Pub Date : 2026-03-01Epub Date: 2025-12-06DOI: 10.1016/j.jii.2025.101036
Wei Liang , Zeqiang Zhang , Dan Ji , Haiye Chen , Yan Li , Qiyao Duan , Zongxing He
Once a disassembly line is constructed, it typically remains unchanged for an extended period. Consequently, the disassembly of end-of-life (EOL) electronic and electrical appliances for subsequent orders must be planned based on the existing production line configuration after completing the previous order of EOL electronic and electrical appliances. To address this challenge, this study proposed a type-II robotic partial disassembly line balancing (II-RPDLB) problem, leveraging advanced robot techniques. In addition, a mixed integer programming (MIP) model was developed according to the characteristics of the II-RPDLB problem. Furthermore, this study designed an MIP-based bi-stage genetic neighborhood search algorithm (bi-GNSA) for solving the II-RPDLB problem. The effectiveness of the proposed MIP-based bi-GNSA was verified by comparing its solutions with those obtained from the MIP model. Additionally, the improvement effect of the designed MIP-based bi-GNSA was verified with the original algorithm. The solution quality of the MIP-based bi-GNSA was validated with the NSGA-II and multi-objective enhanced differential evolution algorithm. Finally, a case study involving the disassembly of an EOL television was conducted to demonstrate the practical applicability of the bi-GNSA on an existing disassembly line.
{"title":"Type-II robotic partial disassembly line balancing problem and MIP-based bi-stage genetic neighborhood search algorithm","authors":"Wei Liang , Zeqiang Zhang , Dan Ji , Haiye Chen , Yan Li , Qiyao Duan , Zongxing He","doi":"10.1016/j.jii.2025.101036","DOIUrl":"10.1016/j.jii.2025.101036","url":null,"abstract":"<div><div>Once a disassembly line is constructed, it typically remains unchanged for an extended period. Consequently, the disassembly of end-of-life (EOL) electronic and electrical appliances for subsequent orders must be planned based on the existing production line configuration after completing the previous order of EOL electronic and electrical appliances. To address this challenge, this study proposed a type-II robotic partial disassembly line balancing (II-RPDLB) problem, leveraging advanced robot techniques. In addition, a mixed integer programming (MIP) model was developed according to the characteristics of the II-RPDLB problem. Furthermore, this study designed an MIP-based bi-stage genetic neighborhood search algorithm (bi-GNSA) for solving the II-RPDLB problem. The effectiveness of the proposed MIP-based bi-GNSA was verified by comparing its solutions with those obtained from the MIP model. Additionally, the improvement effect of the designed MIP-based bi-GNSA was verified with the original algorithm. The solution quality of the MIP-based bi-GNSA was validated with the NSGA-II and multi-objective enhanced differential evolution algorithm. Finally, a case study involving the disassembly of an EOL television was conducted to demonstrate the practical applicability of the bi-GNSA on an existing disassembly line.</div></div>","PeriodicalId":55975,"journal":{"name":"Journal of Industrial Information Integration","volume":"50 ","pages":"Article 101036"},"PeriodicalIF":10.4,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145689909","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 : 2026-03-01Epub Date: 2025-12-01DOI: 10.1016/j.jii.2025.101031
Marcello Braglia , Mohamed Afy-Shararah , Francesco Di Paco , Roberto Gabbrielli , Leonardo Marrazzini
Water management is becoming an increasingly critical challenge for manufacturing industries due to growing environmental concerns, stricter regulatory requirements, and rising pressure from clients demanding more sustainable practices. Efficient and transparent use of water resources is no longer optional but a strategic necessity across industrial sectors. In this paper a new Lean performance indicator for evaluating water usage in industrial processes is presented. The proposed indicator, named Overall Water Effectiveness, aims to systematically assess industrial water performance by quantifying the gap between actual and ideal performance. It builds on the logic of Overall Equipment Effectiveness to identify water-related losses and support informed decision-making for continuous improvement while introducing a comprehensive industrial loss structure specifically designed for water use and consumption. Jointly, two key additional indicators are introduced: one measures how effectively the production process consumes input water, while the other evaluates the dependency on external water sources, taking into account the contributions of recycled and returned water. By translating high-level sustainability goals into actionable operational metrics, this new set of indicators enables the integration of water management into daily industrial operations through a practical, easy-to-use tool. The approach is applied in a major textile manufacturing company, demonstrating its practical utility in evaluating water use and consumption, identifying loss patterns, and leading the identification of improvement actions.
