Minimisation of torque ripple in a switched reluctance motor using a neural network

This paper describes the application of associative memory neural networks to the problem of torque ripple minimisation in a switched reluctance motor (SRM). Torque ripple arises from the failure of simple communication schemes to take account of the non-linear torque production characteristics of the motor phase windings. Initial experiments carried out using a simulation based on actual static torque measurements have been successful in verifying the capability of neural networks to learn the required current profiles. An experiment rig is under construction and the networks used have been implemented using a digital signal processor (DSP). Their speed of operation, including online training has been verified as in excess of that demanded by the application. A field programmable gate array (FPGA) implementation of the networks is under development.