一种生成式船舶薄板产线数字孪生建模方法

doi:10.16183/j.cnki.jsjtu.2025.296

摘要/Abstract

摘要：

在船舶制造中,薄板产线工序复杂、设备多样且数据源异构,传统数字孪生建模依赖人工定义接口与属性,效率低且一致性不足.提出一种面向薄板产线的生成式数字孪生建模方法,通过构建涵盖工艺单元、设备、属性、感控单元与功能行为的5类核心本体,形成统一的语义结构;并设计本体到数字孪生定义语言(DTDL)的映射机制,实现领域知识的标准化表达.进一步引入大语言模型,利用其语义理解与生成能力,将自然语言描述自动转化为符合 DTDL v3 规范的结构化模型,并通过语义一致性校验确保模型逻辑与约束的正确性.实验结果表明,该方法在建模效率、语义准确性和属性覆盖度方面较人工方法分别提升约79.6%、9.4%和28.5%,显著提升了数字孪生建模的自动化与规范化水平.

关键词: 数字孪生, 大语言模型, 数字孪生定义语言, 本体

Abstract:

In shipbuilding, thin-plate production lines are characterized by complex processes, diverse equipment, and heterogeneous data sources. Traditional digital twin modeling relies heavily on manually defined interfaces and attributes, resulting in low efficiency and limited consistency. A generative digital twin modeling method tailored for thin-plate production lines is proposed. This method forms a unified semantic structure by constructing five core ontologies, covering process units, equipment, attributes, sensing-control units, and functional behaviors. Furthermore, a mapping mechanism from ontology to the digital twins definition language (DTDL) is designed to enable standardized representation of domain knowledge. A large language model (LLM) is then introduced to leverage its semantic understanding and generative capabilities, automatically transforming natural language descriptions into structured models compliant with the DTDL v3 specification. Semantic consistency validation is performed to ensure the correctness of model logic and constraints. Experimental results demonstrate that, compared with manual approaches, the proposed method improves modeling efficiency, semantic accuracy, and attribute coverage by approximately 79.6%, 9.4%, and 28.5%, respectively. This significantly enhances the automation and standardization of digital twin modeling.

Key words: digital twin, large language model (LLM), digital twins definition language (DTDL), ontology

中图分类号:

TH181

温晓健, 严哲, 胡佐治, 袁轶, 鲍劲松, 张丹. 一种生成式船舶薄板产线数字孪生建模方法[J]. 上海交通大学学报, 2026, 60(8): 1279-1289.

WEN Xiaojian, YAN Zhe, HU Zuozhi, YUAN Yi, BAO Jinsong, ZHANG Dan. A Generative Digital Twin Modeling Method for Ship Thin-Plate Production Lines[J]. Journal of Shanghai Jiao Tong University, 2026, 60(8): 1279-1289.

