Neural network identifier-based dynamic decoupling control of a novel permanent magnet spherical actuator

Abstract

The real-time torque calculation and dynamic modelling have been the difficulties in the motion control of the permanent magnet spherical actuator. Based on the presented new configuration design, a simplified torque calculation model and a non-linear system dynamic model have been proposed. The perturbation of the system parameters, the model error and the external disturbance result in the actual system uncertainties, which have great effects on the control performance and caused the interferences on three degrees-of-freedom. Based on the modelling, this paper also presents the dynamic decoupling control scheme for the permanent magnet spherical actuator with neural network identifier. The feed forward neural network and adaptive learning algorithm are adopted to construct the neural network identifier for online identification of uncertainties and unmodelled dynamic parts in the motion process. The simulation and experiment results show the effectiveness of the proposed method by alleviating the inter-axis cross-coupling influences and offering high robustness to guarantee the static and dynamic performances.