Pub Date : 2026-01-01Epub Date: 2025-11-21DOI: 10.1016/j.jii.2025.101012
Wei Gong , Shuang Qiao , Chenhong Cao , Shilei Tan , Junliang Ye , Haoxiang Liu , Si Chen , Xuesong Wang
In smart manufacturing, accurate sensor fault diagnosis is essential for operational integrity. However, the direct application of Large Language Models (LLMs) to this task yields unstructured analyses and inefficient resource use. To address these challenges, we propose a novel multi-agent framework that instills a structured, modular, and adaptive reasoning process. The framework features a Reasoning Module to classify problem complexity and a Decision Module that employs a difficulty-aware workflow. Simple problems are resolved directly, while complex cases activate a deliberative debate among multiple agents to form a consensus. Evaluated on the specialized FailureSensorIQ benchmark, our framework significantly boosts the performance of open-source LLMs. For example, Llama3.1-8B-instruct’s accuracy surged from 36.5% to 54.6%—an 18.1 percentage point improvement. Crucially, our method empowers smaller 7B/8B models to surpass larger, proprietary models like GPT-4o-mini. Ablation studies validate that our dynamic routing mechanism provides an optimal trade-off between diagnostic accuracy and computational cost. This work establishes a new paradigm for industrial fault diagnosis, improving accuracy, interpretability, and resource efficiency, thereby paving the way for reliable and accessible AI in critical manufacturing systems.
{"title":"Harnessing collective intelligence of multi-agent LLM systems for sensor failure reasoning in smart manufacturing","authors":"Wei Gong , Shuang Qiao , Chenhong Cao , Shilei Tan , Junliang Ye , Haoxiang Liu , Si Chen , Xuesong Wang","doi":"10.1016/j.jii.2025.101012","DOIUrl":"10.1016/j.jii.2025.101012","url":null,"abstract":"<div><div>In smart manufacturing, accurate sensor fault diagnosis is essential for operational integrity. However, the direct application of Large Language Models (LLMs) to this task yields unstructured analyses and inefficient resource use. To address these challenges, we propose a novel multi-agent framework that instills a structured, modular, and adaptive reasoning process. The framework features a Reasoning Module to classify problem complexity and a Decision Module that employs a difficulty-aware workflow. Simple problems are resolved directly, while complex cases activate a deliberative debate among multiple agents to form a consensus. Evaluated on the specialized FailureSensorIQ benchmark, our framework significantly boosts the performance of open-source LLMs. For example, Llama3.1-8B-instruct’s accuracy surged from 36.5% to 54.6%—an 18.1 percentage point improvement. Crucially, our method empowers smaller 7B/8B models to surpass larger, proprietary models like GPT-4o-mini. Ablation studies validate that our dynamic routing mechanism provides an optimal trade-off between diagnostic accuracy and computational cost. This work establishes a new paradigm for industrial fault diagnosis, improving accuracy, interpretability, and resource efficiency, thereby paving the way for reliable and accessible AI in critical manufacturing systems.</div></div>","PeriodicalId":55975,"journal":{"name":"Journal of Industrial Information Integration","volume":"49 ","pages":"Article 101012"},"PeriodicalIF":10.4,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145567554","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-01Epub Date: 2025-11-10DOI: 10.1016/j.jii.2025.101010
Cheng Wang , Shiyong Wang , Wujie Zhang , Min Xia , Zhenfeng Shi
The perception-control-execution layered architecture, commonly used in current embodied intelligent robot control systems, suffers from inherent latency caused by its serial processing mechanism, which limits a robot's ability to respond to sudden disturbances, such as falls and collisions. To overcome this bottleneck, this study proposes a biomimetic emergency response control architecture for embodied intelligent robots. This architecture is inspired by the collaborative control principles of higher-level central control and spinal reflex mechanisms in the human nervous system. In addition, this architecture decouples the process of conventional decision-making and planning from emergency response mechanisms, thus constructing a four-layer heterogeneous control framework containing a perception-planning layer, a motion control layer, an emergency response layer, and a physical execution layer. The perception-planning layer is responsible for scene understanding and long-term planning. The motion control layer performs precise control of the entire body's posture and motion trajectory. The emergency response layer transmits upper-layer control commands under normal conditions, achieving fine motion control. In the event of sudden disturbances, the emergency response layer receives sensor signals directly, without waiting for the perception and decision results of the perception-planning layer. A lightweight, online-learnable reflex rule base, such as a balance compensation mechanism based on contact force mutation thresholds, enables rapid response to sudden disturbances. The emergency response layer is used as an independent module in the embodied intelligent control architecture, addressing the serial delay problem and offering an innovative solution for improving motion robustness and operational safety of robots in highly dynamic and uncertain environments.
