Advanced Engineering Informatics最新文献_第6页

Robust and interpretable bearing fault diagnosis via multi-scale differential-enhanced convolution with dual-dimensional redundancy suppression and dynamic quaternion transformer 基于二维冗余抑制和动态四元数变换的多尺度差分增强卷积的鲁棒可解释轴承故障诊断

IF 9.9 1区工程技术 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Advanced Engineering Informatics

Pub Date : 2026-01-30 DOI: 10.1016/j.aei.2026.104397

Chuang Chen , Zefeng Wang , Jiantao Shi , Dongdong Yue , Dan Bao , Ge Shi , Cunsong Wang , Tao Xie

Bearings are critical components that ensure the safe and efficient operation of mechanical equipment. However, most existing bearing fault diagnosis models suffer significant performance degradation under strong noise conditions, where fault features are easily overwhelmed by noise. To address this issue, this paper proposes a robust and locally interpretable intelligent bearing fault diagnosis model (named MSDRQF), achieving accurate identification in noisy environments. The proposed model integrates a multi-scale differential enhancement convolution and gated fusion module (MSDEC-GFM), a dual-dimensional redundancy suppression unit with differential-integral interaction (DI-RSU), and a quaternion Transformer with dynamic selector (DynQ-Former). The MSDEC-GFM enhances feature representation through multi-scale convolution and differential structures, while the gating mechanism fuses diverse fault information to strengthen sensitivity and interpretability to subtle fault variations. The DI-RSU collaboratively suppresses redundancy across spatial and channel dimensions by embedding the physical priors of differential enhancement and integral denoising into the network, while achieving a balance between noise suppression and feature enhancement. The DynQ-Former combines quaternion algebra with a dynamic head routing mechanism to optimize attention allocation and simultaneously model local and global features. Experimental results demonstrate that the proposed method achieves superior performance on both the Case Western Reserve University (CWRU) and Nanjing Tech University (NanTech) bearing datasets, maintaining high accuracy and stability under strong noise conditions, while exhibiting a certain degree of interpretability.

轴承是确保机械设备安全高效运行的关键部件。然而，大多数现有的轴承故障诊断模型在强噪声条件下性能下降明显，故障特征容易被噪声淹没。为了解决这一问题，本文提出了一种鲁棒的局部可解释智能轴承故障诊断模型（MSDRQF），实现了噪声环境下的准确识别。该模型集成了一个多尺度差分增强卷积和门控融合模块（MSDEC-GFM）、一个具有微分-积分相互作用的二维冗余抑制单元（DI-RSU）和一个带有动态选择器的四元数变压器（DynQ-Former）。MSDEC-GFM通过多尺度卷积和差分结构增强特征表征，而门控机制融合了多种断层信息，增强了对细微断层变化的敏感性和可解释性。DI-RSU通过在网络中嵌入微分增强和积分去噪的物理先验，协同抑制跨空间和信道维度的冗余，同时实现噪声抑制和特征增强之间的平衡。DynQ-Former将四元数代数与动态头部路由机制相结合，优化注意力分配，同时模拟局部和全局特征。实验结果表明，该方法在凯斯西储大学（CWRU）和南京工业大学（NanTech）轴承数据集上都取得了优异的性能，在强噪声条件下保持了较高的精度和稳定性，同时具有一定的可解释性。

{"title":"Robust and interpretable bearing fault diagnosis via multi-scale differential-enhanced convolution with dual-dimensional redundancy suppression and dynamic quaternion transformer","authors":"Chuang Chen , Zefeng Wang , Jiantao Shi , Dongdong Yue , Dan Bao , Ge Shi , Cunsong Wang , Tao Xie","doi":"10.1016/j.aei.2026.104397","DOIUrl":"10.1016/j.aei.2026.104397","url":null,"abstract":"<div><div>Bearings are critical components that ensure the safe and efficient operation of mechanical equipment. However, most existing bearing fault diagnosis models suffer significant performance degradation under strong noise conditions, where fault features are easily overwhelmed by noise. To address this issue, this paper proposes a robust and locally interpretable intelligent bearing fault diagnosis model (named MSDRQF), achieving accurate identification in noisy environments. The proposed model integrates a multi-scale differential enhancement convolution and gated fusion module (MSDEC-GFM), a dual-dimensional redundancy suppression unit with differential-integral interaction (DI-RSU), and a quaternion Transformer with dynamic selector (DynQ-Former). The MSDEC-GFM enhances feature representation through multi-scale convolution and differential structures, while the gating mechanism fuses diverse fault information to strengthen sensitivity and interpretability to subtle fault variations. The DI-RSU collaboratively suppresses redundancy across spatial and channel dimensions by embedding the physical priors of differential enhancement and integral denoising into the network, while achieving a balance between noise suppression and feature enhancement. The DynQ-Former combines quaternion algebra with a dynamic head routing mechanism to optimize attention allocation and simultaneously model local and global features. Experimental results demonstrate that the proposed method achieves superior performance on both the Case Western Reserve University (CWRU) and Nanjing Tech University (NanTech) bearing datasets, maintaining high accuracy and stability under strong noise conditions, while exhibiting a certain degree of interpretability.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"71 ","pages":"Article 104397"},"PeriodicalIF":9.9,"publicationDate":"2026-01-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146077886","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}

引用次数: 0

Spatiotemporal dynamic modeling approach for distributed thermal processes under digital twin framework 数字孪生框架下分布式热过程的时空动态建模方法

IF 9.9 1区工程技术 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Advanced Engineering Informatics

Pub Date : 2026-01-30 DOI: 10.1016/j.aei.2026.104384

Tianyue Wang , Han-Xiong Li , Lihui Wang , Xi Vincent Wang

Distributed thermal processes are prevalent in a wide range of manufacturing systems. Developing accurate spatiotemporal models for these manufacturing thermal processes is crucial for understanding their mechanisms and enabling effective process monitoring. In recent years, digital twin-based predictive models have emerged as an important approach for enhancing smart manufacturing systems. However, the non-Gaussian distribution of spatiotemporal data caused by non-stationary working conditions poses a challenge to traditional spatiotemporal models. Also, traditional models need to be repeatedly retrained from scratch with all historical data, which leads to substantial calculation. In this paper, a digital twin-based spatiotemporal dynamic modeling framework is proposed for distributed thermal processes in manufacturing. Concretely, a spectral-based physical explainable spatiotemporal model is first conducted to reflect the manufacturing thermal process. Then, a time/space separation-based data basic model is developed to supplement the deficiencies of the physical model. Finally, a dynamic spatiotemporal incremental learning scheme is designed to handle the non-stationary working condition of thermal processes, thereby strengthening the model adaptability. Experiments conducted on the curing oven thermal process confirm the effectiveness of the proposed dynamic modeling approach. The corresponding comparative and ablation experiments confirm the superiority of the proposed module.

