A study on partial discharges pattern recognition under DC voltage through clustering algorithms and cross-correlation filter

It is widely accepted that the presence of defects within insulation systems and the consequent triggering of Partial Discharges (PD) leads to the gradual deterioration of the components up to its failure. Monitoring these phenomena is a widely used strategy to control health and integrity of an insulation system. The classification of the type of PD detected in the component under test is necessary to recognize the defect where the phenomenon is generated. In AC application, a widely used tool for this purpose is the Phase Resolved Partial Discharge (PRPD) pattern. However, the same approach cannot be used in HVDC systems, because the phase reference is missing under DC voltage. Furthermore, an equally powerful technique for the DC case has not been developed yet, although several proposals are present in the literature. The aim of this paper is to present a study for performing PD patterns recognition and noise separation, under DC voltage. The presented work is based on the comparison between a clustering algorithm and a cross-correlation filter applied to the Time-Frequency Map (TF Map), proposed by other researchers. The results show that it is possible to distinguish noise from discharges and evaluate their behavior throughout the measurement phase.