Advanced Engineering Informatics最新文献_第3页

Investigating VR Safety Training Transfer in Construction Hazard Recognition: A Neurocognitive Perspective 研究VR安全培训在建筑危险识别中的转移：神经认知视角

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

Advanced Engineering Informatics

Pub Date : 2026-02-09 DOI: 10.1016/j.aei.2026.104412

Zhaoyang Xiong , Zhikun Ding , Yaning Li , Jianfeng Zhang

Effective construction safety training requires training transfer to enable workers to recognize untrained hazards, a critical capability for adapting to the complex, dynamic environments of construction sites. Current virtual reality (VR) safety training evaluates performance only within trained scenarios, neglecting the mechanisms necessary for transfer and limiting validation of real-world effectiveness. This study utilized magnetoencephalography (MEG) to compare immersive VR and conventional PowerPoint (PPT) training, specifically investigating transfer effectiveness and underlying mechanisms using an untrained hazard scenario protocol. VR training showed significantly better transfer performance (Δd’=0.481 vs. 0.051, p=0.035). This behavioral superiority was linked to distinct neural signatures (248–336 ms post-stimulus) revealing the coordinated emotional-contextual encoding as the core transfer mechanism. These findings challenge rule-based learning assumptions, demonstrating that effective hazard recognition relies on forming robust, generalizable embodied threat memories. Based on these neuroscience-driven insights, we propose the S.A.F.E. neurocognitive framework (Schema matching, Affective salience, Feature detection, Element identification) as the first evidence-based design specification for optimizing VR safety training. This research establishes a methodological foundation for assessing training transfer, identifies its precise neurocognitive mechanisms, and provides design principles to enhance novel hazard recognition in the real world.

有效的建筑安全培训需要培训转移，使工人能够识别未经培训的危险，这是适应建筑工地复杂、动态环境的关键能力。当前的虚拟现实（VR）安全培训仅在训练过的场景中评估性能，忽视了转移所需的机制，并限制了对现实世界有效性的验证。本研究利用脑磁图（MEG）来比较沉浸式VR和传统PPT （PPT）培训，特别研究了使用未经训练的危险场景协议的转移有效性和潜在机制。VR训练表现出更好的迁移表现（Δd ' =0.481 vs. 0.051, p=0.035）。这种行为优势与不同的神经信号（刺激后248-336 ms）有关，揭示了协调的情绪-情境编码是核心转移机制。这些发现挑战了基于规则的学习假设，表明有效的危险识别依赖于形成强大的、可概括的具身威胁记忆。基于这些神经科学驱动的见解，我们提出了S.A.F.E.神经认知框架（图式匹配、情感显著性、特征检测、元素识别），作为优化VR安全培训的第一个循证设计规范。本研究为训练迁移评估奠定了方法学基础，明确了训练迁移的精确神经认知机制，并为增强现实世界中对新危险的识别提供了设计原则。

{"title":"Investigating VR Safety Training Transfer in Construction Hazard Recognition: A Neurocognitive Perspective","authors":"Zhaoyang Xiong , Zhikun Ding , Yaning Li , Jianfeng Zhang","doi":"10.1016/j.aei.2026.104412","DOIUrl":"10.1016/j.aei.2026.104412","url":null,"abstract":"<div><div>Effective construction safety training requires training transfer to enable workers to recognize untrained hazards, a critical capability for adapting to the complex, dynamic environments of construction sites. Current virtual reality (VR) safety training evaluates performance only within trained scenarios, neglecting the mechanisms necessary for transfer and limiting validation of real-world effectiveness. This study utilized magnetoencephalography (MEG) to compare immersive VR and conventional PowerPoint (PPT) training, specifically investigating transfer effectiveness and underlying mechanisms using an untrained hazard scenario protocol. VR training showed significantly better transfer performance (Δd’=0.481 vs. 0.051, p=0.035). This behavioral superiority was linked to distinct neural signatures (248–336 ms post-stimulus) revealing the coordinated emotional-contextual encoding as the core transfer mechanism. These findings challenge rule-based learning assumptions, demonstrating that effective hazard recognition relies on forming robust, generalizable embodied threat memories. Based on these neuroscience-driven insights, we propose the S.A.F.E. neurocognitive framework (Schema matching, Affective salience, Feature detection, Element identification) as the first evidence-based design specification for optimizing VR safety training. This research establishes a methodological foundation for assessing training transfer, identifies its precise neurocognitive mechanisms, and provides design principles to enhance novel hazard recognition in the real world.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"71 ","pages":"Article 104412"},"PeriodicalIF":9.9,"publicationDate":"2026-02-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146188818","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 hierarchical Bayesian approach for the modelling of ice force peaks on ship hull considering ice Thickness, concentration and floe size 考虑冰厚、浓度和浮冰大小的船体冰力峰值分层贝叶斯建模方法

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

Advanced Engineering Informatics

Pub Date : 2026-02-09 DOI: 10.1016/j.aei.2026.104416

Boyang Jiao , Liangliang Lu , Fang Li , Pentti Kujala

Ships navigating in Polar areas are exposed to significant ice loads, characterized by stochasticity stemming from highly variable ice conditions. Accurate prediction of the ice loads is important for ensuring structural integrity and operational safety. A critical aspect of ice load prediction involves establishing reliable relationships between ice conditions and the resulting ice loads. However, inherent uncertainties in ice condition measurements significantly impact the precision of ice load predictions, and the quantitative assessment of this uncertainty remains challenging. This paper introduces a Hierarchical Bayesian Model (HBM) designed to probabilistically model ice force peak distributions as a function of ice thickness, concentration and floe size, accounting for the uncertainties associated with the covariate measurements. The model’s core framework treats the parameters of the ice force distribution as random variables that follow their own probability distributions, allowing uncertainty to be fully propagated through the model’s hierarchical structure. A Generalized Nonlinear Model (GNLM) is also introduced for a comparison purpose. The data from the 2018–2019 Antarctic voyage of S.A. Agulhas II is used to train the models. Results show successful HBM convergence and strong posterior predictive fits to both training and testing data, demonstrating robust generalization. While the GNLM is also evaluated, the comparative analysis shows it provides a poorer fit to the data, particularly for high-magnitude ice forces. In contrast, the HBM provides a robust framework for probabilistic ice force modelling, enhancing predictions under diverse ice conditions to support safer ship design and polar navigation.

