請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/88604
標題: | 基於影像及排程之營建數位孿生自動化進度監控 Automated activity-level progress monitoring from visual data and schedules through digital twin construction |
作者: | 安里達 Aritra Pal |
指導教授: | 謝尚賢 Shang-Hsien HSIEH |
關鍵字: | 深度學習,計算機視覺,活動級進度監控,神經輻射場,自然語言處理,日程更新,數字孿生, Deep learning,computer vision,activity-level progress monitoring,neural radiance field,natural language processing,schedule update,digital twin, |
出版年 : | 2023 |
學位: | 博士 |
摘要: | 營建工程的進度管控是工程成功交付的重要關鍵,而目前影像資料已是瞭解工程進度的重要資訊來源之一。在過去幾十年中,許多研究開發應用電腦視覺於建築施工的自動進度監測的方法,這些方法在監測個別元件(如柱子、樑、牆壁)方面非常有效,但在監測由元件組成之排程工項工程進度(如一樓模板工程、鋼筋綁匝、混凝土澆置)仍有困難。現有方法通常難以推測介於已建或未建之間的未完成進度之工程狀態,因此限制了此技術在分析細節進度資訊方面的應用。本研究旨在透過兩種新的方法解決之研究問題。第一種方法稱為排程工項進度監測系統(ALPMS),旨在監測正在施工的元件之活動級別進度,主要以施工現場影像和四維建築信息模型(BIM)作為輸入,產出數位孿生資訊系統。該系統從影像中生成實境現場的點雲,將其與原排程中的BIM進行比較,並應用基於深度學習的語義分割進行進度推理。因此可估計每個工項的完成百分比,並且為更新專案進度提供有價值的資訊。數位孿生資訊系統同時可將語義資訊整合到實際建造的點雲和BIM中,實現進度狀態的三維可視化。第二種方法研究於在缺乏更新的四維BIM的情況下也能自動比對專案進度與實際模型,該比對方法首先使用三維BIM或控制點將實際模型對應到世界坐標系統,然後應用點雲分割來檢測與特定位置、建築元件和工項相關的進度。使用基於自然語言處理(NLP)的技術從每個工項中提取相關位置、元件和任務的資訊。從實際模型和工項中提取的資料通過基於語意距離的匹配技術進行比對,再將進度資訊與相應的進度活動進行比對。以上提出的方法已在台灣的工程專案上應用及評估並且有展示相關的有效性和適用性,ALPMS成功分析工項等級的進度狀態,平均絕對誤差少於6%,而缺乏四維BIM的情況下仍可準確更新進度資料。這些方法論通過提供有關元件和工項等級進度的分析,為工程進度監測領域進一步的貢獻,也可使人們更容易地理解專案狀態,以實現高效的進度管理和做出明智的決策,進而促進專案成功交付。 Monitoring the progress of construction projects is crucial for ensuring successful project delivery. Visual data, such as images and videos, has emerged as a valuable source of information to understand the status of construction operations. Over the past few decades, several vision-based methods have been developed for automated progress monitoring in building construction. These methods have been effective in monitoring individual elements (e.g., columns, beams, walls) but face challenges in monitoring progress at the schedule activity level (e.g., formwork, reinforcement, concrete). Existing methods often struggle to report progress status beyond a binary form of built or not-built, limiting their usefulness in capturing nuanced progress information. This research focuses on addressing these challenges through two novel methodologies. The first methodology, called the Activity Level Progress Monitoring System (ALPMS), aims to monitor progress at the activity level of under-construction building elements. It takes construction site images and a four-dimensional building information model (BIM) as input and creates a Digital Twin information system. The system generates as-built point clouds from the images, compares them with the as-planned BIM, and applies deep learning-based semantic segmentation for progress reasoning. This enables the estimation of activity-wise completion percentages, providing valuable information for updating project schedules. The DT information system also integrates rich semantic information into the as-built point cloud and BIM, enabling three-dimensional visualization of progress status. The second methodology focuses on automatically aligning project schedules with reality models, even in the absence of an updated 4D BIM. The alignment method starts by aligning reality models to the world coordinate system using a 3D BIM or control points. Point cloud segmentation is then applied to detect progress associated with specific locations, building elements, and tasks. Information about locations, elements, and tasks is extracted from each schedule activity using natural language processing (NLP)-based techniques. Extracted information from the reality models and the schedule activities are matched through a distance-based matching technique, mapping the progress information with the corresponding schedule activities. The proposed methodologies have been applied and evaluated on construction projects in Taiwan, demonstrating their effectiveness and applicability. The ALPMS successfully reports activity level progress status with less than 6% mean absolute error. The automatic alignment method shows promise in accurately updating progress information without relying on an updated 4D BIM. These methodologies contribute to the field of construction progress monitoring by providing accurate and detailed insights into progress at both the element and activity levels. They enable a better understanding of project status, efficient schedule management, and informed decision-making, ultimately facilitating successful project delivery. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/88604 |
DOI: | 10.6342/NTU202302333 |
全文授權: | 同意授權(全球公開) |
顯示於系所單位: | 土木工程學系 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-111-2.pdf | 26.35 MB | Adobe PDF | 檢視/開啟 |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。