Multipolarization Temperature–Frequency Normalization Algorithm for High-Voltage Generator Stator Epoxy–Mica Insulation

Frequency domain spectroscopy (FDS) is widely applied in the insulation condition assessment of high-voltage generator stators. The evaluation process requires frequency–temperature normalization, which is mostly conducted using the “master curve” method. However, the existence of multiple polarization processes inside the dielectric leads to significant frequency shift errors, making it difficult to accurately characterize the insulation aging state. To solve this problem, this paper constructs a multi-polarization decoupling model and adopts combined frequency domain spectroscopy (FDS) and thermally stimulated depolarization current (TSDC) testing to identify the polarization types in the main insulation. The Kramers–Kronig transformation is used to separate the infinite-frequency capacitance and conductivity components to obtain the relaxation polarization component. An improved Havriliak–Negami model coupled with the Non-dominated Sorting Genetic Algorithm III (NSGA-III) is applied to decouple multiple relaxation processes and analyze the temperature-dependent evolution of each polarization component. The energy barrier levels and shift factors of each polarization process are extracted respectively to obtain an accurate frequency–temperature normalization algorithm. Experimental results show that the curve corrected by the multi-polarization normalization method has good consistency with the reference temperature curve, avoiding the errors caused by the one-size-fits-all frequency shift of the “master curve” method.