階層型ニューラルネットワークと計算力学に基づく逆問題解析法 : 学習・推定メカニズムの検討と定量的欠陥同定への適用

概要

論文の詳細を見る
This paper describes an application of neural networks to inverse analyses. The neural networks have been applied not only to qualitative pattern recognition but also to quantitative mapping problems. However, the mechanism of learning and estimating processes, particularly for quantitative mapping, has not been studied so well. On the other hand, nondestructive determination of crack size, which may be regarded as quantitative mapping, has been important in predicting fatigue life as well as in estimating failure probabilities. In the fields of Computational Mechanics and Fracture Mechanics, various identification methods have been suggested so far. However, these methods are not swift and effective for on-line inspection. A BP (Back-Propagation) neural network is applied here to an identification of size and location of a circular defect to demonstrate the features of neural network in quantitative mapping and to investigate its feasibility as an on-line nondestructive inspection.
一般社団法人日本機械学会の論文
1991-08-25