Journal of Molecular Structure 480–481 (1999) 211–217

Analysis of spark decomposition products of SF6 using multivariate

mid-infrared spectrum evaluation
R. Kurte, H.M. Heise*, D. Klockow
Institut für Spektrochemie und Angewandte Spektroskopie, Bunsen-Kirchhoff-Str. 11, D-44139 Dortmund, Germany
Received 25 August 1998; accepted 30 September 1998

Abstract
Infrared spectroscopy is a powerful tool for the analysis of gaseous by-products generated from electrical discharges in sulfur
hexafluoride gas. The spark decomposition of SF6 was studied under pressures of 250 kPa and 300 kPa, which were provided
for the insulating gas inside a discharge chamber constructed from stainless steel and Teflon. The sparks were generated
between a point-plane configuration by discharging a capacitor. The spark energies were in the range between 0.15 and 1.0 J/
spark. The electrode materials used were copper, aluminium and silver. Quantification was carried out using classical least-
squares with spectral data from selected intervals. The main-products found were the sulfuroxyfluorides SOF4 and SOF2 with
concentrations correlated to the spark energy. Moreover, we were able to measure HF at spark energies in the upper range. The
experimental conditions and practical aspects for reliable quantitative analysis of reactive sulfurfluoride species are discussed.
䉷 1999 Elsevier Science B.V. All rights reserved.
Keywords: Fourier transform infrared spectroscopy; Trace gas analysis; Sulfur hexafluoride decomposition; By-product monitoring

1. Introduction Special conditions, i.e., the presence of oxygen and

moisture causing formation of oxygenated species,
The increased application of SF6 in gas-insulated were investigated in the past, see e.g., [1,5]. The
electrical equipment, such as switchgear or transmis- decomposition of SF6 further depends on gas pressure,
sion lines, has led to many recent analytical investiga- electrode materials and the inner surface materials of
tions on the stability and decomposition of SF6 [1–3]. the equipment. In particular, the influence from elec-
Although SF6 shows great chemical inertness under trode materials, such as aluminium, copper and stain-
the absence of electrical stress, in the presence of arcs, less steel, on the production of SF6 break-down and
sparks or partial discharges it decomposes. The condi- reaction products were studied, for example, for
tions for SF6 decomposition were described in detail sparking conditions [1] and for corona discharges
previously [4]. Among the products found are species (aluminium and stainless steel) [6].
such as SOF2, SO2F2, SOF4, S2F10, SO2, HF, SiF4 and Recent assays of decomposition products were
others. Recent research has focused on the formation carried out by gas chromatography [1], gas chromato-
of S2F10, which is a by-product of high toxicity. graphy/mass spectrometry [7] or infrared spectro-
scopy [5,8,9], although a quality assessment of SF6
* Corresponding author. Tel.: ⫹ 49-231-1392215; fax: ⫹ 49-
can also be achieved using ion mobility spectrometry
231-1392120. [10]. Infrared spectroscopy is a splendid analytical
E-mail address: [email protected] (H.M. Heise) tool for a direct and in-situ multicomponent
0022-2860/99/$ - see front matter 䉷 1999 Elsevier Science B.V. All rights reserved.
PII: S0022-286 0(98)00642-5
212 R. Kurte et al. / Journal of Molecular Structure 480–481 (1999) 211–217

Fig. 1. Schematic diagram of the experimental system used for the infrared spectrometric analysis of sparked SF6 gas.

