請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/88879
標題: | 應用基因演算法與深度學習於橋梁最佳感測器配置 Application of Genetic Algorithm and Deep Learning in Optimal Sensor Placement for Bridge |
作者: | 張智傑 Chih-Chieh Chang |
指導教授: | 呂良正 Liang-Jenq Leu |
關鍵字: | 感測器最佳配置,基因演算法,結構健康監測,頻率域分解法,隨機子空間識別法, Optamal sensor placement,Genetic algorithm,Structural health monitoring,Frequency domain decomposition,Stochastic subspace identification, |
出版年 : | 2023 |
學位: | 碩士 |
摘要: | 身處台灣,時常遭受到各項天然災害的侵擾,如颱風、地震等。每當災害發生後,為確保各項民用設施的使用安全,因此,進行結構物健康檢測是必須的;然而,感測器的安設是項耗費時間與金錢成本的作業,為減輕此負擔,本研究旨在使感測器的配置最佳化,希望得到最佳的感測器數量與配置位置。
本研究以SAP 2000 建立有限元素模型,並輸入多筆地震資料以得到結構物中各節點的加速度歷時反應做為資料庫,爾後當給定一組感測器配置組合,便可透過深度學習模型找尋配置節點與剩餘節點間的交互關係,並計算相對應之重建誤差。搭配基因演算法變換配置組合進行最佳化,目的是找出能使未裝設感測器處的反應重建誤差達最小的配置組合,即為最佳配置。 本文以兩個有限元素模型進行案例探討,分別是二維及三維的桁架橋梁。過程中將比較不同深度學習模型對於加速度歷時的重建效果、感測器最佳配置數量的探討以及重建未裝設感測器處的節點加速度歷時之應用,包含結構物模態頻率的識別與結構物損傷偵測。 Living in Taiwan, we often face various natural disasters such as typhoons and earthquakes.After each disaster, ensuring the safety of civil infrastructure is crucial. Therefore,conducting structural health monitoring is necessary. However, installing sensors is a time-consuming and costly operation. To alleviate this burden, this study aims to optimize the configuration of sensors, seeking the optimal number of sensors and their placement. In this study, SAP2000 is used to build finite element models, and multiple sets of seismic data are inputted to obtain the time history response of accelerations at each node in the structure, which serves as the database. Subsequently, given a set of sensor configuration combinations, a deep learning model is employed to explore the interaction relationship between the configured nodes and the remaining nodes,then calculating the corresponding reconstruction error. The genetic algorithm is utilized to optimize the configuration combinations whose objective is to find the configuration that minimizing the reconstruction error at the nodes without installing sensors, which is the optimal configuration. Two finite element models are investigated in this paper: a 2D truss bridge and a 3D truss bridge. The study compares different deep learning models in terms of their effectiveness in reconstructing acceleration time history, explores the optimal number of sensor configurations, and applies the reconstruction of acceleration time history at nodes without sensors, including tasks such as identification of structural modal frequencies and detection of structural damage. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/88879 |
DOI: | 10.6342/NTU202302961 |
全文授權: | 未授權 |
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 目前未授權公開取用 | 5.5 MB | Adobe PDF |
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。