Online Parameter Identification of Permanent Magnet Synchronous Machines under Sensorless Control

This thesis focuses on the online parameter identification of permanent magnet (PM) synchronous machines (PMSMs) under sensorless control at medium and high speeds.
Firstly, under position sensored control of interior PMSMs (IPMSMs), an improved position-offset injection based parameter identification method is proposed to identify dq-axis inductances and PM flux linkage simultaneously by three machine states corresponding to injections of zero, positive, and negative position offsets. Compared with the conventional position-offset injection methods, the proposed method can fully eliminate the influence of inverter nonlinearity and can be applied to both constant and variable speed operations within 5% estimation error.
Then, the position-offset injection method is extended into the position sensorless control of surface-mounted PMSMs (SPMSMs) and IPMSMs, respectively, to solve the influence of position error on parameter identification. In SPMSMs, during injecting square wave position-offset, parameters can be estimated by controlling the q-axis voltage fluctuation. In IPMSMs, due to more parameters than SPMSMs, a dual signal alternate injection method is proposed to simultaneously identify the multi-parameters.
Finally, a novel perspective is introduced to consider the influence of position error on parameter identification under sensorless control. On one hand, the novel virtual back-EMF and virtual flux-linkage are injected into back-EMF observer and flux observer, respectively, which is the indirect method to generate the position-offset response. Consequently, the position error caused by parameter error and inverter nonlinearity keeps constant and is only related to the ratio of q- and d-axis voltage fluctuation. Based on this characteristic, the influence of both position error and inverter nonlinearity on parameter estimation can be completely cancelled. Moreover, the injected virtual flux-linkage is independent of rotor speed, which can enhance the identification performance at low speed to minimum 2% rated speed under flux observer based sensorless control.
All developed methods have been verified by simulation and experiments.