ANALYSIS AND EVALUATION OF DIELECTRIC PARAMETERS FOR DESIGN VERIFICATION AND CALIBRATION OF A NEWLY DEVELOPED DIAGNOSTIC SYSTEM FOR MV POWER CABLES

For the purpose of a condition based asset management in-depth knowledge of the equipment condition is needed. In medium voltage power networks paper-insulated lead covered (PILC) cables are to see as special critical assets. In order to evaluate the most appropriate diagnostic parameters and their characteristic values, multiple PILC cables have been artificially aged at power frequency during a long lasting laboratory aging experiment. Over this process, the relevant parameters like dissipation factor and partial discharge behaviour have been monitored and recorded and a fundamental database was built up. As a next step, field measurements of these values need to follow. Therefore, a mobile measurement system has been furtherly developed and integrated in a cable test van. In order to get improved diagnostic criteria the mobile test system facilitates, that the dielectric properties of the PILC cables are determined at 0.1Hz and 50Hz test voltage frequency. In this paper, the development, the calibration and the adjustment of the measurement system as well as influences on field measurements will be described and discussed.


INTRODUCTION
In European urban medium voltage power networks, there is still a huge amount of PILC cables in operation.In the local network, where this project is executed, this adds up to a total percentage of 56%.These cables were partially installed multiple decades ago and the current asset management strategy is often failure-based.In order to facilitate a condition based replacement strategy, onsite diagnostic measurements need to be carried out.The mobile measurement system, which is necessary for this purpose, has been developed, realized, verified and finally integrated into a cable test van.Based on a previously executed laboratory experiment (Integrated Cable Accelerated Aging System, ICAAS [1]), the cable condition can be estimated and furthermore -founded on additional load assumptions -a lifetime prognosis and further recommendations for maintenance actions can be derived.Within the aging experiment, multiple cables have been artificially aged with thermo-electrical stresses until the instant of failure.Their relevant aging parameters, such as tan and partial discharge behavior, have been monitored and recorded during the whole aging process to a fundamental databank, which serves as a basis for the evaluation and interpretation of the field-measured values [2].

INTEGRATED DIAGNOSTIC SYSTEM
In order to improve the methods and resulting statements of diagnostic measurements and analyses, the commonly used very low frequency tests (VLF) have been extended by the measurement of the dissipation factor also at 50 Hz.The measurement unit is called IDS-1000, for Integrated Diagnostic System at a sample rate of 1000 kHz.There are two different voltage sources inside the cable test van.With the 50Hz source, a compromise between weight respectively physical dimensions and power output had to be found.It consists of a regulating transformer followed by a high voltage transformer, which is fed by the substation and provides a maximum voltage of 14kV at 19kVA.This power output covers measurements at nominal voltage (12kV) of cables of a length up to approximately 1km (at a supposed capacitance per unit length of 0.36µF/km), which is equivalent to 83% of the local power network cable lines.Nevertheless, it is also possible to measure the longer cable lines at lower voltage levels.The 0.1Hz source is an industrial solution from one of the project partners which provides true-sinus © voltages up to 42.5kV, a BAUR viola TD.In the following requirements, challenges and solutions by the development of such a unit are presented and discussed, as well as the calibration of the unit and finally tests and analyses by field measurements.

GENERAL STRUCTURE OF THE MEASUREMENT SYSTEM
In general, the system measures the relevant parameters on high voltage potential directly at the cable inside the switching cell in the substation.Measuring on HVpotential avoids complex recalculations and adjustments concerning e.g. the capacitance of the supply voltage cable.The use of only passive elements instead of converters or amplifiers in the measurement circuit guarantees phase linearity and accuracy.Voltage and current are measured simultaneously on high voltage potential, as to be seen in Fig. 1.Digital signal processing together with a patented algorithm enable the determination of the phase shift between voltage and current resulting in the dissipation factor tan, in a highly accurate way.The FPGA-based measurement unit transfers the data via fibre-optics to the host computer inside the cable test van.For the voltage measurement (VU in Fig. 1) a compensated resistive divider has been developed and set up.This kind of divider ensures a frequency independent transfer function, which is necessary for the two different test voltage frequencies of 0.1Hz and 50Hz.The low voltage side of the divider itself is implemented as another voltage divider in order to enhance the measurable voltage range and also the precision over this whole spread.For the frequency independence the following condition needs to be fulfilled: ...
The used elements in the high-and low voltage parts of the system are from the same model ranges.
The current measurement is realized through a voltage measurement over different selectable precise shunt power resistors.Both of these voltage measurements are realized through a measurement module which provides simultaneous analogue-to-digital converters at a sample rate of 1MS/s.With a maximum voltage input of ±10V and a maximum divider voltage ratio of ~1:6000, voltages up to 42,5kV (RMS) can be measured.
From the input sight, the voltage is measured before the current and therefore needs to be corrected by the current measurement in order to receive the real voltage that is applied to the DUT.Even though the small voltage drop over the shunt resistors seems insignificant compared to the high voltage applied to the input, this correction is of vital importance for the accuracy since voltage-and current-phasors are nearly orthogonal.

