Pub Date : 2025-12-24DOI: 10.1016/j.aei.2025.104258
Hao Lyu , Yanyong Guo , Pan Liu , Nan Zheng , Ting Wang , Quansheng Yue
Mitigating traffic oscillations in mixed flows of connected automated vehicles (CAVs) and human-driven vehicles (HDVs) is critical for enhancing traffic stability. A key challenge lies in modeling the nonlinear, heterogeneous behaviors of HDVs within computationally tractable predictive control frameworks. This study proposes an adaptive deep Koopman predictive control framework (AdapKoopPC) to address this issue. The framework features a novel deep Koopman network, AdapKoopnet, which represents complex HDV car-following dynamics as a linear system in a high-dimensional space by adaptively learning from naturalistic data. This learned linear representation is then embedded into a Model Predictive Control (MPC) scheme, enabling real-time, scalable, and optimal control of CAVs. We validate our framework using the HighD dataset and extensive numerical simulations. Results demonstrate that AdapKoopnet achieves superior trajectory prediction accuracy over baseline models. Furthermore, the complete AdapKoopPC controller significantly dampens traffic oscillations with lower computational cost, exhibiting strong performance even at low CAV penetration rates. The proposed framework offers a scalable and data-driven solution for enhancing stability in realistic mixed traffic environments. The code is made publicly available.1
{"title":"Mitigating traffic oscillations in mixed traffic flow with scalable deep Koopman predictive control","authors":"Hao Lyu , Yanyong Guo , Pan Liu , Nan Zheng , Ting Wang , Quansheng Yue","doi":"10.1016/j.aei.2025.104258","DOIUrl":"10.1016/j.aei.2025.104258","url":null,"abstract":"<div><div>Mitigating traffic oscillations in mixed flows of connected automated vehicles (CAVs) and human-driven vehicles (HDVs) is critical for enhancing traffic stability. A key challenge lies in modeling the nonlinear, heterogeneous behaviors of HDVs within computationally tractable predictive control frameworks. This study proposes an adaptive deep Koopman predictive control framework (AdapKoopPC) to address this issue. The framework features a novel deep Koopman network, AdapKoopnet, which represents complex HDV car-following dynamics as a linear system in a high-dimensional space by adaptively learning from naturalistic data. This learned linear representation is then embedded into a Model Predictive Control (MPC) scheme, enabling real-time, scalable, and optimal control of CAVs. We validate our framework using the HighD dataset and extensive numerical simulations. Results demonstrate that AdapKoopnet achieves superior trajectory prediction accuracy over baseline models. Furthermore, the complete AdapKoopPC controller significantly dampens traffic oscillations with lower computational cost, exhibiting strong performance even at low CAV penetration rates. The proposed framework offers a scalable and data-driven solution for enhancing stability in realistic mixed traffic environments. The code is made publicly available.<span><span><sup>1</sup></span></span></div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"71 ","pages":"Article 104258"},"PeriodicalIF":9.9,"publicationDate":"2025-12-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145842078","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 : 2025-12-24DOI: 10.1016/j.aei.2025.104269
Wei Zhang , Kaicheng Yu , Lijie Su , Yifeng Yao , Lihua Lu , Swee Leong Sing
The efficient design of cancellous bone tissue engineering scaffolds that closely replicate both the histological and mechanical characteristics of native cancellous bone remains a significant challenge in the field. The vast design space of porous scaffolds gives rise to a highly complex, nonlinear relationship between geometric morphology, mechanical properties, and osteogenic performance. Currently, the design and in vitro fabrication of scaffolds for cancellous bone defect repair largely depend on expert-driven forward design methods. These approaches are time-consuming, costly, and often lack reproducibility and controllability, limiting their suitability for clinical applications. To overcome these limitations, this study introduces an innovative inverse design framework – TriTopo-LGDM – which combines topological optimization priors with a latent graph diffusion generative model. Built on a triply aligned dataset encompassing structure, physics, and optimization, the framework establishes a scaffold generation pipeline specifically tailored for cancellous bone defect reconstruction. It enables the efficient generation and accurate modeling of multi-scale functional scaffold structures. Experimental evaluations confirm that TriTopo-LGDM establishes a robust bidirectional mapping between topological parameters and target mechanical properties, significantly reducing design time and cost while improving structural consistency and 3D printability. Mechanical testing and finite element simulations further validate the strong mechanical and morphological resemblance of the generated scaffolds to natural cancellous bone. This work presents a generalizable and efficient strategy for the rapid, patient-specific design of implants that promote cancellous bone regeneration.
