Anal Bioanal Chem (2002) 373 : 639–646
DOI 10.1007/s00216-002-1280-4
S P E C I A L I S S U E PA P E R
R. Kurte · C. Beyer · H. M. Heise · D. Klockow
Application of infrared spectroscopy
to monitoring gas insulated high-voltage equipment:
electrode material-dependent SF6 decomposition
Received: 11 December 2001 / Revised: 12 February 2002 / Accepted: 20 February 2002 / Published online: 13 April 2002
© Springer-Verlag 2002
Abstract Sulfur hexafluoride is a chemically inert gas
which is used in gas insulated substations (GIS) and other
high-voltage equipment, leading to a significant enhance-
ment of apparatus lifetime and reductions in installation
size and maintenance requirements compared to conven-
tional air insulated substations. However, component fail-
ures due to aging of the gas through electrical discharges
may occur, and on-site monitoring for risk assessment is
needed. Infrared spectroscopy was used for the analysis of
gaseous by-products generated from electrical discharges
in sulfur hexafluoride gas. An infrared monitoring system
was developed using a micro-cell coupled to an FTIR
spectrometer by silver halide fibers. Partial least-squares
calibration was applied by using a limited number of opti-
mally selected spectral variables. Emphasis was placed on
the determination of main decomposition products, such
as SOF2, SOF4, and SO2F2. Besides the different electrical
conditions, the material of the plane counter electrode of
the discharge chamber was also varied between silver, alu-
minum, copper, tungsten, or tungsten/copper alloy. For
the spark experiments the point electrode was the same
material as chosen for the plane electrode, whereas for
partial discharges a stainless steel needle was employed.
Complementary investigations on the chemical composi-
tion within the solid counter electrode material by sec-
ondary neutral mass spectrometry (SNMS) were also car-
ried out. Under sparking conditions, the electrode material
plays an important role in the decomposition rates of the
gas-phase, but no relevant material dependence could be
observed under partial discharge conditions.
Dedicated to Professor David M. Hercules on the occasion
of his 70th birthday
R. Kurte · C. Beyer · H.M. Heise (✉) · D. Klockow
Institute of Spectrochemistry and Applied Spectroscopy,
University of Dortmund,
Bunsen-Kirchhoff-Str. 11, 44139 Dortmund, Germany
e-mail: heise@isas-dortmund.de
Keywords Fourier transform infrared spectroscopy ·
Trace gas analysis · Sulfur hexafluoride decomposition ·
By-product monitoring · Partial least-squares calibration
Introduction
Sulfur hexafluoride is a chemically inert gas which is used
in electrical high-voltage equipment due to its insulating
and arc extinguishing properties. During the last two
decades, the use of gas insulated substations (GIS) and
other high-voltage equipment containing SF6 has in-
creased significantly due to their enhanced lifetime and
reduced maintenance needs. Indoor substations are possi-
ble owing to a drastically reduced apparatus size com-
pared to conventional air insulated substations [1]. De-
spite the proven reliability of such components in the
electrical power distribution systems, failures may occur
due to SF6 aging under electrical stress, and on-site moni-
toring of the fill gas is needed for risk assessment. In-
frared spectroscopy is a powerful tool for this purpose,
because of its capacity to identify and determine gaseous
by-products [2].
