Advanced Engineering Informatics最新文献

{"title":"How does contextual fidelity impact how we think, talk, and act in AI-assisted engineering design?","authors":"Shantanu Vyas,&nbsp;Suryapavan Cheruku,&nbsp;Vinayak R. Krishnamurthy","doi":"10.1016/j.aei.2026.104456","DOIUrl":"10.1016/j.aei.2026.104456","url":null,"abstract":"<div><div>In this work, we investigate the role of generative AI agents as reflective partners in engineering design. While such models are increasingly used to generate design solutions, concerns remain about their potential to diminish designers’ critical thinking and reasoning skills. To address this, we develop a mixed-initiative conversational framework that positions large language and vision–language models as reflective thinking partners rather than solution providers. The framework is structured around five contextual information channels, namely task–role context, design representations, historical context, evaluation signals, and target references, that enable AI agents to ask reflective questions and provide explanations and suggestions. To study this framework within a concrete design context, we develop an interactive tool that embodies the notion of contextual fidelity of 2D structure design tasks. We implement varying levels of contextual fidelity, defined by the extent of contextual information available to the agent. We evaluate these levels through a between-subjects study with forty-six participants, comparing a high-fidelity and a low-fidelity agent against a control group without AI support. We examine the impact of the agents on how users think, talk and act, using a comprehensive set of metrics, including coarse-level design objectives (deformation and material usage), solution quality metrics (structural and geometric analysis), process-oriented measures (design space exploration patterns and trajectories, design strategy shifts), conversational dynamics (thematic and temporal analysis), and subjective surveys (NASA-TLX, Cognitive Load Theory, Trust in AI). Our analyses show that while conversational agents do not immediately help improve coarse-level design objectives, they significantly shape nuanced aspects of design processes and outcomes. Interaction with the agents critically influences how users explore the design space, where agent-supported groups exhibited more focused exploration patterns compared to the control group’s broader trial-and-error approaches. Furthermore, interactions with the high-fidelity agent led to solutions with higher symmetry and topological alignment with optimal designs, fostered deeper reflection, reduced mental demand, and supported more deliberate design decisions. Building on these findings, we discuss broader implications of AI agents for problem-solving processes and outline guidelines for designing adaptive and generalizable frameworks for different domains.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"72 ","pages":"Article 104456"},"PeriodicalIF":9.9,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146154282","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}

{"title":"IDS-Net: A novel framework for few-shot photovoltaic power prediction with interpretable dynamic selection and feature information fusion","authors":"Hang Fan ,&nbsp;Weican Liu ,&nbsp;Zuhan Zhang ,&nbsp;Ying Lu ,&nbsp;Wencai Run ,&nbsp;Dunnan Liu","doi":"10.1016/j.aei.2026.104455","DOIUrl":"10.1016/j.aei.2026.104455","url":null,"abstract":"<div><div>With the growing demand for renewable energy, countries are accelerating the construction of photovoltaic (PV) power stations. However, accurately forecasting power data for newly constructed PV stations is extremely challenging due to limited data availability. To this end, we propose a novel interpretable dynamic selection network (IDS-Net) based on feature information fusion to achieve accurate few-shot prediction. This transfer learning framework primarily consists of two parts. In the first stage, we pre-train on the large dataset, utilizing Maximum Mean Discrepancy (MMD) to select the source domain dataset most similar to the target domain data distribution. Subsequently, the ReliefF algorithm is utilized for feature selection, reducing the influence of feature redundancy. Then, the Hampel Identifier (HI) is used for training dataset outlier correction. In the IDS-Net model, we first obtain the initial extracted features from a pool of predictive models. Following this, two separate weighting channels are utilized to determine the interpretable weights for each sub-model and the adaptive selection outcomes, respectively. Subsequently, the extracted feature results from each sub-model are multiplied by their corresponding weights and then summed to obtain the weighted extracted features. Then, we perform cross-embedding on the additional features and fuse them with the extracted weighted features. This fused information is then passed through the MLP (Multi-Layer Perceptron) layer to obtain predictions. In the second stage, we design an end-to-end adaptive transfer learning strategy to obtain the final prediction results on the target dataset. We validate the transfer learning process using two PV power datasets from Hebei province, China, to demonstrate the effectiveness and generalization of our framework and transfer learning strategy.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"72 ","pages":"Article 104455"},"PeriodicalIF":9.9,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146154143","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}

