Professional Documents
Culture Documents
A Novel Fuzzy-Logic Approach For Furan Estimation in Transformer Oil
A Novel Fuzzy-Logic Approach For Furan Estimation in Transformer Oil
net/publication/261045707
CITATIONS READS
16 4,971
3 authors, including:
All content following this page was uploaded by A. Abu-Siada on 10 December 2015.
Abstract—An ultraviolet-to-visible (UV-Vis) spectroscopic transformer oil. Since this method requires specific skills and
analysis based on the fuzzy-logic approach has been developed expensive equipment to conduct, most utility companies out-
for furan content measurement in transformer oil. Following the source the furan test to the relevant service companies. The use
successful identification and quantification of furan derivatives in
transformer oil by the ASTM D 5837 standard, the new approach of new and improved technology, such as spectroscopy, is also
is able to approximate the furan content with a more convenient increasingly important in recent power transformer condition
and economical method. Since the furan concentration level would monitoring [5]. This research paper proposes a very convenient
determine the absorption intensity in the UV-Vis spectral range, alternative to estimate furan concentration in transformer oil
the fuzzy-logic software model developed would exploit this char- using ultraviolet-to-visible (UV-Vis) spectroscopy using the
acteristic to aggregate the furans content level in transformer oil.
The UV-Vis spectral response sensitivity to ambient temperature fuzzy-logic approach. The UV-Vis spectral response of the
was also studied. The proposed technique provide a convenient oil sample gives sufficient information to predict its furan
alternative to conventional techniques, namely: high-performance contents based on the fuzzy-logic approach. A series of tests
liquid chromatography (HPLC) or gas chromatography-mass was conducted on laboratory aged, onsite collected, and diluted
spectrometry (GC/MS) methods. transformer oil samples. A fuzzy-logic model is developed
Index Terms—Furfurals, fuzzy logic, transformer insulation oil, based on the input parameters taken from the UV-Vis spectral
ultraviolet to visible spectroscopy. response to estimate the furan concentration with up to 95%
accuracy compared with conventional measurement methods.
I. INTRODUCTION II. FURAN FORMATION MECHANISM
The furanic compounds in transformer oil that are detectable
(1)
TABLE I
SELECTED OIL SAMPLES’ TEST RESULT BY GC/MS
TABLE II
BANDWIDTH LENGTH AND ABSORPTION PEAK
AT DIFFERENT FURAN CONTENTS
found were almost identical to Fig. 3, but there were slight vari-
ations in the bandwidth and absorption peaks. Changes were
recorded and plotted in Figs. 5 and 6. As shown in Fig. 5, for
Fig. 7. Fuzzy inference flowchart for model development. the same 2-FAL concentration, the ambient temperature incre-
ment from 20 C to 25 C expands the bandwidth by generally
10 nm whereas absorption peaks for the same increment in am-
bient temperature overlap at different furans concentration as
shown in Fig. 6.
V. FUZZY-LOGIC MODELING
Fuzzy logic is “soft-computing” methodology that has vari-
ables whose values are generally words that are based on the
fuzzy IF-THEN rule. Even though words are inherently less pre-
cise than numbers, their use is closer to human intuition because
Fig. 8. Gaussian membership functions for bandwidth wavelength. it is built on rules from human experience and not based on for-
mulas. It exploits the tolerance of imprecision and uncertainty
to achieve tractable and lower solution cost. It is proposed in
the almost parallel relationship between the bandwidth wave- this paper to use fuzzy-logic theory in furans content estimation
length and maximum peak in estimating the furans concentra- because it was found exceptionally useful in estimating furan
tion. Therefore, these two parameters would be reliable indica- contents based on the input parameters of the bandwidth and
tors for furans measurement by UV-Vis spectroscopy. absorption peak.
Higher furan contents contribute to wider absorption band- The model was developed in accordance with the fuzzy in-
width and higher maximum peak. The bandwidth length ference flowchart shown in Fig. 7, which is a process of formu-
increases linearly according to the furans level in the range of lating the mapping from a given input to the designated output.
