Commutating switched reluctance motors efficiently via CMAC neural network with learning rate function

This paper presents a novel control scheme for torque ripple reduction in switched reluctance motors (SRMs) operating at low constant speeds. The control scheme is implemented using a CMAC neural network, trained with a modified LMS algorithm. This algorithm uses a varying learning rate function (LRF) which is defined as a function of the rotor angle of the motor under control. Experimental measurements of the static torque production of a 4 kW, four-phase SRM form the basis of simulation studies of this approach. The simulation results demonstrate that a learned current profile, capable of minimizing torque ripple while having high power efficiency, can be obtained by selecting a LRF with suitable turn-on and turn-off angles during the training of the network.