Partial Discharge Measurement

Abstract

This chapter is devoted to the measurement of partial discharges (PD) originating in weak spots in the insulation of HV apparatus and their components. As most HV equipment used to generate, transmit and distribute electric power are energized by high voltage alternating current (HVAC), this chapter focuses primarily on PD measurements under alternating voltage. However, specific problems which arise for PD tests under DC and impulse voltage will also briefly be considered. Sensitive PD measurements are often interfered by electromagnetic noises in the measuring surroundings. Therefore advanced features developed in the past to cancel disturbing noises will also be highlighted in this chapter as well as in the relevant sub-sections of Sects. 10.3 and 10.4. For a better understanding of the principles, procedures and instrumentation required for the measurement of electrical PD quantities, first some fundamentals of the PD occurrence will be presented, where also the PD quantities recommended in IEC 60270:2000 for the insulation condition assessment are considered. In the following section, the existing PD models proposed to estimate the PD charge transfer will critically be reviewed. Thereafter specific aspects of pulse charge measurements, as specified in IEC 60270:2000, are considered, which include the major components required to design PD measuring systems used to acquire and display the captured PD data and even the procedures employed to calibrate PD measuring systems. The following sections are dealing with the localization of PD faults, the visualization of PD events, and features developed to reduce or cancel electromagnetic noises interfering sensitive PD measurements. Finally, so-called non-conventional PD detection methods will briefly be highlighted, such as the measurement of electromagnetic PD transients up to the ultra-high frequency range as well as the detection of ultrasonic signals emitted from PD sources.

4.1 Fundamentals

4.1.1 PD Occurrence

Partial discharges (PD) are often caused by imperfections in dielectric materials, such as gaseous inclusions in liquid and solid dielectrics as well as sharp edges in ambient air, which cause a local field enhancement, so that the intrinsic field strength may be exceeded. As a consequence, self-sustaining avalanche-discharge may be ignited, where the electrons liberated from neutral gas molecules on account of collision and photo ionization are moving at extremely high drift velocity. This is associated with a very fast displacement current characterized by time parameters in the nanosecond range, which induces a pulse charge at the electrodes. This can thus be detected by means of a coupling device connected to the terminals of the test object or even via field sensors to capture the associated electromagnetic transients radiated from the test object. As defined in the relevant standard IEC 60270:2000, partial discharges are

localized electrical discharges that only partially bridge the insulation between conductors and which can or cannot occur adjacent to a conductor. Partial discharges are in general a consequence of local electrical stress concentrations in the insulation or on the surface of the insulation. Generally such discharges appear as pulses having durations of much less than 1 μs.

The first experimental study of PD signatures dates back to the year 1777 when Lichtenberg discovered dust figures like stars and circles. These appeared on the surface of an amber cake after it has been hidden by spark discharges up to 40 cm in length (Lichtenberg 1777, 1778). In the middle of the nineteenth century, besides the dust-figure technique initially used by Lichtenberg also photographs became a valuable tool to study PD phenomena, such as surface and interfacial discharges (Blake 1870; Toepler 1898; Müller 1927). At the beginning of the last century, when high alternating voltage was increasingly employed for long-distance power transmission links, it became known that partial discharges can be considered as a precursor for an ultimate breakdown, particularly when occurring in gaseous inclusions embedded in solid dielectrics. Since that time various tools, such as optical, chemical, acoustical and electrical methods, were increasingly used to detect partial discharges. Since the 1960s, the electrical PD measurement has become a widely established procedure for quality assurance tests of HV apparatus and their components, as will be discussed in more detail in the following sections.

Basically, partial discharges are caused by the ionization of gas molecules. Thus PD events occur not only in ambient air but also in gaseous inclusions in solid dielectrics or in gas bubbles and even in water vapour in liquid dielectrics. Therefore it is widely accepted that the discharge processes occurring in gaseous inclusions are comparable with those appearing in ambient air, such as Townsend and streamer discharges as well as leader discharges, which have extensively been investigated since the late 19th and early 20th centuries and published in numerous technical papers and text-books (Paschen 1889, Townsend 1915, 1925; Schumann 1923; Meek and Craggs 1953; Loeb 1956; Raether 1964; Park and Cones 1963; Devins 1984).

Photographs of typical discharge filaments observed in ambient air, insulating oil, and in PMMA are exemplarily shown in Fig. 4.1 (Lemke 1967; Hauschild 1970; Pilling 1976).

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig1_HTML.gif — Fig. 4.1
Photographs of discharge channels. a Streamer discharge in air (Lemke 1967). b Leader discharge in oil (Hauschild 1970). c Electrical trees in PMMA (Pilling 1976)

As already mentioned above, the formation of single electron avalanches occurs within the nanosecond range due to the extremely fast drift velocity of the electrons. This is associated with a very fast rising displacement current and induces thus a pulse charge at the electrodes of the test object, which occurs also within the nanosecond range, as has been estimated theoretically by Raether in 1964 and by Bailey in 1966 and confirmed experimentally by Fujimoto and Boggs in 1981 as well as by Boggs and Stone in 1982 using the first available 1 GHz oscilloscope (Fig. 4.2). As the origin of PD defects in the insulation of power equipment is usually not accessible, the true shape of PD current pulses cannot be measured when decoupled from the terminals of the test object. This is because the frequency content of the PD signal is dramatically reduced due to the inevitable attenuation and dispersion of the PD signal when transferred from the PD source to the terminals of the test object.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig2_HTML.gif — Fig. 4.2
Time parameters of PD current pulses. a Theoretical pulse shape calculated by Bailey for small voids embedded in solid dielectrics. b Pulse shapes measured by Boggs and Stone for a sharp point (left) and a floating particle in SF6 (right)

As an example, consider the oscilloscopic screenshot shown in Fig. 4.3. Here, an artificial PD pulse created by a PD calibrator was injected in a power cable of 16 m in length. The distance between injection point and near cable end, where the oscilloscope was connected, was chosen as 4m. Thus the first (direct) pulse traveled 4 m, while the second pulse reflected at the remote cable end traveled in total (16 − 4)m + 16 m = 28 m. Considering Fig. 4.3b, where the bandwidth of the oscilloscope was set to 200 MHz, the peak value of the second (reflected) pulse is substantially lower than the peak value of the first arriving (direct) pulse. Reducing the bandwidth down to 20 MHz, however, the peak value of the first pulse appears also substantially reduced, while the peak value of the second pulse is only marginally affected. With other words: the peak values of both pulses become invariant when the measuring frequency is chosen as low as possible, which is in principle accomplished by a low-pass filtering and thus by a so-called quasi-integration. Under this condition the peak values become proportional to the time integral of the recorded pulses, which can be considered as the background for measuring the PD quantity “ apparent charge ” (Kind 1963; Schon 1986; Zaengl and Osvath 1986). Based on this approach it is recommended in IEC 60270:2000 as well as in the associated Amendment published in 2015, that the upper limit frequency of the entire measuring system should be chosen below 1 MHz to accomplish the desired quasi-integration of the captured PD signal.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig3_HTML.gif — Fig. 4.3
Influence of the measuring frequency on the attenuation of a PD pulse travelling along a XLPE power cable of 16 m in length. a Experimental set-up. b Oscilloscopic record gained at overall bandwidth of 200 MHz. c Oscilloscopic record gained at overall bandwidth of 20 MHz

4.1.2 PD Quantities

As mentioned above, to evaluate the PD severity of HV apparatus in service, chemical, optical, acoustical and electrical PD detection methods are commonly employed. In the following, however, only PD quantities gained in the course of electrical PD measurements will be considered, as defined in IEC 60270:2000 as well as in the Amendment to this standard published in 2015.

1.
Apparent charge

The apparent charge , which can be considered as the main PD quantity, is defined in the relevant IEC 60270:2000 as

that charge which, if injected within a very short time between the terminals of the test object in a specified test circuit, would give the same reading on the measuring instrument as the PD current itself. The apparent charge is usually expressed in picocoulombs (pC).

Moreover it is noted in this standard, that

the apparent charge is not equal to the amount of charge locally involved at the site of the discharge, which cannot be measured directly.

For more information in this respect see the Sects. 4.3 and 4.4, respectively.

2.
PD inception and extinction voltage

The inception voltage V_i and the extinction voltage V_e are usually of interest for HV equipment or its components, which failed the type or development test or even the quality assurance test after manufacturing. Moreover it is a common practice to determine V_i and V_e of HV apparatus, which cannot be designed fully PD-free under operation voltage, such as, for instance, the stator bar insulation of rotating machines. As a practical example, consider the PC screenshots shown in Fig. 4.4, which refer to development tests of a turbo-generator. The voltage profile applied is shown in Fig. 4.4a. Considering Fig. 4.4b, the PD level recorded under rising AC test voltage (red trace) deviates only marginally from that recorded under decreasing test voltage (green trace). Such a low hysteresis is commonly a signature for a “healthy” insulation, which is also manifested by the comparatively low difference between the inception voltage V_i and the extinction voltage V_e. Different to this, harmful PD defects can often be recognized by a clear hysteresis and an extinction voltage being significantly lower than the inception voltage, as exemplarily shown in Fig. 4.4c.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig4_HTML.gif — Fig. 4.4
PC screenshots gained in the course of development tests of turbo-generators, rated voltage 24 kV. (a) Profile of the applied 50 Hz Ac test voltage (b) PD level of a “healthy” stator-bar under raising (red trace) and decreasing (green trace) test voltage (c) PD level of a “faulty” stator-bar

As known, PD pulses are only detectable when the apparent charge exceeds the background noise level, which should thus be kept as low as possible to prevent erroneous test results. For a better understanding consider Fig. 4.5, which refers to a PD test of a defective XLPE cable termination, rated voltage 36 kV. Under laboratory conditions (noise level < 0.5 pC) an inception voltage of 12 kV_peak was measured. Assuming, for instance, a background noise level of 5 pC, the measurable inception voltage would increase up to 25 kV_peak, as indicated in Fig. 4.5.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig5_HTML.gif — Fig. 4.5
Impact of the background noise level on the determination of the inception and extinction voltage (explanation in the text)

3.
Phase-angle

The determination of the phase-angle Φ_i according to IEC 60270:2000 is based on the following equation:

$\Phi _{i} = \frac{{\Delta t_{i} }}{{T_{c} }} \cdot 360^{0} ,$

(4.1)

with ∆t_i the time span elapsing between the negative-to-positive crossing of the applied AC test voltage and that instant at which the PD pulse is ignited, and T_c the cycle duration of the applied AC test voltage.

Performing practical PD tests, however, the actual phase-angle of each individual PD pulse is usually of minor interest, while the qualitative distribution of the PD pulse trains correlated to the phase-angle of the applied AC test voltage is often helpful to identify harmful PD sources. For a better understanding consider the oscilloscopic screenshots shown in Fig. 4.6, which refer to so-called Trichel discharges. Due to the comparatively high pulse repetition rate, the individual charge pulses appeared only time-resolved when the initially chosen time scale of 4 ms/div (Fig. 4.6a) was set to 0.4 ms/div (Fig. 4.6b). As can be seen, the Trichel pulses are randomly distributed around the negative peak region of the applied AC test voltage, i.e. the phase-angle is scattering around 270°.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig6_HTML.gif — Fig. 4.6
Apparent charge pulses (pink traces) of Trichel discharges occurring in a 3.5 mm needle-plane gap under AC test voltage (green trace) displayed at a time base of 4 ms/div (a) and 0.4 ms/div (b), respectively

As another example, consider the oscilloscopic records shown in Fig. 4.7, which refer to surface and cavity discharges. Here the PD pulses are randomly distributed around the zero-crossing of the AC test voltage, i.e. the phase-angle scatters around 0 and 180°, respectively. As can be seen, the PD signature of cavity discharges is qualitatively comparable to that of surface discharges. However, the pulse magnitudes of surface discharges are substantially greater than those of cavity discharges, which offers the opportunity to discriminate between both kinds of discharges.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig7_HTML.gif — Fig. 4.7
Oscilloscopic screenshots of apparent charge pulses (pink traces) gained for surface discharges (a) and cavity discharges (b) under power frequency test voltage (green traces)

4.
Pulse repetition rate

The determination of the PD pulse repetition rate n according to IEC 60270:2000 is based on the following equation:

$n = \frac{{N_{p} }}{\Delta t}$

(4.2)

with N_p the total number of PD pulses and ∆t the recording time interval. In practice only those PD pulses are evaluated, which exceed either a specified magnitude or occur within a specified range of magnitude.

As a practical example consider the oscilloscopic screenshots shown Fig. 4.8, which were gained in the course of on-site PD tests of 110 kV power transformers. Rising the power frequency (50 Hz) test voltage up to only V_rms = 19 kV, a single PD pulse appeared in each half-cycle, where the apparent charge exceeded 60 nC in magnitude, see Fig. 4.8a. Here a recording time of ∆t = 50 ms was chosen, which covers five half-cycles. Thus the pulse repetition amounts n = 5/(50 ms) = 100 s⁻¹. However, increasing the test voltage level further, at least up to the highest phase-to-earth-voltage being V_rms = 71 kV, the magnitude of the apparent charge pulses remained nearly constant, while the PD pulse number increased up to seven per half-cycle, i.e. the repetition rate attained n = 35/(50 ms) = 700 s⁻¹.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig8_HTML.gif — Fig. 4.8
Oscilloscopic screenshot showing the PD signatures of a defective 110 kV transformer bushing at inception voltage V_irms = 19 kV (a) and at highest phase-to-earth voltage being V_rms = 71 kV (b)

Note Based on further experimental studies it has been found that the extremely high magnitudes of the charge pulses displayed in Fig. 4.8 are the result of spark discharges originating in a transformer bushing on account of a poor grounding of the bushing tap terminal.

5.
Accumulated apparent charge

The accumulated apparent charge q_n is defined in the Amendment to IEC 60270:2000 and can be determined using the following expression:

$q_{n} = /q_{1} / + / q_{2} / + / q_{3} / + \cdots + / q_{i} / ,$

(4.3)

with /q₁/, /q₂/, /q₃/ … /q_i/ the absolute values of the apparent charge, and i the number of PD pulses captured within the time interval ∆t_r.

Under practical condition only PD pulses are considered, which exceed either a specified threshold level or occur within specified apparent charge limits. For a better understanding consider Fig. 4.9, which refers to the PD behavior of a defective XLPE cable termination of 24 kV rated voltage. Rising the AC voltage up to a test level of V_rms = 12.8 kV, the maximum apparent charge of the recorded PD pulse trains approached almost 500 pC (pink trace), where the accumulated apparent charge exceeded 7000 pC per cycle (green trace).

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig9_HTML.gif — Fig. 4.9
Oscilloscopic screenshot gained for a defective XLPE cable termination (rated voltage 24 kV) showing besides the apparent charge pulses (pink trace) also the accumulated apparent charge (green trace), where the recording time covers both half-cycles of the applied power frequency (50 Hz) test voltage (blue trace)

6.
Average discharge current

The determination of the average discharge current I_n according to IEC 60270:2000 is based on the following equation:

$I_{n} = \frac{1}{{T_{ref} }} \cdot \left\{ {/q_{1} / + /q_{2} / + /q_{3} / + \cdot \cdot \cdot + /q_{i} /} \right\} = \frac{{q_{n} }}{{T_{ref} }},$

(4.4)

with T_ref the chosen reference time interval and i the number of the captured PD pulses. As a quantitative example consider the above presented Fig. 4.9. Inserting the accumulated charge being 7360 pC in Eq. (4.4) one gets the following average discharge current:

$I_{n} = \frac{7360\,pC}{20\,ms} \approx 0.37 \mu A.$

7.
Discharge power

The determination of the discharge power P_n according to IEC 60270:2000 is based on the following equation:

$P_{n} = \frac{1}{{T_{ref} }} \cdot \left\{ {/q_{1} \cdot v_{1} / + /q_{2} \cdot v_{2} / + /q_{3} \cdot v_{3} / + \cdot \cdot \cdot + /q_{i} \cdot v_{i} /} \right\},$

(4.5)

with v₁, v₂, v₃ … v_i the values of the AC test voltage at instants when the PD pulses of q₁, q₂, q₃ … q_i in magnitude are captured. Basically it has to be taken into account that the phase-angle of the measured AC test voltage is equal to that applied to the terminals of the test object. To determine the discharge power P_n at reasonable measuring accuracy in the real-time mode needs the use of computerized PD measuring systems.

