Journal of Manufacturing Systems最新文献_第3页

Agentic digital twin-embedded maintenance methodology for energy equipment: A self-evolving operational paradigm 能源设备的代理数字双嵌入式维护方法：一种自进化的操作范式

IF 14.2 1区工程技术 Q1 ENGINEERING, INDUSTRIAL

Journal of Manufacturing Systems

Pub Date : 2025-12-09 DOI: 10.1016/j.jmsy.2025.11.020

Wang Cong, Wu Tao, Bao Jinsong

Intelligent operation and maintenance of energy equipment represents a critical component in ensuring the stable performance of new-generation power systems. Faced with complex operational conditions and nonlinear fault characteristics, conventional manual maintenance suffers from delayed perception and ambiguous causality. However, while digital twin technology can establish a virtual-real interaction space, its static modeling approach exhibits prediction failure in dynamic scenarios. To address these challenges, this study proposes an intelligent maintenance methodology based on cognitive agents and virtual-real co-evolution: constructing a dynamic environment representation model to achieve spatiotemporal feature correlation of equipment states and operational condition migration; designing a memory-planning-decision architecture to enhance causal reasoning capabilities for equipment faults and integrating with digital twin models for virtual-real interaction. The methodology is validated through an 18-month case study of a gas-steam boiler in a combined heat and power plant, utilizing 5.2 million historical operational records. Experimental results demonstrate that this approach achieves a 97.3 % accuracy rate in diagnosing non-stationary faults for gas-steam boiler equipment, realizes a 20-fold improvement in knowledge update time (from 48 to 2.3 h), and attains significant performance enhancements including 31.2 % cost efficiency improvement, 3-fold early warning lead time extension (from 24 to 72 h), and 16.2 % overall collaborative performance improvement (from 82.2 % to 95.5 %). The research validates the engineering value of dynamic cognitive paradigms in intelligent maintenance of power equipment, providing a feasible solution for autonomous decision-making in high-real-time scenarios.

能源设备的智能运维是保证新一代电力系统稳定运行的重要组成部分。面对复杂的运行条件和非线性的故障特征，传统的人工维修存在感知延迟和因果关系模糊的问题。然而，虽然数字孪生技术可以建立一个虚拟-真实的交互空间，但其静态建模方法在动态场景中表现出预测失败。针对这些挑战，本研究提出了一种基于认知代理和虚实协同进化的智能维护方法：构建动态环境表示模型，实现设备状态的时空特征关联和运行状态迁移；设计记忆-规划-决策体系结构，增强设备故障的因果推理能力，并与数字孪生模型集成，实现虚实交互。通过对一家热电联产电厂的燃气蒸汽锅炉进行为期18个月的案例研究，利用520万份历史运行记录，验证了该方法的有效性。实验结果表明，该方法对燃气蒸汽锅炉设备非平稳故障的诊断准确率达到97.3% %，知识更新时间从48到2.3 h提高了20倍，成本效率提高了31.2% %，预警提前期延长了3倍（从24到72 h），整体协同性能提高了16.2% %（从82.2 %到95.5 %）。研究验证了动态认知范式在电力设备智能维护中的工程价值，为高实时场景下的自主决策提供了可行的解决方案。

{"title":"Agentic digital twin-embedded maintenance methodology for energy equipment: A self-evolving operational paradigm","authors":"Wang Cong, Wu Tao, Bao Jinsong","doi":"10.1016/j.jmsy.2025.11.020","DOIUrl":"10.1016/j.jmsy.2025.11.020","url":null,"abstract":"<div><div>Intelligent operation and maintenance of energy equipment represents a critical component in ensuring the stable performance of new-generation power systems. Faced with complex operational conditions and nonlinear fault characteristics, conventional manual maintenance suffers from delayed perception and ambiguous causality. However, while digital twin technology can establish a virtual-real interaction space, its static modeling approach exhibits prediction failure in dynamic scenarios. To address these challenges, this study proposes an intelligent maintenance methodology based on cognitive agents and virtual-real co-evolution: constructing a dynamic environment representation model to achieve spatiotemporal feature correlation of equipment states and operational condition migration; designing a memory-planning-decision architecture to enhance causal reasoning capabilities for equipment faults and integrating with digital twin models for virtual-real interaction. The methodology is validated through an 18-month case study of a gas-steam boiler in a combined heat and power plant, utilizing 5.2 million historical operational records. Experimental results demonstrate that this approach achieves a 97.3 % accuracy rate in diagnosing non-stationary faults for gas-steam boiler equipment, realizes a 20-fold improvement in knowledge update time (from 48 to 2.3 h), and attains significant performance enhancements including 31.2 % cost efficiency improvement, 3-fold early warning lead time extension (from 24 to 72 h), and 16.2 % overall collaborative performance improvement (from 82.2 % to 95.5 %). The research validates the engineering value of dynamic cognitive paradigms in intelligent maintenance of power equipment, providing a feasible solution for autonomous decision-making in high-real-time scenarios.</div></div>","PeriodicalId":16227,"journal":{"name":"Journal of Manufacturing Systems","volume":"84 ","pages":"Pages 100-116"},"PeriodicalIF":14.2,"publicationDate":"2025-12-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145786761","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

End-to-end multimodal knowledge graph construction for industrial exploded views via attention-guided expert chains 基于注意力引导专家链的工业爆炸视图端到端多模态知识图谱构建

IF 14.2 1区工程技术 Q1 ENGINEERING, INDUSTRIAL

Journal of Manufacturing Systems

Pub Date : 2025-12-05 DOI: 10.1016/j.jmsy.2025.10.013

Xinxin Liang , Zuoxu Wang , Mingrui Li , Chun-Hsien Chen , Jihong Liu

Industrial exploded views (IEVs) integrate images, text, and part–assembly relations, which is essential for advancing intelligent manufacturing. However, semantic ambiguities, structural inconsistencies, and fragmented annotations hinder effective knowledge extraction and reuse. We cast extraction from IEVs as constrained inference over scene graphs and present a Scene-aware Cascade Expert Chain (SACEC) that incrementally resolves entities, relations, and assembly context. A Visual–Structural–Rule (VSR) validator then enforces domain rules and semantic consistency on every triple. A dynamic triple-cutting strategy selects credible triples by jointly balancing local evidence, contextual coherence, and assembly order, yielding a multimodal knowledge graph (MMKG). We also introduce the Industrial Exploded-View (IEV) dataset, with fine-grained component and relation annotations and assembly-order metadata. Experiments on VRD, VG150, and the IEV dataset demonstrate significant improvements over state-of-the-art baselines, achieving R@100 of 73.2%, 63.9%, and 67.4%, and TripleAcc of 31.8%, 20.2%, and 24.9%. At the triple level, we further obtain P@100 of 54.9%, 39.8%, and 49.6%, and F1@100 of 46.2%, 34.1%, and 45.1%. Against strong path- and context-based baselines, our method improves by up to +7.4 pp in recall@100, +2.7 pp in TripleAcc, +15.8 pp in Precision@100, and +13.5 pp in F1@100. The approach reduces manual annotation and yields interpretable, audit-ready outputs for intelligent design and process planning, offering a practical route to automated and interpretable knowledge extraction in industrial environments.

