PIVにおけるフーリエ変換法のニューラルネットによる性能改善

概要

論文の詳細を見る
In this paper, we present a new Fourier transform method with improved PIV measurement accuracy and processing time. A multiple-layer neural network is used in order to suppress obviously incorrect velocity vectors and to reduce the processing time. The information input to the network is the complex phase extracted from the Fourier transforms of two images obtained in a short time interval, and the output is the mean spatial shift of the image pattern in a sampling window. The learning of the correlation between the input and the output is performed using an error back propagation method, and performance tests are carried out using numerically simulated particle images and images obtained experimentally. The test results show that the present method is more robust against velocity fluctuation than the original one and the CPU time is reduced to about 75% of that of the original method.
一般社団法人日本機械学会の論文
1997-05-25