8.
Quadratic rate

The determination of the quadratic rate D_n according to IEC 60270:2000 is based on the following equation:

$D_{n} = \frac{1}{{T_{ref} }} \cdot \left\{ {q_{1}^{2} + q_{2}^{2} + q_{3}^{2} \cdots + q_{i}^{2} } \right\} .$

(4.6)

To determine this PD quantity in the real-time mode at reasonable measuring accuracy, the use of computerized PD measuring systems is inevitable. This seldom used quantity was introduced to underline the PD severity of high PD pulse magnitudes.

4.2 PD Models

As known, the electromagnetic transients due to PD events are only detectable at the terminals of the test object. Therefore it seems of interest if this external pulse charge is correlated with the internal pulse charge flowing through the PD defect. To analyse the PD charge transfer, instead of realistic dielectric imperfections shown in Fig. 4.10 (Kreuger 1989), simple shapes of gaseous inclusions are commonly considered, such as spherical, elliptical and cylindrical cavities. Generally, it can be distinguished between the network-based PD model represented by a capacitive equivalent circuit, and the dipole-based PD model represented by bipolar space charges deposited at the cavity walls.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig10_HTML.gif — Fig. 4.10
Typical sizes of gaseous inclusions in solid dielectrics according to Kreuger (1989) a Flat cavity perpendicular to the electric field, b Spherical cavity, c Flat cavity aligned with the electric field, d Interfacial cavity

4.2.1 Network-Based PD Model

The origin of the network-based PD model traces back to the year 1928, when Burstyn analyzed theoretically the breakdown sequences per half-cycle of a spark gap connected via a small capacitance to an AC test voltage source. The main aim behind this approach was to underline that the comparatively high loss factor of mass-impregnated power cables is the consequence of spark discharges in gaseous inclusions embedded between the laminated insulation (Burstyn 1928). This simple approach has later been investigated experimentally by Gemant and Philippoff using the network proposed by Burstyn, as shown in Fig. 4.11 (Gemant and Philippoff 1932). To measure the breakdown sequences per half-cycle, the spark gap F was connected in parallel to the horizontal deflection plates of a high-voltage oscilloscope. To change the effective capacitance C₂, which is being discharged via the spark gap, additional capacitors were connected in parallel. The capacitance C₁ and resistance R connected in series, see Fig. 4.11, served for a limitation of the transient current through the spark gap F. Typical oscilloscopic records gained for two different test levels are shown in Fig. 4.12. The studies revealed that the measured breakdown sequence depending on both the test voltage level and the capacitance C₂ being discharged was in reasonable agreement with theoretical estimations.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig11_HTML.gif — Fig. 4.11
Experimental set-up used by Gemant and Philippoff (1932) to measure the breakdown sequence of the spark gap depending on the applied AC test voltage level

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig12_HTML.gif — Fig. 4.12
Oscilloscopic records taken by Gemant and Philippoff (1932) (a) and comparison with theoretical estimations (b)

To calculate not only the breakdown sequences depending on the test voltage level and the capacitance being discharged but also to deduce the detectable external charge from the internal charge flowing through the gas-filled cavity, the circuit shown in Fig. 4.11, has later been modified by Whitehead (1951) and Kreuger (1964), as depicted in Fig. 4.13. In this circuit, C_a represents the capacitance of the bulk dielectric between the electrodes of the test object, while C_b and C_c represent the series connection of the fictive capacitance of the “healthy” dielectric column with the intuitively assumed cavity capacitance, which is being periodically discharged via the parallel connected spark gap F_c. Due to the three characteristic capacitances the equivalent circuit shown in Fig. 4.13 is traditionally referred to as abc-model .

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig13_HTML.gif — Fig. 4.13
Network-based PD model. a Plane-parallel electrodes with capacitive circuit elements. b Capacitive equivalent circuit (abc-model)

To analyse the pulse charge transfer from the PD source to the terminals of the test object, a cylinder-like column of diameter $2r_{b}$ is commonly considered, as illustrated in Fig. 4.13a. To simplify the following treatment the radius of the “healthy” dielectric column is assumed as equal to that of the gas-filled cylindrical cavity. Under this condition the equivalent capacitances C_b and C_c can simply be expressed by

$C_{b} = \varepsilon_{0} \cdot \varepsilon_{r} \cdot \frac{{\pi \cdot r_{b}^{2} }}{{d_{b} }},$

(4.7a)

$C_{c} = \varepsilon_{0} \cdot \frac{{\pi \cdot r_{b}^{2} }}{{d_{c} }},$

(4.7b)

with d_c—the length of the gaseous cylinder representing the PD defect, d_b = d_b1 + d_b2—the entire length of the both solid cylinders (partial lengths d_b1 and d_b2) between gaseous cylinder and electrodes, ε₀—the permittivity of air, and ε_r—the relative permittivity of the solid dielectric. Provided, the spark gap shown in Fig. 4.13a is breaking down at inception field strength E_i, the transient voltage appearing across the cavity capacitance C_c would collapse down to almost zero and thus attain a magnitude of ∆V_c ≈ E_i ∙ d_c. That means, the internal charge q_c stored in the capacitance C_c is entirely neutralized and disappears thus completely. Therefore it can be written:

$q_{c} \approx \Delta V_{c} \cdot C_{c} \approx E_{i} \cdot \varepsilon_{o} \cdot \pi \cdot r_{b}^{2} .$

(4.8)

As indicated in Fig. 4.13b, the above mentioned voltage collapse ∆V_c appears also across the series connection of the equivalent capacitances C_b1 − C_b2 − C_a. Assuming for simplification the relation C_b1 = C_b2 = 2 C_b, one gets the following measurable voltage step, which appears across C_a:

$\Delta V_{a} = \Delta V_{c} \cdot \frac{{C_{b} }}{{C_{a} + C_{b} }}$

(4.9)

Multiplying this with the test object capacitance C_a, the measurable external charge can be expressed as follows:

$q_{a} = \Delta V_{a} \cdot C_{a} = \Delta V_{c} \cdot \frac{{C_{b} \cdot C_{a} }}{{C_{b} + C_{a} }} \approx \Delta V_{c} \cdot C_{b} \approx E_{i} \cdot \varepsilon_{0} \cdot \varepsilon_{r} \cdot \pi \cdot r_{b}^{2} \cdot \frac{{d_{c} }}{{d_{b} }}$

(4.10a)

$q_{a} \approx q_{c} \cdot \frac{{d_{c} }}{{d_{b} }}$

(4.10b)

For technical insulation, the inequality d_c ≪ d_b1 + d_b2 = d_b ≈ d_a is generally satisfied and the relative permittivity $\varepsilon_{r}$ of the used solid dielectrics is commonly below 5. Under this condition follows from Eq. (10):

$\frac{{q_{a} }}{{q_{c} }} \approx \varepsilon_{r} \frac{{d_{c} }}{{d_{a} }}{ \ll }1.$

(4.11)

Due to this relation, the external charge detectable in the connection leads of the test object is referred to as apparent charge and it is noted in IEC 60270:2000 that the apparent charge is not equal to the amount of that charge involved at the site of the discharge, which cannot be measured directly.

4.2.2 Dipole-Based PD Model

The classical apparent charge concept based on the capacitive equivalent circuit according to Fig. 4.13 has been rejected by Pedersen and his co-workers (Pedersen 1986, 1987; Pedersen et al. 1991, 1995). In particular they criticized that the voltage collapsing across the intuitively assumed cavity capacitance is not the result of a spark discharge but caused by the formation of an electrical dipole, which is established by bipolar space charges deposited at the cavity walls. As these bipolar charge carriers are released from neutral gas molecules in the course of collision and photo ionization processes, the number of electrons and negative ions deposited at the anode-side dielectric boundary equals that number of positive ions deposited at the cathode-side dielectric boundary (Fig. 4.14). As the Poissian field between the space charges of opposite polarity opposes the Laplacian (electrostatic) field resulting from the applied test voltage, the ionization of gas molecules will suddenly be quenched. The time interval required for the separation of the bipolar charge carriers lies usually in the nanosecond range, which is thus also representative for the time parameters of the associated current pulse.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig14_HTML.gif — Fig. 4.14
Dipole model according to Pedersen (1986, 1987). a Space charge field due to bipolar point charges deposited at the cavity wall. b Circuit elements

To estimate the transient PD current through the equivalent capacitance C_a and thus through the test object shown in Fig. 4.14, the classical continuity equation applies. That means the conductive current $i_{c} (t)$ through the gaseous inclusion, represented in Fig. 4.14b by the capacitance C_c, equals the displacement current $i_{b} (t)$ through the solid dielectric column represented by the capacitances $C_{b1}$ and $C_{b2}$ , and continues thus also through the test object capacitance C_a, which causes a voltage jump ∆V_a. That means, the external PD charge $q_{a}$ detectable at the terminals of the test sample equals the internal PD charge $q_{c}$ , which is given by the time-integral of the transient current flowing through the gaseous inclusion on account of the motion of the bipolar charge carriers. Obviously, this is in contrast to the classical apparent charge concept deduced from the previously considered network-based PD model.

To estimate the PD charge transfer quantitatively, Pedersen proposed a field theoretical approach, which is based on the Maxwell equations (Pedersen 1986). However, this alternative concept has been ignored in the past, while the traditional abc-model is promoted also nowadays (Achillides et al. 2008, 2013, 2017). The reason for that is, apparently, that a PD charge transfer cannot easily be explained when Pedersens concept is adopted, because this needs an excellent knowledge of the field theory based on the Maxwell equations (Maxwell 1873). However, the analysis of the PD charge transfer can substantially be simplified when instead of spherical, ellipsoidal or cylindrical cavities, as frequently investigated and published in numerous technical papers, the establishment of a dipole moment under quasi-uniform field conditions is considered, i.e. the Laplacian field between the electrodes is assumed as constant for the time span required for the separation of the charge carriers (Lemke 2012, 2016). Therefore, to estimate the PD charge transfer, only the change of the Poissian field caused by the separation of the charge carriers of opposite polarity has to be analyzed.

For a better understanding, consider first the motion of a single electron and the associated positive ion, where the propagation to the electrodes shall be hampered by two dielectric layers arranged at spacing d_c, see Fig. 4.15. Provided, at inception field strength E_i only a single electron carrying the elementary charge—e is liberated from a neutral molecule, which occurs at position x = x_i,, this electron will be attracted by the anode due to the Coulomb force F = −e ∙ E_i. Thus, after travelling the maximum possible distance x_i, the field energy transferred to the electron becomes

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig15_HTML.gif — Fig. 4.15
Parameters used for the estimation of the field energy transferred to an electron and the associated positive ion

$W_{e} = F\int\limits_{{x_{i} }}^{o} {{\text{d}}x = - e \cdot E_{i} \left( {0 - x_{i} } \right) = e \cdot E_{i} \cdot x_{i} .}$

(4.12)

In an analogue manner one gets the following expression for the energy, which is being transferred from the field to the associated positive ion carrying the elementary charge +e:

$W_{p} = F\int\limits_{{x_{i} }}^{o} {{\text{d}}x = e \cdot E_{i} \cdot \left( {d_{c} - x_{i} } \right).}$

(4.13)

Combining the Eqs. (4.12) and (4.13) the entire field energy transferred to both the electron and the positive ion after approaching the solid dielectric boundaries, becomes:

$W_{t} = W_{e} + W_{p} = e \cdot E_{i} \cdot d_{c} .$

(4.14)

Obviously, this expression is independent on the actual site x_i where the electron was released from a neutral molecule. Thus the maximum field energy transferred to an electron avalanche created by the ionization of $n_{i}$ neutral gas molecules can simply be expressed by

$W_{a} = e \cdot n_{i} \cdot E_{i} \cdot d_{c} .$

(4.15)

If one imagines that just before the ignition of a PD event, which occurs at the inception voltage V_i, the test sample is disconnected from the HV test supply, the energy transferred to the moving charge carriers is delivered exclusively from the field energy stored in the test object capacitance $C_{a}$ , see Fig. 4.14b. For simplification it shall be assumed that at instant $t_{d}$ all charge carrier approached the dielectric boundaries. Under this condition the classical energy balance theorem can be written as follows:

$W_{a} = V_{i} \int\limits_{o}^{{t_{d} }} {i_{a} (t) \cdot {\text{d}}t = V_{i} \cdot q_{a} = e \cdot n_{i} \cdot d_{c} \cdot E_{i} = P_{m} \cdot E_{i} ,}$

(4.16)

with $P_{m} = e \cdot n_{i} \cdot d_{c}$ —the dipole moment created by the bipolar charge carriers, and E_i—the inception field strength. Separating the charge $q_{a}$ delivered from the test object capacitance C_a one gets

$q_{a} = e \cdot n_{i} \cdot d_{c} \cdot \frac{{E_{i} }}{{V_{i} }} = P_{m} \cdot \frac{{E_{i} }}{{V_{i} }}.$

(4.17)

In this context it seems worth to notice that a similar approach can also be deduced using the concept of image charges (Shockley 1938; Kapcov 1955; Frommhold 1956). This has often been employed by many researchers to account for the displacement current associated with the motion of charge carriers between plane-parallel electrodes (Meek and Craggs 1953; Raether 1964).

Generally it is supposed that the insulation deterioration caused by PD events is mainly governed by the field energy transferred to the charge carriers. Without going into further details it can thus be stated based on Eqs. (4.15) and (4.16) that the measurable external PD charge $q_{a}$ represents a reasonable quantity to assess the PD severity . So the density of the dielectric flux lines entering the electrodes and thus the danger for an ultimate breakdown increases at increasing cavity length d_c, see Fig. 4.16. This is not reflected by the previously discussed network-based PD model, where the internal charge decreases at increasing cavity length, because this leads to a decrease of the cavity capacitance C_c.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig16_HTML.gif — Fig. 4.16
Distribution of the dielectric flux density depending on the distance d_c between the positive and negative space charge

Note: With reference to the previously considered network-based PD model according to Fig. 4.14 it is noteworthy that the internal PD charge q_c could be measured under the condition when all negative and positive charge carriers would be able to cross the entire distance between the anode and cathode. As can readily be shown under this condition the internal charge q_c becomes equal to the measurable external PD charge q_a. Based on the Eqs. (4.10b) and (4.17) it can thus be written:
$\begin{aligned} & q_{a} \approx q_{c} \cdot \frac{{d_{c} }}{{d_{b} }} = e \cdot n_{i} \cdot \frac{{d_{c} \cdot E_{i} }}{{V_{i} }} = e \cdot n_{i} \cdot \frac{{d_{c} \cdot E_{i} }}{{d_{b} \cdot E_{i} }} = e \cdot n_{i} \cdot \frac{{d_{c} }}{{d_{b} }} \\ & q_{c} \approx e \cdot n_{i} . \\ \end{aligned}$
That means the “fictive” internal PD charge q_c deduced from the network-based PD model equals that charge amount, which is carried by either the positive ions alone or even by the negative electrons alone. However, this is in contrast to the physics of gas discharges due to the fact that in case of a cavity discharge the charge amount carried by the positive ions is equal but of opposite polarity to that charge amount carried by the electrons and negative ions. Due to the very short distance between the bipolar space charges deposited at anode-side and cathode-side dielectric boundary, a great amount of the the positive space charge is compensated by the negative space charge, as can also be deduced from Fig. 4.16a. Consequently, the net charge attains only a low fraction of the unipolar charge carried by the positive ions alone, which is given by e ∙ n_i, with n_i—the number of ionized gas molecules.

