A Review of Core-Loss Modeling in Electrical Machines Under Complex and Rotational Magnetization Conditions

Abstract

Accurate core-loss prediction in electrical machines is difficult when local magnetic flux follows elliptical, circular, irregular, harmonic-rich, or three-dimensional trajectories. This structured review examines empirical and loss-separation formulations, vector and hysteresis models, flux-locus methods, finite-element (FE)-assisted workflows, analytical and magnetic-network field solvers, reduced-order control models, and machine learning (ML) extensions. The literature is compared using common engineering criteria: required material data and parameters, applicable excitation and material conditions, reported error, computational burden, rotational-field capability, and suitability for machine-level design or control. The evidence shows that scalar alternating-field models remain useful for rapid estimation, but their reliability deteriorates under strong rotationality, minor loops, DC bias, saturation, and pulse-width-modulated excitation. Vector and hysteresis models offer stronger physical fidelity but require richer data and substantially greater calibration and computational effort. Recent reduced-order and data-driven approaches can accelerate FE-level prediction; however, their validity is limited to the training or calibration domain. The review therefore recommends a layered framework in which vector-field extraction and flux-locus classification select an appropriate physics-based baseline, while ML is used for residual correction, surrogate prediction, and uncertainty estimation. Priority research needs include standardized multidimensional datasets, anisotropic and grain-oriented material models, transparent quantitative benchmarks, and loss-aware models suitable for optimization and control.