Pub Date : 2025-12-12DOI: 10.1016/j.jmsy.2025.11.022
Yuqi Cheng , Yunkang Cao , Haiming Yao , Wei Luo , Cheng Jiang , Hui Zhang , Weiming Shen
Industrial surface defect detection is vital for upholding product quality across contemporary manufacturing systems. As the expectations for precision, automation, and scalability intensify, conventional inspection approaches are increasingly found wanting in addressing real-world demands. Notable progress in computer vision and deep learning has substantially bolstered defect detection capabilities across both 2D and 3D modalities. A significant development has been the pivot from closed-set to open-set defect detection frameworks, which diminishes the necessity for extensive defect annotations and facilitates the recognition of novel anomalies. Despite such strides, a cohesive and contemporary understanding of industrial defect detection remains elusive. Consequently, this survey delivers an in-depth analysis of both closed-set and open-set defect detection strategies within 2D and 3D modalities, charting their evolution in recent years and underscoring the rising prominence of open-set techniques. We distill critical challenges inherent in practical detection environments and illuminate emerging trends, thereby providing a current and comprehensive vista of this swiftly progressing field.
{"title":"A comprehensive survey for real-world industrial surface defect detection: Challenges, approaches, and prospects","authors":"Yuqi Cheng , Yunkang Cao , Haiming Yao , Wei Luo , Cheng Jiang , Hui Zhang , Weiming Shen","doi":"10.1016/j.jmsy.2025.11.022","DOIUrl":"10.1016/j.jmsy.2025.11.022","url":null,"abstract":"<div><div>Industrial surface defect detection is vital for upholding product quality across contemporary manufacturing systems. As the expectations for precision, automation, and scalability intensify, conventional inspection approaches are increasingly found wanting in addressing real-world demands. Notable progress in computer vision and deep learning has substantially bolstered defect detection capabilities across both 2D and 3D modalities. A significant development has been the pivot from closed-set to open-set defect detection frameworks, which diminishes the necessity for extensive defect annotations and facilitates the recognition of novel anomalies. Despite such strides, a cohesive and contemporary understanding of industrial defect detection remains elusive. Consequently, this survey delivers an in-depth analysis of both closed-set and open-set defect detection strategies within 2D and 3D modalities, charting their evolution in recent years and underscoring the rising prominence of open-set techniques. We distill critical challenges inherent in practical detection environments and illuminate emerging trends, thereby providing a current and comprehensive vista of this swiftly progressing field.</div></div>","PeriodicalId":16227,"journal":{"name":"Journal of Manufacturing Systems","volume":"84 ","pages":"Pages 152-172"},"PeriodicalIF":14.2,"publicationDate":"2025-12-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145786611","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}
Pub Date : 2025-12-12DOI: 10.1016/j.jmsy.2025.12.004
Suyoung Park, Shreyes N. Melkote
Conventional supplier selection methods for assembled products have primarily relied on qualitative or business-level assessments of supplier capabilities, since manufacturing-related metrics such as product geometry, cost, time, and tolerance are heterogeneous and difficult to integrate into a unified evaluation. This reliance makes the identification of suppliers with adequate manufacturing capability particularly challenging as global supply chains grow increasingly complex. To address this gap, we propose the Deep Unsupervised Assembly Supplier Matcher (DU-ASM), an integrated data-driven framework that jointly embeds geometry, topology, and quantitative manufacturing attributes into a unified latent space for assembly-level supplier selection and ranking. Leveraging a graph autoencoder, DU-ASM reconstructs manufacturing attributes and supports robust supplier selection even with incomplete inputs. Experimental validation across multiple case studies demonstrates that DU-ASM achieves over 95 % supplier selection accuracy under complete requirements and over 90 % with partially masked inputs, while attaining mean normalized Discounted Cumulative Gain scores at top-k positions (nDCG@k) exceeding 0.99 in ranking tasks. By linking geometric, topological, and quantitative data, DU-ASM demonstrates both methodological novelty and strong quantitative performance, providing a scalable foundation for supplier matching at the assembly level and supporting multi-tier decision-making in future manufacturing supply networks.
