ニューラルネットワークと計算力学に基づくシステム同定の検討 : 第1報, ホログラフィックニューラルネットワークとバックブロバゲーションニューラルネットワークの比較

概要

論文の詳細を見る
Recently computer simulation plays more and more important role in design stage. But it is difficult to make an accurate model of computer simulation represented by finite element method (FEM), because there are some differences in physical parameters from those of real structure. In this paper, the physical parameters of FEM which are not consistent with those of real structure are treated as damage. The detection of damage location, and its physical values are studied based on neural network and computational mechanics. An approach of damage detection is discussed with four types of structure of neural network as digital type, analog type, coordinate type and hybrid type, by holographic neural network. Comparing to the multiple layer neural network (back propagation training scheme), it is shown that holographic neural network is more accurate and faster in training and detection.
1998-05-25