工业爆炸视图集成了图像、文本和零部件关系，对推进智能制造至关重要。然而，语义歧义、结构不一致和碎片化的注释阻碍了有效的知识提取和重用。我们将evs提取作为场景图上的约束推理，并提出了一个场景感知级联专家链（SACEC），该链可以增量地解析实体、关系和装配上下文。然后，可视化结构规则（VSR）验证器对每个三元组强制执行域规则和语义一致性。动态三重切割策略通过联合平衡局部证据、上下文一致性和装配顺序来选择可信的三元组，从而产生多模态知识图（MMKG）。我们还介绍了工业爆炸视图（IEV）数据集，该数据集具有细粒度的组件和关系注释以及装配顺序元数据。在VRD、VG150和IEV数据集上的实验表明，与最先进的基线相比，有了显著的改进，R@100的效率分别为73.2%、63.9%和67.4%，TripleAcc的效率分别为31.8%、20.2%和24.9%。在三重水平上，我们进一步得到P@100为54.9%、39.8%和49.6%，F1@100为46.2%、34.1%和45.1%。对于基于路径和上下文的强基线，我们的方法在recall@100中提高了+7.4 pp，在TripleAcc中提高了+2.7 pp，在Precision@100中提高了+15.8 pp，在F1@100中提高了+13.5 pp。该方法减少了手工注释，并为智能设计和过程规划提供了可解释的、可审计的输出，为工业环境中自动化和可解释的知识提取提供了一条实用的途径。

{"title":"End-to-end multimodal knowledge graph construction for industrial exploded views via attention-guided expert chains","authors":"Xinxin Liang , Zuoxu Wang , Mingrui Li , Chun-Hsien Chen , Jihong Liu","doi":"10.1016/j.jmsy.2025.10.013","DOIUrl":"10.1016/j.jmsy.2025.10.013","url":null,"abstract":"<div><div>Industrial exploded views (IEVs) integrate images, text, and part–assembly relations, which is essential for advancing intelligent manufacturing. However, semantic ambiguities, structural inconsistencies, and fragmented annotations hinder effective knowledge extraction and reuse. We cast extraction from IEVs as constrained inference over scene graphs and present a Scene-aware Cascade Expert Chain (SACEC) that incrementally resolves entities, relations, and assembly context. A Visual–Structural–Rule (VSR) validator then enforces domain rules and semantic consistency on every triple. A dynamic triple-cutting strategy selects credible triples by jointly balancing local evidence, contextual coherence, and assembly order, yielding a multimodal knowledge graph (MMKG). We also introduce the Industrial Exploded-View (IEV) dataset, with fine-grained component and relation annotations and assembly-order metadata. Experiments on VRD, VG150, and the IEV dataset demonstrate significant improvements over state-of-the-art baselines, achieving R@100 of 73.2%, 63.9%, and 67.4%, and TripleAcc of 31.8%, 20.2%, and 24.9%. At the triple level, we further obtain P@100 of 54.9%, 39.8%, and 49.6%, and F1@100 of 46.2%, 34.1%, and 45.1%. Against strong path- and context-based baselines, our method improves by up to +7.4pp in recall@100, +2.7pp in TripleAcc, +15.8pp in Precision@100, and +13.5pp in F1@100. The approach reduces manual annotation and yields interpretable, audit-ready outputs for intelligent design and process planning, offering a practical route to automated and interpretable knowledge extraction in industrial environments.</div></div>","PeriodicalId":16227,"journal":{"name":"Journal of Manufacturing Systems","volume":"84 ","pages":"Pages 85-99"},"PeriodicalIF":14.2,"publicationDate":"2025-12-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145682162","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Designing Synthetic Active Learning for model refinement in manufacturing parts detection 面向制造零件检测模型精化的综合主动学习设计

IF 14.2 1区工程技术 Q1 ENGINEERING, INDUSTRIAL

Journal of Manufacturing Systems

Pub Date : 2025-12-04 DOI: 10.1016/j.jmsy.2025.11.023

Xiaomeng Zhu , Jacob Henningsson , Pär Mårtensson , Lars Hanson , Mårten Björkman , Atsuto Maki

This paper introduces Synthetic Active Learning (SAL), a fully automatic model refinement strategy for manufacturing parts detection using only synthetic data actively generated with domain randomization for training. SAL iteratively updates the detection model by identifying its weaknesses, such as in specific categories, materials, or object sizes, using custom evaluators, and generating targeted synthetic data to address them; it selectively synthesizes new useful data with respect to active learning, where traditionally humans in the loop select data to label. During each iteration, model training and data generation occur simultaneously to improve efficiency. Evaluated on four use cases from two industrial datasets, SAL achieved mAP@50 improvements of 2 to 6% percentage points over static learning, which refers to training on a fixed, pre-generated dataset. It also showed notable gains in underperforming categories, leading to more balanced performance across classes. Another benefit is that it uses a consistent configuration across multiple use cases, avoiding the need for extensive hyperparameter tuning common in prior domain randomization studies. Given its encouraging performance across diverse scenarios, we believe that SAL can scale to broader industrial applications where training can be fully or mostly based on synthetic data.

