Aged oil-paper classification using statistical parameters and clustering analysis

The paper presents an aging experiment on oil-paper insulation under electrical and thermal stresses. A total of 55 specimens were placed in a designed oil tank for the accelerated aging experiment. Partial discharges and degree of polymerization of specimens were measured during the aging experiment. A total of 27 statistical operators were used for recognition of aged oil-paper insulation. There were 9 principle parameters extracted from the 27 statistical operators by factor analysis. The PD parameter vectors composed of either the statistical operators or the principle parameters were clustered by two types of clustering analysis methods, the A-mean and hierarchical clustering. The results showed that the hierarchical clustering had more advantages over the k-mean clustering for aged oil-paper insulation classification. The principle parameters can be used to obtain the clustering correctness ratios the same as the statistical operators, even though the number of principle parameters was much less than statistical operators.