Partial Discharges under HVDC Conditions

1

Partial Discharges under HVDC Conditions

1M Judd*, 2W H Siew, 2X Hu, 2E Corr, 2M Zhu, 3A Reid, 4O El Mountassir, 5M Urizarbarrena Cristobal, 6R Giussani and 6M Seltzer-Grant

1High Frequency Diagnostics and Engineering Ltd, Glasgow, UK 2University of Strathclyde, Glasgow, UK

3Glasgow Caledonian University, Glasgow, UK 4Offshore Renewable Energy Catapult, Glasgow, UK

5SP Energy Networks, UK 6HVPD Ltd, Manchester, UK

*Corresponding author: [email protected]

Abstract: This paper describes partial discharge (PD) phenomena in HVDC insulation, including diagnostic methods that are being developed. PD detection and analysis is used to assess the condition of electrical insulation in many types of HV equipment during manufacture, acceptance testing and condition assessment of equipment in the field. In HVAC systems, the phase of the power frequency cycle has a dominant influence on PD activity. Consequently, phase-resolved PD patterns form the cornerstone of many diagnostic techniques for HVAC insulation. However, this approach cannot be used for HVDC systems. The increasing use of HVDC in recent years has led to a growing interest in how PD diagnosis techniques developed for AC systems can be adapted for use with HVDC equipment. Under the influence of an electrostatic (DC) field, the flow of PD current is unipolar, which means that mobile charges in the insulation are subject to forces that cause them to migrate from one conductor to the other, with conductivity of the insulation playing a significant role. This can lead to the accumulation of localised regions of trapped charge. While PD pulses tend to occur much less frequently under HVDC conditions than under comparable HVAC excitation, this trapped charge can cause bursts of discharge activity when the steady-state equilibrium is disturbed, most notably during energising and de-energising. Measurements are presented to illustrate PD behaviour in various test objects subjected to HVAC, HVDC and composite (HVDC plus modulation) voltages. Additional tests on defective 33 kV cables are described, which include DC ramp tests, polarity reversal and soak testing at DC overpotentials. Results results demonstrate how PD activity can be influenced by comparatively minor features of the applied HVDC voltage, such as those associated with ripple from converter switching. Finally, PD data acquisition and analysis techniques are discussed with a view to optimising the application of PD detection under HVDC conditions.

1. Introduction The authors of this paper have been working in collaboration for the last two years on a project aimed at developing on-line partial discharge (PD) monitoring techniques for HVDC cables and accessories. The motivation for this work at the outset was the need to reduce the cost of offshore renewable generation. The project set out to develop technology that could contribute to cost reduction by improving the availability of HVDC transmission links through condition monitoring, which would also help to reduce the cost of insuring these key assets. While the outlook for large-scale UK offshore generation has become less certain, PD monitoring of HVDC equipment remains an important goal as DC systems become more pervasive in various MV applications and in HVDC transmission links. This paper reviews why and how PD activity under HVDC excitation differs from the better-understood phenomena that occur in HVAC equipment. While cables are used as a focus for the paper, many of the underlying principles are applicable to other types of HVDC equipment.

2

To date, the UK has installed some 5 GW of offshore wind generation capacity in wind farms that are moderately close to the coast and are connected to the grid using HVAC links. However, some of the best wind resources can be found more than 100 km offshore in locations that are favourable for the implementation of large wind farms capable of generating power capacities in the region of 1 GW. Wind farms on this scale located so far offshore would need HVDC transmission connections, which could be implemented using modern voltage source converter (VSC) technology [1]. At present, only a limited number of offshore wind farms in Germany are using HVDC transmission systems, but further connections are in the planning process or are due to be commissioned all around Europe [2]. Early HVDC projects and schemes such as BorWin1 have experienced a number of technical issues, leading to significant delays and additional costs that have affected investor confidence [3, 4]. Key issues in HVDC transmission systems are mostly associated with the grid connection so it is imperative that the transmission system is reliable. HVDC export cables are critical components of transmission systems for remote locations such as islands and wind farms and usually represent the largest element of capital cost [5]. According to industry reports, cables are the root cause of the most frequent and largest insurance claims relating to offshore wind developments [6]. For example, when a marine cable fails, it is necessary to initially locate the fault and then raise the cable from the seabed in order to carry out repair. These operations are extremely costly [7] and logistically difficult, requiring expert intervention to accurately pinpoint the location of the fault then deploying a cable lifting vessel to make the repair. HVDC cables are thoroughly tested and verified in accordance with international recommendations [8, 9] during the delivery and commissioning process. While CIGRE studies [10] highlight the high reliability of the internal design of HVDC subsea transmission cable systems, some failures have still occurred. Additionally, since extruded HVDC subsea cables are a relatively new transmission technology, there is limited field experience regarding how cables age electrically under HVDC voltage stresses and a lack of data regarding their long-term reliability. While the predominant cause of cable issues so far has been attributed to damage during installation, the complexities of ageing and space charge phenomena for HVDC cables (discussed in Section 2 below) suggest that continuous insulation monitoring techniques have the potential to play a useful diagnostic role.

