Journal of Manufacturing Systems最新文献

{"title":"Assemble like expert: Multitask-meta hierarchical imitation learning algorithm guided by an expert skill model for robot polygonal peg-in-hole assembly","authors":"Hubo Chu ,&nbsp;Tie Zhang ,&nbsp;Yanbiao Zou ,&nbsp;Hanlei Sun","doi":"10.1016/j.jmsy.2025.09.003","DOIUrl":"10.1016/j.jmsy.2025.09.003","url":null,"abstract":"<div><div>Robot polygonal peg-in-hole assembly is still challenging due to the unknown assembly environment and diverse tasks. To equip robots with expert assembly skills, this paper employs a model-guided strategy learning approach and proposes a multitask-meta hierarchical imitation learning algorithm guided by an expert skill model. Specifically, to construct a skill model for guiding strategy learning, a deterministic expert strategy is proposed. Based on this strategy, expert assembly characteristics are analyzed, and an expert skill model is developed to represent these characteristics. Furthermore, to learn experts' skill adjustment and generalization strategies across different tasks, a multitask-meta hierarchical imitation learning algorithm (MMHIL) is proposed. A parallel encoding attention network is designed to assist MMHIL in extracting multi-level skill information and learning assembly actions. A multitask-meta learning generalization framework with a mutual supervised learning optimization mechanism is proposed to enable MMHIL to rapidly adapt to new assembly tasks with limited training data. Comparative verification and polygonal peg-in-hole assembly experiments show that MMHIL has better skill learning effects and higher assembly success rates.</div></div>","PeriodicalId":16227,"journal":{"name":"Journal of Manufacturing Systems","volume":"83 ","pages":"Pages 82-102"},"PeriodicalIF":14.2,"publicationDate":"2025-09-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145020029","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}

{"title":"A surrogate model-driven assembly coordination framework for aircraft components based on cooperative multi-agent deep reinforcement learning","authors":"Yifan Zhang ,&nbsp;Wenxu Luo ,&nbsp;Ye Hu ,&nbsp;Qing Wang ,&nbsp;Liang Cheng ,&nbsp;Yinglin Ke","doi":"10.1016/j.jmsy.2025.08.021","DOIUrl":"10.1016/j.jmsy.2025.08.021","url":null,"abstract":"<div><div>This study presents a multi-agent reinforcement learning (MARL) approach to address coordination challenges in aircraft component assembly. A machine learning–based surrogate model is developed to approximate component deformation, enabling the construction of a realistic and computationally efficient MARL training environment. Within this environment, multiple agents rapidly learn strategies to optimize both individual component deformations and inter-component coordination. The surrogate model compresses the high-dimensional displacement fields into lower-dimensional representations, significantly reducing the complexity of the state space. The reward function combines both local and coordination rewards, where the local reward evaluates manufacturing accuracy at the component level, and the coordination reward assesses alignment accuracy between components. By exchanging local state information during training, agents enhance cooperation, accelerate convergence, and improve overall assembly performance. The effectiveness of the proposed method is demonstrated through a fuselage panel assembly case study, achieving average reductions of 94.91 % in panel deformation and 95.02 % in inter-panel gaps. This framework offers a promising solution for coordinating deformable structures, substantially enhancing both assembly quality and efficiency.</div></div>","PeriodicalId":16227,"journal":{"name":"Journal of Manufacturing Systems","volume":"83 ","pages":"Pages 46-64"},"PeriodicalIF":14.2,"publicationDate":"2025-09-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145010819","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}