分布式热过程在广泛的制造系统中普遍存在。为这些制造热过程开发准确的时空模型对于理解其机制和实现有效的过程监测至关重要。近年来，基于数字孪生的预测模型已成为增强智能制造系统的重要方法。然而，非平稳工作条件导致的时空数据的非高斯分布对传统的时空模型提出了挑战。此外，传统模型需要使用所有历史数据从零开始进行反复的再训练，这导致了大量的计算。本文提出了一种基于数字孪生的制造业分布式热过程时空动态建模框架。具体而言，首先建立了一个基于光谱的物理可解释时空模型来反映制造热过程。在此基础上，建立了基于时/空分离的数据基础模型，弥补了物理模型的不足。最后，设计了一种动态的时空增量学习方案来处理热过程的非平稳工况，从而增强了模型的适应性。对固化炉热过程的实验验证了所提出的动态建模方法的有效性。相应的对比和烧蚀实验证实了该模块的优越性。

{"title":"Spatiotemporal dynamic modeling approach for distributed thermal processes under digital twin framework","authors":"Tianyue Wang , Han-Xiong Li , Lihui Wang , Xi Vincent Wang","doi":"10.1016/j.aei.2026.104384","DOIUrl":"10.1016/j.aei.2026.104384","url":null,"abstract":"<div><div>Distributed thermal processes are prevalent in a wide range of manufacturing systems. Developing accurate spatiotemporal models for these manufacturing thermal processes is crucial for understanding their mechanisms and enabling effective process monitoring. In recent years, digital twin-based predictive models have emerged as an important approach for enhancing smart manufacturing systems. However, the non-Gaussian distribution of spatiotemporal data caused by non-stationary working conditions poses a challenge to traditional spatiotemporal models. Also, traditional models need to be repeatedly retrained from scratch with all historical data, which leads to substantial calculation. In this paper, a digital twin-based spatiotemporal dynamic modeling framework is proposed for distributed thermal processes in manufacturing. Concretely, a spectral-based physical explainable spatiotemporal model is first conducted to reflect the manufacturing thermal process. Then, a time/space separation-based data basic model is developed to supplement the deficiencies of the physical model. Finally, a dynamic spatiotemporal incremental learning scheme is designed to handle the non-stationary working condition of thermal processes, thereby strengthening the model adaptability. Experiments conducted on the curing oven thermal process confirm the effectiveness of the proposed dynamic modeling approach. The corresponding comparative and ablation experiments confirm the superiority of the proposed module.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"71 ","pages":"Article 104384"},"PeriodicalIF":9.9,"publicationDate":"2026-01-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146077878","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}

引用次数: 0

Vision–proprioception fusion with Mamba2 in end-to-end reinforcement learning for motion control 基于Mamba2的视觉本体感觉融合在端到端运动控制强化学习中的应用

IF 9.9 1区工程技术 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Advanced Engineering Informatics

Pub Date : 2026-01-29 DOI: 10.1016/j.aei.2026.104389

Xiaowen Tao , Yinuo Wang , Jinzhao Zhou

End-to-end reinforcement learning (RL) for motion control trains policies directly from sensor inputs to motor commands, enabling unified controllers for different robots and tasks. However, most existing methods are either blind (proprioception-only) or rely on fusion backbones with unfavorable compute–memory trade-offs. Recurrent controllers struggle with long-horizon credit assignment, and Transformer-based fusion incurs quadratic cost in token length, limiting temporal and spatial context. We present a vision-driven cross-modal RL framework built on SSD-Mamba2, a selective state–space backbone that applies state–space duality (SSD) to enable both recurrent and convolutional scanning with hardware-aware streaming and near-linear scaling. Proprioceptive states and exteroceptive observations (e.g., depth tokens) are encoded into compact tokens and fused by stacked SSD-Mamba2 layers. The selective state–space updates retain long-range dependencies with markedly lower latency and memory use than quadratic self-attention, enabling longer look-ahead, higher token resolution, and stable training under limited compute. Policies are trained end-to-end under curricula that randomize terrain and appearance and progressively increase scene complexity. A compact, state-centric reward balances task progress, energy efficiency, and safety. Across diverse motion-control scenarios, our approach consistently surpasses strong state-of-the-art baselines in return, safety (collisions and falls), and sample efficiency, while converging faster at the same compute budget. These results suggest that SSD-Mamba2 provides a practical fusion backbone for resource-constrained robotic and autonomous systems in engineering informatics applications.