{"title":"TriTopo-LGDM: A reverse design method for trabecular bone scaffolds integrating topology optimization and latent graph diffusion models","authors":"Wei Zhang , Kaicheng Yu , Lijie Su , Yifeng Yao , Lihua Lu , Swee Leong Sing","doi":"10.1016/j.aei.2025.104269","DOIUrl":"10.1016/j.aei.2025.104269","url":null,"abstract":"<div><div>The efficient design of cancellous bone tissue engineering scaffolds that closely replicate both the histological and mechanical characteristics of native cancellous bone remains a significant challenge in the field. The vast design space of porous scaffolds gives rise to a highly complex, nonlinear relationship between geometric morphology, mechanical properties, and osteogenic performance. Currently, the design and in vitro fabrication of scaffolds for cancellous bone defect repair largely depend on expert-driven forward design methods. These approaches are time-consuming, costly, and often lack reproducibility and controllability, limiting their suitability for clinical applications. To overcome these limitations, this study introduces an innovative inverse design framework – TriTopo-LGDM – which combines topological optimization priors with a latent graph diffusion generative model. Built on a triply aligned dataset encompassing structure, physics, and optimization, the framework establishes a scaffold generation pipeline specifically tailored for cancellous bone defect reconstruction. It enables the efficient generation and accurate modeling of multi-scale functional scaffold structures. Experimental evaluations confirm that TriTopo-LGDM establishes a robust bidirectional mapping between topological parameters and target mechanical properties, significantly reducing design time and cost while improving structural consistency and 3D printability. Mechanical testing and finite element simulations further validate the strong mechanical and morphological resemblance of the generated scaffolds to natural cancellous bone. This work presents a generalizable and efficient strategy for the rapid, patient-specific design of implants that promote cancellous bone regeneration.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"71 ","pages":"Article 104269"},"PeriodicalIF":9.9,"publicationDate":"2025-12-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145842079","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 : 2025-12-24DOI: 10.1016/j.aei.2025.104281
Rui Huang , Fan Zhang , Shilong Yu , Bo Huang
With the widespread use of computer-aided technologies represented by CAD/CAM/CAPP in product machining, a large amount of process data is continuously generated, and the embedded rich process knowledge provides an enabling means for the data-driven applications in intelligent process planning. At present, the driven geometry construction of the machining operations strongly relies on the experienced process designers to analyze the geometric features of the parts in detail, and this process is not only time-consuming and labor-intensive, but also unable to guarantee the construction of sufficiently optimal driven geometry. To address these limitations, this paper proposes an effective driven geometry construction method for contour finishing of complex parts via integrating graph neural network and reinforcement learning. First, based on the structured process data, a machining tool prediction model based on graph neural network (GNN) is proposed to learn the mapping relationships between machining regions under the contour finishing working step and machining tools. Then, the rules for dividing the machining region to generate the cutting areas that constitute the driven geometry and the constraints for the selection of the cutting area during the generation of the driven geometry are designed to construct an effective driven geometry. Finally, based on the rules and constraints for the generation of the driven geometry, the machining tool and the geometry topology of the machining region, a reinforcement learning method based on bidirectional long short-term memory neural network (Bi-LSTM) is proposed to generate the optimal driven geometry in terms of machining time. The experimental results show that the method can effectively construct the optimal driven geometry for part contour finishing and significantly reduce the machining time.
