Pub Date : 2026-03-01Epub Date: 2026-01-27DOI: 10.1016/j.autcon.2026.106792
Jinxin Yi , Xuan Kong , Hao Tang , Jie Zhang , Zhenming Chen , Lu Deng
Recent advances in computer vision have provided new solutions for intelligent welding. However, existing vision-based weld seam extraction techniques exhibit limited adaptability to various workpieces in unstructured environments. Therefore, this paper proposes a three-dimensional vision-based method tailored for weld seam extraction and path generation. The proposed method synergizes a deep learning-based point cloud segmentation technique with an improved multi-scale point cloud registration algorithm to reconstruct the complete point cloud model of all weld regions in the workpieces. Subsequently, the welding paths and torch poses are calculated using an optimized multi-plane fitting algorithm integrated with geometry model of weld seam. Experimental validation on four workpieces demonstrates that the proposed method achieves good accuracy and outperforms the existing techniques in terms of efficiency and applicability, offering a robust solution for automated welding of steel structures.
{"title":"Weld seam extraction and path generation for robotic welding of steel structures based on 3D vision","authors":"Jinxin Yi , Xuan Kong , Hao Tang , Jie Zhang , Zhenming Chen , Lu Deng","doi":"10.1016/j.autcon.2026.106792","DOIUrl":"10.1016/j.autcon.2026.106792","url":null,"abstract":"<div><div>Recent advances in computer vision have provided new solutions for intelligent welding. However, existing vision-based weld seam extraction techniques exhibit limited adaptability to various workpieces in unstructured environments. Therefore, this paper proposes a three-dimensional vision-based method tailored for weld seam extraction and path generation. The proposed method synergizes a deep learning-based point cloud segmentation technique with an improved multi-scale point cloud registration algorithm to reconstruct the complete point cloud model of all weld regions in the workpieces. Subsequently, the welding paths and torch poses are calculated using an optimized multi-plane fitting algorithm integrated with geometry model of weld seam. Experimental validation on four workpieces demonstrates that the proposed method achieves good accuracy and outperforms the existing techniques in terms of efficiency and applicability, offering a robust solution for automated welding of steel structures.</div></div>","PeriodicalId":8660,"journal":{"name":"Automation in Construction","volume":"183 ","pages":"Article 106792"},"PeriodicalIF":11.5,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146071953","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}
Pub Date : 2026-03-01Epub Date: 2026-01-19DOI: 10.1016/j.autcon.2026.106780
Zhikai Su , Mengnan Shi , Tianyu Gao , Jiaqi Hao , Hongtao Li , Qiang Yao
Infrared imaging is effective for pavement-crack detection under low-illumination conditions, but the scarcity of infrared datasets hinders its broader adoption. This paper proposes a Physics-Informed Diffusion Model to convert readily available visible-light crack images into physically consistent infrared images. The model integrates physical constraints within a Latent Diffusion Model and employs a Channel-Adaptive Dynamic Gamma Correction (CDGC) method to enhance thermally relevant feature representation. Experiments on a ground-truth infrared test set demonstrate that synthetic data generated by the proposed method substantially improves segmentation performance, achieving Pixel Accuracy (PA) of 0.9678 and Frequency-Weighted IoU (FW-IoU) of 0.9459. By obviating the costly, labor-intensive process of infrared dataset collection, the proposed approach facilitates the widespread adoption of infrared machine vision and visible–infrared fusion systems.
