Engineering Applications of Artificial Intelligence最新文献_第8页

Discrete physics-informed neural network with enforced interface constraint for domain decomposition 面向领域分解的具有强制接口约束的离散物理信息神经网络

IF 8 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Engineering Applications of Artificial Intelligence

Pub Date : 2026-02-07 DOI: 10.1016/j.engappai.2026.114065

Jichao Yin , Mingxuan Li , Jianguang Fang , Chi Wu , Hu Wang , Guangyao Li

While domain decomposition method (DDM) constitutes an effective strategy for improving the training efficiency of physics-informed neural network (PINN), the approach simultaneously introduces an increased risk of training instability owing to the additional loss terms introduced. To address this issue, the work proposes an energy-based discrete PINN (dPINN) approach incorporating a proposed enforced interface constraint (EIC) mechanism within the context of the DDM. The dPINN builds upon the DDM with the EIC mechanism and will henceforth be referred to as EIC-DDM-dPINN. Within this framework, the dPINN computes the system energy in an element-wise fashion using Gaussian integration, guided by finite element-inspired formulations. Meanwhile, displacement continuity across subdomain interfaces is explicitly enforced through the EIC mechanism. This enforcement obviates the need to incorporate supplementary loss terms into the loss function, thereby substantially mitigating the risk of training instability. The integration of the EIC-based DDM facilitates simpler and more flexible subdomain mesh partitioning within the EIC-DDM-dPINN framework, thereby reducing the strong dependence on sampling strategies typically required in conventional DDM-based PINN. Beyond improving computational efficiency via parallelization, the DDM also helps decouple the weak spatial constraint (WSC) effect, which can otherwise result in spurious displacement continuity across geometrically discontinuous gaps. Comprehensive numerical experiments in both two- and three-dimensional settings are conducted to assess the accuracy and efficiency of the proposed approach, and the results demonstrate its scalability and robustness, highlighting its potential for application to large-scale problems with complex geometries.

虽然域分解方法（DDM）是提高物理信息神经网络（PINN）训练效率的有效策略，但由于引入了额外的损失项，该方法同时引入了增加训练不稳定性的风险。为了解决这个问题，本研究提出了一种基于能量的离散PINN （dPINN）方法，该方法在DDM上下文中结合了一种拟议的强制接口约束（EIC）机制。dPINN建立在带有EIC机制的DDM之上，因此将被称为EIC-DDM-dPINN。在此框架内，dPINN在有限元启发公式的指导下，使用高斯积分以单元方式计算系统能量。同时，通过EIC机制显式地实现了跨子域接口的位移连续性。这种强制消除了在损失函数中加入补充损失项的需要，从而大大降低了训练不稳定的风险。基于eic的DDM的集成使得EIC-DDM-dPINN框架内的子域网格划分更简单、更灵活，从而降低了传统基于DDM的PINN对采样策略的强烈依赖。除了通过并行化提高计算效率之外，DDM还有助于解耦弱空间约束（WSC）效应，否则会导致在几何不连续的间隙中产生虚假的位移连续性。在二维和三维环境下进行了全面的数值实验，以评估所提出方法的准确性和效率，结果表明其可扩展性和鲁棒性，突出了其应用于具有复杂几何形状的大规模问题的潜力。

{"title":"Discrete physics-informed neural network with enforced interface constraint for domain decomposition","authors":"Jichao Yin , Mingxuan Li , Jianguang Fang , Chi Wu , Hu Wang , Guangyao Li","doi":"10.1016/j.engappai.2026.114065","DOIUrl":"10.1016/j.engappai.2026.114065","url":null,"abstract":"<div><div>While domain decomposition method (DDM) constitutes an effective strategy for improving the training efficiency of physics-informed neural network (PINN), the approach simultaneously introduces an increased risk of training instability owing to the additional loss terms introduced. To address this issue, the work proposes an energy-based discrete PINN (dPINN) approach incorporating a proposed enforced interface constraint (EIC) mechanism within the context of the DDM. The dPINN builds upon the DDM with the EIC mechanism and will henceforth be referred to as EIC-DDM-dPINN. Within this framework, the dPINN computes the system energy in an element-wise fashion using Gaussian integration, guided by finite element-inspired formulations. Meanwhile, displacement continuity across subdomain interfaces is explicitly enforced through the EIC mechanism. This enforcement obviates the need to incorporate supplementary loss terms into the loss function, thereby substantially mitigating the risk of training instability. The integration of the EIC-based DDM facilitates simpler and more flexible subdomain mesh partitioning within the EIC-DDM-dPINN framework, thereby reducing the strong dependence on sampling strategies typically required in conventional DDM-based PINN. Beyond improving computational efficiency via parallelization, the DDM also helps decouple the weak spatial constraint (WSC) effect, which can otherwise result in spurious displacement continuity across geometrically discontinuous gaps. Comprehensive numerical experiments in both two- and three-dimensional settings are conducted to assess the accuracy and efficiency of the proposed approach, and the results demonstrate its scalability and robustness, highlighting its potential for application to large-scale problems with complex geometries.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"169 ","pages":"Article 114065"},"PeriodicalIF":8.0,"publicationDate":"2026-02-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146175403","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Enhanced anomaly interpretation in intelligent vehicles: A causal constraint graph attention network for root cause diagnosis 智能车辆中增强的异常解释：用于根本原因诊断的因果约束图注意网络

IF 8 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Engineering Applications of Artificial Intelligence