{"title":"A unified LLM-KG framework for low‑annotation urban rail transit signal system operation: knowledge acquisition and dynamic update","authors":"Wei Cai ,&nbsp;Xiaomin Zhu ,&nbsp;Zeyu Sun ,&nbsp;Aihui Ye ,&nbsp;Guanhua Fu ,&nbsp;Runtong Zhang","doi":"10.1016/j.aei.2026.104327","DOIUrl":"10.1016/j.aei.2026.104327","url":null,"abstract":"<div><div>Intelligent operation and maintenance (O&amp;M) of urban rail transit signal systems (URTSS) is essential for ensuring train safety and operational efficiency. However, most O&amp;M data exist as unstructured and sparsely labeled texts, posing major challenges for reliable knowledge extraction, semantic reasoning, and dynamic knowledge management. To address these issues, this paper proposes a unified large language model-knowledge graph framework (ULLM-KG) tailored for low-annotation, knowledge-intensive O&amp;M environments. Firstly, a bidirectional knowledge graph construction mechanism (BKGC) is introduced to rapidly build a domain-specific initial knowledge graph. Secondly, a KG-enhanced distantly supervised entity and event extraction method (KG-DS3E) is designed to enhance critical knowledge extraction accuracy from unstructured texts. Thirdly, a prompt-driven knowledge-enhanced reasoning method (PD-KER) is proposed to improve semantic quality in fault diagnosis and maintenance recommendations. Lastly, a dynamic knowledge graph updating mechanism with temporal awareness and conflict resolution (DKG-UCF) is used to ensure efficient and accurate knowledge evolution. Based on real-world URTSS O&amp;M data, experimental evaluations are conducted on state-of-the-art LLMs (GPT-4o, DeepSeek-V3, and Qwen3-32B). On datasets with varying annotation ratios and rare faults, ULLM-KG demonstrates significantly superior performance in knowledge extraction and reasoning tasks compared to other state-of-the-art methods. Its ability to dynamically update knowledge is also verified to be excellent. ULLM-KG provides a general solution for the intelligent O&amp;M of URTSS under low-annotation conditions.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"71 ","pages":"Article 104327"},"PeriodicalIF":9.9,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146078379","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}

{"title":"Tunable plasmonic absorption in metal–dielectric multilayers via FDTD simulations and an explainable machine learning approach","authors":"Emmanuel A. Bamidele","doi":"10.1016/j.aei.2026.104311","DOIUrl":"10.1016/j.aei.2026.104311","url":null,"abstract":"<div><div>Plasmonic devices, fundamental to modern nanophotonics, exploit resonant interactions between light and free electrons in metals to achieve enhanced light trapping and electromagnetic field confinement. However, modeling their complex, nonlinear optical responses remain computationally intensive. In this work, we combine finite-difference time-domain (FDTD) simulations with machine learning (ML) to simulate and predict absorbed power behavior in multilayer plasmonic stacks composed of SiO₂, gold (Au), silver (Ag), and indium tin oxide (ITO). By varying Au and Ag thicknesses (10–50  nm) across a spectral range of 300–1500  nm, we generate spatial absorption maps and integrated power metrics from full-wave solutions to Maxwell’s equations. A multilayer perceptron models global absorption behavior with a mean absolute error (MAE) of 0.0953, while a convolutional neural network predicts spatial absorption distributions with an MAE of 0.0101. SHapley Additive exPlanations identify plasmonic layer thickness and excitation wavelength as dominant contributors to absorption, which peaks between 450 and 850  nm. Gold demonstrates broader and more sustained absorption compared to silver, although both metals exhibit reduced efficiency outside the resonance window. The integrated FDTD–ML framework accelerates plasmonic design while maintaining physical interpretability and predictive accuracy, enabling efficient exploration of tunable optical responses in multilayer nanophotonic systems for applications in optical sensing, photovoltaics, and device engineering.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"71 ","pages":"Article 104311"},"PeriodicalIF":9.9,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145926900","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}