IDS-1000 CALIBRATION IN LABORATORY AND ITS VERIFICATION
For the calibration of the measurement system, different reference systems have been used.There are some industrial solutions for tan measurements, which usually need an extra reference capacitor in order to enable the highly accurate measurement.The different types of high voltage capacitors have different (dis-)advantages, which need to be considered for the particular application.
Due to the high dielectric energy density of oil impregnated paper, capacitors with this dielectric material are available in a relative broad range of values, from some pF up to a range of several 10nF, while staying compact and transportable.Gas-filled capacitors fulfil special requirements for the independence of voltage, temperature and frequency.The low losses of the dielectric material lead directly to low values for the dissipation factor.Disadvantages result only from the use of gaseous dielectric, which limits the realizable capacitances [3].
For laboratory measurements, a SF6-filled high voltage capacitor of 57,543pF has been used as reference object.
For the devices under test, different oil capacitors up to a total capacitance of 45nF were available, which is equivalent to approximately 125 meters of PILC cable.Furthermore, as their tan is in the same value range, these capacitors are a suitable simulation substitute for real PILC cables.
As first step, a 50Hz reference system (OMICRON MI 600) has been used to evaluate the exact values of the divider's high voltage side elements (Elements ,  1−60 in Fig. 1).
The losses of one capacitor can be determined from the measured tan, where  is the circular frequency: , 1 tan One element of the high voltage divider consists of a capacitance Cx with its frequency dependent losses RL,C and the parallel resistance Rx as shown in Fig. 2.

Fig. 2: One element of the high voltage divider
The measured series resistance of the whole divider will result in a slightly reduced value for each of its elements compared to the used components (Rx), which is explained through the dielectric losses of the capacitors., , , The indirect proportionality between the losses and the capacitance together with formulas (1) and (2) show that the actual resistance on the low voltage side is also reduced by the same factor.
Paper 0797 CIRED 2017 3/5 These measured and calculated values are passed on to the software, which is able to model the whole measurement circuit digitally.The resulting transfer function is implemented in a way that the divider behavior and its physical reflection is methodically captured and modeled.
With the calibration at two different system frequencies a widely frequency-independent measurement in amplitude and phase within a broad frequency range is realizable.
In the next step, the measured high voltage amplitude has been matched against another reference system.The measured differences are adjusted by a factor for each of the two selectable divider stages, which is multiplied with the voltage phasor.This is calculated from the time signal by complex transformation, frequency neutralization and periodic filtering.Thus, only the amplitude, not the phase (respectively tan), is being scaled.
The complex input impedance of the analogue measurement module is also modelled in the software.The correlation of this parameter was the only necessary phase adjustment for the 50Hz measurement in order to get the same tan values as measured by the reference system for each one of the divider stages.
The adjustment of the 0.1Hz measurements was performed using the 0.1Hz voltage source, which is also able to measure a cable's tan.The laboratory tests showed only a very slight difference concerning the dissipation factor's voltage dependency of the measurement system IDS-1000 and the reference system viola TD.Only a small angle correction needed to be performed on the insensitive stage of the divider's low voltage side.
As test objects, two capacitors of 10nF each were used in parallel.Varying the value of  , in a scope of ±0.1% the two developments could be set match each other in a range of 1*10 -4 .

INFLUENCES ON FIELD MEASUREMENTS
In order to verify the laboratory adjustments, repeated measurements have been performed in field on a cable line that consists of 7 PILC sections, with a total length of 632m.The 50Hz measurement was again verified by the system used in the laboratory -OMICRON MI 600.Due to the heavy weight and physical dimensions of the available gas filled capacitor, one of the oil capacitors (tan and exact capacitance was previously measured in the laboratory) has been used in field as reference capacitance.Within one day and under comparable conditions, three 50Hz measurement cycles were performed, in which all three phases were measured successively with first increasing and afterwards decreasing voltages.Measurements are presented in Fig. 3-5, where the colours represent the phases, the triangles signify rising, the circles decreasing voltages.In the first and the last cycle, the IDS-1000 has been used (Fig. 3 & 5) and in the cycle in between, the OMICRON reference system (Fig 4).It can be seen, that the measured values of the three cycles are nearly in the same value range, also the differences between the single phases are present in all cycles.This indicates that the laboratory calibration was done properly.Stated that the two systems deliver the same measurement values, there is still an unknown dependency which causes the dissipation factor to rise within 3 hours by about 2.2-2.5%.The influence of the measurement current and the resulting temperature rise is to be seen as insignificant, as it was at a maximum of 1.3A.The temperature of the cable is likely to have decreased within this period of time after disconnecting from the mains, which has a strong influence on the tan behaviour as can be seen in Fig. 7. qualitative insight on the tan development of the two measurements, which were performed within a time lag of 1 month, this second measurement was carried out with only increasing voltage steps and only on phase L1.

