Partial discharge recognition based on pulse waveform using time domain data compression method

A digital integrated system is developed to record partial discharge (PD) pulses and to recognize their patterns by waveforms. Special considerations on the experiment circuit are given to electromagnetic shielding, system bandwidth and pattern variety. Both typical laboratory models and simulation models of stator windings are tested, which represent corresponding PD types in air, transformer oil, artificial cavity of stator bar insulation and along the surfaces of the end-winding. The waveforms of discharge current of distinct models are recorded and their respective features are extracted by segmented time domain data compression method. An artificial neural network is applied to recognize different patterns. Several factors concerning recognition rate are discussed. The training and testing result shows the potentiality of applying waveform analysis in partial discharge recognition.