{"title":"An effective driven geometry construction method for contour finishing of complex parts via integrating graph neural network and reinforcement learning","authors":"Rui Huang , Fan Zhang , Shilong Yu , Bo Huang","doi":"10.1016/j.aei.2025.104281","DOIUrl":"10.1016/j.aei.2025.104281","url":null,"abstract":"<div><div>With the widespread use of computer-aided technologies represented by CAD/CAM/CAPP in product machining, a large amount of process data is continuously generated, and the embedded rich process knowledge provides an enabling means for the data-driven applications in intelligent process planning. At present, the driven geometry construction of the machining operations strongly relies on the experienced process designers to analyze the geometric features of the parts in detail, and this process is not only time-consuming and labor-intensive, but also unable to guarantee the construction of sufficiently optimal driven geometry. To address these limitations, this paper proposes an effective driven geometry construction method for contour finishing of complex parts via integrating graph neural network and reinforcement learning. First, based on the structured process data, a machining tool prediction model based on graph neural network (GNN) is proposed to learn the mapping relationships between machining regions under the contour finishing working step and machining tools. Then, the rules for dividing the machining region to generate the cutting areas that constitute the driven geometry and the constraints for the selection of the cutting area during the generation of the driven geometry are designed to construct an effective driven geometry. Finally, based on the rules and constraints for the generation of the driven geometry, the machining tool and the geometry topology of the machining region, a reinforcement learning method based on bidirectional long short-term memory neural network (Bi-LSTM) is proposed to generate the optimal driven geometry in terms of machining time. The experimental results show that the method can effectively construct the optimal driven geometry for part contour finishing and significantly reduce the machining time.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"71 ","pages":"Article 104281"},"PeriodicalIF":9.9,"publicationDate":"2025-12-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145842080","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}
Cooperative perception technology plays a crucial role in autonomous driving systems by improving safety and enabling real-time decision-making. However, existing LiDAR point cloud processing methods face significant challenges in both local geometric feature extraction and global feature fusion. To address these issues, this paper proposes DF-CoopNet, a cooperative perception framework comprising two core modules: Local Geometry Enhancement (LGE) and Sparse Key Feature Attention (SKFA). The LGE module enhances local geometric representations using a deformable k-nearest neighbor graph structure and adaptive fusion mechanism to effectively detect occluded targets. The SKFA module introduces a hierarchical sparse attention mechanism that balances performance and computational complexity through a Top-k strategy. Extensive experiments on the OPV2V and V2V4Real datasets demonstrate that DF-CoopNet significantly outperforms existing methods while maintaining robust detection performance even with substantially reduced point cloud data, validating its effectiveness for real-world cooperative perception applications.
{"title":"DF-CoopNet: Cooperative perception via local feature enhancement and global sparse attention","authors":"Hui Wu , Yu Xiao , Yisheng Chen , Chongcheng Chen , Ruihai Dong , Ding Lin","doi":"10.1016/j.aei.2025.104282","DOIUrl":"10.1016/j.aei.2025.104282","url":null,"abstract":"<div><div>Cooperative perception technology plays a crucial role in autonomous driving systems by improving safety and enabling real-time decision-making. However, existing LiDAR point cloud processing methods face significant challenges in both local geometric feature extraction and global feature fusion. To address these issues, this paper proposes DF-CoopNet, a cooperative perception framework comprising two core modules: Local Geometry Enhancement (LGE) and Sparse Key Feature Attention (SKFA). The LGE module enhances local geometric representations using a deformable k-nearest neighbor graph structure and adaptive fusion mechanism to effectively detect occluded targets. The SKFA module introduces a hierarchical sparse attention mechanism that balances performance and computational complexity through a Top-k strategy. Extensive experiments on the OPV2V and V2V4Real datasets demonstrate that DF-CoopNet significantly outperforms existing methods while maintaining robust detection performance even with substantially reduced point cloud data, validating its effectiveness for real-world cooperative perception applications.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"71 ","pages":"Article 104282"},"PeriodicalIF":9.9,"publicationDate":"2025-12-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145842081","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 : 2025-12-23DOI: 10.1016/j.aei.2025.104263
Xujie Long , Jing Teng , Zhiwei Zhu , Shaobo Zhao , Mengyang Pu , Ruifeng Shi , You Lv , Jonathan Li , Guoqing Jing
The complex geometries, environmental variability, and inconsistent imaging conditions in shield tunnel linings pose substantial challenges to water leakage detection. Existing models heavily rely on extensive annotated data from diverse environments to ensure reliable performance across varying scenarios, which incurs significant time and labor costs in data annotation. To alleviate the annotation burden, we propose Co-MixPL, a novel semi-supervised learning approach that integrates labeled data with pseudo-labels generated by the Mixed Pseudo Label (MixPL) strategy to iteratively update the teacher-student models. Specifically, Co-MixPL integrates an additional head into the MixPL framework to enhance the encoder’s discriminative capability and introduces a Soft Regression method to mitigate the inherent localization bias in pseudo-labeling, refining the regression loss of pseudo-labels through adaptive reliability scores. Remarkably, experiments on the public “water leakage” dataset, Mendeley Data V1, demonstrate that Co-MixPL approaches state-of-the-art (SOTA) performance using only one-seventh of the training data and outperforms the SOTA by 2.8 AP with merely one-third of the annotations. These findings highlight the effectiveness of Co-MixPL in delivering superior detection performance with significantly reduced annotations, thus better meeting the practical demands of engineering inspection and maintenance. Codes are available at https://github.com/LXJ010/Co-MixPL.