Pub Date : 2026-02-07 DOI: 10.1016/j.engappai.2026.114030

Shixiang Chen , Haigen Min , Xia Wu , Xiangmo Zhao

Intelligent vehicles increasingly rely on multi-sensor data for real-time environmental perception and decision-making, yet vehicular sensors are vulnerable to various disturbances, generating anomalies that threaten safety. Traditional anomaly diagnosis methods, such as graph attention networks, suffer from spurious correlations and fixed thresholds, leading to unreliable diagnoses under distribution shifts. To overcome these limitations, this paper proposes a causal constraint graph attention network for multi-sensor data anomaly detection and root cause diagnosis in intelligent vehicles. First, to address spurious correlations and parameter estimation bias in traditional graph attention networks, a sparse connection attention mechanism is designed that transforms fully-connected attention computation into one guided by node causal relationships. Second, to address the false positives and false negatives issues caused by a fixed threshold under environmental changes, a sliding adaptive threshold calculation method based on data anomaly degree weighting is proposed. Third, to achieve interpretable anomaly analysis, a hierarchical anomaly diagnosis strategy is implemented, which integrates feature reconstruction errors, variable causal relationships, and graph attention network weights for root cause localization and identification in anomaly propagation paths. Experiments with real vehicle data and public datasets demonstrate that the proposed method achieves excellent anomaly detection performance and enables an interpretable root cause diagnosis for six types of anomalies, benefiting from the causal constraint attention mechanism and adaptive threshold. With real-time inference capability, the proposed method provides significant opportunities for practical deployment of anomaly diagnosis in intelligent vehicle applications.

智能汽车越来越依赖多传感器数据进行实时环境感知和决策，但车载传感器容易受到各种干扰，产生威胁安全的异常。传统的异常诊断方法，如图关注网络，存在虚假的相关性和固定的阈值，导致分布变化下的诊断不可靠。为了克服这些局限性，本文提出了一种用于智能汽车多传感器数据异常检测和根本原因诊断的因果约束图关注网络。首先，针对传统图注意网络中存在的虚假关联和参数估计偏差问题，设计了一种稀疏连接注意机制，将全连接注意计算转化为节点因果关系引导下的注意计算。其次，针对环境变化下固定阈值导致的假阳性和假阴性问题，提出了一种基于数据异常度加权的滑动自适应阈值计算方法；第三，为了实现可解释的异常分析，实现了一种分层异常诊断策略，该策略集成了特征重构误差、变量因果关系和图关注网络权重，用于异常传播路径的根本原因定位和识别。在真实车辆数据和公共数据集上进行的实验表明，该方法具有良好的异常检测性能，利用因果约束注意机制和自适应阈值，能够对6类异常进行可解释的根本原因诊断。该方法具有实时推理能力，为智能车辆异常诊断的实际部署提供了重要的机会。

{"title":"Enhanced anomaly interpretation in intelligent vehicles: A causal constraint graph attention network for root cause diagnosis","authors":"Shixiang Chen , Haigen Min , Xia Wu , Xiangmo Zhao","doi":"10.1016/j.engappai.2026.114030","DOIUrl":"10.1016/j.engappai.2026.114030","url":null,"abstract":"<div><div>Intelligent vehicles increasingly rely on multi-sensor data for real-time environmental perception and decision-making, yet vehicular sensors are vulnerable to various disturbances, generating anomalies that threaten safety. Traditional anomaly diagnosis methods, such as graph attention networks, suffer from spurious correlations and fixed thresholds, leading to unreliable diagnoses under distribution shifts. To overcome these limitations, this paper proposes a causal constraint graph attention network for multi-sensor data anomaly detection and root cause diagnosis in intelligent vehicles. First, to address spurious correlations and parameter estimation bias in traditional graph attention networks, a sparse connection attention mechanism is designed that transforms fully-connected attention computation into one guided by node causal relationships. Second, to address the false positives and false negatives issues caused by a fixed threshold under environmental changes, a sliding adaptive threshold calculation method based on data anomaly degree weighting is proposed. Third, to achieve interpretable anomaly analysis, a hierarchical anomaly diagnosis strategy is implemented, which integrates feature reconstruction errors, variable causal relationships, and graph attention network weights for root cause localization and identification in anomaly propagation paths. Experiments with real vehicle data and public datasets demonstrate that the proposed method achieves excellent anomaly detection performance and enables an interpretable root cause diagnosis for six types of anomalies, benefiting from the causal constraint attention mechanism and adaptive threshold. With real-time inference capability, the proposed method provides significant opportunities for practical deployment of anomaly diagnosis in intelligent vehicle applications.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"169 ","pages":"Article 114030"},"PeriodicalIF":8.0,"publicationDate":"2026-02-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174846","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

LLM4CGDS: Large language model-based agents for Chinese graded document simplification LLM4CGDS：基于大型语言模型的中文分级文档简化代理

IF 8 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Engineering Applications of Artificial Intelligence

Pub Date : 2026-02-07 DOI: 10.1016/j.engappai.2026.113905

Dengzhao Fang , Jipeng Qiang , Wenjie Hou , Yi Zhu , Jingtong Gao , Xiangyu Zhao

Graded reading tailors text difficulty to learners’ proficiency by producing multiple versions of the same content—an approach long embraced in language education but still dependent on labor-intensive, expert-driven adaptation. In this paper, we introduce the task of Chinese Graded Document Simplification (CGDS) for non-native learners, which seeks to automate the creation of multi-level reading materials in accordance with established proficiency standards. Guided by the three stages of the Hanyu Shuiping Kaoshi (HSK) 3.0 framework (Levels 1–3 for Advanced, Levels 4–6 for Intermediate, and Levels 7–9 for Beginner learners), we propose Large Language Model for Chinese Graded Document Simplification (LLM4CGDS), a rule-guided, large language model (LLM)-based framework that integrates HSK-level readability constraints and external knowledge retrieval to control document-level simplification without requiring supervised fine-tuning. To foster further research, we construct two complementary datasets: Journey to the West Document Simplification (JWDS) and Multi-Domain Document Simplification (MDDS) that covering diverse genres and difficulty levels. Experimental evaluation on two datasets demonstrates that LLM4CGDS substantially outperforms direct prompting of state-of-the-art LLMs in both readability control and meaning preservation.

