] KANG Y，KIM J. ChatMOF：An artificial intelligence system for predicting and generating metal-organic frameworks using large language models[J]. Nature communications，2024，15(1)：4705.
[22] ZHANG Huan，SONG Yu，HOU Ziyu，et al. HoneyComb：A flexible LLM-based agent system for materials science[J]. ArXiv Preprint，2024，arXiv：2409.00135.
[23] JADHAV Y，FARIMANI AB. Large language model agent as a mechanical designer[J]. ArXiv Preprint，2024，arXiv：2404.17525.
[24] FAN Haolin，HUANG Junlin，XU Jilong，et al. AutoMEX：Streamlining material extrusion with AI agents powered by large language models and knowledge graphs[J]. Materials & Design，2025，251：113644.
[25] YAVARTANOO M，HONG S，NESHATAVAR R，et al. Text2CAD：Text to 3D CAD generation via technical drawings[J]. ArXiv Preprint，2024，arXiv：2411.06206.
[27] WANG Z，LO FP-W，CHEN Q，et al. An LLM-enabled multi-agent autonomous mechatronics design framework[J]. ArXiv Preprint，2025，arXiv：2504.14681.
[28] TAKERNGSAKSIRI W，PASUKSMIT J，THONGT- ANUNAM P，et al. Human-in-the-loop software development agents[J]. ArXiv Preprint，2024，arXiv：2411.12924.
[29] LI Feiyang，FANG Peng，SHI Zhan，et al. CoT-RAG：Integrating chain of thought and retrieval-augmented generation to enhance reasoning in large language models[J]. ArXiv Preprint，2025，arXiv：2504.13534.
[30] 刘怀举，张洪春，魏沛堂，等. 高性能齿轮传动数据库软件设计与开发[J]. 计算机集成制造系统，2023，29(8)：2513-2523. LIU Huaiju，ZHANG Hongchun，WEI Peitang，et al. Design and development of high performance gear transmission database software[J]. Computer Integrated Manufacturing Systems，2023，29(8)：2513-2523.
[31] WU Qingyun，BANSAL G，ZHANG Jieyu，et al. AutoGen：Enabling next-gen LLM applications via multi-agent conversations[J]. ArXiv Preprint，2023，arXiv：2308.08155.
[32] PELLURU K. LangChain & LangGraph in production：Architectures for multi- agent LLM systems[J]. Journal of Data and Digital Innovation，2025，2(3)：1-9.
[33] YANG An，LI Anfeng，YANG Baosong，et al. Qwen3 technical report[J]. ArXiv Preprint，2025，arXiv：2505.09388.
[34] ALHANAHNAH M，HASAN R，XU L，et al. An empirical evaluation of pre-trained large language models for repairing declarative formal specifications[J]. Empirical Software Engineering，2025，30(5)：1-38.
[35] ISO 6336-2019. Calculation of load capacity of spur and helical gears[S]. International Organization for Standardization，2019.
[36] ISO 10300-2023. Calculation of load capacity of bevel gears[S]. International Organization for Standardization，2023.
[37] ISO 281-2007. Rolling bearings-Dynamic load ratings and rating life[S]. International Organization for Standardization，2007.
[38] ISO 16281-2008. Rolling bearings-Methods for calculating the modified reference rating life for universally loaded bearings[S]. International Organization for Standardization，2008.
[39] ISO 76-2006. Rolling bearings-Static load ratings[S]. International Organization for Standardization，2006.
[40] DIN 743-2012. Calculation of load capacity of shafts and axles[S]. German Institute for Standardization，2012.
[41] ISO 4156-2005. Straight cylindrical involute splines-Metric module，side fit[S]. International Organization for Standardization，2005.
[42] GB/T 17855-2017. 花键承载能力计算方法[S]. 中国国家标准化管理委员会，2017. GB/T 17855-2017. Calculation of load capacity of spline[S]. State Administration of the People’s Republic of China，2017.
[43] DIN 6885-2021. Drive type fastenings without taper action，parallel keys，keyways-Deep pattern[S]. German Institute for Standardization，2021.
[44] 成大先. 机械设计手册：第6版[M]. 北京：化学工业出版社，2017. CHENG Daxian. Mechanical design handbook (in Chinese). 6th ed[M]. Beijing：Chemical Industry Press; 2017.
[45] BEDAKA AK，LIN C-Y. CAD-based robot path planning and simulation using OPEN CASCADE[J]. Procedia Computer Science，2018，133：779-785.
[46] GB/T 10609.1-2008. 技术制图标题栏[S]. 中国国家标准化管理委员会，2008. GB/T 10609.1-2008. Technical drawings-Title blocks[S]. State Administration of the People’s Republic of China，2008.
[47] REIN D，HOU BL，STICKLAND AC，et al. GPQA：A graduate-level google-proof q&a benchmark[C]//First Conference on Language Modeling，2024.