{"title":"Co-MixPL: An optimized semi-supervised learning method for tunnel water leakage detection","authors":"Xujie Long , Jing Teng , Zhiwei Zhu , Shaobo Zhao , Mengyang Pu , Ruifeng Shi , You Lv , Jonathan Li , Guoqing Jing","doi":"10.1016/j.aei.2025.104263","DOIUrl":"10.1016/j.aei.2025.104263","url":null,"abstract":"<div><div>The complex geometries, environmental variability, and inconsistent imaging conditions in shield tunnel linings pose substantial challenges to water leakage detection. Existing models heavily rely on extensive annotated data from diverse environments to ensure reliable performance across varying scenarios, which incurs significant time and labor costs in data annotation. To alleviate the annotation burden, we propose Co-MixPL, a novel semi-supervised learning approach that integrates labeled data with pseudo-labels generated by the Mixed Pseudo Label (MixPL) strategy to iteratively update the teacher-student models. Specifically, Co-MixPL integrates an additional head into the MixPL framework to enhance the encoder’s discriminative capability and introduces a Soft Regression method to mitigate the inherent localization bias in pseudo-labeling, refining the regression loss of pseudo-labels through adaptive reliability scores. Remarkably, experiments on the public “water leakage” dataset, Mendeley Data V1, demonstrate that Co-MixPL approaches state-of-the-art (SOTA) performance using only one-seventh of the training data and outperforms the SOTA by 2.8 AP with merely one-third of the annotations. These findings highlight the effectiveness of Co-MixPL in delivering superior detection performance with significantly reduced annotations, thus better meeting the practical demands of engineering inspection and maintenance. Codes are available at <span><span>https://github.com/LXJ010/Co-MixPL</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"71 ","pages":"Article 104263"},"PeriodicalIF":9.9,"publicationDate":"2025-12-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145841974","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 : 2025-12-23DOI: 10.1016/j.aei.2025.104265
Jiang-Jun Hou , Jinyu Lu , Jun Zou , Binglin Lai , Haichen Zhang , Na Li
Architectural free-form surfaces have been increasingly adopted in large-scale public buildings due to their unique and aesthetically appealing forms; the geometric complexity of these surfaces, nevertheless, presents significant challenges for grid partitioning, and a universally efficient method is still lacking. To this end, a novel grid partitioning approach for architectural free-form surfaces, termed GridGAN-TXT, is proposed herein, leveraging Generative Adversarial Networks (GAN), a form of generative artificial intelligence. A distinguishing feature of GridGAN-TXT is its ability to perform grid partitioning without explicit reliance on surface characteristics. Through training data mining and learning, the model autonomously extracts relevant features, enabling automated partitioning of free-form surfaces. Moreover, GridGAN-TXT can simultaneously generate grid structures composed of triangular or quadrilateral elements (via text prompts) — a capability not supported by previous methods, which typically require totally different strategies for each element type. The technical details of GridGAN-TXT are elaborated, and a parametric strategy is proposed for constructing large-scale training datasets. Additionally, a novel grid evaluation metric — similarity evaluation — is introduced to complement the existing geometric evaluation method. The effectiveness and generalizability of GridGAN-TXT are validated through ablation studies, extensive testing, and case analyses. Results demonstrate that GridGAN-TXT exhibits exceptional performance in partitioning grids on architectural free-form surfaces and can flexibly generate grid structures with varying basic elements in response to user text prompts. Such capabilities significantly enhance the efficiency of grid partitioning while simultaneously expanding the range of potential applications, highlighting the method as a strong candidate for practical implementation.