在极地航行的船舶暴露在巨大的冰载荷下，其特点是由高度可变的冰况引起的随机性。准确预测冰荷载对保证结构完整性和运行安全具有重要意义。冰负荷预测的一个关键方面是在冰况和冰负荷之间建立可靠的关系。然而，冰况测量中固有的不确定性极大地影响了冰负荷预测的精度，对这种不确定性的定量评估仍然具有挑战性。本文介绍了一种分层贝叶斯模型（HBM），该模型设计用于将冰力峰值分布作为冰厚度、浓度和浮冰大小的函数进行概率建模，并考虑了协变量测量的不确定性。该模型的核心框架将冰力分布的参数视为遵循其自身概率分布的随机变量，从而允许不确定性通过模型的分层结构充分传播。为了进行比较，还引入了广义非线性模型（GNLM）。“阿古拉斯二号”2018-2019年南极航行的数据用于训练模型。结果表明HBM成功收敛，对训练数据和测试数据具有较强的后验预测拟合，具有鲁棒泛化性。虽然GNLM也进行了评估，但比较分析表明，它提供的数据拟合性较差，特别是对于高强度的冰力。相比之下，HBM为概率冰力建模提供了一个强大的框架，增强了不同冰情下的预测，以支持更安全的船舶设计和极地导航。

{"title":"A hierarchical Bayesian approach for the modelling of ice force peaks on ship hull considering ice Thickness, concentration and floe size","authors":"Boyang Jiao , Liangliang Lu , Fang Li , Pentti Kujala","doi":"10.1016/j.aei.2026.104416","DOIUrl":"10.1016/j.aei.2026.104416","url":null,"abstract":"<div><div>Ships navigating in Polar areas are exposed to significant ice loads, characterized by stochasticity stemming from highly variable ice conditions. Accurate prediction of the ice loads is important for ensuring structural integrity and operational safety. A critical aspect of ice load prediction involves establishing reliable relationships between ice conditions and the resulting ice loads. However, inherent uncertainties in ice condition measurements significantly impact the precision of ice load predictions, and the quantitative assessment of this uncertainty remains challenging. This paper introduces a Hierarchical Bayesian Model (HBM) designed to probabilistically model ice force peak distributions as a function of ice thickness, concentration and floe size, accounting for the uncertainties associated with the covariate measurements. The model’s core framework treats the parameters of the ice force distribution as random variables that follow their own probability distributions, allowing uncertainty to be fully propagated through the model’s hierarchical structure. A Generalized Nonlinear Model (GNLM) is also introduced for a comparison purpose. The data from the 2018–2019 Antarctic voyage of S.A. Agulhas II is used to train the models. Results show successful HBM convergence and strong posterior predictive fits to both training and testing data, demonstrating robust generalization. While the GNLM is also evaluated, the comparative analysis shows it provides a poorer fit to the data, particularly for high-magnitude ice forces. In contrast, the HBM provides a robust framework for probabilistic ice force modelling, enhancing predictions under diverse ice conditions to support safer ship design and polar navigation.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"71 ","pages":"Article 104416"},"PeriodicalIF":9.9,"publicationDate":"2026-02-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146188698","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

Multi-fidelity Kriging method based on active and ensemble learning for structural reliability analysis 基于主动学习和集成学习的结构可靠性分析多保真度Kriging方法

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

Advanced Engineering Informatics

Pub Date : 2026-02-07 DOI: 10.1016/j.aei.2026.104411

Bingyi Li , Qian Zhao , Xiang Jia , Guang Jin

Structural reliability analysis (SRA) plays an important role in uncertainty theory and engineering application. A basic problem is estimating failure probability, and a popular idea is sequential strategy. With digital twin and alternative technologies, multiple data sources, considered as different fidelities, are available, including physical experiments and simulations. Multi-fidelity (MF) data include finite high-fidelity (HF) data and abundant low-fidelity (LF) data. MF methods for SRA tend to focus on single aspects to select more samples, and the TC (termination criterion) are fixed, limiting efficiency and prediction accuracy. To address these issues, an MF-Kriging method based on active learning and ensemble learning is proposed herein. First, a MF-Kriging model is constructed using initial LF and HF samples. Second, a multi-fidelity ensemble learning function (MELF) framework determines the position and fidelity of the next ideal sample. This framework provides rich information on learning functions and three factors connecting different fidelities. The TC based on double stability of relative error and solution convergence process balances accuracy and efficiency. Finally, results of five numerical examples and two application cases show that the proposed method can estimate failure probabilities with high efficiencies, high accuracies, and low costs.

