Analysis of breakdown process at U50 voltage for plane rod discharges by means of Neural Networks

The influence of many parameters on the electric discharges at U50% can only be correlated by means of an analysis of a large amount of experimental data. The 50% probability of breakdown is considered a random process. We have investigated how to find relevant statistical differences leading to breakdown or withstanding during an impulse voltage test under the same conditions for a plane high voltage electrode and a grounded electrode configuration. An analysis based on Neural Network discrimination has been developed which indicates that measurements do contain some relevant information test at early stages in a reduced time frame.