图/表 11

图1

表1

图2

图3

图4

图5

图6

图7

表2

图8

表3

参考文献 24

[1]	TAO F, XIAO B, QI Q L, et al. Digital twin modeling[J]. Journal of Manufacturing Systems, 2022, 64: 372-389. doi: 10.1016/j.jmsy.2022.06.015 URL
[2]	沈慧, 刘世民, 许敏俊, 等. 面向加工领域的数字孪生模型自适应迁移方法[J]. 上海交通大学学报, 2022, 56(1): 70-80. doi: 10.16183/j.cnki.jsjtu.2021.167
	SHEN Hui, LIU Shimin, XU Minjun, et al. Adaptive transferring method of digital twin model for machining domain[J]. Journal of Shanghai Jiao Tong University, 2022, 56(1): 70-80.
[3]	LIU X, JIANG D, TAO B, et al. A systematic review of digital twin about physical entities, virtual models, twin data, and applications[J]. Advanced Engineering Informatics, 2023, 55: 101876. doi: 10.1016/j.aei.2023.101876 URL
[4]	LIU J F, ZHANG Y M, LIU Z Y, et al. Digital twins enable shipbuilding[J]. Alexandria Engineering Journal, 2024, 107: 915-931. doi: 10.1016/j.aej.2024.09.007 URL
[5]	金轩铖, 洪舸, 高硕, 等. 面向船舶大型曲面薄板的装配形变TSM-TLHS预测方法[J]. 上海交通大学学报, 2025, 59(8): 1092-1102. doi: 10.16183/j.cnki.jsjtu.2023.576
	JIN Xuancheng, HONG Ge, GAO Shuo, et al. TSM-TLHS prediction method for assembly deformation of large curved thin plates in shipbuilding[J]. Journal of Shanghai Jiao Tong University, 2025, 59(8): 1092-1102.
[6]	杨润党, 续爱民, 甄希金, 等. 船舶薄板平面分段产线技术发展及应用现状[J]. 船舶工程, 2022, 44(10): 16-23.
	YANG Rundang, XU Aimin, ZHEN Xijin, et al. Development and application status of ship thin plate plane sectional production line technology[J]. Ship Engineering, 2022, 44(10): 16-23.
[7]	WEN X J, SUN Y C, LIU S M, et al. Fine-grained decomposition of complex digital twin systems driven by semantic-topological-dynamic associations[J]. Journal of Manufacturing Systems, 2024, 77: 780-797. doi: 10.1016/j.jmsy.2024.10.023 URL
[8]	SONG J Q, LIU S M, MA T L, et al. Resilient digital twin modeling: A transferable approach[J]. Advanced Engineering Informatics, 2023, 58: 102148. doi: 10.1016/j.aei.2023.102148 URL
[9]	SU C, HAN Y, TANG X, et al. Knowledge-based digital twin system: Using a knowledge-driven approach for manufacturing process modeling[J]. Computers in Industry, 2024, 159/160: 104101. doi: 10.1016/j.compind.2024.104101 URL
[10]	HAN X, LIU X Y, WANG H H, et al. A digital twin modeling approach cooperating design and manufacturing using knowledge graph and multi-physics field modelling[J]. The International Journal of Advanced Manufacturing Technology, 2025, 137(7/8): 3423-3442. doi: 10.1007/s00170-025-15230-6
[11]	JIA W J, WANG W, ZHANG Z Z. From simple digital twin to complex digital twin Part I: A novel modeling method for multi-scale and multi-scenario digital twin[J]. Advanced Engineering Informatics, 2022, 53: 101706. doi: 10.1016/j.aei.2022.101706 URL
[12]	JIA W J, WANG W, ZHANG Z Z. From simple digital twin to complex digital twin part II: Multi-scenario applications of digital twin shop floor[J]. Advanced Engineering Informatics, 2023, 56: 101915. doi: 10.1016/j.aei.2023.101915 URL
[13]	WU Y T, SUN Y C, WEN X J, et al. A generative modeling method for digital twin shop floor[J]. IEEE Internet Computing, 2025, 29(1): 24-31.
[14]	宋吉群. 大规模异构数字孪生资产快速建模与集成方法研究[D]. 上海: 东华大学, 2024.
	SONG Jiqun. Research on rapid modeling and integration method for large-scale heterogeneous digital twin assets[D]. Shanghai: Donghua University, 2024.
[15]	宁方伟, 鲁嘉星, 王一轩, 等. 智能生成式设计: 一种新的机械设计理念[J]. 机械工程学报, 2025, 61(24): 267-284.
	NING Fangwei, LU Jiaxing, WANG Yixuan, et al. Intelligent generative design: A new mechanical design concept[J]. Journal of Mechanical Engineering, 2025, 61(24): 267-284.
[16]	LIN X S, CHEN W Q, ZHOU Z, et al. A five-dimensional digital twin framework driven by large language models-enhanced RL for CNC systems[J]. Robotics and Computer-Integrated Manufacturing, 2025, 95: 103009. doi: 10.1016/j.rcim.2025.103009 URL
[17]	YIN S K, FU C Y, ZHAO S R, et al. A survey on multimodal large language models[J]. National Science Review, 2024, 11(12): nwae403.
[18]	MEN C H, HAN Y, WANG P, et al. The interpretable reasoning and intelligent decision-making based on event knowledge graph with LLMs in fault diagnosis scenarios[J]. IEEE Transactions on Instrumentation and Measurement, 2025, 74: 1-16.
[19]	SHI D C, LIEDL P, BAUERNHANSL T. Interoperable information modelling leveraging asset administration shell and large language model for quality control toward zero defect manufacturing[J]. Journal of Manufacturing Systems, 2024, 77: 678-696. doi: 10.1016/j.jmsy.2024.10.011 URL
[20]	WANG Y X, WU Z Y, YAO J F, et al. TDAG: A multi-agent framework based on dynamic Task Decomposition and Agent Generation[J]. Neural Networks, 2025, 185: 107200. doi: 10.1016/j.neunet.2025.107200 URL
[21]	XIONG X R, CAI H M, YU H, et al. DR-RAG: Domain-Rule-based Retrieval-Augmented Generation for aviation digital model design[J]. Advanced Engineering Informatics, 2025, 68: 103688. doi: 10.1016/j.aei.2025.103688 URL
[22]	ZHANG Q, XU C, LI J, et al. LLM-TSFD: An industrial time series human-in-the-loop fault diagnosis method based on a large language model[J]. Expert Systems with Applications, 2025, 264: 125861. doi: 10.1016/j.eswa.2024.125861 URL
[23]	胡天翔, 叶辉, 杨晓飞. 基于数字孪生的船舶制造车间监视系统的构建[J]. 系统仿真学报, 2025, 37(2): 517-528. doi: 10.16182/j.issn1004731x.joss.23-1175
	HU Tianxiang, YE Hui, YANG Xiaofei. Construction of a digital twin-based ship manufacturing workshop monitoring system[J]. Journal of System Simulation, 2025, 37(2): 517-528. doi: 10.16182/j.issn1004731x.joss.23-1175
[24]	ISO 15926-2: 2003 Industrial automation systems and integration-Integration of life-cycle data for process plants including oil and gas production facilities-Part 2: Data model[S].

