IEEE Transactions on Dielectrics and Electrical Insulation Vol. 26, No.

4; August 2019 1293

UHF PD Signal Transmission in GIS: Effects of 90° Bends

and an L-shaped CIGRE Step 1 Test Section
Glenn Behrmann and Stefan Franz*
ABB Switzerland Ltd.
Brown Boveri Strasse 5
CH8050 Zürich, Switzerland
*now with KPMG AG in Zürich

Jasmin Smajic, Zeljko Tanasic and Reto Christen

HSR, University of Applied Science Rapperswil
IET, Institute for Energy Technology
Oberseestrasse 10
8640 Rapperswil, Switzerland

ABSTRACT
When applying radio-frequency (RF) techniques to measure and monitor partial
discharges (PD), the CIGRE ‘sensitivity verification’ recommendation details a
methodology for achieving a specified equivalent PD charge sensitivity for the
combination of RF PD sensors and the measurement or monitoring system installed in
a gas-insulated substation (GIS). The recommendation allows some flexibility in the
GIS set-up used to perform the test. For the work covered in this paper, second in a
series, a CIGRE step 1 laboratory test was conducted using an ‘L’ -shaped section
between sensors ‘C1’ and ‘C2’, in order to present a more challenging propagation
path between them. RF network analyzer (RFNA) measurements and finite-element
modeling (FEM) were performed on the section and compared. The RF propagation
properties of GIS sections with similar topologies in both lab and field were
investigated through both models and measurements and these results compared with
prior work in the field. Refinements to FEM modeling techniques result in a closer
match to actual measurements, but an exact match probably remains out of reach. We
discuss these results and our recent observations, the implications for PD sensor
placement to satisfy the CIGRE recommendation, and outline anticipated next steps.
Index Terms — partial discharges (PD), gas insulated switchgear (GIS), UHF
method, CIGRE sensitivity verification

1 INTRODUCTION laboratory part and an on-site part, to establish the equivalent

PD charge sensitivity of a UHF method-based system.
THE so-called UHF (ultra-high frequency) method for
However, the CSVP really only explains how to do the tests;
measurement and monitoring of partial discharge (PD) in gas-
effective design and placement of the UHF PD sensors
insulated switchgear (GIS) or lines (GIL) has been covered
requires in-depth understanding of the RF signal propagation
widely in the technical literature beginning in the 1980’s [1–5]
environment within the GIS. Although outside of the specific
through the present day [6–8]. It has been well-known since
scope of this paper, it is important to understand that the
the earliest days that it is impossible to derive the charge
results of the CSVP tests – indeed, any UHF-based PD
magnitude of the PD event from the received RF signal
measurement - depend on the frequency response of the UHF
strength (RSS) of the PD signal [5, 9–11]. This necessitates
PD sensor, the RF propagation (transfer function) along each
careful placement of UHF PD sensors throughout a GIS to
unique sensor-sensor path through the GIS, and the interaction
assure sufficient coverage and sensitivity to pick up any
(RF matching) between the electronics and the sensors.
significant PD activity, e.g. a 2 mm particle, anywhere in the
GIS. This challenge gave rise to the well-known CIGRE At first glance, GIS (here we only deal with single-phase)
sensitivity verification procedure [9] (henceforth abbreviated would appear to present the RF propagation characteristics of
as ‘CSVP’) which sets forth a two-part method, consisting of a well-behaved coaxial waveguide, but the reality is very
different. GIS is purposely designed and built to transmit and
Manuscript received on 21 December 2018, in final form 18 March switch kiloampere currents at kilovolt voltages. All of the
2019, accepted 21 March 2019. Corresponding author: J. Smajic. components required to perform its primary function – the