{"title":"Overall water effectiveness: A new lean indicator for digital evaluation of water efficiency in industrial processes","authors":"Marcello Braglia , Mohamed Afy-Shararah , Francesco Di Paco , Roberto Gabbrielli , Leonardo Marrazzini","doi":"10.1016/j.jii.2025.101031","DOIUrl":"10.1016/j.jii.2025.101031","url":null,"abstract":"<div><div>Water management is becoming an increasingly critical challenge for manufacturing industries due to growing environmental concerns, stricter regulatory requirements, and rising pressure from clients demanding more sustainable practices. Efficient and transparent use of water resources is no longer optional but a strategic necessity across industrial sectors. In this paper a new Lean performance indicator for evaluating water usage in industrial processes is presented. The proposed indicator, named Overall Water Effectiveness, aims to systematically assess industrial water performance by quantifying the gap between actual and ideal performance. It builds on the logic of Overall Equipment Effectiveness to identify water-related losses and support informed decision-making for continuous improvement while introducing a comprehensive industrial loss structure specifically designed for water use and consumption. Jointly, two key additional indicators are introduced: one measures how effectively the production process consumes input water, while the other evaluates the dependency on external water sources, taking into account the contributions of recycled and returned water. By translating high-level sustainability goals into actionable operational metrics, this new set of indicators enables the integration of water management into daily industrial operations through a practical, easy-to-use tool. The approach is applied in a major textile manufacturing company, demonstrating its practical utility in evaluating water use and consumption, identifying loss patterns, and leading the identification of improvement actions.</div></div>","PeriodicalId":55975,"journal":{"name":"Journal of Industrial Information Integration","volume":"50 ","pages":"Article 101031"},"PeriodicalIF":10.4,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145658118","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}
Rapid urbanization and industrial growth have intensified air pollution in metropolitan regions, making accurate and energy-efficient Air Quality Index (AQI) prediction critical for sustainable smart city management. Existing centralized and conventional federated learning approaches suffer from high communication overhead, excessive energy consumption, and privacy risks, limiting their applicability in distributed urban sensing environments. This paper proposes GreenEdge AI, a green federated learning framework integrating a green-aware custom LSTM (GA-CLSTM) model with energy-aware training, adaptive aggregation, and a hybrid loss function for decentralized AQI forecasting. The framework enables edge-level learning across heterogeneous IoT-based air quality and meteorological sensors while preserving data privacy and minimizing cloud dependency. Sustainability is explicitly incorporated through green metrics, including energy consumption, Energy–Delay Product (EDP), Energy Efficiency Ratio (EER), and Power-to-Performance Ratio (PPR), which guide both model optimization and federated aggregation. Experimental results on real-world hourly AQI data from five major metropolitan cities demonstrate that GreenEdge AI achieves up to 60% improvement in prediction accuracy and approximately 37% reduction in energy consumption compared to conventional baseline models, while significantly reducing peak power usage and communication overhead compared to centralized and conventional federated baselines. These findings underscore the practical value of GreenEdge AI for municipalities and environmental agencies, motivating future research on scalable, energy-aware federated intelligence for smart city applications.