{"title":"TSDR-SFE: A prediction model for dam crack width based on two-stage decomposition–reconstruction and spatiotemporal feature extraction","authors":"Yixiang Fang ,&nbsp;Chongshi Gu ,&nbsp;Yangtao Li ,&nbsp;Yiming Wang ,&nbsp;Taiqi Lu ,&nbsp;Lei Shen ,&nbsp;Xiao Sun ,&nbsp;Mingyuan Zhu ,&nbsp;Fuqiang Zhou ,&nbsp;Sitao Fu ,&nbsp;Hao Gu","doi":"10.1016/j.aei.2025.104290","DOIUrl":"10.1016/j.aei.2025.104290","url":null,"abstract":"<div><div>Crack formation and propagation in concrete dams present substantial risks to long-term structural integrity and operational stability. Accurate crack-width prediction models are thus essential for ensuring the continued safe operation of these structures. However, crack width monitoring data are highly nonlinear and non-stationary, limiting the effectiveness of traditional single-technique signal decomposition methods in capturing their complex time–frequency dynamics. To overcome these challenges, this study proposes a predictive model (TSDR-SFE) that integrates two-stage decomposition–reconstruction with spatiotemporal feature extraction. The model begins by applying improved empirical mode decomposition (IEMD) to decompose the crack width time series into intrinsic mode functions (IMFs). Then, spatial post-multiscale fusion entropy (SPMFE) is used to compute the entropy value of each IMF, which subsequently serves as input to the k-Graph clustering algorithm. By constructing a graph-based structure, IMFs are classified and reconstructed into stochastic, periodic, and trend components. Next, the stochastic component, due to its higher complexity, undergoes a second decomposition using successive variational mode decomposition (SVMD). Finally, the trend and periodic components, along with the subsequences of the stochastic component obtained, are separately predicted using the TSMixer, which effectively extracts spatiotemporal features. The final prediction is obtained by aggregating the prediction results of the three components. Using monitoring data collected from seven crack-width meters installed at multiple locations on a concrete dam, the accuracy and generalization capability of TSDR-SFE are evaluated by comparing its predictive performance with ten ablation models and twelve benchmark models. The experimental results show that TSDR-SFE consistently outperforms all comparative models in both fitting accuracy and predictive performance, achieving coefficient of determination (R²) values above 0.97 and exhibiting the most compact residual distributions. These findings confirm that the layered strategy of decomposition, clustering, secondary decomposition, and modeling effectively reduces the complexity of non-stationary time series. This reduction facilitates the extraction of deeper knowledge from simplified components. It provides a robust theoretical basis for diagnosing crack evolution patterns and guiding engineering safety decisions.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"71 ","pages":"Article 104290"},"PeriodicalIF":9.9,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145926894","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}

{"title":"An improved tuna swarm optimization with dimension learning-based hunting for global optimization and real-world engineering applications","authors":"Eda Özkul","doi":"10.1016/j.aei.2025.104298","DOIUrl":"10.1016/j.aei.2025.104298","url":null,"abstract":"<div><div>This study proposes an improved tuna swarm optimization algorithm (I-TSO) for solving global optimization and engineering design problems. However, despite its strong global search ability, tuna swarm optimization (TSO) suffers from trapping in local optima, having premature convergence, and the loss of diversity in the early stage. To eliminate these disadvantages and improve the original TSO, the proposed I-TSO algorithm uses a dimension learning-based hunting (DLH) strategy. DLH constructs a neighborhood for each tuna in the population and uses that information in the optimization process. Thus, it improves population diversity, provides a proper balance between exploration and exploitation, and prevents trapping into local optima. The performance of the proposed algorithm is evaluated on 23 classical benchmark functions, CEC-2017, CEC-2020, and CEC-2022 test suites, and compared it with eight other optimization algorithms. Comparative results demonstrate that I-TSO exhibits stable and effective optimization capabilities. Further, the Friedman test and Wilcoxon signed-rank test are conducted to statistically evaluate the performance of the proposed algorithm, and thus its superiority is statistically confirmed. Moreover, the applicability of the I-TSO in real-world problems is validated on eight engineering design problems. Consequently, the I-TSO algorithm is capable of solving both numerical and engineering design problems with its efficient and superior performance.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"71 ","pages":"Article 104298"},"PeriodicalIF":9.9,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145926970","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}