DOI: 10.1109/TDEI.2019.008005
1294 G. Behrmann et al.: UHF PD Signal Transmission in GIS: Effects of 90° Bends and an L-shaped CIGRE Step 1 Test Section

switching and contact assemblies, voltage and current the CSVP allows flexibility for placing sensors C1 and C2
transformers, 90° bends, T-junctions, changes in diameter, etc. when performing the Step 1 Laboratory Part: “The lowest
– combine to present an extremely challenging propagation possible attenuation (shortest distance) between two sensors
environment at UHF frequencies. The reason is clear since the [C1, C2] is preferable. If not possible, the sensitivity
GIS interior dimensions and UHF wavelengths are of the same verification Step 1 could be performed, without any
order. The interior components and changes in geometry disadvantages, using a complete GIS bay” [9].
manifest themselves as countless RF impedance Our previous work [8] was based on a CSVP Step 1 set-up
discontinuities, giving rise to strong moding effects along with in which sensors C1 and C2 were located in relatively short
constructive and destructive reflections. Instead of a well- compartments, side by side. Recently however, a customer
behaved coaxial waveguide, GIS rather resembles a chain of requested a CSVP Step 1 test to be carried out with a more
very high-Q RF resonant cavities and ‘traps’ (notch filters)
challenging RF propagation path between sensors C1 and C2.
whose resonant frequencies are direct functions of their
Financial and time considerations prevented building up a
dimensions; thus even adjacent phases of the same section of
complete GIS bay for the test, so an ‘L’-section comprising a
GIS often exhibit surprisingly large differences in propagation
straight compensation unit plus a 90° ‘elbow’ bend was agreed
characteristics. The received RF signal strength (RSS)
upon, as shown in Figure 2. This L-shape was chosen with
measured at the PD sensor is determined by the transfer
earlier fine work in mind [6, 15–17] and because experience
function of the RF propagation path back to the PD defect
has shown both 90° elbows and compensator units strongly
(which is unknown); this is the reason it is not possible to
affect UHF signal propagation. Compensator units allow
determine the charge level (pC) [5, 9–11].
limited lateral movement of the GIS; both their inner
The GIS community has been aware of these facts since RF conductor and enclosure are not solid and continuous, but
(VHF, UHF) techniques began to be applied for detecting PD instead consist of separate components which result in
in GIS in the 1980s [1, 2]. Although RF signal attenuation is interruptions of RF surface currents which in turn represent
surprisingly low in long straight sections (GIL), it was soon impedance discontinuities and thus higher loss for the UHF
realized that UHF PD sensors need to be placed throughout a signals. The most obvious discontinuity in the RF path can be
GIS in order to guarantee sufficient sensitivity to detect PD seen in the photograph in Figure 2 by noting the copper bands
defects. Because of the RSS-charge ambiguity problem, a and aluminum plates which carry the enclosure currents across
means was sought to somehow establish a relationship the flexible sections at both ends of the compensator unit. This
between them, whence came the CIGRE sensitivity is a good example of how a GIS component, designed for its
recommendation [9]. This defines a test which establishes an primary purpose of transmitting heavy electrical currents,
equivalence between the RSS for a given GIS, sensor, and nevertheless presents obstacles to RF signals, i.e. higher
measurement equipment set-up to the charge of an actual PD losses. Noting the ‘skin effect’ - the depth of penetration of RF
source, calibrated according to IEC 60270. A second step surface currents (ca. 2 μm at 1 GHz for aluminum [18])
employs this value (pulse generator voltage) to verify together with the inductance of the enclosure bypass
sufficient coverage between the UHF PD sensors in the actual connections (negligible at power frequencies but several
GIS. It is not economically feasible to put sensors in every orders of magnitude higher at UHF), it’s clear how perfectly
compartment, but the complexity of predicting RF propagation normal design features of GIS look different from the
in GIS makes the trade-off between getting high sensitivity perspective of UHF frequencies.
throughout the GIS while keeping costs down difficult to
handle. Typically, GIS OEMs respond to this by devising
guidelines and tools to guide sensor placement. In our case,
these guidelines are based on a combination of RF
transmission (S-parameter) measurements, computer modeling
of GIS RF propagation, and other inputs, such as the
accumulated results of on-site CSVP Step 2 checks.
In recent years, much fine work has been done to improve
understanding of UHF PD signal propagation GIS [6, 7, 12–
17]. This motivated us to undertake our own investigations [8]
to compare with others’ work, further deepen understanding,
and improve sensor placement methodology. That initial work
has now been extended to more complex GIS geometries,
again in the context of the CIGRE recommendation, and these
new results and observations are presented here.