本文介绍了一种基于领域随机化主动生成的合成数据进行训练的制造零件检测的全自动模型优化策略——合成主动学习（SAL）。SAL迭代地更新检测模型，通过识别它的弱点，例如在特定的类别，材料，或对象大小，使用自定义评估器，并生成目标合成数据来解决它们；相对于主动学习，它有选择地合成新的有用数据，传统上，人类在循环中选择数据来标记。在每次迭代中，模型训练和数据生成同时进行，以提高效率。通过对来自两个工业数据集的四个用例进行评估，SAL实现了mAP@50比静态学习提高了2到6%的百分点，静态学习指的是在固定的、预先生成的数据集上进行训练。在表现不佳的类别中也显示出显著的进步，导致各个类别的表现更加平衡。另一个好处是，它在多个用例中使用一致的配置，避免了在先前的领域随机化研究中常见的大量超参数调优的需要。考虑到它在不同场景中令人鼓舞的表现，我们相信SAL可以扩展到更广泛的工业应用，在这些应用中，训练可以完全或主要基于合成数据。

{"title":"Designing Synthetic Active Learning for model refinement in manufacturing parts detection","authors":"Xiaomeng Zhu , Jacob Henningsson , Pär Mårtensson , Lars Hanson , Mårten Björkman , Atsuto Maki","doi":"10.1016/j.jmsy.2025.11.023","DOIUrl":"10.1016/j.jmsy.2025.11.023","url":null,"abstract":"<div><div>This paper introduces Synthetic Active Learning (SAL), a fully automatic model refinement strategy for manufacturing parts detection using only synthetic data actively generated with domain randomization for training. SAL iteratively updates the detection model by identifying its weaknesses, such as in specific categories, materials, or object sizes, using custom evaluators, and generating targeted synthetic data to address them; it selectively synthesizes new useful data with respect to active learning, where traditionally humans in the loop select data to label. During each iteration, model training and data generation occur simultaneously to improve efficiency. Evaluated on four use cases from two industrial datasets, SAL achieved mAP@50 improvements of 2 to 6% percentage points over static learning, which refers to training on a fixed, pre-generated dataset. It also showed notable gains in underperforming categories, leading to more balanced performance across classes. Another benefit is that it uses a consistent configuration across multiple use cases, avoiding the need for extensive hyperparameter tuning common in prior domain randomization studies. Given its encouraging performance across diverse scenarios, we believe that SAL can scale to broader industrial applications where training can be fully or mostly based on synthetic data.</div></div>","PeriodicalId":16227,"journal":{"name":"Journal of Manufacturing Systems","volume":"84 ","pages":"Pages 68-84"},"PeriodicalIF":14.2,"publicationDate":"2025-12-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145682163","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Dynamic data driven uncertain physical information self-awareness method for the aircraft composite component assembly system 飞机复合材料装配系统的动态数据驱动不确定物理信息自感知方法

IF 14.2 1区工程技术 Q1 ENGINEERING, INDUSTRIAL

Journal of Manufacturing Systems

Pub Date : 2025-12-01 DOI: 10.1016/j.jmsy.2025.11.001

Pengbo Yin, Yang Zhang, Jiacheng Cui, Jiangtao Zhao, Yulin Jin, Qihang Chen, Yongkang Lu, Wei Liu

High-precision assembly of large composite components is crucial for aircraft structural safety. To address geometric deviation of components caused by anisotropic material behaviors and time-varying process constraints during dynamic assembly, this paper proposes a dynamic data driven uncertain physical information self-awareness (DDDPIA) method. This approach accurately updates physical models of components by integrating dynamic data of the manufacturing process containing displacement information, load configurations, and model information through three key innovations: (1) A simplified affine mapping method from model parameters to system stiffness that decouples material properties from process constraints in deformation modeling. (2) A multi-source prior data-driven model parameter optimization framework enabling efficient identification of material parameters and process constraints while quantifying measurement uncertainty impacts and maintaining high-precision performance with measurement errors below 0.3 mm. (3) An industrial application-oriented shape regulation platform that leverages the updated physical model for precise load inversion to achieve specified shapes of composite components. Experimental and simulation results verify over 64% displacement error reduction relative to uncalibrated static modeling, while load inversion with sub-0.2 N solution accuracy achieves geometric deviations correction of components. This establishes a closed-loop measurement-data-model-assimilation paradigm, enhancing decision autonomy in aviation intelligent manufacturing systems.

大型复合材料部件的高精度装配对飞机结构安全至关重要。针对动态装配过程中由于材料各向异性行为和时变工艺约束导致的部件几何偏差，提出了一种动态数据驱动的不确定物理信息自我意识（DDDPIA）方法。该方法通过三个关键创新，将包含位移信息、载荷配置和模型信息的制造过程动态数据集成在一起，准确地更新部件的物理模型：(1)从模型参数到系统刚度的简化仿射映射方法，将变形建模中的材料属性与工艺约束解耦。(2)多源先验数据驱动的模型参数优化框架，能够有效识别材料参数和工艺约束，同时量化测量不确定度影响，并保持测量误差小于0.3 mm的高精度性能。(3)面向工业应用的形状调节平台，利用更新的物理模型进行精确载荷反演，实现复合材料部件的指定形状。实验和仿真结果表明，相对于未标定的静态建模，位移误差降低了64%以上，而低于0.2 N溶液精度的载荷反演实现了构件的几何偏差校正。建立了闭环测量-数据-模型-同化模式，提高了航空智能制造系统的决策自主性。

{"title":"Dynamic data driven uncertain physical information self-awareness method for the aircraft composite component assembly system","authors":"Pengbo Yin, Yang Zhang, Jiacheng Cui, Jiangtao Zhao, Yulin Jin, Qihang Chen, Yongkang Lu, Wei Liu","doi":"10.1016/j.jmsy.2025.11.001","DOIUrl":"10.1016/j.jmsy.2025.11.001","url":null,"abstract":"<div><div>High-precision assembly of large composite components is crucial for aircraft structural safety. To address geometric deviation of components caused by anisotropic material behaviors and time-varying process constraints during dynamic assembly, this paper proposes a dynamic data driven uncertain physical information self-awareness (DDDPIA) method. This approach accurately updates physical models of components by integrating dynamic data of the manufacturing process containing displacement information, load configurations, and model information through three key innovations: (1) A simplified affine mapping method from model parameters to system stiffness that decouples material properties from process constraints in deformation modeling. (2) A multi-source prior data-driven model parameter optimization framework enabling efficient identification of material parameters and process constraints while quantifying measurement uncertainty impacts and maintaining high-precision performance with measurement errors below 0.3 mm. (3) An industrial application-oriented shape regulation platform that leverages the updated physical model for precise load inversion to achieve specified shapes of composite components. Experimental and simulation results verify over 64% displacement error reduction relative to uncalibrated static modeling, while load inversion with sub-0.2 N solution accuracy achieves geometric deviations correction of components. This establishes a closed-loop measurement-data-model-assimilation paradigm, enhancing decision autonomy in aviation intelligent manufacturing systems.</div></div>","PeriodicalId":16227,"journal":{"name":"Journal of Manufacturing Systems","volume":"83 ","pages":"Pages 1009-1023"},"PeriodicalIF":14.2,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145680786","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

