Requirements of automated PD diagnosis systems for fault identification in noisy conditions

This paper evaluates different design methods for powerful partial discharge (PD) diagnostics using an automated personal computer system. A comparison of innovative and conventional analysis tools for PD diagnosis is presented. With four defined performance features the quality of eight efficient PD fingerprint extraction methods are investigated. Furthermore, the same evaluation methods are applied to determine the diagnostical potential of neural network, fuzzy and distance classification algorithms. Diagnostic decisions under noisy conditions are discussed using fifteen different defects inside SF₆ and air insulated equipment. It is indicated that an advanced PD diagnosis can be performed independent of the test setup. The results show that the potential of every pattern recognition in PD diagnosis is influenced predominantly by the quality of the PD fingerprint. The choice of the classification method in most cases only influences the PD diagnosis system characterization