分级阅读通过生成相同内容的多个版本来根据学习者的熟练程度来调整文本难度——这是一种长期被语言教育所接受的方法，但仍然依赖于劳动密集型、专家驱动的适应。在本文中，我们介绍了面向非母语学习者的中文分级文档简化（CGDS）任务，该任务旨在根据既定的熟练程度标准自动创建多层次阅读材料。在HSK 3.0框架的三个阶段（高级1-3级、中级4-6级和初级7-9级）的指导下，我们提出了基于规则引导的大型语言模型（LLM4CGDS），该框架集成了HSK级别的可读性约束和外部知识检索来控制文档级别的简化，而无需监督微调。为了促进进一步的研究，我们构建了两个互补的数据集：西游记文献简化（JWDS）和多域文献简化（MDDS），涵盖了不同的体裁和难度级别。在两个数据集上的实验评估表明，LLM4CGDS在可读性控制和意义保存方面都大大优于最先进的llm直接提示。

{"title":"LLM4CGDS: Large language model-based agents for Chinese graded document simplification","authors":"Dengzhao Fang , Jipeng Qiang , Wenjie Hou , Yi Zhu , Jingtong Gao , Xiangyu Zhao","doi":"10.1016/j.engappai.2026.113905","DOIUrl":"10.1016/j.engappai.2026.113905","url":null,"abstract":"<div><div>Graded reading tailors text difficulty to learners’ proficiency by producing multiple versions of the same content—an approach long embraced in language education but still dependent on labor-intensive, expert-driven adaptation. In this paper, we introduce the task of <strong>C</strong>hinese <strong>G</strong>raded <strong>D</strong>ocument <strong>S</strong>implification (CGDS) for non-native learners, which seeks to automate the creation of multi-level reading materials in accordance with established proficiency standards. Guided by the three stages of the Hanyu Shuiping Kaoshi (HSK) 3.0 framework (Levels 1–3 for Advanced, Levels 4–6 for Intermediate, and Levels 7–9 for Beginner learners), we propose Large Language Model for Chinese Graded Document Simplification (LLM4CGDS), a rule-guided, large language model (LLM)-based framework that integrates HSK-level readability constraints and external knowledge retrieval to control document-level simplification without requiring supervised fine-tuning. To foster further research, we construct two complementary datasets: <strong>J</strong>ourney to the <strong>W</strong>est <strong>D</strong>ocument <strong>S</strong>implification (JWDS) and <strong>M</strong>ulti-<strong>D</strong>omain <strong>D</strong>ocument <strong>S</strong>implification (MDDS) that covering diverse genres and difficulty levels. Experimental evaluation on two datasets demonstrates that LLM4CGDS substantially outperforms direct prompting of state-of-the-art LLMs in both readability control and meaning preservation.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"169 ","pages":"Article 113905"},"PeriodicalIF":8.0,"publicationDate":"2026-02-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146175401","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Robust classification method for printed circuit board defects based on virtual and real space cooperative diffusion model 基于虚实空间协同扩散模型的印刷电路板缺陷鲁棒分类方法

IF 8 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Engineering Applications of Artificial Intelligence

Pub Date : 2026-02-07 DOI: 10.1016/j.engappai.2026.114126

Xinyi Yu , Haotian Sun , Yuanfu He , Jinmin Peng , Xianping Zeng , Liangshen Chen

Printed circuit board defect detection based on deep learning is a key technology in intelligent manufacturing. However, datasets collected in industrial environments often suffer from label noise due to inconsistent expert knowledge and ambiguous defect features, which severely degrades the performance of conventional models. To address this challenge, this paper proposes a robust classification framework based on a virtual-real space collaborative diffusion model. The proposed method introduces a distributed guidance mechanism to construct a latent semantic space (virtual space), which provides probabilistic priors for the inverse diffusion process in the real image space. Furthermore, a parameterized label encoding module is designed to mitigate the loss of fine-grained semantics during forward diffusion, and an attribute interaction attention mechanism is proposed to enhance the modeling of key defect attributes. Extensive experiments on the publicly available dataset with varying levels of label noise demonstrate that our method outperforms state-of-the-art approaches in terms of classification accuracy, robustness, and generated image quality, especially under strong noise conditions.

基于深度学习的印刷电路板缺陷检测是智能制造中的一项关键技术。然而，在工业环境中收集的数据集往往由于专家知识不一致和缺陷特征不明确而受到标签噪声的影响，这严重降低了传统模型的性能。为了解决这一挑战，本文提出了一个基于虚拟-真实空间协同扩散模型的鲁棒分类框架。该方法引入分布式引导机制，构建潜在语义空间（虚拟空间），为真实图像空间中的逆扩散过程提供概率先验。此外，设计了参数化标签编码模块以减轻前向扩散过程中细粒度语义的丢失，并提出了属性交互关注机制以增强关键缺陷属性的建模。在具有不同级别标签噪声的公开可用数据集上进行的大量实验表明，我们的方法在分类精度、鲁棒性和生成的图像质量方面优于最先进的方法，特别是在强噪声条件下。

{"title":"Robust classification method for printed circuit board defects based on virtual and real space cooperative diffusion model","authors":"Xinyi Yu , Haotian Sun , Yuanfu He , Jinmin Peng , Xianping Zeng , Liangshen Chen","doi":"10.1016/j.engappai.2026.114126","DOIUrl":"10.1016/j.engappai.2026.114126","url":null,"abstract":"<div><div>Printed circuit board defect detection based on deep learning is a key technology in intelligent manufacturing. However, datasets collected in industrial environments often suffer from label noise due to inconsistent expert knowledge and ambiguous defect features, which severely degrades the performance of conventional models. To address this challenge, this paper proposes a robust classification framework based on a virtual-real space collaborative diffusion model. The proposed method introduces a distributed guidance mechanism to construct a latent semantic space (virtual space), which provides probabilistic priors for the inverse diffusion process in the real image space. Furthermore, a parameterized label encoding module is designed to mitigate the loss of fine-grained semantics during forward diffusion, and an attribute interaction attention mechanism is proposed to enhance the modeling of key defect attributes. Extensive experiments on the publicly available dataset with varying levels of label noise demonstrate that our method outperforms state-of-the-art approaches in terms of classification accuracy, robustness, and generated image quality, especially under strong noise conditions.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"169 ","pages":"Article 114126"},"PeriodicalIF":8.0,"publicationDate":"2026-02-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146175404","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

