Back-EMF Based Rotor Position Estimation for Low Cost PMSM Drive Using Fully Connected Cascade Artificial Neural Networks

Abstract


 Permanent Magnet Synchronous Motors (PMSMs) are widely used mainly due to their high torque per volume, high efficiency and low maintenance cost, among other advantages. To perform the vector control on rotor speed and stator currents, the feedback of those variables is necessary, which can be done directly or by estimation. Measuring rotor position and speed directly requires the use of a mechanical device attached to the motor shaft, increasing the drive system volume and its maintenance cost. To overcome such disadvantages, many sensorless methods for speed estimation have been proposed. Among those methods, various strategies based on Artificial Neural Networks (ANNs) can be found. This paper presents a back-electromotive force estimator based on Fully Connected Cascade ANNs (FCC-ANNs). From the estimator, rotor position and speed can be obtained. Simulation and experimental results using automatically generated C code functions for the FCC-ANNs using fixed point notation provided rotor position estimation with simple implementation.