Abstract
This paper presents a new method for analysing creeping discharges based on information theory as it applies to medical imaging. The analysis of information surface data is used to determine the impact of relaxation time on the characteristic parameters of creeping discharges. The same information is used to make a comparative study of the morphology of discharges propagating in palm kernel oil methyl ester (PKOME) and in mineral oil (MO). Other comparative methods based on fractal analysis and normality hypothesis tests associated with Anderson Darling (AD), Kolmogorov-Smirnoff (KS) and Shapiro-Wilk (SW) statistics are used. The results show that very short relaxation times increase the error on the measurement of the fractal dimension and the maximum extension of the discharges. A growth of the mutual information between 0 and 60% is observed for relaxation times varying between 60s and 420s respectively. For the same time interval, the P-value increases from 0.027 to 0.821 according to the AD statistic, from 0.01 to more than 0.150 according to KS and from 0.083 to more than 0.1 according to SW. This result indicates that the data are from a normal distribution. After 420s of relaxation, the error on the maximum extension measurement is reduced by 94% in PKOME and 92% in MO. Similarly, the error on the mean fractal dimension in MO is reduced by 86.7% for a relaxation time between 301s and 420s, and by 84.6% in PKOME for a time between 180s and 420s. These different results imply that the impact of the discharge can be predicted when it is in its initial phase during which the number of discharge occurrences is reduced. On the other hand, the physicochemical characteristics of the insulating liquid used dictate the relaxation time to be allowed for the laboratory measurements.