结构可靠度分析（SRA）在不确定性理论和工程应用中占有重要地位。一个基本问题是估计故障概率，一个流行的想法是顺序策略。使用数字孪生和替代技术，可以使用多种被认为是不同保真度的数据源，包括物理实验和模拟。多保真度（MF）数据包括有限的高保真度（HF）数据和丰富的低保真度（LF）数据。用于SRA的MF方法往往侧重于单一方面以选择更多的样本，并且TC（终止准则）是固定的，限制了效率和预测精度。为了解决这些问题，本文提出了一种基于主动学习和集成学习的MF-Kriging方法。首先，利用初始LF和HF样本构建MF-Kriging模型。其次，多保真度集成学习函数（MELF）框架确定下一个理想样本的位置和保真度。该框架提供了丰富的学习功能信息和连接不同保真度的三个因素。基于相对误差和解收敛过程双重稳定性的TC平衡了精度和效率。最后，5个数值算例和2个应用实例的结果表明，该方法具有效率高、精度高、成本低等优点。

{"title":"Multi-fidelity Kriging method based on active and ensemble learning for structural reliability analysis","authors":"Bingyi Li , Qian Zhao , Xiang Jia , Guang Jin","doi":"10.1016/j.aei.2026.104411","DOIUrl":"10.1016/j.aei.2026.104411","url":null,"abstract":"<div><div>Structural reliability analysis (SRA) plays an important role in uncertainty theory and engineering application. A basic problem is estimating failure probability, and a popular idea is sequential strategy. With digital twin and alternative technologies, multiple data sources, considered as different fidelities, are available, including physical experiments and simulations. Multi-fidelity (MF) data include finite high-fidelity (HF) data and abundant low-fidelity (LF) data. MF methods for SRA tend to focus on single aspects to select more samples, and the TC (termination criterion) are fixed, limiting efficiency and prediction accuracy. To address these issues, an MF-Kriging method based on active learning and ensemble learning is proposed herein. First, a MF-Kriging model is constructed using initial LF and HF samples. Second, a multi-fidelity ensemble learning function (MELF) framework determines the position and fidelity of the next ideal sample. This framework provides rich information on learning functions and three factors connecting different fidelities. The TC based on double stability of relative error and solution convergence process balances accuracy and efficiency. Finally, results of five numerical examples and two application cases show that the proposed method can estimate failure probabilities with high efficiencies, high accuracies, and low costs.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"71 ","pages":"Article 104411"},"PeriodicalIF":9.9,"publicationDate":"2026-02-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146188823","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

Advancing quality control in off-site construction with large language models enhanced by hybrid retrieval-augmented generation 利用混合检索增强生成增强的大型语言模型推进非现场施工的质量控制

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

Advanced Engineering Informatics

Pub Date : 2026-02-07 DOI: 10.1016/j.aei.2026.104381

Fanfan Meng, Mi Pan

Quality control (QC) is critical for off-site construction (OSC), but it still relies heavily on the knowledge and expertise of inspectors. Projects face challenges involving heterogeneous and fragmented knowledge from multiple stakeholders across different stages, compounded by skilled labor shortage, subjective biases, and human errors. A consistent and reliable approach is needed to guide knowledge-informed QC in OSC, yet currently lacking. This paper aims to develop a novel knowledge-driven framework for advancing off-site construction quality control, empowered by hybrid retrieval-augmented generation (hybrid RAG)-enhanced large language models (LLMs). The hybrid RAG employs a prompt-based approach for entity and relationship extraction to support automated graph construction from unstructured knowledge. Then, a semantic alignment approach is designed to align dense retrieval, sparse retrieval, and subgraph traversal for vector-graph hybrid retrieval, thereby enabling the LLMs to generate more reliable outputs for complex QC decision-making scenarios. Comparative analysis against baseline RAG was conducted on three designed use cases, containing quality information retrieval, quality compliance checking, and quality control task guidance, using three broadly used open-source LLMs, namely DeepSeek-R1-14B, GPT-OSS-20B, and Qwen3-14B. The results demonstrate the superiority of the proposed hybrid RAG in significantly improving model response accuracy, trustworthiness and reliability. This study further demonstrates that medium-sized LLMs can effectively address complex retrieval and generation tasks guided by appropriate approaches. The findings of this study offer valuable insights for advancing construction QC practices, and inform future research in consistent and reliable knowledge retrieval for addressing knowledge-intensive tasks in the architecture, engineering and construction industry.

质量控制（QC）对非现场施工（OSC）至关重要，但它仍然严重依赖于检查员的知识和专业知识。项目面临的挑战包括来自不同阶段的多个涉众的异构和碎片化的知识，以及熟练劳动力短缺、主观偏见和人为错误。在OSC中，需要一个一致和可靠的方法来指导知识灵通的QC，但目前缺乏。本文旨在通过混合检索-增强生成(hybrid RAG)-增强的大型语言模型（llm），开发一种新的知识驱动框架，用于推进非现场施工质量控制。混合RAG采用基于提示的方法进行实体和关系提取，以支持从非结构化知识中自动构建图形。然后，设计了一种语义对齐方法，对向量图混合检索中的密集检索、稀疏检索和子图遍历进行对齐，从而使llm能够为复杂的QC决策场景生成更可靠的输出。采用DeepSeek-R1-14B、GPT-OSS-20B和Qwen3-14B三种广泛使用的开源llm，对包含质量信息检索、质量符合性检查和质量控制任务指导的三个设计用例与基线RAG进行对比分析。结果表明，该算法显著提高了模型响应精度、可信度和可靠性。本研究进一步证明，在适当的方法指导下，中型llm可以有效地解决复杂的检索和生成任务。本研究的发现为推进建筑质量控制实践提供了有价值的见解，并为未来的研究提供了一致和可靠的知识检索，以解决建筑、工程和建筑行业的知识密集型任务。

