Artificial Neural Network-Based PMSM Modeling for the Electric Motor Emulation

Abstract

Lookup table (LUT)-based modeling of electric machines using finite element analysis (FEA) is an accurate technique for high-fidelity real-time emulation of electric motors, however, the cost of a huge computational burden. To overcome this shortcoming and the need for expensive hardware for the motor emulation, an artificial neural network (ANN)-based modeling method is proposed and developed in this paper for a 22-kW permanent magnet synchronous machine (PMSM). The ANN-based machine model is trained using back-propagation algorithms and its weights are optimized for the given arbitrary input(s) to consider the non-linearities in a PMSM, which are supposed to be reflected in the LUTs implemented in the emulation system. A strong correlation with minimal error, after comparing the dq-axis currents and electromagnetic torques extracted from both the LUT-based model and proposed ANN-based model under various loading conditions, confirms the accuracy of the proposed computationally efficient ANN-based modeling method.