Advanced Engineering Informatics最新文献_第2页

RLLM-SS: A knowledge-guided simplex search method integrating large language model and reinforcement learning for injection molding quality control RLLM-SS：一种集成大语言模型和强化学习的知识引导单纯形搜索方法，用于注塑质量控制

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

Advanced Engineering Informatics

Pub Date : 2026-04-01 Epub Date: 2026-01-23 DOI: 10.1016/j.aei.2026.104372

Haipeng Zou , Xinyu Li , Yongkuan Yang , Ke Yao , Xiangsong Kong , Zhijiang Shao , Furong Gao

In injection molding (IM), product quality and process stability are highly dependent on the setting of key process parameters, making efficient parameter tuning essential for achieving reliable and consistent production. However, the tuning process is traditionally guided by expert experience and trial-and-error methods, which often lead to low efficiency and prolonged adjustment cycles. To address this challenge, we propose a knowledge-guided simplex search method that integrates a large language model (LLM) with the Soft Actor–Critic (SAC) reinforcement learning algorithm in a collaborative optimization framework, called RLLM-SS. In RLLM-SS, a quasi-gradient mechanism leverages historical data to dynamically estimate the step size and gradient compensation direction of the simplex search method. These estimated variables, integrated with domain knowledge, are encoded into structured prompts that guide the injection molding quality LLM in dynamically adjusting simplex coefficients through natural language reasoning. This enables the simplex search to overcome fixed-coefficient limitation and avoid local optima to the maximum extent. To mitigate the drawbacks of the LLM, such as its tendency to generate hallucinated outputs and lack of memory of past tuning adjustments, a SAC-based evaluation module is introduced. It assigns rewards based on optimization performance, thereby reinforcing effective strategies and fostering continuous policy improvement when similar conditions recur. Experimental evaluations first verified LLM-SS on standard high-dimensional benchmark functions, confirming its effectiveness in complex search spaces, and were then conducted on an injection molding quality simulation platform built on a neural network trained with practical IM process data. Results show that RLLM-SS outperforms several advanced methods, reducing the average number of iterations by 27.6% and the final Euclidean distance to the target quality curve by 68.3%. It also maintains strong robustness under Gaussian noise perturbations.

在注射成型（IM）中，产品质量和工艺稳定性高度依赖于关键工艺参数的设置，因此有效的参数调整对于实现可靠和一致的生产至关重要。然而，传统的调整过程是由专家经验和试错方法指导的，这往往导致低效率和长时间的调整周期。为了解决这一挑战，我们提出了一种知识引导的单纯形搜索方法，该方法将大型语言模型（LLM）与软行为者-评论家（SAC）强化学习算法集成在一个称为RLLM-SS的协作优化框架中。在RLLM-SS中，拟梯度机制利用历史数据动态估计单纯形搜索方法的步长和梯度补偿方向。这些估计变量与领域知识相结合，被编码成结构化提示，指导注塑质量LLM通过自然语言推理动态调整单纯形系数。这使得单纯形搜索克服了固定系数的限制，最大程度地避免了局部最优。为了减轻LLM的缺点，例如它容易产生幻觉输出和缺乏对过去调谐调整的记忆，引入了一个基于sac的评估模块。它根据优化绩效分配奖励，从而加强有效的策略，并在类似情况再次发生时促进持续的政策改进。实验评估首先验证了LLM-SS在标准高维基准函数上的有效性，验证了其在复杂搜索空间中的有效性，然后在基于实际IM过程数据训练的神经网络的注塑质量仿真平台上进行了实验评估。结果表明，RLLM-SS优于几种先进的方法，平均迭代次数减少27.6%，最终到目标质量曲线的欧几里得距离减少68.3%。在高斯噪声扰动下也保持较强的鲁棒性。

{"title":"RLLM-SS: A knowledge-guided simplex search method integrating large language model and reinforcement learning for injection molding quality control","authors":"Haipeng Zou , Xinyu Li , Yongkuan Yang , Ke Yao , Xiangsong Kong , Zhijiang Shao , Furong Gao","doi":"10.1016/j.aei.2026.104372","DOIUrl":"10.1016/j.aei.2026.104372","url":null,"abstract":"<div><div>In injection molding (IM), product quality and process stability are highly dependent on the setting of key process parameters, making efficient parameter tuning essential for achieving reliable and consistent production. However, the tuning process is traditionally guided by expert experience and trial-and-error methods, which often lead to low efficiency and prolonged adjustment cycles. To address this challenge, we propose a knowledge-guided simplex search method that integrates a large language model (LLM) with the Soft Actor–Critic (SAC) reinforcement learning algorithm in a collaborative optimization framework, called RLLM-SS. In RLLM-SS, a quasi-gradient mechanism leverages historical data to dynamically estimate the step size and gradient compensation direction of the simplex search method. These estimated variables, integrated with domain knowledge, are encoded into structured prompts that guide the injection molding quality LLM in dynamically adjusting simplex coefficients through natural language reasoning. This enables the simplex search to overcome fixed-coefficient limitation and avoid local optima to the maximum extent. To mitigate the drawbacks of the LLM, such as its tendency to generate hallucinated outputs and lack of memory of past tuning adjustments, a SAC-based evaluation module is introduced. It assigns rewards based on optimization performance, thereby reinforcing effective strategies and fostering continuous policy improvement when similar conditions recur. Experimental evaluations first verified LLM-SS on standard high-dimensional benchmark functions, confirming its effectiveness in complex search spaces, and were then conducted on an injection molding quality simulation platform built on a neural network trained with practical IM process data. Results show that RLLM-SS outperforms several advanced methods, reducing the average number of iterations by 27.6% and the final Euclidean distance to the target quality curve by 68.3%. It also maintains strong robustness under Gaussian noise perturbations.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"71 ","pages":"Article 104372"},"PeriodicalIF":9.9,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146023024","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

A general and efficient approach for uncertainty quantification in neural networks: Identifying risky decisions in AI systems 神经网络中不确定性量化的通用有效方法：识别人工智能系统中的风险决策