{"title":"Physics-informed diffusion for visible-to-infrared domain translation of pavement crack images","authors":"Zhikai Su , Mengnan Shi , Tianyu Gao , Jiaqi Hao , Hongtao Li , Qiang Yao","doi":"10.1016/j.autcon.2026.106780","DOIUrl":"10.1016/j.autcon.2026.106780","url":null,"abstract":"<div><div>Infrared imaging is effective for pavement-crack detection under low-illumination conditions, but the scarcity of infrared datasets hinders its broader adoption. This paper proposes a Physics-Informed Diffusion Model to convert readily available visible-light crack images into physically consistent infrared images. The model integrates physical constraints within a Latent Diffusion Model and employs a Channel-Adaptive Dynamic Gamma Correction (CDGC) method to enhance thermally relevant feature representation. Experiments on a ground-truth infrared test set demonstrate that synthetic data generated by the proposed method substantially improves segmentation performance, achieving Pixel Accuracy (PA) of 0.9678 and Frequency-Weighted IoU (FW-IoU) of 0.9459. By obviating the costly, labor-intensive process of infrared dataset collection, the proposed approach facilitates the widespread adoption of infrared machine vision and visible–infrared fusion systems.</div></div>","PeriodicalId":8660,"journal":{"name":"Automation in Construction","volume":"183 ","pages":"Article 106780"},"PeriodicalIF":11.5,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146000931","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}
Pub Date : 2026-03-01Epub Date: 2026-01-24DOI: 10.1016/j.autcon.2026.106799
Difeng Hu , You Dong , Mingkai Li , Hanmo Wang , Tao Wang
BIM-derived point clouds are valuable for semantic segmentation and BIM modeling, but distribution discrepancies between BIM and real-world scans significantly degrade segmentation performance. To mitigate this issue, this paper develops a margin-aware maximum classifier discrepancy (MMCD) method, which extends the conventional MCD framework by incorporating a margin-aware mechanism. Task-specific classifiers act as discriminators to encourage the feature generator to learn domain-invariant yet discriminative features for unlabeled real point clouds, improving BIM-to-scan distribution alignment and segmentation accuracy. A margin-aware discrepancy loss is formulated to enforce sufficient margin between features and classification boundaries, improving robustness to domain shift. In addition, a training strategy is proposed to support MMCD optimization. Finally, a refined RandLA-Net with an attention-based upsampling module is constructed as the backbone for validation. Experiments demonstrate that the proposed approach achieves superior performance, with an IoU of 72.79% and an overall accuracy of 87.99%, outperforming RandLA-Net variants with or without MCD.
{"title":"Margin-aware maximum classifier discrepancy for BIM-to-scan semantic segmentation of building point clouds","authors":"Difeng Hu , You Dong , Mingkai Li , Hanmo Wang , Tao Wang","doi":"10.1016/j.autcon.2026.106799","DOIUrl":"10.1016/j.autcon.2026.106799","url":null,"abstract":"<div><div>BIM-derived point clouds are valuable for semantic segmentation and BIM modeling, but distribution discrepancies between BIM and real-world scans significantly degrade segmentation performance. To mitigate this issue, this paper develops a margin-aware maximum classifier discrepancy (MMCD) method, which extends the conventional MCD framework by incorporating a margin-aware mechanism. Task-specific classifiers act as discriminators to encourage the feature generator to learn domain-invariant yet discriminative features for unlabeled real point clouds, improving BIM-to-scan distribution alignment and segmentation accuracy. A margin-aware discrepancy loss is formulated to enforce sufficient margin between features and classification boundaries, improving robustness to domain shift. In addition, a training strategy is proposed to support MMCD optimization. Finally, a refined RandLA-Net with an attention-based upsampling module is constructed as the backbone for validation. Experiments demonstrate that the proposed approach achieves superior performance, with an IoU of 72.79% and an overall accuracy of 87.99%, outperforming RandLA-Net variants with or without MCD.</div></div>","PeriodicalId":8660,"journal":{"name":"Automation in Construction","volume":"183 ","pages":"Article 106799"},"PeriodicalIF":11.5,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146036034","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}
Pub Date : 2026-03-01Epub Date: 2026-02-04DOI: 10.1016/j.autcon.2026.106807
Ya-Dong Xue , Fei Jia , Wei Luo , Dong-Mei Zhang , Jie Liu , Yong-Fa Guo
With the rapid expansion of large-scale shield tunnel operations, deep learning has been extensively studied for automated defect recognition. This paper provides a comprehensive review of recent research in deep learning-based methods for tunnel defect recognition, organized into three key stages: dataset establishment, model development, and practical implementation. The review first details the acquisition and preprocessing of tunnel lining images obtained from various inspection equipment, followed by the establishment of defect datasets. It then provides a systematic overview of commonly used deep learning models for defect recognition, with a focus on three primary areas: defect detection, semantic, and instance segmentation, summarizing key innovations within each domain. Based on this analysis, current challenges are identified and future research directions are discussed for each stage. This review aims to promote the practical application of deep learning in tunnel engineering and to support the development of predictive and intelligent maintenance for shield tunnels.
