Partial discharge pattern recognition algorithm for turbo generator monitoring

Abstract

This contribution proposes an algorithm for automated partial discharge (PD) pattern recognition for synchronous turbo generators. PD patterns indicate faults in the mica insulation of the stator windings of high voltage synchronous turbo generators. Usually these patterns can only be interpreted by experienced test engineers. The presented diagnostic routine is focused on insulation faults of turbo generators. It is possible to detect slot discharge, inner corona grading faults, outside corona grading faults, surface discharges in the end winding, sparks, and excitation signals, which are not stator faults yet, but appear in PD patterns. PD measuring instruments working with phase resolved partial discharge pattern diagnoses are delivering 3D-patterns and 2D-patterns. The 2D-patterns are evaluated by a diagnostic tool based on a Matlab toolbox. The known reference patterns represent typical shapes which are caused by typical stator faults. The diagnosis is based on the comparison of reference patterns with measurement results. The proposed algorithm is designed to interpret 2D-images quickly (less than 3 minutes on a 3 GHz personal computer) and with a high hit rate. Significant effort has been made to keep the number of reference patterns limited.