The decomposition of SF6 under electrical stress de-
pends on insulation gas pressure, electrode materials, and
the inner surface materials of the equipment. Furthermore,
impurities such as oxygen and water vapor play an impor-
tant role. In particular, the influence of electrode materi-
als, such as aluminum, copper, stainless steel, tungsten,
and a tungsten/copper alloy on the production of SF6 break-
down and reaction products has been studied for either
sparking or partial discharge conditions (also called
corona discharge) by us [3, 4] and others [5, 6]. Parallel
investigations were recently carried out by us on the sur-
face layer composition of the metallic electrodes using the
methods of secondary neutral mass spectrometry (SNMS),
X-ray photoelectron spectrometry (XPS), and scanning
electron microscopy with energy dispersive X-ray analy-
sis (SEM-EDX) [7]. Apart from decomposition of pure
SF6, conditions with a 50% admixture of CF4 were also
investigated by Casanovas et al. [5, 8] using gas chro-
640
matography. As SF6 is a potent greenhouse gas, mixtures
with N2 or other low-risk gases were recently considered
as replacements for pure SF6 to reduce the amounts even-
tually released into the atmosphere. Such mixtures were
submitted to corona discharges [9] or sparking [10]. Within
the latter study, the influence of solid organic insulation
material was also investigated. The effects of the insula-
tor, oxygen, and water on undiluted SF6 and SF6-CF4 mix-
tures have been summarized by Casanovas et al. [11]. An
approach to reaction modeling of a decaying arc plasma
with regard to SF6 decomposition has also been described
[12].
The emphasis in our study was on the determination of
the main decomposition products such as SOF4, SOF2, and
SO2F2. Representative results obtained by various partial
discharge and spark experiments are reported in this pa-
per, extending the electrical stress conditions used previ-
ously. Furthermore, the generation of decomposition
products was investigated including counter electrode ma-
terials not studied earlier, but found in high-voltage equip-
ment.
For gas-phase analysis, infrared spectroscopy can be
used for direct in situ multi-component quantification.
Due to the complexity of the gas composition, multivari-
ate chemometric methods must be applied for trace com-
ponent quantification. Recent assays of decomposition
products relied on a classical least-squares (CLS) fitting
procedure using appropriate infrared reference spectra of
the components, which contribute to the absorbance spec-
trum to be analyzed [2, 3]. Partial least-squares calibra-
tion, as reported for trace gas analysis of indoor or ambi-
ent air [13, 14], was recently applied by us to allow the
fast in situ process monitoring of high-voltage equipment
[15]. The aim of the present work is also to provide addi-
tional information for measurements carried out with a
small number of specific spectral wavelengths for the de-
velopment of simple and inexpensive infrared photome-
ters, which can be based, for example, on narrow-band fil-
ters, grating monochromators with a limited number of
detector elements, or infrared quantum cascade laser
based systems (for recent developments, see [16]). Such
instrumentation must be applicable for routine monitoring
under various electrical stress conditions.
Experimental
The decomposition of SF6 by partial discharges and sparking was
studied under a pressure of 250–300 kPa, which was provided for
insulation inside a discharge chamber with an internal volume of
350 mL, constructed from stainless steel and Teflon® using differ-
ent point-to-plane configurations [17]. The plane counter electrode
material was silver, aluminum, copper, tungsten, or tungsten/cop-
per alloy (elemental mass ratio of 72:28). The pure metals were of
a purity >99.9%.
For partial discharge experiments with a duration of 1 or 3 days,
the point electrode was a stainless steel sewing needle with a point
radius of 30 µm. A gap size of 11 mm was usually applied between
both electrodes. For standardization, the energy involved in the
AC-discharge processes was also recorded; for the measurement
principle, see [18]. For spark experiments, the needle was replaced by
a pin of 5 mm outer diameter with rounded ends (gap size 1–2 mm).
Fig. 1a–c Schematics of the high-voltage equipment used for par-
tial discharges using a point/plate configuration (a), electrode used
for spark experiments (b), and scheme of the counter electrodes
used for all kind of electrical experiments (c); the latter two dia-
grams are enlarged
The pin material was the same as chosen for the counter electrode;
for the experimental set-up, see Fig. 1. Experiments were carried
out accumulating different numbers of sparks at an energy of either
0.15 or 1.0 J spark–1.
Before the spark and partial discharge experiments were
started, the high-voltage chamber was prepared as follows: After
electrode installation, the chamber was flushed three times with
SF6 (99.9%) from Messer Griesheim (Krefeld, Germany) and
evacuated for five minutes resulting in a residual pressure of 100 Pa.