Even if Eq. (4.16) applies to quasi-homogenous field conditions, this is in principle also applicable for technical electrode configurations, provided the cavity length $d_{c}$ in field direction is substantially lower than the electrode spacing (Lemke 2013). As an example, consider Fig. 4.17, which refers to a spherical cavity embedded in the bulk dielectric between coaxial cylinder electrodes, as representative for power cables.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig17_HTML.gif — Fig. 4.17
Parameters used for analyzing the PD charge transfer in extruded power cables

To simplify the following treatment, a virgin (space-charge-free) spherical cavity of radius r_c embedded in the bulk dielectric between plane electrodes of distance d_a shall be considered. This seems reasonable because the charge carriers affecting the field distribution are created just after the instant when the inception field strength E_i is exceeded. Assuming the inequality r_c ≪ d_a, which is generally satisfied for technical insulation, the so-called field enhancement factor k_ε can be approximated as follows (Schwaiger 1925):

$k_{\varepsilon } = \frac{{3\varepsilon_{r} }}{{1 + 2\varepsilon_{r} }}.$

(4.18)

Considering a polyethylene-insulated cable, where the relative dielectric permittivity amounts ε_r = 2.3, one gets k_ε ≈ 1.2. Inserting this in Eq. (4.17), the detectable pulse charge can be estimated using the following approach:

$q_{a} \approx P_{m} \cdot \frac{1.2}{{r_{c} \left[ {\ln \left( {\frac{{r_{a} }}{{r_{i} }}} \right)} \right]}}.$

(4.19)

Determining the dipole moment $P_{m}$ quantitatively, however, is a challenge due to the fact that the number $n_{i}$ of ionized molecules is randomly distributed over an extremely wide range. Thus it seems reasonable to investigate only the worst case, which arises when streamer-like discharges are ignited. For this specific case the dipole moment can be expressed by the following semi-empirical approach (Lemke 2013):

$P_{m} \approx (270\;{\text{pC}}/{\text{mm}}) \cdot d_{c}^{2} \quad {\text{for}}\;0.1\;{\text{mm}} < d_{c} < 2\;{\text{mm}}.$

(4.20)

Introducing this in Eq. (4.19) and inserting the following assumed geometrical parameters r_i = 8.5 mm, r_a = 14 mm and r_c = 10 mm, which are representative for a 20-kV polyethylene-insulated power cable, one gets

$q_{a} \approx \left( {67\;{\text{pC}}} \right) \cdot \left( {\frac{{d_{c} }}{{d_{o} }}} \right)^{2} .$

(4.21)

with d₀ = 1 mm—a reference cavity diameter. The curve $q_{a}$ versus $d_{c}$ following from this approach is plotted in Fig. 4.18. For comparison purposes, experimental data are also plotted in this figure. In this context it should be noted that some experimental data refer to configurations other than the coaxial cylinder electrodes, as has been investigated here. Nevertheless, the calculated curve meets the measured data quite well, where a better agreement cannot be expected due to the inherent large scattering of the PD pulse magnitudes.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig18_HTML.gif — Fig. 4.18
PD pulse charge q_a versus cavity diameter d_c calculated for a 20 kV XLPE cable and experimental data

With respect to quality assurance tests of extruded power cables, the pulse charge $q_{a}$ versus the PD inception voltage V _i is the most interesting function. To solve Eq. 4.17, the ratio $E_{1} /V_{i}$ must be known. An appropriate approach to determine the inception voltage E_i versus the cavity diameter d_c can be deduced from the Schumann curves (Schumann 1923), which reads:

$E_{i} \approx E_{0} \left( {1 + \sqrt {\frac{{d_{r} }}{{d_{c} }}} } \right)\quad {\text{for 0}}.1\;{\text{mm}} < d_{c} < 2\;{\text{mm,}}$

(4.22)

with $E_{0} = 2.47\;{\text{kV/mm}}$ —the static inception field strength of ambient air under atmospheric reference conditions and $d_{r} = 0.82\;{\text{mm}}$ a reference cavity diameter. Inserting these values in Eq. (4.17), one gets the following approach to determine the inception voltage of a virgin spherical cavity embedded in the bulk dielectric between coaxial cylinder electrodes:

$V_{i} = \frac{{E_{0} }}{{k_{\varepsilon } }} \cdot r_{c} \cdot \left[ {\ln \left( {\frac{{r_{a} }}{{r_{i} }}} \right)} \right] \cdot \left( {1 + \sqrt {\frac{{d_{\text{r}} }}{{d_{c} }}} } \right) \approx V_{i} \approx \left( {10.5\;{\text{kV}}} \right) \cdot \left( {1 + \sqrt {\frac{{0.82\;{\text{mm}}}}{{d_{c} }}} } \right).$

(4.23)

Combining the Eqs. (4.21) and (4.23), the detectable PD pulse charge $q_{a}$ versus the inception voltage $V_{i}$ has been calculated for and plotted in Fig. 4.19. In this context it seems worth to notice that in IEC 60502:1997 a maximum test voltage level of 1.73 V₀ is recommended for medium voltage cables with extruded insulation, where V₀ represents the rms value of the phase-to-earth voltage. For the here considered 20 kV cable this is equivalent to a peak voltage of 28 kV. In Fig. 4.19a this value is indicated by a circle, which yields that the cable would pass the PD test only when the PD level remains below 5 pC. This result is in satisfying agreement with the recommendations of IEC 60502:1997, which specifies that the magnitude of the discharge at 1.73 V₀ shall not exceed 10 pC. Moreover, it can be deduced from Eq. (4.23) that a PD inception voltage above 28 kV can only be guaranteed for a cavity diameter d_c < 0.3 mm.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig19_HTML.gif — Fig. 4.19
PD inception voltage versus PD charge (a) and cavity diameter (b) calculated for an XLPE medium voltage cable

Even if the calculated curves plotted in the Figs. 4.18 and 4.19 are in satisfying agreement with practical experience, it must be emphasized here that the quantitative values are only approximations. This is because Eq. (4.17) refers to a virgin cavity filled with ambient air under atmospheric normal conditions. However, for technical insulation, neither the cavity size nor the gas pressure are known. Moreover, the dipole moment established by subsequent PD events may strongly be affected by space charges deposited at the cavity walls on account of preceding PD events. This might also be the reason for the large scattering of the pulse charge magnitudes, as commonly encountered in practice.

4.3 PD Pulse Charge Measurement

4.3.1 Decoupling of PD Signals

As has been discussed already in Sect. 4.1, the frequency content of PD current pulses is substantially reduced when the electromagnetic transients are travelling from the PD site to the terminals of the test object, see Fig. 4.3. Different to this, the time integral of the transient PD current and thus the pulse charge is more or less invariant. For a better understanding, consider Fig. 4.20, which refers to a gas-filled cavity embedded in the bulk dielectric between plane-parallel electrodes, as has already been investigated in Sect. 4.2.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig20_HTML.gif — Fig. 4.20
Principle of pulse charge measurement

As the PD transients are characterized by time parameters down to the nanosecond range and thus by a frequency spectrum up to 100 MHz and even more, the impedance of the HV inductance between HV test supply and test object becomes very high, so that for the duration of each PD event the test object can be considered as disconnected from the HV test supply. Under this condition the charge $q_{c}$ flowing through the PD defect is provided only by the test object capacitance $C_{a}$ , which is associated with a fast voltage collapse $\Delta V_{a}$ across the test object. Following the dipole-based PD model, the pulse charge $q_{c}$ flowing through the PD defect is equal to that charge amount flowing through the capacitance C_a of the test object. Therefore it can be written:

$q_{c} = q_{a} = \Delta V_{a} \cdot C_{a} .$

(4.24)

At instant when the PD process is quenched and the charge carriers are deposited at the cavity walls, the PD current decays to zero so that the frequency content and thus the effective impedance of the HV connection leads decreases accordingly. Therefore the test object capacitance $C_{a}$ will be recharged again by the HV test supply, i.e. the former voltage step $\Delta V_{a}$ appearing across C_a is compensated, which appears at inverted polarity. That means the time integral of the current recharging C_a and thus the apparent charge can also be assessed using Eq. (4.24), even if the actual shape of the recharging current is very different from that of the original PD current. This offers the opportunity to measure the PD pulse charge by means of a measuring impedance , if connecting the low voltage electrode of the test object to earth potential (Fig. 4.20). Additionally the terminals of the test object should be shunted by a HV capacitance $C_{k}$ to ensure that the entire current recharging C_a is flowing through the measuring impedance .

For a direct measurement of the pulse charge, the measuring impedance could be equipped with a measuring capacitance $C_{m}$ . Under this condition, the magnitude of the voltage jump appearing across $C_{m}$ is direct proportional to the pulse charge to be measured. At alternating test voltages, however, the capacitive load current flowing through the test object and thus through the measuring impedance could substantially exceed the signal level caused by the PD events, as exemplarily shown in Fig. 4.21a. To overcome this crucial problem, the measuring capacitance $C_{m}$ is commonly shunted by a measuring resistor $R_{m}$ , which reduces the superimposed AC current accordingly, see Fig. 4.21b and c. Practical experience revealed, however, that a measuring impedance equipped with an RC network shown in Fig. 4.20 is only applicable for low-capacitive test samples used for fundamental PD studies, such as point-to-plane gaps, but not for PD tests of HV equipment and their components.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig21_HTML.gif — Fig. 4.21
Records of typical voltage signals caused by PD events (pink trace), which were captured from an RC measuring impedance connected in series with a point-to-plane gap subject to AC test voltage (green trace)

Even if the highest measuring sensitivity is achieved by connecting the measuring impedance between low voltage side of the test object and ground potential, this approach is not applicable in general. On one hand the ground connection lead of the test object can often not be interrupted and, on the other hand, the measuring impedance must be able to carry the entire capacitive load current through the test object, as discussed above, and additionally the fast transient current that would occur in case of an unexpected breakdown. To overcome this crucial problems, the measuring impedance is usually connected in series with the coupling capacitor $C_{k}$ , as illustrated in Fig. 4.22. Here, the measuring resistor $R_{m}$ is shunted by an inductance L_s, which carries the entire alternating load current trough the coupling capacitor. The over-voltage protection unit $O_{p}$ is required to suppress fast over-voltages due to unexpected breakdowns and prevents thus a damage of the measuring impedance and the measuring instrument, as well. As the PD coupling unit shown in Fig. 4.22 provides a high-pass filter, it has to be taken care that the lower limit frequency $f_{1}$ is chosen as low as possible, preferably below 100 kHz.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig22_HTML.gif — Fig. 4.22
PD measuring circuit, where the measuring impedance is connected in series with the coupling capacitor

Example To design a coupling device according to Fig. 4.22, which is intended for induced voltage tests of power transformers, the following parameters shall be assumed:

$\begin{aligned} & {\text{Lower}}\,{\text{limit}}\,{\text{frequency:}}\quad f_{1} = 50\;{\text{kHz}} \\ & {\text{Effective}}\,{\text{measuring}}\,{\text{impedance:}}\quad R_{m} = 1\;{\text{k}}\varOmega \\ & {\text{Maximum}}\,{\text{applied test}}\,{\text{voltage}}\,{\text{level:}}\quad V_{a} = 200\;{\text{kV}} \\ & {\text{Maximum}}\,{\text{test}}\,{\text{frequency:}}\quad f_{ac} = 400\;{\text{Hz}} \\ \end{aligned}$
As the lower limit frequency is given by
$f_{1} = \frac{1}{{2\pi \cdot C_{k} \cdot R_{m} }} = 50\;{\text{kHz}}$
the minimum capacitance of the coupling capacitor can be determined as follows:
$C_{k} = \frac{1}{{2\pi \cdot f_{1} \cdot R_{m} }} \approx 3.2\;{\text{nF}}.$

Inserting the maximum exciting frequency being f_ac = 400 Hz, one gets the following capacitive current through the coupling capacitor:
$I_{c} = 2\pi \cdot f_{\text{ac}} \cdot C_{k} \cdot V_{\text{ac}} = 1.6\;{\text{A}}.$

As this current flows through the measuring resistor $R_{m} = 1\;{\text{k}}\Omega$ , a voltage drop as high as 1600 V would appear. Of course, this is dangerous for the operator and could also damage the connected PD measuring system. Thus, to reduce this high voltage drop, R_m should be shunted by an inductance L_s (Fig. 4.22). However, applying this option it has to be taken care that the lower limit frequency may significantly exceed the previously mentioned lower limit frequency, which should be chosen below 100 kHz. This condition is accomplished for
$2\pi \cdot f_{m} \cdot L_{s} \ge 5R_{m} ;\quad L_{s} \ge 16\;{\text{mH}}$

For the here considered maximum frequency f_ac = 400 Hz of the applied test voltage the inductive impedance attains
$Z_{l} = 2\pi \cdot f_{\text{ac}} \cdot L_{s} \approx 40\;\varOmega .$

Under this condition, the above mentioned load current of $I_{c} = 1.6\;{\text{A}}$ flowing through the coupling capacitor of $C_{k} = 3.2\;{\text{nF}}$ causes a comparative low voltage drop across $L_{s}$ , which attains 64 V. This value can further be reduced using a high-pass filter of higher order.
To display the PD pulse in a phase-resolved matter by means of an oscilloscope or even a computer-based PD measuring system, the PD coupling unit could further be configured using an additional measuring capacitor C_m, as illustrated in Fig. 4.22. Due to the very different frequency spectra of the PD pulses and the AC test voltage, these both signals can simply be discriminated. Considering Fig. 4.22, these appear at the outputs “PD pulses” and “test voltage”. If, for instance, the above-introduced maximum test voltage level of V_ac = 200 kV should be attenuated down to 50 V, a divider ratio of 1:4000 would be required. Using a coupling capacitor of C_k = 3.2 nF, a low-voltage measuring capacitance of C_m = 12.8 μF would be required.

4.3.2 PD Measuring Circuits According to IEC 60270

The basic PD measuring circuits recommended in IEC 60270:2000 are shown in Fig. 4.23. To avoid any danger for the operator due to hazardous over-voltages in case of an insulation breakdown, the measuring impedance should always be placed inside the HV test area. Moreover, it has to be taken care that the HV connection leads are designed PD-free up to the highest applied test voltage level. The grounding leads used for the current return should be kept as short as possible and made of Cu or AL foil of approx. 100 mm in width to minimize the inevitable inductance in the higher-frequency range and thus to prevent electromagnetic interferences. For more information in this respect see Sects. 4.5.2 and 9.2.2.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig23_HTML.gif — Fig. 4.23
Basic PD measuring circuits recommended in IEC 60270:2000. a Measuring impedance in series with a coupling capacitor used for grounded test objects. b Test object grounded via the measuring impedance. c Bridge circuit recommended for noise reduction

The most commonly employed coupling mode is the use of a coupling capacitor in series with a measuring impedance (Fig. 4.24), as has already been discussed previously based on Fig. 4.22.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig24_HTML.gif — Fig. 4.24
Photograph of a PD measuring circuit designed according to IEC 60270.
Courtesy of Doble Lemke

Electromagnetic noises interfering sensitive PD measurements can also be eliminated to a certain extend using the so-called balanced PD bridge according to Fig. 4.23c. Here, the adjustable measuring impedances Z_m1 and Z_m2 are installed in the ground connection leads of both the test object and the coupling capacitor providing the measuring branch and the reference branch, respectively. Adjusting $Z_{m1}$ and $Z_{m2}$ accordingly to balance the bridge, common mode noises appearing at the high-voltage terminals are more or less suppressed by the differential amplifier. Thus only the PD signal originating in the test object appears at the output of the differential amplifier and is thus measured by the PD instrument. To ensure a high common mode rejection, the bridge should be designed as symmetrical as possible. Thus it is advisable to use instead of the coupling capacitor a complementary PD-free test object as a reference. Despite of the benefits of the balanced bridge for noise suppression, this approach is not generally employed in practice because the design is a challenge due to the fact that both branches must have an equivalent frequency response over the full bandwidth used for the PD signal processing. Moreover it has to be taken into account that the time-delay of the interfering signal traveling along the complementary PD-free test object is equal to that signal traveling along the test object under investigation.

Another option frequently used for PD tests of power transformers is the so-called bushing tap coupling mode, as illustrated in Fig. 4.25. Of course, this coupling mode is only applicable for capacitive-graded bushings equipped with a bushing tap intended for ${\text{C}}/\tan \;\delta$ measurements. Here the HV bushing capacitance C₁ provides in principle the coupling capacitor $C_{k}$ shown in Fig. 4.23a.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig25_HTML.gif — Fig. 4.25
Bushing tap coupling mode commonly used for PD tests of power transformers

4.3.3 PD Signal Processing

As discussed in Sect. 4.2.2, the time integral of the transient PD current flowing through the connection leads of the test object and thus the measurable pulse charge is correlated to that charge amount flowing through the PD defect. Moreover, the pulse charge is more or less invariant even if the PD current pulse may substantially be distorted when travelling from the PD source to the terminals of the test object. Thus the transient voltage appearing across a resistive measuring impedance is commonly integrated to measure the so-called apparent charge. This is conveniently achieved by a band-pass filtering , commonly referred to as quasi-integration. For this purpose the signal processing is performed in a frequency range where the amplitude–frequency spectrum of the captured PD pulses is nearly constant (Kind 1963; Kuffel et al. 2006; Schon 1986; Zaengl and Osvath 1986; König and Narayana 1993; Lemke 1997), as illustrated in Fig. 4.26. Practical experiences revealed that this requirement is accomplished for most test objects when the upper cut-off frequency is limited below 1 MHz, as has also been recommend in the Amendment to IEC 60270:2000, published in 2015.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig26_HTML.gif — Fig. 4.26
Frequency spectrum of PD pulses in comparison to the frequency band recommended for PD pulse charge measurements according to IEC 60270:2000

Depending on the bandwidth ∆f = f₂ − f₁ used for PD signal processing it is generally distinguished between wide-band and narrow-band instruments . For wide-band PD instruments the following frequency parameters are recommended in the Amendment to IEC 60270:2000:

$\begin{array}{*{20}l} {{\text{Lower}}\;{\text{limit}}\;{\text{frequency:}}} \hfill & {30\;{\text{kHz}} \le f_{1} \le 100{\text{ kHz}}} \hfill \\ {{\text{Upper}}\;{\text{limit}}\;{\text{frequency:}}} \hfill & {130\;{\text{kHz}} \le f_{2} \le 1 0 0 0\;{\text{kHz}}} \hfill \\ {\text{Bandwidth:}} \hfill & {100\;{\text{kHz}} \le \Delta f \le 970\;{\text{kHz}}} \hfill \\ \end{array}$

Note According to IEC 60270: 2000, the term “wide-band” refers to the band-pass filter characteristics of the PD processing unit characterized by a bandwidth ∆f = f₂ − f₁, which is substantially greater than the lower limit frequency f₁. In this context, it must be emphasized that this term is not correlated with the frequency spectrum of real PD current pulses, which covers often a frequency range up to the GHz range, see Sect. 4.1.

