Sensorless direct vector control of induction motor using Kalman filter trained neuro observer

This paper demonstrates the development of a neuro observer which makes sim ultaneous estimation of rotor speed, rotor resistance and rotor flux possib le in a direct vector control scheme of an induction motor. The observer co mprises a real time recurrent neural network. Extended Kalman Filter algori thm is employed for training the neural network. The simulation results sho w that the observer identifies the rotor resistance within an iteration, ma king the observer suitable for real time applications. The speed estimation is robust against the variation in rotor resistance as Rr is simultaneousl y estimated. The dynamic behaviour of the motor in the direct vector contro l scheme has been found to be very good as the observer helps in getting pe rfectly decoupled components of current.