{"title":"From brain to reflex: An emergency response control architecture for embodied intelligent robots","authors":"Cheng Wang , Shiyong Wang , Wujie Zhang , Min Xia , Zhenfeng Shi","doi":"10.1016/j.jii.2025.101010","DOIUrl":"10.1016/j.jii.2025.101010","url":null,"abstract":"<div><div>The perception-control-execution layered architecture, commonly used in current embodied intelligent robot control systems, suffers from inherent latency caused by its serial processing mechanism, which limits a robot's ability to respond to sudden disturbances, such as falls and collisions. To overcome this bottleneck, this study proposes a biomimetic emergency response control architecture for embodied intelligent robots. This architecture is inspired by the collaborative control principles of higher-level central control and spinal reflex mechanisms in the human nervous system. In addition, this architecture decouples the process of conventional decision-making and planning from emergency response mechanisms, thus constructing a four-layer heterogeneous control framework containing a perception-planning layer, a motion control layer, an emergency response layer, and a physical execution layer. The perception-planning layer is responsible for scene understanding and long-term planning. The motion control layer performs precise control of the entire body's posture and motion trajectory. The emergency response layer transmits upper-layer control commands under normal conditions, achieving fine motion control. In the event of sudden disturbances, the emergency response layer receives sensor signals directly, without waiting for the perception and decision results of the perception-planning layer. A lightweight, online-learnable reflex rule base, such as a balance compensation mechanism based on contact force mutation thresholds, enables rapid response to sudden disturbances. The emergency response layer is used as an independent module in the embodied intelligent control architecture, addressing the serial delay problem and offering an innovative solution for improving motion robustness and operational safety of robots in highly dynamic and uncertain environments.</div></div>","PeriodicalId":55975,"journal":{"name":"Journal of Industrial Information Integration","volume":"49 ","pages":"Article 101010"},"PeriodicalIF":10.4,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145485560","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-01Epub Date: 2025-11-01DOI: 10.1016/j.jii.2025.100998
Biswajit Sarkar , Rekha Guchhait , Mitali Sarkar
Electric vehicle production has recently gained popularity due to increasing emissions. The research on electric vehicle production concerning technical, economic, and environmental aspects is very less compared to the traditional vehicle. This research studies a mixed-type electric vehicle production system that produces spare parts and finally assembles all spare parts for the vehicle. The spare parts production combines in-line production with returnable items and outsourcing. An automated inspection for both spare parts and vehicles is included within the system. Two different types of machines work for the production process: Machine 1 for spare parts and Machine 2 for vehicles. As the basic purpose is to provide an ecofriendly logistics facility, the manufacturing company takes care of carbon emissions from the system, customer satisfaction, and the green quality of vehicles. Necessary and sufficient conditions of classical optimization find global optimum solutions. Results show that green technology and customer satisfaction are two important factors for vehicle production. Comparative discussions, sensitivity, and robust analysis are provided to validate the theoretical contributions. The proposed mixed-type production model earns 85.32% more profit than a traditional production model. The electric vehicle provides a 96% customer satisfaction with an increase of 68.97% profit without customer satisfaction.
