Research Of Oil-Paper Aging Based On Multivariate Statistical Analysis Of PD Characteristic Parameters

As the main insulation used in oil-immersed transformer, Oil-paper suffers from thermal and electrical aging in the long-term running, which deteriorate the insulating performance and threaten the reliability of transformer directly. Aging research work is of vital importance in insulation subject, which contributes to the insulation diagnosis and lifetime prediction. Especially, with the application of computer technology, insulation diagnosis is experiencing rapid development. There are many advanced, available and reliable systems are built to identify insulation material's aging conditions, which ensure the security service of high voltage equipments.An electrical-thermal aging equipment is designed in this paper, in which lifetime tests of oil-paper samples under different conditions are carried out. Using multiple regression method, the lifetime parameters of Ramu and Fallou models are calculated. By comparing the errors of these two models, it is concluded that the latter one is more suitable to predict the lifetime of oil-paper samples in our test condition.The multivariate statistical method is introduced in this paper to diagnose insulation performance and identify its aging stages. Based on the lifetime model obtained by life tests, the reasonable test condition is chosen to age several oil-paper samples for about 240 hours. Acquisition of partial discharge signal is carried out in the constant interval during the whole aging period. A visual interface designed by Matlab is used to distract characteristic parameters from PD signal, and the results are conserved in a finger database designed by Access. There are total 31 parameters are extracted and 11 are find to be closely correlated to aging stages, which are used as aging characteristic parameters in the followed analysis.In order to identify the proper aging condition, cluster method is used to analyze 30 vectors distracted from 6 samples at 5 different aging stages. It is shown that different samples have the similar performance at the same aging stage. The 30 vectors are used to obtain discriminant functions. The characteristic vector with eleven-dimension is mapped to two-dimension and the other 30 vectors are analyzed using discriminant functions. It is concluded that a reasonable result can be achieved using the discriminant functions.