Some aspects of the statistical modeling of partial discharge inception conditions

Statistical modeling procedures are presented for representing various conditions in different types of samples and equipment at, or near, the onset of partial discharge (PD). Using multiple-regression techniques it was possible to establish trends in PD behavior in terms of the repetition rate. The latter parameter was chosen as being sensitive to changes near inception, and it is measurable by conventional instrumentation (for example a multichannel pulse-height analyzer). The statistical repetition rate models also can be related to the probability of inception. The probabilistic definition derived complies with the simpler criteria often used in partial-discharge testing. Application of the methods is described by statistical models determined for five examples that include resin-insulated and oil-impregnated systems. Computer analysis showed that a high proportion of the variability of discharge behavior near inception could be estimated from the models: in particular, the effect of voltage in combination with increase in time, insulation thickness, and temperature changes. In addition, the models indicated that characteristic PD patterns are identifiable for resin voids and oil wedges. It is concluded that adoption of statistical techniques could result in the specification of PD inception in terms of probability, thus reducing the difficulties in interpreting the results from commercial tests