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

Advanced Engineering Informatics

Pub Date : 2026-04-01 Epub Date: 2026-01-22 DOI: 10.1016/j.aei.2026.104343

Zhao Zhang, Senlin Luo, Xiaolong Wu, Xikai Gao, Jiawei Pi, Limin Pan

Uncertainty quantification in neural networks enables the assessment of predictive reliability in artificial intelligence systems, thereby reducing the risk of unsafe decisions. Existing approaches rely heavily on ensemble construction to sample the model parameter space and capture decision variability. However, under realistic resource constraints, small-scale sampling leads to insufficient evidence sources and inaccurate uncertainty estimates. In addition, the design of uncertainty metrics significantly influences estimation accuracy and may limit applicability across different types of machine learning (ML) tasks. In this paper, a Systematic Reusable Ensemble (SRE) framework is proposed for uncertainty quantification. The approach reuses and shares neural network components during retraining to efficiently generate multiple model instances within a single training process. Furthermore, a compounded ensemble pruning strategy is introduced to promote more uniform sampling in parameter space. A general fusion metric is then developed based on evidence theory with a redesigned trust allocation mechanism. Experimental results demonstrate that the proposed framework systematically reduces ensemble construction overhead while improving the reliability of uncertainty estimation. The generalization capability of the SRE is further validated through its effectiveness in identifying high-risk decisions across at least five categories of ML tasks.

神经网络中的不确定性量化能够评估人工智能系统的预测可靠性，从而降低不安全决策的风险。现有方法严重依赖于集成构造来采样模型参数空间并捕获决策可变性。然而，在现实的资源限制下，小规模抽样导致证据来源不足和不确定性估计不准确。此外，不确定性度量的设计显着影响估计的准确性，并可能限制在不同类型的机器学习（ML）任务中的适用性。提出了一种用于不确定性量化的系统可重用集成（SRE）框架。该方法在再训练过程中重用和共享神经网络组件，从而在单个训练过程中高效地生成多个模型实例。在此基础上，引入复合集合剪枝策略，使采样在参数空间上更加均匀。基于证据理论，提出了一种通用的融合度量，并重新设计了信任分配机制。实验结果表明，该框架系统地降低了集成构建开销，提高了不确定性估计的可靠性。通过在至少五类机器学习任务中识别高风险决策的有效性，进一步验证了SRE的泛化能力。

{"title":"A general and efficient approach for uncertainty quantification in neural networks: Identifying risky decisions in AI systems","authors":"Zhao Zhang, Senlin Luo, Xiaolong Wu, Xikai Gao, Jiawei Pi, Limin Pan","doi":"10.1016/j.aei.2026.104343","DOIUrl":"10.1016/j.aei.2026.104343","url":null,"abstract":"<div><div>Uncertainty quantification in neural networks enables the assessment of predictive reliability in artificial intelligence systems, thereby reducing the risk of unsafe decisions. Existing approaches rely heavily on ensemble construction to sample the model parameter space and capture decision variability. However, under realistic resource constraints, small-scale sampling leads to insufficient evidence sources and inaccurate uncertainty estimates. In addition, the design of uncertainty metrics significantly influences estimation accuracy and may limit applicability across different types of machine learning (ML) tasks. In this paper, a Systematic Reusable Ensemble (SRE) framework is proposed for uncertainty quantification. The approach reuses and shares neural network components during retraining to efficiently generate multiple model instances within a single training process. Furthermore, a compounded ensemble pruning strategy is introduced to promote more uniform sampling in parameter space. A general fusion metric is then developed based on evidence theory with a redesigned trust allocation mechanism. Experimental results demonstrate that the proposed framework systematically reduces ensemble construction overhead while improving the reliability of uncertainty estimation. The generalization capability of the SRE is further validated through its effectiveness in identifying high-risk decisions across at least five categories of ML tasks.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"71 ","pages":"Article 104343"},"PeriodicalIF":9.9,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146023035","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 adaptive spatial–temporal encoder with gated multi-convolutions for remaining useful life prediction 用于剩余使用寿命预测的门控多卷积自适应时空编码器

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

Advanced Engineering Informatics

Pub Date : 2026-04-01 Epub Date: 2026-01-23 DOI: 10.1016/j.aei.2026.104369

Wen Liu , Jyun-You Chiang , Yi Li , Haobo Zhang

Accurate estimation of Remaining Useful Life (RUL) is essential for safe and economical operation of turbofan engines. This paper introduces an Adaptive Spatial-Temporal Encoder with Gated Multi-Convolutions (GMC-ASTE), a novel approach that simultaneously models temporal dynamics and inter-sensor relationships to enhance RUL prediction accuracy. The methodology employs a multi-scale gated convolution module to extract refined features from raw multi-sensor data, effectively reducing noise while retaining transient signals. These features are subsequently processed through an adaptive spatial–temporal encoder, which utilizes graph attention mechanisms to dynamically adjust sensor connections and multi-head temporal attention to capture long-term dependencies parallelly. Extensive experiments on the Commercial Modular Aero-Propulsion System Simulation (C-MAPSS) benchmark demonstrate that GMC-ASTE achieves superior performance, with the lowest Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Scoring metrics across all four sub-datasets. The results confirm the effectiveness and interpretability of the proposed model, providing an advanced framework that improves engine prognostic theory and offering airlines a practical tool to reduce downtime and maintenance costs.