2. HVDC Insulation There is no fundamental difference in structure between AC and DC high voltage cables apart from the formulation of the polymeric insulation, which for DC may include inorganic fillers and the use of other proprietary measures to suppress the formation of space charge [11]. Electric fields in AC cables are predominantly graded by capacitive effects on the field distribution, being controlled by the design of conductor shape and by incorporating materials of different dielectric constant to avoid excessive concentrations of electric stress, as shown in Fig. 1. In AC insulation, reversal of the electric field vector occurs so frequently that conduction via low-mobility electrical charges is usually insignificant. However, for HVDC insulation, the presence of a constant electric field across polymeric insulation causes the migration of charges through the insulation (conduction). Some of this charge becomes trapped in the insulation (space charge) in a manner that modifies the electric field pattern considerably from that expected under AC conditions, where we tend to regard the insulation as ‘charge-free.’ To illustrate this point, consider the measurements presented by Yamanaka et. al. [13], which are illustrated in Fig. 2. This is one of many studies that have shown how the electric field in an HVDC cable changes over time (usually hours) in response to both the applied voltage and temperature (i.e., load current), leading to concentrations of electric stress at unexpected locations in the cable insulation.

3

Fig. 1 Electric field distribution at a cable end using stress control materials is illustrated here by the pattern of

equipotential lines that are graded smoothly between the coaxial cable region at the right and the open HV conductor at the left (after [12]).

Fig. 2 Electric field distribution in the insulation of a 500 kV HVDC cable (after [13]) showing the change after 3

hours caused by the evolution of space charge effects. Note how the location of maximum electric field has migrated from the inner conductor boundary towards a position much closer to the cable screen conductor.

The term space charge, which occurs so frequently in research into polymeric HVDC cables (see chapters 4 & 5 in [11], for example), originates in the historical context of vacuum diodes where it described the electronic charge that accumulates in the space between anode and cathode as a result of electron emission from the cathode [14]. Space charge may also accumulate in a solid dielectric in the presence of a DC field. As it does so, this charge modifies and distorts the electric field pattern. If the space charge density becomes sufficiently high, the local electric field strength can exceed the dielectric strength of the insulation, leading to failure [14]. This is particularly the case following a reversal in polarity of the externally applied voltage because under these circumstances, space charge distributions that were generating a local field in opposition to the external field (thereby diminishing it) will be oriented so as to enhance the externally-generated field following polarity reversal. In recent years, much R&D effort has been invested in the development of HVDC cable technologies that can minimise the accumulation of space charge in the insulation (see Ch. 5 in [11]). Approaches include structural modification of the insulation polymer to reduce its propensity for trapping charges and further reducing the conductivity of the insulation by maintaining extreme cleanliness and material purity during manufacture [15].

4

Cable accessories such as joints and terminations have long been recognised as more likely locations for insulation defects that could give rise to PD in AC networks since they disrupt the uniform cable structure and could allow ingress of contaminants if damaged. Furthermore, there is potential for human error during their assembly, a task that must be performed with great care. Likewise, HVDC cable accessories will face similar challenges due to their complexity, which again arises from the need to control the field grading to avoid excessive concentrations of local electric stress. HVDC accessories may employ nonlinear resistive field grading materials based on ZnO (or sometimes SiC) filler compounds to produce a field-dependent conductivity so that conductivity increases at higher electric fields in order to reduce stress (see section 5.7.3 in [11]). These materials reversibly change their electrical properties from being highly insulating to highly conducting in regions where the field exceeds a critical value [16].

3. Partial Discharges in DC Insulation PD is well-known to be an indicator of defective insulation and a cause of insulation ageing and degradation. On-line PD monitoring of HVAC equipment is widely used as a tool for predictive maintenance [17]. Historically, PD under DC conditions has received less attention for the following reasons [18]:

For main years, HVDC was mainly used for non-energy applications (e.g. television, radar, X-ray equipment, etc.) where the failure of an individual component did not affect large numbers of people.

PD under DC is a less directly associated with causing insulation damage (i.e., it is an indicator of weakness rather than a primary cause of degradation in itself).

One main difference in PD phenomena under DC conditions is that current is only flowing in one direction, which means that the PD are usually all of the same polarity. Another difference is that the build-up of charges either in the bulk material, on surfaces, or at material interfaces can suppress further activity until the externally applied voltage changes. This effect leads to PD pulses occurring much less frequently under DC conditions, which means that the physical damage caused by PD itself may be minimal. However, the build-up of charges in the insulation poses a hidden threat that could lead to sudden breakdown in abnormal circumstances, without exhibiting a prior growth in PD activity that might be expected under AC conditions. Hence, although DC PD pulses may be sparse, they should still be regarded as indicators of potential insulation weakness and techniques must be developed to address the challenging task of interpreting DC PD activity. PD detection and analysis under DC conditions is of growing interest and research progress made in some of the earlier work was summarised in 2005 by Morshuis et al. [19]. For detection purposes, the same types of PD sensor can be used as for AC since the same current pulse is flowing in the external circuit. Hence the mature measurement techniques developed for AC PD testing remain applicable under DC conditions [18]. For DC PD analysis, no standard method of representation exists, unlike the AC case where phase resolved patterns can be obtained for many different tests [19]. There are two available parameters in DC PD tests, namely, the discharge magnitude qi and time interval between two

consecutive pulses Δti, which are analogous to PD magnitude qi and phase different i in AC tests, as illustrated in Fig. 3. In AC testing, these quantities both feed into the “apparent charge” measure of PD magnitude that is defined in the measurement standard IEC 60270. For DC testing, there is not yet an internationally agreed standard for assigning a magnitude to PD severity the PD pulses under steady-state DC conditions may occur very infrequently (for example, many minutes or more apart in time). However, proposals made during ongoing revisions to the IEC 60270 partial discharge standard are mentioned in [20], which describes some methods for evaluating PD data from DC tests. Typically, the pattern of consecutive PD pulses at a constant DC test voltage is recorded over a relatively long test period (typically 30 minutes). From this measurement, the cumulative PD pulse charge plot over the

