Real-time suppression of stochastic pulse shaped noise for on-site PD measurements

Partial Discharge (PD) diagnosis and monitoring systems are state of the art for performing quality assurance or fault identification of high voltage components. However their on-site application is a rather difficult task because PD information is widely superimposed with electromagnetic disturbances. These disturbances affect very negatively all known PD diagnosis systems. Especially the detection and suppression of stochastic pulse shaped noise is a major problem. To determine such noise a fast machine intelligent signal recognition system (NeuroTEK II) was developed. This system is able to distinguish between PD pulses and noise up to a signal to noise ratio of 1:10. Therefore a measuring system acquires the input data in the VHF range (20..100MHz). The discrimination of the pulses is performed in real time in time domain using fast neural network hardware. With that signal recognition system a noise suppressed Phase Resolved Pulse Sequence (PRPS) data set in the CIGRE data format is generated that can be the input source of the most modern PD evaluation software.