Automated PRPD Pattern Analysis using Image Recognition.

This contribution describes an evaluation method for phase resolved patterns of partial discharges over time. Therefore, their specific shape characterizes patterns in question. The method searches for these characteristics within PRPD patterns by comparing characterization and PRPD using 2-dimensional cross-correlation. Crosscorrelation represents a measure for the similarity between different PRPD patterns. A use case determines the usability of the method in practice. The result is the time dependent occurrence of specific patterns, respectively specific PD sources. Long-term monitoring data of an aged generator step-up transformer with several PD sources of varying activity is used as PD dataset. PDs are recorded by the use of the electro magnetic PD measurement method, also known as the UHF-method. [ABSTRACT FROM AUTHOR]