Journal of Manufacturing Systems最新文献

{"title":"Time series prediction for lock nuts production quality driven by information fusion and data-model hybrid","authors":"Tianzi Tian ,&nbsp;Dunwang Qin ,&nbsp;Ning Wang ,&nbsp;Jun Yang ,&nbsp;Kai Wu","doi":"10.1016/j.jmsy.2025.08.005","DOIUrl":"10.1016/j.jmsy.2025.08.005","url":null,"abstract":"<div><div>The locking performance of nuts directly impacts the lifespan and reliability of assembled products. However, certain aerospace nuts undergo over 120 h of rigorous testing per batch, causing increased costs and delayed product delivery. Therefore, accurately predicting production quality and remaining testing time (RTT) is crucial for improving efficiency. Facing this new challenge, this paper proposes a data-model hybrid time series prediction method based on quality information fusion. First, considering that the monitoring data contains two sets of related features, we introduce a multi-task parallel deep learning (MTL) network with a temporal self-attention mechanism (TSAM). The TSAM assigns importance to key degradation information, while MTL leverages shared feature information to capture more accurate long-term trends. Next, considering the multi-stage nature and uncertainty of the degradation process, a semi-empirical physical degradation model is constructed, where stage identification is achieved using the Pruned Exact Linear Time (PELT) method, and uncertainty is estimated through Particle Filtering (PF). The Bayesian framework enables hybrid correction between the data-based and the model-based methods, integrating the strengths of both. Finally, experimental results demonstrate that the proposed method outperforms traditional models, effectively achieving more accurate quality predictions.</div></div>","PeriodicalId":16227,"journal":{"name":"Journal of Manufacturing Systems","volume":"82 ","pages":"Pages 1171-1191"},"PeriodicalIF":14.2,"publicationDate":"2025-08-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144902874","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":"Optimisation method for semiconductor wafer manufacturing system scheduling: Reinforcement learning with decision graph guiding","authors":"Da Chen ,&nbsp;Jie Zhang ,&nbsp;Lihui Wu ,&nbsp;Peng Zhang ,&nbsp;Youlong Lv ,&nbsp;Junliang Wang ,&nbsp;Hong Wang","doi":"10.1016/j.jmsy.2025.08.004","DOIUrl":"10.1016/j.jmsy.2025.08.004","url":null,"abstract":"<div><div>Semiconductor wafer fabrication, as a large-scale and complex discrete manufacturing system, presents significant challenges in shop floor scheduling due to its scale, uncertainty, and re-entrant processing. Additionally, effectively leveraging historical scheduling decision data remains a challenge, limiting the ability of scheduling algorithms to accurately assess the current system state. To address these issues, this paper proposes a reinforcement learning-based optimisation method guided by decision graphs. First, we introduce a multidimensional heterogeneous disambiguation graph to comprehensively represent the operational state of the wafer manufacturing system. Second, we design a graph neural network to characterise the multidimensional disambiguation graph and learn from historical decision-making experiences. Finally, we propose a decision graph-guided action strategy that optimises the reinforcement learning policy, improving training efficiency and the accuracy of action selection. Experimental results demonstrate that our method achieves superior generalisation performance and outperforms traditional approaches. This study provides an effective solution for optimising scheduling in semiconductor wafer manufacturing systems.</div></div>","PeriodicalId":16227,"journal":{"name":"Journal of Manufacturing Systems","volume":"82 ","pages":"Pages 1158-1170"},"PeriodicalIF":14.2,"publicationDate":"2025-08-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144902873","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":"Self-supervised time–frequency feature alignment for process monitoring of cyber–physical CNC machines","authors":"Yadong Xu ,&nbsp;J.C. Ji ,&nbsp;Yuxin Sun ,&nbsp;Sihan Huang ,&nbsp;Zhiheng Zhao ,&nbsp;George Q. Huang","doi":"10.1016/j.jmsy.2025.08.007","DOIUrl":"10.1016/j.jmsy.2025.08.007","url":null,"abstract":"<div><div>Self-supervised learning excels at uncovering latent features from incomplete data, thereby providing robust support for downstream applications. Capitalizing on this strength, a growing number of fault diagnosis models have been