Electrical Insulation Diagnosing using a New Statistical Classification Method

Signal patterns emitted from the electrical insulation apparatuses can easily be detected using some sensors. However, it would be difficult to judge that a detected signal pattern corresponds to which phenomenon such as a severe fault, an abnormal condition with no fault, or a simple harmless noise, because of the issue of the inverse problem. The statistical classification methods can classify the signal patterns clearly into the objective classes with the supervised training data collected in the laboratory with known defects or noise patterns. The important issue in such a classification is, first how we find the effective feature extraction from the signal patterns, and second how we select the efficient classification tools. In this paper, we report the use of the generalized normal distribution function for the feature extraction, and the use of the decision tree method for classification algorithm. The method proposed here is applied to some real data case, and the classification result is compared with that using other features such as the simple moments