{"title":"Deep learning-based computer vision methods for shield tunnel defect recognition","authors":"Ya-Dong Xue , Fei Jia , Wei Luo , Dong-Mei Zhang , Jie Liu , Yong-Fa Guo","doi":"10.1016/j.autcon.2026.106807","DOIUrl":"10.1016/j.autcon.2026.106807","url":null,"abstract":"<div><div>With the rapid expansion of large-scale shield tunnel operations, deep learning has been extensively studied for automated defect recognition. This paper provides a comprehensive review of recent research in deep learning-based methods for tunnel defect recognition, organized into three key stages: dataset establishment, model development, and practical implementation. The review first details the acquisition and preprocessing of tunnel lining images obtained from various inspection equipment, followed by the establishment of defect datasets. It then provides a systematic overview of commonly used deep learning models for defect recognition, with a focus on three primary areas: defect detection, semantic, and instance segmentation, summarizing key innovations within each domain. Based on this analysis, current challenges are identified and future research directions are discussed for each stage. This review aims to promote the practical application of deep learning in tunnel engineering and to support the development of predictive and intelligent maintenance for shield tunnels.</div></div>","PeriodicalId":8660,"journal":{"name":"Automation in Construction","volume":"183 ","pages":"Article 106807"},"PeriodicalIF":11.5,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146185058","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}
Pub Date : 2026-03-01Epub Date: 2026-01-14DOI: 10.1016/j.autcon.2025.106748
Hamed Hasani, Francesco Freddi
This study presents an AI-powered framework for automated structural health monitoring that integrates modal identification, anomaly detection, and damage localization under varying environmental and operational conditions. The approach combines stochastic subspace identification with frequency–spatial domain decomposition for automated modal extraction and a condition-aware anomaly detector based on a conditional variational autoencoder. A secondary SSA–OC-SVM module verifies and localizes damage. The methodology is validated on a laboratory-scale structure through 500 one-hour tests under temperature variations up to 35 °C and diverse loading conditions. The identified modes exhibit MAC = 0.99–1.00, confirming reliable automated identification. The CVAE reconstructs healthy-state modal frequencies with MAPE = 0.23%, RMSE = 0.027 Hz, and = 0.836, effectively distinguishing environmental effects ( pp) from genuine structural changes. The integrated framework further accurately localizes all induced damage scenarios across nine structural zones, demonstrating high accuracy, robustness, and scalability for next-generation SHM automation.
{"title":"Condition-aware AI framework for automated structural health monitoring","authors":"Hamed Hasani, Francesco Freddi","doi":"10.1016/j.autcon.2025.106748","DOIUrl":"10.1016/j.autcon.2025.106748","url":null,"abstract":"<div><div>This study presents an AI-powered framework for automated structural health monitoring that integrates modal identification, anomaly detection, and damage localization under varying environmental and operational conditions. The approach combines stochastic subspace identification with frequency–spatial domain decomposition for automated modal extraction and a condition-aware anomaly detector based on a conditional variational autoencoder. A secondary SSA–OC-SVM module verifies and localizes damage. The methodology is validated on a laboratory-scale structure through 500 one-hour tests under temperature variations up to 35 °C and diverse loading conditions. The identified modes exhibit MAC = 0.99–1.00, confirming reliable automated identification. The CVAE reconstructs healthy-state modal frequencies with MAPE = 0.23%, RMSE = 0.027 Hz, and <span><math><msup><mrow><mi>R</mi></mrow><mrow><mn>2</mn></mrow></msup></math></span> = 0.836, effectively distinguishing environmental effects (<span><math><mrow><mo>≤</mo><mn>0</mn><mo>.</mo><mn>27</mn></mrow></math></span> pp) from genuine structural changes. The integrated framework further accurately localizes all induced damage scenarios across nine structural zones, demonstrating high accuracy, robustness, and scalability for next-generation SHM automation.</div></div>","PeriodicalId":8660,"journal":{"name":"Automation in Construction","volume":"183 ","pages":"Article 106748"},"PeriodicalIF":11.5,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145976055","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}