A Contextual Multimodal Federated Transformer with dual distillation for decentralized chronic obstructive pulmonary disease related lung pathology classification 一个上下文多模式联合变压器与双蒸馏分散慢性阻塞性肺疾病相关的肺病理分类

IF 8 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Engineering Applications of Artificial Intelligence

Pub Date : 2026-02-07 DOI: 10.1016/j.engappai.2026.114046

Ayesha Jabbar, Huang Jianjun, Muhammad Kashif Jabbar

The use of multimodal data within decentralized healthcare presents an opportunity to improve artificial intelligence (AI)–based classification of lung pathology related to chronic obstructive pulmonary disease (COPD) while maintaining patient privacy. Nonetheless, heterogeneous client modalities, non-independent and non-identically distributed (non-IID) data, and equity challenges in aggregating models remain unresolved. To address these issues, we propose CMF-Former-D (Contextual Multimodal Federated Transformer with Dual-Level Distillation), a privacy-preserving decentralized learning method for peer-to-peer training across heterogeneous edge devices. CMF-Former-D employs an attention-based fusion mechanism that integrates audio features from respiratory sounds and image features from chest X-rays. We further introduce PeerMesh-Distill, a server-free protocol that enables decentralized knowledge sharing with equitable distribution of model updates across heterogeneous client sites, and FairWeight-Gossip, a strategy that promotes fair update aggregation among clients. The model is dynamically adapted to client hardware and modality constraints using resource-aware configurations. CMF-Former-D achieves 98.3% accuracy, a macro-averaged F1-score (macro-F1) of 0.987, and an area under the receiver operating characteristic curve (AUROC) of 0.972 on a synthetic proxy-aligned benchmark. End-to-end multimodal inference latency is 150 ms (ms) on a central processing unit (CPU) (batch size = 1), while graphics processing unit (GPU) latency is reported separately on an edge GPU under batch size = 16. Statistical tests indicate significant improvements (

p

-value (p)

< 0.01

), and client-level analysis shows reduced performance disparities across clients.

在分散式医疗保健中使用多模式数据为改进基于人工智能（AI）的与慢性阻塞性肺疾病（COPD）相关的肺部病理分类提供了机会，同时维护了患者隐私。尽管如此，异构客户端模式、非独立和非同分布（非iid）数据以及聚合模型中的公平性挑战仍未得到解决。为了解决这些问题，我们提出了CMF-Former-D (context Multimodal Federated Transformer with Dual-Level Distillation)，这是一种保护隐私的分散学习方法，用于跨异构边缘设备的点对点训练。CMF-Former-D采用基于注意力的融合机制，将呼吸声音的音频特征和胸部x光片的图像特征集成在一起。我们进一步介绍了peermesh -蒸馏，这是一种无服务器协议，可以通过在异构客户端站点之间公平分配模型更新来实现分散的知识共享，以及FairWeight-Gossip，这是一种促进客户端之间公平更新聚合的策略。该模型使用资源感知配置动态地适应客户机硬件和模态约束。在合成代理校准基准上，CMF-Former-D的准确率为98.3%，宏观平均f1评分（macro-F1）为0.987，接收者工作特征曲线下面积（AUROC）为0.972。端到端多模态推理延迟在中央处理单元（CPU）（批大小= 1）上为150毫秒（ms），而在边缘GPU（批大小= 16）上单独报告图形处理单元（GPU）延迟。统计测试表明有显著的改进（p值(p) <0.01），客户级分析表明客户端之间的性能差异减少了。

{"title":"A Contextual Multimodal Federated Transformer with dual distillation for decentralized chronic obstructive pulmonary disease related lung pathology classification","authors":"Ayesha Jabbar, Huang Jianjun, Muhammad Kashif Jabbar","doi":"10.1016/j.engappai.2026.114046","DOIUrl":"10.1016/j.engappai.2026.114046","url":null,"abstract":"<div><div>The use of multimodal data within decentralized healthcare presents an opportunity to improve artificial intelligence (AI)–based classification of lung pathology related to chronic obstructive pulmonary disease (COPD) while maintaining patient privacy. Nonetheless, heterogeneous client modalities, non-independent and non-identically distributed (non-IID) data, and equity challenges in aggregating models remain unresolved. To address these issues, we propose CMF-Former-D (Contextual Multimodal Federated Transformer with Dual-Level Distillation), a privacy-preserving decentralized learning method for peer-to-peer training across heterogeneous edge devices. CMF-Former-D employs an attention-based fusion mechanism that integrates audio features from respiratory sounds and image features from chest X-rays. We further introduce PeerMesh-Distill, a server-free protocol that enables decentralized knowledge sharing with equitable distribution of model updates across heterogeneous client sites, and FairWeight-Gossip, a strategy that promotes fair update aggregation among clients. The model is dynamically adapted to client hardware and modality constraints using resource-aware configurations. CMF-Former-D achieves 98.3% accuracy, a macro-averaged F1-score (macro-F1) of 0.987, and an area under the receiver operating characteristic curve (AUROC) of 0.972 on a synthetic proxy-aligned benchmark. End-to-end multimodal inference latency is 150 ms (ms) on a central processing unit (CPU) (batch size = 1), while graphics processing unit (GPU) latency is reported separately on an edge GPU under batch size = 16. Statistical tests indicate significant improvements (<span><math><mi>p</mi></math></span>-value (p) <span><math><mrow><mo><</mo><mn>0</mn><mo>.</mo><mn>01</mn></mrow></math></span>), and client-level analysis shows reduced performance disparities across clients.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"169 ","pages":"Article 114046"},"PeriodicalIF":8.0,"publicationDate":"2026-02-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174844","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