2 CIGRE STEP 1 ON AN ‘L’-SECTION

Figure 1. Photograph of the CIGRE Step 1 set-ups from our previous work
Because the detailed design and placement possibilities of [8]; the enclosures are ~700 mm long, as is therefore the distance between
UHF PD sensors are dependent on the individual GIS model, sensors.
IEEE Transactions on Dielectrics and Electrical Insulation Vol. 26, No. 4; August 2019 1295

step. In order to always have comparable results, we maintain

the exact same frequency span (0 – 2 GHz) and number of
frequency steps (2000). To generate the CPS value, the
relative attenuation (or signal strength) in dB (dBm) is first
linearized (converted into watts, hence the factor 10) for each
frequency step, and then the resulting ‘power/frequency
blocks’ are summed together. The CPS value enables an easy
way to quantify the difference between the complicated
frequency-response plots typical of GIS propagation paths.
2000
 P /10
cps  10 freq ( n ) (1)
n1

The ‘CPS’ values are calculated in the MATLAB program

which generates the frequency response plots. In Figure 3,
MATLAB produced CPS values of 15.0667 and 0.51248,
Figure 2. Present CIGRE Step 1 set-up; C1a -C2a distance shown, C2b – C1b
distance is 4760 mm + 2210 mm, respectively.
respectively. Dividing these and taking the log10 (power in
dBm or loss in dB), the L-shaped section is shown to exhibit
The difference in GIS geometries between sensors C1 and 14.7 dB more loss compared to the side-by-side set-up in Fig.
C2 in the two set-ups is obvious, and one of the first things 1. This loss value compares well with measurements taken
was to plot their respective S21 transmission frequency during our previous work (not published at the time). That set-
responses using MATLAB to see how they compare, as shown up also happened to include a 90° elbow through which a loss
Figure 3. (All measurements in this paper were taken using a of about -10 dB was measured across the band 0 – 2 GHz. In
Rohde & Scharz ZVL RF network/spectrum analyzer [19]; in this present set-up, the compensator unit also adds some small
all plots, the vertical axis is logarithmic signal strength or additional loss due to its internal construction, as detailed
attenuation, and the horizontal axis is linear frequency, above.
nominally 0 – 2 GHz.) As in our previous work [8], a detailed FEM model of the
C1-C2 path through the L-shaped section was built up,
including fine-scale details of the sensors. Figure 4 shows the
comparison between the S21 frequency response of the FE
model together with RF network analyzer measurement,
measured between PD sensors C1a – C2a (the red sensors in
Figure 2). Often plots match quite well if viewed from a
distance, but closer examination reveals many fine-scale
differences which will now be discussed in more detail.

Figure 3. Comparison of S21 sensor-to-sensor transmission response

measurements for two CIGRE set-ups shown in Figures 1 and 2; Figure 1
(C1-C2 adjacent) = blue trace, Figure 2 (‘L’-section) = green trace.

As would be expected, the attenuation of the set-up with the

‘L’-section is somewhat higher than in the earlier set-up where
sensors C1 and C2 were in adjacent compartments, but it is
virtually impossible to precisely quantify the difference
Figure 4. Comparison of S21 frequency response through the L-section in
between such plots based on simple visual comparison. Figure 2, simulated (red trace) vs. measured (blue trace).
To solve this problem, ‘channelized power summation’
(CPS) is used, shown in Equation (1) and expressed in units of To further underline the complexity of the RF propagation
W/MHz. This comprises the area under the respective environment in even a relatively simple section of GIS, we
frequency-response curve, numerically integrated across the present RF measurements of the signal produced by an actual
measured spectrum, divided by the measurement frequency hopping particle (5 pC per IEC 60270) in the L-shaped
1296 G. Behrmann et al.: UHF PD Signal Transmission in GIS: Effects of 90° Bends and an L-shaped CIGRE Step 1 Test Section

CIGRE test section which are both interesting and counter- placement is to make transmission measurements with an RF
intuitive. network analyzer on actual GIS on site. Recently a large
In the first measurement, the RF signals from sensors C1a 400 kV GIS located in the U.K. offered the opportunity to
(blue trace) and C2a (pink trace) received from the hopping measure two somewhat unusual ‘bridge’ (‘П’) sections
particle are plotted together in Figure 5. Remembering that designed to allow large construction cranes entrance access to
C2a is the ‘remote’ sensor, careful inspection reveals the the site. Photographs of the two ‘П’-sections and their
surprising fact that the particle signal amplitude is actually corresponding S21 frequency response plots are shown in
higher at C2a than it is at sensor C1a for certain discrete Figure 7.
frequencies.

