Application of time-delay neural and recurrent neural networks for the identification of a hingeless helicopter blade flapping and torsion motions

• Autor(a) principal: Marques,F. D.
• Data de Publicação: 2005
• Outros Autores: Souza,L. de F. Rodrigues de, Rebolho,D. C., Caporali,A. S., Belo,E. M., Ortolan,R. L.
• Tipo de documento: Artigo
• Idioma: eng
• Título da fonte: Journal of the Brazilian Society of Mechanical Sciences and Engineering (Online)
• Texto Completo: http://old.scielo.br/scielo.php?script=sci_arttext&pid=S1678-58782005000200001
• topic: System identification; helicopter blade; time delay neural networks; recurrent neural networks

System identification consists of the development of techniques for model estimation from experimental data, demanding no previous knowledge of the process. Aeroelastic models are directly influence of the benefits of identification techniques, basically because of the difficulties related to the modelling of the coupled aero- and structural dynamics. In this work a comparative study of the bilinear dynamic identification of a helicopter blade aeroelastic response is carried out using artificial neural networks is presented. Two neural networks architectures are considered in this study. Both are variations of static networks prepared to accomodate the system dynamics. A time delay neural networks (TDNN) for response prediction and a typical recurrent neural networks (RNN) are used for the identification. The neural networks have been trained by Levemberg-Marquardt algorithm. To compare the performance of the neural networks models, generalization tests are produced where the aeroelastic responses of the blade in flapping and torsion motions at its tip due to noisy pitching angle are presented. An analysis in frequency of the signals from simulated and the emulated models are presented. In order to perform a qualitative analysis, return maps with the simulation results generated by the neural networks are presented.