Optimized instantaneous torque control of switched reluctance motor by neural network

The switched reluctance motor (SRM), owing to its doubly salient pole structure and to its operation in the saturation region, has a highly nonlinear torque characteristics. Therefore, a linear model of a SRM can predict its performance characteristics only for a limited range of operation. An accurate and comprehensive nonlinear model of SRM, however, is extremely complicated and is computationally intensive to be implemented in real time for control purposes. To alleviate some of these difficulties, several simplified models are presented in the literature. These models, however, lack accuracy. Artificial neural networks (ANNs) have been used successfully in the control of nonlinear dynamic systems. This paper presents an ANN based torque control scheme of SRM which generates optimal current profiles to minimize torque pulsation. Theoretically, it is found that for each speed and torque below base speed there are several current profiles which produce the desired torque without any pulsation. This method finds the one which gives the maximum torque per ampere. Unlike the other ANN based nonlinear modeling of SRM presented in the literature which uses static magnetization data for training, the training data for the ANN in the proposed method is obtained from the simulation of a dynamic model of the SRM. The proposed control scheme, therefore, controls torque on an instantaneous basis, thus allowing torque control even during the dynamic operation of the motor. In order to include the effect of the nonlinearity, the dynamic model of the SRM uses static magnetization data generated experimentally. Operation of the SRM from zero speed to the base speed is considered. Simulation results of the optimal torque control scheme are presented. To validate the applicability of the proposed technique, the result of the ANNs will is compared with experimentally measured results