A Comparative Study of Meta-Modeling Approaches for IPMSM Performance Prediction with Neural Networks

Abstract

Improving an Interior Permanent Magnet Synchronous Machine (IPMSM) performance requires advanced approaches, often leading to an optimization problem. A Finite Element Method (FEM)-based numerical evaluation is needed to evaluate the accurate thermally stable output power of the IPMSM, which is computation-expensive. This paper presents and compares three computationally efficient meta-modeling approaches as alternatives to traditional FEM-based evaluation. All three approaches leverage Artificial Neural Networks (ANNs), with the highest accuracy achieved by using three deep ANNs for flux linkage and torque evaluation. These are combined with analytical models to estimate the output power. A multi-objective optimization scenario was conducted to demonstrate the advantages and limitations of the best-performing meta-modeling approach.

Authors 3

1. Mitja Garmut

   University of Maribor

   Affiliation as printed

   FERI, University of Maribor,Institute of Electrical Power Engineering,Maribor,Slovenia

2. Simon Steentjes Aachen Institute of Electrical Machines (IEM)

   RWTH Aachen University

   Affiliation as printed

   RWTH Aachen University,Institute of Electrical Machines,Aachen,Germany

3. Martin Petrun

   University of Maribor

   Affiliation as printed

   FERI, University of Maribor,Institute of Electrical Power Engineering,Maribor,Slovenia