{"title":"Evolution of text mining in construction industry: an LLM-driven analysis of statistical machine learning dominance and internal-external delayed LLM adoption","authors":"Jinjing Li ,&nbsp;Xianbo Zhao ,&nbsp;Haizhe Yu ,&nbsp;Lili Gao ,&nbsp;Xiaopeng Deng ,&nbsp;Bon-Gang Hwang","doi":"10.1016/j.aei.2026.104321","DOIUrl":"10.1016/j.aei.2026.104321","url":null,"abstract":"<div><div>Over 85 % of construction data remains unstructured, creating urgent needs for text mining (TM). While considerable attention has been directed toward the evolution of TM, a critical gap persists in the form of diachronic analysis, with limited exploration of its trajectory in the context of large language models (LLMs). Hence, this research aims to: (1) generate the LLM-based TM framework for construction; (2) explore different evolutions of TM methods in construction; and (3) identify the driving factors for the evolution. To achieve these objectives, two LLM-based TM methods were used to review the TM-related literature. The results reveal a dual delay pattern: internally, statistical machine learning maintains dominance over LLMs in the construction industry, while externally, LLM adoption lags 2–3 years behind sectors such as healthcare and biomedicine. The study extends existing taxonomies by introducing novel data sources (elicited discourse corpora and multimodal data) and establishing software-based analysis as a distinct methodological stage. Moreover, it addresses the research paradigm gap for LLM-based TM, offering enhanced strategic guidance for practitioners in selecting TM tools.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"71 ","pages":"Article 104321"},"PeriodicalIF":9.9,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145926969","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}

