Structure Damage diagnosis by Neural Networks and Optimization Techniques

• Main Authors: Gwo-Shiang, Lee, 李國驤
• Other Authors: Shih-Ming Yang
• Format: Others
• Language: en_US
• Published: 1998
• Online Access: http://ndltd.ncl.edu.tw/handle/87922373242745120234

博士 === 國立成功大學 === 航空太空工程學系 === 87 === Structure integrity has to be inspected periodically to ensure operational safety and avoid catastrophe. Conventional nondestructive evaluation techniques are often labor intensive and local in nature. Efficient global structure damage diagnosis (SDD) is necessary. Two structure damage diagnosis methods, one in pattern recognition by back-propagation neural network and the other in model updating by goal attainment optimization, are developed in this dissertation. These model based methods are conducted with incomplete modal data and without modal expansion or model reduction. In diagnosis by neural networks, it is shown that the damage condition can be identified effectively and efficiently by a well trained network with the poles and zeros of the damaged structure as signatures. In diagnosis by model updating, the 2-stage method is developed in which the most-analogous-substructured method is for damage localization in the first stage and the modal and pole/zero signatures are for damage extent evaluation in the second stage. It is shown that the performance of the 2-stage method can be improved significantly compared with previous methods in the literature. In addition, modeling error and measurement noise can be better controlled.