{"title":"GreenEdge AI: Sustainable federated learning for smart city air quality prediction","authors":"Sweta Dey , Rishi Raina , Sudeepta Mishra , Abhinandan S. Prasad , Ramesh Dharavath","doi":"10.1016/j.jii.2026.101081","DOIUrl":"10.1016/j.jii.2026.101081","url":null,"abstract":"<div><div>Rapid urbanization and industrial growth have intensified air pollution in metropolitan regions, making accurate and energy-efficient Air Quality Index (AQI) prediction critical for sustainable smart city management. Existing centralized and conventional federated learning approaches suffer from high communication overhead, excessive energy consumption, and privacy risks, limiting their applicability in distributed urban sensing environments. This paper proposes <em>GreenEdge AI</em>, a green federated learning framework integrating a green-aware custom LSTM (GA-CLSTM) model with energy-aware training, adaptive aggregation, and a hybrid loss function for decentralized AQI forecasting. The framework enables edge-level learning across heterogeneous IoT-based air quality and meteorological sensors while preserving data privacy and minimizing cloud dependency. Sustainability is explicitly incorporated through green metrics, including energy consumption, Energy–Delay Product (EDP), Energy Efficiency Ratio (EER), and Power-to-Performance Ratio (PPR), which guide both model optimization and federated aggregation. Experimental results on real-world hourly AQI data from five major metropolitan cities demonstrate that <em>GreenEdge AI</em> achieves up to 60% improvement in prediction accuracy and approximately 37% reduction in energy consumption compared to conventional baseline models, while significantly reducing peak power usage and communication overhead compared to centralized and conventional federated baselines. These findings underscore the practical value of <em>GreenEdge AI</em> for municipalities and environmental agencies, motivating future research on scalable, energy-aware federated intelligence for smart city applications.</div></div>","PeriodicalId":55975,"journal":{"name":"Journal of Industrial Information Integration","volume":"50 ","pages":"Article 101081"},"PeriodicalIF":10.4,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146072129","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 : 2026-03-01Epub Date: 2025-12-15DOI: 10.1016/j.jii.2025.101041
Xiaojian Wen , Wenchao Bian , Shimin Liu , Jie Wen , Jinsong Bao , Dan Zhang
With the deepening of digital transformation in the manufacturing industry, digital twin technology has become a key enabler for enhancing the flexibility and intelligent reconfiguration of manufacturing systems. However, the current construction of digital-twin-based production lines still relies heavily on manual expertise and lacks a systematic approach capable of automatically selecting and configuring appropriate components under task constraints. To address this issue, this paper proposes a multi-scale digital twin model reconstruction method based on compatibility and exclusivity mechanisms. The proposed approach establishes a component selection framework that integrates functional, spatial, and associative semantics, and further incorporates the temporal dimension to capture the dynamic evolution of component compatibility. This enables task-driven dynamic reconfiguration and adaptive optimization of spatial layouts. Experimental results demonstrate that the proposed methyod significantly improves component selection accuracy and the level of automation in configuration processes, while ensuring functional compatibility and spatial coordination. The study provides both theoretical support and an engineering-oriented solution for multi-scale intelligent planning and decision optimization in complex manufacturing systems.
{"title":"A multi-scale digital twin model reconstruction method based on compatibility and exclusivity mechanisms","authors":"Xiaojian Wen , Wenchao Bian , Shimin Liu , Jie Wen , Jinsong Bao , Dan Zhang","doi":"10.1016/j.jii.2025.101041","DOIUrl":"10.1016/j.jii.2025.101041","url":null,"abstract":"<div><div>With the deepening of digital transformation in the manufacturing industry, digital twin technology has become a key enabler for enhancing the flexibility and intelligent reconfiguration of manufacturing systems. However, the current construction of digital-twin-based production lines still relies heavily on manual expertise and lacks a systematic approach capable of automatically selecting and configuring appropriate components under task constraints. To address this issue, this paper proposes a multi-scale digital twin model reconstruction method based on compatibility and exclusivity mechanisms. The proposed approach establishes a component selection framework that integrates functional, spatial, and associative semantics, and further incorporates the temporal dimension to capture the dynamic evolution of component compatibility. This enables task-driven dynamic reconfiguration and adaptive optimization of spatial layouts. Experimental results demonstrate that the proposed methyod significantly improves component selection accuracy and the level of automation in configuration processes, while ensuring functional compatibility and spatial coordination. The study provides both theoretical support and an engineering-oriented solution for multi-scale intelligent planning and decision optimization in complex manufacturing systems.</div></div>","PeriodicalId":55975,"journal":{"name":"Journal of Industrial Information Integration","volume":"50 ","pages":"Article 101041"},"PeriodicalIF":10.4,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145785014","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 : 2026-03-01Epub Date: 2025-12-15DOI: 10.1016/j.jii.2025.101039
Lunyong Li , Auwal Haruna , Wanming Ying , Khandaker Noman , Yongbo Li
Aviation equipment fault diagnosis faces significant challenges due to the complexity of systems, the scarcity of high-quality labeled data, and the critical need for interpretability in maintenance decisions. While Knowledge Graph (KG) offers a promising solution for structured knowledge management, their broader application is impeded by limitations in knowledge extraction methods from unstructured texts and inefficient retrieval mechanisms. To address these gaps, this study proposes an innovative KG method that integrates an enhanced joint extraction model with Large Language Model (LLM) for aviation equipment fault diagnosis. To overcome the bottlenecks of high complexity and low efficiency in constructing KGs for aviation fault diagnosis, this study proposes an Aviation Equipment Maintenance Cascade Binary Tagging (AemCASREL) model optimized with Bidirectional Encoder Representation from Transformers (BERT) fine-tuning and attention enhancement. This model extracts fault entities and relations from diverse unstructured sources, such as aircraft maintenance manuals and equipment logs, to build a KG database in Neo4j. Additionally, a method integrating an LLM with the KG database is introduced, enhancing the model’s generation ability, enabling intelligent question-answering, and offering robust domain knowledge support for fault diagnosis. The experimental evaluation, using both self-built and public datasets, demonstrates the improved model's superiority over the baseline. On the self-built dataset, the F1 score rises from 0.907 to 0.968, and on the public dataset, it increases from 0.907 to 0.980. The integration of LLM and KG enhances the accuracy and intelligence of the question-answering system for aircraft fault diagnosis and maintenance, making it more adaptable to complex faults. This study provides a feasible knowledge-driven paradigm for multi-source information fusion and integration in complex industrial scenarios.