准确估计剩余使用寿命（RUL）是保证涡扇发动机安全经济运行的关键。本文介绍了一种具有门控多卷积的自适应时空编码器（gmmc - aste），这是一种同时模拟时间动态和传感器间关系以提高RUL预测精度的新方法。该方法采用多尺度门控卷积模块从原始多传感器数据中提取精细特征，在保留瞬态信号的同时有效地降低噪声。这些特征随后通过自适应时空编码器进行处理，该编码器利用图形注意机制动态调整传感器连接和多头时间注意来并行捕获长期依赖关系。在商用模块化航空推进系统仿真（C-MAPSS）基准测试上进行的大量实验表明，GMC-ASTE在所有四个子数据集上都具有最低的均方根误差（RMSE）、平均绝对误差（MAE）和评分指标，具有卓越的性能。研究结果证实了该模型的有效性和可解释性，为改进发动机预测理论提供了一个先进的框架，并为航空公司提供了一个减少停机时间和维护成本的实用工具。

{"title":"An adaptive spatial–temporal encoder with gated multi-convolutions for remaining useful life prediction","authors":"Wen Liu , Jyun-You Chiang , Yi Li , Haobo Zhang","doi":"10.1016/j.aei.2026.104369","DOIUrl":"10.1016/j.aei.2026.104369","url":null,"abstract":"<div><div>Accurate estimation of Remaining Useful Life (RUL) is essential for safe and economical operation of turbofan engines. This paper introduces an Adaptive Spatial-Temporal Encoder with Gated Multi-Convolutions (GMC-ASTE), a novel approach that simultaneously models temporal dynamics and inter-sensor relationships to enhance RUL prediction accuracy. The methodology employs a multi-scale gated convolution module to extract refined features from raw multi-sensor data, effectively reducing noise while retaining transient signals. These features are subsequently processed through an adaptive spatial–temporal encoder, which utilizes graph attention mechanisms to dynamically adjust sensor connections and multi-head temporal attention to capture long-term dependencies parallelly. Extensive experiments on the Commercial Modular Aero-Propulsion System Simulation (C-MAPSS) benchmark demonstrate that GMC-ASTE achieves superior performance, with the lowest Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Scoring metrics across all four sub-datasets. The results confirm the effectiveness and interpretability of the proposed model, providing an advanced framework that improves engine prognostic theory and offering airlines a practical tool to reduce downtime and maintenance costs.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"71 ","pages":"Article 104369"},"PeriodicalIF":9.9,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146023134","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

Discovering interpretable blast Loading equations from Black-Box Machine learning models 从黑匣子机器学习模型中发现可解释的爆炸加载方程

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

Advanced Engineering Informatics

Pub Date : 2026-04-01 Epub Date: 2025-12-16 DOI: 10.1016/j.aei.2025.104244

Zifan Shi , Qilin Li , Yanda Shao , Ling Li , Hong Hao

Boiling Liquid Expanding Vapour Explosion (BLEVE) is a high-energy event characterised by intense blast waves that pose serious safety risks. Accurate prediction of the resulting overpressure wave is essential for knowledge-intensive engineering analysis and decision support. While empirical methods are available for predicting overpressure in simple BLEVE scenarios, they fail to capture nonlinear relationships in multi-feature and complex conditions. Computational Fluid Dynamics (CFD) methods offer high accuracy in overpressure wave prediction but are computationally intensive, expensive to use and difficult to integrate into automated or real-time engineering workflows. Machine learning models offer a promising alternative for rapid predictions, but their limited interpretability, particularly in deep learning architectures, poses a significant barrier to integration into real-world engineering systems. This study proposes a systematic approach combining machine learning, explainable artificial intelligence, and symbolic regression for BLEVE overpressure prediction. A feedforward neural network model is developed and interpreted using SHapley Additive exPlanations (SHAP). Global SHAP analysis identified nine features with the most significant contributions, which were subsequently used to train a global surrogate model via symbolic regression. This approach yielded an explicit mathematical expression that approximates the behaviour of the original neural network. The derived equation achieved a relative error of 15.73% on simulated data and 35.45% on experimental data, outperforming existing empirical formulas. This research demonstrates the potential of combining black-box machine learning models with xAI techniques to develop interpretable and reliable equations for blast load prediction. More importantly, it introduces a novel data-driven methodology of data-model-interpretation-equation that formalises engineering knowledge by transforming black-box models into explicit and interpretable computational representations.

沸腾液体膨胀蒸汽爆炸（BLEVE）是一种高能量事件，其特征是强烈的冲击波，具有严重的安全风险。准确预测由此产生的超压波对于知识密集型工程分析和决策支持至关重要。虽然经验方法可用于预测简单BLEVE场景下的超压，但它们无法捕捉多特征和复杂条件下的非线性关系。计算流体动力学（CFD）方法在超压波预测方面具有很高的准确性，但计算量大，使用成本高，难以集成到自动化或实时工程工作流程中。机器学习模型为快速预测提供了一个有希望的替代方案，但是它们有限的可解释性，特别是在深度学习架构中，对集成到现实世界的工程系统构成了重大障碍。本研究提出了一种结合机器学习、可解释人工智能和符号回归的BLEVE超压预测系统方法。建立了一个前馈神经网络模型，并使用SHapley加性解释（SHAP）进行了解释。全局SHAP分析确定了贡献最大的九个特征，随后通过符号回归用于训练全局代理模型。这种方法产生了一个显式的数学表达式，近似于原始神经网络的行为。推导出的方程对模拟数据的相对误差为15.73%，对实验数据的相对误差为35.45%，优于现有的经验公式。该研究展示了将黑盒机器学习模型与xAI技术相结合，开发可解释且可靠的爆炸载荷预测方程的潜力。更重要的是，它引入了一种新的数据驱动的数据模型-解释-方程方法，通过将黑箱模型转换为显式和可解释的计算表示来形式化工程知识。