5

test period is constructed. Since the slope of the cumulative charge plot is proportional to the mean PD current, this is a useful quantity because it relates to whether the PD current is growing or diminishing over time. Two simple methods for presenting the distribution of PD pulse amplitudes graphically are also suggested: The first representation counts the total number of PD pulses whose charge lies within consecutive bands of PD magnitude – these are termed the class frequencies of PD pulses. Secondly (and closely related to the first representation), the number of pulses whose apparent charge exceeds a defined series of threshold levels is counted – these are termed the exceeding frequencies of PD pulses. These plots are best appreciated by means of an example, which can be found in Section 4.1 of this paper. An important aspect of HVDC testing that needs to be defined is the ramp test profile [21], for which one option is to use the peak value VR of the AC PD inception voltage as a basis for defining the voltage variation with respect to time. The ramp test consists of three voltage increments, as shown in Fig. 4. Following each increment, the voltage is held constant at VR/2, VR and 3VR/2 respectively for 30 minutes each, after which the recorded PD pulses can be represented using the statistical analysis mentioned above. At the conclusion of the test, the voltage is ramped down (from 3VR/2 to 0 V) within 30 seconds.

Fig. 3 Basic PD parameters for AC (left) and DC (right) excitation voltages [19]. The magnitudes of PD pulses are

represented by qi while the phase and time differences between successive pulses are represented by i and Δti.

Fig. 4 Ramp test profile for DC testing. VR is the peak value of the AC PD inception voltage.

4. Experimental Results In this section, various measurement examples are presented to illustrate some aspects of HVDC partial discharge and the challenges that are posed by the difference in activity that it apparent compared AC tests.

Time (min)

Vo

ltag

e (

V)

3VR/2

VR/2

VR

0.5 32.533

6565.5

97.598

098.5

2.535 67.5

6

4.1 Comparing PD measurements under AC and DC

The test object shown in Fig. 5 is used here to illustrate the different responses of a PD source under AC and DC conditions. The sample consists of a disc of epoxy resin insulation into which a cluster of sub-mm voids were introduced prior to curing the sample. The five voids were measured using a microscope as having diameters of 0.59, 0.46, 0.43, 0.32 and 0.21 mm. The region containing these voids was mounted between brass electrodes attached to the upper and lower faces of the disc using the epoxy resin adhesive. The AC and DC test arrangements are shown in Figs. 6(a) and 6(b) respectively. AC tests were performed using a calibrated IEC60270 standard measurement system (LDS-6) comprising an HV transformer, coupling capacitor CAC = 1 nF and a proprietary measuring impedance Zm. The bandwidth of this measurement system was 100 kHz – 400 kHz. The DC tests used a circuit consisting of an HVDC power

supply, current-limiting resistor Rin = 25 M and an HV coupling capacitor CDC = 1.9 nF. The sensor used in this case was a high frequency current transformer (HFCT) with a nominal transfer impedance of 4.3 Ω and a bandwidth of 100 kHz – 20 MHz. PD pulses were recorded using the same IEC PD measurement system used for the AC tests. Prior to both AC and DC tests, a pulse injection calibration unit was used to calibrate the response of the IEC 60270 measurement system in terms of the apparent charge.

(a) (b)

Fig. 5 (a) Structure of the epoxy resin test object with five small internal voids in the region between electrodes. Diameters of the voids varied between 0.2 mm and 0.6 mm. (b) Photograph of the assembly.

(a)

(b)

Fig. 6 Configurations of test equipment for PD testing of insulation samples under (a) AC, and (b) DC excitation. In both cases capacitance Ca represents the test object while CAC/CDC are the respective coupling capacitors. Zm is the standard measurement impedance for the AC test set. For the DC testing, Rin is a current-limiting

resistor and the PD pulses are measured using an HFCT sensor.

7

AC tests were performed by gradually increasing the applied voltage until sustained PD became apparent at the inception voltage, which was 8 kV rms for this sample. At this voltage, the IEC PD level indicated by the measurement system was 680 pC and the PD activity shown in Fig. 7 in phase-resolved form was recorded over a 10 second period. Note that the PD pulses tend to occur in advance of the test voltage peaks, which is typical for internal discharges. While the PD level appears quite large for small voids, this is because the electrodes are strongly coupled to the PD currents due to their proximity. For DC testing, the ramp method introduced in Fig. 4 was used. The AC inception voltage of 8 kV rms gives a peak voltage of VR = 11.3 kV that was used to define three holding voltages (5.6 kV, 11.3 kV and 17.0 kV) for the ramp profile. However, even after the voltage had reached the highest level of 17 kV, hardly any PD was detected after the 2 minute wait period (which is necessary to ensure any that PD activity detected is not solely caused by the recent voltage change). The voltage was increased further to 20 kV where 33 PD events were recorded over the next 30 minutes – these are shown in Fig. 8(a). Fig. 8(b) is the plot of accumulated charge over the measurement period. To illustrate the statistical methods introduced in Section 3, Fig. 9(a) represents the class frequencies of PD pulses while Fig. 9(b) shows the exceeding frequencies of PD pulses for this measurement.

