Abstract
From the electromagnetic theory, we know the electromagnetic model of the magnetic levitation system is usually strongly nonlinear, depending on the length of the air-gap. This paper presents a feedback error learning control for a 4-point attraction magnetic levitation system. Firstly, taking account of the nonlinear model of the system, we design the stable closed-loop using a PD controller based on the mechanical model and the normal inverse model of the electromagnetic system (IMES). Secondly, by combining the mechanical and electromagnetic characteristics of the magnetic levitation system with the generalized radial basis functions (GRBF), we present a new kind of hybrid neural network (HNN) to learn the inverse model of the nonlinear system. The HNN has the advantage that the learning algorithm is linear and is therefore fast. At last, experimental results are included to show the excellent performance of the designed control system and it is verified that even in the case of abruptly large changes of the system parameters, the control performances are still very satisfactory.