quantification. With different optical pathlengths experiment. The top layer composition of the plane
available, a broad range of concentrations can be metal plates was analyzed by the methods of SNMS,
dealt with. For this work, we employed IR-spectro- XPS and SEM-EDX. These results will be reported
scopy for the simultaneous analysis of several by- elsewhere [12].
products and their stability using multivariate spec- Before the spark experiments were started, the
trum evaluation based on classical least-squares high voltage chamber was prepared in the
fitting. Results from experiments with varying spark following way. After electrode installation, the
energies are presented. The electrode materials cell was filled with 200 kPa SF6 (99.9%) from
studied were copper, aluminium and silver. Besides Messer Griesheim (Krefeld, Germany) and evacu-
the sulfur compounds and SiF4, we were able to ated for 5 min after a short time of SF6 gas expo-
measure HF at spark energies in the upper range, sure. The chamber was refilled with the gas for a
which has not been detected using other methods second time and repeatedly evacuated to a residual
reported so far in the literature. pressure of 1 hPa. Finally, the cell was refilled with
SF6, setting the pressure for the isolating gas to 250
and 300 kPa, respectively.
2. Experimental Within our study, infrared spectra were recorded by
using an IFS 113v FT-IR spectrometer from Bruker
For different high-voltage experiments, a discharge (Karlsruhe, Germany) equipped with a KBr-beams-
chamber from NiCr-stainless steel and Teflon TM was plitter and a mercury cadmium telluride (MCT)
constructed in our laboratory for the simulation of SF6 detector (cut-off at 530 cm ⫺1) operated at liquid
decomposition [11]. The chamber (internal volume nitrogen temperature. To reach the appropriate spec-
350 ml) housed a point-plane electrode configuration tral signal–to–noise ratio a total measurement time of
with an average gap distance of 1.2 mm. The sparks 3.5 min was necessary to accumulate hundred inter-
were generated by discharging a capacitor, and ferograms in total for each spectrum with a spectral
experiments with spark energies in the range between resolution of 0.5 cm ⫺1. The interferograms were
0.15 and 1.0 J/spark were carried out. The plane and Fourier–transformed after correction of the non-linear
counter electrodes were from aluminium, copper or detector response to the photon flux noticed in the
silver. Owing to the deterioration of the point-plane interferogram domain by using the spectrometer soft-
materials, a new set of electrodes was used for each ware. Such data processing is necessary to avoid
 R. Kurte et al. / Journal of Molecular Structure 480–481 (1999) 211–217 213

systematic photometric errors within the multivariate

spectrometric data evaluation [13].
The stressed SF6 gas from the discharge chamber
was directly expanded into an evacuated two-cell
system (see Fig. 1), containing a heatable transmit-
tance cell of 10 cm optical pathlength (cell body
from stainless steel) delivered by SPECAC
(Orpington, UK) and a multiple reflection cell with
a pathlength set to 3 m (cell body from borosilicate
glass; Long Path Mini cell from Infrared Analysis
(Anaheim, USA) allowing the simultaneous analysis
of trace and significantly concentrated components.
A similar set of reference spectra (see Fig. 2) as
previously discussed [8] was used, apart from that
of SO2F2, which had been prepared by Dr. M. Lieb
from the Department of Inorganic Chemistry of the
Ruhr–University Bochum. The quantification of HF
was done by using a spectrum from the HITRAN–
library [15], which was convoluted by a sinc 2 function
to adjust for the experimental spectral resolution.
Fig. 2. Absorbance spectra showing significant absorption bands
suitable for quantitative analysis of by-products in SF6. The spectra
were recorded by means of a 10 cm cell; the following sample
3. Results and discussion
pressures were used: SO2F2 0.20 hPa, SOF4 0.30 hPa, SOF2
1.11 hPa, SO2 2.55 hPa, SF4 0.27 hPa and CF4 0.22 hPa. The sample
pressure for S2F10 (0.92 hPa) was estimated from literature data For process monitoring of fluorinated sulfur
[14]; the spectrum of SiF4 was taken from a commercial library compounds, the mid-infrared spectral range from
[17] (sample pressure equivalent for a pathlength of 10 cm: 1600 – 500 cm ⫺1 is of immense importance, because
0.50 hPa). strong absorbing fundamental and combination vibra-
tion bands exist for these species. Another compound
produced under electrical stress is HF which can be

Fig. 3. Absorbance spectrum of HF (partial pressure: 0.04 hPa) in sparked SF6 (total cell pressure 865 hPa) recorded with an optical pathlength
of 10 cm and a spectral resolution of 0.5 cm ⫺1. The corresponding spark experiment was carried out using copper electrodes and 3000 sparks at
250 kPa of SF6 discharge gas pressure (spark energy ca. 1 J/spark).
214 R. Kurte et al. / Journal of Molecular Structure 480–481 (1999) 211–217

Fig. 4. Gas spectra from a spark experiment recorded at 10 cm

optical pathlength and a total cell pressure of 676 hPa: (A) Absor-
bance spectra of stressed (upper trace, offset) and pure SF6 (lower
trace): copper electrode, 12000 sparks with energy of 0.15 J/spark,
total pressure within discharge chamber: 250 kPa. (B) Difference
spectrum derived showing SF6 decomposition products.