{"title":"Advancing quality control in off-site construction with large language models enhanced by hybrid retrieval-augmented generation","authors":"Fanfan Meng, Mi Pan","doi":"10.1016/j.aei.2026.104381","DOIUrl":"10.1016/j.aei.2026.104381","url":null,"abstract":"<div><div>Quality control (QC) is critical for off-site construction (OSC), but it still relies heavily on the knowledge and expertise of inspectors. Projects face challenges involving heterogeneous and fragmented knowledge from multiple stakeholders across different stages, compounded by skilled labor shortage, subjective biases, and human errors. A consistent and reliable approach is needed to guide knowledge-informed QC in OSC, yet currently lacking. This paper aims to develop a novel knowledge-driven framework for advancing off-site construction quality control, empowered by hybrid retrieval-augmented generation (hybrid RAG)-enhanced large language models (LLMs). The hybrid RAG employs a prompt-based approach for entity and relationship extraction to support automated graph construction from unstructured knowledge. Then, a semantic alignment approach is designed to align dense retrieval, sparse retrieval, and subgraph traversal for vector-graph hybrid retrieval, thereby enabling the LLMs to generate more reliable outputs for complex QC decision-making scenarios. Comparative analysis against baseline RAG was conducted on three designed use cases, containing quality information retrieval, quality compliance checking, and quality control task guidance, using three broadly used open-source LLMs, namely DeepSeek-R1-14B, GPT-OSS-20B, and Qwen3-14B. The results demonstrate the superiority of the proposed hybrid RAG in significantly improving model response accuracy, trustworthiness and reliability. This study further demonstrates that medium-sized LLMs can effectively address complex retrieval and generation tasks guided by appropriate approaches. The findings of this study offer valuable insights for advancing construction QC practices, and inform future research in consistent and reliable knowledge retrieval for addressing knowledge-intensive tasks in the architecture, engineering and construction industry.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"71 ","pages":"Article 104381"},"PeriodicalIF":9.9,"publicationDate":"2026-02-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146188697","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

Reinforcement learning-based dynamic slot allocation in container yards considering container cluster strategy 考虑集装箱集群策略的基于强化学习的集装箱堆场动态槽位分配

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

Advanced Engineering Informatics

Pub Date : 2026-02-07 DOI: 10.1016/j.aei.2026.104414

Dong Tang , Chengji Liang , Rui Zhao , Yu Wang , Panlong Wang , Fengfeng Gu , Lu Han

The continuous growth in global trade places immense pressure on port logistics systems, making the operational efficiency of container terminals a critical determinant of overall port performance. A key challenge is the real-time slot allocation for import containers, which directly impacts terminal productivity by requiring a delicate balance between rapid vessel unloading and the minimization of future rehandling. This complex decision must account for dynamic yard conditions (e.g., equipment status, congestion) and long-term uncertainties (e.g., container pickup times). Conventional heuristic methods often prove inadequate for this large-scale, high-dimensional optimization problem. To address these limitations, this paper proposes an analytical method that integrates a novel Container Cluster strategy with a Hierarchical Deep Reinforcement Learning (HDRL) framework to solve the dynamic slot allocation problem. The problem is modeled as a multi-stage Markov Decision Process (MDP), and a two-level “Block-Slot” HDRL architecture is constructed. In this architecture, the high-level agent employs a Deep Q-Network (DQN) to select the target block, while the low-level agent uses the Deep Deterministic Policy Gradient (DDPG) algorithm to precisely locate the specific slot. Extensive simulations were conducted to validate our method against rule-based heuristics and other reinforcement learning benchmarks. The results demonstrate that our proposed framework significantly reduces container rehandling while maintaining high operational throughput. Furthermore, it exhibits superior convergence speed and decision stability. This research provides a novel technical approach for intelligent container terminal management, holding significant potential to enhance overall port operational efficiency.

全球贸易的持续增长给港口物流系统带来了巨大的压力，使集装箱码头的运营效率成为港口整体绩效的关键决定因素。一个关键的挑战是进口集装箱的实时槽位分配，这直接影响到码头的生产力，因为需要在快速卸船和最小化未来再处理之间取得微妙的平衡。这种复杂的决策必须考虑动态堆场条件（如设备状态、拥堵）和长期不确定性（如集装箱取货时间）。传统的启发式方法常常被证明不适用于这种大规模、高维的优化问题。为了解决这些限制，本文提出了一种将新的容器集群策略与层次深度强化学习（HDRL）框架相结合的分析方法来解决动态槽分配问题。将该问题建模为多阶段马尔可夫决策过程（MDP），构造了两层“块槽”HDRL体系结构。在该体系结构中，高级代理使用深度Q-Network （DQN）来选择目标块，而低级代理使用深度确定性策略梯度（DDPG）算法来精确定位特定槽。进行了大量的模拟，以验证我们的方法对基于规则的启发式和其他强化学习基准。结果表明，我们提出的框架在保持高操作吞吐量的同时显著减少了集装箱再处理。此外，该算法还具有较好的收敛速度和决策稳定性。本研究为智能集装箱码头管理提供了一种新的技术途径，对提高港口整体运营效率具有重要的潜力。