Pub Date : 2026-03-01Epub Date: 2026-01-27DOI: 10.1016/j.autcon.2026.106791
Yuandong Pan , Mudan Wang , Linjun Lu , Rabindra Lamsal , Erika Pärn , Sisi Zlatanova , Ioannis Brilakis
Digital twins are increasingly used in the Architecture, Engineering, and Construction (AEC) industry, but their adoption is often hindered by the need for specialised knowledge, such as database querying. This paper presents Graph-DT-GPT, a multi-agent framework that integrates Large Language Models (LLMs) with graph-based digital twins to enable natural language interaction. The framework is designed with modular agents, including decision, query generation, and answer extraction, and grounds all LLMs’ outputs in structured graph data to improve response reliability and reduce hallucinations. The framework is evaluated on two use cases: a city-level graph with over 40,000 building nodes and room-level apartment layout graphs. Graph-DT-GPT achieves 100% and 95.5% answer correctness using Claude Sonnet 4.5 and GPT-4o, respectively, in the city-scale case, and 100% correctness in the room-level case, significantly outperforming baseline methods including LangChain Neo4j pipelines by approximately 40% and 10%, respectively. These results demonstrate its scalability and potential to enhance accessible, accurate information retrieval in AEC digital twin applications.
{"title":"LLM-enabled multi-agent framework for natural language interaction with graph-based digital twins","authors":"Yuandong Pan , Mudan Wang , Linjun Lu , Rabindra Lamsal , Erika Pärn , Sisi Zlatanova , Ioannis Brilakis","doi":"10.1016/j.autcon.2026.106791","DOIUrl":"10.1016/j.autcon.2026.106791","url":null,"abstract":"<div><div>Digital twins are increasingly used in the Architecture, Engineering, and Construction (AEC) industry, but their adoption is often hindered by the need for specialised knowledge, such as database querying. This paper presents Graph-DT-GPT, a multi-agent framework that integrates Large Language Models (LLMs) with graph-based digital twins to enable natural language interaction. The framework is designed with modular agents, including decision, query generation, and answer extraction, and grounds all LLMs’ outputs in structured graph data to improve response reliability and reduce hallucinations. The framework is evaluated on two use cases: a city-level graph with over 40,000 building nodes and room-level apartment layout graphs. Graph-DT-GPT achieves 100% and 95.5% answer correctness using Claude Sonnet 4.5 and GPT-4o, respectively, in the city-scale case, and 100% correctness in the room-level case, significantly outperforming baseline methods including LangChain Neo4j pipelines by approximately 40% and 10%, respectively. These results demonstrate its scalability and potential to enhance accessible, accurate information retrieval in AEC digital twin applications.</div></div>","PeriodicalId":8660,"journal":{"name":"Automation in Construction","volume":"183 ","pages":"Article 106791"},"PeriodicalIF":11.5,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146071735","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}
This paper proposes a lightweight semantic segmentation framework utilizing 3D point cloud data to enable automatic and rapid construction progress monitoring in high-rise building projects. This study centers on developing an efficient L-PointNet++ model that integrates self-attention mechanisms and MobileNetV3 modules, significantly reducing computational complexity and achieving a 95.63 % reduction in total training time compared to traditional PointNet++. A dual-stage training strategy is adopted to effectively address class imbalance, resulting in high segmentation accuracy with mean Intersection over Union (mIoU) values of 0.9308 for edge points and 0.9300 for corner points. Experimental results indicate that the developed framework can significantly enhance the speed and adaptability of as-built BIM model reconstruction and provide substantial improvements in decision-making efficiency and project management through the implementation of a visualization-based progress monitoring and early-warning system. Overall, the proposed approach demonstrates notable advantages in 3D reconstruction accuracy, speed, and project control, providing a robust solution for real-time construction progress monitoring applications.
