A Novel Approach for Development of Neural Network based Electrical Machine Models for HEV System-level Design Optimization

Abstract

To find the optimal system-level design of hybrid electric vehicles (HEVs), component models are used in simulations to evaluate a large number of different designs within a high dimensional design space. As the electrical machine (EM) represents a key component of the HEV powertrain in terms of energy consumption, models require scalability and sufficient accuracy with manageable computational effort. This paper presents a novel approach for the development of scalable EM models based on Neural Networks (NN). The models are trained with data derived by a Finite Element Analysis (FEA) based scaling procedure and capable to represent the characteristics of a wide range of EM designs without the incorporation of further details. Once a model is trained, it can be directly used in system-level design optimization. The practicality of the model is proven within an exemplary simulation study and its goodness of fit to the training data is validated by a statistical analysis. This approach can help to reduce the computational effort of EM efficiency maps calculation, since only a small number of time-consuming FEA based scaling simulations must be performed prior to the optimization.