Fig. 6: Comparison of tan development of measurement and reference system
First of all is striking, that the reference system shows values slightly differing from the IDS-1000.This is likely to result from the used reference capacitor, which is to see as not ideal concerning ambient conditions and their influence on the capacitor's relevant parameters (capacitance, tan).
Assuming that the variances can therefore be neglected, the absolute values are still in another value range compared to the previous measurement (Fig. 3-5).These deviations strengthen the suspicion that the (50Hz-) tan is highly dependent on yet to be determined parameters.

Fig. 7:
Normalized tan over temperature development of differently pre-aged cable samples [4], [5] One parameter strongly influencing the tan measurement is the temperature of the test object (respective ground temperature).The first measurement was performed in the beginning of November, the second in the beginning of December, therefore the latter was carried out under longer-term colder conditions.Fig. 7 shows the tantemperature dependency and also, that selectivity is given in a lower (<20°C) and a higher (>50°C) temperature range.
It is principally not possible to determine the actual cable temperature comparable to the aging experiment under laboratory conditions, where the values in Fig. 7 originate from.The best estimation seems to be the soil temperature in the depth of about 60-80 cm, in which the cables are supposed to be usually laid [6].There is no exact data on the temperature in the particular area where the cable is installed, but the data from a local weather station gives a hint on the development during this time of the year.The available temperature values are measured in a depth of 1m and 0.5m, the calculated value in between is taken as a reference.
Fig. 8: Soil temperature in Nuremberg in a depth of 1m, 0.5m and their average; Data source: Deutscher Wetterdienst DWD -German weather service Fig. 8 shows that the average soil temperature dropped from 10.2 °C to 4.4 °C in the time between the two measurements.The mean values of the November's measurements of L1increasing voltage (blue lines with triangles in Fig. 3-5) compared to the December's measurement with the IDS-1000 (Fig. 6 -orange line) results in a mean tan increase of 39%.
Based on a linear approximation in the lower temperature range of Fig. 7, the steepest tan development (1951) would lead to a tan increase of 29% for the temperature drop from 10.2 °C to 4.4 °C.The previously determined increase of 39% therefore indicates that this cable might not be in a good condition (worse as the one from 1951), caused e.g. by different load profiles, which are not to be determined.Another difficulty is, that the precise cable temperature cannot be determined, as it is strongly dependent on its location, surrounding soil temperature (which is not constant along the cable length), previously applied load and the elapsed time between unloading and measuring.This also compounds the direct interpretation of absolute (field-measured 50Hz-) tan values.One of the 0.1Hz reference field measurements can be seen in Fig. 9.The small deviation of the two developments may be owed to the fact, that for both systems different reference elements were used for their first calibration.The principal accordance in the progression however is clearly given.

CONCLUSION
A new measurement system for the offline diagnosis of medium voltage PILC cables, comprising two different voltage sources, has been developed, realized and integrated in a cable test van.The implementation of the necessary measurement equipment needs to be frequencyindependent.This was realized by a compensated resistive high-voltage divider as well as with the measurement unit itself, being able to model the accurate frequency behavior of all divider components.The system calibration was performed in laboratory with different reference systems and for the two test voltage frequencies.The accuracy under laboratory conditions could thereby be approved.
Influences on field measurements are problematic in the case of the 50Hz-tan, as e.g. the cable temperature cannot be exactly determined, as under laboratory conditions.The 0.1Hz-tan showed a strong dependency from external partial discharges, which could be eliminated by the application of field control elements.

Fig. 1 :
Fig. 1: Simplified schematic of the voltage and current measurement

Fig. 9 :
Fig.9: Comparison of 0.1Hz tan behavior of reference system viola TD and IDS-1000 Fig.10shows two 0.1Hz measurements of the same phase of one cable, whereas one was performed with, and the other without field control (Fig.11).

Fig. 10 :
Fig. 10: 0.1Hz measurements with and without field controlThe higher VLF-tan values for rising voltages are influenced by PD inside the switching cell.The external partial discharges can be visualized with a special corona camera (Fig.11).

Fig. 11 :
Fig. 11: Partial discharges inside switching cell (left side: without field control, right side: with field control) Both images of Fig. 11 were taken at 29kV at 0.1Hz.The left side directly shows the PD-sources (temporarily superimposed by the corona camera).The according tan development is shown in Fig. 10, orange line.The blue line shows the tan development with applied field control.It is clearly visible, that the partial discharges have a strong influence on the tan.This shows that field control elements should always be applied on the switchgears as to be seen in Fig.10, right side, if possible, on both ends of the cable.
without field control 0.1Hz with field control