ニューラルネットワークと計算力学に基づく構造同定 : 一次元はり問題への適用 (<小特集>ファジィ/ニューラルネットのダイナミックスと制御への応用)

概要

論文の詳細を見る
This paper describes an application of hierarchical neural networks to structural identification. The present procedure consists of three phases : ( a ) preparation of learning data, in which the learning data are produced through a number of FEM calculations with various identification parameters, ( b ) training of the neural network and ( c ) utilization of the neural network, i. e., structural identification. As the first step of the present research, this procedure is applied to the identification of material properties of an one-dimensional beam from its vibration characteristics such as eugenmode, and displacement, and acceleration distributions. In addition, the analog, digital and hybrid output methods are compared with one another from the viewpoints of efficiency and accuracy.
一般社団法人日本機械学会の論文
1992-06-25