A neural network based surrogate model for predicting noise in synchronous reluctance motors

This paper proposes a neural network (NN) based noise prediction model for electric machines, applied to the case of synchronous reluctance motors (SynRMs). The natural frequencies of various vibration modes for a SynRM stator with generalized tooth geometry and slot numbers have been obtained using structural FEA based computations and then used to build a NN based surrogate model. The accuracy of the surrogate model has been tested and applied to predict the noise level in SynRMs. Also, varying trends in the noise levels for single-barrier SynRMs have been analyzed as a function of the rotor's flux carrier and barrier widths using the natural frequency prediction model.