Reliable iron-loss evaluation under noisy conditions: Advantages of Bayesian estimation over conventional time averaging

Accurate evaluation of iron loss in soft magnetic materials is often hindered by low signal-to-noise ratio (SNR) conditions, such as at low frequencies or in materials with low permeability. The conventional averaging method, which corresponds to ensemble averaging of repeated single-cycle waveforms, is effective for noise suppression but is time-consuming and susceptible to errors due to environmental drift, while filtering and frequency-domain denoising techniques improve SNR at the cost of waveform fidelity. We propose a Bayesian inference framework integrated with the two-coil method for physics-oriented analysis, which reconstructs current and voltage waveforms and directly provides amplitudes, phases, and noise levels together with credible intervals. Experimental validation demonstrated that the proposed approach achieves more reliable loss estimation with far fewer signal averages than the conventional methods. The framework provides a robust and efficient solution for magnetic-material characterization and has the potential to be extended beyond sinusoidal excitation to more complex excitation conditions in future studies.