Figure 6. A pulse of 10 V is injected at sensor C1a and the resulting signal

measured at sensors C1b (blue trace) and C2b (maroon trace). The red ovals
Figure 5. Here the output signals at the two type ‘WB3’ sensors C1a (blue mark tones where the received RF signal level at the distant sensor (C2b)
trace) & C2a (pink trace) are compared (the 5 pC particle was the signal exceeds the level at the near sensor (C2a). See text for details.
source). The red ovals mark tones where the received RF signal strength at the
distant sensor (C2a) exceeds the level at the near sensor (C2b). See text for
details.

In the second example, a fast rise-time pulse generator

(Power Diagnostix CAL2B) was used to inject a signal at the
(red) sensor C1a (shown in Figure 2), and the signals received
at sensors C1b and C2b (‘b’ indicates another type of sensor
built into the disconnector module shown in the photo in
Figure 2) are plotted together for comparison in Figure 6.
Again, signals at the distant sensor C2b can be seen whose
amplitudes exceed those at the ‘near’ sensor C1b at certain
discrete frequencies.
In both Figures 5 and 6, the spectra from the sensors nearer
to the signal source (either the 5 pC particle or the pulse
generator) mostly exhibit the higher amplitude, which would
be intuitively expected, but it is also clearly visible that signals Figure 7. Photos of ‘maintenance bridge’ П-sections with their respective S21
at certain discrete frequencies exhibit higher levels. This effect frequency-response plots; arrows indicate the PD sensor locations, dotted line
boxes indicate broadband notch in their RF transmission spectra. See text.
was not staged! - it will occur in any GIS set-up. We like to
call this ‘signal tunneling’; such effects have been hinted at in
Both geometries exhibited the remarkable broadband notch
others’ work [17], and again demonstrate the extremely
behavior clearly visible in both plots. Equivalent transmission
complex nature of the RF propagation environment in GIS.
behavior has since been observed in similar GIS sections. This
These effects are shown to again underline the fact that it is striking behavior motivated us to investigate further. Both
impossible to estimate the coulombic charge of a PD source space and financial constraints prevented setting up full-scale
from the received RF signal strength, or to assign a discrete ‘П-bridges’ in our laboratory, but it was possible to investigate
attenuation value for a particular GIS module or path. one side of the ‘П’-section as shown in Figure 8. Its S21
One of the best ways to gain insight into UHF signal transmission response is plotted together with that of a simple
propagation and improve engineering guidelines for PD sensor straight 3 m long bus section for comparison in Figure 9.
IEEE Transactions on Dielectrics and Electrical Insulation Vol. 26, No. 4; August 2019 1297

frequency range enclosed by the dotted line box in Figure 9,

and also includes the on-site measurement from one of the
‘П’-sections shown in Figure 7.
Again, FE models of the two ‘П-bridges’ were built up
using the actual mechanical CAD data of the GIS; Figure 11
shows that small geometrical details of all of the internal
components are discretized with a very fine mesh. This is a
decisive parameter for achieving a high level of agreement
with actual measurements, but at the same time considerably
increasing memory requirements and CPU time. The
transmission frequency responses measured on site are plotted
together with those calculated from the FEM models for
comparison.

Figure 8. Photograph of the '└┐'-section laboratory set-up.

Figure 11. Drawing of the FE model showing mesh.

Figure 9. S21 transmission response of └┐-section (blue trace) vs. 3 m long

straight section of simple bus (green).

Although the dramatic broadband notch seen in the on-site

measurements (Figure 7) does not appear here, closer
inspection reveals a region in the RF spectrum which suggests
unusual behavior compared to the simple 3 m long straight
section. To see this more clearly, Figure 10 zooms in on the

Figure 12. Plot of S21 transmission response for └┐-section shown in Figure
8; network analyzer measurement (aquamarine) vs. FEM model (brown).