AR-assisted human-robot collaborative assembly system: Integrating visual language model and deep reinforcement learning for task planning and seamless interactive guidance ar辅助人机协同装配系统：集成视觉语言模型和深度强化学习，实现任务规划和无缝交互引导

IF 14.2 1区工程技术 Q1 ENGINEERING, INDUSTRIAL

Journal of Manufacturing Systems

Pub Date : 2025-11-29 DOI: 10.1016/j.jmsy.2025.11.019

Changchun Liu , Dunbing Tang , Haihua Zhu , Zequn Zhang , Liping Wang , Qingwei Nie

In the human-oriented context of Industry 5.0, human-robot collaboration (HRC) has become the core driving force for innovating the production model of assembly systems by integrating human flexibility ability with the precision and repeatability of robots. As the core link to achieve efficient collaboration, task planning is facing multiple challenges in dynamic and complex scenarios. On the one hand, it is hard to perform cognition of multimodal human-robot-environment in HRC assembly scenarios. On the other hand, heterogeneous capabilities of humans and robots (e.g., flexible decision-making by humans and precise execution by robots) are hard to be fully used to achieve reasonable task allocation and timing optimization in dynamic and complex HRC assembly scenarios. To address these issues, a Vision Language Model (VLM)-enhanced deep reinforcement learning-driven task planning approach is proposed towards Augmented Reality (AR)-assisted HRC assembly system. Firstly, a self-trained VLM is proposed through the integration of domain-specific knowledge and real-time situational data to enable context-aware in HRC assembly system. Through the fine-tuning of role configuration parameters for the pre-constructed VLM via prompt engineering, VLM can possess cognition of multi-dimensional assembly scenario elements. Reinforcement learning model can be endowed with the eyes to perceive HRC assembly scenarios through VLM-enhanced cognition of the dynamic HRC environment. Based on the VLM-enhanced cognition of the dynamic HRC environment, an improved multi-agent reinforcement learning-based HRC assembly task planning model is established to achieve humanized task planning, which can consider the competitive relationship between humans and robots with multi-agent conflict mechanism. Based on the HRC assembly task planning result, AR can enable operators to accomplish visual HRC assembly guidance through virtual-real mapping of HRC assembly information (e.g., HRC assembly procedures) and interact seamlessly with the VLM. Finally, experimental results show that the proposed method can improve the efficiency and well-being of HRC in human-centric assembly systems.

在工业5.0以人为本的背景下，人机协作（human-robot collaboration， HRC）将人的柔性能力与机器人的精度和可重复性相结合，成为创新装配系统生产模式的核心驱动力。任务规划作为实现高效协同的核心环节，在动态复杂的场景下面临着多重挑战。一方面，在HRC装配场景中，很难对多模态人-机器人-环境进行认知。另一方面，在动态复杂的HRC装配场景中，人与机器人的异构能力（如人的灵活决策和机器人的精确执行）难以充分发挥，难以实现任务的合理分配和时间优化。针对这些问题，提出了一种基于视觉语言模型（VLM）的深度强化学习驱动任务规划方法，用于增强现实（AR）辅助HRC装配系统。首先，通过整合领域知识和实时情景数据，提出了一种自训练的VLM，实现了HRC装配系统的上下文感知；通过快速工程化对预构建VLM的角色配置参数进行微调，使VLM具备对多维装配场景要素的认知能力。通过vlm增强对动态HRC环境的认知，可以赋予强化学习模型感知HRC装配场景的眼睛。基于vlm增强的HRC动态环境认知，建立了一种改进的基于多智能体强化学习的HRC装配任务规划模型，以实现任务规划的人性化，该模型考虑了具有多智能体冲突机制的人与机器人之间的竞争关系。基于HRC装配任务规划结果，AR可以使操作员通过HRC装配信息（如HRC装配过程）的虚实映射实现可视化的HRC装配指导，并与VLM实现无缝交互。最后，实验结果表明，该方法可以提高以人为中心的装配系统中HRC的效率和福祉。

{"title":"AR-assisted human-robot collaborative assembly system: Integrating visual language model and deep reinforcement learning for task planning and seamless interactive guidance","authors":"Changchun Liu , Dunbing Tang , Haihua Zhu , Zequn Zhang , Liping Wang , Qingwei Nie","doi":"10.1016/j.jmsy.2025.11.019","DOIUrl":"10.1016/j.jmsy.2025.11.019","url":null,"abstract":"<div><div>In the human-oriented context of Industry 5.0, human-robot collaboration (HRC) has become the core driving force for innovating the production model of assembly systems by integrating human flexibility ability with the precision and repeatability of robots. As the core link to achieve efficient collaboration, task planning is facing multiple challenges in dynamic and complex scenarios. On the one hand, it is hard to perform cognition of multimodal human-robot-environment in HRC assembly scenarios. On the other hand, heterogeneous capabilities of humans and robots (e.g., flexible decision-making by humans and precise execution by robots) are hard to be fully used to achieve reasonable task allocation and timing optimization in dynamic and complex HRC assembly scenarios. To address these issues, a Vision Language Model (VLM)-enhanced deep reinforcement learning-driven task planning approach is proposed towards Augmented Reality (AR)-assisted HRC assembly system. Firstly, a self-trained VLM is proposed through the integration of domain-specific knowledge and real-time situational data to enable context-aware in HRC assembly system. Through the fine-tuning of role configuration parameters for the pre-constructed VLM via prompt engineering, VLM can possess cognition of multi-dimensional assembly scenario elements. Reinforcement learning model can be endowed with the eyes to perceive HRC assembly scenarios through VLM-enhanced cognition of the dynamic HRC environment. Based on the VLM-enhanced cognition of the dynamic HRC environment, an improved multi-agent reinforcement learning-based HRC assembly task planning model is established to achieve humanized task planning, which can consider the competitive relationship between humans and robots with multi-agent conflict mechanism. Based on the HRC assembly task planning result, AR can enable operators to accomplish visual HRC assembly guidance through virtual-real mapping of HRC assembly information (e.g., HRC assembly procedures) and interact seamlessly with the VLM. Finally, experimental results show that the proposed method can improve the efficiency and well-being of HRC in human-centric assembly systems.</div></div>","PeriodicalId":16227,"journal":{"name":"Journal of Manufacturing Systems","volume":"84 ","pages":"Pages 40-67"},"PeriodicalIF":14.2,"publicationDate":"2025-11-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145682161","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Intention recognition and task allocation in human–robot collaborative assembly based on adaptive networks and DT belief space-GA 基于自适应网络和DT信念空间-遗传算法的人机协同装配意图识别与任务分配