developed to monitor CNC machine tools, which are essential to modern manufacturing. These machines operate under demanding conditions – characterized by high speeds and heavy loads – and consequently generate mechanical signals with pronounced nonlinearity. Such inherent nonlinearity poses significant challenges for conventional feature extraction methods, necessitating advanced self-supervised techniques to effectively capture and interpret the underlying fault-related features for reliable condition monitoring. In this research, we introduce a self-supervised time–frequency feature alignment (STFA) algorithm for monitoring the manufacturing processes of industrial CNC machine tools. The STFA algorithm initially employs two domain-specific modules to extract time–frequency features from surveillance signals. A modern CNN is utilized to extract spatiotemporal information from the time domain, while a multi-scale CNN captures multi-granular features from the frequency domain. Subsequently, a dedicated time–frequency feature alignment module (TFAM) maps these features into a unified space, thereby exploiting their complementarity and enabling a more comprehensive representation. The STFA algorithm is trained through a dual-stage process—first, a pre-training phase to establish robust feature representations from unlabeled data, followed by a fine-tuning stage using a limited number of labeled samples to adapt the model for precise fault diagnosis. The effectiveness of the proposed STFA algorithm is validated using two manufacturing datasets collected from industrial CNC machine tools.</div></div>","PeriodicalId":16227,"journal":{"name":"Journal of Manufacturing Systems","volume":"82 ","pages":"Pages 1145-1157"},"PeriodicalIF":14.2,"publicationDate":"2025-08-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144896625","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":"Cerebrum twin: A 6D semantic digital twin of multi-lobe digital brain functions for human-centric Industry 5.0","authors":"Hanwei Teng ,&nbsp;Shuo Chen ,&nbsp;Changping Li ,&nbsp;Shujian Li ,&nbsp;Rendi Kurniawan ,&nbsp;Moran Xu ,&nbsp;Jielin Chen ,&nbsp;Tae Jo Ko","doi":"10.1016/j.jmsy.2025.08.009","DOIUrl":"10.1016/j.jmsy.2025.08.009","url":null,"abstract":"<div><div>Industry 5.0 highlights the need for human-centric and adaptive intelligence in smart manufacturing. This paper proposes the Cerebrum Twin (CT), a brain-inspired, six-dimensional semantic digital twin (SDT) system that unifies five human-like senses, including listening, speaking, reading, writing, and looking, within a cohesive multi-lobe digital brain framework. CT integrates real-time physical signals from force, vibration, and vision sensors by leveraging a synergistic ensemble of advanced artificial intelligence (AI) modules, such as Extreme Gradient Boosting (XGBoost), ConvNeXt V2, Efficient Sub-Pixel Convolutional Networks (ESPCN), stacked sparse autoencoder with supervision (SSAES), large language models (LLM), and reinforcement learning (RL). Uniquely, CT establishes a closed-loop semantic feedback mechanism, enabling dynamic perception, multimodal semantic abstraction, signal-driven prediction, adaptive parameter optimization, and intuitive voice-based human interaction. This holistic integration bridges the physical, semantic, and cognitive layers of CNC machining, supporting robust, transparent, and operator-oriented decision-making. The proposed system was validated through ultrasonic vibration-assisted blade dicing (UVABD) experiments. CT reduced dicing force prediction error by 39.86 %, improved tool wear prediction accuracy by 29.59 %, and decreased edge chipping severity by 60.47 % compared to the baseline model. These results demonstrate that a semantically empowered, multisensory digital twin (DT), enabled by real-time physical–semantic–AI fusion and human-in-the-loop optimization, can significantly enhance intelligent manufacturing performance and fulfill the vision of Industry 5.0.