The powerful effects 90° elbows exert on RF signal

propagation is surprising in light of their relatively small
physical size. FEM RF models make it easy to visualize these
transmission effects at specific discrete frequencies. Figures
15 and 16 show snapshots from animated visualizations of the
RF signal flowing through 90° elbows at two different discrete
frequencies, one corresponding to a peak in the S21
transmission response and the other to one of the deep troughs.
The relative field strength is indicated in the vertical stripe
legend on the right; dark blue/black indicates low field
Figure 10. Zoom of dotted box from Fig. 9 with the on-site ‘П’-section
‘bridge’ superimposed for comparison (green trace); note this plot is from 0 – strength, red indicates high field strength. (The full animations
600 MHz. can be seen on-line at [Link]/CAEM/GIS.)
1298 G. Behrmann et al.: UHF PD Signal Transmission in GIS: Effects of 90° Bends and an L-shaped CIGRE Step 1 Test Section

Figure 16. FEM of └┐-section, f = 260 MHz; blue = ‘cold’ = no signal.

recommendation was just being drafted [9], and people began

making RF transmission measurements across sections of GIS
to improve understanding of propagation effect [4, 20, 21].
Figure 13. S21 transmission frequency response of the ‘П-section bridge’ from The technology of making network analyzer (or tracking
Figure 7 left; measured (aquamarine) vs. FEM (brown). generator) measurements has not changed that much (although
the instrument OEMs might disagree strenuously with that
statement), but there has been a revolution in the capabilities
of computational modeling technology, which in turn has been
greatly enhanced by the concurrent dramatic increase in
computing power and memory density. The RF modeling suite
used here (COMSOL [22]) allows directly importing the
detailed CAD construction parameters of the individual GIS
components on which to base the FEM models.
Models with this level of detail, precision, and 3D spatial
resolution were simply not achievable 20+ years ago, and
should theoretically allow very close agreement to actual
measurement data. However, after painstakingly modeling a
section of GIS and plotting its frequency response, it is often
still quite discouraging to see large discrepancies to actual
measurements. Despite recent advances, there are serious
impediments involved with this endeavor, for example:
Figure 14. S21 transmission frequency response of the ‘П-section bridge’ from 1. Correct entry of the GIS CAD construction data is very
Figure 7 right; measured (acquamarine) vs. FEM (brown). time-consuming and requires specialized know-how.
2. The computing times necessary to achieve high frequency
resolution are very resource-intensive. The table below
gives some statistics for the computing resources and run-
times for the model of the └┐-shape in Figure 8.

Table 1. Computation time with different set-ups.

Specification RAM Runtime CPU-time
8 nodes, 8*96 GB 25h, 10min 4027h,
Cluster
160 cores = 768 GB 28s 54min 40s
Work- 1 socket, 256 GB 137h, 1378h,
station 10 cores (max) 50min, 10s 21min, 20s
Work- 2 sockets, 256 GB 69h, 24min, 1388h,
Figure 15. FEM of └┐-section, f = 540 MHz; the signal passes with little station 20 cores (max) 58s 19min, 20s
loss.
3. Recalling that the penetration depth of surface currents at
3 DISCUSSION 1 GHz is on the order of a few microns, it is clear that the
RF methods for measuring PD in GIS, usually referred to ‘macroscopic’ CAD construction data for the GIS is
as the ‘UHF technique’ (or ‘method’), have been being simply not resolved to that level, and it would be
applied since the late 1980s. The references show only a extremely costly to model such ‘microscopic’ resolution
small selection of efforts made to better understand and (the run-times would also increase exponentially).
quantify UHF PD signal behavior in GIS, starting over 20 4. The GIS components and their surface boundaries are also
years ago [3, 4, 13, 20]. In this same period, the original CIGRE not precision machined to such high tolerances; in
IEEE Transactions on Dielectrics and Electrical Insulation Vol. 26, No. 4; August 2019 1299