A new distance approach to measure the evidence conflict in orderable set and its applications 一种新的测量有序集证据冲突的距离方法及其应用

IF 8 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Engineering Applications of Artificial Intelligence

Pub Date : 2026-02-07 DOI: 10.1016/j.engappai.2026.114112

Xiaoyan Su, Yanying Yu

Dempster–Shafer (D–S) evidence theory can effectively represent and handle uncertainty, and has been widely applied across various domains. In D–S evidence theory, distance is a key tool for quantifying the conflict between basic probability assignments (BPAs). Such measurement is crucial in artificial intelligence (AI) tasks such as clustering and pattern recognition under uncertainty, and plays an important role in ensuring the reliability of multi-source information fusion. In this paper, a new distance metric is proposed to account for the ordered frame of discernment (FOD). A relative position coefficient is first defined to capture the intrinsic ordinal relationships among focal elements. Based on this coefficient, a relative position matrix is then constructed to characterize the overall distributional differences between two pieces of evidence. The proposed distance satisfies the axioms of non-negativity, symmetry, and triangle inequality. It avoids counter-intuitive results and improves conflict quantification. For engineering applications, the method is validated through two case studies: the reliability assessment of vehicle roll angle sensors and the expert group decision-making process. The results indicate that the proposed distance not only advances the modeling of ordered uncertainty, but also provides practical benefits for improving system reliability in real-world engineering scenarios. Overall, the research offers a theoretically sound and practically effective framework for handling ordered uncertainty, enabling more reliable conflict analysis.

Dempster-Shafer （D-S）证据理论能够有效地表示和处理不确定性，在各个领域得到了广泛的应用。在D-S证据理论中，距离是量化基本概率分配之间冲突的关键工具。这种度量在不确定条件下的聚类和模式识别等人工智能任务中至关重要，对保证多源信息融合的可靠性起着重要作用。本文提出了一种新的距离度量来考虑有序识别框架（FOD）。首先定义了相对位置系数来捕捉焦点元素之间的内在顺序关系。基于该系数，然后构建一个相对位置矩阵来表征两个证据之间的总体分布差异。提出的距离满足非负性公理、对称性公理和三角不等式公理。它避免了反直觉的结果，并改善了冲突的量化。在工程应用方面，通过车辆侧倾角传感器可靠性评估和专家组决策过程两个案例对该方法进行了验证。结果表明，所提出的距离不仅推进了有序不确定性的建模，而且对提高实际工程场景下的系统可靠性具有实际意义。总体而言，该研究为处理有序不确定性提供了理论健全和实践有效的框架，使冲突分析更加可靠。

{"title":"A new distance approach to measure the evidence conflict in orderable set and its applications","authors":"Xiaoyan Su, Yanying Yu","doi":"10.1016/j.engappai.2026.114112","DOIUrl":"10.1016/j.engappai.2026.114112","url":null,"abstract":"<div><div>Dempster–Shafer (D–S) evidence theory can effectively represent and handle uncertainty, and has been widely applied across various domains. In D–S evidence theory, distance is a key tool for quantifying the conflict between basic probability assignments (BPAs). Such measurement is crucial in artificial intelligence (AI) tasks such as clustering and pattern recognition under uncertainty, and plays an important role in ensuring the reliability of multi-source information fusion. In this paper, a new distance metric is proposed to account for the ordered frame of discernment (FOD). A relative position coefficient is first defined to capture the intrinsic ordinal relationships among focal elements. Based on this coefficient, a relative position matrix is then constructed to characterize the overall distributional differences between two pieces of evidence. The proposed distance satisfies the axioms of non-negativity, symmetry, and triangle inequality. It avoids counter-intuitive results and improves conflict quantification. For engineering applications, the method is validated through two case studies: the reliability assessment of vehicle roll angle sensors and the expert group decision-making process. The results indicate that the proposed distance not only advances the modeling of ordered uncertainty, but also provides practical benefits for improving system reliability in real-world engineering scenarios. Overall, the research offers a theoretically sound and practically effective framework for handling ordered uncertainty, enabling more reliable conflict analysis.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"169 ","pages":"Article 114112"},"PeriodicalIF":8.0,"publicationDate":"2026-02-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146175350","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Compact axial attention with detail enhancement for visual object tracking 紧凑轴向注意与细节增强视觉目标跟踪

IF 8 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Engineering Applications of Artificial Intelligence

Pub Date : 2026-02-07 DOI: 10.1016/j.engappai.2026.114121

Yuanyun Wang , Geng Gu , Chao Tang , Jun Wang

The current Transformer-based trackers are mainly divided into two types, Convolutional Neural Networks-Transformer-based and fully Transformer-based. Due to the local receptive field of Convolutional Neural Networks, Convolutional Neural Networks-Transformer-based trackers still struggle to capture long-range dependencies and global context, which can degrade tracking performance in complex scenarios such as occlusion and large viewpoint change. The fully Transformer-based uses Transformer for feature extraction and relationship modeling. This type of tracker is often a single-stream single-stage tracking framework. This not only brings huge computational costs, but also loses the template features cached in the test phase of the two-stage trackers. In response to this, we design a compact axial attention with a detail enhancement module, which compresses the features onto a single axis to reduce computational complexity without compromising global information. Based on this, a novel Transformer-based feature extraction backbone is designed. Further, we propose a new tracker, compact axial attention with detail enhancement for visual object tracking, which achieves the state-of-the-art performance on six challenging tracking benchmarks. In particular, the proposed tracker achieves an area under curve (AUC) score of 65.8% on large-scale single object tracking (LaSOT), a normalized precision score of 86.8% on the large-scale dataset and benchmark for object tracking in the wild (TrackingNet), and a precision score of 90.0% on unmanned aerial vehicle dataset (UAV123).

