Performance Analysis Of A Neural Flux Observer For A Bearingless Induction Machine With Divided Windings

Abstract

This article presents a system of vector speed control using neural estimation of rotor flux for a 1.1 kWatt winding bearingless three-phase induction machine. The neural estimator is composed of two multilayer feedforward linear networks and substitutes the estimator based on the inverse model of the machine, in which undesirable characteristics such as non-linearity and parametric variations that generate flux estimation errors are observed. The inputs of the developed networks are the currents in vector coordinates of the rotor flux and the mechanical speed of the machine, and their outputs are the angular speed of the rotor flux and the magnetization current. The vector controller operates in conjunction with radial position and current controllers. The control algorithm is implemented in a Digital Signal Processing (DSP) with six PWM monophasic inverters at 10 kHz switch frequency.1