Modeless Magnetic Bearing System Tracking Using an Adaptive Fuzzy Hermite Neural Network Method

Abstract

This paper proposes a reference model adaptive fuzzy Hermite neural-network controller (RAFHNC) for position tracking of a modeless magnetic bearing system (MBS). Magnetic bearings evolved from traditional bearings and can eliminate the shortcomings of traditional bearings by using electromagnetic forces to reduce friction during use. The MBS is nonlinear, for which system parameters are difficult to obtain accurately. This paper proposes the RAFHNC for precision position tracking in an MBS. The RAFHNC has three important characteristics. First, the RAFHNC does not require the MBS parameters to track a reference model accurately. Second, the RAFHNC uses Hermite functions to replace the Gaussian function, this reduces training time in the membership level. Third, the RAFHNC uses an online self-tuning fuzzy Hermite neural network estimator to resolve lumped uncertainty in the MBS. The fuzzy Hermite neural network incorporates expert knowledge, reducing training time on controller parameters. Finally, the Lyapunov stability and system adaptive laws are used to guarantee convergence. The output responses of the RAFHNC and adaptive reference model sliding-mode controller (ARSC) methods were compared. The RAFHNC method exhibited improved output responses compared with the ARSC method in terms of root mean squared error (RMSE) index. It can reduce problems associated with external loads and system lumped uncertainty in MBS.