This thesis investigates intelligent partial discharge (PD) diagnosis for the condition monitoring of turbogenerators. The research focuses on graphical classification methods for PD activities, particularly addressing challenges with small and incomplete PD databases. It begins with an exploration of feature extraction techniques for various PD patterns and automated pattern recognition using conventional classifiers and neural networks. Laboratory tests were conducted to generate a PD database of typical types observed on industrial model bars, leading to the presentation of a Hybrid Clustering Method (HCM) and an advanced Self-Organizing Map (SOM) for visualizing the relationships between new and historical PD samples.

The research presented in this thesis confirms the utility of graphical classification methods for diagnosing partial discharge in turbogenerators. These methods, whether used independently or in conjunction with other techniques, are shown to provide reliable diagnostic information. The work contributes to the field of condition monitoring by offering advanced approaches for analyzing PD data, particularly in scenarios with limited datasets. The development of the Hybrid Clustering Method and the advanced Self-Organizing Map provides novel tools for visualizing and understanding PD patterns.