{"title":"Vision–proprioception fusion with Mamba2 in end-to-end reinforcement learning for motion control","authors":"Xiaowen Tao ,&nbsp;Yinuo Wang ,&nbsp;Jinzhao Zhou","doi":"10.1016/j.aei.2026.104389","DOIUrl":"10.1016/j.aei.2026.104389","url":null,"abstract":"<div><div>End-to-end reinforcement learning (RL) for motion control trains policies directly from sensor inputs to motor commands, enabling unified controllers for different robots and tasks. However, most existing methods are either <em>blind</em> (proprioception-only) or rely on fusion backbones with unfavorable compute–memory trade-offs. Recurrent controllers struggle with long-horizon credit assignment, and Transformer-based fusion incurs quadratic cost in token length, limiting temporal and spatial context. We present a vision-driven cross-modal RL framework built on <em>SSD-Mamba2</em>, a selective state–space backbone that applies <em>state–space duality (SSD)</em> to enable both recurrent and convolutional scanning with hardware-aware streaming and near-linear scaling. Proprioceptive states and exteroceptive observations (e.g., depth tokens) are encoded into compact tokens and fused by stacked SSD-Mamba2 layers. The selective state–space updates retain long-range dependencies with markedly lower latency and memory use than quadratic self-attention, enabling longer look-ahead, higher token resolution, and stable training under limited compute. Policies are trained end-to-end under curricula that randomize terrain and appearance and progressively increase scene complexity. A compact, state-centric reward balances task progress, energy efficiency, and safety. Across diverse motion-control scenarios, our approach consistently surpasses strong state-of-the-art baselines in return, safety (collisions and falls), and sample efficiency, while converging faster at the same compute budget. These results suggest that SSD-Mamba2 provides a practical fusion backbone for resource-constrained robotic and autonomous systems in engineering informatics applications.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"71 ","pages":"Article 104389"},"PeriodicalIF":9.9,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146077879","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"FACTS: Training-free zero-shot diffusion framework for facade texture restoration in 3D urban models","authors":"Juexiao Cheng ,&nbsp;Xiangru Huang ,&nbsp;Guanzhou Chen ,&nbsp;Tong Wang ,&nbsp;Jiaqi Wang ,&nbsp;Xiaoliang Tan ,&nbsp;Aiyi Jiang ,&nbsp;Xiaodong Zhang","doi":"10.1016/j.aei.2026.104385","DOIUrl":"10.1016/j.aei.2026.104385","url":null,"abstract":"<div><div>High-fidelity facade texture restoration is crucial for the realism and utility of 3D urban models in digital twin applications. Low-quality textures can compromise visualization, simulation accuracy, and decision-making. This challenge is particularly evident in Level of Detail 1 and 2 (LoD-1 and LoD-2) models, which represent buildings as basic massing models. In these models, textures baked from complex 3D mesh sources often suffer from geometric distortions, occlusions, and inconsistent illumination. To address these issues, we introduce FACTS (Facade Automated Correction and Texture Synthesis), a novel zero-shot, training-free framework for facade texture restoration. FACTS operates as an automated pipeline, taking 3D Mesh as input and producing geometrically and photometrically corrected models. Its key innovations are as follows: (1) a prompt-guided, occlusion-aware inpainting module that uses semantic guidance to repair missing texture regions; (2) a multi-scale edge-feature-guided diffusion process that enforces geometric consistency by leveraging structural priors extracted from the image; and (3) an efficient illumination harmonization method in the CIELAB color space to resolve lighting inconsistencies across texture patches. Recognizing that conventional metrics fail to assess architectural integrity, we propose three novel metrics: the Edge Straightness Score (ESS), Hough Transform Line Consistency (HTLC), and Linearity Index (LI). Our experiments on the SFDB and RUF-3D datasets show significant improvements over baselines. Specifically, FACTS improved ESS, HTLC, and LI scores on degraded textures by 40.69%, 11.16%, and 54.76%, respectively. The framework processes 2.5-megapixel texture in approximately 58.8 s on a single consumer-grade GPU. This work provides a scalable and interpretable solution for the automated restoration of defective facade textures, thereby enhancing the visual realism and structural accuracy of existing 3D urban models. Code and data available at <span><span>https://github.com/CVEO/FACTS</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"71 ","pages":"Article 104385"},"PeriodicalIF":9.9,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146023021","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}

{"title":"Data-driven modelling of unloading hours using explainable gradient boosting models","authors":"Celal Cakiroglu ,&nbsp;Najat Almasarwah ,&nbsp;Mehmet Hakan Özdemir ,&nbsp;Batin Latif Aylak ,&nbsp;Manjeet Singh ,&nbsp;Muhammet Deveci","doi":"10.1016/j.aei.2026.104353","DOIUrl":"10.1016/j.aei.2026.104353","url":null,"abstract":"<div><div>Unloading processes denote the extraction of finished goods and raw materials from transport units and their subsequent conveyance to designated locations. The efficiency of unloading processes is vital in supply chain and logistics management, regarded as an essential component. Delays in unloading operations result in numerous challenges, including heightened operational expenses, diminished labour efficiency, and supply chain bottlenecks. Consequently, it is essential to ascertain unloading times beforehand to mitigate these challenges, resulting in diminished idle time, enhanced overall efficiency, and optimized scheduling. Therefore, precise prediction of unloading times is critically significant. The novelty of this study lies in the application of machine learning techniques to improve operational efficiency by accurately predicting unloading time. To that end, this study employed LightGBM and XGBoost to predict the unloading time in a real case. The unloading time can be predicted with <span><math><msup><mrow><mi>R</mi></mrow><mo>2</mo></msup></math></span> score greater than 0.99 utilizing both models. Subsequently, the SHapley Additive exPlanations (SHAP) methodology was used to ascertain how each input feature contributed to the model’s output. The load of leg significantly influences the unloading time more than the gross weight of truck and the leg distance.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"71 ","pages":"Article 104353"},"PeriodicalIF":9.9,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146023023","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}