自励振動を有する2次元チャンネル内噴流の数値解析 : 第2報, ニューラルネットワークの適用

抄録

An application of a neural network to jets with self-induced oscillation exhausting into a two-dimensional channel is presented. No established method nor suitable parameter to control flip-flop jets is known. One new approach for neural networks is a recurrent network. We find that if the flip-flop is periodic, the recurrent network can be useful to reproduce the original periodic orbit obtained by the time delay method. Another application of the recurrent network employed the flow pattern from velocity fields of the neural network. In this case, the results depend on the number of reference points and on the number of teaching patterns in computed fields given. A promising result is obtained for the control of flip-flop jets using a recurrent network.