Active Hybrid Bearing Using Neural Network Applied Model Predictive Control
Henry Pizarro Viveros1; Rodrigo Nicoletti2
1 University of Sao Paulo, Sao Carlos School of Engineering; 2 University of Sao Paulo, Sao Caelos School of Engineering
doi:10.20906/CPS/COB-2015-1129
Resumo
Rotating machines are essential elements in the industry production chain. Hence, they must present high performance and availability. It is a crucial parameter to keep production and avoid financial losses. The fact that rotating machines could present instability under some operating conditions led to studies focusing on increasing the machine-damping reserve and the enlargement of stability ranges. The Hybrid Hydrodynamic-Magnetic Bearing (HHMB) is a hydrodynamic bearing with embedded electromagnetic actuator a new generation of active hybrid bearing. Although traditional control techniques are known applications for these systems, the identification of nonlinearity of the process to use them in neural network controller and attenuation of lateral shaft vibration by active hybrid bearings are still limited. In this paper describes the control of a shaft for a SISO process through a Neural Network applied Model Predictive Control (NNMPC) thus, so are attenuate the lateral vibration of a shaft through HHMB. For that, a recurrent neural network identifies the nonlinear system, this identification is done offline using the Levenberg-Marquardt learning algorithm. Thus, the controller calculate future control signals solving online at each sampling instant a quadratic optimization equations based on the error between the nonlinear prediction and the reference trajectory on a prediction horizon. Numerical results show the identification process present a high generalization ability of the nonlinear process, the error for the training data set MSEtrain = 3.077e-10 and for the test data set MSEtest = 5.328e-10. Two cases were tested: without external disturbance (unbalance force) and with external disturbance for three different rotating speeds (20, 30 and 40Hz). In both cases, we have a good response because the displacement of the shaft was successfully follow the changes of reference trajectory. The displacement of the shaft is controlled successfully, hence increases the dynamics coefficients of hybrid bearing (equ
Palavras-chave: Rotor Dynamics; Neural Network; Active Bearings; Model Predictive Control; Vibration; Control System