Verfahren zur automatisierten Teilentladungsdiagnostik von Energiekabeln (English shop)

Preview

Table of Contents, Datei (68 KB)
Extract, Datei (430 KB)

This work presents methods suitable for the automated processing and evaluation of partial discharge measurement data on power cable systems. This makes it possible to achieve improved results in both offline and online cable PD diagnostics and ultimately to save costs, since the methods make the operator’s task easier.

One focus of this work was placed on alternative methods for determining the arrival time of pulses, which is required for location by means of time domain reflectometry (TDR). Due to dispersion on the cable, differences in pulse rise times occur that can lead to systematic location errors if only the pulse maxima are used for location (peak-peak method). A further problem in offline diagnostics is the dead length. Here, pulse superpositions occur near the cable ends, which complicates both accurate charge determination and location. Here, too, improvements can be achieved by means of special algorithms. Methods were investigated that output the front or the foot point of pulses (EC, AIC, SM), that locate via changes in phase (PM), or that carry out a pulse spacing measurement by means of autocorrelation (AK). Not all methods are equally well suited for this purpose. However, there are indeed robust alternatives to the peak-peak method that can be selected depending on the application.

Particularly in PD measurements on installed cable systems under on-site conditions, strong interference signal couplings occur that differ considerably depending on the measurement environment. In offline diagnostics, in which the location is ultimately determined via a pulse reflection at the open-circuited cable end, the additional difficulty arises that large dynamic differences occur between the individual pulses to be evaluated. As a consequence, in addition to the coupled interference, quantisation and amplifier noise must be expected in the range of the pulses to be identified and located. The goal of reducing such interfering signal components in the measurement data was realised through two approaches. On the one hand, improvements were made to the measurement hardware by applying shielding and mains filtering at the high-voltage coupling capacitor. Great importance was attached here to practical applicability (mobility and flexibility), since local conditions vary considerably. Experience for this was gathered in the course of participating in measurements on cable systems in distribution networks. On the other hand, noise and interference levels in the measurement signal were considerably reduced by suitable signal processing methods. Both widely used methods based on the wavelet transform (DWT, SGWT) and a novel method, empirical mode decomposition (EMD), which was applied to cable PD measurement data for the first time in this work, were employed. Through appropriate use of these methods it is possible, depending on the interference environment, to remove components of the interference signals to such an extent that a useful signal can be recognised as such and PD location can be carried out.

A further focus of the work was on providing a cable PD model that makes it possible to model offline cable PD reflectograms. These can in turn serve either for the location or for the identification of measured reflectograms. Using such a model, cable systems consisting of several cable segments of different types or containing branches can also be modelled. Here, additional reflections can occur in the measurement signal that make conventional TDR location very difficult. A basic prerequisite for such a model is data on the geometric structure of the cables and materials that is as accurate as possible. In this work the model was verified on an XLPE cable system in the laboratory. Taking into account the transfer behaviour and the input impedance of the measurement setup, a very high degree of congruence between the modelled and the measured reflectogram was achieved. Model-based location was carried out by means of a correlation method and an artificial cable PD defect was located with a very small location error.

Various methods were compared with one another for determining the charge of cable PD pulses. It becomes apparent that the conventional methods investigated – integration in the time domain, quasi-integration and charge determination by means of pulse amplitude – in some cases yield very different values. The indicated charge depends both on the selected data window and on the origin of the PD, since effects such as attenuation, losses and partial reflections at joints can influence the respective calculation. These problems were solved by a dynamic adaptation of the integration windows as well as by an attenuation-compensated charge determination. An improved charge determination is thus possible.

The methods presented were implemented in a MATLAB toolbox and successfully applied both to the evaluation of PD signals from measurements on cable systems in the laboratory and on selected PD-affected cable systems installed in medium-voltage networks.