Speed control accuracy and robustness in sensorless drives based on non-salient PMSMs, especially in the low-speed region, are critical issues mainly due to the low amplitude of the back-EMF, being the main information that state observers are normally exploiting, [1]-[4]. Open-loop control (e.g. constant amplitude rotating current space vector), is normally employed to start the motor up to an enough high speed, where sensorless closed-loop control is possible. Extension of the closed-loop control range towards zero is however strongly desired. Besides the adopted estimation techniques, it has been shown experimentally that adding an arbitrary direct-axis current component to the current reference leads to an important improvement in the feedback control capability, [2][5]. This phenomenon can be intuitively explained by considering that the rotor magnetic axis tends to align to the current vector angle, then the estimated direct-axis current forces convergence of estimated and actual position. In this paper, an unpublished analytical explanation of this effect is presented, while extensive simulation and experimental investigations demonstrate the effectiveness of the adopted approach and the resulting advantages in an actual sensorless industrial drive system.
