Soft starter of an induction motor using neural network based feedback estimator

This paper presents the neural network based soft starter which uses two neural networks; one with radial basis function which estimates the electromagnetic torque, and rotor fluxes and angles while the other uses feed forward back propagation algorithms to decide firing angle of thyristors in AC voltage controller. The neural networks were trained with simulation data. ANN models have the ability to learn from input and output samples in defined boundaries off-line as well as online. Radial basis function was preferred because of low training time requirement for large number of samples. DSP estimator was also implemented to check the validity of the designed ANN models.