The PD pulse response of a band-pass filter having the upper and lower limit frequency of $f_{2} = 320{\mkern 1mu} {\text{kHz}}$ and $f_{1} = 40{\mkern 1mu} {\text{ kHz}},$ respectively, is shown in Fig. 4.27a. Based on the network theory it can be stated that the peak value of the output pulse is proportional to the time integral of the input pulse where the duration of the output pulse is substantially greater than that of the input pulse.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig27_HTML.gif — Fig. 4.27
PD pulse responses of a wide-band (a) and a narrow-band (b) processing unit. a Lower limit frequency $f_{1} = 40\;{\text{kHz}}$ and upper limit frequency $f_{2} = 320\;{\text{kHz}}$ . b Mid-band frequency $f_{0} = 780\;{\text{kHz}}$ and bandwidth ∆f = 9 kHz

$$ f_{1} = 40\;{\text{kHz}} $$ — Fig. 4.27
PD pulse responses of a wide-band (a) and a narrow-band (b) processing unit. a Lower limit frequency $f_{1} = 40\;{\text{kHz}}$ and upper limit frequency $f_{2} = 320\;{\text{kHz}}$ . b Mid-band frequency $f_{0} = 780\;{\text{kHz}}$ and bandwidth ∆f = 9 kHz

In this context it should be noted that the integration performance is only governed by the upper limit frequency f2 and not by the lower limit frequency f₁. Thus also narrow-band filters could in principle be used to accomplish a quasi-integration. For this purpose the bandwidth $\Delta f$ must be chosen substantially lower than the center frequency $f_{0} = \left( {f_{2} - f_{1} } \right)/2$ . As a typical measuring example, consider Fig. 4.27b, which refers to the PD pulse response of a narrow-band amplifier characterized by a center frequency of $f_{0} = 780\;{\text{kHz}}$ and a bandwidth of $\Delta f = 9\;{\text{kHz}}$ . Here the maximum peak-to-peak value of the oscillating response is direct proportional to the pulse charge injected in the the narrow-band amplifier, which follows also from the classical network theory (Schon 1986). In IEC 60270:2000 the following frequency parameters are recommended for narrow-band PD instruments:

${\text{Center frequency}}:\quad 50\,{\text{kHz}} \le f_{0} \le 1000\,{\text{kHz}}$

${\text{Bandwidth}}:\quad 9\,{\text{kHz}} \le \Delta f \le 30\,kHz$

The main advantage of narrow-band amplifiers is the noise immunity, i.e. continuous appearing high-frequency interferences received from radio broadcasting stations can effectively be canceled by tuning the center frequency $f_{0}$ accordingly, see Sect. 4.5. In this context it must be emphasized, however, that fatal superposition errors may appear due to the comparatively long duration of the oscillating response, which exceeds often several hundreds of μs. As an example consider Fig. 4.28, which refers to the double-pulse response of a narrow-band amplifier. Here the time interval between two subsequent appearing impulses was reduced from originally 1.4 µs down to 0.7 µs. As can be seen, the peak-to-peak values of the oscillating signal decreases substantially under this condition. Such a behavior is typical for PD pulses decoupled from power cables due to their reflection at the cable ends. Moreover, superposition errors may also occur on account of oscillations excited by PD events in the windings of rotating machines and power transformers. This is the reason why narrow-band amplifiers are in general not recommended for measuring the apparent charge of PD pulses in terms of pC.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig28_HTML.gif — Fig. 4.28
Impact of the double-pulse distance on the oscillation magnitude of a narrow-band PD processing unit

4.3.4 PD Measuring Instruments

4.3.4.1 General

Schering bridges in combination with oscilloscopes can be considered as the first instruments used for the electrical PD detection. However, the measuring sensitivity was comparatively low. This was substantially increased in the 1920s when the first super-heterodyne receivers equipped with narrow-band amplifiers were available (Armann and Starr 1936; Dennhardt 1935; Koske 1938; Lloyd and Starr 1928; Müller 1934; Schering 1919). To ensure comparable and reproducible PD measurements, the requirements for such instruments were first specified in the USA and North America in 1940, when the standard “Methods for Measuring Radio Noise” was published by the “National Electrical Manufactures Association NEMA”. This standard was later revised by the NEMA Publication 107 “Methods of Measurement of Radio Influence Voltage (RIV) of High-Voltage Apparatus” and issued in 1964. An equivalent standard for RIV measurements of HV apparatus was also edited in Europe by the “Comité International Spécial des Perturbation Radioélectrique (CISPR)”, published in 1961.

Radio interference voltages (RIV) are commonly measured in terms of µV and weighted according to the acoustical noise impression of the human ear. Therefore it cannot be expected that these are correlated to the apparent charge of PD pulses measured in terms of pC, as can readily be proven experimentally (Harrold and Dakin 1973; Vaillancourt et al. 1981). Moreover, fatal superposition errors might appear at high PD pulse repetition rate or even in case of reflections and oscillations excited by the fast PD transients in cables and inductive components, as has been discussed previously. As a consequence, the “International Technical Commission (IEC), Technical Committee No. 42: High-Voltage Testing and Measuring Technique” decided the edition of a separate standard on PD measurements. The first edition of “IEC Publication 270” appeared in 1968, where besides the definition of the PD quantity “ apparent charge ” as well as the inception and extinction voltage, several other PD quantities were introduced, such as the repetition rate as well as the power of consecutive PD pulses. Additionally, rules for calibrating PD measuring circuits were specified, and guidelines were attached which supported the identification of typical PD defects under AC test voltage based on oscilloscopic records using either the elliptical or the linear time base to record the characteristic PD pulse trains in a phase-resolved manner.

The second edition of IEC Publication 270, published in 1981, contained more details on the calibration procedure. Additionally, the PD quantity “largest repeatedly occurring PD charge” was specified. Based on this standard, the electrical PD measurement became an indispensable tool for tracing dielectric imperfections in HV apparatus, which might be caused by design failures as well as by a poor assembling work. Therefore, the measurement of partial discharges was increasingly requested with respect to increased quality requirements, which was also forced by the enhancement of the design field strength and, last but not least, by demands concerning the enlargement of the lifetime of HV equipment.

The following treatment is based on the third edition of IEC 60270 published in 2000, which can be considered as an extensive revision of the second edition. The specification covers besides the traditional analogue PD signal processing also the advanced digital acquisition of the captured PD pulses. Moreover, a section has been added, which refers to maintaining the characteristics of PD measuring systems and the associated calibrators, as will be considered more in detail in Sect. 4.3.7.

4.3.4.2 Analogue PD Instruments

A simplified block diagram of analogue PD instruments is shown in Fig. 4.29. The input of the device is commonly equipped with an attenuator to adjust the desired measuring sensitivity as well as a fast over-voltage protection unit to avoid a damage of the instrument in case of an unexpected breakdown of the test object. The desired integration of the captured PD signal is commonly performed by the use of a band-pass amplifier, as mentioned previously. As an alternative, a wide-band amplifier in combination with an electronic integrator can also be used (Lemke 1969), as illustrated in Fig. 4.29. This concept offers the opportunity to record the true shape of the captured PD pulses, which enables flight-of-time measurements used for the localization of the PD site, for instance, in power cables, as will described more in detail in Sect. 4.4. Another benefit of this concept is that pulse-shaped noises can effectively be rejected using various features for gating and windowing, as will be presented in Sect. 4.5.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig29_HTML.gif — Fig. 4.29
Simplified block diagram of an analogue PD instrument

The quasi-peak detector in combination with the reading instrument is intended for measuring the “largest repeatedly occurring PD magnitude” according to IEC 60270:2000. This unit averages the reading, which is particularly useful to weight randomly distributed PD pulse magnitudes, see Fig. 4.30. Another benefit of this unit is that stochastically appearing noise pulses of comparatively low repetition rate are either not indicated or their magnitude is substantially reduced. As specified in IEC 60270:2000, the charging and discharging time constant of the quasi-peak detector should be chosen as $\tau_{1} \le 1\;{\text{ms}}$ and $\tau_{2} \approx 440\;{\text{ms}},$ respectively, to accomplish the pulse train response according to Fig. 4.31. In this context it should be emphasized that this approach ensures more or less reproducible test results under power frequency (50/60 Hz) AC voltages, but not when the test frequency is changed, as usual for quality assurance tests of power transformers and instrument transformers, where the test frequency is occasionally increased up to 400 Hz. This has also to be taken into account when resonance test sets are used for on-site PD tests of power cables, where the test frequency is often varied between 20 Hz and 300 Hz (Rethmeier et al. 2012).

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig30_HTML.gif — Fig. 4.30
Screenshots gained by means of a computerized PD measuring system showing the PD level of a MV power cable termination under AC (50 Hz) voltage (test level 38 kV, recording time 120 s). a Peak values of each individual charge pulse. b Largest repeatedly occurring PD magnitude weighted according to IEC 60270:2000

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig31_HTML.gif — Fig. 4.31
Pulse train response specified in IEC 60270:2000 to measure the largest repeatedly occurring PD magnitude

To evaluate PD test results, it is highly recommended to display the phase-resolved PD patterns (PRPDP), which supports the identification of potential PD defects and enables often the discrimination of disturbing noises from real PD events. For this purpose, either the built-in oscilloscope or even an external connected multichannel oscilloscope as well as a computerized measuring system can be used.

As discussed previously, narrow-band instruments are commonly not capable of measuring the apparent charge of PD pulses in terms of pC. Nevertheless, such devices, which are commonly designed to measure radio interference voltages (RIV) as well as to investigate the electromagnetic capability (EMC) of electronic devices, are nowadays also widely employed for PD measurements, particularly for PD diagnosis tests of HV apparatus under on-site condition due to their excellent noise immunity. Even if such instruments measure the captured PD signal in terms of μV and not in terms of pC, there is no doubt that the PD inception and extinction voltage as well as the change and the trend of the PD activity can well be determined by means of narrow-band instruments.

4.3.4.3 Digital PD Instruments

Due to the recent achievements in micro-electronics and in particular in digital signal processing (DSP), the traditional analogue PD instruments are nowadays increasingly replaced by advanced digital PD measuring systems. The first concept of a computerized PD measuring instrument has been presented by Tanaka and Okamoto (1978). After that time, various solutions for computer-based PD measuring systems have been proposed, for instance, by Kranz (1982), Haller and Gulski (1984), Okamoto and Tanaka (1986), van Brunt (1991), Gulski (1991), Kranz and Krump (1992), Fruth and Gross (1994), Shim (2000), Lemke et al. (2002), Plath et al. (2002).

Currently, two basic measuring principles are frequently employed. The first one is illustrated in Fig. 4.32a, which performes an analogue pre-processing of the captured PD pulses to establish the apparent charge pulses using a band-pass amplifier, as treated previously. Thereafter a digital signal processing is performed, where the A/D conversion requires a comparatively low sampling rate. Another option is the use of a very fast A/D converter to digitize the short PD pulses captured from the test object. The band-pass filtering required for the quasi-integration of the PD pulses is commonly performed by an adjustable digital filter and a numerical integrator. The main advantage of digital PD measuring systems is the ability to acquire, store and visualize the following characteristic PD quantities:

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig32_HTML.gif — Fig. 4.32
Block diagram of digital PD instruments. a Analogue pre-processing of the PD signal followed by an A/D conversion of the apparent charge pulses. b Direct A/D conversion of the wide-band amplified PD pulses

$t_{i}$: instant of PD occurrence
$q_{i}$: pulse charge at instant $t_{i}$
$u_{i}$: test voltage value at instant $t_{i}$
$\phi_{i}$: phase angle at instant $t_{i}$

These parameters ensure not only an evaluation of all PD quantities recommended in IEC 60270:2000, see Sect. 4.1.2, but also an in-depth analysis of the very complex PD occurrence using the following features:

Statistical analysis based on phase-resolved 2D and 3D patterns and pulse sequence pattern to classify and identify PD sources as well as to cancel electromagnetic noises.
Clustering the PD pulses in homogenous families, based on waveform analysis and spectral amplitude diagrams in order to separate the characteristic patterns of PD events originating in different dielectric imperfections.
Localization of PD faults in power cables and GIS using time-domain reflectometry as well as in the windings of rotating machines and power transformers using multichannel techniques.

More details concerning the capabilities of computerized (digital) PD measuring systems, as exemplarily shown in Fig. 4.33, are discussed in the Sects. 4.4–4.8.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig33_HTML.gif — Fig. 4.33
View of a computerized PD measuring system.
*Courtesy* of Doble Lemke

4.3.5 Calibration of PD Measuring Circuits

The aim behind the calibration of PD measuring circuits is to determine the scale factor $S_{f}$ required to measure the apparent charge $q_{a}$ of PD pulses, which is deduced from the reading $M_{p}$ of the PD measuring instrument or even from the signal magnitude displayed on the screen of an oscilloscope or a computer. For this purpose, a well known calibrating charge $q_{0}$ is injected between the terminals of the test object. Provided this causes the reading M₀, the apparent charge can be calculated for using the following relation:

$q_{a} = \frac{{q_{0} }}{{M_{0} }} \cdot M_{p} = S_{f} \cdot M_{p} .$

(4.25)

The calibration procedure is based on the fact that each PD event, which transfers a pulse charge from the PD site to the test object capacitance C_a, causes a transient voltage step $\Delta V_{a}$ across C_a detectable between the terminals of the test object, see Sect. 4.2. An equivalent response appears when a calibrating charge is injected between the terminals of the test object, as illustrated in Fig. 4.34 (Lemke et al. 1996; Lukas et al. 1997).

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig34_HTML.gif — Fig. 4.34
Principle of the calibration procedure used to determine the scale factor of PD measuring systems

Example Assuming a calibrating charge of q₀ = 20 pC, which is injected between the terminals of the test object and causes a deflection of 5.4 divisions on the display of an oscilloscope, which is connected to the output of the PD instrument. Thus, the scale factor becomes S_f = 20 pC/ 5.4 div = 3.7 pC/ div. Performing an actual PD test, where a recorded PD pulse causes a maximum deflection of 8.6 div, the apparent charge amounts q_a = (3.7 pC/ div) × 8.6 div ≈ 32 pC.

