透過整合建築資訊模型、深度學習與網路地圖服務提升地震後建築復原系統

Abstract

地震後的建築復原是一項高度複雜且時間緊迫的過程，需要創新方法來提升效率和精確性。本研究聚焦於兩種主要方法，以實現整合尖端技術的策略，包括網路地圖服務 (WMS)、深度學習 (DL) 和建築資訊模型 (BIM)，以改善並加速地震後建築復原過程。第一種方法著重於開發基於BIM的應用程式，作為建築復原計劃的規劃平台。該平台功能涵蓋檢查每個結構元素的狀態、制定修復計劃、估算總復原成本，並根據復原計劃更新BIM模型。在此過程中，DL用於結構元素評估，自動化損壞分類和損壞區域量化，有助於提升建築復原規劃的精確性和效率。第二種方法則結合BIM、DL與WMS，以分析結構元素損壞並為長期建築修復做準備。透過WMS整合地理空間數據，此方法加強了受損區域的情境分析，並支援可持續的修復策略。採用Vision Transformer模型進行圖像分類，並利用Detectron2進行分割，可精確檢測和評估裂縫、剝落及裸露鋼筋等問題，並直接連接至BIM數據，以實現全面的復原規劃。案例研究強調了這些方法的實際應用，顯示了計劃速度和資源分配的改善，計劃時間縮短了約10-15%。此整合框架的重要優勢在於，通過考量低碳材料和節能技術，提升可持續性，加速復原計劃的制定，並提高損壞分類的精準度。儘管目前數據集的限制影響了DL模型的精度，隨著數據的持續收集和系統改進，結果預期會逐步提升。總體而言，本研究強調了結合BIM應用開發和整合BIM、DL與WMS的雙方法在地震後復原中的變革潛力。不僅支援有效且可持續的建築復原，所建議的方法亦為未來朝向完全自動化、數據驅動的智慧復原系統奠定基礎，符合現代建築目標和可持續發展策略。

Building recovery after an earthquake is a highly complicated and time-sensitive process requiring novel approaches to improve efficiency and accuracy. The study concentrates on two primary approaches to accomplish an integrated strategy that makes use of cutting-edge technologies: Web Map Services (WMS), deep learning (DL), and Building Information Modeling (BIM) in order to improve and expedite the post-earthquake building recovery process. The first methodology focuses on developing a BIM-based application to serve as a platform for planning building recovery. This platform's features involve reviewing each structural element's status, generating rehabilitation plans, estimating the total cost of building recovery, and updating the BIM model following the recovery plan. This approach, which DL supports for structural element evaluation, improves the precision and effectiveness of building recovery planning by automating damage classification and quantifying damage areas. The second methodology integrates BIM, DL, and WMS to analyze structural element damage and prepare for long-term building rehabilitation. By facilitating the incorporation of geospatial data using WMS, this method improves the contextual analysis of damaged regions and supports sustainable rehabilitation strategies. Vision Transformer models for image classification and Detectron2 for segmentation allow for the exact detection and evaluation of problems such as cracks, spalling, and exposed rebar, which can then be connected directly to BIM data for comprehensive recovery planning. Case studies highlight the practical implementation of these approaches, demonstrating gains in planning speed and resource allocation, with possible reductions in planning time of 10-15%. Significant advantages of this integrated framework include improved sustainability by considering low-carbon materials and energy-efficient techniques, quicker recovery plan formulation, and more precise damage classification. Even though dataset limits now impact the accuracy of DL models, these results should eventually increase due to ongoing data collecting and system improvements. Overall, this study emphasizes the transformational potential of a dual-methodological approach to post-earthquake recovery that combines BIM-based application development with integrating BIM, DL, and WMS. Along with supporting effective and sustainable building recovery, the suggested approaches lay the foundation for future developments in the direction of fully automated, data-driven intelligent recovery systems that complement contemporary construction goals and sustainable development strategies.