A comparison between directly parametrised observers and extended Kalman filters for sensorless induction motor drives

This paper presents a performance comparison between the directly parametrised full-order observer (for flux and speed estimation) presented by L. Harnefors and H.P Nee (see Proc. Eur. Conf. Power Electron. Appl., Trondheim, Norway, vol.3, p.375-81, 1997) and the extended Kalman filter (EKF). It is shown that, although the noise content of the flux estimate of the EKF is lower than that of the directly parametrised observer, it tends to be less accurate in the presence of erroneous model machine parameters. Furthermore, the computational requirements of the directly parametrised observer are vastly lower than those of the EKF, while it also is easier to incorporate gain scheduling and/or adaptation of machine parameters.