In this work, a methodology is presented to assess the thermal aging condition and moisture content of cross-linked polyethylene (XLPE) cable insulation from a single polarization current measurement. For this purpose, polarization current measurements are performed on several specimens of 11 kV XLPE having varied degrees of moisture content and thermal aging. The moisture content is introduced in the aged and unaged specimens through very small artificial holes of 0.5 mm diameter. From the polarization current measurements, various features are extracted using a five-parameter power law-based dielectric response modeling and a modified version of isothermal relaxation current analysis. Following this, an efficient feature selection algorithm analysis of variance (ANOVA) is used to select the best features prior to classification based on thermal aging using three well-known machine learning (ML) classifiers, namely, cubic support vector machine (SVM), extreme gradient boost (XGBoost), and random forest algorithm (RFA). All samples, irrespective of the moisture content, are classified into five classes with an accuracy of 99.5%, 97.5%, and 98%, respectively. After that, a moisture-sensitive parameter is identified that is capable of predicting the moisture content quantitatively in each of the aging classes with satisfactory performance.
