Sensorless adaptive neural network control of permanent magnet synchronous motors

Rotor speed and position estimations in permanent magnet synchronous motors (PMSM) suffer from accuracy due to variations of machine parameters such as torque constant, stator resistance and inductance, especially at low speeds. Also, conventional linear estimators are not adaptive. In this paper, two neural network-based model reference adaptive systems (MRAS) for position and speed estimation are presented for PMSM drives. The first network estimates the rotor speed and adapts online to any change in stator resistance. The second network estimates the rotor position and adapts online for any changes in the torque constant. A q-axis model for stator inductance based on current is used. In both cases, the MRAS adjusts the neural weights to give optimal performance over a wide speed range. Simulation results showed the estimation adjusts well to changes in the motor operating points