Search Results

The control and analysis of magnetic bearings has been primarily based upon classical linear control theory. This approach does not allow for some important system complexities and nonlinearities to be taken into account. The resulting simplifications degrade the overall system performance. This paper investigates the use of a neural network to control a magnetic bearing flywheel energy storage system. A plant simulation is developed as well as a neural network emulator and controller.

AMBER (Active Magnetic Bearing Evaluation Routine) is a computer algorithm developed for the University of Maryland pancake magnetic bearing, which supports and controls a flywheel in a kinetic energy storage system. Because of the gap growth due to centrifugal forces at high speed, the bearing axial load capability degrades and the axial characteristics become critical in the bearing design. AMBER applies magnetic circuit theory, magnetic material saturation curves, coenergy theory, and finite permeance-based elements to solve the air gap flux density and coenergy over a series of incremental axial displacements. Differentiation of the coenergy of the magnetic field yields axial force and stiffness characteristics. An axial test machine is constructed to conduct experiments to verify the flux distribution and axial forces predicted by the model. User interaction with AMBER allows modification of the bearing geometry and composition to optimize future prototypes.