Statistical Analysis of Impulsive Flashover Voltages Across Solid-Air Interfaces

Abstract

Understanding of the influence of experimental conditions on the breakdown voltage of composite insulation is important to facilitate optimal design of high voltage systems. This includes ensuring that the statistical analysis performed on breakdown voltage data is relevant, in providing extra information on the failure voltages. Therefore, in order to investigate the applicability of statistical techniques in aiding to elucidate further detail about the breakdown process, multiple statistical methods were applied and analysed, in order to find the most suitable for the data found during a specific set of breakdown tests. Normal, lognormal, 2-parameter Weibull and 3-parameter Weibull fittings are discussed herein, as applied to the authors' experimental data on the flashover voltages across solid-air interfaces, subjected to 100/700 ns impulse voltages. Negative impulse voltages were applied to composite insulation, consisting of one of three different solid materials in air - Delrin (Polyoxymethylene), Ultem (Polyetherimide) and HDPE (High Density Polyethylene). Samples of each material were machined to a smooth finish. The environmental conditions used in this paper were a fixed air pressure of −0.5 bar gauge, and relative humidity (RH) levels of <10% RH, ~50% RH and >90% RH. The data points used in the statistical analyses were from the average flashover voltages found using the ASTM D3426-97 ‘step up’ testing procedure. Fitting the normal, log-normal, 2-parameter Weibull and 3-paramter Weibull distributions to breakdown voltage data obtained for each set of test conditions allowed for the relative quality of fit of each to be directly compared. The Kolmogorov-Smirnov (K-S) test was deployed in order to compare the maximum distance between the experimental data and the theoretical cumulative distribution function, thus verifying the most accurate method of statistical analysis for a given dataset. It was found that the overall best fit to the experimental data was given by the 3-parameter Weibull distribution.