請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/52504
標題: | 鋼筋與裂縫敲擊回音訊號之辨識 Recognition of Reinforcing Bars and Cracks Based on Impact-Echo Phase Analysis |
作者: | Cheng-Yu Yeh 葉承瑜 |
指導教授: | 劉佩玲(Pei-Ling Liu) |
關鍵字: | 結構安全評估,非破壞檢測,敲擊回音法,相位,混凝土, Safety Evaluation,Nondestructive Testing,Impact-echo,Phase,Concrete, |
出版年 : | 2015 |
學位: | 碩士 |
摘要: | 敲擊回音法已廣泛的應用在混凝土內部缺陷的檢測,利用試體表面敲擊產生的應力波,將試體表面位移經由傅立葉轉換頻譜中的回音尖峰判讀缺陷位置。然而由回音尖峰無法顯示反射界面是鋼筋還是裂縫所造成。
本研究主要提出一個自動化判別混凝土內鋼筋及裂縫的方法,在敲擊回音試驗中,試體表面產生的壓力波遭遇自由界面會產生張力回波,遭遇固定界面則會產生壓力回波,故兩種不同界面所產生的敲擊回音相位不同。根據此原理,本研究逐以試體表面位移訊號之傅立葉相位頻譜發展鋼筋與裂縫訊號的判別方法。 本研究以有限元素分析內含8到12公分深裂縫及4到6公分深鋼筋的混凝土之敲回音反應,並對二塊混凝土試體進行模型試驗,各含有12公分深裂縫及6公分深的鋼筋。將模擬與模型訊號之相位頻譜中對應於回音頻率的相位及相位斜率畫在相位—相位斜率的圖上,可以明顯的看出裂縫與鋼筋分佈在兩個分隔的區域,裂縫相位較接近0,鋼筋相位較接近pi/2,且鋼筋相位斜率偏低。因此本研究便以回音相位及相位斜率作為判斷反射界面類別的特徵參數。 本研究利用線性及類神經網路兩種分類器對前述訊號進行分類。線性分類器是直接在相位—相位斜率圖上以一垂直線(一元區分法)及一斜線(二元區分法)將裂縫及鋼筋兩群資料分開,此二直線均可作為決策邊界,可擇一使用。當產生新的敲擊回音訊號時,可將其回音相位及相位斜率直接點在圖中,看此點落於決策邊界那一側來判斷反射界面的種類。 本研究之類神經網路分類器是以一個倒傳遞網路作為基礎,以回音相位及相位斜率作為輸入特徵,反射界面種類作為輸出(裂縫為1,鋼筋為0),再以前述之數值模擬及模型試驗數據對網路進行訓練。當產生新的敲擊回音訊號時,只要將其回音相位及相位斜率輸入此類神經網路分類器,便可由輸出值判斷反射界面的種類。 為測試前述分類器之準確性,本研究中另內含傾斜裂縫及3公分深鋼筋的模型試驗訊號,以及內含25公分深水平裂縫及15公分深圓柱鋼筋的模擬及模型試驗訊號代入各分類器,發現一元、二元線性分類器及類神經網路皆能正確判別反射界面種類,顯示本研究方法之有效性。 Impact-echo is widely used on inspecting the defect in concrete. Through the Fourier transform of surface vibration signal, inspector can use peak frequency to calculate the depth of the defect. However, peak frequency cannot determine whether the reflector is a crack or steel. The study develops a method to automatically determine whether the echo comes from a crack or a steel bar. As an impact is applied on the concrete surface, compression waves will be generated. As a compression wave encounters a free interface, it is reflected as a tension wave. On the other hand, if it encounters a fixed boundary, it is reflected as compression wave. Hence, the phase of the impact echo signal contains the information about the type of the interface. Therefore, this study tries to develop a reflector classifier based on the phase spectrum of the impact echo signal. Finite element analysis and model tests were conducted on specimens containing cracks or steel bars at different depth, ranging from 8 to 12 cm for cracks and 4 to 6 cm for steel bars. Fourier transforms were applied to the response signals and the phase spectrum for each signal was obtained. A 2D scatter plot was generated using the phase and the slope of phase at the echo frequency. It is found that the data points were divided into two separated groups. Hence, the phase and the slope of phase at the echo frequency were used as the input features for the classifier in this study. Two types of classification methods were used in this research, namely, the linear classification and the artificial neural network (ANN). In the linear classifier, a straight line is drawn in the phase |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/52504 |
全文授權: | 有償授權 |
顯示於系所單位: | 應用力學研究所 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-104-1.pdf 目前未授權公開取用 | 2.27 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。