目前基于变压器的跟踪器主要分为基于卷积神经网络的变压器跟踪器和完全基于变压器跟踪器的跟踪器两种。由于卷积神经网络的局部接受场，基于卷积神经网络-变压器的跟踪器仍然难以捕获远程依赖关系和全局上下文，这可能会降低在遮挡和大视点变化等复杂场景下的跟踪性能。完全基于Transformer的使用Transformer进行特征提取和关系建模。这种类型的跟踪器通常是单流单阶段跟踪框架。这不仅带来了巨大的计算成本，而且还丢失了缓存在两阶段跟踪器测试阶段的模板特征。针对这一点，我们设计了一个紧凑的轴向关注和细节增强模块，该模块将特征压缩到单个轴上，以降低计算复杂性，同时不影响全局信息。在此基础上，设计了一种基于变压器的特征提取主干。此外，我们提出了一种新的跟踪器，紧凑的轴向注意力与细节增强视觉目标跟踪，它在六个具有挑战性的跟踪基准上实现了最先进的性能。特别是，该跟踪器在大规模单目标跟踪（LaSOT）上的曲线下面积（AUC）得分为65.8%，在大规模数据集和野外目标跟踪基准（TrackingNet）上的归一化精度得分为86.8%，在无人机数据集（UAV123）上的精度得分为90.0%。

{"title":"Compact axial attention with detail enhancement for visual object tracking","authors":"Yuanyun Wang , Geng Gu , Chao Tang , Jun Wang","doi":"10.1016/j.engappai.2026.114121","DOIUrl":"10.1016/j.engappai.2026.114121","url":null,"abstract":"<div><div>The current Transformer-based trackers are mainly divided into two types, Convolutional Neural Networks-Transformer-based and fully Transformer-based. Due to the local receptive field of Convolutional Neural Networks, Convolutional Neural Networks-Transformer-based trackers still struggle to capture long-range dependencies and global context, which can degrade tracking performance in complex scenarios such as occlusion and large viewpoint change. The fully Transformer-based uses Transformer for feature extraction and relationship modeling. This type of tracker is often a single-stream single-stage tracking framework. This not only brings huge computational costs, but also loses the template features cached in the test phase of the two-stage trackers. In response to this, we design a compact axial attention with a detail enhancement module, which compresses the features onto a single axis to reduce computational complexity without compromising global information. Based on this, a novel Transformer-based feature extraction backbone is designed. Further, we propose a new tracker, compact axial attention with detail enhancement for visual object tracking, which achieves the state-of-the-art performance on six challenging tracking benchmarks. In particular, the proposed tracker achieves an area under curve (AUC) score of 65.8% on large-scale single object tracking (LaSOT), a normalized precision score of 86.8% on the large-scale dataset and benchmark for object tracking in the wild (TrackingNet), and a precision score of 90.0% on unmanned aerial vehicle dataset (UAV123).</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"169 ","pages":"Article 114121"},"PeriodicalIF":8.0,"publicationDate":"2026-02-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146175405","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Structural-aware key node identification in hypergraphs via representation learning and fine-tuning 基于表征学习和微调的超图结构感知关键节点识别

IF 8 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Engineering Applications of Artificial Intelligence

Pub Date : 2026-02-07 DOI: 10.1016/j.engappai.2026.114108

Xiaonan Ni , Guangyuan Mei , Su-Su Zhang , Yang Chen , Xin Xu , Chuang Liu , Xiu-Xiu Zhan

The ability to pinpoint strategically important nodes plays a decisive role in shaping diffusion outcomes and maintaining the stability of complex systems. Yet, most existing approaches remain rooted in pairwise interaction assumptions, making them ill-suited for systems where collective participation and attribute-sharing give rise to higher-order structures. In this work, we introduce AHGA, a learning-driven framework that leverages autoencoder-based representations, hypergraph neural network pre-training, and an active learning mechanism to uncover nodes that jointly influence propagation dynamics and structural cohesion. Rather than relying on handcrafted descriptors, AHGA learns informative higher-order features and progressively refines node importance through selective supervision. Evaluations on eight empirical hypergraphs show that this strategy leads to substantially more reliable rankings, with improvements of up to 36.8% over classical baselines. Beyond ranking accuracy, nodes prioritized by AHGA exhibit pronounced structural leverage: their removal triggers an accelerated loss of network efficiency, reaching 0.6628, markedly exceeding the disruptive effect achieved by competing methods. These findings demonstrate that AHGA not only advances higher-order node identification methodology, but also offers practical guidance for intervention strategies in scenarios such as misinformation containment and infrastructure robustness.

精确定位战略重要节点的能力在形成扩散结果和维持复杂系统的稳定性方面起着决定性作用。然而，大多数现有的方法仍然植根于成对交互假设，这使得它们不适合集体参与和属性共享产生高阶结构的系统。在这项工作中，我们引入了AHGA，这是一个学习驱动的框架，它利用基于自编码器的表示、超图神经网络预训练和主动学习机制来发现共同影响传播动态和结构内聚的节点。AHGA不依赖于手工制作的描述符，而是学习信息丰富的高阶特征，并通过选择性监督逐步优化节点重要性。对八个经验超图的评估表明，这种策略导致了更可靠的排名，比经典基线提高了36.8%。除了排序精度之外，AHGA优先排序的节点表现出明显的结构杠杆：它们的移除会加速网络效率的损失，达到0.6628，明显超过竞争方法所达到的破坏效果。这些发现表明，AHGA不仅推进了高阶节点识别方法，而且为错误信息遏制和基础设施鲁棒性等情况下的干预策略提供了实用指导。