{"title":"Reinforcement learning-based dynamic slot allocation in container yards considering container cluster strategy","authors":"Dong Tang , Chengji Liang , Rui Zhao , Yu Wang , Panlong Wang , Fengfeng Gu , Lu Han","doi":"10.1016/j.aei.2026.104414","DOIUrl":"10.1016/j.aei.2026.104414","url":null,"abstract":"<div><div>The continuous growth in global trade places immense pressure on port logistics systems, making the operational efficiency of container terminals a critical determinant of overall port performance. A key challenge is the real-time slot allocation for import containers, which directly impacts terminal productivity by requiring a delicate balance between rapid vessel unloading and the minimization of future rehandling. This complex decision must account for dynamic yard conditions (e.g., equipment status, congestion) and long-term uncertainties (e.g., container pickup times). Conventional heuristic methods often prove inadequate for this large-scale, high-dimensional optimization problem. To address these limitations, this paper proposes an analytical method that integrates a novel Container Cluster strategy with a Hierarchical Deep Reinforcement Learning (HDRL) framework to solve the dynamic slot allocation problem. The problem is modeled as a multi-stage Markov Decision Process (MDP), and a two-level “Block-Slot” HDRL architecture is constructed. In this architecture, the high-level agent employs a Deep Q-Network (DQN) to select the target block, while the low-level agent uses the Deep Deterministic Policy Gradient (DDPG) algorithm to precisely locate the specific slot. Extensive simulations were conducted to validate our method against rule-based heuristics and other reinforcement learning benchmarks. The results demonstrate that our proposed framework significantly reduces container rehandling while maintaining high operational throughput. Furthermore, it exhibits superior convergence speed and decision stability. This research provides a novel technical approach for intelligent container terminal management, holding significant potential to enhance overall port operational efficiency.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"71 ","pages":"Article 104414"},"PeriodicalIF":9.9,"publicationDate":"2026-02-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146188691","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 novel hybrid neural network for high-accuracy vehicle-to-infrastructure network traffic prediction 基于混合神经网络的车辆与基础设施网络流量高精度预测

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

Advanced Engineering Informatics

Pub Date : 2026-02-06 DOI: 10.1016/j.aei.2026.104423

Xiaosheng Ni , Jingpu Duan , Xiong Li , Xin Zhang

To address the challenges in Vehicle-to-Infrastructure (V2I) network traffic prediction, this study proposes an innovative solution. We first establish a novel paradigm that integrates physical models to systematically convert publicly available vehicle trajectory data into V2I traffic data. On this basis, a gCNN–BiLSTM–MHA deep learning model is constructed, whose core advantage lies in its use of a lightweight GhostNet-based convolutional network (gCNN) to improve computational efficiency, while leveraging the synergistic effect of a bidirectional long short-term memory network (BiLSTM) and a multi-head attention mechanism (MHA) to effectively balance prediction efficiency and accuracy. The model’s superiority is comprehensively validated: compared to baseline models like LSTM, it demonstrates significant advantages across a series of key evaluation metrics — including running time, MBD, MAE, MAPE, RMSE, and

R^{2}

— achieving an overall balanced performance. Furthermore, the model exhibits excellent performance on multiple benchmark datasets, confirming its strong robustness and high applicability for complex V2I network traffic prediction tasks.

为了应对车辆到基础设施（V2I）网络流量预测中的挑战，本研究提出了一种创新的解决方案。我们首先建立了一个新的范例，该范例集成了物理模型，系统地将公开可用的车辆轨迹数据转换为V2I交通数据。在此基础上，构建了gCNN - BiLSTM - MHA深度学习模型，其核心优势在于利用基于ghostnet的轻量级卷积网络（gCNN）提高计算效率，同时利用双向长短期记忆网络（BiLSTM）和多头注意机制（MHA）的协同效应，有效平衡预测效率和准确性。该模型的优势得到了全面验证：与LSTM等基线模型相比，它在一系列关键评估指标（包括运行时间、MBD、MAE、MAPE、RMSE和R2）上显示出显著优势，实现了整体平衡性能。此外，该模型在多个基准数据集上表现出优异的性能，证实了其较强的鲁棒性和对复杂V2I网络流量预测任务的高适用性。

{"title":"A novel hybrid neural network for high-accuracy vehicle-to-infrastructure network traffic prediction","authors":"Xiaosheng Ni , Jingpu Duan , Xiong Li , Xin Zhang","doi":"10.1016/j.aei.2026.104423","DOIUrl":"10.1016/j.aei.2026.104423","url":null,"abstract":"<div><div>To address the challenges in Vehicle-to-Infrastructure (V2I) network traffic prediction, this study proposes an innovative solution. We first establish a novel paradigm that integrates physical models to systematically convert publicly available vehicle trajectory data into V2I traffic data. On this basis, a gCNN–BiLSTM–MHA deep learning model is constructed, whose core advantage lies in its use of a lightweight GhostNet-based convolutional network (gCNN) to improve computational efficiency, while leveraging the synergistic effect of a bidirectional long short-term memory network (BiLSTM) and a multi-head attention mechanism (MHA) to effectively balance prediction efficiency and accuracy. The model’s superiority is comprehensively validated: compared to baseline models like LSTM, it demonstrates significant advantages across a series of key evaluation metrics — including running time, MBD, MAE, MAPE, RMSE, and <span><math><msup><mrow><mi>R</mi></mrow><mrow><mn>2</mn></mrow></msup></math></span> — achieving an overall balanced performance. Furthermore, the model exhibits excellent performance on multiple benchmark datasets, confirming its strong robustness and high applicability for complex V2I network traffic prediction tasks.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"71 ","pages":"Article 104423"},"PeriodicalIF":9.9,"publicationDate":"2026-02-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146188825","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

Multivariate forecasting of energy demand using recurrent neural networks 基于递归神经网络的能源需求多元预测

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

Advanced Engineering Informatics

Pub Date : 2026-02-06 DOI: 10.1016/j.aei.2026.104419

Mariela N. Uhrig , Leandro D. Vignolo , Omar V. Müller

Accurate short-term electricity demand forecasting remains challenging due to complex relationships between demand and meteorological drivers, lagged effects acting at different temporal scales, and increased uncertainty during peak-demand periods. This paper addresses the design of recurrent neural networks for predicting electricity demand based on multiple input variables. Given the pressing need to enhance energy efficiency and reduce environmental impact, deep learning methods offer a promising approach to address these challenges.

