Soft starter of an induction motor using adaptive neuro fuzzy inference system and back propagation based feedback estimator

Induction motor is most widely used motor in the industry. It requires sophisticated control of speed, inrush current and pulsations in the electromagnetic torque developed at the starting. This paper presents the neural network based soft starter with feed back estimator to support the on-line training of the network. Thyristor-based controller is used whose firing angles are adjusted by adaptive neuro fuzzy inference system (ANFIS) whose rule base was developed with the experience of the experts and off-line simulation data. A neural network based estimator is designed using back propagation based training algorithms to compute the electromagnetic torque, rotor angles and fluxes fed to ANFIS to adjust the firing angle of the thyristors. Back propagation based neural network was found to be the best among many others because this algorithm requires very small number of neurons. The presented approach can be used with off-line training as well as with on-line training and hence solve the problem of on-line computation of firing angle. Estimator developed was compared in results with the DSPbased estimator and results are shown.