{"title":"Towards instructional collaborative robots: From video-based learning to feedback-adapted instruction","authors":"Jinyi Huang,&nbsp;Xun Xu,&nbsp;Jan Polzer","doi":"10.1016/j.jmsy.2025.08.018","DOIUrl":"10.1016/j.jmsy.2025.08.018","url":null,"abstract":"<div><div>Collaborative robots (cobots) enhanced by artificial intelligence (AI) are enabling intelligent, human-centric manufacturing environments. These dynamic settings require cobots with cognitive intelligence, i.e., capabilities covering perception, learning, decision-making, and adaptation. Such intelligence enables proactive collaboration that integrates bidirectional instructional and cooperative competencies. However, while extensive research has focused on improving the performance of robot collaborative skills, systematic investigations into the instructional capabilities of cobots remain notably limited. To lay the technological foundation for addressing this gap, this survey adopts a multimodal perspective to review three essential aspects of this field: (1) robot learning from video (LfV) for instructional capabilities acquisition, (2) robot-guided instruction methodologies, and (3) feedback-driven adaptation. We present a systematic review of technologies for representing human actions, object states, and human-object interactions (HOI), with a particular focus on multimodal data sources from video. Furthermore, we analyze diverse instructional strategies, including visual guidance, auditory directives, robot-performed actions, emphasizing their effectiveness in robot-guided instruction. A significant focus is placed on feedback-driven adaptation mechanisms, which enable cobots to dynamically refine their instructional capabilities based on user feedback. We identify key challenges such as environmental complexity, user variability, real-time processing constraints, and trust-building requirements, while also highlighting emerging opportunities in multimodal integration, AI-powered robots, and collaborative learning systems. Finally, we underscore the transformative potential of instructional cobots in smart manufacturing and emphasize the necessity for further research.</div></div>","PeriodicalId":16227,"journal":{"name":"Journal of Manufacturing Systems","volume":"83 ","pages":"Pages 29-45"},"PeriodicalIF":14.2,"publicationDate":"2025-09-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145005074","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}

{"title":"Modeling stochastic dynamics of manufacturing processes with manifold signals: A harmonic analysis approach with NP-ODEs","authors":"Yu Wang ,&nbsp;Bo Yang ,&nbsp;Shilong Wang ,&nbsp;Zhengping Zhang ,&nbsp;Yucheng Zhang ,&nbsp;Haijian Liu","doi":"10.1016/j.jmsy.2025.08.010","DOIUrl":"10.1016/j.jmsy.2025.08.010","url":null,"abstract":"<div><div>Modern manufacturing increasingly leverages intricate signals that reflect the complex geometries of produced parts. These signals hold vital insights into the quality and working status of the underlying manufacturing systems. While conventional deep learning approaches excel with structured data like time series processing or image vision tasks, they can struggle to model these geometrically complex signals, which are inherently in a non-Euclidean manifold domain and exhibit stochastic dynamical evolution behaviors over time. In this paper, we present a novel methodology that leverages the principles of harmonic analysis with the potential of continuous-time dynamics modeling and stochastic process representation. Specifically, our contributions are twofold: (1) we present a general and flexible framework for modeling the stochastic dynamics of manifold signals within manufacturing processes. This framework uniquely synergizes Neural ODEs and Neural Processes (NPs), enhanced by a neural numerical integration scheme for computational efficiency; (2) we develop a rigorous and tailored representation approach rooted in harmonic analysis, enabling the construction of learnable wavelet filters for multiscale pattern analysis on manifold signals, with Geometric Deep Learning (GDL) principles ensuring compatibility with modern deep learning architectures. We comprehensively validate our methodology through simulations and a real-world automotive case study. Results demonstrate its effectiveness in modeling the complex stochastic dynamics inherent in manifold signals found in practical manufacturing settings, highlighting its potential for industrial applications.</div></div>","PeriodicalId":16227,"journal":{"name":"Journal of Manufacturing Systems","volume":"83 ","pages":"Pages 12-28"},"PeriodicalIF":14.2,"publicationDate":"2025-09-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144934094","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}

