以數值方法模擬振動訊號配合類神經網路在結構破壞檢測之應用

Abstract

本文嘗試以PC版ANSYS5.0套裝軟體, 利用節點移動方式產生奇異元, 而以 有限元素法模擬建構出含各種裂縫之鋼樑結構, 並求得結構模型之動態特 性, 因裂縫位置及深度的不同, 結構反應出來之動態行為也隨著改變, 累 積此一差異性, 我們就可據以預測判斷出裂縫之位置及深度.由本文的結 果顯示出, 以移動結點至徑向 1/4處產生之奇異性元素, 做為裂縫尖端之 分割元素, 可迤得到與試驗相當近似之模擬效果, 同時, 以類神經網路中 之倒傳遞網路, 來判斷單一裂縫之無束制自由鋼樑之裂縫位置及深度, 可 得到滿意之辨識結果, 因此以有限元素法配合類神經網路做為非破壞性檢 測之工具, 將是一個可行的方法. This research used the node moving method form the commertcial F.E.M package,ANSYS PC version 5.0 , to deduce singular element. F.E.M used to nodel the cracked steel structure and fine the dynamic characteritics of this cracking structure. The inflwence of the position and depth are compled each orther. Suming up these characteritics, the position and the depth of the crack can be predicted in the crack tip. In this paper, the singular element was fromed by moving the middle node in the radian direction to the quarter point. The position and depth of the crack can be identified by the BP model of neural networks excellently. Hence, combining F.E.M and neural networks as a tool of Nnondestructive Testing will be a appropriate approach .