Session: 17-01-01 Research Posters

Paper Number: 69771

Start Time: Thursday, 02:25 PM

69771 - A Time-Frequency Domain Adaptive Control Approach for Vibration of Active Magnetic Bearing System 

Active magnetic bearings (AMBs), which are emerging suspension technology, are getting more applications in rotating machinery. The features such as non-contact and active control, make AMBs having several advantages over conventional mechanical bearings. However, it is challenging for control system implement because of its nonlinear electromagnetic characteristic as well as the complex rotor dynamics. Thus the control system is the key factor of AMB.

Various control strategies and algorithms have been applied and designed for AMB system. Classical linear control methods based on PID control or robust control are widely used and are proven effective in industry applications. With the development of advanced control theory as well as the improvement of processor performance, various complex methods and algorithms such as fuzzy logic control, sliding mode control and neural network, are also popular in this research area. Moreover, the combinations of different methods such as and further improve the control effect. Besides, adaptive feedforward control based on LMS (Least mean square) algorithm, which is common in active vibration control, and some novel algorithms are applied for AMB control as well.

Although these control methods achieve significant effect, one factor should not be neglected. Most control schemes are in time domain by processing data of single time, while the control in frequency domain, which may cause potential problems, is rarely considered. If control can also consider the frequency domain, the stability of system and the control precision could improve simultaneously. Tang and Lee used short-time Fourier transform (STFT) and filtered-x LMS algorithm to implement a hybrid time-frequency domain control scheme for active noise control. Luo et. al. proposed an adaptive fuzzy dividing frequency-control method for hybrid active power filter. Their method consists of two control units: a generalized integrator control unit and fuzzy adjuster unit. Simulation and experimental results shows its advantages of shorter response time and higher control precision. Wavelet, which is an indispensable tool in signal processing and time-frequency analysis, is also an attractive approach for control in time-frequency domain. A wavelet-PID controller is used for interior permanent-magnet motor drives, and comparison at different dynamic operating conditions proves the superiority of the proposed controller over conventional fixed-gain controllers. Daya et. al. proposed a multiresolution controller which is a hybrid of discrete wavelet transform (DWT) and self-tuning fuzzy logic algorithm, and proved its effectiveness. In AMB application, Cade et. al. utilized a technique based on multi-level wavelet coefficient analysis with transient system dynamics. Liu and Suh proposed an adaptive feekforward controller based on DWT and filtered-x LMS algorithm to minimize the vibration of AMB system. Although these wavelet hybrid approaches work well, the linear control methods like PID or fuzzy-PID limit the effect for complex nonlinear system.

In this study, a time-frequency domain control approach is proposed for AMB-rotor system. The control scheme is implemented using wavelet theory and deep learning theory. First, the operation principle of a 4-DOF AMB-rotor system is analyzed and its model is established. Then a novel controller is proposed and its operation algorithms is designed. The controller consists of 2 main part: a filter bank for DWT to obtain time-frequency domain signal, and a deep neural network (DNN) for multi-dimension adaptive nonlinear control. Finally, the rotor dynamics are simulated and the vibration is analyzed. The instantaneous frequency via Hilbert-Huang Transform (HHT) is utilized for time-frequency analysis. Simulation results on vibration are comparedto indicate the advantages of the proposed control approach in both time domain and in frequency domain.

Presenting Author: Xuan Yao Harbin Institute of Technology

Authors:

Xuan Yao Harbin Institute of Technology
Zhaobo Chen School of Mechatronics Engineering, Harbin Institute of Technology