contrast to expensive RF components, there is no way to would not completely trust a model to precisely determine RF
know exactly how RF surface currents flow at GIS transmission performance of an unknown and untested section
interfaces, e.g. through contact assemblies. Just of GIS. Further insight was gained about the strong effects of
disassembling and reassembling the GIS would change 90° bend geometries on UHF PD signals and this information
these microscopic interface parameters as well. has been integrated in our sensor placement guidelines. We
5. Last, but not least, the models presented here are bounded note that all of the 90° bends discussed here are simple bends
and do not take into account the exact RF impedances of in the same plane, i.e. they are 2-dimensional geometries, or
the adjacent sections of GIS – again, it would be ‘co-planar’. However, ‘non-coplanar’, i.e. ‘back-to-back’ 90°
prohibitively expensive to do so. Instead, we terminate the elbows are often required in actual GIS. Recently, the CIGRE
ends in ‘ZGIS’, a gross simplification ignoring the fact that Step 2 failed on-site for an exit feeder containing back-to-
the RF behavior of each GIS module is affected by its back, non-coplanar 90° elbows (the problem was subsequently
neighboring modules. For this same reason it is solved). As a result, we have now begun analyzing these
impossible to assign a fixed attenuation value for a given structures, and an upcoming GIS project will offer the
GIS module; they are embedded in a complex system. opportunity to make S-parameter measur-ements on several
such sections. We plan to present that work in our next
In summary, despite great advances in computing power
publication.
and in the modeling packages (and how to apply them),
constructing an FEM model of GIS which exactly reflects its
effects on RF signals with the accuracy considered routine in ACKNOWLEDGMENT
the world of industrial RF circuit design remains out of reach. The authors wish to thank Markus Schraudolph (ABB) and
Efforts continue to better understand how to ‘tune’ the FE Daniel Treyer (Paul Scherer Institute) for valuable technical
models, including which aspects are being overlooked or discussions and Urs Kruesi (ABB) for his continued support.
missed, which parameters require more detailed modeling, and
which, if any, may be neglected. While very useful as a
REFERENCES
scientific tool to gain insight into GIS signal propagation as
[1] S.A. Boggs, “Electromagnetic Techniques for Fault and Partial
we have shown here, modeling an entire GIS remains Discharge Location in Gas-Insulated Cables and substations,” IEEE
economically infeasible, at least for now. Other workers in the Trans. Power App. Sys., vol. PAS-101, no. 7, pp. 1935–1941, July 1982.
field have come to similar conclusions [19]. The best solution [2] B. F. Hampton, R.J. Meats, “Diagnostic Measurements at UHF in Gas
for understanding UHF PD sensor placement remains carrying Insulated Substations,” IEE Proc., vol. 135, part C, no. 2, pp.137–144, 1988.
[3] M. D. Judd, O. Farish, and B.F. Hampton, “The Excitation of UHF
out the on-site step 2 part of the CIGRE Sensitivity Signals by Partial Discharges in GIS,” IEEE Trans. Dielectr. Electr.
Verification, taking careful measurements of as many Insul., vol. 3, no. 2, pp. 213–228, Apr. 1996.
candidate paths as possible, and carefully correlating the [4] G.J. Behrmann, S.M. Neuhold, R. Pietsch, “Results of UHF
Measurements in a 220 kV GIS Substation during On-site
results. Commissioning Tests,” IEEE Int. Symp. High Voltage Engineering
The simulation results presented here were obtained using a (ISH), 1997, pp. 451–455.
[5] CIGRE WG D1.33, Tech. Brochure 444, Guidelines for Unconventional
modern 3-D vector, full-Maxwell FEM solver in the frequency Partial Discharge Measurements, 2010.
domain [22, 23]. This simulation technology allows [6] S. Okabe, G. Ueta, H. Hama, T. Ito, M. Hikita, H. Okubo, New Aspects
performing 3-D simulations of GIS sections in their full of UHF PD Diagnostics on Gas-insulated Systems, IEEE Trans.
Dielectr. Electr. Insul., vol. 21, no. 5, pp. 2245–2258, Oct. 2014.
complexity, i.e. no simplifications are needed [8]. This is a [7] S. M. Hoek, A. Kraetge, M. Koch, O. Kessler, M. Heindl, Emission and