The analysis of daily electricity demand data from the Province of Entre Ríos (Argentina), including a rich set of 21 meteorological, temporal, and energy-related variables reveals significantly higher demand during extreme temperature events, weekdays, and a sustained long-term growth trend. Capturing these patterns is essential for robust operational forecasting. This study proposes a multivariate LSTM-based forecasting model capable of learning complex nonlinear temporal dependencies from heterogeneous inputs. LSTM networks are able to jointly model short- and long-term dynamics without manual feature engineering, while maintaining a reasonable architecture suitable for multi-horizon forecasting.

Results show that the proposed LSTM model clearly outperforms a baseline statistical approach and a state-of-the-art model based on gated recurrent neural networks, providing more accurate and stable forecasts across different prediction horizons, including during peak-demand periods. These findings highlight the effectiveness of parsimonious deep learning architectures for operational electricity demand forecasting and support their adoption in real-world power system planning and decision-making.

由于需求与气象驱动因素之间的复杂关系、不同时间尺度的滞后效应以及高峰需求期间不确定性的增加，准确的短期电力需求预测仍然具有挑战性。本文研究了基于多输入变量的递归神经网络预测电力需求的设计。鉴于提高能源效率和减少环境影响的迫切需要，深度学习方法为解决这些挑战提供了一种有希望的方法。对恩特雷省Ríos（阿根廷）每日电力需求数据的分析，包括一组丰富的21个气象、时间和能源相关变量，表明在极端温度事件、工作日和持续的长期增长趋势期间，需求显著增加。捕获这些模式对于稳健的操作预测至关重要。本研究提出了一种基于多元lstm的预测模型，该模型能够从异构输入中学习复杂的非线性时间依赖性。LSTM网络能够在不需要人工特征工程的情况下联合建模短期和长期动态，同时保持适合多水平预测的合理架构。结果表明，所提出的LSTM模型明显优于基线统计方法和基于门控递归神经网络的最先进模型，在不同的预测范围（包括高峰需求期）提供更准确和稳定的预测。这些发现强调了精简的深度学习架构在运营电力需求预测方面的有效性，并支持将其应用于现实世界的电力系统规划和决策。

{"title":"Multivariate forecasting of energy demand using recurrent neural networks","authors":"Mariela N. Uhrig , Leandro D. Vignolo , Omar V. Müller","doi":"10.1016/j.aei.2026.104419","DOIUrl":"10.1016/j.aei.2026.104419","url":null,"abstract":"<div><div>Accurate short-term electricity demand forecasting remains challenging due to complex relationships between demand and meteorological drivers, lagged effects acting at different temporal scales, and increased uncertainty during peak-demand periods. This paper addresses the design of recurrent neural networks for predicting electricity demand based on multiple input variables. Given the pressing need to enhance energy efficiency and reduce environmental impact, deep learning methods offer a promising approach to address these challenges.</div><div>The analysis of daily electricity demand data from the Province of Entre Ríos (Argentina), including a rich set of 21 meteorological, temporal, and energy-related variables reveals significantly higher demand during extreme temperature events, weekdays, and a sustained long-term growth trend. Capturing these patterns is essential for robust operational forecasting. This study proposes a multivariate LSTM-based forecasting model capable of learning complex nonlinear temporal dependencies from heterogeneous inputs. LSTM networks are able to jointly model short- and long-term dynamics without manual feature engineering, while maintaining a reasonable architecture suitable for multi-horizon forecasting.</div><div>Results show that the proposed LSTM model clearly outperforms a baseline statistical approach and a state-of-the-art model based on gated recurrent neural networks, providing more accurate and stable forecasts across different prediction horizons, including during peak-demand periods. These findings highlight the effectiveness of parsimonious deep learning architectures for operational electricity demand forecasting and support their adoption in real-world power system planning and decision-making.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"71 ","pages":"Article 104419"},"PeriodicalIF":9.9,"publicationDate":"2026-02-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146188122","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

Quality-aware conditional modality gating attention network for robust multimodal underwater propeller diagnosis 基于质量感知的条件模态门控注意网络的多模态水下螺旋桨鲁棒诊断

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

Advanced Engineering Informatics

Pub Date : 2026-02-06 DOI: 10.1016/j.aei.2026.104396

Wenfeng Zhao, Ze Yu, Bo He

Reliable fault diagnosis for autonomous underwater vehicle (AUV) propellers is critical yet highly challenging in complex underwater environments. Diagnostic methods based on single optical video or electrical signals exhibit inherent limitations. Moreover, traditional multimodal fusion strategies suffer from a sharp performance decline when facing dynamic degradation of optical data quality (e.g., blur, low-light) due to the lack of a reliability assessment mechanism. To address this bottleneck, this paper proposes the quality-aware conditional modality gating attention network (QACMANet). This network introduces a dual-dimensional reliability assessment framework that integrates external objective image-quality cues with internal subjective model uncertainty. The mechanism synergistically evaluates data quality and model confidence, adaptively reducing reliance on optical data when it is unreliable and intelligently shifting reliance to the more stable electrical signals. Extensive experiments on a multimodal dataset featuring 11 test conditions demonstrate that QACMANet achieves an average accuracy of 92.86%, outperforming the best baseline by a significant margin of 7.78 percentage points, with the advantage being particularly pronounced under severe visual degradation. This research provides a robust solution for multimodal diagnosis in non-ideal underwater environments and validates the critical value of explicit reliability assessment to enhance the environmental adaptability of the system.

