The Application of Neural Network Metamodels Interior Permanent Magnet Machine Performance Prediction

Abstract

To increase the computational efficiency of electrical machine optimization and to utilize transfer learning from one metamodel to another, metamodels based on neural networks seem to be a promising solution. This paper presents a methodology of applying neural networks for developing metamodel for the prediction of interior permanent magnet machine performance. Furthermore, it provides procedures and guidelines on design space sampling and developing neural-network-based metamodels to achieve good predicting performance. The proposed approach has been tested on a case of a six-phase 200 kW IPM motor.