Finally, the cell was refilled with SF6, adjusting the pressure of the
insulating gas to 250 or 300 kPa. This cell preparation removed
H2O that could otherwise be desorbed from the electrodes and the
cell walls during the tests. The residual mass fractions of impuri-
ties in SF6 (99.9%) were below the following limits (in µg g–1): air
= 500, H2O = 15, CF4 = 500, and HF = 1.
Within our study, infrared spectra were recorded by using an
IFS 113v FT-IR spectrometer (Bruker, Ettlingen, Germany)
equipped with a KBr beamsplitter and a mercury cadmium tel-
luride (MCT) detector (cut-off at 530 cm–1) operated at liquid ni-
trogen temperature. The stressed SF6 gas from the discharge cham-
ber was directly expanded into a stainless steel heatable cell of
10 cm optical pathlength (SPECAC, Orpington, UK) with an aver-
age cell pressure of 73.5 kPa. To reach the appropriate spectral sig-
nal-to-noise ratio, a hundred interferograms were accumulated for
each spectrum with a spectral resolution of 0.5 cm–1 within a mea-
surement time of 3.5 min. The interferograms were Fourier trans-
formed after correction of the non-linear detector response to the
photon flux in the interferogram domain by using the spectrometer
software to avoid systematic photometric errors within the multi-
variate spectrometric data evaluation.
Alternatively, gas-phase measurements were carried out using a
lightpipe of 560 µL inner volume, coupled to an FTIR spectrometer
(Bruker IFS 66) and MCT detector (Infrared Associates, Inc., Stewart,
FL, USA) by silver halide fibers of 0.7 mm outer diameter, in order
to illustrate the potential for remote sensing of decomposition prod-
ucts close to the gas insulated high-voltage equipment.
The PLS calibration was based on our PC software package
programmed in MATLAB (The Mathworks, South Natick, MA,
USA). For the SF6 decomposition product analysis based on clas-
sical least-squares (CLS), a similar set of reference spectra with
optimum data intervals was used as previously discussed [2]. The
PLS reference gas concentration values were provided by the spec-
tral CLS evaluation. A special wavelength selection procedure was

chosen to set up PLS calibration models that are close to multiple
linear regression models, which use the classical least-squares es-
timator. Our scheme is based on a pair-wise selection of spectral
variables suggested by the absolute weights of the optimum PLS
regression vector calculated for appropriate broad, component spe-
cific spectral intervals [19]. Standard errors of prediction (SEP) are
reported from leave-one-out cross-validation.
Fig. 2a,b a SF6 transmission spectrum recorded using a lightpipe
as gas cuvette (inner volume 0.5 mL, coupled to the interferometer
by silver halide fibers), and b difference absorbance spectrum ob-
tained for a gaseous sample after partial discharges within the dis-
charge chamber housing a silver electrode (total discharge energy
50 mJ, duration 3 days); experimental spectroscopic conditions: ef-
fective optical pathlength l = 22 cm, total gas pressure p = 101.3 kPa;
spectral resolution 0.5 cm–1, 100 scans co-added
Table 1 Characterization of the SF6 decomposition products within the sample population studied and results of different PLS calibra-
tions
Compound Spectral intervala (cm–1) Nb, Mc
SOF4 830– 806 94, 8
SOF2 1340–1330 89, 4
SF4 746– 722 61, 4
SO2F2 1507–1470 72, 6
SiF4 1045–1010 84, 6
aWavenumbers used for individual PLS calibrations (first set is for
spectral variables with positive, and the second set for those with
negative PLS regression weights; values in cm–1): SOF4: 829.7,
820.6, 819.4, 806.1; 825.4, 824.4, 823.2, 817.2; SOF2: 134.5,
1330.6; 1334.0, 1331.4; SF4: 736.9, 729.7; 746.3, 722.9; SO2F2:
1504.2, 1502.8, 1500.8; 1506.6, 1471.4, 1470.0; SiF4: 1031.0,
1029.6, 1027.4; 1045.0, 1031.7, 1010.0
bN number of calibration samples
641
The microscopic SEM images of the metallic electrodes and
counter electrode surfaces were recorded using a SEM-EDX in-
strument model JEOL JSM 6400 (Jeol, Tokyo, Japan). Sputtering
and analysis of elemental composition of the latter materials for
depth profiling measurements were carried out by an SNMS-in-
strument model INA 3 (Leybold Heraeus; now supported by Specs
in Berlin, Germany). The diameter of the analyzed area was 3 mm.