IF 14.2 1区工程技术 Q1 ENGINEERING, INDUSTRIAL

Journal of Manufacturing Systems

Pub Date : 2025-11-29 DOI: 10.1016/j.jmsy.2025.11.018

Jiawei He , Chaoyang Zhang , Yichen Wang , Pai Zheng , Juchen Zhang

Human–robot collaboration (HRC) is pivotal in human-centered intelligent manufacturing, leveraging the complementary strengths of human adaptability and robotic efficiency to improve assembly quality and productivity. However, accurately recognizing human assembly intentions and optimizing task allocation remain key challenges. To address these issues, this study proposes an intention recognition-driven digital twin task allocation method, integrating an improved two-stream self-enhancing adaptive graph convolutional network (2S-SEAGCN) and a Belief Space genetic algorithm (BS-GA). First, a lightweight skeletal point detection model is developed to recognize human assembly intentions. Subsequently, the digital twin task allocation method applies the BS-GA algorithm to optimize resource allocation and enhance decision-making efficiency. The algorithm integrates a custom encoding scheme, belief space modeling, elitism, and local search mechanisms to avoid local optima, maintain optimization stability, and enhance global optimization. Finally, integrating reinforcement learning models to achieve end-to-end collaboration between robots and human operators, forming a closed-loop interactive assembly system. Experimental results demonstrate that 2S-SEAGCN outperforms traditional convolutional neural networks (CNN) and graph convolutional networks (GCN) in assembly intentions recognition, balancing accuracy and real-time performance. Moreover, the BS-GA algorithm achieves a higher average fitness than conventional genetic and heuristic algorithms in optimizing complex assembly tasks. In addition, reinforcement learning models demonstrate strong adaptability in dynamic and uncertain assembly scenarios. This closed-loop system provides practical solutions for real-world HRC applications, significantly improving production efficiency and quality, thereby demonstrating the substantial potential of industrial deployment.

人机协作（HRC）是以人为中心的智能制造的关键，利用人的适应性和机器人效率的互补优势来提高装配质量和生产率。然而，准确识别人类装配意图和优化任务分配仍然是关键的挑战。为了解决这些问题，本研究提出了一种意图识别驱动的数字孪生任务分配方法，该方法集成了改进的双流自增强自适应图卷积网络（2S-SEAGCN）和信念空间遗传算法（BS-GA）。首先，开发了一种轻量级的骨骼点检测模型来识别人体装配意图。随后，数字孪生任务分配方法采用BS-GA算法优化资源分配，提高决策效率。该算法集成了自定义编码方案、信念空间建模、精英主义和局部搜索机制，避免了局部最优，保持了优化稳定性，增强了全局最优性。最后，整合强化学习模型，实现机器人与人类操作者端到端的协作，形成闭环交互装配系统。实验结果表明，2S-SEAGCN在装配意图识别、平衡精度和实时性等方面均优于传统卷积神经网络（CNN）和图卷积网络（GCN）。此外，BS-GA算法在优化复杂装配任务时，比传统的遗传算法和启发式算法具有更高的平均适应度。此外，强化学习模型在动态和不确定的装配场景中表现出较强的适应性。该闭环系统为现实世界的HRC应用提供了实用的解决方案，显著提高了生产效率和质量，从而展示了工业部署的巨大潜力。

{"title":"Intention recognition and task allocation in human–robot collaborative assembly based on adaptive networks and DT belief space-GA","authors":"Jiawei He , Chaoyang Zhang , Yichen Wang , Pai Zheng , Juchen Zhang","doi":"10.1016/j.jmsy.2025.11.018","DOIUrl":"10.1016/j.jmsy.2025.11.018","url":null,"abstract":"<div><div>Human–robot collaboration (HRC) is pivotal in human-centered intelligent manufacturing, leveraging the complementary strengths of human adaptability and robotic efficiency to improve assembly quality and productivity. However, accurately recognizing human assembly intentions and optimizing task allocation remain key challenges. To address these issues, this study proposes an intention recognition-driven digital twin task allocation method, integrating an improved two-stream self-enhancing adaptive graph convolutional network (2S-SEAGCN) and a Belief Space genetic algorithm (BS-GA). First, a lightweight skeletal point detection model is developed to recognize human assembly intentions. Subsequently, the digital twin task allocation method applies the BS-GA algorithm to optimize resource allocation and enhance decision-making efficiency. The algorithm integrates a custom encoding scheme, belief space modeling, elitism, and local search mechanisms to avoid local optima, maintain optimization stability, and enhance global optimization. Finally, integrating reinforcement learning models to achieve end-to-end collaboration between robots and human operators, forming a closed-loop interactive assembly system. Experimental results demonstrate that 2S-SEAGCN outperforms traditional convolutional neural networks (CNN) and graph convolutional networks (GCN) in assembly intentions recognition, balancing accuracy and real-time performance. Moreover, the BS-GA algorithm achieves a higher average fitness than conventional genetic and heuristic algorithms in optimizing complex assembly tasks. In addition, reinforcement learning models demonstrate strong adaptability in dynamic and uncertain assembly scenarios. This closed-loop system provides practical solutions for real-world HRC applications, significantly improving production efficiency and quality, thereby demonstrating the substantial potential of industrial deployment.</div></div>","PeriodicalId":16227,"journal":{"name":"Journal of Manufacturing Systems","volume":"84 ","pages":"Pages 20-39"},"PeriodicalIF":14.2,"publicationDate":"2025-11-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145617340","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Aircraft assembly process planning based on knowledge graph constructed by integrating LLMs and SLMs 基于集成llm和slm构建的知识图谱的飞机装配工艺规划