</div></div>","PeriodicalId":16227,"journal":{"name":"Journal of Manufacturing Systems","volume":"82 ","pages":"Pages 1125-1144"},"PeriodicalIF":14.2,"publicationDate":"2025-08-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144864294","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":"KGESM: A knowledge graph embedding-based similarity matching model for intelligent assembly process generation","authors":"Youzi Xiao ,&nbsp;Shuai Zheng ,&nbsp;Hucheng Feng ,&nbsp;Yejia Huang ,&nbsp;Jiewu Leng ,&nbsp;Jun Hong","doi":"10.1016/j.jmsy.2025.08.002","DOIUrl":"10.1016/j.jmsy.2025.08.002","url":null,"abstract":"<div><div>The assembly process knowledge graph is an important carrier for assembly process knowledge management and reuse in manufacturing enterprises. Its important application scenario is to generate assembly process. However, the primary method is to perform simple retrieval on the knowledge graph using the graph database tool. This method can only retrieve the identical assembly process due to the lack of deep semantics for process knowledge, which restricts the flexibility of assembly process generation. Since there may be different representations and descriptions of identical process knowledge, it is necessary to mine deep semantic information to achieve process generation. To address these challenges, we propose a knowledge graph embedding-based similarity matching model for intelligent assembly process generation. First, we build a knowledge graph embedding-based similarity matching model called KGESM. Then, we construct a dataset consisting of a series of assembly process knowledge pairs extracted from actual electronic equipment manufacturing documents. Finally, the trained model is used to generate assembly processes according to new manufacturing needs. We conduct comprehensive experiments on the electronic equipment assembly process knowledge graph, where the mean square error of similarity matching achieves <span><math><mrow><mn>1</mn><mo>.</mo><mn>200</mn><mspace></mspace><mo>×</mo><mspace></mspace><mn>1</mn><msup><mrow><mn>0</mn></mrow><mrow><mo>−</mo><mn>3</mn></mrow></msup></mrow></math></span>. Unlike traditional knowledge graph retrieval, similarity matching based on assembly process knowledge graph embedding has the advantage of fusing the features of assembly process nodes and assembly relations. Furthermore, examples of electronic equipment assembly processes are generated, and the highest similarity score of the generated assembly processes is 0.939, which proves the feasibility of our method in the equipment manufacturing field.</div></div>","PeriodicalId":16227,"journal":{"name":"Journal of Manufacturing Systems","volume":"82 ","pages":"Pages 1110-1124"},"PeriodicalIF":14.2,"publicationDate":"2025-08-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144852035","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":"Lead frame surface defect detection based on geometric constraints and adaptive dual-branch normality recall for semiconductor manufacturing","authors":"Tingrui Sun ,&nbsp;Zhiwei Li ,&nbsp;Xinjie Xiao ,&nbsp;Xiangyang Hu ,&nbsp;Zhihong Sun","doi":"10.1016/j.jmsy.2025.06.016","DOIUrl":"10.1016/j.jmsy.2025.06.016","url":null,"abstract":"<div><div>Surface defect detection in lead frames using machine vision is crucial for ensuring the quality and reliability of semiconductor manufacturing. While high-resolution industrial cameras are widely used, detecting tiny defects across hundreds of unit cells remains a significant challenge. Additionally, imbalanced defect data distributions in industrial environments exacerbate the detection challenge. To address these, a novel framework specifically designed for lead frame surface defect detection is proposed. The framework consists of two main components: multi-scale geometric constraint matching (MGCM) and dual-branch attention-based normality recall (DB-ANR). MGCM is a specialized algorithm designed for array-structured detection, efficiently extracting unit cells from high-resolution images by leveraging geometric constraints and eliminating redundant matching points, ensuring stable and precise results. DB-ANR addresses data imbalance and normal pattern forgetting by training on normal data to store patterns, which are recalled during inference to enhance detection accuracy. Additionally, the adaptive local–global dual attention module dynamically balances the contributions of local and global features, enabling robust detection across various defect types. Experiments on a self-constructed dataset with five types of lead frames demonstrate the effectiveness of the proposed framework. MGCM reliably extracts unit cells from high-resolution images, while DB-ANR achieves an AUC of 97.76%, PRO of 89.12%, and F1-score of 81.6%. The model also demonstrates efficient memory usage and fast inference speed, meeting deployment requirements in real industrial scenarios. Furthermore, the proposed array-oriented detection framework is not limited to lead frames and can be extended to other semiconductor applications, such as chip and wafer quality detection, where similar array-structured patterns exist.