A neural learning algorithm for online rotor resistance estimation in sensorless induction motor drive systems

Abstract

This research proposes an advanced artificial neural network (ANN) framework optimized for the dynamic, real-time identification of rotor resistance (Rr) in sensorless induction motor (IM) drive systems. The proposed architecture introduces a self-tuning momentum factor within the neural learning update rule, which is adaptively modulated at each sampling interval. This modulation is governed by a Mamdani-based fuzzy inference system to ensure accelerated convergence and enhanced stability of the estimation process. Concurrently, the motor's angular velocity is estimated through a parallel ANN observer. Reliable identification of the time-varying rotor resistance is pivotal for compensating parametric sensitivity in flux observers, thereby optimizing the drive's control fidelity under varying thermal and load conditions. Comprehensive simulation and hardware-in-the-loop experimental results confirm that the proposed estimator tracks the actual Rr with high precision, maintaining steady-state errors within a 5% threshold.