<![CDATA[This study addresses the limitations of traditional Model Reference Adaptive Systems (MRAS) in sensorless induction motor (IM) control, particularly the degraded performance at low speeds and under dynamic load conditions. The main objective is to enhance speed and torque estimation accuracy by replacing the classical proportional-integral (PI) adaptation mechanism with an adaptive neuro-fuzzy architecture. The research contribution lies in developing and experimentally validating two intelligent adaptation schemes: one based on fuzzy logic and another combining fuzzy inference with a recurrent neural network (RNN) within a sensorless field-oriented control (FOC) framework. The proposed system integrates a fuzzy logic-based estimator and an RNN-driven torque predictor to improve tracking precision and robustness. Real-time implementation was carried out on a 1.1 kilowatt, 1430 revolutions per minute induction motor using a dSPACE DS1104 platform. Comparative experiments were conducted under two challenging benchmark profiles that include load disturbances, parameter mismatches, and full-speed reversals. Results showed that the hybrid neuro-fuzzy controller reduced the steady-state speed error by 91 %, from 0.65 rad/s to 0.08 rad/s, and improved torque estimation accuracy by 42%, reducing SMAPE from 45.2 % to 26.3 %, compared to the PI-based MRAS. It also outperformed the standalone fuzzy and neural MRAS controllers in rise time, tracking error, overshoot suppression, and adaptation quality. These findings confirm that the proposed method provides improved estimation fidelity, enhanced control robustness, and reliable sensorless operation suitable for real-time industrial applications. The study concludes that the integration of neuro-fuzzy intelligence into MRAS-based control structures offers a technically effective and scalable solution for advanced IM drives.]]>
