Multivariable Identification of Active Magnetic Bearing Systems

Abstract

In many applications of magnetic bearings, high system performance (disturbance rejection) is required. Often several eigenfrequencies of the rotor must be actively damped. Advanced controller design is then necessary, which in ture relies on accurate plant models. System identification is therefore an important prerequisite for fast and reliable commissioning. Magnetic bearing systems are unstable MIMO*1 plants. In case of current control, they are characterized by a set of poles on the positive and negative real axis (rigid body modes of the rotor), and a set of poorly damped poles along the imaginary axis (flexible modes). Previously described identification algorithms fail with this kind of system. Therefore a novel algorithm has been developed. The system poles are identified from the determinant of the measured frequency response function matrix. The algorithm provides a state-space model of pre-defined order and structure, suited for controller design and verification. Experimental results obtained using measurement data from a magnetic bearing system with a flexible rotor are presented.