{"title":"Lightweight semantic segmentation for construction progress monitoring using 3D point clouds","authors":"Jinting Huang , Zhonghua Xiao , Ankang Ji , Limao Zhang","doi":"10.1016/j.autcon.2026.106765","DOIUrl":"10.1016/j.autcon.2026.106765","url":null,"abstract":"<div><div>This paper proposes a lightweight semantic segmentation framework utilizing 3D point cloud data to enable automatic and rapid construction progress monitoring in high-rise building projects. This study centers on developing an efficient L-PointNet++ model that integrates self-attention mechanisms and MobileNetV3 modules, significantly reducing computational complexity and achieving a 95.63 % reduction in total training time compared to traditional PointNet++. A dual-stage training strategy is adopted to effectively address class imbalance, resulting in high segmentation accuracy with mean Intersection over Union (mIoU) values of 0.9308 for edge points and 0.9300 for corner points. Experimental results indicate that the developed framework can significantly enhance the speed and adaptability of as-built BIM model reconstruction and provide substantial improvements in decision-making efficiency and project management through the implementation of a visualization-based progress monitoring and early-warning system. Overall, the proposed approach demonstrates notable advantages in 3D reconstruction accuracy, speed, and project control, providing a robust solution for real-time construction progress monitoring applications.</div></div>","PeriodicalId":8660,"journal":{"name":"Automation in Construction","volume":"183 ","pages":"Article 106765"},"PeriodicalIF":11.5,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145995538","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}
Pub Date : 2026-03-01Epub Date: 2026-01-19DOI: 10.1016/j.autcon.2026.106768
Zijian Wang , Ronen Barak , Rafael Sacks , Sitsofe K. Yevu , Arnon Bentur , Georgios M. Hadjidemetriou
Although digital technologies are increasingly studied in construction, their specific impacts on productivity remain partially understood. This review aims to investigates the relationship between digital technologies and construction productivity. The methodology comprises a bibliometric analysis and a systematic literature review of studies published over the past decade. Scopus was selected as the primary database for data retrieval, with 346 publications across 16 journals being identified and analyzed. The bibliometric analysis reveals publication trends and technology interrelations, highlighting AI and optimization as central to a cohesive ecosystem involving BIM, digital twins, sensors, and robotics. The systematic literature review is structured around categorising the use of technology for productivity into four dimensions: measurement, estimation, optimisation, and enhancement. Despite this qualitative synthesis being influenced by the authors' judgement and subjectivity, it highlights the practical benefits such as improved prediction and automation, alongside challenges including data standardization, integration, and workforce adaptation.
{"title":"Construction productivity and digital technologies","authors":"Zijian Wang , Ronen Barak , Rafael Sacks , Sitsofe K. Yevu , Arnon Bentur , Georgios M. Hadjidemetriou","doi":"10.1016/j.autcon.2026.106768","DOIUrl":"10.1016/j.autcon.2026.106768","url":null,"abstract":"<div><div>Although digital technologies are increasingly studied in construction, their specific impacts on productivity remain partially understood. This review aims to investigates the relationship between digital technologies and construction productivity. The methodology comprises a bibliometric analysis and a systematic literature review of studies published over the past decade. Scopus was selected as the primary database for data retrieval, with 346 publications across 16 journals being identified and analyzed. The bibliometric analysis reveals publication trends and technology interrelations, highlighting AI and optimization as central to a cohesive ecosystem involving BIM, digital twins, sensors, and robotics. The systematic literature review is structured around categorising the use of technology for productivity into four dimensions: measurement, estimation, optimisation, and enhancement. Despite this qualitative synthesis being influenced by the authors' judgement and subjectivity, it highlights the practical benefits such as improved prediction and automation, alongside challenges including data standardization, integration, and workforce adaptation.</div></div>","PeriodicalId":8660,"journal":{"name":"Automation in Construction","volume":"183 ","pages":"Article 106768"},"PeriodicalIF":11.5,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146000934","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}
This paper integrates AI image recognition and BIM technology to develop a prototype system that achieves automation and visualization of construction progress control. The system supports using a BIM model for planning deployment of multiple surveillance cameras to encompass the entire construction site. The real-time images captured by these cameras are processed using object detection technology to locate all actively constructed objects in the images and identify their respective construction phases. By integrating the perspectives of these cameras into the BIM model, the AI detection results from each camera image are automatically inputted into corresponding components of the BIM model. Subsequently, the real-time site progress information stored in the BIM model is compared with the planned schedule, and the comparative results are visually presented on the BIM model components in different colors. Through visualization, this approach enables management personnel to control progress in a specific and intuitive manner in real-time.