在复杂的水下环境中，自主水下航行器（AUV）螺旋桨的可靠故障诊断至关重要，但也极具挑战性。基于单个光学视频或电信号的诊断方法具有固有的局限性。此外，由于缺乏可靠性评估机制，传统的多模态融合策略在面对光学数据质量的动态退化（如模糊、弱光）时，性能会急剧下降。为了解决这一瓶颈，本文提出了质量感知条件模态门控注意网络（QACMANet）。该网络引入了一个二维可靠性评估框架，该框架集成了外部客观图像质量线索和内部主观模型不确定性。该机制协同评估数据质量和模型置信度，在光数据不可靠时自适应减少对光数据的依赖，并智能地将依赖转移到更稳定的电信号上。在包含11个测试条件的多模态数据集上进行的大量实验表明，QACMANet的平均准确率为92.86%，比最佳基线高出7.78个百分点，在严重的视觉退化情况下优势尤其明显。本研究为非理想水下环境下的多模态诊断提供了鲁棒性解决方案，验证了显式可靠性评估的临界值，提高了系统的环境适应性。

{"title":"Quality-aware conditional modality gating attention network for robust multimodal underwater propeller diagnosis","authors":"Wenfeng Zhao, Ze Yu, Bo He","doi":"10.1016/j.aei.2026.104396","DOIUrl":"10.1016/j.aei.2026.104396","url":null,"abstract":"<div><div>Reliable fault diagnosis for autonomous underwater vehicle (AUV) propellers is critical yet highly challenging in complex underwater environments. Diagnostic methods based on single optical video or electrical signals exhibit inherent limitations. Moreover, traditional multimodal fusion strategies suffer from a sharp performance decline when facing dynamic degradation of optical data quality (e.g., blur, low-light) due to the lack of a reliability assessment mechanism. To address this bottleneck, this paper proposes the quality-aware conditional modality gating attention network (QACMANet). This network introduces a dual-dimensional reliability assessment framework that integrates external objective image-quality cues with internal subjective model uncertainty. The mechanism synergistically evaluates data quality and model confidence, adaptively reducing reliance on optical data when it is unreliable and intelligently shifting reliance to the more stable electrical signals. Extensive experiments on a multimodal dataset featuring 11 test conditions demonstrate that QACMANet achieves an average accuracy of 92.86%, outperforming the best baseline by a significant margin of 7.78 percentage points, with the advantage being particularly pronounced under severe visual degradation. This research provides a robust solution for multimodal diagnosis in non-ideal underwater environments and validates the critical value of explicit reliability assessment to enhance the environmental adaptability of the system.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"71 ","pages":"Article 104396"},"PeriodicalIF":9.9,"publicationDate":"2026-02-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146188821","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

Mitigating class imbalance in deep learning-based multi-class structural damage recognition using an informatics-oriented data augmentation framework 基于信息学的数据增强框架缓解基于深度学习的多类结构损伤识别中的类不平衡

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

Advanced Engineering Informatics

Pub Date : 2026-02-06 DOI: 10.1016/j.aei.2026.104430

Pa Pa Win Aung , Almo Senja Kulinan , Minsoo Park , Dongyoung Ko , Gichun Cha , Seunghee Park

Deep learning has advanced automated structural damage recognition; however, real-world datasets remain severely imbalanced, with cracks dominating while critical defects such as spalling, efflorescence, and leakage are underrepresented. This scarcity leads to model bias and significantly reduces generalization and reliability. To address this issue, this paper proposes an informatics-oriented data augmentation framework that employs high-fidelity virtual environments to systematically balance multi-class structural damage data. The framework integrates class-targeted balancing strategies with customized multilabel annotation to control the generation and distribution of augmented samples, ensuring balanced representation and eliminating labor-intensive manual labeling. Experimental evaluation demonstrates substantial performance improvements for minority classes: achieving F1-scores up to 92.95% for leakage and 87.16% for spalling in segmentation tasks, representing relative improvements of 26.72% and 15.56%, respectively, over the real-only baseline, while maintaining stable accuracy for majority classes. Comparative results further show that algorithmic balancing methods such as Focal Loss struggle under severe imbalance, confirming that informatics-driven data augmentation provides a reliable, scalable, and methodologically essential foundation for robust deep learning in structural health monitoring and smart infrastructure inspection.

深度学习促进了自动结构损伤识别；然而，现实世界的数据集仍然严重不平衡，裂缝占主导地位，而诸如剥落、开花和泄漏等关键缺陷的代表性不足。这种稀缺性导致模型偏差，并显著降低了泛化和可靠性。为了解决这一问题，本文提出了一种面向信息学的数据增强框架，该框架采用高保真虚拟环境来系统地平衡多类结构损伤数据。该框架将针对类别的平衡策略与定制的多标签标注相结合，以控制增强样本的生成和分布，确保平衡表示并消除劳动密集型的人工标注。实验评估表明，少数类的性能有了实质性的提高：在分割任务中，泄漏和剥落的f1得分分别达到了92.95%和87.16%，相对于真实基线分别提高了26.72%和15.56%，同时对大多数类保持了稳定的准确率。对比结果进一步表明，算法平衡方法（如Focal Loss）在严重失衡的情况下难以实现，证实了信息学驱动的数据增强为结构健康监测和智能基础设施检查中的稳健深度学习提供了可靠、可扩展和方法上必不可少的基础。