{"title":"GridGAN-TXT: An intelligent approach to partitioning architectural free-form surfaces with text prompts","authors":"Jiang-Jun Hou , Jinyu Lu , Jun Zou , Binglin Lai , Haichen Zhang , Na Li","doi":"10.1016/j.aei.2025.104265","DOIUrl":"10.1016/j.aei.2025.104265","url":null,"abstract":"<div><div>Architectural free-form surfaces have been increasingly adopted in large-scale public buildings due to their unique and aesthetically appealing forms; the geometric complexity of these surfaces, nevertheless, presents significant challenges for grid partitioning, and a universally efficient method is still lacking. To this end, a novel grid partitioning approach for architectural free-form surfaces, termed GridGAN-TXT, is proposed herein, leveraging Generative Adversarial Networks (GAN), a form of generative artificial intelligence. A distinguishing feature of GridGAN-TXT is its ability to perform grid partitioning without explicit reliance on surface characteristics. Through training data mining and learning, the model autonomously extracts relevant features, enabling automated partitioning of free-form surfaces. Moreover, GridGAN-TXT can simultaneously generate grid structures composed of triangular or quadrilateral elements (via text prompts) — a capability not supported by previous methods, which typically require totally different strategies for each element type. The technical details of GridGAN-TXT are elaborated, and a parametric strategy is proposed for constructing large-scale training datasets. Additionally, a novel grid evaluation metric — similarity evaluation — is introduced to complement the existing geometric evaluation method. The effectiveness and generalizability of GridGAN-TXT are validated through ablation studies, extensive testing, and case analyses. Results demonstrate that GridGAN-TXT exhibits exceptional performance in partitioning grids on architectural free-form surfaces and can flexibly generate grid structures with varying basic elements in response to user text prompts. Such capabilities significantly enhance the efficiency of grid partitioning while simultaneously expanding the range of potential applications, highlighting the method as a strong candidate for practical implementation.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"71 ","pages":"Article 104265"},"PeriodicalIF":9.9,"publicationDate":"2025-12-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145841908","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 : 2025-12-23DOI: 10.1016/j.aei.2025.104278
Zihang Dong , Cheng Hu , Xi Zhang , Yifan Shen , Xiaojun Shen , Jose I. Leon
Heating, ventilation, and air conditioning (HVAC) systems are significant contributors to energy consumption in buildings, directly affecting energy efficiency and occupant comfort. Traditional physics-based temperature modeling and centralized control frameworks often struggle to effectively balance the challenges of scaling across multiple buildings while minimizing operational costs. To address these challenges, this paper proposes a novel data-driven distributed control framework for the economical and resilient operation of the building community. Specifically, the framework employs artificial neural networks (ANNs) to capture multi-time-step nonlinear temperature dynamics, enhancing predictive accuracy for energy management across interconnected buildings. A distributed economic model predictive control (EMPC) strategy is developed, enabling local controllers to coordinate HVAC schedules in each building iteratively. This strategy minimizes energy shortages, optimizes overall community energy costs, and ensures thermal comfort. In addition, by facilitating energy interaction between HVAC systems, distributed energy resources (DERs), and storage units, the framework ensures electrical energy supply and demand balance during power outages. Simulation results demonstrate that the proposed strategy improves cost efficiency, resilience, and multi-step prediction accuracy, outperforming traditional physics-based EMPC approaches in coordination across multiple buildings.