348 to 470 nm whereas the absorption peaks range is relatively Inputs variable for the model are the values of bandwidth wave-
small, and some of them were intersecting with each other. length and the maximum absorption peak collected from test
Therefore, the peak is described up to four decimal places to results in Table II. A variation in these parameters at different
increase the accuracy of reading. All of the maximum peaks ambient temperatures as found in Figs. 5 and 6 is included in the
occur in the range of 220 to 230 nm. The details of the zero design of the model. The output from the model is the estimated
crossing point at the end of each spectra bandwidth and its furans content.
maximum absorption peak for each furan concentration are The model was built using the graphical user interface tools
available in Table II. provided by MATLAB. Each input was fuzzified into 10 sets of
Due to the fact that changes in spectral wavelength preci- the Gaussian combination membership function (MF) governed
sion can occur because of the variation of ambient temperature by
[9], [11], all oil samples were rescanned for UV-Vis spectral re-
sponse at a room temperature of 25 C. The waveform shapes (4)
ABU-SIADA et al.: A NOVEL FUZZY-LOGIC APPROACH FOR FURAN ESTIMATION IN TRANSFORMER OIL 473
TABLE III
COMPARISON BETWEEN ASTM D 5837 AND
UV-VIS SPECTROSCOPY WITH FUZZY LOGIC
(5)
(6)
The function set developed for each input is a combination
of and parameters, where the first function 1, specified where is the membership function of the output.
by and , determines the shape of the leftmost curve, and Fig. 11 shows the fuzzy model to estimate furan concentra-
the second function 2, specified by and , determines tion where variables BW and Ab represent the spectral-response
the shape of the rightmost curve. The corresponding Gaussian bandwidth and peak absorbance, respectively, as input data to
function curve for each input variable is shown in Figs. 8 and 9. the fuzzy model and t is the step time for the fuzzy model
The result of fuzzification from each input was then applied simulation.
with the fuzzy operator in the antecedent and related to the con- The model developed is able to estimate the furan concentra-
sequence, which is the output, by the application method. In tion with more than 95% accuracy compared to measurement
contrast with the input variable, the triangular-shaped member- results from Table II. Fig. 12 shows the developed set of fuzzy
ship function is selected to describe the output function. Each rules relating the input variables to the output. The mapping
474 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 27, NO. 2, APRIL 2012
from the inputs of bandwidth and absorption peak to the furan [12] A. M. Emsley, X. Xiao, R. J. Heywood, and M. Ali, “Degradation of
concentration can be illustrated in a 3-D graph shown in Fig. 13. cellulosic insulation in power transformers. Part 2: Formation of furan
products in insulating oil,” Proc. Inst. Elect. Eng., Sci., Meas. Technol.,
, vol. 147, pp. 110–114, 2000.
VI. COMPARISON BETWEEN ASTM D 5837 AND UV-VIS [13] J. S. S. Levchik, G. Camino, W. Tumiatti, and M. Avidano, “De-
ploymerization process in the thermal degradation of cellulosic paper
SPECTROSCOPY WITH FUZZY LOGIC insulation in electrical transformers,” in Polymer Degrad. Stability,
ASTM D 5837 is widely adopted for accurate identification Feb. 13, 1997, pp. 507–511.
[14] P. Verma, M. Roy, R. K. Tiwari, and S. Chandra, “Generation of fu-
and quantification of furan derivatives available in transformer ranic compounds in transformer oil under accelerated thermal and elec-
oil. However, testing with HPLC or GC/MS requires trained trical stress,” in Proc. Elect. Insul. Conf. Elect. Manuf. Expo., 2005, pp.
personnel, tedious pretreating of test samples, and very expen- 112–116.
[15] J. Scheirs, G. Camino, W. Tumiatti, and M. Avidano, “Study of the
sive equipment. By proposing UV-Vis spectroscopy and the mechanism of thermal degradation of cellulosic paper insulation in
fuzzy-logic model, utility companies could use this alternative electrical transformer oil,” Die Angewandte Makromolekulare Chemie,
to estimate the overall furans condition in the oil before it is vol. 259, pp. 19–24, 1998.
[16] H. Li and M. M. Gupta, “Fuzzy logic and intelligent systems,” in In-
found necessary to be sent for a standard test. The comparison ternational Series in Intelligent Technologies. Norwell, MA: Kluwer,
of both methods is summarized in Table III. 1995.