{"title":"Discovering interpretable blast Loading equations from Black-Box Machine learning models","authors":"Zifan Shi , Qilin Li , Yanda Shao , Ling Li , Hong Hao","doi":"10.1016/j.aei.2025.104244","DOIUrl":"10.1016/j.aei.2025.104244","url":null,"abstract":"<div><div>Boiling Liquid Expanding Vapour Explosion (BLEVE) is a high-energy event characterised by intense blast waves that pose serious safety risks. Accurate prediction of the resulting overpressure wave is essential for knowledge-intensive engineering analysis and decision support. While empirical methods are available for predicting overpressure in simple BLEVE scenarios, they fail to capture nonlinear relationships in multi-feature and complex conditions. Computational Fluid Dynamics (CFD) methods offer high accuracy in overpressure wave prediction but are computationally intensive, expensive to use and difficult to integrate into automated or real-time engineering workflows. Machine learning models offer a promising alternative for rapid predictions, but their limited interpretability, particularly in deep learning architectures, poses a significant barrier to integration into real-world engineering systems. This study proposes a systematic approach combining machine learning, explainable artificial intelligence, and symbolic regression for BLEVE overpressure prediction. A feedforward neural network model is developed and interpreted using SHapley Additive exPlanations (SHAP). Global SHAP analysis identified nine features with the most significant contributions, which were subsequently used to train a global surrogate model via symbolic regression. This approach yielded an explicit mathematical expression that approximates the behaviour of the original neural network. The derived equation achieved a relative error of 15.73% on simulated data and 35.45% on experimental data, outperforming existing empirical formulas. This research demonstrates the potential of combining black-box machine learning models with xAI techniques to develop interpretable and reliable equations for blast load prediction. More importantly, it introduces a novel data-driven methodology of data-model-interpretation-equation that formalises engineering knowledge by transforming black-box models into explicit and interpretable computational representations.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"71 ","pages":"Article 104244"},"PeriodicalIF":9.9,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145792146","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

Integration of LiDAR scan-to-IFC and UWB real-time positioning for automated construction monitoring: a precast module case study 集成激光雷达扫描到ifc和超宽带实时定位的自动化施工监控：预制模块案例研究

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

Advanced Engineering Informatics

Pub Date : 2026-04-01 Epub Date: 2025-12-19 DOI: 10.1016/j.aei.2025.104247

Maggie Y. Gao, Chengjia Han, Zhen Peng, Yiqing Dong, Robert L.K. Tiong, Yaowen Yang

This paper presents a novel framework for construction monitoring, focusing on efficient as-built model registration through the integration of LiDAR scanning and Ultra-Wideband (UWB) positioning technology. The proposed approach leverages UWB positioning data as preliminary spatial references for precise alignment of as-built model for precast module and components from LiDAR point cloud processing. This integrated framework addresses the critical gap in construction monitoring by integrating multiple technologies into a cohesive system, overcoming the limitations of fragmented approaches. During these procedures, the study presents a systematic targeted partial transformation to correct angular misalignments during point cloud registration. This framework employs BIMCrossNet, a custom deep learning architecture specifically designed for point cloud segmentation in construction scenarios. At last, the study enables automated updating of semantic enriched as-built BIM models with real-time validation of component placement, making it particularly valuable for quality control and progress monitoring in modular construction applications.

本文提出了一种新的建筑监测框架，通过激光雷达扫描和超宽带（UWB）定位技术的集成，重点关注有效的建成模型配准。所提出的方法利用超宽带定位数据作为激光雷达点云处理中预制模块和组件的建成模型精确对准的初步空间参考。这个集成框架通过将多种技术集成到一个有凝聚力的系统中，克服了分散方法的局限性，解决了施工监测中的关键差距。在这些过程中，研究提出了一个系统的有针对性的部分变换，以纠正点云配准过程中的角度失调。该框架采用BIMCrossNet，这是一种专门为建筑场景中的点云分割而设计的定制深度学习架构。最后，该研究能够自动更新语义丰富的建成BIM模型，实时验证组件放置，使其对模块化建筑应用中的质量控制和进度监控特别有价值。

{"title":"Integration of LiDAR scan-to-IFC and UWB real-time positioning for automated construction monitoring: a precast module case study","authors":"Maggie Y. Gao, Chengjia Han, Zhen Peng, Yiqing Dong, Robert L.K. Tiong, Yaowen Yang","doi":"10.1016/j.aei.2025.104247","DOIUrl":"10.1016/j.aei.2025.104247","url":null,"abstract":"<div><div>This paper presents a novel framework for construction monitoring, focusing on efficient as-built model registration through the integration of LiDAR scanning and Ultra-Wideband (UWB) positioning technology. The proposed approach leverages UWB positioning data as preliminary spatial references for precise alignment of as-built model for precast module and components from LiDAR point cloud processing. This integrated framework addresses the critical gap in construction monitoring by integrating multiple technologies into a cohesive system, overcoming the limitations of fragmented approaches. During these procedures, the study presents a systematic targeted partial transformation to correct angular misalignments during point cloud registration. This framework employs BIMCrossNet, a custom deep learning architecture specifically designed for point cloud segmentation in construction scenarios. At last, the study enables automated updating of semantic enriched as-built BIM models with real-time validation of component placement, making it particularly valuable for quality control and progress monitoring in modular construction applications.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"71 ","pages":"Article 104247"},"PeriodicalIF":9.9,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145792145","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

Dynamic reliability informed adaptive task scheduling for multirobot manufacturing system 基于动态可靠性的多机器人制造系统自适应任务调度

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

Advanced Engineering Informatics

Pub Date : 2026-04-01 Epub Date: 2026-01-13 DOI: 10.1016/j.aei.2026.104335

Jian Zhou , Hang Zhang , Buyun Tang , Lianyu Zheng , Yiwei Wang

Efficient and reliable scheduling is critical for multirobot manufacturing systems, yet existing performance diagnosis and reliability evaluation methods fail to capture the dynamic evolution of system states, and current scheduling often neglects reliability information. This paper proposes a collaborative scheduling method integrating dynamic reliability assessment. First, multisource operational signals are collected and fused to identify degradation stages. Based on a classification probability mapping mechanism, the service state is then converted into probabilistic attributes applicable to modeling and scheduling. Subsequently, a system model incorporating structure, state, and behavior is constructed, and reliability indicators of the system are dynamically evaluated through logical model evolution. On this basis, a heuristic multiagent reinforcement learning scheduling algorithm is designed, using reliability attributes and constraint graph structures as inputs to achieve collaborative scheduling of multiple robots. Finally, during task execution, real-time state changes are dynamically perceived, and the scheduling plan is adaptively updated by triggering rescheduling based on the real-time evaluation results, thus forming a reliability-informed closed-loop scheduling mechanism. Case studies demonstrate 18.2% reduction in task completion time for static allocation compared to four baseline methods, along with 17.1% improvement for dynamic rescheduling against engineering practices. These quantitative results confirm the method’s significant enhancements in scheduling responsiveness to degradation, adaptive task optimization, and overall system stability and efficiency.