{"title":"A sustainable electric vehicle smart production with work-in-process inventory of outsourced spare parts","authors":"Biswajit Sarkar , Rekha Guchhait , Mitali Sarkar","doi":"10.1016/j.jii.2025.100998","DOIUrl":"10.1016/j.jii.2025.100998","url":null,"abstract":"<div><div>Electric vehicle production has recently gained popularity due to increasing emissions. The research on electric vehicle production concerning technical, economic, and environmental aspects is very less compared to the traditional vehicle. This research studies a mixed-type electric vehicle production system that produces spare parts and finally assembles all spare parts for the vehicle. The spare parts production combines in-line production with returnable items and outsourcing. An automated inspection for both spare parts and vehicles is included within the system. Two different types of machines work for the production process: Machine 1 for spare parts and Machine 2 for vehicles. As the basic purpose is to provide an ecofriendly logistics facility, the manufacturing company takes care of carbon emissions from the system, customer satisfaction, and the green quality of vehicles. Necessary and sufficient conditions of classical optimization find global optimum solutions. Results show that green technology and customer satisfaction are two important factors for vehicle production. Comparative discussions, sensitivity, and robust analysis are provided to validate the theoretical contributions. The proposed mixed-type production model earns 85.32% more profit than a traditional production model. The electric vehicle provides a 96% customer satisfaction with an increase of 68.97% profit without customer satisfaction.</div></div>","PeriodicalId":55975,"journal":{"name":"Journal of Industrial Information Integration","volume":"49 ","pages":"Article 100998"},"PeriodicalIF":10.4,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145424057","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-01Epub Date: 2025-11-09DOI: 10.1016/j.jii.2025.101009
Hongyi Qu , Luo Fang , Jinbiao Tan
In the process of multi-cavity hot runner injection molding, the issue of mold filling imbalance caused by uneven temperature distribution significantly affects the quality of precision products such as optical lenses. Traditional methods primarily rely on mold thermal structure design and lack dynamic optimization strategies aimed at product quality. This paper proposes an embodied intelligent online optimization method integrated with digital twin technology, which fundamentally overcomes the limitations of traditional fixed-temperature control and offline optimization by enabling dynamic, data-driven adjustment of process parameters. By utilizing real-time process information from sensor readings within a batch, along with product quality data obtained through machine vision inspection after each batch, and employing a ‘mutual feedback’ sharing mechanism for multi-cavity process information, a ‘time-batch’ dual-scale real-time iterative learning and updating framework is established for the digital twin model. This approach enables closed-loop adaptive optimization of the mold filling state. Experimental results show that this method significantly outperforms traditional fixed temperature setting controls in terms of profile accuracy, offering an innovative solution for high-precision injection molding.
{"title":"An embodied intelligence-based online optimization methodology for injection molding process using multi-cavity hot-runner","authors":"Hongyi Qu , Luo Fang , Jinbiao Tan","doi":"10.1016/j.jii.2025.101009","DOIUrl":"10.1016/j.jii.2025.101009","url":null,"abstract":"<div><div>In the process of multi-cavity hot runner injection molding, the issue of mold filling imbalance caused by uneven temperature distribution significantly affects the quality of precision products such as optical lenses. Traditional methods primarily rely on mold thermal structure design and lack dynamic optimization strategies aimed at product quality. <em>This paper proposes an embodied intelligent online optimization method integrated with digital twin technology, which fundamentally overcomes the limitations of traditional fixed-temperature control and offline optimization by enabling dynamic, data-driven adjustment of process parameters</em>. By utilizing real-time process information from sensor readings within a batch, along with product quality data obtained through machine vision inspection after each batch, and employing a ‘mutual feedback’ sharing mechanism for multi-cavity process information, a ‘time-batch’ dual-scale real-time iterative learning and updating framework is established for the digital twin model. This approach enables closed-loop adaptive optimization of the mold filling state. Experimental results show that this method significantly outperforms traditional fixed temperature setting controls in terms of profile accuracy, offering an innovative solution for high-precision injection molding.</div></div>","PeriodicalId":55975,"journal":{"name":"Journal of Industrial Information Integration","volume":"49 ","pages":"Article 101009"},"PeriodicalIF":10.4,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145473276","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-01Epub Date: 2025-11-24DOI: 10.1016/j.jii.2025.101019
Robinson Garcés-Marín , José Arias-Pérez , Camilo Restrepo-Estrada
In a world facing pressing environmental challenges like climate change and resource scarcity, Artificial Intelligence (AI) is widely regarded as a powerful tool to enhance and support sustainability goals via Circular Economy Capability (CEC). The organizational capacity to leverage this technology, Artificial Intelligence Capability (AIC), is conceptualized through the lens of the Resource-Based Theory (RBT) as the capacity to effectively implement and utilize AI to generate strategic value. However, the direct relationship between AIC and CEC is not straightforward. The purpose of this research is to investigate this nuanced relationship by examining how socio-technical factors such as Data-Driven Insights (DDI)—actionable inferences derived from analytics over data—and AI Anxiety—stemming from employees' fear of job loss—shape the relationship between AIC and CEC. Using a moderated mediation model and Partial Least Squares Structural Equation Modeling (PLS-SEM), we analyzed data from firms with moderate to high technology maturity. While the study’s results are primarily based on context-specific evidence, which invites further investigation into generalizability to other settings, our findings suggest that the direct effect of AIC on CEC is not significant. Instead, DDI significantly mediate the relationship, confirming that AIC must be bundled with actionable insights to create value. Crucially, AI anxiety negatively moderates the effect of DDI on CEC. This means that while organizations may generate valuable insights, employee resistance and fear hinder their effective translation into sustainability practices. This study highlights the critical socio-technical barriers to AI adoption and their impact on achieving sustainability goals.