IF 14.2 1区工程技术 Q1 ENGINEERING, INDUSTRIAL

Journal of Manufacturing Systems

Pub Date : 2025-11-26 DOI: 10.1016/j.jmsy.2025.11.016

Yunfei Ma , Shuai Zheng , Zheng Yang , Pai Zheng , Jiewu Leng , Jun Hong

In commercial aircraft manufacturing, process planning serves as a crucial bridge between design and production, ensuring the accurate realization of design concepts and significantly improving manufacturing efficiency and product quality. With the development of knowledge graph technologies, significant progress has been made in using historical process documentation for commercial aircraft manufacturing process planning. However, traditional deep learning-based methods for constructing knowledge graph heavily rely on manual object selection and label assignment, making the process highly time-consuming. Additionally, the methods often face challenges in the field of process planning, including low domain-specific terminology recognition rates and incomplete entity extraction. To tackle these challenges, this paper introduces a hybrid approach that integrates large and small language models to construct an aircraft process planning knowledge graph. Initially, clustering-based multi-agent approach is employed to pre-annotate the process planning dataset, with domain experts re-annotate the defect data to create a high-quality process planning dataset. Subsequently, a knowledge extraction framework for aircraft process planning, KE-LSM, was constructed using the small language model trained on this dataset, together with the LLM. Experimental results show that KE-LSM outperforms existing named entity recognition models. Finally, KE-LSM is applied in a commercial aircraft manufacturing company, accompanied by the development of a prototype system designed to facilitate intelligent process planning. It is hoped that the research can provide valuable insights and support for the application of LLM-based solutions in the field of aircraft manufacturing.

在商用飞机制造中，工艺规划是连接设计和生产的重要桥梁，保证了设计理念的准确实现，显著提高了制造效率和产品质量。随着知识图谱技术的发展，利用历史工艺文件进行商用飞机制造工艺规划取得了重大进展。然而，传统的基于深度学习的知识图构建方法严重依赖于人工对象选择和标签分配，使得该过程非常耗时。此外，这些方法在过程规划领域经常面临挑战，包括特定领域术语识别率低和实体提取不完整。为了解决这些问题，本文介绍了一种集成大、小语言模型的混合方法来构建飞机工艺规划知识图。首先，采用基于聚类的多智能体方法对工艺规划数据集进行预标注，由领域专家对缺陷数据进行重新标注，生成高质量的工艺规划数据集。随后，利用在该数据集上训练的小语言模型和LLM，构建了飞机工艺规划知识提取框架KE-LSM。实验结果表明，KE-LSM优于现有的命名实体识别模型。最后，将KE-LSM应用于某商用飞机制造公司，并开发了用于智能工艺规划的原型系统。希望本研究能为基于llm的解决方案在飞机制造领域的应用提供有价值的见解和支持。

{"title":"Aircraft assembly process planning based on knowledge graph constructed by integrating LLMs and SLMs","authors":"Yunfei Ma , Shuai Zheng , Zheng Yang , Pai Zheng , Jiewu Leng , Jun Hong","doi":"10.1016/j.jmsy.2025.11.016","DOIUrl":"10.1016/j.jmsy.2025.11.016","url":null,"abstract":"<div><div>In commercial aircraft manufacturing, process planning serves as a crucial bridge between design and production, ensuring the accurate realization of design concepts and significantly improving manufacturing efficiency and product quality. With the development of knowledge graph technologies, significant progress has been made in using historical process documentation for commercial aircraft manufacturing process planning. However, traditional deep learning-based methods for constructing knowledge graph heavily rely on manual object selection and label assignment, making the process highly time-consuming. Additionally, the methods often face challenges in the field of process planning, including low domain-specific terminology recognition rates and incomplete entity extraction. To tackle these challenges, this paper introduces a hybrid approach that integrates large and small language models to construct an aircraft process planning knowledge graph. Initially, clustering-based multi-agent approach is employed to pre-annotate the process planning dataset, with domain experts re-annotate the defect data to create a high-quality process planning dataset. Subsequently, a knowledge extraction framework for aircraft process planning, KE-LSM, was constructed using the small language model trained on this dataset, together with the LLM. Experimental results show that KE-LSM outperforms existing named entity recognition models. Finally, KE-LSM is applied in a commercial aircraft manufacturing company, accompanied by the development of a prototype system designed to facilitate intelligent process planning. It is hoped that the research can provide valuable insights and support for the application of LLM-based solutions in the field of aircraft manufacturing.</div></div>","PeriodicalId":16227,"journal":{"name":"Journal of Manufacturing Systems","volume":"84 ","pages":"Pages 1-19"},"PeriodicalIF":14.2,"publicationDate":"2025-11-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145594700","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Deep reinforcement learning for event-driven predictive–reactive multi-objective scheduling in dynamic flexible job shop 动态柔性作业车间事件驱动预测反应多目标调度的深度强化学习

IF 14.2 1区工程技术 Q1 ENGINEERING, INDUSTRIAL

Journal of Manufacturing Systems

Pub Date : 2025-11-21 DOI: 10.1016/j.jmsy.2025.11.014

Chunyan Duan, Yuxin Mo, Zehao Zhang

In the era of intelligent manufacturing, dynamic scheduling problems characterized by high variability and complexity mandate prompt solutions. Deep Reinforcement Learning (DRL) is particularly effective for addressing these challenges through their formulation as Markov Decision Processes (MDPs). However, most DRL-based dynamic scheduling methods primarily adopt reactive strategies, thereby restricting their capacity to achieve global optimality. This paper addresses the Dynamic Multi-Objective Flexible Job Shop Scheduling Problem (DMFJSP) by introducing a comprehensive event-driven predictive–reactive scheduling framework designed to balance computational efficiency and solution quality. The approach employs a deterministic scheduling model that integrates the proposed Multi-Objective Hybrid Deep Q-Network (MOHDQN) as both the schedule generator and the baseline, enabling adaptive adjustment of objective weights. The Markov Jump Decision Process (MJDP) is incorporated to better manage event-driven predictive–reactive scheduling. Furthermore, seven rescheduling rules are developed to update schedules through a hybrid deep recurrent Q-network in response to dynamic events. Extensive numerical experiments across diverse problem scales demonstrate that the proposed method achieves a favorable balance between scheduling performance and computational efficiency. Finally, the most suitable rescheduling strategies for different enterprises are analyzed with respect to their specific operational complexities, and future development directions are also discussed.

