An Enhanced Dataset for Axial Flux Machine Performance Study

Axial Flux Machines (AFMs) are gaining attention in modern electric vehicle applications due to their high power density, compactness, and material efficiency. The performance of AFMs can be analyzed using either high-fidelity (HF) three-dimensional (3D) Finite Element Method (FEM) or low-fidelity (LF) quasi-3D FEM, depending on the specific research requirements. This paper presents comparative numerical studies to assess the gap between different HF and LF simulation models. Then, a comprehensive dataset is generated with hybrid HF and LF FEM simulations for surrogate modelling. This dataset enables the training, validation, and testing of surrogate models with minimal computational time and resources, thereby accelerating the performance evaluation process for AFMs. The numerical results of a baseline with a Multilayer Perceptron (MLP) model demonstrate that it accurately captures the relationship between design variables and key performance indicators (KPIs), which can be integrated into functional procedures to speed the analysis processing bypass the invoking of FEM solver.