Identification and Control of Switched Reluctance Motors Using Artificial Neural Networks and a Priori Knowledge
Date of Award
Dissertation - Restricted
Doctor of Philosophy (PhD)
Electrical and Computer Engineering
Brown, Ronald H.
Abdul-Rahman, A. A.
Belfore, Lee A.
The identification and control of switched reluctance motors (SRMs) is of primary interest in the area of controls because of their reliability and high torque to weight ratio. This makes them ideal for use in many applications. However, these distinctive characteristics are not without drawbacks. Currently, no precise on-line methods have been developed for identification and control of these motors because of the highly nonlinear way that static torque varies with rotor position and stator current. Artificial neural networks (ANNs) can be treated as nonlinear function approximators. Given the proper weight updating techniques and network architectures, artificial neural networks can represent a computational model of an unknown plant. By merging the attractive features of artificial neural networks and the requirements needed for control of switched reluctance motors, an identification and control scheme can be developed and applied to switched reluctance motors. This dissertation, by means of simulated implementation, shows the feasibility of using artificial neural networks to identify and control switched reluctance motors. The main sections of this dissertation are as follows: I. Expand the current status of the 'gray layer technology'. This involves ANN architecture development. A novel modification to the gradient descent algorithm is also discussed in light of any inherent periodicity known to exist a priori in the plant that the ANN is modeling. 2. Form an appropriate computational model of an SRM and drive circuit for implementation in the simulated environment. 3. Achieve an identification model of an SRM using ANNs and available a priori information about SRMs. 4. Develop a control scheme suitable for the insertion of an SRM, again including a priori information, and demonstrate the results.