{"title":"Knowledge graph-driven fault diagnosis for aviation equipment: Integrating improved joint extraction with large language model","authors":"Lunyong Li , Auwal Haruna , Wanming Ying , Khandaker Noman , Yongbo Li","doi":"10.1016/j.jii.2025.101039","DOIUrl":"10.1016/j.jii.2025.101039","url":null,"abstract":"<div><div>Aviation equipment fault diagnosis faces significant challenges due to the complexity of systems, the scarcity of high-quality labeled data, and the critical need for interpretability in maintenance decisions. While Knowledge Graph (KG) offers a promising solution for structured knowledge management, their broader application is impeded by limitations in knowledge extraction methods from unstructured texts and inefficient retrieval mechanisms. To address these gaps, this study proposes an innovative KG method that integrates an enhanced joint extraction model with Large Language Model (LLM) for aviation equipment fault diagnosis. To overcome the bottlenecks of high complexity and low efficiency in constructing KGs for aviation fault diagnosis, this study proposes an Aviation Equipment Maintenance Cascade Binary Tagging (AemCASREL) model optimized with Bidirectional Encoder Representation from Transformers (BERT) fine-tuning and attention enhancement. This model extracts fault entities and relations from diverse unstructured sources, such as aircraft maintenance manuals and equipment logs, to build a KG database in Neo4j. Additionally, a method integrating an LLM with the KG database is introduced, enhancing the model’s generation ability, enabling intelligent question-answering, and offering robust domain knowledge support for fault diagnosis. The experimental evaluation, using both self-built and public datasets, demonstrates the improved model's superiority over the baseline. On the self-built dataset, the F1 score rises from 0.907 to 0.968, and on the public dataset, it increases from 0.907 to 0.980. The integration of LLM and KG enhances the accuracy and intelligence of the question-answering system for aircraft fault diagnosis and maintenance, making it more adaptable to complex faults. This study provides a feasible knowledge-driven paradigm for multi-source information fusion and integration in complex industrial scenarios.</div></div>","PeriodicalId":55975,"journal":{"name":"Journal of Industrial Information Integration","volume":"50 ","pages":"Article 101039"},"PeriodicalIF":10.4,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145785016","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 : 2026-03-01Epub Date: 2025-12-27DOI: 10.1016/j.jii.2025.101050
Jing Long , Jiahao Zeng , Zhifei Yan , Min Shi , Kun Xie , Meng Shen , Naixue Xiong
In smart manufacturing systems, interconnected systems composed of equipment nodes such as intelligent machine tools and sensors can be abstracted as attributed networks. However, such networks are vulnerable to security risks like cyber-attacks and equipment failures, which directly threaten the stable operation of smart manufacturing systems. In the unsupervised setting, existing anomaly detection models intrinsically lean towards fitting the overwhelming majority of normal patterns during training. However, they cannot escape being influenced by anomalous characteristics, which degrades the detection of anomaly patterns in smart manufacturing environments. To address this challenge, this paper proposes a novel anomaly detection method called ARENA for attributed networks in smart manufacturing, which adopts two-stage dynamic reconstruction bias learning. Firstly, a graph autoencoder uncovers latent data patterns in smart manufacturing scenarios by minimizing reconstruction error. Then, the dynamic reconstruction biased learning module adjusts the training process in two stages to filter out pseudo-normal nodes and pseudo-anomalous nodes, enabling the model to adaptively fine-tune, mitigating the impact of anomalous data during training. Finally, the classification module further amplifies the anomaly score, making abnormal patterns more pronounced and easier to detect. The overall anomaly score is calculated by combining the results of the graph reconstruction and classification modules. Experimental results show that the ARENA method significantly improves performance, with an increase of 3.73% in AUC and 21.1% in AUPRC, including the success of the case study, providing strong support for the intelligent operation and maintenance of equipment in industrial manufacturing systems.