{"title":"The interplay of data-driven insights and AI anxiety in shaping the impact of AI capabilities on circular economy capability","authors":"Robinson Garcés-Marín , José Arias-Pérez , Camilo Restrepo-Estrada","doi":"10.1016/j.jii.2025.101019","DOIUrl":"10.1016/j.jii.2025.101019","url":null,"abstract":"<div><div>In a world facing pressing environmental challenges like climate change and resource scarcity, Artificial Intelligence (AI) is widely regarded as a powerful tool to enhance and support sustainability goals via Circular Economy Capability (CEC). The organizational capacity to leverage this technology, Artificial Intelligence Capability (AIC), is conceptualized through the lens of the Resource-Based Theory (RBT) as the capacity to effectively implement and utilize AI to generate strategic value. However, the direct relationship between AIC and CEC is not straightforward. The purpose of this research is to investigate this nuanced relationship by examining how socio-technical factors such as Data-Driven Insights (DDI)—actionable inferences derived from analytics over data—and AI Anxiety—stemming from employees' fear of job loss—shape the relationship between AIC and CEC. Using a moderated mediation model and Partial Least Squares Structural Equation Modeling (PLS-SEM), we analyzed data from firms with moderate to high technology maturity. While the study’s results are primarily based on context-specific evidence, which invites further investigation into generalizability to other settings, our findings suggest that the direct effect of AIC on CEC is not significant. Instead, DDI significantly mediate the relationship, confirming that AIC must be bundled with actionable insights to create value. Crucially, AI anxiety negatively moderates the effect of DDI on CEC. This means that while organizations may generate valuable insights, employee resistance and fear hinder their effective translation into sustainability practices. This study highlights the critical socio-technical barriers to AI adoption and their impact on achieving sustainability goals.</div></div>","PeriodicalId":55975,"journal":{"name":"Journal of Industrial Information Integration","volume":"49 ","pages":"Article 101019"},"PeriodicalIF":10.4,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145593631","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-01Epub Date: 2025-10-25DOI: 10.1016/j.jii.2025.100992
Pallavi Tiwari , S. Srinivasan
The proliferation of cloud computing and web-based services has led to a significant increase in the number and complexity of online web services. As a result, discovering appropriate services that meet user requirements has become a challenging task. Traditional web services discovery techniques often lack the efficiency and adaptability needed to handle user expectations in a dynamic environment. Additionally, it may struggle with limited scalability when dealing with large service sets. This results in suboptimal service selection, reduced user satisfaction, and increased latency. To address this challenge, a user requirement-oriented web services discovery approach based on Petri Nets and optimized Reinforcement (PN-ODRL) was proposed, aimed at improving the efficiency of agent-based services composition. Initially, service composition combines several atomic services related to specific tasks to fulfill user requirements. After that, a reinforcement learning-based Q-learning approach is utilized to choose the web services required by the user. Next, the Petri Net model is used to define RL actions by creating new finite action groups. A series of transitions within each action group identifies the best services, which are then recommended to the user. Then, Puffer Fish Optimization (PFO) is utilized to tune the learning rate and discount parameter present in the Q-learning algorithm, thereby enhancing the response time, cost, and reliability of the proposed approach. Experimental result for the proposed approach has an 85 % user satisfaction rate, 9ms of service discovery efficiency, 15.3Mbps of throughput, 97 % of availability, 24.6s of computational time, 18.3s of response time, 21.3s of processing time, 12.4s of mean residence time, 68.8s of execution time, and 93 % reliability. This approach reduced the response and processing time, enabling quicker service execution. Additionally, it could enhance user satisfaction with the system.