In practice the capacitances $C_{01}$ and $C_{02}$ shown in Fig. 4.34 are substituted by only a single calibrating capacitor $C_{0}$ , which is commonly connected between the internal pulse generator of the calibrator and the HV terminal of the test object in order to inject the calibrating charge, see Fig. 4.35.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig35_HTML.gif — Fig. 4.35
PD Calibrator connected to a 20 kV instrument transformer

As a measuring example, consider the oscilloscopic records shown in Fig. 4.36b. Here a step voltage (CH1), which is commonly caused by a PD event, was injected at the HV site of a 110-kV transformer bushing, as illustrated in Fig. 4.36a. This signal appeared differentiated when decoupled from the bushing tap terminal (CH2), which is due to the high-pass filter characteristics following from the series connection of the HV bushing (capacitance being close to 200 pF) and the measuring impedance (equipped with a 50 Ω measuring resistor). Under this condition the characteristic time constant attains nearly 10 ns. This signal was integrated (CH3) by means of a PD measuring instrument to reproduce a signal, which is proportional to the magnitude of the injected step pulse, and could in principle also be caused by a real PD event originating in a transformer under test.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig36_HTML.gif — Fig. 4.36
Set-up to demonstrate the step pulse response of a transformer bushing (a) and characteristic signals injected at the HV terminal (CH1) and decoupled from the bushing tap (CH2) as well as from the output “apparent charge” of the connected PD measuring instrument (CH3) (b)

To ensure reproducible PD test results, the step voltage shape of PD calibrators is specified in the Amendment to IEC 60270:2000 as follows (Fig. 4.37):

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig37_HTML.gif — Fig. 4.37
Step voltage parameters specified for PD calibrators in the Amendment to IEC 60270:2000, published in 2015

$\begin{array}{*{20}l} {{\text{rise}}\;{\text{time:}}} \hfill & {t_{r} \le 60\;{\text{ns}}} \hfill \\ {{\text{time}}\;{\text{to}}\;{\text{steady}}\;{\text{state:}}} \hfill & {t_{s} \le 200\;{\text{ns}}} \hfill \\ {{\text{step}}\;{\text{voltage}}\;{\text{duration:}}} \hfill & {t_{d} \ge 5\;\upmu{\text{s}}} \hfill \\ {{\text{absolute}}\;{\text{voltage}}\;{\text{deviation:}}} \hfill & {\Delta V \le 0.03\;{\text{V}}_{o} } \hfill \\ \end{array}$

To minimize distortions of the step voltage shape, which may occur at comparatively high capacitive load, the calibrating capacitor $C_{0}$ should be chosen not greater than 200 pF. Additionally, the condition $C_{0} \le 0.1\;C_{a}$ should be satisfied to accomplish the complete charge transfer to the test object capacitance C_a via the calibrating capacitance C₀.

4.3.6 Performance Tests of PD Calibrators

To verify the characteristic time and voltage parameters of PD calibrators specified in the Amendment to IEC 60270:2000, a performance test is required. The simplest approach would be the injection of the charge $q_{0}$ created by the calibrator into a measuring capacitor C_m, see Fig. 4.38a. To accomplish the entire charge transfer from the calibrator to C_m, which is required to apply the relation $q_{0} = C_{m} \cdot \Delta V_{m} ,$ the condition $C_{m} \ge 100\;C_{0}$ must be satisfied. If, for instance, the calibrator is equipped with a capacitor of $C_{0} = 100\;{\text{pF}}$ , the measuring capacitor should be chosen as high as $C_{m} \ge 10\;{\text{nF}}$ . Applying a conventional foil capacitor to guarantee a high-temperature stability, however, the front of the step pulse may be distorted drastically on account of the inherent inductive component, as exemplarily shown in Fig. 4.38b. This record refers to a calibrating charge of $q_{0} = 120\;{\text{pC}}$ , which was injected via a 100 pF series capacitor into a 20 nF measuring capacitor. To overcome this crucial problem, an appropriate damping resistor $R_{d}$ should be connected between calibrator and measuring capacitor, which leads to a response comparable to that shown in Fig. 4.38c. Another option to reduce the effective inductance is the use of numerous measuring capacitors connected in parallel.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig38_HTML.gif — Fig. 4.38
Set-up for measuring the calibrating charge (a) and typical oscilloscopic records gained by the use of damping resistors by R_d = 50 Ω (b) and R_d = 390 Ω (c), respectively

The main obstacle of the calibrating circuit shown in Fig. 4.38a is the fact that calibrating charges below about 50 pC can hardly be measured by means of conventional digital oscilloscopes, because a calibrating charge of $q_{0} = 50\;{\text{pC}}$ injected in a 10 nF measuring capacitor causes a voltage jump as low as $\Delta V_{m} = 5\;{\text{mV}},$ i.e. a change in voltage as low as 0.2 mV must be recognized to achieve an appropriate measuring accuracy. To overcome this crucial problem, an electronic integrator could be employed to enhance the measuring sensitivity accordingly (Lemke 1996). A schematic diagram of such a circuit is shown in Fig. 4.39, which ensures the complete charge transfer from the calibrator to the capacitor $C_{m} ,$ even if the capacitance of C_m is reduced 100 times, i.e. from originally 10 nF down to about 100 pF, where the latter equals the value of the calibrating capacitor C₀. As obvious from the oscilloscopic records shown in Fig. 4.39, which refers to a calibrating charge of approx. 3 pC, a sufficient high-voltage step of approx. 30 mV is achieved, which can conviently be measured at desired accuracy, in particular when the oscilloscope is equipped with an averaging tool. In this context it should also be noted that despite the comparatively low integrating capacitance of only 100 pF the decay time constant of the measured signal is sufficiently high (Fig. 4.39c), so that the recorded voltage signal CH2 decreases only marginally during the recording time, which is chosen as 5 μs and corresponds thus to the minimum duration $t_{d}$ of the step voltage response specified in the Amendment to IEC 60270:2000.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig39_HTML.gif — Fig. 4.39
Principle of an electronic integrator for measuring the charge of PD calibrators (a) and typical signals recorded at time scale of 100 ns/div (b) and 1 μs (c), respectively

Another option recommended in IEC 60270:2000 is the injection of the calibrating charge into a measuring resistor $R_{m}$ , as illustrated in Fig. 4.40a. Obviously, the series connection of $C_{0}$ and $R_{m}$ represents a high-pass filter, so that the time-dependent voltage $v_{m} (t)$ appearing across $R_{m}$ must be integrated again to determine the charge $q_{0}$ created by the calibrator. For this purpose, a digital oscilloscope equipped with a feature for numerical integration could be be employed, where the A/D converter should have a vertical resolution not lower than 10 bits at 50 MS/s sampling rate to achieve an adequate resolution of the input signal. Additionally, the analogue bandwidth should be not lower than 50 MHz. Generally, digital oscilloscopes calibrated in compliance with IEC 61083-1:2002 should only be employed.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig40_HTML.gif — Fig. 4.40
Set-up recommended for a numerical integration of the calibrating charge $q_{0}$ injected in a measuring resistor of R_m = 50 Ω (a) and typical oscilloscopic records (b) gained for a calibrating charge of $q_{0} = 150\;{\text{pC}}$

$$ q_{0} $$ — Fig. 4.40
Set-up recommended for a numerical integration of the calibrating charge $q_{0}$ injected in a measuring resistor of R_m = 50 Ω (a) and typical oscilloscopic records (b) gained for a calibrating charge of $q_{0} = 150\;{\text{pC}}$

4.3.7 Maintaining the Characteristics of PD Measuring Systems

The third edition of IEC 60270:2000 recommends the following three levels for maintaining the characteristics of PD measuring facilities composed of the coupling device, the PD measuring instrument, and the PD calibrator including the necessary connection leads:

1.
The routine calibration of the complete PD measuring system connected to the HV test circuit. This should be performed just prior a PD test, where the calibration provides the scale factor S_f of the entire measuring system to be used in the actual PD test. Nowadays, this procedure is mainly used to adjust the reading of the PD measuring instrument to obtain a direct reading of the PD magnitude, i.e. S_f should satisfy preferable values (e.g. 1, 2, 5, 10, 20…). For this routine calibration, there are no major changes as compared to the IEC 270 edited in 1981.
2.
The determination of the specified characteristics of the complete PD measuring system should be performed at least once a year or after major repair.
3.
The calibration of the PD calibrator itself, as presented above.

In general, manufacturers of PD measuring devices have to provide the necessary guidelines required for the verification of the specified technical parameters. Independent on such guidelines, the current (third) edition of IEC 60270:2000 recommends additional test procedures, where the results have to be recorded in a “Record of Performance (RoP)” to be established and maintained by the user. The RoP should include the following information:

Nominal characteristics (identification; operation conditions, measuring range, supply voltage)
Type test results
Routine test results
Performance test results (date and time)
Performance check results (date and time; result: passed/failed: if failed: action taken)

Verifications of PD measuring systems and PD calibrators shall be performed once as acceptance tests. Performance tests should be performed annually or after any major repair, but at least every 5 years. Performance checks have to be performed at least once a year. To maintain the characteristics of PD measuring instruments , the following tests should be performed:

(a)
Type tests are to be done by the manufacturer and shall be performed for one PD measuring system of a series and shall at least include the determination of the following parameters:
- The frequency-dependent transfer impedance Z(f) as well as the lower and upper limit frequencies f₁ and f₂ over a frequency range in which it has dropped to 20 dB from the peak band-pass value;
- The scale factor k to calibrating pulses of at least three different pulse charge magnitudes ranging between 10 and 100% of the full reading at a pulse repetition rate n around 100 s⁻¹. In order to prove the linearity of the PD measuring instrument, the variation of k shall be less than 5%;
- The pulse resolution time T_r by applying calibration pulses of constant magnitude but decreasing time interval between consecutive pulses;
- The pulse train response for pulse repetition rates N ranging between 1 s⁻¹ and >100 s⁻¹.
(b)
Routine tests are to be done by the manufacturer and shall include all tests required in a performance test as listed below. Routine tests shall be performed for each measuring system of a series. If the test results are not available from the manufacturer, the required tests shall be arranged by the user.
(c)
Performance tests shall include the determination of the following parameters:
- The frequency-dependent transfer impedance Z(f) as well as the lower and upper limit frequencies f ₁ and f ₂ over a frequency range in which it has dropped down to 20 dB from the peak band-pass value.
- The linearity of the scale factor k to be verified between 50% of the lowest and 200% of the highest specified PD magnitude. Using calibrating pulses of adjustable magnitude having a repetition rate of approximately n = 100 s⁻¹, the scale factor k shall vary not more than 5%.
(d)
Performance checks shall include the determination of the transfer impedance Z(f) at one frequency selected in the band-pass range in order to verify that the value deviates not more than 10% from that one recorded in the performance test.

To maintain the characteristics of PD calibrators , the following tests should be performed:

(a)
Type tests are to be done by the manufacturer and shall be performed for one PD calibrator of a series. Type tests shall include at least all tests required in a performance test. If results of type tests are not available from the manufacturer, the required tests for verification the technical parameters of PD calibrators shall be arranged by the user.
(b)
Routine tests are to be done by the manufacturer and shall include all tests required in a performance test. Routine tests are to be performed by the manufacturer for each measuring system of a series. If the test results are not available from the manufacturer, the required tests shall be arranged by the user.
(c)
Performance tests shall include the determination of the following parameters:
- The actual magnitude of the pulse charge q₀ for all nominal settings, where a measuring uncertainty within 5% or 1 pC, whichever is greater, is acceptable.
- Rise time t_r of the voltage step U₀, where a measuring uncertainty within 10% is acceptable.
- Pulse repetition frequency N, where a measuring uncertainty within 1% is acceptable.
(d)
Performance checks include the determination of the actual magnitude of the calibrating charge q₀ for all nominal settings, where a measuring uncertainty within 5% or 1 pC, whichever is greater, is accepted.

4.3.8 PD Test Procedure

The main aim behind PD tests according to IEC 60270:2000 is to prove the integrity of insulation systems of HV apparatus and their components. The test procedures applied for quality assurance tests after manufacturing and repair as well as the test voltage levels and the limits of tolerated pulse charge magnitudes deduced from long-term experience are specified for each kind of HV apparatus by the relevant Technical Committee. As the test procedures vary for different HV equipment, only one typical application shall be presented in the following, which refers to a PD test of a single-phase power transformer under induced voltage based on the test circuit sketched in Fig. 4.41.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig41_HTML.gif — Fig. 4.41
Set-up applied for a PD tests of a power transformer under induced voltage using the bushing tap coupling mode

The PD test procedure commonly applied can generally be divided into the following steps:

(a)
Configuration of the HV test circuit:

To minimize the impact of electromagnetic noises, the transformer under test should well be grounded. Moreover, the use of a low-pass filter at the LV site is highly recommended. Using the bushing tap coupling mode, the measuring impedance should be located as close to the test object and connected to the PD measuring instrument not only via the measuring cable but additionally via a parallel connected ground connection lead made of Cu or Al foil. Occasionally shielding electrodes should be arranged on the top electrodes of the bushings to prevent corona discharges, see Figs. 4.41 and 4.42. Moreover, it has to be taken care that the bushing is cleaned and dried to prevent surface discharges. To record the phase-resolved PD patterns, additionally to the PD signal an AC voltage should be recorded as reference, which could be taken from the measuring impedance connected to the bushing tap, as reported in Sect. 4.3.1 and illustrated in Fig. 4.22.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig42_HTML.jpg — Fig. 4.42
Shielding electrodes assembled at the *top* of the bushings of a 500 kV single-phase transformer intended to prevent disturbing corona discharges

(b)
Adjustment of the measuring frequency range :

For tall and inductive test objects, such as power transformers, the frequency content of the detectable PD signal travelling from the PD site to the bushing tap is drastically reduced. Thus, the upper limit frequency $f_{2}$ should be selected preferably below 300 kHz, while the lower limit frequency $f_{1}$ should be adjusted below 100 kHz to ensure a PD signal processing in a wide-band range. Reducing $f_{1}$ substantially below 100 kHz, however, serious disturbances might appear due a possible iron core saturation as well as other core-related noises and even harmonics superimposed on the exciting AC voltage.

The main objective of the calibration procedure is to determine the ratio between calibrating charge $q_{0}$ injected into the top electrode of the bushing and the reading $M_{0}$ of the PD instrument, which provides the scale factor $S_{f}$ , as pointed out in Sect. 4.3.5. The ground connection lead between calibrator and transformer tank should be of low inductance and should thus be kept as short as possible. To prove the linearity of the PD measuring system, calibrating charges of different magnitudes should be injected. If, for instance, the full reading of the PD instrument is obtained for a calibrating charge of 200 pC, it is recommended to reduce thereafter the calibrating charge down to 100 pC and at least to 50 pC. Under this condition the deviation of the reading from that to be expected for a linear operating instrument should be below 10% of the full reading. After the calibration procedure has been finished it has to be taken care that the calibrator is removed from the test object to avoid any damage when the HV test voltage is switched on.

(d)
Actual PD test under HVAC voltage:

The test voltage profile to be applied for quality assurance tests is specified in the relevant apparatus standards. A typical measuring example is shown in Fig. 4.43, which refers to the IEEE standard C57.113: 2010. First, the induced AC test voltage level has to be raised up to approx. 50% of the rated voltage $V_{1}$ to determine the “energized background noise level” in terms of pC. This should not exceed the 50% value of the specified apparent charge, which is being accepted. Thereafter, the test voltage has to be raised up to the specified 1 h test value $V_{2}$ and held constant for few minutes to verify whether there are any PD problems. If not, the test voltage has then to be raised up to the enhancement (withstand) level $V_{3}$ and held constant for 7200 cycles in order to observe the PD trend.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig43_HTML.gif — Fig. 4.43
Profile of the HVAC test voltage recommended in the IEEE Standard C57.113 for PD testing of liquid-filled power transformers and shunt reactors

Note As HVAC test voltages substantially higher than the rated voltage are applied, the exciting frequency must be enhanced accordingly, which can conveniently be realized by the use of HVAC test sets of variable test frequency. For this reason, the duration of the PD test period at enhancement voltage level is expressed in terms of cycles. If, for instance, a test frequency of 120 Hz is applied, 7,200 cycles are equivalent to a test period of 60 s. For more details in this respect see Sect. 3.2.5.

Finally, the applied HVAC test voltage is reduced down to the 1 h test voltage level $V_{2}$ . Under this condition, the apparent charge is recorded at subsequent five-minute intervals, where the recording time can be limited to about 1 min at each interval.

(e)
Evaluation of PD test results:

As stated in the above mentioned IEEE Standard C57.113, the transformer has passed the PD test when the mean value of the maximum PD pulse magnitudes expressed in terms of pC and recorded during the 1 h test interval

is below a specified value in terms of pC,
is within a specified tolerance band,
does not exhibit any steadily rising trend, and
does not suddenly increase during the last 20 min of the 1 h test period.

In this context, it has to be taken into account that during the above mentioned 5-min test intervals sporadic noises may be encountered, for instance, by the switching of cranes. Provided, the test results do not comply with the specified limits, the PD-tested transformer should not warrant immediate rejection but lead to consultation between purchaser and manufacturer to decide further actions.