considerable step forward compared to previous simulation Propagation Mechanisms of PD Pulses for UHF and Traditional
work presented in [6, 12–17], where the Finite Difference Electrical Measurements, IEEE Int. Conf on Cond. Monitoring &
Time Domain (FDTD) was employed. Bound to the FDTD Diagnosis (ICCMD), 2012, pp. 23–27.
[8] G. Behrmann, J. Smajic, “RF PD Signal Propagation in GIS: Comparing
and its regular mesh discretization, earlier authors were S-parameter Measurements with an RF Transmission Model for a Short
constrained to relatively simplified versions of GIS 90° bends, Section of GIS,” IEEE Trans. Dielectr. Electr. Insul., vol. 23, no. 3, pp.
T-junctions, and insulator/contact assemblies. Building on that 1331–1337, June 2016.
[9] CIGRÉ TB 654 (WG 1.25); UHF PD Detection System for GIS:
earlier excellent work, the goal was to gain more exact models Application Guide for Sensitivity Verification, April 2016. (Update/rev.
by use of more advanced FEM techniques as shown here. of CIGRE TF 15/33.03.05, ÉLECTRA no. 183, pp. 74–87, April 1999.)
[10] High-voltage test techniques – Measurement of partial discharge by
electromagnetic and acoustic methods, IEC 62478, IEC 62478, 2016.
4 CONCLUSIONS [11] CIGRÉ TB 662 (WG 1.37), Guidelines for PD Detection using
Conventional and Unconventional Methods, Aug 2016.
Frequency response results derived from FEM models have [12] H. Muto, M. Doi, H. Fujii and M. Krunei, “Resonance characteristics
been compared with actual S-parameter measurements for and identification of modes of electromagnetic wave excited by partial
some relatively simple sections of 400 kV GIS, including a discharges in GIS,” IEEE Int. Symp. High Voltage Eng. (ISH), 1999, vol.
small L-shaped set-up employed for a CIGRE Sensitivity 5, pp. 70–73.
[13] Meijer, S., Smit, J. J., Girodet A., “Estimation of UHF signal
Verification Step 1 test, a ‘П-shaped’ ‘bridge’ section of an propagation through GIS for sensitive PD detection,” IEEE Int. Electr
actual GIS on site, and a laboratory emulation of one end of Insul, 2002.
the latter geometry. While clear progress has been made [14] S. Okabe, S. Yuasa, S. Kaneko, M. Yoshimura, H. Muto, H. Yoshiyasu,
C. Nishida, M. Kamei, “Simulation of Propagation Characteristics of
getting the FE models to more closely match actual Higher Order Mode Electro-magnetic Waves in GIS,” IEEE Trans.
measurement data, much work remains to be done; we still Dielectr. Electr. Insul., vol. 13, pp. 855–861, Aug 2006.
1300 G. Behrmann et al.: UHF PD Signal Transmission in GIS: Effects of 90° Bends and an L-shaped CIGRE Step 1 Test Section
[15] M. Hikita, S. Ohtsuka, T. Teshima, S. Okabe, S. Kaneko, “Examination Zeljko Tanasic received his [Link]. and [Link]. degree
of Electromagnetic Mode Propagation Characteristics in Straight and L- from the University of Applied Science Rapperswil
Section GIS Model Using FD-TD Analysis,” IEEE Trans. Dielectr. Elec. (Switzerland) in 2012 and 2016 respectively on the topic
Insul., vol 14, no. 6, pp. 1477–1483, Dec 2007. of coupled electro-magnetic analysis of GCB earthing
[16] K. Omori; M. Sawada; S. Isejima; S. Ohtsuka; H. Ikeda; M. Hikita; G. switches.
Ueta; S. Okabe; T. Hoshino; S. Maruyama; T. Sakakibara, “Propagation From 2012 to 2018 he was working for the
properties of PD-induced electromagnetic wave through L-shaped Computational and Applied Electromagnetics Group of
structure in model GIS measured with UHF sensors,” IEEE Int. Conf. the University of Applied Sciences Rapperswil. His work
Cond. Mon. Diag., 2008, pp. 1044–1047. covered a wide range of projects in the field of
[17] M. Hikita, S. Otsuka, J. Wada, S. Okabe, T. Hoshino, and computational electromagnetics and coupled problems such as effects of short
S. Maruyama., “Propagation Properties of PD-induced Electromagnetic circuit forces on transformers and circuit breakers. Additionally working on
Wave in 66 kV GIS Model with L Branch Structure,” IEEE, Trans. projects concerning the development and optimization of AC welding
Dielectr. Electr. Insul., vol.18, no.5, pp.1678–1685, 2011. transformers as well as numerical simulations of very fast electromagnetic
