類神經網路於含斜裂縫桿件破壞檢測之應用

Abstract

本研究嘗試用ANSYS電腦程式內提供之裂縫特殊元素，以有限元素法模擬 出鋼梁結構各種位置、深度及角度之通用斜裂縫，並且求得不同結構模型 之力學動態特性。且為了評估使用電腦模擬結構物動態特性的可行性，而 以真實鋼梁配合振動訊號的實驗，以確定電腦模擬的可靠性。本研究主要 目的是希望用類神經網路(Artificial Neural Networks)中之倒傳遞網 路(Back-Propagation Network,BPN)，進行網路訓練的工作。因不同型式 之裂縫其動態特性有顯著的差別，累積此一差異性，藉由網路的學習，就 可以判斷結構物裂縫的位置、深度及角度。由研究結果顯示，以類神經網 路作為含斜裂縫桿件破壞檢測的方法，有不錯的辨識效果。 This reaserch applied the singular element from the ANSYS FEM package to simulate the general crack in different locations , depths and slanting angles in steel structures and found the dynamic characteristics of the cracked structures. The results from modal experiments validated the reliability of singular element of the FEM package in simulating the artificial crack. The goal of this research is to train the networks by the Back- Propagation model of Artificial Neural Networks. The influences of the locations, depths and slanting angles are coupled each other. Summing up those dynamic characteristics, the location, depth and slanting angle of a crack can be predicted. In this research, the member including single slanting crack can be identified by Artificial Neural Networks excellentlly. Hence, we will study the problem by the above method in order to make the N.D.T. of structures more popular.