高效、可靠的调度对多机器人制造系统至关重要，但现有的性能诊断和可靠性评估方法无法捕捉到系统状态的动态演变，且当前的调度往往忽略了可靠性信息。提出了一种集成动态可靠性评估的协同调度方法。首先，采集并融合多源操作信号，识别退化阶段；基于分类概率映射机制，将服务状态转换为适用于建模和调度的概率属性。随后，构建了包含结构、状态和行为的系统模型，并通过逻辑模型演化对系统的可靠性指标进行动态评估。在此基础上，设计了启发式多智能体强化学习调度算法，以可靠性属性和约束图结构为输入，实现多机器人协同调度。最后，在任务执行过程中，动态感知实时状态变化，并根据实时评估结果触发重调度，自适应更新调度计划，形成一种可靠性知情的闭环调度机制。案例研究表明，与四种基线方法相比，静态分配的任务完成时间减少了18.2%，与工程实践相比，动态重新调度的任务完成时间提高了17.1%。这些定量结果证实了该方法在调度响应退化、自适应任务优化以及整体系统稳定性和效率方面的显著增强。

{"title":"Dynamic reliability informed adaptive task scheduling for multirobot manufacturing system","authors":"Jian Zhou , Hang Zhang , Buyun Tang , Lianyu Zheng , Yiwei Wang","doi":"10.1016/j.aei.2026.104335","DOIUrl":"10.1016/j.aei.2026.104335","url":null,"abstract":"<div><div>Efficient and reliable scheduling is critical for multirobot manufacturing systems, yet existing performance diagnosis and reliability evaluation methods fail to capture the dynamic evolution of system states, and current scheduling often neglects reliability information. This paper proposes a collaborative scheduling method integrating dynamic reliability assessment. First, multisource operational signals are collected and fused to identify degradation stages. Based on a classification probability mapping mechanism, the service state is then converted into probabilistic attributes applicable to modeling and scheduling. Subsequently, a system model incorporating structure, state, and behavior is constructed, and reliability indicators of the system are dynamically evaluated through logical model evolution. On this basis, a heuristic multiagent reinforcement learning scheduling algorithm is designed, using reliability attributes and constraint graph structures as inputs to achieve collaborative scheduling of multiple robots. Finally, during task execution, real-time state changes are dynamically perceived, and the scheduling plan is adaptively updated by triggering rescheduling based on the real-time evaluation results, thus forming a reliability-informed closed-loop scheduling mechanism. Case studies demonstrate 18.2% reduction in task completion time for static allocation compared to four baseline methods, along with 17.1% improvement for dynamic rescheduling against engineering practices. These quantitative results confirm the method’s significant enhancements in scheduling responsiveness to degradation, adaptive task optimization, and overall system stability and efficiency.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"71 ","pages":"Article 104335"},"PeriodicalIF":9.9,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145977718","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

A transformer-based framework for cross-material in situ monitoring in extrusion-based bioprinting 一种基于变压器的框架，用于挤压生物打印中交叉材料的原位监测

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

Advanced Engineering Informatics

Pub Date : 2026-04-01 Epub Date: 2026-01-16 DOI: 10.1016/j.aei.2026.104323

Jiayi Zhang , Kaicheng Yu , Yifeng Yao , Lihua Lu , Qiang Gao , Peng Zhang , Guoyin Shang , Swee Leong Sing

Additive manufacturing is advancing toward intelligent and functionally reliable fabrication, particularly in biomedical applications. In extrusion-based three-dimensional (3D) bioprinting, machine learning (ML)-enabled in situ monitoring is crucial for improving print quality and ensuring the functional performance of tissue engineering constructs. This study proposes a transformer-based transfer learning framework for cross-material monitoring that efficiently transfers knowledge across diverse polymer systems under limited-data conditions. The model extracts geometric features of extruded filaments from in situ monitoring images and achieves 99.55% classification accuracy on PLCL and over 98% accuracy for PCL/

β

-TCP and GelMA datasets with less than 0.1% trainable parameters. Beyond visual monitoring, the predicted filament process states were quantitatively correlated with downstream mechanical performance, demonstrating a 24.6% improvement in tensile strength and enhanced geometric fidelity under optimal-heating conditions. Furthermore, in vivo wound-healing experiments using the bioprinted constructs verified the biological relevance and translational potential of the proposed monitoring strategy. Constructs fabricated under optimal conditions promoted accelerated tissue regeneration and vascularization, achieving faster wound closure within 10 days compared with suboptimal printing conditions. Overall, the proposed transformer-based cross-material framework establishes a generalizable and biologically validated paradigm for vision-guided process monitoring, providing a key step toward intelligent and adaptive bioprinting.

