Robust ANN-based nonlinear speed observer for permanent magnet synchronous machine drives

This paper introduces a robust artificial neural network (ANN) based nonlinear speed observer for permanent magnet synchronous machines (PMSMs). A multilayer perception is trained online using back-propagation learning algorithm to estimate the rotor speed without any a priori dynamics knowledge. Thus, the proposed observer is able to cope with higher degrees of nonlinearity since it is not based on a linear-in-parameters model, unlike many neural network observers. Therefore, robustness to parameter variations is achieved. Simulation results for different situations highlight the performance of the proposed observer in the presence of high parametric uncertainties. The proposed observer is reliable and effective for PMSM drives.