Speed sensorless control system of induction machine using adaptive rotor resistance observer

Due to quick torque response and robustness against parameter variation, the direct torque control system has been widely utilized in the industrial production. To improve its dynamic performance, a novel approach using wavelet network for identifying stator resistance on-line is proposed. The wavelet transform can accurately detect and localize signal characteristic in time-frequency domains, where the real-time stator resistance variation can be tracked. The wavelet network provides a sufficient structure that guarantees the approximation precision. Comparing with neural network, the wavelet network has four merits: self-construction network, partial retrieval of approximated function, fast convergence and escaping local minima, improving dynamic performance of direct torque control in low speed. The input nodes of wavelet network have two variables: one is stator current error, and the other is variance ratio of stator current error. The output node of wavelet network is stator resistance error. The improved least squares algorithm is used to complete the network parameter initialization, which would have good convergence property. The simulation results prove that the proposed method can efficiently reduce the torque ripple and current ripple, optimizing the inverter control strategy.