[18] John Kraus & Keith Carver, Electromagnetics, McGraw-Hill 1973 transients in x-ray applications and gas-insulated switchgears. In 2018 he
(newer versions exist) started working as a Simulation Engineer at Hyundai Electric’s R&D centre in
[19] Agilent Application Note: “Agilent Network Analyzer Basics,” pub. no. Switzerland.
5965-7917E, Agilent Technologies Inc., CA, USA, 2004.
[20] R. Kurrer, K. Feser, “Attenuation Measurements of Ultra High
Frequency Partial Disdtarge Signals in Gas Insulated Substations,” IEEE Reto Christen received his [Link]. and [Link]. degree
Int. Symp. High Voltage Eng. (ISH), 1997, vol. 2, pp. 161–164. from the Faculty of Electrical Engineering in
[21] M.D. Judd and O. Farish, “Transfer functions for UHF partial discharge Rapperswil (Switzerland) in 2015 and 2018,
signals in GIS,” IEEE Int. Symp. High Voltage Eng (ISH), 1999, vol. 5, respectively. The main topic of his [Link]. studies was
pp. 74–77. a voltage regulator module for data centers.
[22] COMSOL Multiphysics Modeling Software, [Link], 2015. In 2015 he joined the Computational and Applied
[23] J. Smajic, W. Holaus, J. Kostovic, U. Riechert, “3D Full-Maxwell Electromagnetics Group of the University of Applied
Simulations of Very Fast Transients in GIS,” IEEE Trans. Mag., vol. 47, Sciences Rapperswil. His work covers a wide range of
no. 5, pp. 1154–1517, May 2011. projects in the field of computational and applied
electromagnetics such as LI-modeling and simulation of transformer
windings, fast electromagnetic transients in power- and distribution
transformers, very fast electromagnetic transients in gas insulated switchgears,
Glenn Behrmann (S’78, M’79) received a B.S.E.E. modeling and simulation of solid-state transformers, and simulating and
from Union College, Schenectady, NY, in 1979. He designing power electronics.
began working in RF engineering and signal
processing at MIT Lincoln Laboratory in Lexington,
MA, USA. He continued in RF and EMC, joining Stefan Franz received his [Link] and [Link] degree
ABB Corporate Research in Baden, Switzerland in from the Dept. of Information Technology and
1989. At ABB his work focuses on detection, Electrical Engineering from the Swiss Federal
monitoring, and assessment of PD signals in GIS Institute of Technology in Zurich (Switzerland) in
using both conventional and UHF techniques. He has 2014 and 2017, respectively, on the topic of
authored papers and holds patents in the field, and is active in CIGRE D1 and optimization of HVDC disconnectors for bus-transfer
IEC TC42. switching. From 2015 to 2017, he was working as
intern in the teams of research and development and
IT and projects at ABB Switzerland. During this
Jasmin Smajic (M‘07) received his [Link]. and Ph.D. internship, Stefan investigated, inter alia, the
degree from the Faculty of Electrical Engineering propagation of electromagnetic waves in switchgear. In 2017, he started
and Computing in Zagreb (Croatia) in 1998 and working at KPMG Switzerland and is still part of the cyber security team.
2001, respectively, on the topic of numerical
computing and optimization of static and time-
varying electromagnetic fields.
From 2002 to 2004 he was a postdoctoral research
fellow in the Laboratory for Electromagnetic Fields
and Microwave Electronics at the ETH Zurich
(Switzerland). The main topics of his post-doctoral
studies were full-Maxwell electromagnetic simulations of photonic crystals
and similar electromagnetic meta-materials. In 2004 he took a position of
scientist at the ABB Corporate Research Centre in Baden-Dättwil
(Switzerland) where he stayed until 2011. His work in ABB covered a wide
range of projects in the field of computational and applied electromagnetics
such as LI-modeling and simulation of transformer windings, fast
electromagnetic transients in power- and distribution transformers, coupled
electromagnetic-mechanical and electromagnetic-thermal analysis of
transformers and circuit breakers, and very fast electromagnetic transients in
gas insulated switchgear. In 2011, he was appointed professor of electrical
engineering at University of Applied Sciences Eastern Switzerland (HSR).
Jasmin Smajic is a member of CIGRE and IEEE and he has authored over
one hundred scientific publications, several books, and dozens of patents.