NN-MRAS based speed estimator vs. RF-MRAS one: Design and comparison

This paper is concerned by the use of neural networks and for controlling a non-linear process namely MRAS based speed estimator of induction machine (IM). In the first case study, principals of a conventional rotor flux based MRAS (RF-MRAS) speed estimator will be designed. In the second case study, the design procedure uses a neural model trained with the inverse model of the process. Thus, the overall controlled system is formed using this inverse model. This work analyses each estimator in terms of tracking and regulation. It is shown that the neural network based MRAS (NN-MRAS) speed estimator is slightly better with respect to the conventional MRAS in the transient while they have quite similar behavior in the steady-state regime.