Eventos Anais de eventos
COBEM 2021
26th International Congress of Mechanical Engineering
Time domain identification of dynamic parameters of magnetic bearings using statistical learning methods
Submission Author:
Felipe Vieira , RJ , Brazil
Co-Authors:
Felipe Vieira, Diego Alejandro Godoy Diaz, Fernando Augusto de Noronha Castro Pinto
Presenter: Felipe Vieira
doi://10.26678/ABCM.COBEM2021.COB2021-0434
Abstract
Magnetic bearings are used to control and monitor rotating machinery, and modelling its behavior is essential for understanding the limitations of the system in operation. However, the experimental conditions for identification of their parameters cannot be maintained for long. The noise in observations is usually bypassed by the use of methods in frequency domain and the acquisition of data during multiple periods. In this work, we propose the use of Savitzky-Golay filter for noise attenuation and two methods for time domain identification of dynamic coefficients. The first is based on optimization algorithms adopted by machine learning literature. The second one is a linear regression that favors online estimation of parameters. Both methods demonstrated similar results when compared to a baseline frequency-domain method, when the system is excited by harmonic signals. The use of linear regression with multisine excitation was also analyzed.
Keywords
Magnetic Bearings, System Identification, dynamic stiffness, Savitzky-Golay, Simple Linear Regression
