Unscented Kalman filter for non-linear estimation of induction machine parameters

Abstract

In a previous authors' work, a short circuit and start-up tests have been used with the so-called two steps identification method to compute the induction machine's (IM) electrical and mechanical parameters. This latter approach was unable to compute IM parameters with a single dynamic test and could not be applied when IM identification data are noise corrupted. In the present study, the unscented Kalman filter (UKF) is used as non-linear optimal predictor for the saturated electromechanical stochastic dynamic model of IM. In order to overcome the uncertainty on the knowledge of noise-model parameters, the maximum likelihood estimation algorithm is combined to the UKF to compute IM parameters. The technique is successfully applied for the parameters estimation of a 2-kW, 4-pole, 10-A, 60-Hz laboratory induction motor using a start-up test noisy corrupted data. Furthermore, a cross-validation of the estimated model using short-circuit test greatly attests to the effectiveness and validity of the estimated IM model in a wide range of applications.