標題: 應用類神經網路於橋梁檢測之研究--以彰化縣橋梁為例
A Study of Artificial Neural Network Model for Bridge Inspection in Chung-Hua County
作者: 葉行健
Yeh, Hsing-Chien
洪士林
Hung, Shih-Lin
工學院工程技術與管理學程
關鍵字: 類神經網路;橋樑;橋樑管理系統;Artificial Neural Network;Bridges;Bridge Management System
公開日期: 2009
摘要: 台灣地區經濟的發展,交通運輸系統佔有重要的地位,而橋樑工程更是整個交通運輸系統的重要角色,所以不論高山或平地皆有橋梁工程的蹤跡。所以,橋樑功能若受損,勢必造成社會、民生與經濟之重大衝擊;尤其是台灣地區天然災害頗多,更凸顯出橋樑養護工作之重要性。因此,如何維護數量相當的橋樑並發揮應有功能變得相當重要。另外,維護此一數量龐大的橋樑勢必花費相當的金錢與人力,也因此,如何有效率並在有限人力與經費下,維護橋樑並避免劣化,係為橋樑管理單位首要課題。 目前國內針對橋樑管理建立了一套橋樑管理系統,並由全國各橋樑維護機關使用當中,然台灣地區地形高山至平原皆有,各縣市位處地形之並非完全相似,因此本研究以地勢較平坦地區之橋樑(彰化縣之橋樑)做為研究對象,並以該縣利用DERU橋梁目視檢測之463筆成果與應用類神經網路之學習及預測能力,檢討地勢平坦地區較可能之橋樑損壞模式。研究結果顯示,相較於DERU檢測法,類神經網路模式對訓練與驗證案例均可提供準確評估(預測)橋損壞等級;亦證實利用類神經網路模式於橋梁損壞模式評估之可行性。
Transportation system plays the most important role for the rapidly economic developing in Taiwan. In which, bridges take part in the major function in the system; hence, there are numerous bridges located among plant and mountain area of Taiwan. Therefore, the damage of bridges may result in inconvenient and distress in many points, such as social, civil, and economic concern. The issue is more essential in Taiwan since Taiwan locates in frequently natural disaster region. Hence, how to maintain these numerous bridges health in well function with low cost and less manpower is vital. Currently, a bridge management system based on the evaluation of DERU has been developed and employed in various bridge management bureaus in Taiwan. However, the locations and types of bridges are deviation county to county. This work aims to develop an artificial neural network (ANN) bridge damage detection model based on the investigation data by established DERU approach. The target of this study focus on plate or beam type bridges in plant area, says Chang-Hua County. There are 463 investigated data used in this work and divided into training and verify instances. The optimal topology of ANN model is studied first. The verified results illustrate that ANN model can accurately detect the degree of damage for both training and verification cases. The performance of ANN model is better than conventional DERU approach by comparing with detecting accuracy. The work also confirms that the feasibility of applying ANN bridge damage detection models in different locations and types of bridges.
URI: http://140.113.39.130/cdrfb3/record/nctu/#GT079370505
http://hdl.handle.net/11536/40686
Appears in Collections:Thesis


Files in This Item:

  1. 050501.pdf