{"title":"Online sequential decision making of multi-stage assembly process parameters based on deep reinforcement learning and its application in diesel engine production","authors":"Yi-Tian Song ,&nbsp;Yan-Ning Sun ,&nbsp;Li-Lan Liu ,&nbsp;Jie Wu ,&nbsp;Zeng-Gui Gao ,&nbsp;Wei Qin","doi":"10.1016/j.jmsy.2025.08.012","DOIUrl":"10.1016/j.jmsy.2025.08.012","url":null,"abstract":"<div><div>Maintaining fixed parameters during batch assembly of complex mechanical products often results in quality inconsistencies due to time-varying operational conditions, including equipment performance degradation, production environment disturbance, and operator skill variations. This operational reality necessitates online parameter adaptation mechanisms to counteract progressive quality deviations. While complex assemblies inherently involve sequential multi-stage workflows across distributed stations, conventional optimization strategies often employ monolithic parameter adjustments that neglect error propagation effects and inter-stage quality interdependencies. To address the dual challenges of dynamic operating conditions and multi-stage coordination, this study proposes an online sequential decision-making framework based on deep reinforcement learning. First, a causal inference model for assembly quality prognosis is constructed by integrating the greedy equivalence search algorithm with domain-specific expert knowledge, enabling systematic modeling of multi-stage quality dependencies. Subsequently, the multi-stage parameters optimization problem is formalized as a Markov decision process, with innovatively defined state space as assembly progress, action space as adjusted parameters range, and physics-informed reward function derived from quality inference results. Building on this, the proximal policy optimization algorithm is improved by stage-aware experience replay and gradient alignment constraints to learn the optimal policy, and then select the optimal action. Experiments on a real-world diesel engine assembly dataset demonstrate a 17.16 % improvement in product qualification probability, significantly outperforming conventional methods. The proposed framework effectively captures time-varying assembly characteristics and achieves cross-stage parameter coordination through sequential decision-making, offering a novel data-driven solution for quality control in complex product assembly systems.</div></div>","PeriodicalId":16227,"journal":{"name":"Journal of Manufacturing Systems","volume":"82 ","pages":"Pages 1252-1268"},"PeriodicalIF":14.2,"publicationDate":"2025-09-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144932313","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}

{"title":"Step-time measurement: A scalable sub-cycle time defining methodology for anomaly detection and predictive maintenance in sequential production lines","authors":"Jon Zubieta ,&nbsp;Unai Izagirre ,&nbsp;Luka Eciolaza ,&nbsp;Asier Saez de Buruaga ,&nbsp;Lander Galdos","doi":"10.1016/j.jmsy.2025.08.011","DOIUrl":"10.1016/j.jmsy.2025.08.011","url":null,"abstract":"<div><div>Sub-cycle time periods from machines in production lines offer valuable insights into component-level health. They enable data-driven condition monitoring without the need for additional sensors. However, the lack of a standardized methodology for defining these sub-cycle time periods limits the practicality and scalability of this approach in real-world applications. We propose a scalable methodology to define sub-cycle time periods within the machine cycle time, using Programmable Logic Controllers (PLCs) programmed in compliance with the IEC 60848 standard. To achieve scalability, the proposed methodology makes sub-cycle time period definition automatic, simple and thus, fast. This is achieved by defining each sub-cycle time period as the total activation time of a Step. For this reason, the sub-cycle time periods defined with this methodology are named “Step-time”s. Because the methodology does not depend on the type of action or actuator involved, and because it can be applied to any step without requiring changes to the overall program structure, it can be easily replicated across multiple steps, modules, or even machines. This modularity enables a scalable deployment of Step-time measurements, whether for a few components or across entire production lines. Moreover, our methodology offers deeper insights into machine behavior by distinguishing between different operational contexts for the same component. To assess its feasibility in industrial production environments, we developed two implementation approaches, one based on Structured Text (ST) and another using Sequential Function Charts (SFC). The results demonstrate that machine anomalies such as air leaks, pressure drops and fluctuations in pneumatic circuits, are accurately reflected in Step-times. This confirms the high resolution of the Step-times and highlights its potential for powering data-driven condition monitoring systems in future works. Finally, the data acquisition results indicate that the proposed methodology has minimal impact on the PLC scan-cycle, making it suitable for most industrial use cases.</div></div>","PeriodicalId":16227,"journal":{"name":"Journal of Manufacturing Systems","volume":"83 ","pages":"Pages 1-11"},"PeriodicalIF":14.2,"publicationDate":"2025-09-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144934093","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}