{"title":"Agent based web service composition using Q-learning algorithm with puffer fish optimization and petri net model","authors":"Pallavi Tiwari , S. Srinivasan","doi":"10.1016/j.jii.2025.100992","DOIUrl":"10.1016/j.jii.2025.100992","url":null,"abstract":"<div><div>The proliferation of cloud computing and web-based services has led to a significant increase in the number and complexity of online web services. As a result, discovering appropriate services that meet user requirements has become a challenging task. Traditional web services discovery techniques often lack the efficiency and adaptability needed to handle user expectations in a dynamic environment. Additionally, it may struggle with limited scalability when dealing with large service sets. This results in suboptimal service selection, reduced user satisfaction, and increased latency. To address this challenge, a user requirement-oriented web services discovery approach based on Petri Nets and optimized Reinforcement (PN-ODRL) was proposed, aimed at improving the efficiency of agent-based services composition. Initially, service composition combines several atomic services related to specific tasks to fulfill user requirements. After that, a reinforcement learning-based Q-learning approach is utilized to choose the web services required by the user. Next, the Petri Net model is used to define RL actions by creating new finite action groups. A series of transitions within each action group identifies the best services, which are then recommended to the user. Then, Puffer Fish Optimization (PFO) is utilized to tune the learning rate and discount parameter present in the Q-learning algorithm, thereby enhancing the response time, cost, and reliability of the proposed approach. Experimental result for the proposed approach has an 85 % user satisfaction rate, 9ms of service discovery efficiency, 15.3Mbps of throughput, 97 % of availability, 24.6s of computational time, 18.3s of response time, 21.3s of processing time, 12.4s of mean residence time, 68.8s of execution time, and 93 % reliability. This approach reduced the response and processing time, enabling quicker service execution. Additionally, it could enhance user satisfaction with the system.</div></div>","PeriodicalId":55975,"journal":{"name":"Journal of Industrial Information Integration","volume":"49 ","pages":"Article 100992"},"PeriodicalIF":10.4,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145383857","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-11-01Epub Date: 2025-09-02DOI: 10.1016/j.jii.2025.100943
Changchun Liu , Dunbing Tang , Haihua Zhu , Zequn Zhang , Liping Wang , Yi Zhang
Recently, embodied intelligence has emerged as a viable approach to achieving human-like perception, reasoning, decision-making, and execution capacities within human-robot collaborative (HRC) assembly contexts. Due to the lack of generalized enabling technologies and disconnections from physical control systems, embodied intelligence requires repetitive training of various functional models to operate in dynamic HRC scenarios, thereby struggling to adapt effectively to complex and evolving HRC environments. Hence, this study proposes a vision-language model (VLM)-enhanced embodied intelligence framework for digital twin (DT)-assisted human-robot collaborative assembly. Initially, the mapping between embodied agents and physical robots is established to achieve the encapsulation of embodied agents. Building upon the agent-based architecture, a VLM driven by both domain knowledge and real-time scenario data is constructed with sensory capabilities. Based on this, rapid recognition and response to dynamic HRC environments can be realized. Leveraging the strong generalization of VLMs, repetitive training of multiple perception models is circumvented. Furthermore, by utilizing the cognitive learning and intelligent reasoning capabilities of VLMs, an expert knowledge system for assembly processes is developed to provide task-oriented assistance and solution generation. To enhance the adaptability and generalization of complex HRC decision-making, VLMs support reinforcement learning through flexible configuration of HRC assembly state information processing, decision-action generation and guidance, and reward function design. In addition, a DT model of the HRC scenario is constructed to provide a simulation and deduction engine (i.e., embodied brain) for mitigating collision accidents. The decision results are then fed into the VLM as invocation parameters for corresponding sub-function code modules, generating complete collaborative robot action code to form the embodied neuron. Finally, compared with traditional decision methods (e.g., MA-A2C, DQN and GA) and VLM-enhanced MA-A2C, a series of comparative experiments conducted in a real-world HRC assembly scenario demonstrate that the proposed framework exhibits competitive advantages.
