Algorithms for Partial Discharge Monitoring of Medium Voltage Cable Plugs, Using a Multi-Sensor Expert System

The occurrence of partial discharge (PD) is a good indicator for the integrity of insulating materials and thus suitable for the condition assessment of cable terminations. PD emits high-frequency electromagnetic signals in the form of stochastic pulse sequences, which can be distinguished from background noise signals by various methods of machine learning. Up to now, common PD detection methods have been based on single-channel measurements with a specific sensor or on a specific measurement method, e.g. the conventional method according to IEC 60270. This makes data acquisition susceptible to interference from unknown noise sources and PD or interference pulses occurring outside the cable termination. To improve diagnostic performance, especially for on-site testing and monitoring, a non-conventional PD measurement system has been developed that uses several sensors for the detection of PD in different frequency ranges. In this paper, different simple and robust classification algorithms are compared with respect to their efficiency to correctly identify PD signals from different PD sources and to locate faulty cable connectors. Furthermore, their ability to discriminate or tolerate external noise is investigated.