Mehrotra, Prashant (1999) Artificial neural networks in induction motor speed estimation and control. Doctoral (PhD) thesis, Memorial University of Newfoundland.
- Accepted Version
Available under License - The author retains copyright ownership and moral rights in this thesis. Neither the thesis nor substantial extracts from it may be printed or otherwise reproduced without the author's permission.
The squirrel-cage induction motor has various inherent advantages not present in other types of ac motors, and is widely used in the industry. Its usage is expected to go up because of possible applications like electric vehicles, which require a light and efficient motor drive. However, the induction motor has a complex and non-linear structure which makes precise control a complicated and expensive process. Added to this complexity is the fact that the motor parameters undergo a variation during regular operation, chiefly due to a change in temperature and nonlinear magnetic characteristics. This variation reduces the efficacy of the control technique, though its effect can be mitigated with the help of robust control techniques. Also, most control techniques require speed feedback from a shaft encoder and these devices have various disadvantages and are considered undesirable for a number of applications. Thus, present day research in this area is mostly focussed on obtaining speed sensorless and robust induction motor drives. -- Artificial neural networks (ANNs) have shown great promise in image processing and control applications where robustness is desirable. However, these are at the stage of infancy in the area of induction motor control. The ability of ANNs to map arbitrary nonlinear functions has been used to advantage by many researchers. The motivation behind this work was to investigate the possibility of using ANNs to eventually come up with an ANN based sensorless induction motor drive. This central idea was broken down into two major components - speed estimation of induction motors using ANNs, and control of induction motors using ANNs. Both these areas have attracted attention in recent years, though very little work has been done so far. Because of the complexity of the problem, researchers have been unable to come up with a satisfactory solution. -- This work makes an important contribution to the area of induction motor drives, by presenting for the first time, off-line trained ANN speed estimators. Using the d-q axis dynamic equations of the squirrel-cage induction motor, four methods are proposed whereby an ANN is trained off-line to estimate the speed of the motor. The results presented in the thesis indicate that the proposed schemes are able to track the speed under load variations. The effectiveness and superiority of the fourth method is further demonstrated under vector control conditions in the presence of an inverter. This method has also been experimentally verified. -- A novel strategy for control of induction motors using just one off-line trained ANN is also presented. The control strategy employs the magnitude and frequency of the d-q axis quantities to simplify the off-line training of the ANN and allow the ANN to mimic a vector controller. This scheme has the added benefit that subsequent to off-line training, the ANN can be on-line trained for improved performance and robustness, though it can function well without any on-line training also. Simulation results show that after off-line training the ANN is able to run the induction motor for various changes in speed reference and load torque, and the network is able to generalize effectively. Further simulation results are presented to show the robustness of the control strategy under induction motor parameter variation when the ANN controller is functioning under on-line control. An off-line trained ANN is particularly useful for real-time implementation, because of the reduced computational burden. -- Though the problem of obtaining a robust and sensorless induction motor drive is by no means completely solved, the results obtained as part of this work point in a promising direction.
|Item Type:||Thesis (Doctoral (PhD))|
|Additional Information:||Bibliography: pages 179-186.|
|Department(s):||Engineering and Applied Science, Faculty of|
|Library of Congress Subject Heading:||Squirrel cage motors; Electric motors--Electronic control; Neural networks (Computer science)|
Actions (login required)