Hybrid Neural Sliding Mode Observer for Speed-Sensorless Control of Induction Motor

The state of the art demonstrates that Sliding Mode-based Observers (SMO) can effectively meet the speed-sensorless control requirements of induction machines. Even so, parametric uncertainties adversely affect the switching operation of the SMO and intensify the chattering, which appears as variable frequency noises in the observed speed. The chattering stimulates the ignored high-frequency dynamics of the system and leads to poor control performance and instability of the drive system under dynamic conditions. Concerning this issue, an appropriate filtering strategy is required to filter out the chattering in the SMO-based sensorless control of induction machines. However, filtering limits the bandwidth of the speed observer and weakens the dynamic performance of the controller and drive. To address this issue, a novel hybrid Neural Sliding Mode Observer (NSMO) is proposed and developed in this paper. In the NSMO, an artificial neural network is effectively trained to estimate a chattering-free speed signal based on the stator currents, reference speed, and observed rotor flux. In addition, the proposed NSMO is not sensitive to the motor parameters variation, thanks to the model-free structure of the ANN-based speed observer. In the end, the simulation results of various dynamic and comparative test scenarios performed by MATLAB verify the superiority of the proposed technique over an SMO technique.