{"title":"Optimal energy management of buildings using neural network-based thermal prediction and economic model predictive control","authors":"Zihang Dong , Cheng Hu , Xi Zhang , Yifan Shen , Xiaojun Shen , Jose I. Leon","doi":"10.1016/j.aei.2025.104278","DOIUrl":"10.1016/j.aei.2025.104278","url":null,"abstract":"<div><div>Heating, ventilation, and air conditioning (HVAC) systems are significant contributors to energy consumption in buildings, directly affecting energy efficiency and occupant comfort. Traditional physics-based temperature modeling and centralized control frameworks often struggle to effectively balance the challenges of scaling across multiple buildings while minimizing operational costs. To address these challenges, this paper proposes a novel data-driven distributed control framework for the economical and resilient operation of the building community. Specifically, the framework employs artificial neural networks (ANNs) to capture multi-time-step nonlinear temperature dynamics, enhancing predictive accuracy for energy management across interconnected buildings. A distributed economic model predictive control (EMPC) strategy is developed, enabling local controllers to coordinate HVAC schedules in each building iteratively. This strategy minimizes energy shortages, optimizes overall community energy costs, and ensures thermal comfort. In addition, by facilitating energy interaction between HVAC systems, distributed energy resources (DERs), and storage units, the framework ensures electrical energy supply and demand balance during power outages. Simulation results demonstrate that the proposed strategy improves cost efficiency, resilience, and multi-step prediction accuracy, outperforming traditional physics-based EMPC approaches in coordination across multiple buildings.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"71 ","pages":"Article 104278"},"PeriodicalIF":9.9,"publicationDate":"2025-12-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145841973","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 : 2025-12-22DOI: 10.1016/j.aei.2025.104256
Tianyang Zhang, Weizhi Xu, Shuguang Wang, Dongsheng Du
This study presents an integrated generative framework for the inverse design of damping microstructures in energy-dissipating steel walls (EDSWs) for seismic applications, establishing a seamless pipeline from large-scale pixel-based datasets to latent-space representation, three-dimensional reconstruction, industrial fabrication, and finite element analysis (FEA) verification. Starting from over 140,000 boundary-identical microstructures, a variational autoencoder-based TopoFormer compresses geometric features into latent codes, enabling over 90% reduction in generation complexity while maintaining high reconstruction fidelity. Representative structures are selected via k-means clustering in the latent space and analyzed through nonlinear FEA under shear and compression to construct a performance-labeled dataset. A conditional latent diffusion transformer (DiT) is then trained to map complete nonlinear mechanical performance curves to manufacturable geometries, thus achieving a one-to-many correspondence between target responses and structural configurations. Comparative evaluations show that the proposed DiT framework surpasses multiple CondUNet baselines, achieving the lowest FID (11.367) and the highest SSIM (0.676) with balanced coverage and precision. Experimental validation using laser-cut low-yield-point steel specimens under low-cycle reciprocating loading demonstrates close agreement between generated and target hysteresis curves, confirming both geometric fidelity and mechanical reliability. The results establish a scalable, high-accuracy, and experimentally validated approach for automated, performance-driven microstructure design, providing a practical pathway for incorporating generative artificial intelligence into the engineering development of next-generation seismic energy-dissipation systems. The related codes are available at https://github.com/AshenOneme/DiT-Based-Microstructures-Design.