用于运动控制的端到端强化学习（RL）直接从传感器输入到电机命令训练策略，为不同的机器人和任务实现统一的控制器。然而，大多数现有的方法要么是盲目的（只有本体感觉），要么依赖于融合主干，不利于计算内存的权衡。循环控制器与长期信用分配斗争，基于变压器的融合在令牌长度上产生二次成本，限制了时间和空间背景。我们提出了一个基于SSD- mamba2的视觉驱动的跨模态RL框架，这是一个选择性的状态空间主干，它应用状态空间对偶性（SSD）来实现循环扫描和卷积扫描，具有硬件感知流和近线性缩放。本体感觉状态和外感受观察（如深度标记）被编码成紧凑的标记，并通过堆叠的SSD-Mamba2层融合。选择性状态空间更新保留了长期依赖关系，比二次型自关注具有明显更低的延迟和内存使用，支持更长的前瞻性、更高的令牌分辨率和有限计算下的稳定训练。策略在随机化地形和外观并逐渐增加场景复杂性的课程中进行端到端训练。紧凑的、以国家为中心的奖励平衡了任务进度、能源效率和安全。在不同的运动控制场景中，我们的方法在回报、安全性（碰撞和坠落）和样本效率方面始终超过最先进的基线，同时在相同的计算预算下收敛得更快。这些结果表明，SSD-Mamba2为工程信息学应用中资源受限的机器人和自主系统提供了实用的融合骨干。

{"title":"Vision–proprioception fusion with Mamba2 in end-to-end reinforcement learning for motion control","authors":"Xiaowen Tao , Yinuo Wang , Jinzhao Zhou","doi":"10.1016/j.aei.2026.104389","DOIUrl":"10.1016/j.aei.2026.104389","url":null,"abstract":"<div><div>End-to-end reinforcement learning (RL) for motion control trains policies directly from sensor inputs to motor commands, enabling unified controllers for different robots and tasks. However, most existing methods are either <em>blind</em> (proprioception-only) or rely on fusion backbones with unfavorable compute–memory trade-offs. Recurrent controllers struggle with long-horizon credit assignment, and Transformer-based fusion incurs quadratic cost in token length, limiting temporal and spatial context. We present a vision-driven cross-modal RL framework built on <em>SSD-Mamba2</em>, a selective state–space backbone that applies <em>state–space duality (SSD)</em> to enable both recurrent and convolutional scanning with hardware-aware streaming and near-linear scaling. Proprioceptive states and exteroceptive observations (e.g., depth tokens) are encoded into compact tokens and fused by stacked SSD-Mamba2 layers. The selective state–space updates retain long-range dependencies with markedly lower latency and memory use than quadratic self-attention, enabling longer look-ahead, higher token resolution, and stable training under limited compute. Policies are trained end-to-end under curricula that randomize terrain and appearance and progressively increase scene complexity. A compact, state-centric reward balances task progress, energy efficiency, and safety. Across diverse motion-control scenarios, our approach consistently surpasses strong state-of-the-art baselines in return, safety (collisions and falls), and sample efficiency, while converging faster at the same compute budget. These results suggest that SSD-Mamba2 provides a practical fusion backbone for resource-constrained robotic and autonomous systems in engineering informatics applications.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"71 ","pages":"Article 104389"},"PeriodicalIF":9.9,"publicationDate":"2026-01-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146077879","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}

引用次数: 0

Game-theoretic motion planning method for dynamic interactive vehicles based on intent-awareness and multimodal trajectory prediction 基于意图感知和多模态轨迹预测的动态交互车辆博弈运动规划方法

IF 9.9 1区工程技术 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Advanced Engineering Informatics

Pub Date : 2026-01-29 DOI: 10.1016/j.aei.2026.104383

Weihe Liang , Wenhe Cao , Chunyan Wang, Ziyu Zhang, Wenbin Zhang, Wanzhong Zhao

At unsignalized intersections, the inherent high uncertainty of vehicle movements and the strong coupling of dynamic interaction mechanisms often lead to a decline in traffic throughput efficiency and a sharp increase in safety risks. To address these challenges, this paper proposes a novel, highly integrated game-theoretic motion-planning method for dynamic interactive vehicles based on intent awareness and multimodal trajectory prediction. To mitigate the adverse effects of surrounding vehicle motion uncertainty on ego vehicle’s planning, the proposed method first employs a dynamic interaction graph combined with a graph-attention mechanism and an LSTM network to accurately identify the lateral and longitudinal driving intentions of nearby vehicles. Based on the identified driving intentions, a collaborative perception-based multimodal trajectory prediction method is proposed, which integrates driving risk fields and spatiotemporal attention mechanisms to decouple and generate clusters of multimodal feasible trajectories for surrounding vehicles, thereby significantly reducing their motion uncertainty. Finally, the predicted intentions and multimodal trajectories of surrounding vehicles are used to construct an interactive action space between the autonomous vehicle and other vehicles, within which right-of-way conflicts at unsignalized intersections are resolved in real time through Nash equilibrium analysis. Simulation results across various scenarios, together with hardware-in-the-loop experiments and a small-scale real-vehicle experiment, demonstrate that the proposed method effectively enhances traffic efficiency and safety for autonomous vehicles at unsignalized intersections.

在无信号交叉口，固有的车辆运动的高度不确定性和动态相互作用机制的强耦合往往导致交通吞吐量效率下降和安全风险急剧增加。为了解决这些问题，本文提出了一种基于意图感知和多模态轨迹预测的新型、高度集成的博弈论动态交互车辆运动规划方法。为了减轻周围车辆运动不确定性对自我车辆规划的不利影响，该方法首先采用结合图-注意机制和LSTM网络的动态交互图来准确识别附近车辆的横向和纵向驾驶意图。基于识别出的驾驶意图，提出了一种基于协同感知的多模态轨迹预测方法，该方法将驾驶风险场与时空注意机制相结合，对周围车辆解耦并生成多模态可行轨迹簇，从而显著降低了其运动不确定性。最后，利用预测的意图和周围车辆的多模态轨迹，构建自动驾驶车辆与其他车辆之间的交互动作空间，通过纳什均衡分析实时解决无信号交叉口的路权冲突。各种场景仿真、硬件在环实验和小规模实车实验结果表明，该方法有效提高了自动驾驶车辆在无信号交叉口的交通效率和安全性。