{"title":"Vision language model-enhanced embodied intelligence for digital twin-assisted human-robot collaborative assembly","authors":"Changchun Liu , Dunbing Tang , Haihua Zhu , Zequn Zhang , Liping Wang , Yi Zhang","doi":"10.1016/j.jii.2025.100943","DOIUrl":"10.1016/j.jii.2025.100943","url":null,"abstract":"<div><div>Recently, embodied intelligence has emerged as a viable approach to achieving human-like perception, reasoning, decision-making, and execution capacities within human-robot collaborative (HRC) assembly contexts. Due to the lack of generalized enabling technologies and disconnections from physical control systems, embodied intelligence requires repetitive training of various functional models to operate in dynamic HRC scenarios, thereby struggling to adapt effectively to complex and evolving HRC environments. Hence, this study proposes a vision-language model (VLM)-enhanced embodied intelligence framework for digital twin (DT)-assisted human-robot collaborative assembly. Initially, the mapping between embodied agents and physical robots is established to achieve the encapsulation of embodied agents. Building upon the agent-based architecture, a VLM driven by both domain knowledge and real-time scenario data is constructed with sensory capabilities. Based on this, rapid recognition and response to dynamic HRC environments can be realized. Leveraging the strong generalization of VLMs, repetitive training of multiple perception models is circumvented. Furthermore, by utilizing the cognitive learning and intelligent reasoning capabilities of VLMs, an expert knowledge system for assembly processes is developed to provide task-oriented assistance and solution generation. To enhance the adaptability and generalization of complex HRC decision-making, VLMs support reinforcement learning through flexible configuration of HRC assembly state information processing, decision-action generation and guidance, and reward function design. In addition, a DT model of the HRC scenario is constructed to provide a simulation and deduction engine (i.e., embodied brain) for mitigating collision accidents. The decision results are then fed into the VLM as invocation parameters for corresponding sub-function code modules, generating complete collaborative robot action code to form the embodied neuron. Finally, compared with traditional decision methods (e.g., MA-A2C, DQN and GA) and VLM-enhanced MA-A2C, a series of comparative experiments conducted in a real-world HRC assembly scenario demonstrate that the proposed framework exhibits competitive advantages.</div></div>","PeriodicalId":55975,"journal":{"name":"Journal of Industrial Information Integration","volume":"48 ","pages":"Article 100943"},"PeriodicalIF":10.4,"publicationDate":"2025-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145007638","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-11-01Epub Date: 2025-10-19DOI: 10.1016/j.jii.2025.100985
Wenyan Zhao, Yaguang Yuan, Cong Cheng, Wenheng Liu
The rapid advancement of e-commerce has driven unprecedented expansion in urban logistics networks, where their sustainability is constrained by multifaceted factors including strict time-bound service requirements, employee’s satisfaction, traffic congestion, and carbon emission regulations. Among these critical elements, employee’s satisfaction reflected by the workload balance not only influences task execution quality but also affects long-term operational sustainability for logistics enterprises, rendering its enhancement an urgent priority in contemporary urban logistics practices. This paper thus investigates a sustainable urban logistics vehicle routing problem mainly focusing on this perspective. Initially, a bi-objective mixed-integer programming model is formulated to simultaneously minimize total delivery cost and workload balance. Subsequently, a hybrid metaheuristic algorithm combining path relinking (PR) with multi-directional local search framework is developed. The adaptive large neighborhood search is adopted to facilitate the intensive local exploration, while PR techniques enhance global search capabilities through systematic solution space diversification. The algorithm's validity is rigorously verified through comparative analyses with state of art multi-objective optimization algorithms using adapted benchmark instances. Computational results demonstrate the algorithmic effectiveness and efficiency, accompanied by detailed analyses of approximate Pareto front and model’s sensitivity. These findings advance the field of urban delivery and provide practical insights for implementing efficient and sustainable urban logistic systems.