{"title":"Latent diffusion–driven inverse design of damping microstructures with multiaxial nonlinear mechanical targets","authors":"Tianyang Zhang, Weizhi Xu, Shuguang Wang, Dongsheng Du","doi":"10.1016/j.aei.2025.104256","DOIUrl":"10.1016/j.aei.2025.104256","url":null,"abstract":"<div><div>This study presents an integrated generative framework for the inverse design of damping microstructures in energy-dissipating steel walls (EDSWs) for seismic applications, establishing a seamless pipeline from large-scale pixel-based datasets to latent-space representation, three-dimensional reconstruction, industrial fabrication, and finite element analysis (FEA) verification. Starting from over 140,000 boundary-identical microstructures, a variational autoencoder-based TopoFormer compresses geometric features into latent codes, enabling over 90% reduction in generation complexity while maintaining high reconstruction fidelity. Representative structures are selected via k-means clustering in the latent space and analyzed through nonlinear FEA under shear and compression to construct a performance-labeled dataset. A conditional latent diffusion transformer (DiT) is then trained to map complete nonlinear mechanical performance curves to manufacturable geometries, thus achieving a one-to-many correspondence between target responses and structural configurations. Comparative evaluations show that the proposed DiT framework surpasses multiple CondUNet baselines, achieving the lowest FID (11.367) and the highest SSIM (0.676) with balanced coverage and precision. Experimental validation using laser-cut low-yield-point steel specimens under low-cycle reciprocating loading demonstrates close agreement between generated and target hysteresis curves, confirming both geometric fidelity and mechanical reliability. The results establish a scalable, high-accuracy, and experimentally validated approach for automated, performance-driven microstructure design, providing a practical pathway for incorporating generative artificial intelligence into the engineering development of next-generation seismic energy-dissipation systems. The related codes are available at <span><span><strong><em>https://github.com/AshenOneme/DiT-Based-Microstructures-Design</em></strong></span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"71 ","pages":"Article 104256"},"PeriodicalIF":9.9,"publicationDate":"2025-12-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145841911","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 : 2025-12-22DOI: 10.1016/j.aei.2025.104232
Kaiyu Hu , Aili Yusup , Panfei Yan
This paper is devoted to designing a hybrid adaptive fault diagnosis (FD) and fault-tolerant control (FTC) scheme for nonlinear non-Gaussian stochastic systems with feedback packet loss and multi-mode variable structure. In order to solve the problem of inaccurate fault estimation caused by packet loss, the packet loss compensation method is proposed for adaptive learning FD. Hence the fault with incipient and large amplitudes are accurately estimated by prey adaptive strategy algorithm. Then, the second-order sliding mode FTC scheme is improved by adaptive harmonic learning functions with variable structure perturbations. Combining the nonlinearity and fault estimated by FD, this hybrid adaptive active FTC achieves the stable fault self-repair for the variable structure nonlinear non-Gaussian systems. Take the papermaking process system as an example, Lyapunov functions proves the stability, the effectiveness and superiority of the scheme are verified by numerical simulation.
{"title":"Hybrid adaptive fault-tolerant control of variable structure non-gaussian stochastic systems with feedback packet loss","authors":"Kaiyu Hu , Aili Yusup , Panfei Yan","doi":"10.1016/j.aei.2025.104232","DOIUrl":"10.1016/j.aei.2025.104232","url":null,"abstract":"<div><div>This paper is devoted to designing a hybrid adaptive fault diagnosis (FD) and fault-tolerant control (FTC) scheme for nonlinear non-Gaussian stochastic systems with feedback packet loss and multi-mode variable structure. In order to solve the problem of inaccurate fault estimation caused by packet loss, the packet loss compensation method is proposed for adaptive learning FD. Hence the fault with incipient and large amplitudes are accurately estimated by prey adaptive strategy algorithm. Then, the second-order sliding mode FTC scheme is improved by adaptive harmonic learning functions with variable structure perturbations. Combining the nonlinearity and fault estimated by FD, this hybrid adaptive active FTC achieves the stable fault self-repair for the variable structure nonlinear non-Gaussian systems. Take the papermaking process system as an example, Lyapunov functions proves the stability, the effectiveness and superiority of the scheme are verified by numerical simulation.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"71 ","pages":"Article 104232"},"PeriodicalIF":9.9,"publicationDate":"2025-12-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145841910","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 : 2025-12-22DOI: 10.1016/j.aei.2025.104262
Chuang Liang , Xuelin Mu , Ende Wang , Xiaoguang Zhang , Chengcheng Wang , Yubo Shao