{"title":"Game-theoretic motion planning method for dynamic interactive vehicles based on intent-awareness and multimodal trajectory prediction","authors":"Weihe Liang , Wenhe Cao , Chunyan Wang, Ziyu Zhang, Wenbin Zhang, Wanzhong Zhao","doi":"10.1016/j.aei.2026.104383","DOIUrl":"10.1016/j.aei.2026.104383","url":null,"abstract":"<div><div>At unsignalized intersections, the inherent high uncertainty of vehicle movements and the strong coupling of dynamic interaction mechanisms often lead to a decline in traffic throughput efficiency and a sharp increase in safety risks. To address these challenges, this paper proposes a novel, highly integrated game-theoretic motion-planning method for dynamic interactive vehicles based on intent awareness and multimodal trajectory prediction. To mitigate the adverse effects of surrounding vehicle motion uncertainty on ego vehicle’s planning, the proposed method first employs a dynamic interaction graph combined with a graph-attention mechanism and an LSTM network to accurately identify the lateral and longitudinal driving intentions of nearby vehicles. Based on the identified driving intentions, a collaborative perception-based multimodal trajectory prediction method is proposed, which integrates driving risk fields and spatiotemporal attention mechanisms to decouple and generate clusters of multimodal feasible trajectories for surrounding vehicles, thereby significantly reducing their motion uncertainty. Finally, the predicted intentions and multimodal trajectories of surrounding vehicles are used to construct an interactive action space between the autonomous vehicle and other vehicles, within which right-of-way conflicts at unsignalized intersections are resolved in real time through Nash equilibrium analysis. Simulation results across various scenarios, together with hardware-in-the-loop experiments and a small-scale real-vehicle experiment, demonstrate that the proposed method effectively enhances traffic efficiency and safety for autonomous vehicles at unsignalized intersections.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"71 ","pages":"Article 104383"},"PeriodicalIF":9.9,"publicationDate":"2026-01-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146077881","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}

引用次数: 0

An intelligent fault diagnosis approach for construction machinery hydraulic systems based on knowledge graph 基于知识图的工程机械液压系统智能故障诊断方法

IF 9.9 1区工程技术 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Advanced Engineering Informatics

Pub Date : 2026-01-29 DOI: 10.1016/j.aei.2026.104360

Yuming Xu , Tao Peng , Ray Y. Zhong , Kendrik Lim

The safety risks of construction machinery are crucial for construction, most of which stem from hydraulic system faults. However, the knowledge required for such fault diagnosis is extremely complex, besides current knowledge-based methods heavily rely on engineering experience, which are time-consuming and error-prone. To tackle this challenge, we propose an intelligent fault diagnosis approach for construction machinery hydraulic systems based on knowledge graph, incorperating emerging technologies, active learning, pre-trained language models, and graph neural networks, to significantly enhance the performance of knowledge graph. Firstly, active learning-based BERT + Bi-LSTM + CRF fault knowledge extraction was developed to address the challenge of obtaining a large and manually-labeled corpus. Secondly, word embedding-based fault entity alignment resolves the issue of discrepant expression of the same or similar fault knowledge. Thirdly, knowledge graph embedding-based fault knowledge reasoning tackles the omissions and errors in complex, multi-hop causation chains. Finally, word embedding-based fault knowledge retrieval addresses the inconsistent user inputs. The proposed methodology extends fault coverage and enhances the completeness and accuracy of diagnostic results as compared to existing fault diagnosis approaches. The improved performance is validated via a case study featuring a concrete spreader.

工程机械的安全隐患对工程施工至关重要，其中大部分源于液压系统故障。然而，这种故障诊断所需的知识极其复杂，而且现有的基于知识的方法严重依赖工程经验，耗时长且容易出错。为了解决这一问题，我们提出了一种基于知识图的工程机械液压系统智能故障诊断方法，结合新兴技术、主动学习、预训练语言模型和图神经网络，显著提高了知识图的性能。首先，提出了基于主动学习的BERT + Bi-LSTM + CRF故障知识提取方法，解决了获取大型人工标注语料库的难题；其次，基于词嵌入的故障实体对齐解决了相同或相似故障知识表达不一致的问题。第三，基于知识图嵌入的故障知识推理解决了复杂的多跳因果链中的遗漏和错误。最后，基于词嵌入的故障知识检索解决了用户输入不一致的问题。与现有的故障诊断方法相比，该方法扩展了故障覆盖范围，提高了诊断结果的完整性和准确性。通过混凝土摊铺机的案例研究验证了改进后的性能。

{"title":"An intelligent fault diagnosis approach for construction machinery hydraulic systems based on knowledge graph","authors":"Yuming Xu , Tao Peng , Ray Y. Zhong , Kendrik Lim","doi":"10.1016/j.aei.2026.104360","DOIUrl":"10.1016/j.aei.2026.104360","url":null,"abstract":"<div><div>The safety risks of construction machinery are crucial for construction, most of which stem from hydraulic system faults. However, the knowledge required for such fault diagnosis is extremely complex, besides current knowledge-based methods heavily rely on engineering experience, which are time-consuming and error-prone. To tackle this challenge, we propose an intelligent fault diagnosis approach for construction machinery hydraulic systems based on knowledge graph, incorperating emerging technologies, active learning, pre-trained language models, and graph neural networks, to significantly enhance the performance of knowledge graph. Firstly, active learning-based BERT + Bi-LSTM + CRF fault knowledge extraction was developed to address the challenge of obtaining a large and manually-labeled corpus. Secondly, word embedding-based fault entity alignment resolves the issue of discrepant expression of the same or similar fault knowledge. Thirdly, knowledge graph embedding-based fault knowledge reasoning tackles the omissions and errors in complex, multi-hop causation chains. Finally, word embedding-based fault knowledge retrieval addresses the inconsistent user inputs. The proposed methodology extends fault coverage and enhances the completeness and accuracy of diagnostic results as compared to existing fault diagnosis approaches. The improved performance is validated via a case study featuring a concrete spreader.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"71 ","pages":"Article 104360"},"PeriodicalIF":9.9,"publicationDate":"2026-01-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146077887","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}

引用次数: 0

Diagnosis for transformer winding fault via deep fusion of 2-D images of FRA and SwinLCGnet framework 基于FRA和SwinLCGnet框架的二维图像深度融合诊断变压器绕组故障

IF 9.9 1区工程技术 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Advanced Engineering Informatics

Pub Date : 2026-01-29 DOI: 10.1016/j.aei.2026.104361

Anzhi Fu , Dongyang Wang , Hongbo Wang , Hao Chen , Yusong Zheng , Guochao Qian , Lijun Zhou