Fig. 7 Phase-resolved plot of PD activity accumulated over 10 seconds at 8 kV rms for

the epoxy resin test object containing voids.

Fig. 8 PD activity of the epoxy resin test object with voids at 20 kV DC. (a) Individual PD pulses recorded

over the 30 minute test period, and (b) the corresponding cumulative charge curve.

8

Fig. 9 Representation of the DC PD activity of Fig. 8(a) according the methods suggested in [20]:

(a) class frequencies of PD pulses, and (b) exceeding frequencies of PD pulses.

The hypothesis is that statistical distribution parameters representing the profile of plots such as those in Fig. 9 can be extracted as elements of feature vectors that will allow discrimination between different types of HVDC PD phenomena. As experience is gained with online monitoring of DC insulation in service, techniques will be developed to better assess the severity of incipient defects and judge the extent to which they are evolving towards potential failure. Two issues that need to be addressed during the development of these capabilities are:

How the data should be normalised with respect to charge magnitude, given that the maximum PD level could vary considerably between different tests.

How many classes the PD amplitude range should be divided into. Using too few divisions will degrade resolution of the distribution shape, while using too many could lead to some classes containing insufficient data to be of use.

4.2 Influence of voltage ripple Harmonic ripple is an undesirable phenomenon in HVDC systems but its presence may be of value for PD diagnostics since variations in the applied voltage resulting from ripple may influence PD activity, as will be demonstrated here. Harmonics produced by the voltage source converter (VSC) are unique to the specific VSC technology and its modulation techniques and switching frequencies. The higher the ‘pulse number’ of the converter (more frequent switching between phases within a cycle), the lower the harmonic distortion in the DC terminal voltage. For simplicity, harmonic orders based on a simple 6-pulse converter are used, based on the harmonic content listed in Table 1.

Table 1 Relative line-voltage harmonic amplitudes produced by an idealized 6-pulse voltage source converter [22].

Harmonic order Harmonic frequency Amplitude relative to DC

6th 300 Hz 4.04 % 12th 600 Hz 0.99 % 18th 900 Hz 0.44 % 24th 1200 Hz 0.25 %

9

In the test setup of Fig. 10, HV waveforms were applied to insulation samples by first defining a low voltage waveform using an arbitrary waveform generator and then using a high voltage amplifier to boost this signal to a level suitable for the PD test object. Results from three different PD samples are presented, all of which had the same general construction of the sample in Fig. 5. The first contained an embedded sharp metallic point on one of the electrodes; the second contained a single 0.6 mm diameter cavity in the epoxy resin; the third produced surface discharge on the epoxy sample. PD pulses were again measured using an HFCT (transfer impedance 4.3 Ω, bandwidth 100 kHz – 20 MHz) clamped around the earth lead from the test object. The voltage monitor output of the HV amplifier allows a low voltage version of the amplifier’s output (3000:1 divider ratio) to be fed into the oscilloscope for the purpose of correlating voltage levels and phase of the HV ripple with PD pulse activity.

Fig. 10 Experimental configuration for generating and measuring the effects of VSC ripple

on the PD characteristics of various test objects.

With the application of harmonic ripple of increasing magnitude relative to the HVDC voltage, PD repetition period shifted from the expected quasi-static pattern (with some stochastic variation) to a repetition interval that tended to synchronise with the period of the dominant harmonic frequency. This is illustrated in Fig. 11, where the interval between groups of PD activity increases from about 0.7 ms with a pure DC voltage to 3.3 ms (corresponding to the period of the 6th harmonic at 300 Hz) when the 6-pulse converter ripple harmonics were present. This change was accompanied by a significant increase in pulse amplitude as the pulse spacing increased. Phase synchronisation with the harmonic ripple became increasingly pronounced as the amplitude of the dominant harmonic was increased. While PD still remained active at the pure HVDC voltage (no ripple), its amplitude was significantly lower than when harmonics were present for all the PD samples tested. Phase-resolved partial discharge (PRPD) patterns were then compared for the three test objects under application of both AC and superimposed VSC-type ripple to establish similarity, or otherwise, in PD phase-synchronisation characteristics. Figs. 12 – 14 show test results for samples tested consecutively at the same peak applied voltage. These plots use persistence to show in grey the envelope of the cumulative discharge data that forms characteristic patterns with respect to the AC cycle or harmonic ripple. In each case, a single sweep data record of PD pulses has been superimposed in black on the plots. As expected under purely AC conditions, PD may occur at voltage peaks of either polarity, or on the rising or falling slopes of the power frequency cycle (depending on the defect type). With the application of harmonic ripple on a DC voltage, for all defect topologies tested, PD pulses tend to cluster around the harmonic peaks where the composite applied voltage is at a maximum. The most obvious difference in PD activity between HVAC and HVDC with superimposed ripple occurs in the case of the void (Fig. 13), where PD is concentrated in the 1st and 3rd quadrants of the AC sinusoid, but around the voltage peaks and shortly afterwards for the HVDC with ripple.