{"title":"Two-stage dynamic reconstruction biased learning for anomaly detection in attributed networks of smart manufacturing","authors":"Jing Long , Jiahao Zeng , Zhifei Yan , Min Shi , Kun Xie , Meng Shen , Naixue Xiong","doi":"10.1016/j.jii.2025.101050","DOIUrl":"10.1016/j.jii.2025.101050","url":null,"abstract":"<div><div>In smart manufacturing systems, interconnected systems composed of equipment nodes such as intelligent machine tools and sensors can be abstracted as attributed networks. However, such networks are vulnerable to security risks like cyber-attacks and equipment failures, which directly threaten the stable operation of smart manufacturing systems. In the unsupervised setting, existing anomaly detection models intrinsically lean towards fitting the overwhelming majority of normal patterns during training. However, they cannot escape being influenced by anomalous characteristics, which degrades the detection of anomaly patterns in smart manufacturing environments. To address this challenge, this paper proposes a novel anomaly detection method called ARENA for attributed networks in smart manufacturing, which adopts two-stage dynamic reconstruction bias learning. Firstly, a graph autoencoder uncovers latent data patterns in smart manufacturing scenarios by minimizing reconstruction error. Then, the dynamic reconstruction biased learning module adjusts the training process in two stages to filter out pseudo-normal nodes and pseudo-anomalous nodes, enabling the model to adaptively fine-tune, mitigating the impact of anomalous data during training. Finally, the classification module further amplifies the anomaly score, making abnormal patterns more pronounced and easier to detect. The overall anomaly score is calculated by combining the results of the graph reconstruction and classification modules. Experimental results show that the ARENA method significantly improves performance, with an increase of 3.73% in AUC and 21.1% in AUPRC, including the success of the case study, providing strong support for the intelligent operation and maintenance of equipment in industrial manufacturing systems.</div></div>","PeriodicalId":55975,"journal":{"name":"Journal of Industrial Information Integration","volume":"50 ","pages":"Article 101050"},"PeriodicalIF":10.4,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145845506","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 : 2026-03-01Epub Date: 2026-01-09DOI: 10.1016/j.jii.2026.101061
Shupeng Yu , Xiang Li , Yaguo Lei , Bin Yang , Naipeng Li , Ke Feng
Large language models (LLMs) have been showing growing potential in the field of intelligent operation and maintenance, due to their strong capabilities in understanding and generating knowledge across data in multiple modalities. However, in operation and maintenance, time-series signals are among the most critical monitoring data, and their unique formats and high dimensionality pose significant challenges for direct application of LLMs. To address this limitation, we propose a novel large multimodal model for fault diagnosis (LMM-FD), which is a key problem in operation and maintenance. The proposed large multimodal framework effectively aligns time-series vibration signals with textual fault diagnosis knowledge, enabling interpretable and generalized fault diagnosis. The framework includes signal preprocessing, cross-modal alignment through a knowledge graph and graph neural networks, and automated generation of textual diagnostic reports. Extensive experiments on machinery condition monitoring datasets demonstrate that LMM-FD consistently outperforms existing baselines by leveraging multimodal data and constructed triplet-based knowledge graph. The proposed model obtains fairly high accuracy on multiple fault diagnosis scenarios, while achieving strong zero-shot generalization capabilities to unseen compound faults. Furthermore, by bridging numerical sensor data with textual knowledge, LMM-FD provides interpretable fault descriptions, highlighting its potential for practical industrial applications.