Winding fault detection is essential to ensure the safe and reliable operation of power transformers. This paper conducts a diagnostic study on winding fault type identification, degree assessment, and spatial localization using frequency response analysis (FRA) test results, a widely used engineering test method. Firstly, we propose the Gramian angular field with frequency-proximity guidance (GAF_f) method, which converts 1-D FRA curves into 2-D images by comprehensively exploiting global and local signal features, thereby achieving improved visualization of winding fault characteristics. Thereafter, the task-customized mixture of adapters (TC-MoA) is applied to fuse GASF_f and GADF_f feature images for the first time, effectively combining the advantages of dual coding methods to enhance multimodal complementary features. Finally, a two-branch diagnostic framework named SwinLCGnet is proposed to conduct collaborative extraction and modeling of global–local features to accomplish three winding diagnostic tasks: fault type classification, degree quantification, and spatial localization. Experimental data and field cases show that the proposed diagnostic framework can effectively perform all winding fault diagnostic tasks, with diagnostic accuracy and F1 score of up to 98.08% or more, which is a significant advantage compared with other mainstream frameworks. This innovative framework enables highly accurate identification of winding faults, contributes to enhanced intelligent diagnostic capabilities of power transformers, and offers an efficient and automated diagnostic approach to help maintain the reliability of power systems.

绕组故障检测是保证电力变压器安全可靠运行的关键。本文利用广泛应用的工程试验方法频响分析（FRA）试验结果，对绕组故障类型识别、程度评估和空间定位进行诊断研究。首先，我们提出了Gramian角场频率邻近制导（GAFf）方法，该方法综合利用全局和局部信号特征，将1-D FRA曲线转换为2-D图像，从而实现了绕组故障特征的可视化。随后，首次采用任务定制混合适配器（TC-MoA）对GASFf和GADFf特征图像进行融合，有效结合了双编码方法的优势，增强了多模态互补特征。最后，提出了SwinLCGnet两分支诊断框架，对全局-局部特征进行协同提取和建模，完成故障类型分类、程度量化和空间定位三个绕组诊断任务。实验数据和现场案例表明，所提出的诊断框架能够有效地完成所有绕组故障诊断任务，诊断准确率和F1分数高达98.08%以上，与其他主流诊断框架相比具有显著优势。这种创新的框架能够高度准确地识别绕组故障，有助于增强电力变压器的智能诊断能力，并提供有效和自动化的诊断方法，以帮助保持电力系统的可靠性。

{"title":"Diagnosis for transformer winding fault via deep fusion of 2-D images of FRA and SwinLCGnet framework","authors":"Anzhi Fu , Dongyang Wang , Hongbo Wang , Hao Chen , Yusong Zheng , Guochao Qian , Lijun Zhou","doi":"10.1016/j.aei.2026.104361","DOIUrl":"10.1016/j.aei.2026.104361","url":null,"abstract":"<div><div>Winding fault detection is essential to ensure the safe and reliable operation of power transformers. This paper conducts a diagnostic study on winding fault type identification, degree assessment, and spatial localization using frequency response analysis (FRA) test results, a widely used engineering test method. Firstly, we propose the Gramian angular field with frequency-proximity guidance (GAF<sub>f</sub>) method, which converts 1-D FRA curves into 2-D images by comprehensively exploiting global and local signal features, thereby achieving improved visualization of winding fault characteristics. Thereafter, the task-customized mixture of adapters (TC-MoA) is applied to fuse GASF<sub>f</sub> and GADF<sub>f</sub> feature images for the first time, effectively combining the advantages of dual coding methods to enhance multimodal complementary features. Finally, a two-branch diagnostic framework named SwinLCGnet is proposed to conduct collaborative extraction and modeling of global–local features to accomplish three winding diagnostic tasks: fault type classification, degree quantification, and spatial localization. Experimental data and field cases show that the proposed diagnostic framework can effectively perform all winding fault diagnostic tasks, with diagnostic accuracy and F1 score of up to 98.08% or more, which is a significant advantage compared with other mainstream frameworks. This innovative framework enables highly accurate identification of winding faults, contributes to enhanced intelligent diagnostic capabilities of power transformers, and offers an efficient and automated diagnostic approach to help maintain the reliability of power systems.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"71 ","pages":"Article 104361"},"PeriodicalIF":9.9,"publicationDate":"2026-01-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146077880","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}

引用次数: 0

PE-MPINN: A parameters-enhanced multiphysics-informed neural network for data assimilation of seepage-consolidation coupling problems in spatially variable soils PE-MPINN：用于空间变土渗流-固结耦合问题数据同化的参数增强多物理场信息神经网络

IF 9.9 1区工程技术 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Advanced Engineering Informatics

Pub Date : 2026-01-29 DOI: 10.1016/j.aei.2026.104376

Mingyue Sun , Gang Ma , Tongming Qu , Shaoheng Guan , Jiangzhou Mei , Jingzhou Wang , Wei Zhou

Physics-Informed Neural Networks (PINNs) have been widely applied for solving inverse problems due to the powerful capability of integrating physical laws with data-driven learning. However, in the context of multi-field coupling processes in spatially variable soils, conventional PINNs still struggle to explicitly identify uncertain parameters and remain underexplored in engineering-scale applications. To overcome these limitations, this study proposes a Parameters-Enhanced Multiphysics-Informed Neural Network (PE-MPINN) framework for data assimilation of seepage-consolidation problems in spatially variable soils. The framework integrates three-dimensional consolidation theory with random fields, using trainable Karhunen-Loève Expansion vectors to infer heterogeneous soil parameters. Furthermore, a parameters-enhanced subnetwork is introduced to iteratively refine these vectors to continuously improve the representation of soil variability. The proposed approach is validated on a core wall rockfill dam. Results show that PE-MPINN successfully assimilates monitoring and testing data, and accurately predicts pore water pressure, earth pressure, hydraulic conductivity, and compression modulus at engineering scales. Moreover, PE-MPINN demonstrates strong robustness under sparse, noisy data and varying heterogeneity conditions, while achieving superior spatiotemporal extrapolation accuracy. This study highlights the value of integrating physical knowledge with data assimilation by offering a novel, efficient framework for real-time seepage-consolidation analysis and geotechnical digital twin applications.