10

Fig. 11 Effect of 6-pulse converter harmonic ripple amplitude on PD activity for a sharp protrusion source.

(a) DC only; (b) 5% dominant harmonic; (c) 10 % dominant harmonic, and (d) 15% dominant harmonic.

Fig. 12 Sharp protrusion discharge sample: (a) AC PRPD pattern synchronised with the 50 Hz power cycle;

(b) DC + harmonic PRPD pattern synchronised with the 6th harmonic period.

Fig. 13 Void discharge sample: (a) AC PRPD pattern synchronised with the 50 Hz power cycle;

(b) DC + harmonic PRPD pattern synchronised with the 6th harmonic period.

11

Fig. 14 Surface discharge sample: (a) AC PRPD pattern synchronised with the 50 Hz power cycle;

(b) DC + harmonic PRPD pattern synchronised with the 6th harmonic period.

4.3 Testing MV cable samples A number of cable test objects were prepared using 33 kV XLPE AC cables with a 185 mm2 aluminium conductor. Each was fitted with air terminations and defective joints were fabricated at each cable’s mid-point. One of these samples is shown in Fig. 15. Some views of the defects during construction are shown in Fig. 16. These include a rough cutting of the outer semiconducting layer, as shown in Fig. 16(a), and the introduction of an annular void region into the insulation, as shown in Fig. 16(b). Experiments on the cable samples were carried out in a laboratory environment using the test facility shown in Fig. 17. This can provide an AC supply of up to 100 kV rms or a DC supply of either polarity up to 135 kV in a low-noise environment that supports a PD detection threshold of < 2 pC.

Fig. 15 Example of typical test object constructed using 33 kV XLPE-insulated cable.

(a) (b) Fig. 16 Internal views of defects in cable joints: (a) ‘Rough cut’ screen (semicon layer) incorporated into the

joint. (b) ‘Void’ in the insulation created by cutting an annular ring in the XLPE material.

12

Fig. 17 Cable sample (bottom left) under test at the HVDC test facility, University of Strathclyde.

The HV transformer (bottom right) feeds a half-wave rectifier and smoothing circuit. The PD measurements presented here will cover two of the 33 kV cable samples, the first having a screen cut in the joint and the second having an insulation void, as per Figs. 16(a) and 16(b) respectively. The nominal phase-to-earth voltage for this cable is U0 = 33/√3 = 19 kV rms. PD measurements under 50 Hz AC excitation are described first: Fig. 18 shows the phase-resolved PD measurement for the screen-cut cable at the inception voltage of 35 kV rms (which is nearly double the nominal operating voltage). This indicates that the physical defect is not severe enough to cause PD under normal operating conditions. Fig. 19 shows the much greater level of PD activity generated in the cable with the insulation void. Note that this measurement was recorded at 18 kV rms, which is slightly below the nominal rated voltage of the cable and reflects the severity of the insulation damage.

Fig. 18 Phase resolved PD plot for the 33 kV cable with rough screen cut inside the joint. This measurement

was recorded at a phase-to-earth voltage of 35 kV rms and the apparent charge measured according to IEC 60270 was 26 pC.

Fig. 19 Phase resolved PD plot for the 33 kV cable with insulation void inside the joint. This measurement

was recorded at a phase-to-earth voltage of 18 kV rms and the apparent charge measured according to IEC 60270 was 558 pC.

13

The screen cut cable was then subjected to a DC ramp test that involved considerable over-stressing of the insulation. The results of a test conducted with a negative DC voltage reaching -80 kV over a 30 minute period are shown in Fig. 20(a). Only 5 PD pulses occurred during this time, which means that the data available to populate the class frequency histogram of Fig. 20(b) is far from ideal. Since the damage to this cable screen does not seem to have caused serious weakness to the insulation, it was tested further under a harsher DC operating regime. This involved cycling the negative DC potential between -80 kV and -4 kV (the minimum voltage setting of the supply without switching off). Fig. 21 shows the measured PD data, which is still very sparse. Note that, in the majority of cases, a negative PD pulse occurs during the ramp up from -4 kV to -80 kV. However, in one case there is a positive PD pulse that occurs during the decreasing voltage ramp. Finally, the cable with screen cut was subjected to a sequence of ramp stress tests up to 70 kV DC potential that involved polarity reversal in the sequence +, -, +, as shown in Fig. 22. Note that the all of the larger PD pulses occur at the times of energising and removal of the test voltage and that the polarity of the PD pulses at these times has no obvious correlation with the applied DC polarity. During ramping, PD pulses are sparse and of low magnitude (< 50 pC). Since the increases in voltage are more modest at each change of potential in this case, there is usually a delay of some tens of seconds before a resulting PD pulse occurs.

(a) (b)

Fig. 20 (a) PD test results for the screen cut cable sample as an increasing negative DC overvoltage was applied. (b) The histogram of class frequencies of PD pulses is compromised by the paucity of data in this case.

(a) (b)

Fig. 21 (a) PD test results for the screen cut cable sample as the applied negative DC voltage was being ramped repetitively between -4 kV and -80 kV. (b) The histogram of class frequencies of PD pulses

is again of questionable value because there is so little PD data.

