Barrier Effect On The Dielectric Strength of The Transformer Insulting Oils
Barrier Effect On The Dielectric Strength of The Transformer Insulting Oils
Barrier Effect On The Dielectric Strength of The Transformer Insulting Oils
The 20 International Symposium on High Voltage Engineering, Buenos Aires, Argentina, August 27 – September 01, 2017
Abstract: The barriers in transformer insulating oil play an important role in prohibiting
the particles movement in addition to reducing the partial discharge activity, therefore,
enhancing the dielectric strength of the insulating oils. In this paper, a Response Surface
Methodology (RSM) based on Box-Benken Design (BBD) is used to construct a
prediction formula to illustrate the impact of the gap space with barrier between two plates
(d), the barrier placed between the two plates relative to gap space (a/d)% and
eventually, the barrier diameter (D) on the breakdown strength of the insulating oils. The
effect of the above-mentioned factors for the plate to plate gap under alternating current
voltage is demonstrated. The benefit to using BBD is a reduction in the experiments
numbers that in all cases is very costly and all design points are in the safe operating
zone. The experimental works are performed to collect the required data (for BBD, 15
experiments data is sufficient). The results explain the ability of the proposed method to
give a prediction formula with a minimum number of required experiments and minimum
errors, in addition, the optimal gap space (d), relative placement of barrier relative to gap
space (a/d)% and the diameter of barrier (D) is also obtained.
Levels
Input
Minimum Mean Maximum
variables
(-1) (0) (1)
d (cm) 1 3 5 (a) (b)
(a/d)% 25 50 100
D (cm) 5 7 10 Figure 2: (a) the tank with plate to plate electrodes
before filling the oil completely, (b) Schematic
Table 2: Coded values for three factors of BBD
diagram for the oil tank
Act.
Case d (a/d)% D Act. d Act. D Table 3: The output responses according to each
(a/d)%
1 -1 -1 0 1 25 7 run in Table 2
2 -1 1 0 1 100 7 Case VBD(kV)
3 1 -1 0 5 25 7 1 20
4 1 1 0 5 100 7 2 12
5 -1 0 -1 1 50 5 3 87
6 -1 0 1 1 50 10 4 79
7 1 0 -1 5 50 5 5 11
8 1 0 1 5 50 10 6 11
9 0 -1 -1 3 25 5 7 76
10 0 -1 1 3 25 10 8 81
11 0 1 -1 3 100 5 9 59
12 0 1 1 3 100 10 10 64
13 0 0 0 3 50 7 11 52
14 0 0 0 3 50 7 12 60
15 0 0 0 3 50 7 13 55
14 55
4 THE PREDICTION EQUATION 15 55
According to the results of the experiment tests,
The results of the 15 experiment tests according to the quadratic equation that predict the relation
the roles as in Table 2 are described in Table 3. between the breakdown voltage in kV and the
The breakdown voltages (kV) in each run are three input variables (d, (a/d)%, and D) is
registered for each input variables as in Table 2. explained in eqn.3;
The linear form of the empirical model that
developed based on BBD rules which demonstrate VBD=56+33.625d-3.375(a/d)+2.25D+2.55*1014d
the relation between the input variables and a (a/d)+1.25dD+0.75 (a/d) D-10.25d2+3.75 (a/d)2D2
certain output response is given by Eqn. 1;
(3)
Y=b0+b1.X1+b2.X2+b3.X3+b12.X1.X2+b23.X2.X3+
b13.X1.X3+b123.X1.X2.X3+ (1) Equation 3 illustrates that the great effect on the
breakdown voltage comes from the gap space (d)
Where, Y is the response, X1, X2, and X3 are the between the two plates in addition to the barrier
input variables that affect the response. X1X2, place (a/d) % in the gap space. The significant
X1X3, and X2X3 are the two-way terms which its effect of the input variables can be also
number is calculated as (N(N-1)/2, where, N is the demonstrated via statistical study such as student
number of input variables) refer to interaction terms test coefficient (tstat) and p-value which remove
that maybe probably effect on the response. the null hypothesis of the applied data. The
X1X2X3 is the three-way term, b0 is the constant following Table 4 contains the tstat for each
parameter and when tstat value is greater than tcrit , Table 4: statistical parameters for the experimental
therefore, the associated variable is significant. In data
addition to tstat, Table 4 contains also the p-value b SE tstat p-value
which removes the null hypothesis of the
(Intercept) 56 0.93986 59.583 2.52E-08
experiment data. The parameter SE refers to the
standard error which results from the standard d 33.625 0.57554 58.423 2.78E-08
deviation divided by the number of data. a/d -3.375 0.57554 -5.864 0.002045
D 2.25 0.57554 3.9094 0.011302
The risk interval () which can be assumed as 0.05
d:a/d 2.55E-14 0.81394 3.13E-14 1
and with the number of freedom (n-p), the critical
student test value (tcrit) can be computed as in Eqn. d:D 1.25 0.81394 1.5357 0.1852
(4) [8]. ad:D 0.75 0.81394 0.92144 0.3991
2
d -10.25 0.84718 -12.099 6.81E-05
tcrit=(, n-p) (4) 2
(a/d) 3.75 0.84718 4.4265 0.006851
2
A tcrit helps in determining whether the null D -1 0.84718 -1.1804 0.29094
hypothesis is rejected or not. A tcrit is used to
determine the significantly of the influencing Table 5: Values of squared R, adjusted R-square and
variables. Comparing tstat with tcrit explain which F-statistic
input variables influence on the response, R-squared 0.999
therefore, if tstat is greater than tcrit, the variable is Adjusted R-Squared 0.996
significant and vice versa. The critical t can be F-statistic vs. constant model 405
calculated by Eqn. (4), therefore, if the number of Table 6: Validation of the constructed model
experiments (n) for BBD is 15, and the number of Act. pred.