4.4 PD Fault Localization

To assess the PD severity, besides the apparent charge and the PD pulse repetition rate also the origin of the PD source should be known. This is particularly of importance for high-polymeric power cables, where the insulation may irreversibly be deteriorated by PD events having a magnitude of only few pC. Therefore, the localization of PD failures became a well established method since the 1970s, when high-polymeric power cables where increasingly used in distribution networks (Eager and Bahder 1967, 1969; Lemke 1975, 1979; Beinert 1977; Kadry et al. 1977; Beyer and Borsi 1977). The main benefit of the PD fault localization in power cables is on one hand, that the reason for typical PD defects can be clarified so that the technology for manufacturing such cables can be improved. On the other hand, the entire cable length must not be replaced in case of a single PD failure, but rather that short section containing the PD defect. In this context it should be noted that in case of routine tests most of the recognized PD failures appear at the cable ends, mainly due to a poor assembling of the stress cones required for a field grading.

Using the so-called time-domain reflectometry for the PD fault localization, first the travelling wave velocity $v_{c}$ has to be determined. As a measuring example consider Fig. 4.44a. Here, a calibrating pulse was injected in that cable end where the coupling unit is connected, which is usually referred to as near cable end. As electrically long power cables behave like an electromagnetic waveguide, the pulse injected at the near end travels first at wave velocity $v_{c}$ towards the far or remote cable end, where the signal is reflected, and thereafter towards the near cable end. Consequently, a second pulse is also detectable at the near cable end after a time span t_c required for travelling twice the entire cable length, i.e. 2 ∙ $l_{c}$ . Hence the travelling wave velocity can simply be accounted for using the following relation

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig44_HTML.gif — Fig. 4.44
Principle of PD fault localization in electrically long power cables. a Determination of the travelling wave velocity. b Using the time-domain reflectometry (*TDR*) to determine the PD site

$v_{c} = \frac{{2l_{c} }}{{t_{c}^{.} }}$

(4.26)

Provided, a PD fault is located at distance x_n from the near cable end, the PD pulse would travel from the site of origin into both directions, i.e. directly towards the near cable end within the time span t_n and also towards the remote cable end within the time span t_r, where it is reflected, see 4.44b. Thus, the reflected pulse travels twice the distance $x_{r}$ between PD site and remote cable end and thereafter the distance x_n, which is equal to that travelled by the direct PD pulse. Based on this it can be written:

$x_{r} = 0.5 \cdot v_{c} \cdot t_{r} .$

(4.27)

$l_{c} - x_{r} = 0.5 \cdot v_{c} \left( {t_{c} - t_{r} } \right).$

(4.28)

Nowadays available computerized PD measuring systems are mostly equipped with features for time-domain reflectometry (TDR) to localize PD failures in power cables (Lemke et al. 1996, 2001). The main challenge is, however, to measure the time difference $t_{r}$ between the direct and the reflected PD pulse as accurately as possible, which requires an A/D conversion at sampling rate not lower than 100 MS/s and a signal resolution of 10 bit. To record the complete PD data stream occurring during a single half-cycle of a 50-Hz test voltage, the memory depth should be in the GByte range. The overall bandwidth should cover a frequency range between about 50 kHz and 20 MHz.

Another practical example is shown in Fig. 4.45, which refers to an on-site PD test of a XLPE power cable subjected to a so-called DAC test voltage (Lemke et al. 2001). Here the localization of the PD faults was performed by means of a computerized PD measuring system, which covers the following steps:

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig45_HTML.gif — Fig. 4.45
Screenshots of a computer-based PD fault localization system (explanation in the text)

(a)
Inserting the cable data: Besides the fundamental cable parameters (manufacturer, type and insulation of the cable, rated voltage, operation voltage, recently performed tests, etc.) this should include the test voltage parameters to be applied (test voltage levels, number of shots at each test voltage level), also the cable length and the positions of the accessories (joints and terminations), which are especially of interest to localize the PD defects as accurate as possible.
(b)
Calibration : This includes the determination of both the measuring sensitivity and the travelling wave velocity of the PD pulses. As shown in Fig. 4.45a, besides the calibrating pulse injected at the near cable end, several pulse reflections might occur, where only the first one is of interest. Therefore, this signal is zoomed, as obvious from Fig. 4.45b, and the cursors are set accordingly by the computer software to determine the time interval $t_{c}$ and thus the travelling wave velocity $v_{c}$ based on Eq. 4.26 as precisely as possible.
(c)
PD measurement : Recording of the consecutive PD pulses occurring within a pre-selected time interval (Fig. 4.45c) and evaluation the of the PD pulse magnitudes. For this purpose each PD pulse magnitude is initially indicated in terms of Volts, and based on this the pulse charge appearing at each test voltage application is calculated for by the software and stored in the computer memory to perform a statistical analysis of the data stream.
(d)
PD fault localization : To apply the time-domain reflectometry (TDR), only those PD pulses showing typical reflections within the time interval t_r ≤ 2 t_c are extracted. Thereafter, these pulses are zoomed and the cursors are set accordingly to measure the time interval between each direct and the associated reflected pulse, see Fig. 4.45d. Based on this, the distance between PD source and either the near or the remote cable end is determined. This procedure is repeated several times in order to perform an averaging and thus to enhance the measuring accuracy. Occasionally, a digital filtering of the captured signal may be performed to minimize the impact of radio interference voltages on the test results.
(e)
PD mapping : Displaying all determined fault positions and the associated pulse charge magnitudes along the cable length. A typical measuring example for this is shown in Fig. 4.45e, which refers to a 20 kV XLPE cable of 480 m in length having two potential PD defects. The first one is located at 3 m from the near end and the second one at 452 m. Usually, the PD fault localization is performed automatically by the software. Only if a repair of an identified joint or termination is decided, the manual feature should additionally be applied to prove the validity of the automatically located PD sites. For more information in this respect see also the Sects. 7.1.3 and 10.2.2.2.

For some HV equipment other than power cables, the time of arrival measurement can also be applied to identify defective components of a three-phase system. An example for this is shown in Fig. 4.46, which refers to a power transformer. Here the wide-band PD signal was decoupled simultaneously from the bushing taps of all three phases using measuring impedances of 20 MHz bandwidth. The oscilloscopic records reveal that a potential PD source appears in the phase “T”, because the transient PD signal appeared first and showed a substantial higher magnitude if compared with those signals decoupled from the other both phases “R” and “S”, respectively. This was confirmed by a visual inspection.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig46_HTML.gif — Fig. 4.46
PD signals decoupled simultaneously from the three phases (R, S, T) of a power transformer using the bushing tap coupling mode

The synchronous three-phase PD measurement has also been proven as a feasible tool for in-service diagnostics of power cable terminations, as shown in Fig. 4.47. Recording only the time-dependent PD level, it could be supposed that PDs appear in all three terminations. However, displaying the phase-resolved PD pulses versus the recording time, the defective termination could clearly be identified in the phase “red” due to the highest pulse magnitudes. Moreover, the PD pulses decoupled from all three terminations appeared at phase angles equal for all three terminations, which means that the captured signal was radiated from only one single PD source. This conclusion was finally confirmed by a visual inspection after the identified termination has been de-assembled. Replacing this cable termination by a new one, no any critical PD level was recognized.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig47_HTML.gif — Fig. 4.47
Characteristic PD signatures captured from the cable terminations connected to a three-phase gas-insulated switchgear

Another approach to distinguish between different PD sources is the presentation of typical clusters in a 3-phase amplitude relation diagram, which is based on a synchronous multichannel PD measurement (Emanuel et al. 2002). An enhancement of this method is the presentation of so-called three-center-frequency relation diagrams where three different frequencies selected from the complete spectrum of a single PD pulse are evaluated and displayed on the computer screen. This feature provides not only valuable information on the discharge nature itself but can also be used to localize the origin of PD defects (Rethmeier 2009). For more details in this respect see Sect. 4.6.

Another promising tool proposed for the localization of potential PD defects in HV equipment is the so-called pulse waveform analysis. This is based on the extraction of a set of PD pulse parameters, such as the rise time and the decay time as well as the PD pulse width (Montenari 2009). Displaying the characteristic clusters like star diagrams, multiple PD failures can also be recognized, as presented also in Sect. 4.6.

In this context, it should be noted that besides the above-described electrical methods, also the acoustic emission (AE) technique is widely used, in particular to localize PD defects in metal-encapsulated HV apparatus, such as gas-insulated switchgears (GIS) and gas-insulated lines (GIL) as well as large power transformers. The combination of both the electrical and acoustic method can also be very effective, for instance, to enhance the signal-to-noise ratio. For more information in this respect see Sect. 4.8.

Even if the localization of PD faults is nowadays performed by means of advanced computerized PD measuring systems, it should not be overlooked that commercially available digital oscilloscopes can also conveniently be employed for this purpose. In this context it should also be emphasized that a great deal of practical experience is required to decide if a HV equipment showing a high PD activity should really be taken out of order or even kept in service and PD monitored permanently to recognize a sudden increase of the PD activity and thus to prevent an unexpected breakdown. For more information in this respect see Sect. 10.3.

4.5 Noise Reduction

4.5.1 Sources and Signatures of Noises

The PD signal level to be detected is often in the mV range and below and may thus be disturbed by electromagnetic noises in the measuring surroundings. To discriminate such interferences from the PD signal, the sources and signatures of typical noises must be known. Depending on the mode of propagation it is generally distinguished between radiated noises and conducted noises .

Noises radiated from radio broadcast stations appear usually modulated and enter the test area via the electromagnetic field, where the HV electrodes and measuring loops of the PD test circuit act like antennas. Moreover, high-frequency transients associated with corona discharges igniting in the vicinity of the test area at sharp edges and protrusions on the surface of HV electrodes can also be classified as radiated noises.

Electromagnetic noises may also originate in the mains or even in the HV test facility itself du to switching processes, which appear thus stochastically or even periodically. These are transmitted via conductors to the PD test area and hence classified as conductive noises. This refers also to a sparking between metallic parts on floating potential, for instance, due to a poor grounding. Screenshots of typical conductive noises, often encountered in PD test laboratories, are displayed in Figs. 4.48 and 4.49.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig48_HTML.gif — Fig. 4.48
Signatures of stochastically appearing pulse-shaped noises. a Maintenance work (drilling worker), b Car starting nearby, c Switching of a crane in the test lab

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig49_HTML.gif — Fig. 4.49
Signatures of periodically appearing pulse-shaped noises. a Protrusion at the surface of a HV shielding electrode. b Sharp edge of a metallic structure on ground potential. c Sparking between metallic parts on floating potential. d Frequency converter feeding a resonant test set of variable frequency. e Defective xenon lamp in the control room

4.5.2 Noise Reduction Tools

To minimize the impact of radiated noises it is a common practice to erect electromagnetically well-shielded test laboratories , as described in Sect. 9.2.2, where the fundamental laws of HF technology have to be taken into consideration. Particularly wire loops acting as inductive antennas should be kept as low as possible in cross-section to minimize the induction of interferring voltages on account of radiated noises. Moreover, the ground connection leads should also be of low inductance which is best accomplished by using Cu or Al foil.

If PD test laboratories are not carefully shielded against radiated electromagnetic noises, it could be helpful under certain conditions to use the balanced bridge circuit according to Fig. 4.23c. Practical experiences revealed that for comparatively small test objects, such as instrument transformers and bushings, the signal-to-noise ratio can be enhanced by a factor up to 10. However, for tall test objects, such as power transformers, this method is commonly not effective.

An option of the bridge method is the pulse polarity discrimination , originally proposed by Black (1975), where a balance procedure is not required. As the PD pulses decoupled from both bridge branches appear at opposite polarity, these can conveniently be discriminated from radiated electromagnetic noises because these appear unipolar. For tall test objects, however, this method is commonly not applicable because the PD pulses transmitted via both bridge branches might excite oscillations, so that the true polarity is lost.

From a theoretical point of view, the signal-to-noise ratio can substantially be enhanced when the PD signal processing is performed in a higher frequency range, i.e. at frequencies substantially greater than the 1 MHz limit, as specified in the Amendment to IEC 60270:20060270. This concept has originally been adopted by Lemke (1968), where the PD pulses where first amplified by means of a 20 MHz wide-band amplifier and thereafter integrated by means of an electronic integrator in order to establish the pulse charge. A comparison of this alternative concept with the classical method is shown in Fig. 4.50, where the oscilloscopic records refer to 20 pC calibrating pulses (CH1) injected in a 20 kV XLPE cable. As can be seen, the charge pulse gained on account of a low-pass filtering (f₂ = 600 kHz) cannot clearly be identified (CH2, green trace), while the charge pulse gained by an initial wide-band amplification (f₂ = 20 MHz) and a following electronic integration is not significantly interfered (CH3, pink trace).

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig50_HTML.gif — Fig. 4.50
Oscilloscopic screenshots gained for a 20-pC calibrating pulse injected in a XLPE cable: CH1: Input signal captured from a wide-band measuring impedance; CH2: Signal after conventional band-pass filtering; CH3: Signal after non-conventional wide-band amplification $\left( {f_{2} = 20\;{\text{MHz}}} \right)$ followed by an electronic integration

$$ \left( {f_{2} = 20\;{\text{MHz}}} \right) $$ — Fig. 4.50
Oscilloscopic screenshots gained for a 20-pC calibrating pulse injected in a XLPE cable: CH1: Input signal captured from a wide-band measuring impedance; CH2: Signal after conventional band-pass filtering; CH3: Signal after non-conventional wide-band amplification $\left( {f_{2} = 20\;{\text{MHz}}} \right)$ followed by an electronic integration

The basic concept of the above mentioned non-conventional PD measuring instrument equipped with various tools for noise rejection is illustrated in Fig. 4.51 (Lemke 1979; Hauschild et al. 1981; Lemke 1981), where the operation principle covers the following features:

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig51_HTML.gif — Fig. 4.51
Block diagram of a non-conventional wide-band PD measuring system equipped with various de-noising tools, as described in the text

Wide-band amplification of the PD signal captured from the measuring impedance, where a bandwidth up to about 20 MHz seems reasonable.
Automatic gating of pulse-shaped noises appearing periodically and even stochastically. To receive the noisy signals for triggering the gating unit, the rod or loop antennas should be installed as close as possible to the supposed noise sources.
Canceling of radio interference voltages. This is achieved by adjusting the threshold level for passing the RIV rejection unit slightly above the noise level, which is controlled automatically.
Electronic integration of the de-noised PD signal to measure the pulse charge of the wide-band amplified PD signal.

Typical oscilloscopic records illustrating the operation principles of various noise reduction tools are presented in Fig. 4.52. The screenshots displayed in Figs. 4.52a, b refer to the rejection of radio frequency (RF) noises, as already discussed previously based on Fig. 4.50. In this context it should be mentioned that the above mentioned noise canceling method, which is based on a wide-band pre-amplification of the captured PD pulses followed by an electronic integration, was also used for the design of a hand-held, battery-powered PD probe intended for on-site PD diagnostics of HV apparatus in service (Lemke 1985, 1988, 1991). Pulse-shaped noises appearing periodically or even stochastically can also effectively be canceled using a windowing and gating unit, which is triggered by the noisy pulses itself captured by means of an antenna, see Figs. 4.52c, d.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig52_HTML.gif — Fig. 4.52
Oscilloscopic records gained for a wide-band PD measuring system equipped with an electronic integrator and various tools for de-noising interfered PD signals (description in the text.)

Measuring errors due to the superposition of reflected on direct pulses as consequence of PD events in power cables can also be prevented, as illustrated in Figs. 4.52e, f. Here the superimposed reflected pulse recorded in Fig. 4.46e is canceled by means of an electronic reflection suppressor, as recommended in IEC 60885-3:2003. For this purpose a gating unit is triggered by the direct (first arriving) PD pulse, which closes the gate and prevents thus a passing of the reflected (second) PD pulse. This tool operates reliable for double pulses having distances greater than 0.2 µs (Lemke 1979, 1981).