14

Fig. 22 DC stress test on the cable sample with screen cut in the joint. The top plot show the applied voltage and

PD activity for the first, positive polarity test. The middle plot is for the next, negative polarity test and the bottom plot show the final positive stress test. Note that PD activity remains sparse,

except for the times of switch-on and switch-off of the HVDC supply.

When recording PD from the cable sample with a joint containing the insulation void, PD activity during HVDC testing was much more vigorous, reflecting the rather extreme nature of the damage. This is illustrated by the soak test results shown in Fig. 23, in which the cable was energised at -70 kV for a period of 10 hours. Some of the larger pulses were clipped in amplitude by the measurement system, but this was regarded as an acceptable compromise in order to avoid operating at a range that would give poor resolution for the majority of lower-level PD pulses. While most of the PD pulses have the same polarity as the applied voltage, there are occasional bursts of activity that indicate current flow in the opposite direction. These are attributed to localised electric field conditions due to space charge accumulation that can result in reversal of the expected electric field direction. The complete dataset for this soak test contains 35,752 pulses detected over the experiment duration of 36,331 s. Hence the frequency with which detectable PD pulses are generated is about 1 Hz in this case. Fig. 24(a) shows the

magnitude of charge transferred by PD over the test period. This reaches just over 5 C after 10 hours, corresponding to an average current of 140 pA. While there is some fluctuation in the average current (evidenced by deviations of the curve from linear), the overall trend suggests that the PD current is showing no obvious signs of growth or decline. Fig. 24(b) shows the class frequency histogram for this data, which is much more convincing than for the screen cut cable due to the volume of pulse data now available. This type of data and its trending over time is likely to represent one of the more robust options for PD diagnostics on HVDC insulation.

15

Fig. 23 PD activity recorded during a 10-hour test at -70 kV DC on the cable sample with an insulation void.

Although the PD activity appears quite dense, on average there was about 1 pulse per second, still much less frequent than the rate of PD pulses observed under 50 Hz AC excitation.

(a) (b)

Fig. 24 Analysis of the PD data of Fig. 23: (a) Cumulative charge plot, the slope of which gives an indication of periods of higher or lower current flow. (b) The histogram of class frequencies in this case is more

representative of the PD activity due to the volume of PD pulse data.

4.4 Using HVAC cables for HVDC The results above have demonstrated that PD activity is much less frequent under DC conditions than would be the case under a comparable AC operating regime. The cable with a screen cut defect proved able to withstand a DC potential of more than 4 times its nominal phase-to-earth AC rms rating with very little PD activity being apparent. This observation concurs with field tests that have been carried out at a substation where two 3-phase 33 kV AC cables are being evaluated for potential conversion to HVDC use. As part of a rigorous series of measurements, all six of the cable phases withstood a 15 min test at 38 kV DC (2U0), during which an OLPD monitor equipped with HFCT sensors was used to capture any resulting PD activity. There was no evidence of PD in any of the cables under these conditions, but when the cables were in circuit at 50 Hz rated voltage, PD of up to 250 pC in magnitude was generated within one of the cable phases. This illustrates that AC testing is a better option for identifying insulation weaknesses because of the difficulty of initiating PD under DC potential unless the insulation damage is quite severe.

16

5. Discussion of PD Monitoring for HVDC equipment In the context of testing HVDC components, conventional AC PD measurements may retain a role as part of the manufacturer’s quality assurance. However, once an HVDC system has been constructed in the field, it will be of value to apply monitoring as far as is possible, practical and economical during the various phases of its life. Doing so is expected to offer the following benefits:

Ability to detect and rectify manufacturing or assembly defects at commissioning, before they lead to component failure or secondary damage. This might include, for example, detecting partial discharges at inception, below the normal operating voltage during commissioning tests.

Ability to detect equipment defects or deficiencies during the early part of the ‘bathtub’ curve, reducing costs to the operator by ensuring that the benefits of the initial warranty period are maximised and reducing risk to insurers by ensuring that equipment more likely to fail early is brought up to standard before the expiry of warranty.

Ability to reduce operators’ repair costs and regulatory fines (imposed as a result of outages or power quality issues) through enabling condition-based and predictive maintenance to be exploited to a greater extent using the holistic monitoring system approach.

Ability to plan capital investment for replacement or upgrade of assets on the basis of a better understanding of plant health and prognosis of its future rate of deterioration.

Given the characteristics of PD activity under HVDC conditions that have been reported in the literature and observed in the experiments described here, it is possible to outline an approach to the monitoring of PD in HVDC equipment under various headings, as follows:

