Application of computational intelligence to high-performance electric drives

This dissertation investigates the application of computational intelligence techniques to two open problems in high performance electromechanical drives. The objective is to develop non-linear state estimation methods and control systems for the problems under investigation.; The performance of speed sensorless control of induction motor drives currently suffers due to the absence of an accurate non-linear state estimator. An ideal estimator has to be robust to large motor parameters variations and should be able to account for non-linearities. Neural networks offer many advantages over mathematical models including the ability to learn from training examples and approximate any complex non-linear function. A new speed estimator for induction motor drives is presented here along with simulation results. Extended Kalman filters are widely used as non-linear state estimators when the model for a non-linear plant is known. A new adaptive, non-linear state estimator is developed in this work which combines neural networks with an Extended Kalman filter. Simulation results for this adaptive estimator are presented. The developed state estimator does not specifically depend on any properties of induction motors and can be used with a class of non-linear systems.; The second problem investigated here is Precision Position Control using Elastic Links. At present, rigid transmission elements such as lead screws are used in applications requiring a positioning accuracy better than 50 microns. Elastic transmission elements such as belt drives can be used in such applications if the control system is improved to obtain better accuracy. The use of belt drives makes the plant non-linear and reduces the accuracy that can be obtained using linear control systems. A non-linear controller is required to obtain high accuracy control using such elements. In this work, a combination of non-linear feedback control using fuzzy logic and non-linear feedforward control using evolutionary algorithm based system identification is developed. Experimental results show the efficacy of the control system in obtaining a positioning accuracy of 0.5 microns. The control system developed can be applied to other elastic systems without significant design changes. A self-tuning algorithm using evolutionary programming is developed which gives better results than a manually tuned controller.