Abstract
For sensorless controlled salient-pole brushless DC motor drives, a new AC current injection method which estimates the position angle with d and q inductances was proposed. However, the estimated position angles include the errors which are casued by an unsymmetrical construction based on the mechanical or magnetic tolerance on the motor. It is difficult to correct the errors with hardwares like filters.
An error correcting technique of an estimated position angle and rotating speed is proposed. The errors which are casued by the unsymmetrical construction appear as a modeling error of the motor periodically. Then, a neuro-like fuzzy technique can be applied to correct the error of the estimated position angle. When the motor is controlled at a commanded speed, the speed of the motor contains unexpected cyclic speed errors. The fuzzy model which estimates these errors is learned with a back propagation method which uses the cyclic errors as a teacher signal. Then, the system corrects the estimated position angles with the estimated errors. From the experimental results, the proposed sensorless control system could estimate more accurate position angle than conventional systems could.