Sensors Partial discharge sensors should be installed at suitable, accessible locations on the HVDC network to be monitored, in accordance with practice for HVAC equipment. Sensors can include HFCTs, coupling capacitors, TEV detectors and RF antennas, all of which must be appropriately rated for HVDC use and comply with safety requirements to ensure that HV clearances are not compromised. In this respect, it should be noted that ideally, facilities for condition monitoring sensors should be included at the design stage for new equipment as this will help to ensure that monitoring equipment can be installed more effectively than might be the case in a retro-fit scenario. Signal acquisition The challenge in terms of sampling rate for signals from the PD sensors is to optimise the trade-off between hardware cost, volume of data and signal fidelity, all of which increase with sampling rate. Since an accurate representation of the PD pulse detail in the time-domain is preferable to enable discrimination between different types of pulses from different sources, it is considered that a sampling rate of 100 MHz is optimal at present. This may of course change in future, as analogue-to-digital sampling technology continues to advance. As described in the discussions of ‘System triggering’ and ‘Voltage measurement’ below, the length of sampled time-domain data recorded in a single acquisition could benefit from being dynamically variable and governed by measurements of HVDC system parameters, as well as the characteristics of the PD activity itself. System triggering The most significant difference between HVAC and HVDC OLPD (On-Line Partial Discharge) systems is that the latter should normally be ‘event driven’, rather than triggering repetitively on the AC power frequency cycle. This change is necessary due to the scarcity of PD pulses under DC conditions, which means that the conventional AC operation of the digitising, data transfer and re-arming cycle of the acquisition hardware could lead to a high probability of PD pulses being missed. In addition, the conventional approach would lead to the processing of large numbers of ‘empty’ data sets. Instead, the monitoring system should be set to trigger on the detection of a PD pulse and then to record and transfer the signal data only for the period of time necessary for the PD event(s) to be captured in full with sufficient fidelity for pulse characterisation.

17

Synchronous time-stamping A valuable feature expected of HVDC-OLPD monitoring system functionality is that PD signals acquired from all of the sensors in a particular locality of the HVDC network should be captured in a time-synchronised manner. This enables the use of pulse amplitude and time-of-arrival comparisons among sensor outputs when the detected pulses can be attributed to a common source. Both of these comparison methods are important tools for locating the origin of pulses and discriminating between different signal sources (be they PD or interference). An extension of this requirement is that the recorded timing of PD pulses should not simply be traceable across other sensors but should also be known relative to the phase of the AC cycle (or switching regime) of the converter station (where this is nearby and accessible). During monitoring of an HVDC sub-sea transmission cable, this functionality has been found to offer a considerable improvement in the discrimination between PD and non-PD transients detected by the sensors. In addition, it provides diagnostic information concerning PD that is associated with repetitive, system-generated transients. Voltage measurement The HVDC-OLPD system should preferably have access to a measurement channel / sensor that allows it to monitor the instantaneous voltage level on the HVDC system. This voltage sensing capability should have the maximum bandwidth that can practically be achieved (given the limitations imposed by the physical size of HVDC voltage dividers), because its value extends beyond the measurement of the average DC voltage. Superimposed on the HVDC conductor will be ripple and transients, which are known to play a significant role in influencing PD activity in HVDC systems. The incorporation of an instantaneous voltage measurement channel into the OLPD monitor enables dynamic optimisation of signal acquisition settings. This allows the monitoring system to operate in an adaptive mode, in which system triggering and record length are dynamically optimised to capture PD activity in a manner appropriate to changing system conditions, such as: voltage ramping (increasing / decreasing during energising /de-energising), VSC ripple, voltage spikes, switching transients, etc. Environmental sensors Since trends and cycles in parameters such as temperature and humidity have the potential to modify both PD activity and PD-like interference signals, it is important to log such parameters so that their influence can be taken into account within the holistic diagnostic process. While ambient sensors have considerable value, it is equally important to have additional sensing where local effects are dominant (for example, load-dependent heating of components such as cables). Additional data sources As with environmental parameters, knowledge of system operational events and their timing with respect to PD activity will play the same valuable role in the condition monitoring of HVDC equipment as it does for AC. For example, profiling of detected pulses against the timeline of switching operations and the load profile of the circuit being monitoring will allow the diagnosis of incoming data to be carried out with the full picture of possible initiatory and contributing factors. This functionality will help to discount certain signals from the sensors that might otherwise cause undue concern, while allowing attention to be focused on those that are genuinely caused by incipient insulation defects.

18

6. Conclusions Increasing use of HVDC equipment means that PD monitoring techniques need to be adapted from their conventional focus on diagnosis of AC insulation. Although the same types of sensors can be used, interpretation methods need to be quite different due to the absence of a cyclic phase reference and the fact that PD activity is often very intermittent under HVDC conditions. Particularly in solid insulation, PD phenomena under HVDC stress are complicated by the build-up of space charge, which modifies the electric field in the insulation and leads to localised distortion and regions of high electric stress. Standards for PD testing of HVDC equipment are still in development, but key aspects of these tests will include ramp-and-hold type voltage regimes where PD data may have to be gathered over test periods of 30 minutes or more to be sufficient for any kind of evaluation. The results on 33 kV cable samples presented in this paper have indicated that PD pulses can be very rare, even for quite harsh DC overpotential test regimes. In addition to PD being initiated by changes in the level of applied DC potential, experiments have shown that the harmonic ripple superimposed on the DC can cause PD activity to synchronise with the dominant ripple frequency. This may be a useful phenomenon to assist with diagnosis in some circumstances, such as when there is a significant level of VSC ripple present on the HVDC conductor. Statistical approaches that have been proposed for the purpose of classifying HVPD discharges have been applied in this paper, but it is clear that these will be of marginal value when there is very little PD data despite the long duration of testing. Nevertheless, they do appear promising when there is a reasonable rate of PD activity under DC energisation, but certain issues remain to be addressed in terms of normalising the pulse data to allow comparison across different measurement arrangements. The best knowledge is likely to be developed through experience gained during on-line monitoring in the field, which has already been initiated. On the basis of investigations carried out to date, we have been able to propose the functionality required of an on-line PD monitor for HVDC, taking into account the influencing factors outlined in this paper.

