A Time-Frequency Domain Adaptive Control Approach for Vibration of Active Magnetic Bearing System

Abstract Active magnetic bearings (AMBs) have several advantages such as non-contact and active control, and are getting more applications in rotating machinery. Various control strategies have been applied and designed for this nonlinear system with complex rotor dynamics. Most control schemes are in time domain, while the control in frequency domain, which is also essential for stability, is rarely considered. In this paper, a time-frequency domain control approach is proposed for AMB-rotor system. The control scheme is implemented using wavelet theory and deep learning theory. The controller consists of 2 main parts: a filter bank for discrete wavelet transform (DWT) to obtain time-frequency signal, and a deep neural network (DNN) for nonlinear adaptive control. A 4-DOF AMB-rotor system is analyzed and its model is established. The rotor dynamics are simulated and the results are compared. Simulation results demonstrate that the proposed approach has an obvious control effect in improving precision in time domain and stability in frequency domain. This research provides a new adaptive control approach for AMBs, and this approach can also be adopted in other multi-dimension vibration control, especially in multi-frequency applications.