{"title":"Deep unsupervised learning-based supplier selection and ranking for assembly manufacturing","authors":"Suyoung Park, Shreyes N. Melkote","doi":"10.1016/j.jmsy.2025.12.004","DOIUrl":"10.1016/j.jmsy.2025.12.004","url":null,"abstract":"<div><div>Conventional supplier selection methods for assembled products have primarily relied on qualitative or business-level assessments of supplier capabilities, since manufacturing-related metrics such as product geometry, cost, time, and tolerance are heterogeneous and difficult to integrate into a unified evaluation. This reliance makes the identification of suppliers with adequate manufacturing capability particularly challenging as global supply chains grow increasingly complex. To address this gap, we propose the Deep Unsupervised Assembly Supplier Matcher (DU-ASM), an integrated data-driven framework that jointly embeds geometry, topology, and quantitative manufacturing attributes into a unified latent space for assembly-level supplier selection and ranking. Leveraging a graph autoencoder, DU-ASM reconstructs manufacturing attributes and supports robust supplier selection even with incomplete inputs. Experimental validation across multiple case studies demonstrates that DU-ASM achieves over 95 % supplier selection accuracy under complete requirements and over 90 % with partially masked inputs, while attaining mean normalized Discounted Cumulative Gain scores at top-k positions (nDCG@k) exceeding 0.99 in ranking tasks. By linking geometric, topological, and quantitative data, DU-ASM demonstrates both methodological novelty and strong quantitative performance, providing a scalable foundation for supplier matching at the assembly level and supporting multi-tier decision-making in future manufacturing supply networks.</div></div>","PeriodicalId":16227,"journal":{"name":"Journal of Manufacturing Systems","volume":"84 ","pages":"Pages 173-188"},"PeriodicalIF":14.2,"publicationDate":"2025-12-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145786758","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}
Cyber–Physical Systems (CPSs) tightly interconnect digital and physical operations within production environments, enabling real-time monitoring, control, optimization, and autonomous decision-making that directly enhance manufacturing processes and productivity. The inherent complexity of these systems can lead to faults that require robust and interpretable diagnoses to maintain system dependability and operational efficiency. However, manual modeling of faulty behaviors requires extensive domain expertise and cannot leverage the low-level sensor data of the CPS. Furthermore, although powerful, deep learning-based techniques produce black-box diagnostics that lack interpretability, limiting their practical adoption. To address these challenges, we set forth a method that performs unsupervised characterization of system states and state transitions from low-level sensor data, uses several process mining techniques to model faults through interpretable stochastic Petri nets, simulates such Petri nets for a comprehensive understanding of system behavior under faulty conditions, and performs Petri net-based fault diagnosis. The method begins with detecting collective anomalies involving multiple samples in low-level sensor data. These anomalies are then transformed into structured event logs, enabling the data-driven discovery of interpretable Petri nets through process mining. By enhancing these Petri nets with timing distributions, the approach supports the simulation of faulty behaviors. Finally, faults can be diagnosed online by checking collective anomalies with the Petri nets and the corresponding simulations. The method is applied to the Robotic Arm Dataset (RoAD), a benchmark collected from a robotic arm deployed in a scale-replica smart manufacturing assembly line. The application to RoAD demonstrates the method’s effectiveness in modeling, simulating, and classifying faulty behaviors in CPSs. The modeling results demonstrate that our method achieves a satisfactory interpretability-simulation accuracy trade-off with up to 0.676 arc-degree simplicity, 0.395 R, and 0.088 RMSE. In addition, the fault identification results show that the method achieves an F1 score of up to 98.925%, while maintaining a low conformance checking time of 0.020 s, which competes with other deep learning-based methods.