增材制造正朝着智能和功能可靠的制造方向发展，特别是在生物医学应用方面。在基于挤压的三维（3D）生物打印中，支持机器学习（ML）的原位监测对于提高打印质量和确保组织工程结构的功能性能至关重要。本研究提出了一种基于变压器的跨材料监测迁移学习框架，该框架可在有限数据条件下有效地跨不同聚合物系统迁移知识。该模型从现场监测图像中提取挤压细丝的几何特征，在可训练参数小于0.1%的情况下，对PLCL的分类准确率达到99.55%，对PCL/β-TCP和GelMA数据集的分类准确率达到98%以上。除了视觉监测外，预测的长丝工艺状态与下游机械性能定量相关，显示在最佳加热条件下拉伸强度提高24.6%，几何保真度增强。此外，使用生物打印构建体的体内伤口愈合实验验证了所提出的监测策略的生物学相关性和转化潜力。在最佳条件下制造的结构促进了组织再生和血管形成，与次优打印条件相比，在10天内实现了更快的伤口愈合。总的来说，提出的基于变压器的跨材料框架为视觉引导的过程监测建立了一个可推广的和生物验证的范例，为智能和自适应生物打印提供了关键的一步。

{"title":"A transformer-based framework for cross-material in situ monitoring in extrusion-based bioprinting","authors":"Jiayi Zhang , Kaicheng Yu , Yifeng Yao , Lihua Lu , Qiang Gao , Peng Zhang , Guoyin Shang , Swee Leong Sing","doi":"10.1016/j.aei.2026.104323","DOIUrl":"10.1016/j.aei.2026.104323","url":null,"abstract":"<div><div>Additive manufacturing is advancing toward intelligent and functionally reliable fabrication, particularly in biomedical applications. In extrusion-based three-dimensional (3D) bioprinting, machine learning (ML)-enabled <em>in situ</em> monitoring is crucial for improving print quality and ensuring the functional performance of tissue engineering constructs. This study proposes a transformer-based transfer learning framework for cross-material monitoring that efficiently transfers knowledge across diverse polymer systems under limited-data conditions. The model extracts geometric features of extruded filaments from in situ monitoring images and achieves 99.55% classification accuracy on PLCL and over 98% accuracy for PCL/<span><math><mi>β</mi></math></span>-TCP and GelMA datasets with less than 0.1% trainable parameters. Beyond visual monitoring, the predicted filament process states were quantitatively correlated with downstream mechanical performance, demonstrating a 24.6% improvement in tensile strength and enhanced geometric fidelity under optimal-heating conditions. Furthermore, <em>in vivo</em> wound-healing experiments using the bioprinted constructs verified the biological relevance and translational potential of the proposed monitoring strategy. Constructs fabricated under optimal conditions promoted accelerated tissue regeneration and vascularization, achieving faster wound closure within 10 days compared with suboptimal printing conditions. Overall, the proposed transformer-based cross-material framework establishes a generalizable and biologically validated paradigm for vision-guided process monitoring, providing a key step toward intelligent and adaptive bioprinting.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"71 ","pages":"Article 104323"},"PeriodicalIF":9.9,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145977715","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

Causal feature selection framework for stable soft sensor modeling based on time-delayed cross mapping 基于时延交叉映射的稳定软传感器建模因果特征选择框架

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

Advanced Engineering Informatics

Pub Date : 2026-04-01 Epub Date: 2026-01-17 DOI: 10.1016/j.aei.2026.104337

Shi-Shun Chen , Xiao-Yang Li , Enrico Zio

Soft sensor modeling plays a crucial role in process monitoring. Causal feature selection can enhance the performance of soft sensor models in industrial applications. However, existing methods ignore two critical characteristics of industrial processes. Firstly, causal relationships between variables always involve time delays, whereas most causal feature selection methods investigate causal relationships in the same time dimension. Secondly, variables in industrial processes are often interdependent, which contradicts the decorrelation assumption of traditional causal inference methods. Consequently, soft sensor models based on existing causal feature selection approaches often lack sufficient accuracy and stability. To overcome these challenges, this paper proposes a causal feature selection framework based on time-delayed cross mapping. Time-delayed cross mapping employs state space reconstruction to effectively handle interdependent variables in causality analysis, and considers varying causal strength across time delay. Time-delayed convergent cross mapping (TDCCM) is introduced for total causal inference, and time-delayed partial cross mapping (TDPCM) is developed for direct causal inference. Then, in order to achieve automatic feature selection, an objective feature selection strategy is presented. The causal threshold is automatically determined based on the model performance on the validation set, and the causal features are then selected. Two real-world case studies show that TDCCM achieves the highest average performance, while TDPCM improves soft sensor stability and performance in the worst scenario. On average over the two cases, TDCCM decreases root mean square error (RMSE) by about 8.93% compared with the best existing feature selection method, and TDPCM further decreases RMSE in the worst scenario by about 7.69% relative to TDCCM. The code is publicly available at https://github.com/dirge1/TDPCM.

软测量建模在过程监控中起着至关重要的作用。在工业应用中，因果特征选择可以提高软测量模型的性能。然而，现有的方法忽略了工业过程的两个关键特征。首先，变量之间的因果关系总是涉及到时间延迟，而大多数因果特征选择方法在同一时间维度上考察因果关系。其次，工业过程中的变量往往是相互依存的，这与传统因果推理方法的去相关假设相矛盾。因此，基于现有因果特征选择方法的软测量模型往往缺乏足够的精度和稳定性。为了克服这些挑战，本文提出了一种基于时滞交叉映射的因果特征选择框架。时滞交叉映射在因果分析中采用状态空间重构来有效处理相互依赖的变量，并考虑了不同时间延迟的因果强度。在全因果推理中引入了延时收敛交叉映射（TDCCM），在直接因果推理中发展了延时部分交叉映射（TDPCM）。然后，为了实现自动特征选择，提出了一种客观特征选择策略。根据模型在验证集上的性能自动确定因果阈值，然后选择因果特征。两个实际案例研究表明，TDCCM实现了最高的平均性能，而TDPCM在最坏的情况下提高了软传感器的稳定性和性能。在两种情况下，TDCCM平均比现有最佳特征选择方法降低了约8.93%的均方根误差（RMSE）， TDPCM在最坏情况下比TDCCM进一步降低了约7.69%的RMSE。该代码可在https://github.com/dirge1/TDPCM上公开获得。