在智能制造时代，高变异性、高复杂性的动态调度问题要求快速解决。深度强化学习（DRL）通过将其表述为马尔可夫决策过程（mdp）来解决这些挑战特别有效。然而，大多数基于drl的动态调度方法主要采用响应策略，从而限制了它们实现全局最优的能力。本文通过引入一个综合的事件驱动预测反应调度框架来平衡计算效率和解决方案质量，解决了动态多目标柔性作业车间调度问题。该方法采用确定性调度模型，将所提出的多目标混合深度q网络（MOHDQN）作为调度生成器和基线，实现目标权值的自适应调整。将马尔可夫跳跃决策过程（Markov Jump Decision Process， MJDP）纳入其中，以更好地管理事件驱动的预测响应调度。在此基础上，提出了7条重调度规则，通过混合深度循环q网络对动态事件进行调度更新。在不同问题尺度上的大量数值实验表明，该方法在调度性能和计算效率之间取得了良好的平衡。最后，针对不同企业的具体经营复杂性，分析了最适合不同企业的重调度策略，并对未来的发展方向进行了探讨。

{"title":"Deep reinforcement learning for event-driven predictive–reactive multi-objective scheduling in dynamic flexible job shop","authors":"Chunyan Duan, Yuxin Mo, Zehao Zhang","doi":"10.1016/j.jmsy.2025.11.014","DOIUrl":"10.1016/j.jmsy.2025.11.014","url":null,"abstract":"<div><div>In the era of intelligent manufacturing, dynamic scheduling problems characterized by high variability and complexity mandate prompt solutions. Deep Reinforcement Learning (DRL) is particularly effective for addressing these challenges through their formulation as Markov Decision Processes (MDPs). However, most DRL-based dynamic scheduling methods primarily adopt reactive strategies, thereby restricting their capacity to achieve global optimality. This paper addresses the Dynamic Multi-Objective Flexible Job Shop Scheduling Problem (DMFJSP) by introducing a comprehensive event-driven predictive–reactive scheduling framework designed to balance computational efficiency and solution quality. The approach employs a deterministic scheduling model that integrates the proposed Multi-Objective Hybrid Deep Q-Network (MOHDQN) as both the schedule generator and the baseline, enabling adaptive adjustment of objective weights. The Markov Jump Decision Process (MJDP) is incorporated to better manage event-driven predictive–reactive scheduling. Furthermore, seven rescheduling rules are developed to update schedules through a hybrid deep recurrent Q-network in response to dynamic events. Extensive numerical experiments across diverse problem scales demonstrate that the proposed method achieves a favorable balance between scheduling performance and computational efficiency. Finally, the most suitable rescheduling strategies for different enterprises are analyzed with respect to their specific operational complexities, and future development directions are also discussed.</div></div>","PeriodicalId":16227,"journal":{"name":"Journal of Manufacturing Systems","volume":"83 ","pages":"Pages 992-1008"},"PeriodicalIF":14.2,"publicationDate":"2025-11-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145568614","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Efficient human–robot collaborative manipulation of planar deformable objects 平面可变形物体的高效人机协同操作

IF 14.2 1区工程技术 Q1 ENGINEERING, INDUSTRIAL

Journal of Manufacturing Systems

Pub Date : 2025-11-19 DOI: 10.1016/j.jmsy.2025.11.011

Enrico Villagrossi , Paolo Franceschi , Giorgio Nicola , Nicola Pedrocchi

This paper presents an efficient method to improve the productivity and the accuracy of human–robot collaboration in transporting large, planar, deformable objects, specifically during the production of parts made of advanced composite materials. The proposed approach utilises an industrial robot to assist operators in transporting and handling carbon fibre and fibreglass plies during draping. A consumer vision system feeds a data-driven model that estimates the material’s deformation from depth images. This deformation data, transformed into force/torque information via a virtual spring, informs a Human–Robot Role Arbitration (RA) algorithm that dynamically adjusts leadership between humans and robots based on context, enhancing safety and efficiency. Inspired by game theory, the approach adapts to cooperative and non-cooperative scenarios, demonstrating significant productivity gains over traditional algorithms used for the same scope. The paper also compares the use of the RA algorithms with the current industrial practice, which relies entirely on manual production. Company operators, working in a production site, performed the experimental comparison producing a real boat propeller.

本文提出了一种有效的方法来提高大型、平面、可变形物体的人机协作的生产率和精度，特别是在先进复合材料零件的生产过程中。提出的方法利用工业机器人来协助操作员在悬垂过程中运输和处理碳纤维和玻璃纤维层。消费者视觉系统提供一个数据驱动的模型，该模型可以从深度图像中估计材料的变形。这些变形数据通过虚拟弹簧转换为力/扭矩信息，通知人-机器人角色仲裁（RA）算法，该算法根据上下文动态调整人和机器人之间的领导，从而提高安全性和效率。受博弈论的启发，该方法适用于合作和非合作场景，与用于相同范围的传统算法相比，显示出显著的生产力提高。本文还将RA算法的使用与目前完全依赖手工生产的工业实践进行了比较。公司操作人员在生产现场进行了实际船舶螺旋桨的实验比较。