基于物理信息的神经网络（pinn）由于将物理定律与数据驱动学习相结合的强大能力，在求解逆问题方面得到了广泛的应用。然而，在空间可变土壤中多场耦合过程的背景下，传统的PINNs仍然难以明确识别不确定参数，并且在工程规模应用中仍未得到充分的探索。为了克服这些限制，本研究提出了一个参数增强的多物理场信息神经网络（PE-MPINN）框架，用于空间可变土壤的渗流固压问题的数据同化。该框架将三维固结理论与随机场相结合，利用可训练的karhunen - lo扩展向量来推断非均质土壤参数。在此基础上，引入参数增强子网络对这些向量进行迭代细化，不断提高土壤变异性的表征。该方法在某心墙堆石坝上得到了验证。结果表明，PE-MPINN能有效地同化监测和测试数据，并能准确预测工程尺度下的孔隙水压力、土压力、导水率和压缩模量。此外，PE-MPINN在稀疏、噪声数据和变化异质性条件下表现出较强的鲁棒性，同时具有优异的时空外推精度。本研究通过为实时渗流固结分析和岩土工程数字孪生应用提供一个新颖、高效的框架，强调了将物理知识与数据同化相结合的价值。

{"title":"PE-MPINN: A parameters-enhanced multiphysics-informed neural network for data assimilation of seepage-consolidation coupling problems in spatially variable soils","authors":"Mingyue Sun , Gang Ma , Tongming Qu , Shaoheng Guan , Jiangzhou Mei , Jingzhou Wang , Wei Zhou","doi":"10.1016/j.aei.2026.104376","DOIUrl":"10.1016/j.aei.2026.104376","url":null,"abstract":"<div><div>Physics-Informed Neural Networks (PINNs) have been widely applied for solving inverse problems due to the powerful capability of integrating physical laws with data-driven learning. However, in the context of multi-field coupling processes in spatially variable soils, conventional PINNs still struggle to explicitly identify uncertain parameters and remain underexplored in engineering-scale applications. To overcome these limitations, this study proposes a Parameters-Enhanced Multiphysics-Informed Neural Network (PE-MPINN) framework for data assimilation of seepage-consolidation problems in spatially variable soils. The framework integrates three-dimensional consolidation theory with random fields, using trainable Karhunen-Loève Expansion vectors to infer heterogeneous soil parameters. Furthermore, a parameters-enhanced subnetwork is introduced to iteratively refine these vectors to continuously improve the representation of soil variability. The proposed approach is validated on a core wall rockfill dam. Results show that PE-MPINN successfully assimilates monitoring and testing data, and accurately predicts pore water pressure, earth pressure, hydraulic conductivity, and compression modulus at engineering scales. Moreover, PE-MPINN demonstrates strong robustness under sparse, noisy data and varying heterogeneity conditions, while achieving superior spatiotemporal extrapolation accuracy. This study highlights the value of integrating physical knowledge with data assimilation by offering a novel, efficient framework for real-time seepage-consolidation analysis and geotechnical digital twin applications.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"71 ","pages":"Article 104376"},"PeriodicalIF":9.9,"publicationDate":"2026-01-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146077883","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}

引用次数: 0

Socio-technical assessment of generative AI integration in architecture, engineering, and construction (AEC) workflows: An empirical study using O*NET occupational taxonomy 建筑、工程和施工（AEC）工作流程中生成式AI集成的社会技术评估：使用O*NET职业分类的实证研究

IF 9.9 1区工程技术 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Advanced Engineering Informatics

Pub Date : 2026-01-28 DOI: 10.1016/j.aei.2026.104392

Ruoxin Xiong , Yael Netser , Pingbo Tang , Beibei Li , Joonsun Hwang

Generative artificial intelligence (GAI) has the potential to reshape workflows across the Architecture, Engineering, and Construction (AEC) sector. While previous research has offered valuable technical demonstrations and conceptual analyses, empirical evidence quantifying GAI-related impacts across AEC occupations and systematic assessment of adoption readiness remain limited. This study develops a domain-specific socio-technical evaluation framework that provides occupational-level analysis of technical capabilities, social risks, and adoption barriers across thirteen O*NET-defined AEC occupations. Data were collected through a six-month survey of 162 AEC professionals, complemented by six expert interviews and a systematic literature review. The findings reveal: (1) Technical Capability, measured using exposure scores ranging from −1 (low applicability) to +1 (high applicability), shows moderate applicability in design-oriented roles (e.g., architectural drafters: 0.16) and minimal alignment for site-based and manual activities (e.g., construction laborers: −0.89). (2) Social Risks, assessed on a 0–1 scale of concern, identify hallucinations (0.71), data privacy (0.70), and intellectual property issues (0.69) as critical concerns. (3) Socio-Technical Adoption highlights limited technical expertise (26.0%) and uncertain return on investment (16.8%) as primary barriers, while respondents emphasized the need for usage guidelines and standards (29.6%) and targeted training (29.2%) to facilitate responsible integration. Based on these findings, the study outlines strategic priorities for responsible GAI deployment, including AEC-specific standards, targeted workforce training, human-in-the-loop validation mechanisms, and domain-tailored digital infrastructure. The framework and empirical evidence provide a foundation for researchers, practitioners, and policymakers seeking to guide the safe and effective integration of GAI into AEC workflows.