Even if today’s digital PD measuring systems are commonly equipped with various de-noising software tools, it should not be overlooked that the “windowing” and “gating” hardware, originally developed for the traditional analogue PD signal processing, is also applicable for digital PD instruments, where the noise-canceling is performed just prior the analog/digital conversion (Lemke 1996; Lemke and Strehl 1999). A practical example for this is shown in Fig. 4.53, which refers to PD tests of a defective instrument transformer of 110 kV rated voltage. Applying an AC test voltage of variable frequency, the phase-resolved PD patterns yields the superposition of two characteristic noise signatures, see Fig. 4.53a. Here the randomly distributed dots, which are scattering in magnitude and phase angle, are caused by pulse-shaped noises. As can be seen, these are not correlated with the applied 92-Hz test frequency, but apparently with the frequency of the mains, which is being 60 Hz. To cancel this noisy signal, an inductive sensor has been attached close to the power supply of the HV test facility to trigger the gating unit, which is implemented in the computerized PD measuring system. Under this condition, the randomly distributed pulses could entirely be eliminated, see Fig. 4.53b. Moreover, an additional inductive sensor has been installed close to the frequency converter in order to capture the switching pulses from the IGBTs used for triggering a second gating unit and thus to reject also the heavy noises originating in the frequency converter, see Fig. 4.53c.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig53_HTML.gif — Fig. 4.53
Phase-resolved PD pattern of a defective 110-kV instrument transformer interfered by pulse-shaped noises originating in the mains (60 Hz) and in the frequency converter (92 Hz) of the applied resonant test set. a PD signatures without noise canceling. b Gating of noisy pulses originating in the mains. c Additional gating of the noisy pulses originating in the IGBTs of the frequency converter

Note In practical tests using ACRF test systems (see Sect. 3.2.3) the noisy pulses caused by the frequency converter appear often at stable phase angle and can thus simply by identified Under this condition it is not absolutely necessary to perform a noise pulse gating, which might lead to an information loss on the PD occurrence

Different to the above presented measuring examples, the noisy pulses shown in Fig. 4.53 b can also be eliminated off-line, i.e. after the actual PD measurement has been finished. For this purpose the software tool “ windowing ” has been developed. As an example consider Fig. 4.54 which refers to the classical 2D display mode and the more sophisticated 3D display mode, as well. In this context it should be noted that most of the nowadays available digital PD measuring systems are equipped with software packages to discriminate pulse-shaped noises from real PD events, while hardware tools also suitable for this purpose is only rarely implemented in computerized PD instruments.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig54_HTML.gif — Fig. 4.54
Screenshots of phase-resolved PD patterns and synchronously appearing pulse noises originating in the frequency converter (a) and de-noising the PD patterns by gating (b)

Due to the recent advancements in digital signal processing, very promising de-noising software tools have been developed, mostly based on a cluster separation approach. An example for this is the establishment of star diagrams using synchronous three-phase PD measurements (Plath 2002; Kaufhold et al. 2006), as exemplarily shown in Fig. 4.55, which refers to a PD test of a power transformer. Here, external noises due to corona discharges show almost equal magnitudes, which were detected in all three phases and indicated by the three blue clusters in Fig. 4.55. However, PD pulses originating in the “Yellow” phase created a different colored cluster. Using the classical phase-resolved PD pattern (PRPDP) recognition, a potential PD defect was identified in this phase. In principle, the vectors obtained by the three-phase PD measurement could also be added to establish a three-phase amplitude relation diagram. Under this condition, only one single noise cluster would be established. This is located close to the center point, whereas PDs in the single phase appear outside the center point, which leads to the conclusion that disturbing cross-talking phenomena can be neglected.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig55_HTML.gif — Fig. 4.55
Three-phase star diagram showing three typical clusters (blue) for each phase due to external noises as well as a single cluster (colored), indicating that a potential PD defect is located in phase “Yellow” of the investigated power transformer (Plath 2002)

Another approach based on the cluster separation is the decomposition of the acquired PD pulse waveforms. For this purpose characteristic PD pulse parameters are used in either the frequency or the time domain, such as the rise and decay time as well as the pulse width (Cavallini et al. 2002; Rethmeier et al. 2008). A typical measuring example is illustrated in Fig. 4.56, which refers to a defective power cable termination, where the PD measurement was carried out in a poor-screened test laboratory. After acquiring the complete PD data stream captured from the test object, the strongly interfered phase-resolved PD pattern shown in Fig. 4.56a was obtained. Analysing the pulse shapes in the time domain according to the test series shown in Fig. 4.56b, the characteristic PD patterns could clearly be separated from the noise signatures, see Fig. 4.56c. In principle, this approach is also applicable for multi-source PD separation (Plath 2005).

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig56_HTML.gif — Fig. 4.56
De-noising of PD signals using cluster separation according to Cavallini and Montenari (2007). a Acquired data stream captured during a testing time of 10 min. b Waveform analysis. c Cluster separation

In this context it should be emphasized that advanced de-noising tools can successfully be applied only by experienced test engineers, who must be familiar not only with the fundamentals of PD measurements but also with the operation principle as well as the capabilities and obstacles of the de-noising tools adopted. Even if sophisticated de-noising software is often implemented in advanced computerized PD measuring systems, it should not be overlooked that electromagnetic interferences can often simply be discriminated from PD pulses by means of multichannel digital oscilloscopes and thus conveniently be canceled using classical analogue features, such as windowing and gating, as discussed previously.

4.6 Visualization of PD Events

The main aim behind the visualization of phase-resolved PD patterns (PRPDP) by means of oscilloscopes or computerized PD measuring systems is the recognition and identification of typical PD sources. The first tool applied for this purpose was the electron-beam tube, also known as Braun tube, which has been employed by Lloyd and Starr already in 1928. Connecting the AC test voltage signal to the horizontal deflection plates and bridging the vertical deflection plates by a measuring capacitance, so-called Lissajous figures could be recorded, as exemplarily shown in Fig. 4.57a. In this context, it seems worth to notice that this circuit represents in principle an integrating bridge, which enables the measurement of the power losses of discharges based on the so-called parallelogram method, which has first been used by Dakin and Malinaric in 1960.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig57_HTML.gif — Fig. 4.57
So-called Lissajous figure technique used since the 1960 s to display the charge of PD pulses occurring within a single cycle of the applied AC test voltage. a Voltage appearing across the measuring capacitance of an integrating bridge. b Traditional elliptical display mode

When the first PD detectors were available (Arman and Starr 1936; Mole 1954), the classical Lissajous figure technique has also been modified in order to display the pulse charge trains superimposed on an elliptical loop, where the time base covers a single AC cycle, see Fig. 4.57b. Later the use of a linear time-base became a common practice in order to record the pase-resolved PD patterns (PRPDP), as exemplarily shown in Fig. 4.58. This presentation refers to a needle-plane test sample, which is often used for fundamental PD studies, in particular to classify typical PD defects.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig58_HTML.gif — Fig. 4.58
Oscilloscopic screenshots of phase-resolved PD patterns gained for needle-to-plane test samples under power frequency (50 Hz) AC voltage. a Discharges in air at inception voltage. b Discharges in air at test level substantially above inception voltage. c Cavity discharges in XLPE at inception voltage. d Surface discharges at inception voltage

Further essential steps in the evaluation of PD events where achieved in 1969, when Bartnikas and Levi presented a pulse-height analyser for PD rate measurements, and in 1978, when Tanaka and Okamoto presented the first minicomputer-based PD measuring system. At the very beginning of computerized PD measurements, PD pulse trains appearing within individual AC cycles have often been displayed like waterfall diagrams, see Fig. 4.59.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig59_HTML.gif — Fig. 4.59
PD signatures of defective cable accessories displayed like waterfall diagrams

Today’s digital PD measuring systems acquire and store the vector $\left[ {q_{i} ;u_{i} ;t_{i} ;\varphi_{i} } \right]$ for each captured PD pulse. Here are:

$q_{i}$: the pulse charge of the individual PD current pulse
$u_{i}$: the instantaneous value of the applied test voltage
$t_{i}$: the instant of PD occurrence
$\varphi_{i}$: the phase angle at instant of PD occurrence

Based on this, the cumulative (integral) phase-resolved PD pattern is commonly displayed, as illustrated in Fig. 4.60. Since the 1980s, this display mode is widely used for the so-called phase-resolved PD pattern recognition (Kranz and Krump 1988; Ward 1992; Fruth and Gross 1994). As the PD data stream acquired in the real-time mode can completely be stored in the computer memory, these can be recalled again and visualized in a so-called replay-mode like real PD events, even after the actual PD test has been finished, as exemplarily shown in Fig. 4.61 (Lemke et al. 1996; Lemke and Strehl 1999).

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig60_HTML.gif — Fig. 4.60
Principle for displaying cumulative phase-resolved PD patterns

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig61_HTML.gif — Fig. 4.61
Using the replay-mode to visualize phase-resolved PD patterns originally acquired and stored in the computer memory in the course of development tests of stator bars of a hydro-generator visualized by the use of the replay-mode. a Long-term PD test under AC voltage (test level 18 kV, test duration 10 min, which equals 30,000 AC cycles). b Short-term PD test under enhanced AC voltage (test level 24 kV, test duration 2 min, which equals 6000 cycles)

Using the replay-mode besides the traditional two-dimensional graphs shown in Fig. 4.61, the 3D presentation is frequently used, which is in particular useful to discriminate electromagnetic interferences from real PD events, as has already been discussed based on the Figs. 4.48 and 4.49. Moreover, the phase-resolved PD patterns can be displayed in a traditional manner using either the elliptical or linear time base, as shown in Fig. 4.62.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig62_HTML.gif — Fig. 4.62
Using the replay-mode to visualize phase-resolved PD pulses occurring within a single cycle of the applied 50-Hz AC test voltage (see cursor position obvious from the test voltage profile). a Elliptical time base. b Linear time base

Another approach sometimes adopted for PD pattern recognition is the analysing of PD pulse sequences, as proposed by Hoof and Patsch (1994). The algorithm is based on the evaluation of the voltage differences measured between three subsequent PD pulses, as illustrated in Fig. 4.63a. That means, the voltage difference $\Delta V_{n - 1}$ between the reference pulse “n” and the previous occurring pulse “n − 1”, as well as the associated voltage difference $\Delta V_{n}$ between the reference pulse “n” and the following pulse “n + 1”, is plotted in a graph displaying the value pairs $\Delta V_{n}$ versus $\Delta V_{n - 1}$ . A measuring example for this is shown in Fig. 4.63b which refers to slot discharges detected in a rotating machine.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig63_HTML.gif — Fig. 4.63
Principle of PD pulse sequence analysis (a) and measuring example (b) gained for cavity discharges in a power cable termination

To classify typical discharge sources, Tanaka and Okamoto proposed (1986) the establishment of PD fingerprints based on the following statistical operators:

standard deviation,
skewness,
kurtosis and
cross-correlation.

Later the feasibility of such statistical operators to establish PD fingerprints and to identify and classify typical PD sources, has been confirmed by Gulski and his co-workers (Gulski 1991). To create characteristic PD fingerprints, the following statistical parameters shown in Fig. 4.64 are commonly displayed:

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig64_HTML.gif — Fig. 4.64
PD fingerprint of a discharge source recognized in a HV cable termination in service

H_n (phi): Number of PD pulses occurring within each phase window versus the phase angle,
H_q (phi)_peak: Peak values of PD pulses occurring within each phase window versus the phase angle,
H_q (phi)_mean: Mean values of PD pulses occurring within each phase window versus the phase angle. This quantity is deduced from the total charge amount within each phase window divided by the pulse number occurring in this phase window.

Based on PD fingerprints established for various types of failures and stored in the computer memory, these can be compared with each other, where the test conditions have also to be taken into consideration. Practical experiences revealed that graphs according to Fig. 4.65 provide a valuable tool for the insulation condition assessment and thus for maintenance decisions.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig65_HTML.gif — Fig. 4.65
Comparison of the PD fingerprint gained for stator bar insulation with earlier established PD fingerprints gained for various test samples

4.7 PD Detection in the VHF/UHF Range

4.7.1 General

Performing PD measurements in compliance with IEC 60270:2000, the upper cut-off frequency has to be limited below 1 MHz, as recommended in the Amendment to this standard and discussed more in detail in Sect. 4.3. However, under this condition the signal magnitude is substantially attenuated, so that well shielded test laboratories are required to perform sensitive PD tests. Obviously, the signal-to-noise ratio can significantly be enhanced when the PD signal is captured in the VHF/UHF range. This non-conventional method has first been employed for quality assurance tests of gas-insulated substations by Fujimoto and Boggs (1981). Later the benefits of this approach has successfully been proven for on-site PD diagnosis tests of HV/EHV cable accessories (Pommerenke et al. 1995) and even for PD diagnostics of power transformers under on-site condition (Judd et al. 2002). Moreover, the capability and limits of the non-conventional PD detection in the VHF/UHF range has extensively been investigated by various CIGRE Working Groups. The main issues of these studies are summarized in the Technical Brochures No. 444 (2010) and No. 502 (2012). Based on these publications, the IEC 62478:2015 provides valuable recommendations for the design of VHF/UHF PD measuring systems including the sensors required to capture the PD transients radiated from the test object.

Due to the extremely wide frequency spectrum of PD pulses, which covers the ranges of radio frequency (RF: 3–30 MHz), very high frequency (VHF: 30–300 MHz), and ultra-high frequency (UHF: 300–3000 MHz), various kinds of PD couplers have been developed in the past, which are commonly classified as capacitive, inductive and electromagnetic sensors, as will briefly be presented in the following.

4.7.2 Design of PD Couplers

4.7.2.1 Capacitive PD Couplers

To capture the PD signal from the terminals of the test object, the classical coupling device recommended in IEC 60270:2000 provides a HV coupling capacitor connected in series with a measuring impedance. The upper cut-off frequency of such a coupling device (Fig. 4.66a) is commonly limited below 10 MHz, which is equivalent to a rise time close to 30 ns. To increase the upper limit frequency, it is necessary to decrease the value of the coupling capacitor, because this is associated with a reduction in the internal inductance, which determines the achievable upper cut-off frequency (Fig. 4.66b). Consequently, the highest measuring frequency is achievable by means of capacitive sensors providing a simple metallic disc. The feasibility of this simple approach has initially been proven for the detection and pinpointing of PD sources by means of a hand-held PD probe, as shown in Fig. 4.66c and described also in Sect. 10.4.4 (Fig. 10.43). Such kinds of capacitive sensors, often referred to as C-sensors, receive the electric field component of electromagnetic PD transients.

Figure 4.67 shows a sketch of a coaxial C-sensor designed for PD detection in power cable joints. Here a section of the outer semiconductive layer coating the cable insulation is removed to receive the electric field component, which is radiated from the inner cable conductor due to travelling waves excited by PD events. The achievable measuring sensitivity is mainly governed by the effective capacitance $C_{s}$ between sensor electrode and inner cable conductor and lies in the pC range.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig66_HTML.gif — Fig. 4.66
Step voltage response measured for capacitive PD couplers. a High-capacitive coupling capacitor designed according to IEC 60270:2000 (capacitance 2000 pF, rated voltage 24 kV). b Low-capacitive coupling device (capacitance 50pF, rated voltage 12 kV). c Disc-shaped C-sensors intended for PD probing (capacitance <5 pF)

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig67_HTML.gif — Fig. 4.67
Sketch of a capacitive PD coupler attached to a power cable

Example Consider a polyethylene-insulated power cable of dielectric permittivity of ε_r = 2.2. Assuming a ratio between outer and inner cable conductor of r_a/r_i = e ≈ 2.7, one gets for a coaxial C-sensor of length l_a = 100 mm the following approximation:

$C_{s} \approx 2 \cdot \pi \cdot \varepsilon_{0} \cdot \varepsilon_{r} \cdot l_{a} \approx 12\;{\text{pF}}$

Provided the received PD signal is transmitted via a measuring cable matched by its characteristic impedance Z_m, the peak voltage V_p appearing across Z_m and thus at the input of the peak detector can roughly be accounted for using the following approach:

$V_{p} \approx C_{s} \cdot Z_{c} \cdot Z_{m} \cdot \frac{{I_{p} }}{{t_{r} }}.$

Here are I_p and t_r the peak value and the rise time of the PD pulse current, respectively, and Z_c is the characteristic impedance of the power cable. Assuming, for instance, a cavity discharge creates a current pulse of rise time t_r = 1 ns and peak value I_p = 1 mA, one gets for the above-introduced circuit parameters (C_s = 12 pF, Z_c = 30 Ω, Z_m = 50 Ω) a detectable peak voltage of V_p = 18 mV, which is well measurable by means of digital oscilloscopes.

4.7.2.2 Inductive PD Couplers

The operation principle of inductive PD couplers, commonly referred to as L-sensors or or even “yoke coils”, is comparable with that of high-frequency pulse transformers (HFCT) where the primary coil is formed by a conductor, which belongs to the test object, i.e. the turn number is n = 1. To capture the complete magnetic flux surrounding the primary conductor, the windings of the secondary coil are usually wounded around a high-permeable ferrite core. Under this condition, the transient PD current i_p (t) through the primary conductor induces a transient voltage v_p(t) in the secondary coil (Fig. 4.68). In this context the similarity to Rogowski coils should be underlined. The only difference is that the windings of the Rogowski coil are not wound around a permeable core (see Sect. 7.5.2).

