Assessment of Aging Condition for Oil-Immersed Nomex Insulation Paper Based on Frequency-Domain Dielectric Response

To assess the insulation condition of oilimpregnated Nomex paper within transformers during thermal aging, this study proposes an assessment method incorporating the thermal aging effect of oil-impregnated insulation paper. The approach is based on the non-destructive testing technique of frequency-domain dielectric spectroscopy (FDS) combined with a random forest model. Accelerated thermal aging tests were conducted to prepare Nomex paper samples with varying aging states. The changes in their FDS characteristics were investigated, and characteristic parameters reflecting the insulation aging state were extracted. The results show that the dielectric loss factor significantly increases within the frequency range of $\left[10^{-3}, 10^{0}\right] \mathrm{Hz}$ as aging progresses, indicating a close relationship between the degree of material aging and FDS. Furthermore, by fitting the characteristic parameter curves, the dataset was expanded. Employing k-fold cross-validation for model training and testing yielded an assessment result with an average accuracy of $96.5 \%$, alongside average precision, average recall, and F1-scores of $97.4 \%, 96.6 \%$, and $96.8 \%$, respectively. These results demonstrate that the constructed model is efficient and reliable for assessing the insulation condition of oil-paper insulation in power transformers, providing a significant technological foundation for their safe operation.