Robust vector control without speed sensor of an induction motor using neural network and extended Kalman filter

The rotor circuit time constant is an important parameter for indirect field oriented control. Incorrect estimation of the rotor time constant also leads to incorrect flux angle calculations and can cause significant performance deterioration if no means for compensation or identification is applied. The magnetic saturation, the operating temperature and difficulties in using sensors for speed measurement are one of the sources of parameters variations, caused by non-linear nature of the magnetizing curve and the variations of the rotor resistance. The classical indirect field-oriented control is highly sensitive to the inductance values decrease when increasing the saturation level. To solve this problem, this paper presents a detailed study of the application of extended Kalman Filter (EKF) to parameters estimation for speed sensor less and neural network (ANNs) of an induction motor. The general structure of the EKF is reviewed and the various systems vectors and matrices are defined. The elements of the covariance matrices are properly selected. The EKF associated to the neural network (EKF-ANNs) trained off line algorithm are used to estimate the rotor resistance, the main inductance and the rotor speed. By including these parameters as states variables, the EKF equations are established from a discrete two-axis model of the three-phase induction motor. The proposed EKF-ANNs compensation has shown a good performance in both the transient and steady state operations even in the presence of noise, and also at either variable speed operation in the field-weakening region or in the load variation.