Magnetic bearing rotor displacement estimation using O-RLS-SVM

A new approach for magnetic bearing rotor dis-placement estimation using on-line recursive least squares support vector machine (O-RLS-SVM) is proposed. The basic premise of the method is that an O-RLS-SVM forms a very efficient mapping structure for the non-linear magnetic bearing. Through measurement of the phase flux linkages and phase currents the O-RLS-SVM is able to estimate the rotor displacement, thereby facilitating elimination of the rotor displacement sensor. The O-RLS-SVM training data set is comprised of magnetization data for the magnetic bearing with equivalent flux linkage and equivalent current as inputs and the corresponding displacement as output. Given an updating training sample set, the O-RLS-SVM could build up an appropriate adaptive SVM architecture to express the dynamic behavior of magnetic bearing. This paper presents the development, implementation, and operation of O-RLS-SVM-based displacement estimator for a three-phase AC active magnetic bearing.