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig68_HTML.gif — Fig. 4.68
L-sensor attached to a power cable termination

A simple PD coupling unit is shown in Fig. 4.68, where a L-sensor is attached around the ground connection lead of a power cable terminal. To characterize the dynamic behaviour, it is recommended to determine the step current response in the time domain according to Fig. 4.69. Comparing the oscilloscopic records shown in Figs. 4.69c and d it can be concluded that the transmitted pulse length is drastically reduced by decreasing the turn number of the secondary coil, i.e. from originally n = 10 down to n = 1, as has to be expected. From this follows also that by means of classical L-sensors a pulse length shorter than 1 μs is hardly achievable. Thus only PD signals in the RF range (3–30 MHz) but not in the VHF/UHF (>30 MHz) are transmitted.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig69_HTML.gif — Fig. 4.69
Set-up for measuring the step current response of inductive PD sensors (a, b) and oscilloscopic records gained for pulse transformers having n = 10 windings (c) resp. n = 1 winding (d)

4.7.2.3 Electromagnetic PD Couplers

The operation principle of electromagnetic (EM) PD couplers is in principle comparable to that of antennas operating in the near-field region. That means the output signal is determined by both vectors representing the electric field component $\overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\rightharpoonup}$}} {E}$ and the magnetic field component $\overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\rightharpoonup}$}} {H}$ , as given by the Maxwell equations:

${\text{rot}}\,\overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\rightharpoonup}$}} {H} = \varepsilon \cdot \frac{{\delta \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\rightharpoonup}$}} {E} }}{\delta t},\quad {\text{rot}}\,\overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\rightharpoonup}$}} {E} = - \mu \cdot \frac{{\delta \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\rightharpoonup}$}} {H} }}{{{\text{d}}t}}.$

(4.29)

Depending on the geometric configuration of the test object, various kinds of EM sensors are employed to detect PD signals in the VHF/UHF range, such as rod, disc or conical antennas. The latter are commonly used for PD diagnostics of gas-insulated substations (Pearson et al. 1991). A sketch of a conical UHF PD sensor installed in a flange of a GIS compartment is shown in Fig. 4.70. Generally such antennas are capable of receiving PD signals of frequency content up to about 1.5 GHz. The upper cut-off frequency is inversely proportional to the characteristic time constant, which follows from the inevitable stray capacitance between sensor electrode and ground flange if multiplied by the characteristic impedance of the connected measuring cable (Meinke and Gundlach 1968; King 1983; Küpfmüller 1984).

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig70_HTML.gif — Fig. 4.70
Sketch of an UHF sensor installed in a GIS flange

To detect PD faults in power cable accessories, the feasibility of so-called directional coupler sensors (DCS), operating in the frequency range between 2 and 500 MHz, has also successfully been proven (Pommerenke et al. 1995). The design principle is comparable to that of capacitive sensors. The only difference is that the PD signal is captured from both sensor ends, which are commonly referred to as “ports”. Installing a pair of DCS at both joint sides, the PD signal originating inside the joint can be discriminated from noise and even PD signals originating in the power cables connected to both sides. This is because a PD event inside the joint causes pulses at the ports “B” and “C” whose magnitudes are significantly higher than those occurring at the ports “A” and “D” (Fig. 4.71). Another benefit is that one single sensor can be used to calibrate the other one. Practical experience revealed that even under noisy on-site condition a measuring sensitivity in the pC range is achievable.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig71_HTML.gif — Fig. 4.71
Operation principle of a pair of directional coupler sensors (*DCS*) installed at both sides of a power cable joint

To capture the PD signal from grounding leads of the test object, high-frequency current transformers (HFCTs) can advantageously be used, as discussed previously. One obstacle of such kinds of PD couplers is, however, that the measuring frequency is commonly limited to the RF range (3–30 MHz). A promising alternative is the use of so-called pulse transformers, which are based on the concept of transmission line inverters, as illustrated in Fig. 4.72 (Lemke et al. 2003). Here, a coaxial cable of few cm in length is used, where the inner conductor is terminated at the output with its characteristic impedance, while the outer conductor is terminated at the input. Moreover, the inner and outer conductor are grounded at the input and output, respectively. Under this condition both the magnetic and electric field components are transmitted at comparatively low attenuation up to the UHF-range (Lewis 1959), where the pulse polarity appears inverted. The feasibility of PD couplers based on this concept have successfully been proven in practice, especially for on-site PD monitoring of cable joints and terminations (Fig. 4.73).

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig72_HTML.gif — Fig. 4.72
UHF PD coupler based on the concept of transmission line inverters. a Equivalent circuit. b Technical design. c PD pulse response in the time domain recorded at time base of 1 ns/div. d Transfer function in the frequency domain

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig73_HTML.gif — Fig. 4.73
UHF-PD coupler attached to a GIS-cable termination. a Flexible PD sensor used for periodical PD monitoring. b Fixed PD sensor used for continuous PD monitoring

4.7.3 Basic Principles of PD Detection in the VHF/UHF Range

As discussed previously in Sect. 4.7.1, the main benefit of the PD detection in the VHF/UHF range is the comparatively high signal-to-noise ratio. Using this technology, it can basically be distinguished between wide-band and narrow-band PD detection methods. Wide-band VHF/UHF measuring systems are equipped with a high-sensitive wide-band amplifier in combination with a very fast peak detector to evaluate the crest value of the amplified PD signal, as exemplarily shown in Fig. 4.74a. Using the narrow-band method, damped oscillations are excited, where also a fast peak detector is used to evaluate the maximum magnitude of the envelope, see Fig. 4.74b.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig74_HTML.gif — Fig. 4.74
PD pulse response of the investigated UHF amplifiers (violet traces) and the associated peak detector (green traces). a Wide-band measuring system. b Narrow-band measuring system

Peak detectors used for both the wide-band and narrow-band PD signal processing in the VHF/UHF range elongate the input signal, whose duration is usually in the ns range, up to the µs range. Under this condition, the further signal processing can be performed by means of classical PD measuring systems, so that phase-resolved PD patterns can conveniently be displayed. A survey on the basic principles commonly used for the PD detection in the VHF/UHF range is shown in Fig. 4.75.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig75_HTML.gif — Fig. 4.75
Survey on UHF/VHF PD detection principles

In this context it should also be mentioned that for a wide-band PD detection in the VHF/UHF range also digital oscilloscopes can be applied, because these are nowadays commercially available for real-time measurements up to the GHz range. For narrow-band PD detection classical spectrum analysers are applicable, where either the full-span mode or even at zero-span mode can be used, see Fig. 4.76.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig76_HTML.gif — Fig. 4.76
Oscilloscopic screenshots gained by means of a spectrum analyser using the full-span mode (50–300 MHz). a Background noise level of the measuring surroundings. b Frequency response against a calibrating pulse

Using the full-span mode, the frequency spectrum of the captured PD signal as well as the superimposed noise is recorded for the pre-selected start and stop frequencies. To discriminate disturbing noises from the signal caused by PD events, the background noise level is initially recorded just prior the actual PD test is performed, see Fig. 4.76e. The main obstacle of the full-span mode is that the classical phase-resolved PD pattern cannot be displayed. To overcome this crucial problem, the zero-span mode is often preferred, which is in principle comparable with that technique used for radio interference voltage (RIV) measurements (Sect. 4.3.3). That means the center frequency is adjusted such that the noise level becomes a minimum, which can conveniently be determined using the full-span mode.

4.7.4 Comparability and Reproducibility of UHF/VHF PD Detection Methods

As discussed previously, the main benefit of non-conventional UHF/VHF PD detection methods is the considerable enhancement of the signal-to-noise ratio if compared to the traditional apparent charge measurement in compliance with IEC 60270:2000. This offers the opportunity to perform sensitive PD diagnosis tests of HV apparatus under noisy on-site condition, which is commonly impossible by using the IEC method, which recommends a limitation of the upper limit frequency below 1 MHz. In this context it must be emphasized, however, that the magnitude of the PD pulses appearing at the output of PD instruments operating in the VHF/UHF range is not correlated with the magnitudes of the apparent charge pulses. This is underlined by the measuring example shown in Fig. 4.77, which refers to a defective power cable termination.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig77_HTML.gif — Fig. 4.77
Oscilloscopic screenshots of phase-resolved PD pulses captured from a defective power cable termination which were measured simultaneously by means of a VHF measuring system (pink trace) and a PD instrument designed according to IEC 60270 (green trace). a Positive half-cycle. b Negative half-cycle

As can be seen, the magnitudes of the PD pulses appearing at the output of the VHF measuring system (pink trace) are not proportional to those of the apparent charge pulses (green trace). This is underlined by the graphical presentation shown in Fig. 4.78, which results from 10 subsequent screenshots according to Fig. 4.77. Here the peak value of each current pulses appearing at the output of the UHV measuring system was plotted versus the magnitude of the associated apparent charge pulse.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig78_HTML.gif — Fig. 4.78
Results of comparative PD studies showing the magnitudes of the PD current pulses evaluated by the VHF method versus the magnitudes of the apparent charge pulses

Despite the drawback that the UHF/VHF PD detection method cannot be calibrated in terms of pC, there are also various benefits. So the signal-to-noise ratio is essentially enhanced if compared to the IEC method, as mentioned previously. This offers the opportunity to determine the PD inception voltage as well as the PD trend under noisy on-site conditions. Moreover, this technique can advantageously be used for the localization of potential PD defects in geometrical extended HV apparatus, such as GIS, using the time-of-flight measurement (Pearson 1991), as will be described in Sect. 10.4.1.

4.8 Acoustic PD Detection

PD events radiate not only electromagnetic waves but emit also acoustic pressure waves, where the acoustic signal covers a frequency range between some kHz and several hundreds of kHz. The main benefit of the detection of acoustic emitted (AE) waves is their immunity against electromagnetic interferences. To prevent an impact of other mechanical vibrations caused by pumps and fans as well as acoustic noises emanated from the iron core of transformers on account of magnetostriction and Barkhausen effect, commonly the ultrasonic frequency range, preferably between 40 kHz and few 100 kHz, is chosen to capture and acquire AE signals.

The ultrasonic PD detection has initially been employed to localize airborne noises due to corona discharges, for instance, to identify disturbing discharges at shielding electrodes of HV test facilities as well as harmful discharges igniting in broken cap-and-pin insulators. A photograph of a hand-held battery-powered ultrasonic PD detector designed for this purpose is shown in Fig. 4.79.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig79_HTML.gif — Fig. 4.79
Photograph of an ultrasonic PD detector.
Courtesy of Doble Lemke

In the late 1950s, the ultrasonic PD detection technique was also employed to recognize and localize structure-borne noises emitted from PD defects in HV apparatus (Anderson 1956). Thereafter this technology became a widely established tool for preventive PD diagnostics of gas-insulated substations (Graybill 1974; Lundgaard et al. 1990; Albiez and Leijon 1991) and even for the localization of PD faults in large power transformers (Harrold 1975; Nieschwitz and Stein 1976; Howels and Norton 1978; Lundgard et al. 1989; Fuhr et al. 1993). Besides the magnitude, also the shape of the ultrasonic signal could be very informative to identify and localize potential PD defects due to the fact that the frequency content as well as the magnitude of the received acoustic signal appears considerably attenuated at increasing distance to the PD source.

Using only a single ultrasonic transducer, however, the localization procedure is extremely time-consuming, particularly in case of intermitting PD events. As an alternative, the so-called triangulation has nowadays become a common practice. For this purpose three or even more transducers are used to perform time-of-flight measurements , as illustrated in Fig. 4.80. For a homogenous medium, the distances x₁, x₂ and x₃ between the PD source and the AE transducers are proportional to the measured time-of-flight denoted as t₁, t₂ and t₃, which is deduced from the oscilloscopic records. Thus, the trajectories shown in Fig. 4.80 are crossing that point where the PD source is located.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig80_HTML.gif — Fig. 4.80
Principle of triangulation used for localization the PD site

To minimize the localization uncertainty, it is a common practice to use the ultrasonic technique in conjunction with electrical PD measurements. This offers the opportunity to trigger the oscilloscope by an electrical signal at instant when the PD event ignites, see Fig. 4.81a. As the time lag of the captured electrical signal is below the μs range, this can be neglected if compared to the flight time of the acoustic signal, because this travels in oil only 1.2 mm per µs. Under noisy condition, the signal-to-noise ratio can considerably be enhanced by the use of the so-called averaging mode, as illustrated in Fig. 4.81b. The combined acoustic-electrical method confirms moreover that indeed a PD defect has been detected and not a disturbing acoustic noise.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig81_HTML.gif — Fig. 4.81
Principle of time-of-flight measurement, where the acoustic signal is received by three ultrasonic transduces attached to the tank of a 110-kV instrument transformer. Here the oscilloscope was triggered by an electrical PD signal. a Single-pulse triggering. b Multi-pulse triggering (averaging)

Under laboratory condition, the electric signal required for triggering the oscilloscope is commonly captured from the test object via a coupling capacitor or even from the bushing tap, if available. Under noisy on-site condition, however, it is more beneficial to use the VHF/UHF technique to enhance the signal-to-noise ratio, as discussed in Sect. 4.7. In this context it must be emphasized, however, that the triangulation according to Fig. 4.80 provides only reasonable results for acoustic waves travelling in a continuum, where the velocity of acoustic pressure waves remains constant. For HV equipment of complex design, such as power transformers, the wave velocity is strongly affected by the very different construction materials, such as copper, steel, wood, pressboard and insulating oil. Thus, instead of the direct sound wave, propagating the shortest distance between PD source and ultrasonic transducer, two wave fronts of very different velocities have to be taken into consideration. These are commonly referred to as longitudinal (pressure) and transversal (shear) wave. Without going into further details, it should be mentioned that the shortest path is often not the fastest one, as illustrated in Fig. 4.82. This is due to the different velocities of the acoustic waves, which attain, for instance, approx. 1.25 mm/μs in oil and 5.1 mm/μs in steel. To solve the very complex equations gained for the wave velocity in real HV equipment, nowadays advanced computer systems equipped with sophisticated software packages are available.

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig82_HTML.gif — Fig. 4.82
Shortest and fastest path between a PD source in oil and an acoustic sensor placed on a power transformer tank

The facility required for the acoustic detection and location of PD sources in HV equipment comprises besides an array of AE transducers, a signal transmission unit (cabling or fiber optic link) and an acquisition system (digital oscilloscope or computer-based measuring system) to perform the signal processing as well as the visualization and even storage of the captured ultrasonic data. For this purpose the following types of ultrasonic transducers are commonly applied:

Piezo-electric transducers,
Structure-borne sound resonance transducers,
Accelerometers,
Condenser microphones, and
Electro-optic transducers.

As the acoustic impedance of the transducers is very different from that of the metallic enclosure of the HV apparatus under investigation, the transducer surface is usually covered with hard epoxy resin to ensure an efficient signal transmission. This provides additionally the required insulation between transducer and metallic parts of the test object. Moreover, special attention should be paid to the coupling method due to the fact that the emitted acoustic wave might be reflected at the interface between transducer and the enclosure of the HV equipment. Thus, acoustic couplant gel or grease should be used to minimize the impact of reflections.

Generally, it seems beneficial to integrate a pre-amplifier in the ultrasonic transducer in order to enhance the signal-to-noise-ratio. As mentioned above, disturbing mechanical vibrations caused by pumps and fans as well as noises emanated from the iron core of transformer on account of magnetostriction and Barkhausen effect, can effectively be rejected, because such acoustic noises are not correlated to acoustic pressure waves emitted from real PD sources. Using narrow-band amplifiers operating at center frequencies around 40 kHz might also contribute to reasonable test results. Under this condition the pulse response is characterized by oscillations where the envelope covers often a time span longer than hundreds of μs. Thus, a superposition of subsequent acoustic signals might appear, as obvious from Fig. 4.83. Among others, this is the reason why the acoustic PD detection method is not capable of evaluating the PD magnitude quantitatively. Another drawback is the strong attenuation and dispersion of ultrasonic signals if travelling through various insulation structures because these feature a low-pass filter characteristics, where the signal is attenuated nearly proportional to the square of the characteristic frequency (Beyer 1987).

../images/214133_2_En_4_Chapter/214133_2_En_4_Fig83_HTML.gif — Fig. 4.83
PD pulse response of a narrow-band ultrasonic measuring system having a center frequency of 42 kHz and a bandwidth of 800 Hz. a PD pulse train leading to a superposition of the acoustic signal. b Response of the acoustic measuring system against a single PD pulse

4. Partial Discharge Measurement