Rotating machinery is the core equipment of the manufacturing industry, and its stability directly determines the operation of industrial systems. As a key component of rotating machinery, the accuracy of bearing fault diagnosis is particularly critical. Recently, deep learning (DL) has achieved remarkable results in mechanical fault diagnosis. However, traditional convolutional neural networks (CNNs) still have obvious limitations, which are difficult to effectively capture high-order nonlinearity features in the time–frequency domain and also show insufficient ability to decouple fault features in strong noise environments, which seriously restricts the improvement of diagnostic accuracy and model interpretability. To address these problems, a hybrid time–frequency spectral feature enhancement and attention fusion network (TFSAF-Net) is proposed. First, a time–frequency spectral feature enhancement module (TFSFEM) is designed. The TFSFEM employs learnable weight parameters to perform quadratic convolution nonlinear transformation on the wavelet time–frequency map of the signal in order to enhance its ability to extract higher-order fault features. Simultaneously, by integrating physical-driven feature decoupling module to extract envelope-related features and to utilize adaptive norm ratio-based feature metrics to improve the distinction between fault features and noise. Moreover, a convolutional multi-scale attention fusion module (CMSAFM) is developed. The CMSAFM introduces efficient multi-scale attention, which achieves precise focusing on key features through grouped feature interactions and adaptive weight allocation. Further, it realizes the effective integration of local details and global time–frequency distribution information through parallel extraction and weighted fusion of multi-scale features. Finally, a self-built engineering application datasets and PU dataset are implemented to validate the effectiveness and superiority of the TFSAF-Net. The experimental results demonstrate that our proposed method effectively captures high-order nonlinear features in the time–frequency domain and achieves efficient separation of fault characteristics from noise in highly noisy environments, and thereby reaches a higher diagnostic accuracy rate. Meanwhile, in terms of TFSAF-Net interpretability and generalization, it provides valuable insights for exploring similar problems within the field.
{"title":"TFSAF-Net: a hybrid network integrating time-frequency spectral feature enhancement and attention for fault diagnosis of rotating machinery","authors":"Chuang Liang , Xuelin Mu , Ende Wang , Xiaoguang Zhang , Chengcheng Wang , Yubo Shao","doi":"10.1016/j.aei.2025.104262","DOIUrl":"10.1016/j.aei.2025.104262","url":null,"abstract":"<div><div>Rotating machinery is the core equipment of the manufacturing industry, and its stability directly determines the operation of industrial systems. As a key component of rotating machinery, the accuracy of bearing fault diagnosis is particularly critical. Recently, deep learning (DL) has achieved remarkable results in mechanical fault diagnosis. However, traditional convolutional neural networks (CNNs) still have obvious limitations, which are difficult to effectively capture high-order nonlinearity features in the time–frequency domain and also show insufficient ability to decouple fault features in strong noise environments, which seriously restricts the improvement of diagnostic accuracy and model interpretability. To address these problems, a hybrid time–frequency spectral feature enhancement and attention fusion network (TFSAF-Net) is proposed. First, a time–frequency spectral feature enhancement module (TFSFEM) is designed. The TFSFEM employs learnable weight parameters to perform quadratic convolution nonlinear transformation on the wavelet time–frequency map of the signal in order to enhance its ability to extract higher-order fault features. Simultaneously, by integrating physical-driven feature decoupling module to extract envelope-related features and to utilize adaptive norm ratio-based feature metrics to improve the distinction between fault features and noise. Moreover, a convolutional multi-scale attention fusion module (CMSAFM) is developed. The CMSAFM introduces efficient multi-scale attention, which achieves precise focusing on key features through grouped feature interactions and adaptive weight allocation. Further, it realizes the effective integration of local details and global time–frequency distribution information through parallel extraction and weighted fusion of multi-scale features. Finally, a self-built engineering application datasets and PU dataset are implemented to validate the effectiveness and superiority of the TFSAF-Net. The experimental results demonstrate that our proposed method effectively captures high-order nonlinear features in the time–frequency domain and achieves efficient separation of fault characteristics from noise in highly noisy environments, and thereby reaches a higher diagnostic accuracy rate. Meanwhile, in terms of TFSAF-Net interpretability and generalization, it provides valuable insights for exploring similar problems within the field.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"71 ","pages":"Article 104262"},"PeriodicalIF":9.9,"publicationDate":"2025-12-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145841913","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