{"title":"Process mining-driven modeling and simulation to enhance fault diagnosis in cyber–physical systems","authors":"Francesco Vitale , Nicola Dall’Ora , Sebastiano Gaiardelli , Enrico Fraccaroli , Nicola Mazzocca , Franco Fummi","doi":"10.1016/j.jmsy.2025.12.005","DOIUrl":"10.1016/j.jmsy.2025.12.005","url":null,"abstract":"<div><div>Cyber–Physical Systems (CPSs) tightly interconnect digital and physical operations within production environments, enabling real-time monitoring, control, optimization, and autonomous decision-making that directly enhance manufacturing processes and productivity. The inherent complexity of these systems can lead to faults that require robust and interpretable diagnoses to maintain system dependability and operational efficiency. However, manual modeling of faulty behaviors requires extensive domain expertise and cannot leverage the low-level sensor data of the CPS. Furthermore, although powerful, deep learning-based techniques produce black-box diagnostics that lack interpretability, limiting their practical adoption. To address these challenges, we set forth a method that performs unsupervised characterization of system states and state transitions from low-level sensor data, uses several process mining techniques to model faults through interpretable stochastic Petri nets, simulates such Petri nets for a comprehensive understanding of system behavior under faulty conditions, and performs Petri net-based fault diagnosis. The method begins with detecting collective anomalies involving multiple samples in low-level sensor data. These anomalies are then transformed into structured event logs, enabling the data-driven discovery of interpretable Petri nets through process mining. By enhancing these Petri nets with timing distributions, the approach supports the simulation of faulty behaviors. Finally, faults can be diagnosed online by checking collective anomalies with the Petri nets and the corresponding simulations. The method is applied to the Robotic Arm Dataset (RoAD), a benchmark collected from a robotic arm deployed in a scale-replica smart manufacturing assembly line. The application to RoAD demonstrates the method’s effectiveness in modeling, simulating, and classifying faulty behaviors in CPSs. The modeling results demonstrate that our method achieves a satisfactory interpretability-simulation accuracy trade-off with up to 0.676 arc-degree simplicity, 0.395 R<span><math><msup><mrow></mrow><mrow><mn>2</mn></mrow></msup></math></span>, and 0.088 RMSE. In addition, the fault identification results show that the method achieves an F1 score of up to 98.925%, while maintaining a low conformance checking time of 0.020 s, which competes with other deep learning-based methods.</div></div>","PeriodicalId":16227,"journal":{"name":"Journal of Manufacturing Systems","volume":"84 ","pages":"Pages 189-206"},"PeriodicalIF":14.2,"publicationDate":"2025-12-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145786757","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}
Pub Date : 2025-12-10DOI: 10.1016/j.jmsy.2025.11.024
Liguo Zhang , Qinghua Song , Haifeng Ma , Zhanqiang Liu
Deep learning-based tool wear monitoring methods often suffer from poor interpretability, parameter redundancy, and limited cross-domain generalization, especially under variable operating conditions and when only small sample sizes are available. These limitations hinder their practical deployment in practical manufacturing environments. To overcome these challenges, this study proposes a highly interpretable and lightweight tool wear monitoring framework tailored for small-sample, variable-condition scenarios. The method randomly extracts fixed-length segments from vibration signals to construct small-sample datasets, utilizing both signal envelope spectrum and cutting time as input features. Guided by the degradation law of tool wear, explicit physical constraints are imposed on the solution space of a Kolmogorov‑Arnold Fourier neural network, yielding a physics‑informed data‑driven model. SHAP analysis is employed to quantify the contribution of each feature, enhancing model transparency. Validation on public datasets under both single‑ and multi‑condition settings demonstrates that the proposed method delivers excellent performance across diverse operating conditions, achieving a stable prediction R² of up to 95 %, an inference latency of only 2 ms, and a reduction of approximately 90 % in model parameters. This solution can be integrated into the edge computing platform of CNC systems, making it particularly suitable for machining scenarios with high real-time requirements. It offers a lightweight, precise, and efficient monitoring capability for smart factories, contributing simultaneously to improvements in product quality and manufacturing efficiency.
