This paper presents a novel neural adaptive Kalman filter for speed sensorless field oriented vector control of induction motor. The adaptive observer proposed here is based on MRAS (model reference adaptive system) technique, where the linear Kalman filter calculate the stationary components of stator current and the rotor flux and the rotor speed is calculated with an adaptive mechanism. Moreover, to improve the performance of the PI classical controller under different conditions, a novel adaptation scheme based on ADALINE (ADAptive LInear NEuron) neural network is used. It offers a solution to the PI parameters to stabilize automatically about their optimum values and speed estimation to converge quicker to the real. The proposed adaptive Kalman filter represents a good comprise between estimation accuracy and computationally intensive. The simulation results showed the robustness, efficiency, and superiority of the proposed scheme compared to the classical method even in low speed region.

No abstract

No abstract

This article develops a new observer to achieve highly speed estimation for induction motor control. Its principle consists of estimating flux and speed separately, by using linear Luenberger as flux observer and intelligent PI controller as rotor speed estimator. Considering that is generally challenging to established proper PI parameters, the ADALINE algorithm is integrated as a trainable module to automatically calculate the KP and the KI gains in each iteration during all operation induction motor. This algorithm helps to search on the one hand the best regulation of the controller in real-time, taking into account the objectives to achieve (an accurate estimation) and the respect constraints (heavy calculation). On the other hand, it makes it possible to improve the adaptive Luenberger estimation problem in the unobservable zone (low speed). The experimental comparison of the observers demonstrates the visible improvement of the enhanced observer. It can converge to real speed in a short time with less static and transient error.