NEURAL NETWORK BASED MRAS FOR SENSORLESS INDUCTION MOTOR DRIVES TO IMPROVE PERFORMANCE AT LOW SPEEDS

Abstract

This paper proposes a new Neural Network based MRAS speed observer for sensorless vector controlled induction motor drive. This neural network replaces reference model (voltage model) in conventional MRAS. In conventional MRAS, reference model equations depends on stator resistance that changes with temperature during running condition. This change in stator resistance is predominant in low and zero speed operation. In proposed MRAS, all drive non-linearities are included. Hence need for separate stator resistance estimator and integrator problem are eliminated in the proposed neural network based MRAS. Simulation work is done in various operating conditions using MATLAB/Simulink software. Better steady state and dynamic performances are achieved with proposed neural network based MRAS.