Low speed operation improvement of MRAS sensorless vector control induction motor drive using neural network flux observers

This paper presents a novel neural network-based flux observer to solve the low speed problems associated with a model reference adaptive speed estimation scheme which is based on rotor flux. A multilayer feedforward artificial neural network is proposed for rotor flux estimation which is more robust to noise and stator resistance variation and does not have DC-drift problems which are usually associated with these adaptive schemes. A comparison between the performance of the neural network based strategy and conventional scheme is carried out using a validated simulation of an indirect vector controlled induction motor drive working at a low speed