Results and discussion
Spectroscopic measurements and PLS calibrations
For process monitoring of fluorinated sulfur compounds,
the mid-infrared spectral range from 1600–500 cm–1 is of
immense importance, because strong absorbing funda-
mental and combination vibration bands exist for these
molecular species. For illustration, a transmission spec-
trum of pure SF6 is presented in Fig. 2a. The opaque inter-
vals in the SF6 spectrum due to strong absorption bands
are not accessible for quantitative analysis. The lower
trace is the difference spectrum of a contaminated SF6
sample from a partial discharge experiment with a total
discharge energy of 50 mJ after scaled absorbance sub-
traction of the dominating SF6 signatures, providing an
overview of the absorption bands of components found at
significant concentrations under the experimental condi-
tions chosen (see Fig. 2b).
With the procedure used to date, the stressed SF6 gas
was directly expanded into an evacuated cell outside the
Faraday cage that housed the high-voltage testing equip-
ment. Such a procedure requires a test gas volume of
about 1 L at 100 kPa. This sample amount can be consid-
erably reduced using the alternative gas cell equipment
based on a silver halide fiber-coupled, internally gold-
coated lightpipe. Compared to a conventional 10 cm gas
cuvette, the optical pathlength of our mini-cell is larger by
more than a factor of two, which is advantageous for trace
gas analysis. However, the opaque spectral intervals due
to the strong SF6 absorption bands are slightly increased.
For quantitative analysis of the main trace gases from
SF6 decomposition products, PLS calibrations were car-
cmaxd (Pa) cave (Pa) σpopf (Pa) SEPg (Pa)
322 33 73 1.09
145 35 34 0.83
1.89 0.19 0.37 2.55×10–2
0.52 0.15 0.16 1.17×10–2
34.7 2.6 4.4 2.08×10–1
cM number of spectral variables used for quantification
dc
max maximum partial pressure
ec average partial pressure
av
fσ
pop standard deviation of the concentration
values within the cal-
ibration sample population
gSEP standard error of prediction obtained by leave-one-out cross-
validation
642

ried out making use of only a small number of spectral SF6 decomposition under partial discharge conditions
variables. Broad spectral intervals were selected on the
basis of the most intense component absorption bands that The metallic materials tested by us are of special interest
were not overlapped by strong SF6 absorptions [15]. for their usage in high-voltage equipment. Therefore, the
Further optimization was carried out with emphasis on influence of different electrode surfaces on the formation
a selection of a reduced spectral variable set, using the of reactive sulfuroxyfluorides and SFx gases during dis-
strategy of extreme values of the regression vectors [15, charges and sparks under an SF6 atmosphere was studied.
19]. The resulting narrow spectral intervals taken into ac- In the past, interesting dependencies were found for sulfur
count for each component PLS calibration are listed in and fluorine concentrations in the gaseous compartment
Table 1.
The results of such PLS regressions including standard
errors of prediction (SEP) are summarized in Table 1.
These values when doubled also provide estimates for the
infrared spectroscopic detection limits based on a 95%
confidence level. In particular for the compounds of SOF4
and SO2F2, exemplary scatter plots resulting from leave-
one-out cross-validation are shown in Fig. 3. For SOF4
(see Fig. 3a) a clustering is obvious for samples produced
under partial discharge conditions, which resulted in much
larger concentrations than obtained during sparking.