{"title":"Bi-objective sustainable urban logistics vehicle routing problem with workload balance","authors":"Wenyan Zhao, Yaguang Yuan, Cong Cheng, Wenheng Liu","doi":"10.1016/j.jii.2025.100985","DOIUrl":"10.1016/j.jii.2025.100985","url":null,"abstract":"<div><div>The rapid advancement of e-commerce has driven unprecedented expansion in urban logistics networks, where their sustainability is constrained by multifaceted factors including strict time-bound service requirements, employee’s satisfaction, traffic congestion, and carbon emission regulations. Among these critical elements, employee’s satisfaction reflected by the workload balance not only influences task execution quality but also affects long-term operational sustainability for logistics enterprises, rendering its enhancement an urgent priority in contemporary urban logistics practices. This paper thus investigates a sustainable urban logistics vehicle routing problem mainly focusing on this perspective. Initially, a bi-objective mixed-integer programming model is formulated to simultaneously minimize total delivery cost and workload balance. Subsequently, a hybrid metaheuristic algorithm combining path relinking (PR) with multi-directional local search framework is developed. The adaptive large neighborhood search is adopted to facilitate the intensive local exploration, while PR techniques enhance global search capabilities through systematic solution space diversification. The algorithm's validity is rigorously verified through comparative analyses with state of art multi-objective optimization algorithms using adapted benchmark instances. Computational results demonstrate the algorithmic effectiveness and efficiency, accompanied by detailed analyses of approximate Pareto front and model’s sensitivity. These findings advance the field of urban delivery and provide practical insights for implementing efficient and sustainable urban logistic systems.</div></div>","PeriodicalId":55975,"journal":{"name":"Journal of Industrial Information Integration","volume":"48 ","pages":"Article 100985"},"PeriodicalIF":10.4,"publicationDate":"2025-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145416550","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
During the revolution of Industry 4.0, digital twin technology is transforming industrial operations by creating digital models of physical assets, processes, and systems. This innovation enables real-time monitoring, predictive maintenance, and enhanced decision-making capabilities. However, as digital twins become integral to industrial environments, they also introduce new cybersecurity challenges, particularly in the form of distributed denial-of-service (DDoS) attacks, which can disrupt operations and compromise data integrity. This study investigates the resilience of digital twin-based industrial organizations in cyberattack scenarios, specifically focusing on the impacts of DDoS attacks on functional and financial performance. In this paper, a hybrid DDoS attack detection model is introduced, integrating multiple techniques for data preprocessing, feature selection, dimensionality reduction, and classification . To address the class imbalance issue,Synthetic Minority Over-sampling Technique (SMOTE) is applied during preprocessing. Feature selection is performed using filter-based methods, including Information Gain, Gain Ratio, ANOVA F-statistic, Pearson Correlation, and the technique for order preference by similarity to ideal solution (TOPSIS), a multi-criteria decision-making method. To enhance computational efficiency, principal component analysis (PCA) is used for dimensionality reduction, preserving critical information while reducing redundancy. For classification, an extreme learning machine (ELM) is optimized using the particle swarm optimization (PSO) algorithm, improving generalization, preventing overfitting, and ensuring faster convergence. The experiment is conducted using the publicly available CICDDoS2019 dataset in both standalone and cloud-based environments with configurations of vCPU-4, vCPU-8, and vCPU-16. Additionally, a 5-fold stratified cross-validation approach is employed to enhance the model’s generalization performance and ensure robustness across different data distributions. The experimental results indicate that the proposed model achieves a 99.97% detection accuracy and an AUC score of 0.99 in the cloud environment with vCPU-16 and 64GB RAM, outperforming traditional algorithms in DDoS detection. The experimental study finds that increased computational resources improve performance, indicating the model’s adaptability. As digital twins rely on seamless physical-virtual communication, DDoS attacks threaten synchronization, latency, and reliability. The proposed detection approach enhances resilience, minimizes downtime, and preserves process integrity, contributing to secure and robust digital twin architectures in Industry 4.0.
{"title":"An optimized learning approach for enhancing the security of digital twin-enabled industrial systems from distributed denial-of-service attacks","authors":"Debendra Muduli, Rahul Kumar Gupta, Samir Kumar Majhi, Binayak Ojha, Banshidhar Majhi","doi":"10.1016/j.jii.2025.100960","DOIUrl":"10.1016/j.jii.2025.100960","url":null,"abstract":"<div><div>During the revolution of Industry 4.0, digital twin technology is transforming industrial operations by creating digital models of physical assets, processes, and systems. This innovation enables real-time monitoring, predictive maintenance, and enhanced decision-making capabilities. However, as digital twins become integral to industrial environments, they also introduce new cybersecurity challenges, particularly in the form of distributed denial-of-service (DDoS) attacks, which can disrupt operations and compromise data integrity. This study investigates the resilience of digital twin-based industrial organizations in cyberattack scenarios, specifically focusing on the impacts of DDoS attacks on functional and financial performance. In this paper, a hybrid DDoS attack detection model is introduced, integrating multiple techniques for data preprocessing, feature selection, dimensionality