reliably quantified using data within a narrow spectral

interval around 4000 cm ⫺1. An exemplary spectrum Fig. 5. Concentration dependency of by-products on the residence
of this compound, as obtained for spark energies in the time of the sparked SF6 gas inside the discharge chamber (spark
upper range around 1 J/spark, is shown in Fig. 3. The energy ca. 1 J/spark): (A) main decomposition products SOF2 and
spectrum between 4095 and 3800 cm ⫺1 was used for SOF4, (B) SO2F2 and S2F10, (C) most reactive compounds SiF4, SF4
and HF.
least-squares fitting. A recent assay for vapour phase
HF using FT-IR spectroscopy with single line evalua-
tion was described [16]. However, the multivariate of 10 cm optical pathlength. The lower diagram gives
approach is favoured by us, because it gives a reduced the difference spectrum after scaled absorbance
uncertainty estimate for the HF concentrations subtraction of the dominating SF6 absorbance signa-
compared to a single line evaluation. It is certainly tures. Clearly the absorption bands originating from
an advantage of fast scanning FT-IR spectrometers the components with significant concentrations can be
that the whole mid-infrared spectral range can be seen. The spectral intervals for the quantitative deter-
measured simultaneously. mination were as follows: SO2F2 1535–1450 cm ⫺1;
An exemplary set of sample and reference spectra, SO2 1357–1330 cm ⫺1; CF4 1300–1280 cm ⫺1; SiF4
i.e., contaminated and pure SF6, is shown in the upper 1100–1013 cm ⫺1; SOF2, SOF4 and S2F10 830–
diagram of Fig. 4, which was recorded using the cell 690 cm ⫺1.
 R. Kurte et al. / Journal of Molecular Structure 480–481 (1999) 211–217

Fig. 6. Results obtained for the infrared spectrometric determination of SOF2 (1st column), SOF4 (2nd column) and SO2F2 (3rd column) in dependence of the electrode material:
aluminium (1st row), silver (2nd row) and copper (3rd row); spark energy ca. 0.15 J/spark, the pressure of the insulating SF6 gas was set to 250 kPa and 300 kPa, respectively.
215
216 R. Kurte et al. / Journal of Molecular Structure 480–481 (1999) 211–217

The stability of generated by-products was studied given for the experiments with low energy sparks.
over a residence time of the sparked SF6 gas within the Components such as SOF2, SOF4 and SO2F2 show
discharge chamber up to one hour. Exemplarily, rather high concentrations varying with the number
results are shown in Fig. 5 using copper electrodes. of sparks and different electrode materials. For alumi-
The upper diagram illustrates the time dependency for nium it was found that a passivation of the metal
the highly concentrated species SOF2 and SOF4. The surface by AlF3 arises after about 3000 sparks [12],
uncertainties are calculated for a 95% confidence limit preventing further spark generation. In particular, the
as derived from the standard deviation of the spectral concentrations of SOF2 and SO2F2 are much higher
residue after fitting the spectra. The general trend is with aluminium than in the other experiments using
that a decrease of SOF2 is followed by an increase in silver and copper, respectively. This increased
SOF4. Moreover, the generation of SO2F2 is a result of concentration of reaction products in the gas phase
hydrolysis of the latter compound. There is a slight correlates well with the reduced concentrations of
decrease noticeable for the S2F10 concentration, which sulfur and fluorine found in the corroded electrode
decays into SF6 and SF4 (see Fig. 5B). The concentra- surfaces for aluminium as opposed to copper and
tion values dependent on the residence time of the especially silver [12]. Finally, a similar result for the
most reactive species, i.e., HF, SF4 and SiF4, are production rates of gaseous SOF2 and SF4, as well as
presented in Fig. 5C. The formation of SiF4 was SO2F2 and SOF4 for aluminium and copper electrodes
surprising and possibly originated from the silicon has recently been found by Pradayrol et al., [1] in their
within the stainless steel. The 1.4301 steel used has a spark decomposition experiments.
silicon content ⱕ1%. The increase of SiF4 is negatively
correlated with the decrease of the other two
4. Conclusions
compounds. Further investigations have to be carried
out on the time dependent mass balances for the reac-
The decomposition of SF6 under sparking condi-
tions to ascertain the dynamics of composition changes.
tions was investigated using FT-IR spectroscopy.
The formation of SO2 was postulated by many
This technique provides the sensitivity necessary to
researchers, because it is supposed to be a stable end
determine reliably concentrations in the low ppm and,
product from hydrolysis reactions based on S(IV)
occasionally, also in the sub-ppm range. By
chemistry. However, after our spark experiments, no
connecting the discharge chamber directly to the spec-
traces of SO2 were detected in any of the investiga- trometric measurement system, sample manipulation
tions of stressed SF6. An interesting experiment was
could be minimized. Owing to the very strong SF6
carried out using an SF6 gas sample containing an
absorption spectrum, it was important to have
impurity of SO2 owing to its enrichment when
different optical pathlengths available, to achieve
emptying the commercial gas cylinder (original SF6
high sensitivity and broad spectral coverage free
purity 99.9%). The contaminant volume fraction was
from SF6 matrix interference. The main by-products
23 ppm, as determined from its infrared spectrum
found during our discharge experiments were SOF2,
(spectral range 1360–1325 cm ⫺1). The strongest SO2
SOF4 and SO2F2 which showed the expected linear
absorption band is at 1360 cm ⫺1 (Q-branch), and just concentration dependence with spark number and
the P-branch rotational fine structure could be seen in
corresponding integral energy. Owing to the reactivity
the spectrum of the SF6 gas contaminated by SO2, but
of the components studied, the gas sampling is very
other branches were superimposed by a strong SF6
important. Rapid or in-situ analysis must be recom-
absorption band. Such a contaminated SF6 sample
mended, if one is interested in the true component
was employed for spark experiments with 1200 and
concentration profiles of decomposition products
3000 sparks using copper electrodes. In this case, the
within stressed SF6 insulation gas.
SO2 was decomposed forming sulfuroxyfluorides.
This is astonishing, since the SO2 compound is
considered to be a stable end product within the Acknowledgements
decomposition process.
In Fig. 6 a summary of the analytical results is The authors acknowledge gratefully the financial
 R. Kurte et al. / Journal of Molecular Structure 480–481 (1999) 211–217 217

