Sensorless control of single switch based switched reluctance motor drive using neural network

Abstract

Neural networks (NNs) have proved to be useful in approximating non-linear systems and in many applications including motion control. Hitherto NNs advocated in switched reluctance motor (SRM) control have a large number of neurons in the hidden layer. This has impeded their real time implementation with DSPs at high speeds because of the high number of operations required by the NN controller and insufficiency of available time between two sampling intervals for computation and control. One of the ideal applications of NNs in SRM control is in rotor position estimation using only SRM current and or voltage signals. Elimination of rotor position sensors is absolutely required for high volume, high speed and low cost applications of SRM, say, in home appliances such as in vacuum cleaners. In this paper, through simulation and analysis, it is derived and demonstrated that a minimal NN configuration is attainable to implement rotor position estimation in SRM drives. The neural network was trained and implemented with an inexpensive DSP microcontroller for performance evaluation. Neural network training data, current i, and flux-linkage /spl lambda/, has been obtained directly from the system during its operation and was verified using finite element analysis (FEA) tools. Further the chosen method is implemented on a single switch converter driven SRM with two phases. This configuration of the motor drive is chosen because it is believed that this is the lowest cost variable speed machine system available. The theoretical results are correlated experimentally with this converter and machine configuration in order to demonstrate the viability of the proposed approach for the development of low cost motor drives.