Acknowledgements Some of the results presented in this paper were obtained in the context of the recent project “On-Line HVDC Cable Monitor Proposal (OLPD-HVDC) for Offshore Wind Component Technologies Development and Demonstration Scheme” supported by the UK Government Department of Energy & Climate Change. The project has also been supported by Scottish Power Energy Networks through NIA funding. The PhD research of Edward Corr is funded by the EPSRC, project reference number EP/G037728/1.

19

References [1] P Bresesti, W L Kling, R L Hendriks and R Vailati, “HVDC connection of offshore wind farms to the

transmission system”, IEEE Trans. Energy Conversion, Vol. 22, No. 1, pp. 37-43, 2007

[2] www.4coffshore.com/windfarms/converters.aspx (accessed 04/11/15)

[3] www.4coffshore.com/windfarms/hvdc-converter-borwin1-converter-cid1.html (accessed 02/10/15)

[4] Tijdo van der Zee, “Offshore substations should have a standard size of 900MW”, The Offshore Wind Industry, Vol. 3, July 2012

[5] A Madariaga, O El Mountassir and N MacLeod, “Offshore HVDC Risk Perceptions: Are they Driving Investments Appropriately?” 14th Wind Integration Workshop (Brussels, Belgium), Paper No. WIW15-161, October 2015

[6] Carbon Trust, “Cable Burial Risk Assessment Methodology: Guidance for the Preparation of Cable Burial Depth of Lowering Specification”, Doc. Ref. CTC835, February 2015 – available from www.carbontrust.com (accessed 02/10/15)

[7] BBC News, “Moyle Interconnector: Repairs Cost £31M”, www.bbc.co.uk/news/uk-northern-ireland-19969955 (accessed 02/10/2015)

[8] CIGRE Working Group B1.32, “Recommendations for testing DC extruded cable systems for power transmission at a rated voltage up to 500 kV”, Technical Brochure 496, CIGRE, April 2012

[9] CIGRE Working Group 21.02, “Recommendations for tests of power transmission DC cables for a rated voltage up to 800 kV”, CIGRE, Elektra No. 189, pp. 39-55, April 2000

[10] CIGRE Study Committee B1, “Update of service experience of HV underground and submarine cable systems”, Technical Brochure 379, CIGRE, April 2009

[11] G Mazzanti and M Marzinotto, Extruded Cables for High-Voltage Direct-Current Transmission: Advances in Research and Development, IEEE Press Series on Power Engineering, 2013

[12] A Eigner and S Semino, “50 years of electrical-stress control in cable accessories”, IEEE Electrical Insulation Magazine, Vol. 29, No. 5, pp. 47-55, Sept/Oct 2013

[13] T Yamanaka, S Maruyama and T Tanaka, “The development of DC+/-500 kV XLPE cable in consideration of the space charge accumulation”, Proc. 7th Int. Conf. on Properties and Applications of Dielectric Materials, (Nagoya, Japan), Vol. 2, pp. 689-694, June 2003

[14] T L Hanley, R P Burford, R J Fleming and K W Barber, “A general review of polymeric insulation for use in HVDC cables”, IEEE Electrical Insulation Magazine, Vol. 19, No. 1, pp. 13-24, Jan/Feb 2003

[15] A Farkas, C-O Olsson, G Dominguez, V Englund, P-O Hagstrand and U H Nilsson, “Development of high performance polymeric materials for HVDC cables”, Paper A.2.5, 8th Int. Conf. on Insulated Power Cables (Jicable), (Versailles, France), June 2011

[16] H Ghorbani, M Jeroense, C-O Olsson and M Saltzer, “HVDC Cable Systems—Highlighting Extruded Technology”, IEEE Trans. Power Delivery, Vol. 29, No. 1, pp. 414-421, February 2014

[17] L A Renforth, P S Hamer, D Clark, S Goodfellow and R Tower, “Continuous Remote Online Partial Discharge Monitoring of HV Ex/ATEX Motors in the Oil and Gas Industry”, IEEE Trans. Industry Applications, Vol. 51, No. 2, pp. 1326-1332, 2015

[18] P Morshuis, M Jeroense and J Beyer, “Partial discharge. Part XXIV: The analysis of PD in HVDC equipment”, IEEE Electrical Insulation Magazine, Vol. 13, No. 2, pp. 6-16, March/April 1997

[19] P H F Morshuis and J J Smit, “Partial discharges at DC voltage: their mechanism, detection and analysis”, IEEE Trans. Dielectrics and Electrical Insulation, Vol. 12, No. 2, pp. 328-340, April 2005

[20] High-Voltage Test and Measuring Techniques, W Hauschild and E Lemke, Springer, Section 6.5 “PD Measurement at DC Test Voltages”, pp. 279-283, 2014

[21] R S Bever, “Ramp Technique for Partial Discharge Testing”, IEEE Trans. Electrical Insulation, Vol. 20, No. 1, pp. 38-46, February 1985

[22] HVDC for Beginners and Beyond, ALSTOM Grid UK, Publication Ref. ‘Systems-L4-HVDC Basics’ (2165-V2-EN), July 2010