{"title":"GRU-based real-time scheduling method for production-logistics collaboration in digital twin workshop","authors":"Wenchao Yang ,&nbsp;Boxuan Zhang ,&nbsp;Guofu Luo ,&nbsp;Linli Li ,&nbsp;Xiaoyu Wen ,&nbsp;Hao Li ,&nbsp;Haoqi Wang","doi":"10.1016/j.jmsy.2025.08.013","DOIUrl":"10.1016/j.jmsy.2025.08.013","url":null,"abstract":"<div><div>In modern workshops with high customization requirements, production is typically conducted under a small-batch, multi-variety order mode. Under such conditions, random order arrivals and fuzzy manufacturing times, caused by fluctuations in workshop conditions, present significant challenges to real-time scheduling and control. To address these issues, this study proposes a real-time scheduling method for production-logistics collaboration (RT-SMPLC) based on gated recurrent units (GRUs) in a digital twin (DT) workshop. Firstly, a comprehensive RT-SMPLC framework was constructed. Leveraging virtual-physical interaction, a dynamic mapping environment is established to capture the real-time status information of production elements. Secondly, the scheduling process is guided by a task priority index that facilitates the selection of the optimal production-logistics resource group for each task. This priority index is iteratively optimized through virtual evolution and GRU-based prediction. Finally, the operation assignment result is fed back to the physical workshop for execution in real time via industrial communication protocols and networks, enabling closed-loop control through virtual-to-physical interaction. The proposed method was validated on a DT-based experimental platform using real production cases. Comparative experiments across three different-scale scenarios and three algorithms demonstrate that RT-SMPLC effectively reduces makespan, energy consumption, and tardiness, while exhibiting robust real-time responsiveness.</div></div>","PeriodicalId":16227,"journal":{"name":"Journal of Manufacturing Systems","volume":"82 ","pages":"Pages 1269-1289"},"PeriodicalIF":14.2,"publicationDate":"2025-09-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144932245","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}

{"title":"Grating interferometer: The dominant positioning strategy in atomic and close-to-atomic scale manufacturing","authors":"Can Cui ,&nbsp;Xinghui Li ,&nbsp;Xiaohao Wang","doi":"10.1016/j.jmsy.2025.08.008","DOIUrl":"10.1016/j.jmsy.2025.08.008","url":null,"abstract":"<div><div>The rapid evolution of manufacturing technologies has entered its third phase, with a focus on atomic and close-to-atomic scale manufacturing (ACSM), a pivotal advancement driving progress in precision fabrication. One of the cores of ACSM is ultra-precise positioning technologies, which are critical for achieving the required precision and efficiency in nanoscale manufacturing. Grating interferometer has emerged as the leading strategy due to its accuracy, scalability, and stability. Recent advancements in this technology have further solidified its role as a dominant solution in ultra-precision metrology. This review provides an overview of grating interferometer as a positioning tool for ACSM, starting with its application domains and advantages, followed by a detailed explanation of its fundamental principles. We then present a comprehensive comparison of different representative grating interferometers. Additionally, we perform quantitative multi-source measurement error analysis, discuss methods for compensating errors, and explore various phase measurement techniques. Finally, we conclude with a forward-looking perspective on the future development of grating interferometer, highlighting emerging trends and potential breakthroughs.</div></div>","PeriodicalId":16227,"journal":{"name":"Journal of Manufacturing Systems","volume":"82 ","pages":"Pages 1227-1251"},"PeriodicalIF":14.2,"publicationDate":"2025-08-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144913028","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}

