Effects of Distribution Approximations to Expanded Uncertainty of Breakdown Strength of Mineral Insulating Oil in IEC 60156

Abstract

Breakdown Strength test identifies the health condition of oil-immersed power transformers. The erratic and non-homogeneity behavior affects much the results of breakdown strength test. The inclusion of expanded uncertainty in test results of breakdown strength bounds the compensated test results in a defined confidence level, which is not defined in IEC 60156. This paper presents the analysis of the effects of distribution approximations used in the estimation of expanded uncertainty. The study involved Normal distribution (Gaussian), t-distribution (student distribution), Rectangular distribution and Triangular distribution in estimating an expanded uncertainty which incorporated Type A and Type B errors. The Expanded Uncertainty was based on a standard uncertainty multiplied by a coverage factor of k = 2, which provided a minimum confidence level of 95% in all approximated distributions. Normal distribution (Gaussian) showed the validity in the estimation of expanded uncertainty for a sufficiently large number of test results (of up to 25 tests per sample) while Student-t distribution showed invalidity for all test results. Since it is not a good practice to test insulating oil up to 25 tests, expanded uncertainty based on Normal distribution seemed to be valid for any test results less than 25. The inclusion of expanded uncertainty may predict the validity and avoid re-running of the test.