Electrical machine design faces growing constraints from power density, wide operating ranges, thermal and mechanical limits, acoustics, manufacturability, cost, and critical material availability. While finite element and multi-physics simulations remain essential, their direct use in population-based or multi-objective optimization is often computationally prohibitive.
[...] Read more.
Electrical machine design faces growing constraints from power density, wide operating ranges, thermal and mechanical limits, acoustics, manufacturability, cost, and critical material availability. While finite element and multi-physics simulations remain essential, their direct use in population-based or multi-objective optimization is often computationally prohibitive. AI, spanning surrogate modeling, machine learning, deep learning, physics-informed networks, Bayesian optimization, and emerging generative methods, is increasingly used to accelerate analysis, enlarge design spaces, and support inverse or multi-objective tasks. This review examines AI-assisted electrical machine design from a workflow perspective, distinguishing functional approximation, performance prediction, topology-aware learning, physics-informed modeling, active learning, and robust optimization under uncertainty. It highlights current limitations, including narrow topology coverage, reliance on FEM-generated data, weak extrapolation, limited uncertainty reporting, scarce experimental validation, and insufficient attention to manufacturability and sustainability. A design-readiness framework and minimum reporting checklist are proposed to improve trustworthiness and reusability. The review concludes that AI should serve as a physics-aware, validation-dependent accelerator, complementing, not replacing, electromagnetic expertise, multi-physics simulation, and prototype testing.
Full article