{"title":"Structural-aware key node identification in hypergraphs via representation learning and fine-tuning","authors":"Xiaonan Ni , Guangyuan Mei , Su-Su Zhang , Yang Chen , Xin Xu , Chuang Liu , Xiu-Xiu Zhan","doi":"10.1016/j.engappai.2026.114108","DOIUrl":"10.1016/j.engappai.2026.114108","url":null,"abstract":"<div><div>The ability to pinpoint strategically important nodes plays a decisive role in shaping diffusion outcomes and maintaining the stability of complex systems. Yet, most existing approaches remain rooted in pairwise interaction assumptions, making them ill-suited for systems where collective participation and attribute-sharing give rise to higher-order structures. In this work, we introduce AHGA, a learning-driven framework that leverages autoencoder-based representations, hypergraph neural network pre-training, and an active learning mechanism to uncover nodes that jointly influence propagation dynamics and structural cohesion. Rather than relying on handcrafted descriptors, AHGA learns informative higher-order features and progressively refines node importance through selective supervision. Evaluations on eight empirical hypergraphs show that this strategy leads to substantially more reliable rankings, with improvements of up to 36.8% over classical baselines. Beyond ranking accuracy, nodes prioritized by AHGA exhibit pronounced structural leverage: their removal triggers an accelerated loss of network efficiency, reaching 0.6628, markedly exceeding the disruptive effect achieved by competing methods. These findings demonstrate that AHGA not only advances higher-order node identification methodology, but also offers practical guidance for intervention strategies in scenarios such as misinformation containment and infrastructure robustness.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"169 ","pages":"Article 114108"},"PeriodicalIF":8.0,"publicationDate":"2026-02-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174974","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Cross-domain few-shot hyperspectral classification via orthogonal feature disentanglement 基于正交特征解纠缠的跨域少射高光谱分类

IF 8 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Engineering Applications of Artificial Intelligence

Pub Date : 2026-02-07 DOI: 10.1016/j.engappai.2026.114071

Yurong Zhang, Jinrong He, Yuhang Li

In cross-domain few-shot hyperspectral image classification, the limited availability of labeled target-domain samples renders the model highly sensitive to domain shifts. Existing feature disentanglement approaches struggle to simultaneously suppress domain-specific noise and retain cross-domain discriminative semantics. This leads to the entanglement of domain-shared and domain-specific components, increases the risk of negative transfer, and ultimately becomes the bottleneck limiting further improvements in accuracy and generalization. To address these challenges, this paper presents an Orthogonal Feature Disentanglement Network (OFD-Net). Using orthogonal subspace decomposition, OFD-Net projects features into two mutually exclusive subspaces: domain-shared and domain-specific. The domain-shared subspace focuses on extracting cross-domain invariant features, while the domain-specific subspace retains local domain discriminative information. This dual-stream architecture effectively suppresses interference from irrelevant inter-domain features. Additionally, feature orthogonality constraints enhance the model's adaptability to target domain shifts. OFD-Net also adopts a multi-task learning framework. Cross-domain alignment loss ensures the consistency of shared feature distributions between the source and target domains, while inter-class discriminative loss improves the class separability of specific features, creating a hierarchical feature optimization mechanism. On four public benchmark datasets including Indian Pines, Pavia University, Salinas, and Houston, OFD-Net achieves Overall Accuracy of 80.33%, 85.80%, 93.33%, and 80.05% respectively. Its performance outperforms existing state-of-the-art methods, demonstrating superior cross-domain transfer robustness and feature discriminative capability.

在跨域少镜头高光谱图像分类中，标记目标域样本的有限可用性使得模型对域漂移高度敏感。现有的特征解纠缠方法难以同时抑制特定领域的噪声和保留跨领域的判别语义。这导致了领域共享和领域特定组件的纠缠，增加了负迁移的风险，并最终成为限制准确性和泛化进一步提高的瓶颈。为了解决这些问题，本文提出了正交特征解纠缠网络（OFD-Net）。利用正交子空间分解，OFD-Net将特征划分为两个相互排斥的子空间：域共享和域特定。领域共享子空间侧重于提取跨领域的不变特征，而特定于领域的子空间则保留了局部领域的判别信息。这种双流结构有效地抑制了不相关域间特征的干扰。此外，特征正交性约束增强了模型对目标域位移的适应性。OFD-Net还采用了多任务学习框架。跨域对齐损失保证了源域和目标域共享特征分布的一致性，而类间判别损失提高了特定特征的类可分性，创建了层次化的特征优化机制。在Indian Pines、Pavia University、Salinas和Houston四个公共基准数据集上，OFD-Net的总体准确率分别为80.33%、85.80%、93.33%和80.05%。它的性能优于现有的最先进的方法，表现出优越的跨域转移鲁棒性和特征判别能力。

{"title":"Cross-domain few-shot hyperspectral classification via orthogonal feature disentanglement","authors":"Yurong Zhang, Jinrong He, Yuhang Li","doi":"10.1016/j.engappai.2026.114071","DOIUrl":"10.1016/j.engappai.2026.114071","url":null,"abstract":"<div><div>In cross-domain few-shot hyperspectral image classification, the limited availability of labeled target-domain samples renders the model highly sensitive to domain shifts. Existing feature disentanglement approaches struggle to simultaneously suppress domain-specific noise and retain cross-domain discriminative semantics. This leads to the entanglement of domain-shared and domain-specific components, increases the risk of negative transfer, and ultimately becomes the bottleneck limiting further improvements in accuracy and generalization. To address these challenges, this paper presents an Orthogonal Feature Disentanglement Network (OFD-Net). Using orthogonal subspace decomposition, OFD-Net projects features into two mutually exclusive subspaces: domain-shared and domain-specific. The domain-shared subspace focuses on extracting cross-domain invariant features, while the domain-specific subspace retains local domain discriminative information. This dual-stream architecture effectively suppresses interference from irrelevant inter-domain features. Additionally, feature orthogonality constraints enhance the model's adaptability to target domain shifts. OFD-Net also adopts a multi-task learning framework. Cross-domain alignment loss ensures the consistency of shared feature distributions between the source and target domains, while inter-class discriminative loss improves the class separability of specific features, creating a hierarchical feature optimization mechanism. On four public benchmark datasets including Indian Pines, Pavia University, Salinas, and Houston, OFD-Net achieves Overall Accuracy of 80.33%, 85.80%, 93.33%, and 80.05% respectively. Its performance outperforms existing state-of-the-art methods, demonstrating superior cross-domain transfer robustness and feature discriminative capability.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"169 ","pages":"Article 114071"},"PeriodicalIF":8.0,"publicationDate":"2026-02-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146175402","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Predicting stress in two-phase random materials and super-resolution method for stress images by embedding physical information 两相随机材料应力预测及嵌入物理信息的应力图像超分辨方法