support by the Ministerium für Schule und [7] O. Kóréh, T. Rikker, G. Molnár, B.M. Mahara, K. Torkus, J.
Weiterbildung, Wissenschaft und Forschung des Borossay, Rapid Comm. Mass Spectrometry 11 (1997) 1643.
[8] H.M. Heise, R. Kurte, P. Fischer, D. Klockow, P.R. Janissek,
Landes Nordrhein–Westfalen and the Bundesmi- Fresenius J. Anal. Chem. 358 (1997) 793.
nisterium für Bildung, Wissenschaft, Forschung [9] H.M. Heise, P.R. Janissek, R. Kurte, P. Fischer, S.M.A.
und Technologie. Dr. M. Lieb from the Ruhr– Segundo, D. Klockow, Mikrochim. Acta [Suppl.] 14 (1997)
Universität Bochum is thanked for the supply of 193.
sulfurylfluoride. [10] O. Soppart, J.I. Baumbach, S.M. Alberti, D. Klockow, Conf.
Proc. 10th Int. Symp. High Voltage Engineering, Montreal,
Canada, Vol. 4 (1997) 147.
[11] M. Held, M. Kurrat, F. Rickert, J.I. Baumbach, S.M.A.
References Segundo, O. Soppart, D. Klockow, Elektrizitätswirtschaft 96
(1997) 290.
[1] C. Pradayrol, A.M. Casanovas, A. Hernoune, J. Casanovas, J. [12] C. Beyer, H. Jenett, D. Klockow, IEEE Trans. Electr. Insul.,
Phys. D: Appl. Phys. 29 (1996) 1941. submitted.
[2] C. Pradayrol, A.M. Casanovas, C. Aventin, J. Casanovas, J. [13] H.M. Heise, P.R. Janissek, P. Fischer, Mikrochim. Acta
Phys. D: Appl. Phys. 30 (1997) 1356. [Suppl.] 14 (1997) 301.
[3] A. Belarbi, C. Pradayrol, J. Casanovas, A.M. Casanovas, J. [14] J.K. Wilmshurst, H.J. Bernstein, Can. J. Chem. 35 (1957) 191.
Appl. Phys. 77 (1995) 1398. [15] USF HITRAN-PC, Optical Spectra Data Base, ONTAR
[4] F.Y. Chu, IEEE Trans. Electr. Insul. EI-21 (1986) 693. Corporation, North Andover, MA (USA).
[5] M. Piemontesi, R. Pietsch, W. Zaengl, Proc. IEEE Int. Symp. [16] K.R. Brenneman, R.J. Donohue, Appl. Spectrosc. 48 (1994)
Electr. Insul., Pittsburgh, PA (1994) 499. 808.
[6] C. Pradayrol, A.M. Casanovas, A. Belarbi, J. Casanovas, J. [17] QASOFT, Infrared spectra for quantitative analysis of gases,
Phys. III France 5 (1995) 389. Infrared Analysis, Inc., Anaheim, CA 92801 (USA).