生成式人工智能（GAI）有可能重塑建筑、工程和施工（AEC）领域的工作流程。虽然以前的研究提供了有价值的技术演示和概念分析，但量化人工智能对AEC职业的影响和系统评估采用准备程度的经验证据仍然有限。本研究开发了一个特定领域的社会技术评估框架，提供了13个O* net定义的AEC职业的技术能力、社会风险和采用障碍的职业层面分析。通过对162名AEC专业人员进行为期6个月的调查，并辅以6次专家访谈和系统的文献综述，收集了数据。研究结果显示：(1)技术能力，使用从−1（低适用性）到+1（高适用性）的暴露分数进行测量，显示出在面向设计的角色（例如，建筑起草人：0.16）中的中等适用性，以及对基于现场和手工活动（例如，建筑工人：- 0.89）的最小一致性。(2)社会风险，以0-1的关注程度进行评估，认为幻觉（0.71）、数据隐私（0.70）和知识产权问题（0.69）是关键问题。(3)社会技术采用强调有限的技术专业知识（26.0%）和不确定的投资回报（16.8%）是主要障碍，而受访者强调需要使用指南和标准（29.6%）和有针对性的培训（29.2%）来促进负责任的整合。基于这些发现，该研究概述了负责任的GAI部署的战略重点，包括aec特定标准、有针对性的劳动力培训、人在环验证机制和领域定制的数字基础设施。该框架和经验证据为研究人员、从业者和政策制定者寻求指导GAI安全有效地整合到AEC工作流程中提供了基础。

{"title":"Socio-technical assessment of generative AI integration in architecture, engineering, and construction (AEC) workflows: An empirical study using O*NET occupational taxonomy","authors":"Ruoxin Xiong , Yael Netser , Pingbo Tang , Beibei Li , Joonsun Hwang","doi":"10.1016/j.aei.2026.104392","DOIUrl":"10.1016/j.aei.2026.104392","url":null,"abstract":"<div><div>Generative artificial intelligence (GAI) has the potential to reshape workflows across the Architecture, Engineering, and Construction (AEC) sector. While previous research has offered valuable technical demonstrations and conceptual analyses, empirical evidence quantifying GAI-related impacts across AEC occupations and systematic assessment of adoption readiness remain limited. This study develops a domain-specific socio-technical evaluation framework that provides occupational-level analysis of technical capabilities, social risks, and adoption barriers across thirteen O*NET-defined AEC occupations. Data were collected through a six-month survey of 162 AEC professionals, complemented by six expert interviews and a systematic literature review. The findings reveal: (1) <em>Technical Capability</em>, measured using exposure scores ranging from −1 (low applicability) to +1 (high applicability), shows moderate applicability in design-oriented roles (e.g., architectural drafters: 0.16) and minimal alignment for site-based and manual activities (e.g., construction laborers: −0.89). (2) <em>Social Risks</em>, assessed on a 0–1 scale of concern, identify hallucinations (0.71), data privacy (0.70), and intellectual property issues (0.69) as critical concerns. (3) <em>Socio-Technical Adoption</em> highlights limited technical expertise (26.0%) and uncertain return on investment (16.8%) as primary barriers, while respondents emphasized the need for usage guidelines and standards (29.6%) and targeted training (29.2%) to facilitate responsible integration. Based on these findings, the study outlines strategic priorities for responsible GAI deployment, including AEC-specific standards, targeted workforce training, human-in-the-loop validation mechanisms, and domain-tailored digital infrastructure. The framework and empirical evidence provide a foundation for researchers, practitioners, and policymakers seeking to guide the safe and effective integration of GAI into AEC workflows.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"71 ","pages":"Article 104392"},"PeriodicalIF":9.9,"publicationDate":"2026-01-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146048981","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}

引用次数: 0

Explainable topological-disordered 3D lattice model for hydration-temperature coupled fields of early-age concrete 早期混凝土水化-温度耦合场的可解释拓扑无序三维点阵模型

IF 9.9 1区工程技术 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Advanced Engineering Informatics

Pub Date : 2026-01-28 DOI: 10.1016/j.aei.2026.104358

Shujie Han, Xiao Zhang, Xuejiang Lan, Xiaohui Chang

The inherent structural disorder in early-age concrete significantly impacts computational efficiency and simulation accuracy in three-dimensional (3D) lattice modeling, while modeling parameters may demonstrate non-negligible effects on numerical outcomes. To achieve high-fidelity simulation of coupled hydration-temperature fields, this study proposes an automated modeling framework for topological-disordered 3D lattice structures. Parametric studies revealed that randomness and grid resolution critically influence bulk thermal conductivity. Optimal parameters were systematically determined through systematic simulation results and mechanistic interpretation of elemental heat flux distributions using automated algorithms. A macroscopic statistical relationship between modeling parameters and effective thermal conductivity of the lattice system was established by integrating stochastic analysis with the thermal resistance grid method. Validation demonstrated that this approach effectively captures modeling parametric effects on thermal conduction behavior. Finally, the proposed 3D lattice model’s capability to simulate coupled hydration-thermal fields of early-age concrete was rigorously verified against published experimental datasets under two environmental conditions. Explainable machine learning(ML) analysis was conducted on early-age temperature field simulation results of multiple concrete mixtures to investigate the influential significance of various raw material parameters on early-age temperature development. Furthermore, the parametric analysis framework and statistical methodology offer optimized solutions for disorder lattice modeling in cementitious materials, improving both computational efficiency and physical accuracy.

早期混凝土固有的结构无序性显著影响三维（3D）点阵建模的计算效率和模拟精度，而建模参数可能对数值结果产生不可忽视的影响。为了实现水化-温度耦合场的高保真仿真，本研究提出了一种拓扑无序三维晶格结构的自动建模框架。参数研究表明，随机性和网格分辨率对体导热系数有重要影响。通过系统模拟结果和自动化算法对元素热流密度分布的机理解释，系统地确定了最优参数。将随机分析与热阻网格法相结合，建立了模型参数与晶格系统有效导热系数之间的宏观统计关系。验证表明，该方法有效地捕获了建模参数对热传导行为的影响。最后，根据已发表的实验数据集，在两种环境条件下严格验证了所提出的三维晶格模型模拟早期混凝土水化-热耦合场的能力。对多种混凝土混合料早期温度场模拟结果进行可解释性机器学习（ML）分析，探讨各种原材料参数对早期温度发展的影响意义。此外，参数分析框架和统计方法为胶凝材料的无序晶格建模提供了优化的解决方案，提高了计算效率和物理精度。