IF 8 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Engineering Applications of Artificial Intelligence

Pub Date : 2026-02-07 DOI: 10.1016/j.engappai.2026.114087

Tengfei Xing , Xiaodan Ren , Jie Li

Stress analysis is essential in material design. In materials with complex microstructures, such as two-phase random materials (TRMs), failure is typically associated with stress concentration at phase interfaces that govern mechanical performance. Existing Image Super-Resolution (ISR) methods are mainly data-driven and rely on supervised learning, where the achievable magnification of stress images is tightly constrained by the resolution of the training dataset. This limits their applicability for TRMs, where high-resolution stress information at phase interfaces is essential but often unavailable. In practical engineering, limited pixel resolution in microstructure images further constrains stress image clarity and hinders the observation of stress concentration regions. To address this research gap, we propose a stress prediction framework tailored for TRMs that combines microstructural information with physics-based constraints. First, the framework employs a Multiple Compositions U-net (MC U-net) to predict stress in low-resolution material microstructures. By incorporating phase interface information, the MC U-net effectively reduces prediction errors at phase interfaces. Secondly, a Mixed Physics-Informed Neural Network (MPINN)-based stress ISR method (SRMPINN) is introduced. Unlike conventional ISR methods, SRMPINN leverages physical constraints to achieve stress image super-resolution without requiring paired high-resolution training images, enabling stress images to be generated at substantially high magnification factors, including non-integer scales, with magnification ratios not restricted by the training dataset. Finally, transfer learning is applied to perform stress analysis on TRMs with different loading states and anisotropy. The results demonstrate that the proposed framework achieves high accuracy, generalization, and flexibility, particularly in resolving stress concentrations at phase interfaces.

应力分析在材料设计中是必不可少的。在具有复杂微观结构的材料中，如两相随机材料（TRMs），失效通常与控制力学性能的相界面应力集中有关。现有的图像超分辨率（ISR）方法主要是数据驱动的，依赖于监督学习，其中应力图像的可实现放大程度受到训练数据集分辨率的严格限制。这限制了它们对trm的适用性，在trm中，相位界面的高分辨率应力信息是必不可少的，但通常不可用。在实际工程中，微观结构图像像素分辨率的限制进一步限制了应力图像的清晰度，阻碍了对应力集中区域的观察。为了解决这一研究缺口，我们提出了一种针对trm的应力预测框架，该框架将微观结构信息与基于物理的约束相结合。首先，该框架采用多组分U-net （MC U-net）来预测低分辨率材料微结构中的应力。MC - U-net通过引入相界面信息，有效地降低了相界面处的预测误差。其次，介绍了一种基于混合物理信息神经网络的应力ISR方法（SRMPINN）。与传统的ISR方法不同，SRMPINN利用物理约束来实现应力图像的超分辨率，而不需要配对的高分辨率训练图像，使应力图像能够以非常高的放大因子生成，包括非整数尺度，放大倍率不受训练数据集的限制。最后，应用迁移学习对具有不同加载状态和各向异性的trm进行应力分析。结果表明，该框架具有较高的精度、通用性和灵活性，特别是在解决相界面应力集中方面。

{"title":"Predicting stress in two-phase random materials and super-resolution method for stress images by embedding physical information","authors":"Tengfei Xing , Xiaodan Ren , Jie Li","doi":"10.1016/j.engappai.2026.114087","DOIUrl":"10.1016/j.engappai.2026.114087","url":null,"abstract":"<div><div>Stress analysis is essential in material design. In materials with complex microstructures, such as two-phase random materials (TRMs), failure is typically associated with stress concentration at phase interfaces that govern mechanical performance. Existing Image Super-Resolution (ISR) methods are mainly data-driven and rely on supervised learning, where the achievable magnification of stress images is tightly constrained by the resolution of the training dataset. This limits their applicability for TRMs, where high-resolution stress information at phase interfaces is essential but often unavailable. In practical engineering, limited pixel resolution in microstructure images further constrains stress image clarity and hinders the observation of stress concentration regions. To address this research gap, we propose a stress prediction framework tailored for TRMs that combines microstructural information with physics-based constraints. First, the framework employs a Multiple Compositions U-net (MC U-net) to predict stress in low-resolution material microstructures. By incorporating phase interface information, the MC U-net effectively reduces prediction errors at phase interfaces. Secondly, a Mixed Physics-Informed Neural Network (MPINN)-based stress ISR method (SRMPINN) is introduced. Unlike conventional ISR methods, SRMPINN leverages physical constraints to achieve stress image super-resolution without requiring paired high-resolution training images, enabling stress images to be generated at substantially high magnification factors, including non-integer scales, with magnification ratios not restricted by the training dataset. Finally, transfer learning is applied to perform stress analysis on TRMs with different loading states and anisotropy. The results demonstrate that the proposed framework achieves high accuracy, generalization, and flexibility, particularly in resolving stress concentrations at phase interfaces.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"169 ","pages":"Article 114087"},"PeriodicalIF":8.0,"publicationDate":"2026-02-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174845","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0