{"title":"Mitigating class imbalance in deep learning-based multi-class structural damage recognition using an informatics-oriented data augmentation framework","authors":"Pa Pa Win Aung , Almo Senja Kulinan , Minsoo Park , Dongyoung Ko , Gichun Cha , Seunghee Park","doi":"10.1016/j.aei.2026.104430","DOIUrl":"10.1016/j.aei.2026.104430","url":null,"abstract":"<div><div>Deep learning has advanced automated structural damage recognition; however, real-world datasets remain severely imbalanced, with cracks dominating while critical defects such as spalling, efflorescence, and leakage are underrepresented. This scarcity leads to model bias and significantly reduces generalization and reliability. To address this issue, this paper proposes an informatics-oriented data augmentation framework that employs high-fidelity virtual environments to systematically balance multi-class structural damage data. The framework integrates class-targeted balancing strategies with customized multilabel annotation to control the generation and distribution of augmented samples, ensuring balanced representation and eliminating labor-intensive manual labeling. Experimental evaluation demonstrates substantial performance improvements for minority classes: achieving F1-scores up to 92.95% for leakage and 87.16% for spalling in segmentation tasks, representing relative improvements of 26.72% and 15.56%, respectively, over the real-only baseline, while maintaining stable accuracy for majority classes. Comparative results further show that algorithmic balancing methods such as Focal Loss struggle under severe imbalance, confirming that informatics-driven data augmentation provides a reliable, scalable, and methodologically essential foundation for robust deep learning in structural health monitoring and smart infrastructure inspection.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"71 ","pages":"Article 104430"},"PeriodicalIF":9.9,"publicationDate":"2026-02-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146188824","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

Real-time multimodal fusion and semantic mapping for robotic tower crane perception 机器人塔吊感知的实时多模态融合与语义映射

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

Advanced Engineering Informatics

Pub Date : 2026-02-05 DOI: 10.1016/j.aei.2026.104373

Yifan Lu , Xiuzhi Deng , Peter E.D. Love , Wei Zhou , Weili Fang

Robotic tower crane operation requires real-time perception of complex and rapidly changing construction environments. Conventional Simultaneous Localization and Mapping (SLAM) methods assume smooth sensor motion and emphasize geometry over semantics, limiting their suitability for crane-mounted sensing affected by vibration, rotation, and intermittent movement. This research proposes a multimodal perception framework that integrates Light Detection and Ranging (LiDAR), camera, and Inertial Measurement Unit (IMU) data within a tightly coupled fusion and semantic reconstruction pipeline. A Mahony-filter-based attitude optimization module stabilizes high-frequency vibrations, while a Fast LiDAR-Inertial Odometry (FAST-LIVO2)-inspired LiDAR–visual–inertial fusion strategy achieves centimeter-level three-dimensional (3D) mapping. To enhance scene understanding, an improved Random Sampled and Lightweight Aggregated Network (RandLA-Net) jointly exploits geometric and visual cues for point-level semantic segmentation, with color-aware spatial encoding. Field deployment on an operational tower crane demonstrates superior performance, yielding the lowest global reconstruction errors and highest semantic accuracy. The framework provides a robust perception foundation for autonomous planning, safety monitoring, and intelligent lifting assistance.

机器人塔式起重机操作需要实时感知复杂和快速变化的施工环境。传统的同步定位和测绘（SLAM）方法假设传感器运动平滑，强调几何而不是语义，限制了它们对受振动、旋转和间歇性运动影响的起重机安装传感的适用性。本研究提出了一个多模态感知框架，该框架将光探测和测距（LiDAR）、相机和惯性测量单元（IMU）数据集成在一个紧密耦合的融合和语义重建管道中。基于mahoney滤波器的姿态优化模块可稳定高频振动，而受Fast - livo2启发的快速激光雷达-惯性里程计（Fast - livo2）激光雷达-视觉-惯性融合策略可实现厘米级三维（3D）映射。为了增强场景理解，改进的随机采样和轻量级聚合网络（RandLA-Net）结合颜色感知空间编码，共同利用几何和视觉线索进行点级语义分割。在塔式起重机上的现场部署显示出卓越的性能，产生最低的全局重建误差和最高的语义精度。该框架为自主规划、安全监控和智能起重辅助提供了强大的感知基础。

{"title":"Real-time multimodal fusion and semantic mapping for robotic tower crane perception","authors":"Yifan Lu , Xiuzhi Deng , Peter E.D. Love , Wei Zhou , Weili Fang","doi":"10.1016/j.aei.2026.104373","DOIUrl":"10.1016/j.aei.2026.104373","url":null,"abstract":"<div><div>Robotic tower crane operation requires real-time perception of complex and rapidly changing construction environments. Conventional Simultaneous Localization and Mapping (SLAM) methods assume smooth sensor motion and emphasize geometry over semantics, limiting their suitability for crane-mounted sensing affected by vibration, rotation, and intermittent movement. This research proposes a multimodal perception framework that integrates Light Detection and Ranging (LiDAR), camera, and Inertial Measurement Unit (IMU) data within a tightly coupled fusion and semantic reconstruction pipeline. A Mahony-filter-based attitude optimization module stabilizes high-frequency vibrations, while a Fast LiDAR-Inertial Odometry (FAST-LIVO2)-inspired LiDAR–visual–inertial fusion strategy achieves centimeter-level three-dimensional (3D) mapping. To enhance scene understanding, an improved Random Sampled and Lightweight Aggregated Network (RandLA-Net) jointly exploits geometric and visual cues for point-level semantic segmentation, with color-aware spatial encoding. Field deployment on an operational tower crane demonstrates superior performance, yielding the lowest global reconstruction errors and highest semantic accuracy. The framework provides a robust perception foundation for autonomous planning, safety monitoring, and intelligent lifting assistance.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"71 ","pages":"Article 104373"},"PeriodicalIF":9.9,"publicationDate":"2026-02-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146188700","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