The performances of a model predictive control algorithm largely depend on the knowledge of the system model. A model-free predictive control approach skips all the effects of parameters variations or mismatches, as well as of model nonlinearity and uncertainties. A finite-set model-free current predictive control is proposed in this paper. The current variations predictions induced by the eight base inverter voltage vectors are estimated by means of the previous measurements stored into lookup tables. To keep the current variations information up to date, the three current measurements due to the three most recent feeding voltages are combined together to reconstruct all the others. The reconstruction is performed by taking advantage of the relationships between the three different base voltage vectors involved in the process. In particular, 210 possible combinations of three-state voltage vectors can be found, but they can be gathered together in six different groups. A light and computationally fast algorithm for the group identification is proposed in this paper. Finally, the current reconstruction for the prediction of future steps is thoroughly analyzed. A compensation of the motor rotation effect on the input voltages is proposed, too. The control scheme is evaluated by means of both simulation and experimental evidences on two different synchronous reluctance motors.

Due to their robustness and adaptability, position estimators based on the extended Kalman filter have been used in permanent magnet synchronous motors for decades. The time has come to extend their use to reluctance motors as well and this work focuses on the elements that hinder the transition. All passes through the availability of an accurate and analytical magnetic model, which is obtained by Artificial Intelligence tools. It is proved that the sensorless control of synchronous reluctance motors using the extended Kalman filter is possible over broad speed and torque ranges. The experimental session compares different implementation possibilities, concluding with the proposal of a new hybrid algorithm that greatly reduces the computational load.

Sensorless drives for permanent magnet (PM) and reluctance synchronous motors are quite interesting options in modern ac drives. Fundamental-frequency observers provide a rotor position estimate, to be used instead of that measured by delicate sensors. The estimation algorithms are based on either the extended electromotive force or active flux concepts. They feature easy implementation, although their reliability may be undermined by both corrupted measurements and parameter variations, especially in the case of anisotropic rotor structures. Through an analytical sensitivity analysis, this article investigates the influence of model uncertainty on the observed position. The substantial relationships between cause (parameter uncertainty or measurement error) and effect (position error) are validated by experimental results, which also highlights the effects of the q-axis inductance and resistance values on the electric measurement sensitivities. The resulting functions and graphic plots are useful to draw interesting considerations about the motor selection and the control design of the sensorless drive.

The general trend toward more intelligent energy-aware ac drives is driving the development of new motor topologies and advanced model-based control techniques. Among the candidates, pure reluctance and anisotropic permanent magnet motors are gaining popularity, despite their complex structure. The availability of accurate mathematical models that describe these motors is essential to the design of any model-based advanced control. This paper focuses on the relations between currents and flux linkages, which are obtained through innovative radial basis function neural networks. These special drive-oriented neural networks take as inputs the motor voltages and currents, returning as output the motor flux linkages, inclusive of any nonlinearity and cross-coupling effect. The theoretical foundations of the radial basis function networks, the design hints, and a commented series of experimental results on a real laboratory prototype are included in this paper. The simple structure of the neural network fits for implementation on standard drives. The online training and tracking will be the next steps in field programmable gate array based control systems.