本体类别	对应DTDL元素	映射说明
工艺单元本体	Interface, Relationship	工艺单元映射为接口,工序依赖映射为关系,保证工艺链条语义完整
设备本体	Interface, Property, Telemetry, Component, Relationship	设备为接口,静态/配置参数为属性,动态监测为遥测,可复用模块为组件,设备连接为关系
感控单元本体	Telemetry, Command	传感器信号建模为遥测,控制命令建模为命令
功能行为本体	Command, Component	原子行为映射为命令(一次性动作、无内部拆分),复合行为或可复用逻辑映射为组件(由多个动作组成、需在多处复用)
属性本体	Property, Telemetry	静态/配置参数为属性,实时测量为遥测;复杂对象属性映射为 Object schema

评估维度	指标说明	人工建模方法	LLM驱动建模框架(平均)	相对提升幅度/%
建模效率	平均完成一个完整DTDL模型所需时间/min	105.3	21.5	79.6
实体识别准确率	识别出的核心构件实体与预期实体的匹配率/%	83.7	91.6	9.4
属性覆盖度	所识别实体中包含的属性字段对预定义本体的覆盖比/%	68.4	87.9	28.5
模型一致性评分	专家打分,衡量模型结构闭合性和语义一致性(满分10)	6.8	8.4	23.5

LLM	语义完整性/%	结构准确性/%	本体一致性/%	平均响应时间/s
ChatGPT-4	93.2	91.5	96.7	10.3
DeepSeek-Coder- 33B-Instruct	92.6	87.1	92.3	6.9
Qwen-72B-Chat	90.1	87.4	94.8	9.8