Fig. 4a–c Amount of decomposition products obtained by partial

Fig. 3a,b a Calibration results for SOF4 and b for SO2F2. Scatter
plots of predicted versus reference component partial pressure val-
ues for PLS calibration models with optimally selected wavenum-
ber variables (results are for experiments with sparking and partial
discharges under variable conditions; prediction by leave-one-out
cross-validation)
discharges with varying total discharge energy using different
metallic counter electrode materials (SF6 pressure inside the dis-
charge chamber: 250 kPa): a Results for the main products SOF4,
SO2F2, and SOF2 from experiments based on the following condi-
tions: total discharge energy 17 mJ, duration 24 h, b results for the
main products SOF4, SO2F2, and SOF2 from experiments based on
discharge conditions with a total energy of 54 mJ (duration 3 days),
c results for the trace compounds SiF4, S2F10, and SO2 from exper-
iments based on the same discharge conditions as in b
 643

and in the solid electrodes [7]. For corona decomposition,

the passivation of aluminum surfaces was also found by
other authors [6] to lead to very low production rates of
SOF2 above a certain transported electric charge. Results
for gaseous SF6 decomposition products are reported by
us using materials that were previously not considered.
The duration of the experiments was either 24 h with a to-
tal discharge energy of 17 mJ on average for the experi-
ments using counter electrodes of different metallic mate-
rials, or three days to reach a total energy of 54 mJ.
The results for the major decomposition products from
gas-phase measurements with the conventional gas cell of
10 cm optical pathlength are shown in Fig. 4. The ratios
between the concentration values of the compounds SOF4,
SOF2, and SO2F2 were found to be rather independent of
the plane electrode material [4], which is in contrast to
previous results obtained under sparking conditions [3].
There was some scatter in the decomposition rates depend-
ing on the material, and the largest by-product amounts
were seen for tungsten, which is different from the other
materials, presumably because WF6 can be formed which
is known to be volatile (see also Fig. 4a). A smaller vari-
ance in decomposition rates was observed when the by-
product amounts were determined after the partial dis-
charge experiments conducted over three days (see also
Fig. 4b). The amount values found for other trace com-
pounds are shown in Fig. 4c. Owing to the reduced scatter,
mean values and standard deviations have been calculated
which can be used for the calculation of component spe-
cific generation rates (Table 2).
A comparison of the generation of the main decompo-
sition products was made for the low- and high-energy
discharge experiments. If the rates to be expected are Fig. 5a,b a SEM image of a needle of stainless steel after a partial
based on the mean discharge energies, a factor of 3.1 is discharge experiment with a duration of 24 h, b surface of an alu-
calculated (based on discharge duration, the factor is 3), minum counter electrode under the same discharge conditions as
applied in a (magnification can be assessed from the 100 µm
whereas the concentration ratio between the results from scales)
the higher versus lower discharge energy experiments
yields a value of 2.1 for SOF2, 2.5 for SOF4, and 2.4 for
SO2F2. One plausible explanation is that some loss of the this is the slight increase noticed for SO2 concentrations
decomposition products occurs due to water adsorbed on and the significantly larger production rate for SiF4 that is
the walls of the discharge equipment. An indication for formed from silicon of the stainless steel material of the
discharge chamber when exposed to the hydrolysis prod-
Table 2 Amount of several decomposition products obtained for uct of SOF2, namely HF. Furthermore, large amounts of
partial discharge experiments with different counter electrode ma- HF were observed for experiments using partial discharge
terials such as Al, Cu, Ag, W, and W/Cu, respectively; total mean conditions [15]. The concentration increase of S2F10 was
discharge energies were either 17 mJ over a duration of 1 day or in line with the increase seen for SOF4 and SO2F2.