reduction, and classification . To address the class imbalance issue,Synthetic Minority Over-sampling Technique (SMOTE) is applied during preprocessing. Feature selection is performed using filter-based methods, including Information Gain, Gain Ratio, ANOVA F-statistic, Pearson Correlation, and the technique for order preference by similarity to ideal solution (TOPSIS), a multi-criteria decision-making method. To enhance computational efficiency, principal component analysis (PCA) is used for dimensionality reduction, preserving critical information while reducing redundancy. For classification, an extreme learning machine (ELM) is optimized using the particle swarm optimization (PSO) algorithm, improving generalization, preventing overfitting, and ensuring faster convergence. The experiment is conducted using the publicly available CICDDoS2019 dataset in both standalone and cloud-based environments with configurations of vCPU-4, vCPU-8, and vCPU-16. Additionally, a 5-fold stratified cross-validation approach is employed to enhance the model’s generalization performance and ensure robustness across different data distributions. The experimental results indicate that the proposed model achieves a 99.97% detection accuracy and an AUC score of 0.99 in the cloud environment with vCPU-16 and 64GB RAM, outperforming traditional algorithms in DDoS detection. The experimental study finds that increased computational resources improve performance, indicating the model’s adaptability. As digital twins rely on seamless physical-virtual communication, DDoS attacks threaten synchronization, latency, and reliability. The proposed detection approach enhances resilience, minimizes downtime, and preserves process integrity, contributing to secure and robust digital twin architectures in Industry 4.0.</div></div>","PeriodicalId":55975,"journal":{"name":"Journal of Industrial Information Integration","volume":"48 ","pages":"Article 100960"},"PeriodicalIF":10.4,"publicationDate":"2025-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145158368","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-11-01Epub Date: 2025-11-06DOI: 10.1016/j.jii.2025.101004
Yao Shan, Jindong Zhao, Yongchao Song, Haojun Teng, Wenhan Hou, Zhaowei Liu
Modern information and communication technologies have propelled transformative modernization of Industrial Control Systems (ICSs) while exacerbating cybersecurity risks. Federated Learning (FL) offers a privacy-preserving framework for collaborative development of intrusion detection models among distributed participants. However, its effectiveness is significantly limited by inherent model divergence caused by non-independent and identically distributed (Non-IID) data characteristics. Moreover, direct implementation of FL in ICS environments faces critical challenges due to insufficient capabilities in network traffic feature representation and device concealment. To address these challenges, we propose CoperFed, a covert personalized FL framework that generates unique intrusion detection models for individual participants. First, we developed Gicsmeter, a multi-dimensional ICS traffic representation tool for all participants, to enhance model performance at the data level. Second, we designed a personalized update algorithm based on key model parameters to improve collaboration among similar participants. By integrating global knowledge during model aggregation, this algorithm equips the model with local and global scenario detection capabilities. Finally, we designed a covert federated communication scheme for ICS that can effectively conceal the federated training process within regular ICS traffic and reduce the exposure risk of FL participants. Experiments show that CoperFed outperforms baseline methods in intrusion detection and robustness and can effectively divert attackers’ attention from FL participants.
{"title":"CoperFed: A covert personalized federated learning framework for Industrial Control Systems intrusion detection","authors":"Yao Shan, Jindong Zhao, Yongchao Song, Haojun Teng, Wenhan Hou, Zhaowei Liu","doi":"10.1016/j.jii.2025.101004","DOIUrl":"10.1016/j.jii.2025.101004","url":null,"abstract":"<div><div>Modern information and communication technologies have propelled transformative modernization of Industrial Control Systems (ICSs) while exacerbating cybersecurity risks. Federated Learning (FL) offers a privacy-preserving framework for collaborative development of intrusion detection models among distributed participants. However, its effectiveness is significantly limited by inherent model divergence caused by non-independent and identically distributed (Non-IID) data characteristics. Moreover, direct implementation of FL in ICS environments faces critical challenges due to insufficient capabilities in network traffic feature representation and device concealment. To address these challenges, we propose CoperFed, a covert personalized FL framework that generates unique intrusion detection models for individual participants. First, we developed Gicsmeter, a multi-dimensional ICS traffic representation tool for all participants, to enhance model performance at the data level. Second, we designed a personalized update algorithm based on key model parameters to improve collaboration among similar participants. By integrating global knowledge during model aggregation, this algorithm equips the model with local and global scenario detection capabilities. Finally, we designed a covert federated communication scheme for ICS that can effectively conceal the federated training process within regular ICS traffic and reduce the exposure risk of FL participants. Experiments show that CoperFed outperforms baseline methods in intrusion detection and robustness and can effectively divert attackers’ attention from FL participants.</div></div>","PeriodicalId":55975,"journal":{"name":"Journal of Industrial Information Integration","volume":"48 ","pages":"Article 101004"},"PeriodicalIF":10.4,"publicationDate":"2025-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145461741","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","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