{"title":"Automated construction progress monitoring and control through AI-based image recognition and BIM integration","authors":"Chang-Cheng Hsieh, Hung-Ming Chen, Wan-Yu Chen, Ting-Yu Wu","doi":"10.1016/j.autcon.2026.106783","DOIUrl":"10.1016/j.autcon.2026.106783","url":null,"abstract":"<div><div>This paper integrates AI image recognition and BIM technology to develop a prototype system that achieves automation and visualization of construction progress control. The system supports using a BIM model for planning deployment of multiple surveillance cameras to encompass the entire construction site. The real-time images captured by these cameras are processed using object detection technology to locate all actively constructed objects in the images and identify their respective construction phases. By integrating the perspectives of these cameras into the BIM model, the AI detection results from each camera image are automatically inputted into corresponding components of the BIM model. Subsequently, the real-time site progress information stored in the BIM model is compared with the planned schedule, and the comparative results are visually presented on the BIM model components in different colors. Through visualization, this approach enables management personnel to control progress in a specific and intuitive manner in real-time.</div></div>","PeriodicalId":8660,"journal":{"name":"Automation in Construction","volume":"183 ","pages":"Article 106783"},"PeriodicalIF":11.5,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146000936","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}
Pub Date : 2026-03-01Epub Date: 2026-01-26DOI: 10.1016/j.autcon.2026.106794
Yuxin Zhang, Xi Wang, Mo Hu, Zhenyu Zhang
Information retrieval and question answering from safety regulations are essential for automated construction compliance checking but are hindered by the linguistic and structural complexity of regulatory text. Many queries are multi-hop, requiring synthesis across interlinked clauses. To address the challenge, this paper introduces BifrostRAG, a dual-graph retrieval-augmented generation (RAG) system that models both linguistic relationships and document structure. The proposed architecture supports a hybrid retrieval mechanism that combines graph traversal with vector-based semantic search, enabling large language models to reason over both the content and the structure of the text. On a multi-hop question dataset, BifrostRAG achieves 92.8% precision, 85.5% recall, and an F1 score of 87.3%. These results significantly outperform vector-only and graph-only RAG baselines, establishing BifrostRAG as a robust knowledge engine for LLM-driven compliance checking. The dual-graph, hybrid retrieval mechanism presented in this paper offers a transferable blueprint for navigating complex technical documents across knowledge-intensive engineering domains.
{"title":"Bridging dual knowledge graphs for multi-hop question answering in construction safety","authors":"Yuxin Zhang, Xi Wang, Mo Hu, Zhenyu Zhang","doi":"10.1016/j.autcon.2026.106794","DOIUrl":"10.1016/j.autcon.2026.106794","url":null,"abstract":"<div><div>Information retrieval and question answering from safety regulations are essential for automated construction compliance checking but are hindered by the linguistic and structural complexity of regulatory text. Many queries are multi-hop, requiring synthesis across interlinked clauses. To address the challenge, this paper introduces BifrostRAG, a dual-graph retrieval-augmented generation (RAG) system that models both linguistic relationships and document structure. The proposed architecture supports a hybrid retrieval mechanism that combines graph traversal with vector-based semantic search, enabling large language models to reason over both the content and the structure of the text. On a multi-hop question dataset, BifrostRAG achieves 92.8% precision, 85.5% recall, and an F1 score of 87.3%. These results significantly outperform vector-only and graph-only RAG baselines, establishing BifrostRAG as a robust knowledge engine for LLM-driven compliance checking. The dual-graph, hybrid retrieval mechanism presented in this paper offers a transferable blueprint for navigating complex technical documents across knowledge-intensive engineering domains.</div></div>","PeriodicalId":8660,"journal":{"name":"Automation in Construction","volume":"183 ","pages":"Article 106794"},"PeriodicalIF":11.5,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146048561","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}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号