54 mJ over 3 days; ratio is calculated from mean amounts at 54
and 17 mJ; SF6 insulation gas pressure 250 kPa (mean results and An inspection of the point electrode showed that the
standard deviations σ are from 5 experiments) needle tip suffered severe wear-off (Fig. 5a), and that after
24 h the surface of the counter electrode was flecked by
Decom- Amount ± σ Amount ± σ Ratio small spots of compounds containing SxFy fragments (see
position (µmol) (µmol)
Product (17 mJ discharge (54 mJ discharge Fig. 5b). The spot sizes did not increase significantly when
energy) energy) the duration of partial discharge experiments was tripled.
Systematic studies on gaseous compound production
SOF2 25.2 ± 5.6 53.7 ± 10.2 2.13 rates were carried out recently by Casanova et al. under
SOF4 73.2 ± 13.1 179.7 ± 6.9 2.45 various conditions using aluminum or stainless steel, ap-
SO2F2 24.0 ± 8.5 56.7 ± 7.2 2.36
plying negative DC or 50 Hz AC corona discharges and
SO2 6.0 ± 2.0 16.2 ± 1.3 2.69
different moisture levels [9, 20]. The latter authors ob-
S2F10 11.0 ± 1.8 27.2 ± 3.5 2.46
served no dependence of product formation on the plane
SiF4 12.9 ± 2.3 49.4 ± 5.8 3.84
electrode material either.
644

SF6 decomposition under spark conditions

The by-product generation is completely different under

sparking conditions, under which the counter electrode
material is much more affected than under partial dis-
charge conditions. In a recent paper [3], we reported on
SF6 decomposition using aluminum, silver, and copper
electrodes under single spark energies of about 0.15 J
spark–1. An electrically insulating passivation of the metal
surface after application of 3000 sparks was observed
only for aluminum. In contrast, a linear dependence of
SOF2 generation on the number of sparks (0.15 J spark–1)
could be observed, for example, in the case of copper used
as the counter electrode material (see Fig. 6a).
To understand the different material behavior, investi-
gations on the chemistry at the electrode surfaces were
made applying depth profiling of the fluorine and sulfur
content below the metallic surfaces with SNMS, XPS, and
SEM-EDX methodologies [7]. As an example, the SNMS
depth profiles obtained for fluorine in copper after exper-
iments with a varying number of sparks are shown in Fig. 6b.
For aluminum, fluorine is the dominating element with a
lower sulfur content found below a thin top-layer of AlF3.

Fig. 7 Amount of the main composition products SOF2, SOF4, and

SO2F2 in SF6 samples generated by experiments with a spark en-
ergy of 0.15 J spark–1 using tungsten and a tungsten/copper alloy
as electrode materials versus the total number of sparks with a
pressure of 250 kPa for the insulating SF6 gas

Sulfur was found in low concentrations in all metallic ma-

Fig. 6a,b a Amount of gaseous SOF2 found after experiments
with a varying number of sparks with an average energy of 0.15 J
spark–1 (mean values of experiments using a total SF6 pressure of
250 and 300 kPa, respectively), b depth profiles of fluorine content
in a copper counter electrode after a varying number of sparks ob-
tained by sputtering and SNMS analysis
terials. On copper, sulfur was detected only in the topmost
layer, whereas for silver a thicker mixed zone containing
sulfur and fluorine followed.
Our studies were extended to materials that are more
relevant for switch-gear electrodes, that is tungsten and a
tungsten/copper alloy. The generated amounts of the main
products of SOF2, SOF4, and SO2F2 are presented in Fig. 7
for these two materials. Similar to our previous results,
there is a linear relationship between the formation of de-
composition products and the number of sparks, which
may be influenced by the electrode chemistry as observed
previously, for example for aluminum and silver.