{"title":"Causal feature selection framework for stable soft sensor modeling based on time-delayed cross mapping","authors":"Shi-Shun Chen , Xiao-Yang Li , Enrico Zio","doi":"10.1016/j.aei.2026.104337","DOIUrl":"10.1016/j.aei.2026.104337","url":null,"abstract":"<div><div>Soft sensor modeling plays a crucial role in process monitoring. Causal feature selection can enhance the performance of soft sensor models in industrial applications. However, existing methods ignore two critical characteristics of industrial processes. Firstly, causal relationships between variables always involve time delays, whereas most causal feature selection methods investigate causal relationships in the same time dimension. Secondly, variables in industrial processes are often interdependent, which contradicts the decorrelation assumption of traditional causal inference methods. Consequently, soft sensor models based on existing causal feature selection approaches often lack sufficient accuracy and stability. To overcome these challenges, this paper proposes a causal feature selection framework based on time-delayed cross mapping. Time-delayed cross mapping employs state space reconstruction to effectively handle interdependent variables in causality analysis, and considers varying causal strength across time delay. Time-delayed convergent cross mapping (TDCCM) is introduced for total causal inference, and time-delayed partial cross mapping (TDPCM) is developed for direct causal inference. Then, in order to achieve automatic feature selection, an objective feature selection strategy is presented. The causal threshold is automatically determined based on the model performance on the validation set, and the causal features are then selected. Two real-world case studies show that TDCCM achieves the highest average performance, while TDPCM improves soft sensor stability and performance in the worst scenario. On average over the two cases, TDCCM decreases root mean square error (RMSE) by about 8.93% compared with the best existing feature selection method, and TDPCM further decreases RMSE in the worst scenario by about 7.69% relative to TDCCM. The code is publicly available at <span><span>https://github.com/dirge1/TDPCM</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"71 ","pages":"Article 104337"},"PeriodicalIF":9.9,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145977713","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

Asynchronous multi-agent based differential evolution for assembly hybrid differentiation flowshop scheduling with variable sub-lot and limited buffer 基于异步多智能体的可变子批有限缓冲装配混合微分流水车间调度

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

Advanced Engineering Informatics

Pub Date : 2026-04-01 Epub Date: 2026-01-12 DOI: 10.1016/j.aei.2025.104266

Yiling Lu , Qiuhua Tang

Lot-streaming enables a discrete supply of components to multiple assembled products, while the limited buffer controls the fluency of the production flow. Thus, integrating differentiation processing with assembly, this work addresses an assembly hybrid differentiation flowshop scheduling with variable sub-lot and limited buffer (AHDFSP-VS-LB). Focusing on makespan minimization, a mixed-integer linear programming model is established and an asynchronous multi-agent based differential evolution (AMDE) is developed. Constrained by the limited buffer capacity, the AMDE incorporates a deadlock pre-detection strategy to quickly identify infeasible solutions during encoding and a dynamic adjustment strategy to ensure a high-quality feasible solution during decoding. Further, derived from processing-assembly coordination and buffer constraint, four problem-specific properties are extracted and embedded into the initialization and improvement strategy to optimize the performance of the algorithm. To achieve fast and sufficient convergence of the algorithm, an asynchronous multi-agent cooperative learning mechanism is designed to dynamically control the evolution power and direction for each individual by identifying the critical encoding layer and its suitable parameters at different search phases. Comprehensive experiments demonstrate that the designed operators excel in efficiency and coordination, and the proposed algorithm is superior to five state-of-the-art algorithms in solving this new problem.

批量流能够为多个组装产品提供离散的组件供应，而有限的缓冲区控制着生产流程的流畅性。因此，将微分处理与装配相结合，本研究解决了具有可变子批和有限缓冲的装配混合微分流水车间调度（AHDFSP-VS-LB）。以最大时间跨度最小化为目标，建立了混合整数线性规划模型，提出了基于异步多智能体的差分进化算法。由于缓冲容量有限，AMDE采用了死锁预检测策略来快速识别编码过程中的不可行解，并采用动态调整策略来确保解码过程中的高质量可行解。此外，从加工装配协调和缓冲约束出发，提取了四个特定于问题的属性，并将其嵌入到初始化和改进策略中，以优化算法的性能。为了实现算法的快速和充分收敛，设计了异步多智能体合作学习机制，通过识别不同搜索阶段的关键编码层及其合适参数，动态控制每个个体的进化能力和进化方向。综合实验表明，所设计的算子具有较好的效率和协调性，在求解该新问题方面优于现有的5种算法。

{"title":"Asynchronous multi-agent based differential evolution for assembly hybrid differentiation flowshop scheduling with variable sub-lot and limited buffer","authors":"Yiling Lu , Qiuhua Tang","doi":"10.1016/j.aei.2025.104266","DOIUrl":"10.1016/j.aei.2025.104266","url":null,"abstract":"<div><div>Lot-streaming enables a discrete supply of components to multiple assembled products, while the limited buffer controls the fluency of the production flow. Thus, integrating differentiation processing with assembly, this work addresses an assembly hybrid differentiation flowshop scheduling with variable sub-lot and limited buffer (AHDFSP-VS-LB). Focusing on makespan minimization, a mixed-integer linear programming model is established and an asynchronous multi-agent based differential evolution (AMDE) is developed. Constrained by the limited buffer capacity, the AMDE incorporates a deadlock pre-detection strategy to quickly identify infeasible solutions during encoding and a dynamic adjustment strategy to ensure a high-quality feasible solution during decoding. Further, derived from processing-assembly coordination and buffer constraint, four problem-specific properties are extracted and embedded into the initialization and improvement strategy to optimize the performance of the algorithm. To achieve fast and sufficient convergence of the algorithm, an asynchronous multi-agent cooperative learning mechanism is designed to dynamically control the evolution power and direction for each individual by identifying the critical encoding layer and its suitable parameters at different search phases. Comprehensive experiments demonstrate that the designed operators excel in efficiency and coordination, and the proposed algorithm is superior to five state-of-the-art algorithms in solving this new problem.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"71 ","pages":"Article 104266"},"PeriodicalIF":9.9,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145978251","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

Efficiency-aware seismic fragility analysis of super-high arch dam using unsupervised ground motion clustering with probabilistic representation 基于概率表示的无监督地震动聚类的超高拱坝地震易损分析

