Título: Implementação do Servocontrole Auto-Sintonizado em Tempo-Real Utilizando Rede Perceptron Multicamadas
Autores: Vieira, Zenilda Pedrosa; Bauchspiess, Adolfo
Resumo: This paper shows first results of the identification and compensation, using artificial neural networks, of one of the main non-linearities that occurs in servo positioning systems: the operation point dependency of the system gain. In the data acquisition a DC current motor with velocity and position feedback was used. The dead-zone and the saturation are treated using standard non-linear techniques. A multilayer perceptron network (MLP) trained with the classical back-propagation algorithm was used to indentify these non-linearities. The control-loop was closed using a standard discrete PID. The results shows that the network accomplish this task in a satisfactory way, linearizing the system in the dynamic operation range, off course inside the saturation limits.
Artigo em pdf: 4cbrn_072.pdf
Arquivo BibTex: 4cbrn_072.bib