{"title":"Multimodal data-enabled large model for machine fault diagnosis towards intelligent operation and maintenance","authors":"Shupeng Yu , Xiang Li , Yaguo Lei , Bin Yang , Naipeng Li , Ke Feng","doi":"10.1016/j.jii.2026.101061","DOIUrl":"10.1016/j.jii.2026.101061","url":null,"abstract":"<div><div>Large language models (LLMs) have been showing growing potential in the field of intelligent operation and maintenance, due to their strong capabilities in understanding and generating knowledge across data in multiple modalities. However, in operation and maintenance, time-series signals are among the most critical monitoring data, and their unique formats and high dimensionality pose significant challenges for direct application of LLMs. To address this limitation, we propose a novel large multimodal model for fault diagnosis (LMM-FD), which is a key problem in operation and maintenance. The proposed large multimodal framework effectively aligns time-series vibration signals with textual fault diagnosis knowledge, enabling interpretable and generalized fault diagnosis. The framework includes signal preprocessing, cross-modal alignment through a knowledge graph and graph neural networks, and automated generation of textual diagnostic reports. Extensive experiments on machinery condition monitoring datasets demonstrate that LMM-FD consistently outperforms existing baselines by leveraging multimodal data and constructed triplet-based knowledge graph. The proposed model obtains fairly high accuracy on multiple fault diagnosis scenarios, while achieving strong zero-shot generalization capabilities to unseen compound faults. Furthermore, by bridging numerical sensor data with textual knowledge, LMM-FD provides interpretable fault descriptions, highlighting its potential for practical industrial applications.</div></div>","PeriodicalId":55975,"journal":{"name":"Journal of Industrial Information Integration","volume":"50 ","pages":"Article 101061"},"PeriodicalIF":10.4,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145957254","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 : 2026-03-01Epub Date: 2026-01-15DOI: 10.1016/j.jii.2026.101070
Chongxin Wang , Xiaojun Liu , Feixiang Wang , Fengyi Feng , Lv Feng
This paper presents an innovative approach to optimizing connection models in complex digital twin systems. Traditional digital twin systems are often hindered by inefficient connection models, resulting in excessive thread and memory consumption and conflicts during functional expansion. To address these challenges, we propose a Virtual Sensor-based dual strategy combining merging and assembly techniques within the framework of the five-dimensional digital twin model. The merging strategy groups and merges similar models to eliminate redundancies, reducing model complexity and resource consumption. The assembly strategy integrates multiple sub-connection models into a more complex, scalable model. This ensures dynamic adjustment and synchronization of information across various system dimensions. A case study in a packaging production line demonstrates an over 40% reduction in connection models. Due to the deployed stateless singleton architecture, this structural simplification directly translates into a proportional decrease in resource consumption, specifically reducing active thread occupation by approximately 40% and substantially lowering memory usage. These results confirm the proposed method's effectiveness in enhancing scalability and resource efficiency, highlighting its significant industrial applicability.
{"title":"Optimization of connection models in digital twin systems: Efficient merging and assembly strategy for enhanced scalability and resource optimization","authors":"Chongxin Wang , Xiaojun Liu , Feixiang Wang , Fengyi Feng , Lv Feng","doi":"10.1016/j.jii.2026.101070","DOIUrl":"10.1016/j.jii.2026.101070","url":null,"abstract":"<div><div>This paper presents an innovative approach to optimizing connection models in complex digital twin systems. Traditional digital twin systems are often hindered by inefficient connection models, resulting in excessive thread and memory consumption and conflicts during functional expansion. To address these challenges, we propose a Virtual Sensor-based dual strategy combining merging and assembly techniques within the framework of the five-dimensional digital twin model. The merging strategy groups and merges similar models to eliminate redundancies, reducing model complexity and resource consumption. The assembly strategy integrates multiple sub-connection models into a more complex, scalable model. This ensures dynamic adjustment and synchronization of information across various system dimensions. A case study in a packaging production line demonstrates an over 40% reduction in connection models. Due to the deployed stateless singleton architecture, this structural simplification directly translates into a proportional decrease in resource consumption, specifically reducing active thread occupation by approximately 40% and substantially lowering memory usage. These results confirm the proposed method's effectiveness in enhancing scalability and resource efficiency, highlighting its significant industrial applicability.</div></div>","PeriodicalId":55975,"journal":{"name":"Journal of Industrial Information Integration","volume":"50 ","pages":"Article 101070"},"PeriodicalIF":10.4,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145995696","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 : 2026-03-01Epub Date: 2026-01-17DOI: 10.1016/j.jii.2026.101073