{"title":"Large language model-enabled cognitive agent for self-aware manufacturing","authors":"Shanhe Lou,&nbsp;Runjia Tan,&nbsp;Yanxin Zhou,&nbsp;Ziyue Zhao,&nbsp;Yiran Zhang,&nbsp;Chen Lv","doi":"10.1016/j.jmsy.2025.08.015","DOIUrl":"10.1016/j.jmsy.2025.08.015","url":null,"abstract":"<div><div>Although industrial automation has advanced significantly at the level of manufacturing units and production lines, system-level automation remains constrained by the limited cognitive abilities of current manufacturing systems. To address this challenge, this work proposes a cognitive agent (CA) that leverages a large language model (LLM) as its core to facilitate self-aware manufacturing. The cognitive capabilities of CA are facilitated through the combination of retrieval-augmented generation (RAG) and in-context learning. RAG allows CA to retrieve relevant subgraphs from an industrial knowledge graph (IKG) after interpreting natural language commands, thereby establishing focused context awareness and autonomously generating executable manufacturing instructions. Meanwhile, in-context learning enables CA to adapt to specific requirements based on contextual examples without retraining. These techniques empower CA with domain-specific cognition, fostering self-awareness in a flexible and cost-effective manner. Two case studies on pick-and-place and disassembly validate CA's effectiveness in task planning within a lab-scale manufacturing unit. The results demonstrate that the proposed approach surpasses traditional LLM-based methods in task executability and goal achievement, offering a novel perspective on advancing manufacturing systems.</div></div>","PeriodicalId":16227,"journal":{"name":"Journal of Manufacturing Systems","volume":"82 ","pages":"Pages 1213-1226"},"PeriodicalIF":14.2,"publicationDate":"2025-08-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144913026","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}

{"title":"Tool wear prediction in milling process using physics-informed machine learning and thermo-mechanical force model with monitoring applications","authors":"Farzad Pashmforoush ,&nbsp;Arash Ebrahimi Araghizad ,&nbsp;Erhan Budak","doi":"10.1016/j.jmsy.2025.08.014","DOIUrl":"10.1016/j.jmsy.2025.08.014","url":null,"abstract":"<div><div>Accurate wear estimation of milling tools is critical for enhancing the productivity and reliability of machining processes, ensuring consistent product quality while minimizing unexpected tool failure, downtime and machining costs. Traditional approaches, often based on pure experimental and data-driven machine learning (ML) methods, demand extensive, costly wear testing to gather the necessary datasets, which limits their utility in practical industrial monitoring. To address this gap, this work presents a novel physics-informed machine learning (PIML) approach of wear estimation by integrating analytical models with ML techniques. The PIML model utilizes a wear-inclusive thermo-mechanical model to estimating cutting forces considering flank wear and edge forces, with special focus on its adaptation to milling operations and addressing the complexities of milling dynamics. The methodology is demonstrated on Steel 1050, a widely used medium-carbon steel alloy in industrial machining applications. As shown by the results, this hybrid model shows high predictive accuracy, achieving R² values exceeding 98 % for force prediction and 95 % for tool wear estimation, with corresponding RMSE values below 14 N and 8 µm, respectively. Notably, the use of the PIML framework improved tool wear prediction accuracy by over 16 % compared to using ML alone. Another important finding is the significant role of edge forces under severe wear conditions, with their contribution to average cutting forces increasing from 40 % to 57 % at low feed rates, and from 27 % to 45 % at higher feed rates. Using this enhanced model, a simulation-based dataset was generated to train an inverse ML model for estimating tool wear considering milling forces and cutting parameters. The inverse ML model exhibited robust predictive performance, offering a practical and accurate solution for tool wear estimation. This study emphasizes the promising potential of integrating thermo-mechanical model with ML algorithms in machining applications, establishing a foundation of tool wear condition monitoring through milling force data. The presented approach can contribute to enhanced process control, optimized tool usage, and reduced operational costs. Furthermore, it supports the transition to Industry 4.0 by enabling automation and unsupervised manufacturing, where real-time tool wear monitoring and adaptive control can be achieved with minimal human intervention, driving more intelligent and efficient manufacturing systems.</div></div>","PeriodicalId":16227,"journal":{"name":"Journal of Manufacturing Systems","volume":"82 ","pages":"Pages 1192-1212"},"PeriodicalIF":14.2,"publicationDate":"2025-08-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144913027","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}