Parameter extraction by parallel neural networks

Article type: Research Article

Authors: Wong, Y.S.^a | Lee, B.H.K.^b | Wong, T.K.S.^a

Affiliations: [a] Department of Mathematical Sciences, University of Alberta, Edmonton, Canada. E-mail: [email protected]; URL: http://www.math.ualberta.ca/People/Facultypages/Wong.Y.html | [b] Institute for Aerospace Research, National Research Council, Ottawa, Canada. E-mail: [email protected]

Abstract: This paper presents a combined wavelet – neural networks model to extract damping coefficients and modal frequency values from flight flutter data. Wavelet transform is introduced, not only it is used to filter noise from the original raw data, but it is applied so that the original multi-mode signal is decomposed into a sequence of single-mode signals. Consequently, the parameters can be extracted effectively by parallel neural networks. To improve the efficiency in training the neural network, only a small number of non-zero wavelet coefficients are taken as the input to the neural network. Application of the wavelet-neural networks to simulated flutter data are reported, and it is concluded that the proposed model appears to be effective and accurate for parameter extraction.

DOI: 10.3233/IDA-2001-5105

Journal: Intelligent Data Analysis, vol. 5, no. 1, pp. 59-71, 2001