{"title":"Efficient human–robot collaborative manipulation of planar deformable objects","authors":"Enrico Villagrossi , Paolo Franceschi , Giorgio Nicola , Nicola Pedrocchi","doi":"10.1016/j.jmsy.2025.11.011","DOIUrl":"10.1016/j.jmsy.2025.11.011","url":null,"abstract":"<div><div>This paper presents an efficient method to improve the productivity and the accuracy of human–robot collaboration in transporting large, planar, deformable objects, specifically during the production of parts made of advanced composite materials. The proposed approach utilises an industrial robot to assist operators in transporting and handling carbon fibre and fibreglass plies during draping. A consumer vision system feeds a data-driven model that estimates the material’s deformation from depth images. This deformation data, transformed into force/torque information via a virtual spring, informs a Human–Robot Role Arbitration (RA) algorithm that dynamically adjusts leadership between humans and robots based on context, enhancing safety and efficiency. Inspired by game theory, the approach adapts to cooperative and non-cooperative scenarios, demonstrating significant productivity gains over traditional algorithms used for the same scope. The paper also compares the use of the RA algorithms with the current industrial practice, which relies entirely on manual production. Company operators, working in a production site, performed the experimental comparison producing a real boat propeller.</div></div>","PeriodicalId":16227,"journal":{"name":"Journal of Manufacturing Systems","volume":"83 ","pages":"Pages 963-975"},"PeriodicalIF":14.2,"publicationDate":"2025-11-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145568610","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Neural surface partitioner: A physics-informed neural network for five-axis machining by a non-spherical cutting tool 神经曲面分割器：一种基于物理信息的神经网络，用于非球面刀具的五轴加工

IF 14.2 1区工程技术 Q1 ENGINEERING, INDUSTRIAL

Journal of Manufacturing Systems

Pub Date : 2025-11-19 DOI: 10.1016/j.jmsy.2025.11.015

Jiancheng Hao , Haokun Chen , Pengcheng Hu , Dong He , Yamin Li , Xiaoke Deng , Tak Yu Lau , Fan Shi , Yanglong Lu

The machining performance of five-axis machining can be significantly improved by partitioning the part surface into subregions, each of which is machined using an adaptive iso-scallop height method. This approach is particularly beneficial when employing non-spherical cutting tools, as it allows for a diverse range of effective cutting radii that can be optimally tailored to the geometry of each subregion. Current surface partition-based five-axis machining methods begin by partitioning the surface into patches based on geometric constraints, followed by generating the tool path for each partitioned patch. However, this stepwise approach fails to incorporate critical objectives such as machining efficiency, smooth tool orientation, and interference avoidance during the surface partitioning process. Consequently, both the surface partitioning results and the subsequently generated tool paths tend to be sub-optimal. In this research, we introduce a novel Physics-Informed Neural Network (PINN) called the Neural Surface Partitioner (NSP) designed to jointly and near-optimally partition surfaces while generating tool orientations for non-spherical cutting tools. By directly integrating comprehensive issues into the NSP’s loss function, we create a surface partitioning matrix that enables effective partitioning of freeform surfaces. Simultaneously, the NSP generates a tool orientation scalar field, which is employed to produce iso-scallop Cutter Contact (CC) curves while ensuring smooth and interference-free tool orientation. To validate the advantages of the tool paths generated by the proposed NSP, we conducted both computer simulations and physical cutting experiments. The results demonstrate that the average cutting width achieved by our method significantly exceeds that of two benchmark methods, leading to drastically reduced path lengths of up to 42.9 % and machining times of up to 44.8 %.

采用自适应等扇贝高度法对零件表面进行划分，可显著提高五轴加工的加工性能。当使用非球面切削工具时，这种方法特别有益，因为它允许各种有效切削半径范围，可以根据每个子区域的几何形状进行最佳定制。目前基于曲面划分的五轴加工方法首先是根据几何约束将曲面划分为多个小块，然后为每个小块生成刀具轨迹。然而，这种渐进式方法未能在表面划分过程中纳入加工效率、平滑刀具定向和避免干涉等关键目标。因此，表面划分结果和随后生成的刀具路径都趋向于次优。在这项研究中，我们引入了一种新的物理信息神经网络（PINN），称为神经表面分割器（NSP），用于在生成非球面切削刀具的刀具方向时联合和接近最优地分割表面。通过将综合问题直接集成到NSP的损失函数中，我们创建了一个表面划分矩阵，可以有效地划分自由曲面。同时，NSP生成刀具取向标量场，该标量场用于生成等扇形刀具接触（CC）曲线，同时保证刀具取向光滑、无干涉。为了验证由NSP生成的刀具轨迹的优势，我们进行了计算机模拟和物理切削实验。结果表明，该方法获得的平均切削宽度显著超过两种基准方法，使路径长度大幅减少42.9% %，加工次数减少44.8% %。

{"title":"Neural surface partitioner: A physics-informed neural network for five-axis machining by a non-spherical cutting tool","authors":"Jiancheng Hao , Haokun Chen , Pengcheng Hu , Dong He , Yamin Li , Xiaoke Deng , Tak Yu Lau , Fan Shi , Yanglong Lu","doi":"10.1016/j.jmsy.2025.11.015","DOIUrl":"10.1016/j.jmsy.2025.11.015","url":null,"abstract":"<div><div>The machining performance of five-axis machining can be significantly improved by partitioning the part surface into subregions, each of which is machined using an adaptive iso-scallop height method. This approach is particularly beneficial when employing non-spherical cutting tools, as it allows for a diverse range of effective cutting radii that can be optimally tailored to the geometry of each subregion. Current surface partition-based five-axis machining methods begin by partitioning the surface into patches based on geometric constraints, followed by generating the tool path for each partitioned patch. However, this stepwise approach fails to incorporate critical objectives such as machining efficiency, smooth tool orientation, and interference avoidance during the surface partitioning process. Consequently, both the surface partitioning results and the subsequently generated tool paths tend to be sub-optimal. In this research, we introduce a novel Physics-Informed Neural Network (PINN) called the Neural Surface Partitioner (NSP) designed to jointly and near-optimally partition surfaces while generating tool orientations for non-spherical cutting tools. By directly integrating comprehensive issues into the NSP’s loss function, we create a surface partitioning matrix that enables effective partitioning of freeform surfaces. Simultaneously, the NSP generates a tool orientation scalar field, which is employed to produce iso-scallop Cutter Contact (CC) curves while ensuring smooth and interference-free tool orientation. To validate the advantages of the tool paths generated by the proposed NSP, we conducted both computer simulations and physical cutting experiments. The results demonstrate that the average cutting width achieved by our method significantly exceeds that of two benchmark methods, leading to drastically reduced path lengths of up to 42.9 % and machining times of up to 44.8 %.</div></div>","PeriodicalId":16227,"journal":{"name":"Journal of Manufacturing Systems","volume":"83 ","pages":"Pages 976-991"},"PeriodicalIF":14.2,"publicationDate":"2025-11-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145568615","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0