Hybrid clustering method for partial discharge diagnosis of large generators

Abstract

On the way of realizing condition-based maintenance for large generators of power system, partial discharge (PD) diagnosis that can predict insulation problems of the stator windings is necessary to implement. While online PD tests have been made for over 40 years, effective diagnostic methods are still under development. In this paper a practical diagnosis method is proposed focusing on small size and incomplete databases that often exist in a factory environment. A novel hybrid clustering method (HCM) is introduced for classification and diagnosis. Experimental PD data of industrial model bars are used and some results are presented to illustrate and validate the diagnosis method. An example is given for applying this method to investigate the TGA (turbo generator analyzer) data provided by a power plant of British Nuclear Fuels Limited. Diagnosis results are included to demonstrate that the new PD measurement can be identified as new PD type or belonging to an existing type in the database. The relationship between new data and the historical data can be visualized and abundant diagnostic information can be provided to users.