Study of Artificial Neural Network and Observer-Based High Performance Induction Motor Drives

Abstract

The main subject matter of this dissertation is to study the performance of artificial neural network and observer based high performance induction motor drive. Four suitable flux observers compatible with drive control law are discussed and flux estimation with these observers along with effectiveness is studied. Study of the artificial neural networks for flux estimation with baekpropagation training algorithm for simulation is presented in this dissertation. It also presents the general idea about feedforward neural networks, mapping and training of an artificial neural network. The direct and indirect field orientation control methods of induction motor For variable operating conditions are evaluated in this study. In the direct method, flux estimation is applied for vector rotators which controls drive current or voltage magnitude as well as position SO that the rotor flux can be kept constant. In the indirect method flux estimation is used for parameter compensation. Digital simulation procedures are presented to study the performance of these observerbased field oriented induction motor drives. Speed of an induction machine is also estimated with full order observer and parameter adaptation is also presented for sensorless field orientation control. 1'he main cirawbuck of indirect method of field orientation is due to Variation of rotor resistance that degrades performance and requires tuning. Observers are used for detecting the parameter mismatch condition and correcting the controller resistance. By flux feedback the rotor resistance is adapted and the effectiveness of observers is also examined. Reduced order observer in generalized form is used for parameter adaptation of current source inverter fed system. ix Flux estimation with artificial neural network has been carried out and extended to direct field orientation of voltage source inverter fed induction motor system. Finally, comparison with the results obtained by artificial neural network is given.