Table 3 Amount of several
decomposition products ob- Decomposition Al
tained for spark experiments Product (µmol)
with different counter electrode 0.15 J spark–1
materials such as Al, Cu, Ag,
W, and W/Cu, respectively; SOF2 5.26 ± 0.18
total mean energies were either SOF4 0.86 ± 0.07
0.15 or 1.0 J spark–1; regres- SO2F2 0.22 ± 0.01
sion results are for 1000 sparks;
SF6 insulation gas pressure 1.0 J spark–1
250 kPa SOF2 53.1 ± 3.5
SOF4 3.36 ± 0.63
SO2F2 0.60 ± 0.15
Whereas the generation of SOF2 and SO2F2 runs com-
pletely parallel for tungsten and the respective tungsten/
copper alloy material with a larger rate always found for
tungsten electrodes, a larger SOF4 production rate was ob-
served when the alloy counter electrode material was em-
ployed, consistently for all experiments with 1000–12,000
sparks per experiment.
Further experiments were carried out with a much higher
energy of 1 J spark–1, since SF6 decomposition yields
were previously found to be energy-dependent [5]. The
total number of sparks was set to 3000, and further exper-
iments with 300, 600, and 1200 sparks in total were also
considered. A significantly larger amount of SOF2 was
found using aluminum electrodes versus silver or copper,
similar to the results obtained under low energy sparking
conditions. The amount of the other two products, SOF4
and SO2F2, was less affected by the electrode material
(see Table 3).
The regression results from the straight line fit, as
passed through the origin for the generated amounts of the
main decomposition products versus number of sparks,
are summarized in Table 3, which also contains our results
from previous experiments with aluminum, silver, and
copper [3], using low spark energies as for the experi-
ments with tungsten and the tungsten/copper alloy that are
shown here. A similar pattern of material dependence of
the SOF2, SOF4, and SO2F2 formation was observed for
the low energy sparks, as discussed for aluminum, silver,
and copper in the preceding paragraph. When the concen-
tration ratios for the individual components are calculated
using the two different spark energies, the higher spark
energy yields much more SOF2 than expected from the in-
crease in spark energy, whereas much less SOF4 and
SO2F2 are formed (only half of the amount to be expected
on the basis of spark energy). These results show the com-
plex nature of SF6 decomposition, where different electri-
cal conditions and the materials involved have a signifi-
cant impact on the production rates.
The production rates found by us can be compared
with those recently published by Pradayrol et al. [5] ob-
tained for conditions with capacitor discharging (3.6 J
spark–1). The authors used SF6 that contained the follow-
ing impurities: [H2O] <92 µg L–1 and [O2] <6.5 µg L–1.
Two different cell preparations were taken into account: 1)
“very clean cell” conditions with several heating-
pumping cycles, or 2) “clean cell” conditions with just an
645
Ag Cu W W/Cu
(µmol) (µmol) (µmol) (µmol)
1.11 ± 0.04 2.23 ± 0.06 2.87 ± 0.33 2.03 ± 0.10
0.84 ± 0.08 1.01 ± 0.03 0.79 ± 0.04 1.22 ± 0.07
0.07 ± 0.01 0.14 ± 0.01 0.31 ± 0.02 0.22 ± 0.02
9.64 ± 0.35 12.3 ± 0.8
3.48 ± 0.25 3.00 ± 0.32
0.45 ± 0.03 0.44 ± 0.04
evacuation to about 1 Pa before filling was applied, which
is close to our experimental conditions. The data obtained
for plane electrodes of aluminum using gas chromatogra-
phy are in excellent agreement with our results, express-
ing the production rates in nmol J–1 (the results of Praday-
rol et al. [5] are given in parentheses): SOF2 53.1 (49.8 in-
cluding traces of SF4), SOF4 (our value 3.36), and SO2F2
(our value 0.6) giving a total production rate of 3.96
(3.84). The formation of S2F10 was very low under the
„clean cell“ condition, which is in line with our findings
for the case of sparks (below our IR spectroscopic detec-
tion limits). For the other electrode materials investigated
by us, data from the literature was not available for com-
parison.
