Partial Discharge Analysis and Recognition in Cables with Medium Voltage DC

With the current state of power electronic conversion operating at Direct Current (DC) can in some cases be more efficient than ordinary Alternating Current (AC). DC voltage energy transmission and distribution is, thus, playing an increasingly important role. The continuity of the electrical energy supply is essential for a well-functioning society. At AC voltage problems with cables can in most cases be prematurely detected with Partial Discharge (PD) measurements. Partial discharges represent defects in the insulation system of electrical cables and devices. They can be categorized in three groups: corona, surface and internal discharges. By knowing the source of the partial discharges and quantifying the discharges, the dielectric condition of a cable can be estimated. At AC voltage the various patterns of partial discharges, also known as fingerprints, can already be recognized and the source of the defect can be determined. The trend for electrical power distribution is, thus, to use DC voltage instead of AC voltage. However, one problem that should be tackled is that partial discharges under DC voltage form a challenge in recognizing the source of the discharge. Therefore, it is still difficult to properly estimate the dielectric condition of the cable based on DC voltage partial discharges. To help overcome this challenge this master thesis research will try to find a relation between measured medium voltage DC partial discharge patterns and the actual source of the discharges. The influence of noise and data segregation of the measurements will be investigated. A method to separate multiple types of partial discharge from a single measurement will also be discussed. Furthermore, limits will be determined in how long a measurement needs to take in order to reliably detect the source of a certain defect. Four artificial PD defects were used to acquire specific data on the individual types of discharges at DC voltage. PD analysis has been done with a self-written Matlab program. By creating classification figures trends have been found that help recognizing the type of defect in a DC PD measurement. The influence of noise and data segregation on DC PD measurements has also been evaluated. Noise can significantly influence the classification figure. Therefore, noise should be prevented or filtered out prior to the classification process. A single noise waveform can already cause a large standard deviation in the classification figures. However, the calculated average values keep displaying the same trend. Thus, a single noise waveform does not hinder the recognition of the PD source. If due to high background noise lower amplitude PDs are lost, data segregation occurs. The physical relationships of time and charge between succeeding discharges will be distorted. To achieve a successful classification a minimum amount of PD waveforms is necessary, which has been investigated. For positive corona at least 250 PDs should be acquired to get the same trends in the classification figure as with 1000 PD waveforms. To be certain of a negative corona discharge, at least 300 PD-pulses should be recorded, instead of the reference measurement, which contained 903. The reference measurement with the internal discharge sample comprises of 578 waveforms. For proper classification a minimum of 300 discharges should be obtained. In order to be certain of surface discharges, 450 out of 999 waveforms are sufficient to create the same trends in the classification figure. To separate and classify individual discharge types from a multiple PD source measurement, cluster techniques are used. The cluster techniques used in this thesis are based on the (Energy versus Charge) and (Equivalent time versus Equivalent bandwidth) of a waveform. It has proven to be possible to classify and identify individual discharge sources. A classification of a certain cluster showed the same trends as in a reference surface discharge measurement. For negative and positive corona clusters 3 out of 4 classification figures comply with the reference measurements.Electrical Sustainable Energy trackElectrical Sustainable EnergyElectrical Engineering, Mathematics and Computer Scienc