{"title":"Explainable topological-disordered 3D lattice model for hydration-temperature coupled fields of early-age concrete","authors":"Shujie Han, Xiao Zhang, Xuejiang Lan, Xiaohui Chang","doi":"10.1016/j.aei.2026.104358","DOIUrl":"10.1016/j.aei.2026.104358","url":null,"abstract":"<div><div>The inherent structural disorder in early-age concrete significantly impacts computational efficiency and simulation accuracy in three-dimensional (3D) lattice modeling, while modeling parameters may demonstrate non-negligible effects on numerical outcomes. To achieve high-fidelity simulation of coupled hydration-temperature fields, this study proposes an automated modeling framework for topological-disordered 3D lattice structures. Parametric studies revealed that randomness and grid resolution critically influence bulk thermal conductivity. Optimal parameters were systematically determined through systematic simulation results and mechanistic interpretation of elemental heat flux distributions using automated algorithms. A macroscopic statistical relationship between modeling parameters and effective thermal conductivity of the lattice system was established by integrating stochastic analysis with the thermal resistance grid method. Validation demonstrated that this approach effectively captures modeling parametric effects on thermal conduction behavior. Finally, the proposed 3D lattice model’s capability to simulate coupled hydration-thermal fields of early-age concrete was rigorously verified against published experimental datasets under two environmental conditions. Explainable machine learning(ML) analysis was conducted on early-age temperature field simulation results of multiple concrete mixtures to investigate the influential significance of various raw material parameters on early-age temperature development. Furthermore, the parametric analysis framework and statistical methodology offer optimized solutions for disorder lattice modeling in cementitious materials, improving both computational efficiency and physical accuracy.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"71 ","pages":"Article 104358"},"PeriodicalIF":9.9,"publicationDate":"2026-01-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146077877","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}

引用次数: 0

AQUADA-UNI: A super-efficient unified cascade multimodal model for thermographic wind turbine blade segmentation and damage detection AQUADA-UNI：一种用于热成像风力涡轮机叶片分割和损伤检测的超高效统一叶栅多模态模型

IF 9.9 1区工程技术 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Advanced Engineering Informatics

Pub Date : 2026-01-28 DOI: 10.1016/j.aei.2026.104387

Xiaodong Jia, Xiao Chen

Structural damage inspection of wind turbine rotor blades is crucial for preventing catastrophic failures. AI-driven automated approaches are increasingly being used in blade damage inspection due to their high efficiency and consistency. This paper proposes a novel AI-driven cascade multimodal model to perform thermographic blade segmentation and blade damage detection in one single step with exceptional computational efficiency. For blade segmentation, this model leverages RGB images to enhance thermographic blade segmentation, enabling multimodal complementarity and improving segmentation accuracy. For blade damage detection, the model employs hard example mining and curriculum learning to significantly enhance its ability to distinguish anomalous boundaries between normal and anomaly image patches. Utilizing a computationally efficient cascade network architecture, this model unifies conventional multi-step tasks into one single step for AI-based blade damage detection. Compared to the state-of-the-art method (i.e., the second-best method in our benchmark study), the proposed model halves the computational cost and nearly triples the Frames Per Second (FPS) while maintaining equivalent or slightly higher prediction accuracy, particularly in the test cases where the blade damage is small and its thermal contrast is low, making analysis with conventional AI-based methods extremely challenging. We also shared the dataset with the public to support further study.

风力发电机转子叶片结构损伤检测是防止灾难性失效的关键。人工智能驱动的自动化方法由于其高效率和一致性，越来越多地用于叶片损伤检测。本文提出了一种新的人工智能驱动的叶栅多模态模型，该模型可以一步完成热成像叶片分割和叶片损伤检测，并且具有优异的计算效率。对于叶片分割，该模型利用RGB图像增强热成像叶片分割，实现多模态互补，提高分割精度。在叶片损伤检测方面，该模型采用了硬例挖掘和课程学习的方法，显著增强了正常和异常图像斑块之间异常边界的区分能力。该模型利用计算效率高的级联网络架构，将传统的多步骤任务统一为一个步骤，用于基于人工智能的叶片损伤检测。与最先进的方法（即在我们的基准研究中排名第二的方法）相比，所提出的模型将计算成本减半，将每秒帧数（FPS）提高近三倍，同时保持同等或略高的预测精度，特别是在叶片损伤较小且热对比较低的测试用例中，这使得使用传统的基于人工智能的方法进行分析极具挑战性。我们还与公众分享了数据集，以支持进一步的研究。

{"title":"AQUADA-UNI: A super-efficient unified cascade multimodal model for thermographic wind turbine blade segmentation and damage detection","authors":"Xiaodong Jia, Xiao Chen","doi":"10.1016/j.aei.2026.104387","DOIUrl":"10.1016/j.aei.2026.104387","url":null,"abstract":"<div><div>Structural damage inspection of wind turbine rotor blades is crucial for preventing catastrophic failures. AI-driven automated approaches are increasingly being used in blade damage inspection due to their high efficiency and consistency. This paper proposes a novel AI-driven cascade multimodal model to perform thermographic blade segmentation and blade damage detection in one single step with exceptional computational efficiency. For blade segmentation, this model leverages RGB images to enhance thermographic blade segmentation, enabling multimodal complementarity and improving segmentation accuracy. For blade damage detection, the model employs hard example mining and curriculum learning to significantly enhance its ability to distinguish anomalous boundaries between normal and anomaly image patches. Utilizing a computationally efficient cascade network architecture, this model unifies conventional multi-step tasks into one single step for AI-based blade damage detection. Compared to the state-of-the-art method (i.e., the second-best method in our benchmark study), the proposed model halves the computational cost and nearly triples the Frames Per Second (FPS) while maintaining equivalent or slightly higher prediction accuracy, particularly in the test cases where the blade damage is small and its thermal contrast is low, making analysis with conventional AI-based methods extremely challenging. We also shared the dataset with the public to support further study.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"71 ","pages":"Article 104387"},"PeriodicalIF":9.9,"publicationDate":"2026-01-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146077882","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}

引用次数: 0