A comparison between model-based observer and neural network for induction motor rotor flux estimation

This paper proposes the use of a neural network (NN) for the induction motor rotor flux estimation and presents a comparison with two model-based observers. The paper deals with a neural network having eight inputs, two outputs, and one hidden layer with six nodes using sigmoidal-functions. The NN outputs are the α,β components of the rotor flux. The NN is recurrent, that is the two past outputs at time n are fed back into the network. The other inputs are the α,β components of stator voltages at time n, and the α,β components of stator currents at time n and n-1. Once sufficiently well trained the NN improves the drive robustness when parameter variations occur. The standard observers used for comparison are based on the current model. Simulations are presented to show the advantages of the NN over the standard observers in the vector control of induction motor drives.