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

Advanced Engineering Informatics

Pub Date : 2026-04-01 Epub Date: 2026-01-15 DOI: 10.1016/j.aei.2026.104318

Yingbo Chen , Mingchao Li , Qiubing Ren , Zhiyong Qi , Hui Liang

Machine Learning (ML)-driven approaches have been employed to replace computationally intensive seismic simulations of hydraulic engineering structures. For the complex seismic responses of arch dams, constructing a metamodel that captures the nonlinear relationship between ground motion inputs and structural response outputs using a limited set of numerical simulations can significantly reduce the computational cost. However, conventional deterministic predictions and fragility analyses fail to account for the high aleatory and epistemic uncertainties inherent in the seismic response of arch dams. To this end, this paper proposes an efficient fragility analysis method for arch dams that integrates probabilistic ML algorithms with the traditional Incremental Dynamic Analysis. By constructing a Natural Gradient Boosting (NGBoost) metamodel for the arch dam dynamic response, not only can the predicted mean value of each response sample be obtained, but also its conditional probability distribution. Superimpose the simulation data with the response distribution predicted by NGBoost, and the binary parameters of the fragility function are estimated, thereby generating both the fragility curve and the uncertain fragility interval of arch dam. Additionally, representative Ground Motion Records (GMRs) for the arch dam are selected using the Partitioning Around Medoids (PAM) unsupervised clustering technique, determining the minimum subset proportion that effectively represents the whole GMR dataset. The effectiveness of the proposed method is validated in a super-high arch dam. The 40% GMR proportion is found to adequately reproduce the fragility curves of the whole dataset, with the reference curve falling within the derived uncertainty interval, achieving a 56.8% reduction in computational cost. The 60% GMR proportion ensured fragility curves with balanced accuracy and effectiveness, exhibiting maximum mean differences of 0.058 and maximum standard deviation differences of 0.031 from reference curves across all damage levels, while reducing computational cost by 39.7%. Comparative results demonstrate the superiority of NGBoost and PAM over existing deterministic metamodels and GMRs selection techniques, respectively. The efficient fragility analysis method proposed in this study ultimately enables the direct characterization of uncertainties in arch dam seismic responses.

机器学习（ML）驱动的方法已被用于取代水力工程结构的计算密集型地震模拟。对于拱坝复杂的地震反应，利用有限的数值模拟，构建一个能够捕捉地震动输入与结构反应输出之间非线性关系的元模型，可以显著降低计算成本。然而，传统的确定性预测和脆弱性分析无法解释拱坝地震反应中固有的高度随机性和认知不确定性。为此，本文提出了一种将概率ML算法与传统增量动力分析相结合的拱坝易损性分析方法。通过构建拱坝动力响应的自然梯度增强元模型（NGBoost），不仅可以得到各响应样本的预测均值，还可以得到其条件概率分布。将模拟数据与NGBoost预测的响应分布叠加，估计出易损性函数的二值参数，从而生成拱坝易损性曲线和不确定易损性区间。此外，利用无监督聚类技术（PAM）选择具有代表性的拱坝地震动记录（GMRs），确定有效代表整个GMR数据集的最小子集比例。在某超高层拱坝工程中验证了该方法的有效性。发现40%的GMR比例可以充分再现整个数据集的脆弱性曲线，参考曲线落在推导的不确定性区间内，计算成本降低56.8%。60%的GMR比例保证了脆性曲线的准确性和有效性的平衡，在所有损伤级别上与参考曲线的最大平均差异为0.058，最大标准差差异为0.031，同时减少了39.7%的计算成本。对比结果表明，NGBoost和PAM分别优于现有的确定性元模型和gmr选择技术。本研究提出的高效易损分析方法最终能够直接表征拱坝地震反应的不确定性。

{"title":"Efficiency-aware seismic fragility analysis of super-high arch dam using unsupervised ground motion clustering with probabilistic representation","authors":"Yingbo Chen , Mingchao Li , Qiubing Ren , Zhiyong Qi , Hui Liang","doi":"10.1016/j.aei.2026.104318","DOIUrl":"10.1016/j.aei.2026.104318","url":null,"abstract":"<div><div>Machine Learning (ML)-driven approaches have been employed to replace computationally intensive seismic simulations of hydraulic engineering structures. For the complex seismic responses of arch dams, constructing a metamodel that captures the nonlinear relationship between ground motion inputs and structural response outputs using a limited set of numerical simulations can significantly reduce the computational cost. However, conventional deterministic predictions and fragility analyses fail to account for the high aleatory and epistemic uncertainties inherent in the seismic response of arch dams. To this end, this paper proposes an efficient fragility analysis method for arch dams that integrates probabilistic ML algorithms with the traditional Incremental Dynamic Analysis. By constructing a Natural Gradient Boosting (NGBoost) metamodel for the arch dam dynamic response, not only can the predicted mean value of each response sample be obtained, but also its conditional probability distribution. Superimpose the simulation data with the response distribution predicted by NGBoost, and the binary parameters of the fragility function are estimated, thereby generating both the fragility curve and the uncertain fragility interval of arch dam. Additionally, representative Ground Motion Records (GMRs) for the arch dam are selected using the Partitioning Around Medoids (PAM) unsupervised clustering technique, determining the minimum subset proportion that effectively represents the whole GMR dataset. The effectiveness of the proposed method is validated in a super-high arch dam. The 40% GMR proportion is found to adequately reproduce the fragility curves of the whole dataset, with the reference curve falling within the derived uncertainty interval, achieving a 56.8% reduction in computational cost. The 60% GMR proportion ensured fragility curves with balanced accuracy and effectiveness, exhibiting maximum mean differences of 0.058 and maximum standard deviation differences of 0.031 from reference curves across all damage levels, while reducing computational cost by 39.7%. Comparative results demonstrate the superiority of NGBoost and PAM over existing deterministic metamodels and GMRs selection techniques, respectively. The efficient fragility analysis method proposed in this study ultimately enables the direct characterization of uncertainties in arch dam seismic responses.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"71 ","pages":"Article 104318"},"PeriodicalIF":9.9,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145978250","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