Eventos Anais de eventos
COBEM 2023
27th International Congress of Mechanical Engineering
Comparative Study of Artificial Neural Networks and Augmented Kalman Filter models applied to balancing of Flexible Rotors without test mass
Submission Author:
Raimundo Neto , RN , Brazil
Co-Authors:
Raimundo Neto, Stanley Washington Ferreira Rezende, Leonardo Cabral, Aldemir Ap Cavalini Jr, Arinan Dourado, Valder Steffen Jr
Presenter: Raimundo Neto
doi://10.26678/ABCM.COBEM2023.COB2023-0368
Abstract
Among the main failures that occur in rotating machines, unbalance stands out, which occurs due to the irregular distribution of mass along the rotor shaft, that is, when the center of mass of the shaft does not coincide with its geometric center. To mitigate this inconvenience, balancing techniques based on vibration responses were developed, whether in the time or frequency domain. Among the main contributions in this field, the Modal Balancing and the Influence Coefficients stand out. Despite their wide application in the industry, these techniques present limitations regarding the requirement for trial weights. In this context, the main objective of this work is to carry out a comparative study between two rotor balancing techniques in which the trial weights are not required. The first one is based on the application of Neural Networks (NN), which seeks to estimate the correction masses and the corresponding angular positions using the vibration responses of the rotor operating under normal conditions. It is worth mentioning that the training process of the NN is performed by using the vibration responses of the unbalanced rotor as inputs and the correction masses and the corresponding angular positions obtained with the application of a regular balancing technique, like the Modal Balancing and the Influence Coefficients, are the outputs. Its application for balancing purposes is performed using the vibration responses of the unbalanced rotor. The second methodology is based on the implementation of the Augmented Kalman Filter (AKF) together with the optimization of the uncertainty matrices. Once the unbalance forces are estimated, the application of the associated correction masses are carried out. Both methods are compared with the Influence Coefficients approach. The results showed that both methodologies applying NN and AKF were able to perform the balancing satisfactorily, however, the NN presented better results. Furthermore, the NN demonstrated to be more interesting regarding its computational costs as it does not require the mathematical model of the rotor to be applied.
Keywords
Augmented Kalman Filter Algorithm, Artificial neural networks (ANN), Rotating Machine Balancing
