Decision tree rules for insulation condition assessment of pre-molded power cable joints with artificial defects

Abstract

A decision tree method is presented that can judge the initial and final stages of insulation degradation for pre-molded power cable joints with air gap and void defects. According to the partial discharge (PD) of eight experimental subjects, it is validated that the discharge phase changes toward the voltage zero-crossing area and becomes wider. This well-known phenomenon can be applied to assess the transition of insulation status by the cluster theory. Subsequently, the Gini index coefficient and information gain of the decision tree algorithm yield the insulation status diagnostic rules. The result indicates an overall recognition rate of 85% in training data. Furthermore, the proposed diagnostic method has successfully detected the alert of the final stage prior to insulation breakdown for two additional experimental subjects. These decision tree rules are feasible and economical for an online PD monitoring device to achieve condition-based maintenance in distribution power cables.