小波理論與類神經網路在橋梁非破壞檢測之應用

Application of wavelet theory and artificial neural network on bridge nondestructive testing

林昆達(Kun-Ta Lin)

指導教授 ： 王安培

中原大學/工學院/土木工程研究所/碩士(2002年)

https://doi.org/10.6840/cycu200200369

摘要

橋梁系統在交通建設中具有極重要的地位，本文期望建立橋梁非破壞檢測技術，利用小波理論分析及類神經網路識別診斷橋梁的不同破壞程度及檢測破壞點位置，提供快速而簡便的橋樑非破壞檢測方法。
首先製造各種不同裂縫大小及位置之橋面板模型，藉由量測儀器以震波量測法擷取各種裂縫大小及損壞位置的訊號，經由小波理論作訊號處理，擷取訊號的特徵，建立資料庫(database)，再將訊號的特徵作為類神經網路輸入的特徵向量。經由類神經網路的訓練後，可判別橋面板不同裂縫大小及損壞位置。
試驗資料經小波理論及類神經網路識別分析結果發現裂縫大小識別的準確率高達96.3%，破壞點位置的識別可達90 %準確率。由此可知，小波理論在訊號處理及特徵擷取是一個相當不錯的方法，可建立識別資料庫，再加上類神經網路的訓練後可檢測橋梁之破壞程度和破壞點位置。

關鍵字

特徵擷取 ； 類神經 ； 小波理論

並列摘要

Bridge system is a very important status in traffic flow. The purpose of this paper hopes to establish technology of nondestructive testing (NDT) by wavelet theory analysis and artificial neural network (ANN) identification which diagnoses different destruction degree and breakpoint position of the bridge, and providing a fast and convenient way of nondestructive testing.
Firstly, a crack different size and breakpoint position is initiated on the bridge floor model. Then, signals of different crack size and breakpoint position are generated by seismic wave measure method. Using the wavelet theory we can extract features of signal and establish database, which forms the input eigenvector of ANN. After ANN training, we can judge different crack size and breakpoint position of the bridge floor model.
By experiment datum of wavelet theory and the results of ANN, we found that the identification accuracy of crack size achieved 96.3% and the identification accuracy of breakpoint position achieved 90%. It is obvious that wavelet theory on signal processing and feature extraction is a good methodology and after ANN training, which can diagnoses destruction degree and breakpoint position of bridge floor model.

並列關鍵字

wavelet theory ； feature extraction ； ANN

參考文獻

1.McCulloch, W. S., and Pitts, W., “A Logical Calculus of the Ideas Immanent in Nervous activity,” Bulletin of Mathematical Biophysics 5: pp.115-133,1943.

2.Hopfield, J. J., “Neural Networks and Physical Systems with Emergent Collective Computational Abilities,” Proc. Natl. Acad. Sci. USA, 79, pp.2554-2558, 1982.

3.Grossmann, A., Morlet, J., “Decomposition of Hardy function into square integrable wavelets of constant shape,” SIAM J. Math. Anal. 15(4): pp. 736-783, 1984.

4.Rumelhard, D. E., Hintion, G. E. and Willians, R. J., “Learning Representations by Back-Propagating errors,” Nature, 323, pp.533-536, 1986.

5.Kohonen, T., “Self-organization and Associative Memory,” New York: Springer-Verlag, pp.155-157, 1988.