</div></div>","PeriodicalId":16227,"journal":{"name":"Journal of Manufacturing Systems","volume":"82 ","pages":"Pages 1091-1109"},"PeriodicalIF":14.2,"publicationDate":"2025-08-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144842094","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 review of architecture features for distributed and resilient industrial cyber–physical systems","authors":"Jorge Alves ,&nbsp;Pedro Sousa ,&nbsp;Tiago Cruz ,&nbsp;Jérôme Mendes","doi":"10.1016/j.jmsy.2025.07.012","DOIUrl":"10.1016/j.jmsy.2025.07.012","url":null,"abstract":"<div><div>Industrial processes increasingly demand the design of reliable systems with robustness, resilience, and autonomy capabilities. Distributed cyber–physical systems have been seen as reliable and potential solutions to these demands, and they are being more prepared and aligned to the incorporation of new technologies and methodologies. Some technologies and methodologies are service-oriented architectures, multi-agent systems, IEC 61499 Standard, virtualization, and autonomous capabilities. These technologies ensure adaptative dynamic systems that detect and correct changes or faults autonomously, maximizing resource use in dynamic environments and the operation time of industry processes. The fusion of technologies and methodologies allows the optimization of industrial Cyber–Physical Systems (CPSs), making them more flexible, agile, secure, scalable, reliable, and efficient. This work aims to identify fundamental ideas from the recent literature on distributed CPS architectures for industrial environments. For this, this review studies distributed architectures and the fundamental/essential technologies and methodologies contributing to their maturation and growth. The paper studies the fundamental concepts, applications, and integration of communication protocols, IEC 61499 Standard, virtualization, multi-agent, fault-tolerant, and autonomous methodologies on distributed CPS. The review presents the different features addressed to develop a well-defined and optimized distributed architecture and the integration and relationship of different methodologies. Also, the respective strengths, shortcomings, and opportunities for future work are identified. The contribution of the presented work is to provide the crucial technologies and essential concepts to develop a distributed and resilient industrial CPS in the future.</div></div>","PeriodicalId":16227,"journal":{"name":"Journal of Manufacturing Systems","volume":"82 ","pages":"Pages 1069-1090"},"PeriodicalIF":14.2,"publicationDate":"2025-08-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144831266","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 set of novel Mixed-Integer Linear Programming models for robotic disassembly sequence planning in sustainable remanufacturing","authors":"Miguel Reula ,&nbsp;Consuelo Parreño-Torres ,&nbsp;F. Javier Ramírez","doi":"10.1016/j.jmsy.2025.07.014","DOIUrl":"10.1016/j.jmsy.2025.07.014","url":null,"abstract":"<div><div>Disassembly is a necessary and critical step in the remanufacturing process of end-of-life products. The goal is to support decision-making in the sequencing of robotic disassembly, the selection of appropriate tools and the final use of the disassembled components (reuse, remanufacturing, recycling or disposal). While this problem has been widely studied by researchers and practitioners worldwide, much of the focus has been on heuristic and metaheuristic approaches rather than on the development of mathematical models. This study proposes a set of novel Mixed Integer Linear Programming models that completely describes the problem and generalizes those already present in the literature. The optimal solution obtained by the models combines the optimal sequence planning and the most suitable recovery option for each component, achieving the maximum profit from the disassembly process. Moreover, the formulations can be easily adapted to solve different disassembly modes: complete, partial or selective, as well as other specific variants. Computational experiments based on two industrial gear pumps are carried out and, as will be shown, the results demonstrate that the mathematical models are able to reach optimal solutions for the complete disassembly sequence planning problem in a short computational time.