Conclusions
Infrared spectroscopy was successfully employed for the
assessment of SF6 decomposition under different high-
voltage conditions with the simulation of partial dis-
charges and sparks. Main products and trace components
could be measured reliably using our spectroscopic set-up
and measurement settings. Therefore, the infrared method
can also be used for monitoring the cleaning and recycling
process of contaminated SF6. The application of IR fibers
is especially interesting, since it allows the use of hollow
tubes as miniature cells and a simple optical coupling to
the interferometer and its detector. Furthermore, it gives
the flexibility for remote sensing when a spatial separa-
tion of the spectrometer from the high-voltage equipment
to be monitored is needed.
The material dependence of the SF6 decomposition
was studied systematically, but an impact was only found
for sparking conditions. There is a severe increase for
SOF2 using aluminum as counter electrode material, lead-
ing to more gas-phase products which runs parallel to a
reduced solid-phase chemistry turn-over rate. Partial dis-
charges can be recognized by gas-phase analysis, based
on a main decomposition product concentration ratio of
SOF2 and SOF4, whereas under sparking conditions the
electrode materials have a significant influence on the
product generation rates for gaseous compounds, which
gives a larger scatter of concentration ratios.
Acknowledgements The financial support from the Ministerium
für Schule, Wissenschaft und Forschung des Landes Nordrhein-
646
Westfalen and the Bundesministerium für Bildung und Forschung
is gratefully acknowledged.
References
1. Christophorou LG, Olthoff JK, van Brunt RJ (1997) IEEE
Electrical Insulation Mag 13:20–24
2. Heise HM, Kurte R, Fischer P, Klockow D, Janissek PR (1997)
Fresenius’ J Anal Chem 358:793–799
3. Kurte R, Heise HM, Klockow D (1999) J Mol Struct 480/481:
211–217
4. Kurte R, Heise HM, Klockow D (1999) In: Itoh K, Tasumi M
(eds) Fourier transform spectroscopy. 12th International Con-
ference, Waseda University Press, Tokyo, pp 269–270
5. Pradayrol C, Casanovas AM, Hernoune A, Casanovas J (1996)
J Phys D: Appl Phys 29:1941–1951
6. Belarbi A, Pradayrol C, Casanovas J, Casanovas AM (1995)
J Appl Phys 77:1398–1406
7. Beyer C, Jenett H, Klockow D (2000) IEEE Transact Dielectr
Electr Insulation 7:234–240
8. Pradayrol C, Casanovas AM, Aventin C, Casanovas J (1997)
J Phys D: Appl Phys 30:1356–1369
9. Vial L, Casanovas AM, Coll I, Casanovas J (1999) J Phys D:
Appl Phys 32:1681–1692
10. Coll I, Casanovas AM, Vial L, Gleizes A, Casanovas J (2000)
J Phys D: Appl Phys 33:1348–1359
11. Casanovas AM, Coll I, Pradayrol C, Casanovas J (1998) J Phys
D: Appl Phys 31:2835–2845
12. Coll I, Casanovas AM, Vial L, Gleizes A, Casanovas J (2000)
J Phys D: Appl Phys 33:221–229
13. Johansen I-R, Lines GT, Honne A, Midtgaard T (1997) Appl
Spectrosc 51:1540–1546
14. Hart BK, Berry RJ, Griffiths PR (1999) Field Anal Chem Tech-
nol 3:117–130
15. Kurte R, Heise HM, Klockow D (2001) J Mol Struct 565/566:
505–513
16. Kosterev AA, Curl RF, Tittel FK, Gmachl C, Capasso F, Sivco
DL, Baillargeon JN, Hutchinson AL, Cho AY (2001) Laser
Physics 11:39–49
17. Held M, Kurrat M, Rickert F, Baumbach JI, Segundo SMA,
Soppart O, Klockow D (1997) Elektrizitätswirtschaft 96:290-
293
18. Küchler A (1996) Hochspannungstechnik, Grundlagen – Tech-
nologie – Anwendungen, VDI-Verlag, Düsseldorf
19. Heise HM, Bittner A (1997) Fresenius’ J Anal Chem 359:93-
99
20. Belarbi A, Pradayrol C, Casanovas J, Casanovas AM (1995)
J Appl Phys 77:1398–1406