Sensorless position estimation for switched reluctance motors using artificial neural networks

This doctoral thesis presents a new approach to the sensorless operation of the Switched Reluctance Motor (SRM). The premise of the method is that an Artificial Neural Network (ANN) forms an efficient mapping structure for the SRM. Through measurement of the phase flux linkages and phase currents the ANN is able to estimate the rotor position. The ANN training data set includes the SRM magnetization characteristic of which flux linkage (lambda) and current (i) as inputs and the corresponding position (theta) as output.; The suggested method does not require any SRM model to construct the relation between the electric and mechanic variables. The measurement of these variables is the only requirement. From this point of view, the method can be classified as model-free rotor position estimation scheme.; The first step in the investigation was the selection of an ANN. Then the preparation of the training data set was studied. After that, an off-line case study was performed. This study showed that the ANN is capable of estimating rotor position by using SRM magnetization data. Before on-line case study, ANN hardware implementation issues and rotor position estimator requirements were worked out.; At the next step, the possibilities of using the ANN-based estimator in the real time operation of the SRM were examined. For this purpose, first the real time operation of the same 20 kW SRM with shaft encoder was simulated by using Simulink(TM) and then ANN-based estimator was integrated into the simulation to see the on-line operation performance. Various operating conditions were tested from zero to full speed with a successful start-up sequence.; The final milestone is to conduct experimental studies. The first task was the collection of the training data while the SRM is running with shaft encoder. The next step was the preprocessing of these data and to train an appropriate ANN. Then the trained ANN was emulated in a Digital Signal Processor (DSP). Finally, real-time data were fed to the DSP-based ANN and stable sensorless operation was achieved for various operating conditions.