</div></div>","PeriodicalId":16227,"journal":{"name":"Journal of Manufacturing Systems","volume":"82 ","pages":"Pages 1046-1068"},"PeriodicalIF":14.2,"publicationDate":"2025-08-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144831263","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":"Task allocation based on single-limb actions for augmented reality-assisted human-robot collaboration","authors":"Kai-Wen Tien ,&nbsp;Yu-Jen Lu ,&nbsp;Chih-Hsing Chu","doi":"10.1016/j.jmsy.2025.08.006","DOIUrl":"10.1016/j.jmsy.2025.08.006","url":null,"abstract":"<div><div>Human-robot collaboration (HRC) is a key technology for enabling the human-centric vision of Industry 5.0. Collaborative robots have been deployed on the shop floor to support manual operations and enhance overall productivity. However, poor coordination between robotic systems and human workers may compromise collaborative performance and raise safety risks. This study proposes a new task allocation scheme based on single-limb actions to enhance process efficiency in HRC. Each single-limb task, composed of basic motion elements identified through Therblig analysis, is assigned to either a human or a robotic agent based on individual capabilities and spatial proximity. The scheme is formulated as a mixed-integer programming problem and solved using a Random-Key Genetic Algorithm. The allocation result is validated through a collaborative assembly process by comparing it with a traditional method that does not differentiate between the hands during task assignment. An augmented reality (AR)-assisted tool is developed to support participants in performing their assigned tasks with enhanced situational awareness during an actual experiment. Experimental results indicate that the assembly sequence generated by the proposed scheme leads to a shorter makespan. This study demonstrates that fine-grained planning enables more efficient utilization of human and robotic resources, and highlights the potential of AR to facilitate the practical implementation of complex HRC processes.</div></div>","PeriodicalId":16227,"journal":{"name":"Journal of Manufacturing Systems","volume":"82 ","pages":"Pages 1000-1019"},"PeriodicalIF":14.2,"publicationDate":"2025-08-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144831265","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":"LLM-based multi-agent task planning for human-robot collaborative assembly balancing operator experience and efficiency","authors":"Binbin Wang ,&nbsp;Lianyu Zheng ,&nbsp;Yiwei Wang ,&nbsp;Zhonghua Qi","doi":"10.1016/j.jmsy.2025.08.003","DOIUrl":"10.1016/j.jmsy.2025.08.003","url":null,"abstract":"<div><div>In human-robot collaborative assembly (HRCA), systematic task planning is required to enhance the coordination between human and robot, prevent execution conflicts, and improve assembly efficiency. However, traditional HRCA task planning methods are often tailored to specific tasks, lacking generality and requiring significant manual involvement. Meanwhile, overemphasis on efficiency neglects the work experience of operators. This paper proposes MATP, a multi-agent task planning method for HRCA based on large language models (LLMs), aimed at enhancing human-robot collaboration (HRC), avoiding execution conflicts, and balancing operator experience with production efficiency. The method creates multiple agents, each of which explicitly defines its distinct role and responsibilities such that they can collaboratively plan HRCA tasks. A standardized and automated planning process is developed where the input is the assembly task and the output is the generated optimal human-robot task allocation sequence. Specifically, MATP firstly decomposes assembly tasks into action-level subtasks. Then, it evaluates the states of both the operator and the robot from multiple perspectives including fatigue, postural comfort and human-robot trust. Task allocation is finally achieved through deep collaboration between the LLM and the genetic algorithm (GA). Validation in the electronic product assembly scenario demonstrate that MATP outperforms single-agent and traditional method in HRCA task planning. In addition, it effectively balances operator experience and assembly efficiency, significantly enhancing planning efficiency and dynamic adaptability.</div></div>","PeriodicalId":16227,"journal":{"name":"Journal of Manufacturing Systems","volume":"82 ","pages":"Pages 1020-1045"},"PeriodicalIF":14.2,"publicationDate":"2025-08-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144831262","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}