Xin Zhao , Wenjie Liu , Jianhua Shi , Yangyu Zhao , Zikang Li
Long-term operation of mining diesel engines with high power density within a complex working environment of open-pit mines causes them to suffer from compound faults and difficult diagnosis. Therefore, a compound fault diagnosis method that combines a Conditional Deep Convolutional Generative Adversarial Network (CDCGAN) and Multistage Transfer Learning (MTL) is proposed in this paper. This method overcomes different issues arising in compound fault detection, such as sample scarcity, insufficient single signal characterization, and low distinguishability of one-dimensional vibration signal features. The Continuous Wavelet Transform (CWT) and CDCGAN are introduced to process the one-dimensional raw data. An improved Transfer Learning (TL) algorithm based on an MTL strategy is also proposed by incorporating pretraining, fine-tuning, and feature fusion techniques. A ResNetCBAM model integrating the Residual Neural Network (ResNet) with the Convolutional Block Attention Module (CBAM) is trained based on the algorithm. Validation experiments are performed on real diesel engine fault data to evaluate the method’s performance. It is shown that the proposed method’s accuracy improves by 13.75%, 8.75%, 6.37%, and 4.58%, compared with four baseline methods including a one-dimensional convolutional neural network (1D-CNN) with raw one-dimensional vibration signals, a two-dimensional convolutional neural network (2D-CNN) with time-frequency images obtained via the CWT, ResNetCBAM with CDCGAN-augmented data, and ResNetCBAM with conventional TL, respectively. The proposed method achieves 100% diagnostic accuracy on the test data, thus establishing a reliable theoretical basis for the intelligent compound fault diagnosis in diesel engines.
{"title":"Compound fault diagnosis of diesel engines by combining CDCGAN and multistage transfer learning","authors":"Xin Zhao , Wenjie Liu , Jianhua Shi , Yangyu Zhao , Zikang Li","doi":"10.1016/j.jii.2026.101073","DOIUrl":"10.1016/j.jii.2026.101073","url":null,"abstract":"<div><div>Long-term operation of mining diesel engines with high power density within a complex working environment of open-pit mines causes them to suffer from compound faults and difficult diagnosis. Therefore, a compound fault diagnosis method that combines a Conditional Deep Convolutional Generative Adversarial Network (CDCGAN) and Multistage Transfer Learning (MTL) is proposed in this paper. This method overcomes different issues arising in compound fault detection, such as sample scarcity, insufficient single signal characterization, and low distinguishability of one-dimensional vibration signal features. The Continuous Wavelet Transform (CWT) and CDCGAN are introduced to process the one-dimensional raw data. An improved Transfer Learning (TL) algorithm based on an MTL strategy is also proposed by incorporating pretraining, fine-tuning, and feature fusion techniques. A ResNetCBAM model integrating the Residual Neural Network (ResNet) with the Convolutional Block Attention Module (CBAM) is trained based on the algorithm. Validation experiments are performed on real diesel engine fault data to evaluate the method’s performance. It is shown that the proposed method’s accuracy improves by 13.75%, 8.75%, 6.37%, and 4.58%, compared with four baseline methods including a one-dimensional convolutional neural network (1D-CNN) with raw one-dimensional vibration signals, a two-dimensional convolutional neural network (2D-CNN) with time-frequency images obtained via the CWT, ResNetCBAM with CDCGAN-augmented data, and ResNetCBAM with conventional TL, respectively. The proposed method achieves 100% diagnostic accuracy on the test data, thus establishing a reliable theoretical basis for the intelligent compound fault diagnosis in diesel engines.</div></div>","PeriodicalId":55975,"journal":{"name":"Journal of Industrial Information Integration","volume":"50 ","pages":"Article 101073"},"PeriodicalIF":10.4,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145995692","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}
{"title":"AI for information integration and processing in digital twins (AI4IIP-DT)","authors":"Hervé Panetto , Michele Dassisti , Qing Li , Yannick Naudet","doi":"10.1016/j.jii.2026.101066","DOIUrl":"10.1016/j.jii.2026.101066","url":null,"abstract":"","PeriodicalId":55975,"journal":{"name":"Journal of Industrial Information Integration","volume":"50 ","pages":"Article 101066"},"PeriodicalIF":10.4,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145957252","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