Dynamic decoupling control of bearingless switched reluctance motors based on neural network inverse system

Bearingless switched reluctance motors should remain stable levitation force and rotation force in different positions. Not only are the two forces nonlinear functions of positions, but also the levitation forces in two degrees of freedom have strongly coupled nonlinear relationship. Furthermore, the nonlinear coupled relationship exits between the levitation force and rotation force too. In order to realize the stable levitation and controlled rotation of bearingless switched reluctance motors, the first step is to dynamically decouple the levitation forces in different positions and to search for control laws in different positions. Based on basic electromagnetism theory, a radial force and position model of a bearingless switched reluctance motor is presented. Aimed at the nonlinear and strongly coupled characteristics, the model is analyzed with reversibility and proved to be reversible. The nonlinear and strongly coupled multi-variables system can be decoupled and transformed into two linear subsystems without position coupling to each other by connecting a neural network inverse system before a bearingless switched reluctance motor. This neural network inverse system consists of a static neural network (MLN or RBF network) and two integrators, where the static neural network represents the nonlinear mapping relation of the inverse system and the integrators represent the dynamic characteristics of the inverse system. Consequently, the high performance control of the original nonlinear and coupled system can be realized under the help of linear closed-loop controllers for each decoupled subsystem. The results of simulation show that this system can realize the stable levitation of bearingless switched reluctance motors.