Advanced LARC Strategy of Maglev Planar Motor with GRU Neural Network Prediction

Abstract

To achieve high motion control accuracy and performance robustness simultaneously, this paper proposes an advanced learning adaptive robust control (LARC) strategy with gated recurrent unit (GRU) neural network prediction for magnetically levitated (maglev) planar motor. Adaptive model compensation and robust feedback control is firstly applied to guarantee robustness against parameter uncertainties and unknown disturbances. A GRU neural network is then trained with dataset collected from a practical maglev planar motor control system. The accurate predicted tracking error by the trained GRU neural network is compensated into the reference trajectory, which forms the proposed LARC strategy. Comparative experimental investigation validates that the proposed LARC strategy achieves comparable motion accuracy to iterative learning control (ILC) while avoiding undesired time-consuming iterations. Additionally, the proposed strategy outperforms ILC due to that satisfying control performance can be preserved even in the presence of trajectory variation, parameter uncertainty and unknown external disturbance.