本文期望建立一套快速且準確的橋樑識別系統,藉由模擬橋面板的實驗,配合頻譜分析以及類神經理論,對橋樑進行破壞程度與位置的識別診斷,以小波轉換和希爾伯特-黃轉換(HHT)為頻譜分析的工具,期望藉由兩者應用於所設計的橋樑模式,同時比較兩者的優缺點。 試驗步驟首先於橋面板模型設計不同大小的裂縫與敲擊距離裂縫不同距離的面板區塊,藉由量測儀器以震波量測法擷取各種不同情況的振動訊號,經由HHT和小波理論分別作訊號處理,經頻譜分析理論各自獲得不同情況下的振動特徵,再與模糊理論結合,將特徵訊號做較佳的分類與修正。將此特徵訊號作爲類神經網路輸入向量,經由訓練後,可識別出橋面板的損傷程度與破裂位置。 由希爾伯特頻譜圖以及連續小波轉換之圖形得知兩者皆能解析出訊號的特徵反應,而且在解析上希爾伯特頻譜圖優於連續小波轉換之圖形;然而在橋面板的損傷程度與破裂位置上的識別,採用小波轉換的頻譜特徵輸入經由類神經網路之識別結果優於HHT方式的特徵輸入。
The purpose of this paper is to establish an accurate and fast bridge identification system. By bridge floor simulation experience, we combine theories of spectrum analysis and artificial neural networks to identify bridge destruction and its position. The spectrum analysis theories applied here is wavelet theory and Hilbert-Huang transform, and some results are compared in both theories. First, some different crack sizes and breakpoint positions are set, and by using instrument to measure and get vibration signals. The signals are analyzed by both in Hilbert-Huang transform and wavelet theory, and different vibration features are obtained. The vibration features that are classified and modified by fuzzy theory would be taken as input data of artificial neural network. The different crack sizes and breakpoint positions of bridge floor could be identified after ANN’s training. From figures of Hilbert spectrum and figures of continuous wavelet transform, we know both wavelet theory and Hilbert-Huang transform can solve responses of signals. As the results show, the figures of Hilbert spectrum are better than of continuous wavelet transform, but the features inputs of wavelet theory are better than of Hilbert-Huang transform on identifying different crack sizes and breakpoint positions.