{"title":"PI-KAF: A physics-informed constrained online interpretable monitoring method for tool wear","authors":"Liguo Zhang , Qinghua Song , Haifeng Ma , Zhanqiang Liu","doi":"10.1016/j.jmsy.2025.11.024","DOIUrl":"10.1016/j.jmsy.2025.11.024","url":null,"abstract":"<div><div>Deep learning-based tool wear monitoring methods often suffer from poor interpretability, parameter redundancy, and limited cross-domain generalization, especially under variable operating conditions and when only small sample sizes are available. These limitations hinder their practical deployment in practical manufacturing environments. To overcome these challenges, this study proposes a highly interpretable and lightweight tool wear monitoring framework tailored for small-sample, variable-condition scenarios. The method randomly extracts fixed-length segments from vibration signals to construct small-sample datasets, utilizing both signal envelope spectrum and cutting time as input features. Guided by the degradation law of tool wear, explicit physical constraints are imposed on the solution space of a Kolmogorov‑Arnold Fourier neural network, yielding a physics‑informed data‑driven model. SHAP analysis is employed to quantify the contribution of each feature, enhancing model transparency. Validation on public datasets under both single‑ and multi‑condition settings demonstrates that the proposed method delivers excellent performance across diverse operating conditions, achieving a stable prediction R² of up to 95 %, an inference latency of only 2 ms, and a reduction of approximately 90 % in model parameters. This solution can be integrated into the edge computing platform of CNC systems, making it particularly suitable for machining scenarios with high real-time requirements. It offers a lightweight, precise, and efficient monitoring capability for smart factories, contributing simultaneously to improvements in product quality and manufacturing efficiency.</div></div>","PeriodicalId":16227,"journal":{"name":"Journal of Manufacturing Systems","volume":"84 ","pages":"Pages 135-151"},"PeriodicalIF":14.2,"publicationDate":"2025-12-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145786760","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}
Pub Date : 2025-12-10DOI: 10.1016/j.jmsy.2025.11.017
Yang Zhang, Xu Wang, Jiacheng Cui, Yulin Jin, Qihang Chen, Yongkang Lu, Wei Liu
Real-time, high-fidelity, and interactive monitoring of global mechanical responses during the assembly of large-scale, flexible aerospace structures remains a critical and unresolved challenge. Here, we present a perceptive assembly framework that integrates a distributed edge vision network, physics-informed sparse sensing, and immersive augmented reality (AR) visualization to enable full-field structural state monitoring. A modular edge sensing system is deployed to achieve fast, high-precision measurement of distributed displacements across meter-scale components. To overcome view discontinuities, a hierarchical coordinate transformation pipeline is introduced for global registration under non-overlapping camera views. Building on sparse displacement data, we develop a constrained sensor optimization strategy that enables real-time reconstruction of global displacement and strain fields. Through HoloLens 2, the system provides intuitive AR overlays that deliver immersive, in-situ mechanical feedback during assembly. Validation experiments on composite panels demonstrate sub-millimeter reconstruction accuracy and real-time performance, significantly enhancing transparency and decision-making in the assembly process. This work establishes a scalable AR-based perception infrastructure for next-generation intelligent manufacturing of large aerospace structures.
{"title":"Towards perceptive assembly: An edge vision network-enabled augmented reality (AR) monitoring method for global shape and mechanical information in large aerospace components","authors":"Yang Zhang, Xu Wang, Jiacheng Cui, Yulin Jin, Qihang Chen, Yongkang Lu, Wei Liu","doi":"10.1016/j.jmsy.2025.11.017","DOIUrl":"10.1016/j.jmsy.2025.11.017","url":null,"abstract":"<div><div>Real-time, high-fidelity, and interactive monitoring of global mechanical responses during the assembly of large-scale, flexible aerospace structures remains a critical and unresolved challenge. Here, we present a perceptive assembly framework that integrates a distributed edge vision network, physics-informed sparse sensing, and immersive augmented reality (AR) visualization to enable full-field structural state monitoring. A modular edge sensing system is deployed to achieve fast, high-precision measurement of distributed displacements across meter-scale components. To overcome view discontinuities, a hierarchical coordinate transformation pipeline is introduced for global registration under non-overlapping camera views. Building on sparse displacement data, we develop a constrained sensor optimization strategy that enables real-time reconstruction of global displacement and strain fields. Through HoloLens 2, the system provides intuitive AR overlays that deliver immersive, in-situ mechanical feedback during assembly. Validation experiments on composite panels demonstrate sub-millimeter reconstruction accuracy and real-time performance, significantly enhancing transparency and decision-making in the assembly process. This work establishes a scalable AR-based perception infrastructure for next-generation intelligent manufacturing of large aerospace structures.</div></div>","PeriodicalId":16227,"journal":{"name":"Journal of Manufacturing Systems","volume":"84 ","pages":"Pages 117-134"},"PeriodicalIF":14.2,"publicationDate":"2025-12-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145786759","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}
Pub Date : 2025-12-09DOI: 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.
{"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}
Pub Date : 2025-12-05DOI: 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.
{"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 +7.4<!--> <!-->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.</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}
Pub Date : 2025-12-04DOI: 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.
{"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}
Pub Date : 2025-12-01DOI: 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.
{"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}
Pub Date : 2025-11-29DOI: 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.
{"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}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号