coefficients (p) is 10 variables, then, the degree of Case d (a/d)% D error%
VBD(kV) VBD
freedom=n–p = 15 − 10 = 5; hence tcrit (two-tailed)= 1 -1 -1 0 20 19.25 3.75
(0.05, 5) = 2.5706.
2 -1 1 0 12 12.5 4.17
Therefore, the significant variables for Eqn. 3 are 3 1 -1 0 87 86.5 0.57
d, (a/d)%, d2 and (a/d)2 in addition to the constant 4 1 1 0 79 79.75 0.95
term (b0). Also, this fact can be reached through 5 -1 0 -1 11 10.125 7.95
the values of p-value that are shown in column 5 in
6 -1 0 1 11
Table 4. When the p-value is lower than 0.05 then, 12.125 10.22
the null hypothesis can be rejected. Therefore, the 7 1 0 -1 76 74.875 1.48
approximated equation of the breakdown voltage 8 1 0 1 81 81.875 1.08
can be rewritten as follows;
9 0 -1 -1 59 60.625 2.75
2 2 10 0 -1 1 64
VBD =56+33.625d-3.375(a/d) -10.25d +3.75 (a/d) 63.625 0.59
11 0 1 -1 52 52.375 0.72
(5) 12 0 1 1 60 58.375 2.71
Other statistical parameters can be used to explain
13 0 0 0 55 56 1.81
the stability of the prediction equation such as
squared R, adjusted R-squared and F-statistic. 14 0 0 0 55 56 1.81
These parameters are shown in Table 5. The 15 0 0 0 55 56 1.81
squared R and the adjusted R-squared are 0.999 Average error % 2.82
and 0.996 respectively which refer to a good
relation between the predicted and actual results of The validation of the model is done by comparing
the model. The F-statistic is very important the actual value of the measured breakdown
parameter since it demonstrates the relation voltage at values of the input variable with the
between the model mean square and the error calculated breakdown voltage that developed using
mean square and the value of f-statistic is 405 Eqn. 3 at the same input variables. The error
which is a very large value that refers the high percentage between the actual and predicted
accuracy of the constructed model. breakdown voltage refer to only two runs are
higher (10.22% and 7.95%) but all the rest runs are
below 5% which is considered acceptable. The
5 VALIDATION OF THE CONSTRUCTED
average error percentage for the 15 runs is 2.82%
PREDICTION MODEL
which is very acceptable value and it refers to the
ability of the constructed model to predict the
The constructed BBD model to assign the relation
breakdown voltage with a minimum number of
between the response and the input variable
experiment tests. The higher value of the
should be validated.
percentage error occurs at the smallest gap space
(1 cm) and due to the small value of the actual
breakdown voltage and any small variation in the Electrical Insulation and Dielectric Phenomena,
predicted breakdown voltage will cause an Minneapolis, October 19-22, 1997, pp.660-663.
increase in the error percentage. [5] A. Zouaghi and A. Beroual, "Barrier Effect on
the ielectrie Strength of Oil Gaps under DC
Voltage", Conference Record of the 1998 IEEE
6 CONCLUSIONS
International Symposium on Electrical
Insulation, Arlington, Virginia, USA, June 7-10,
BBD is one of the DOE techniques that are used to 1998, pp.640-643.
construct a relation between a certain response [6] Y. Julliard, R. Badent and A. J. Schwab,
and some of influencing input variables. The "Behavior of multiple barrier insulation systems
benefits of BBD over the other DOE technique under impulse conditions," 2001 Annual Report
such as CCF are the less number of experiments Conference on Electrical Insulation and
test that the BBD require constructing the model, in Dielectric Phenomena (Cat. No.01CH37225),
addition to all design points locate in the safe Kitchener, Ont., 2001, pp. 540-543.
operating zone and eventually, all factors are not [7] A. Kara, E. Onal, O. Kalenderli, and K.
set at the high levels at the same time. The results Mardikyan, "The Effect of Insulating Barriers on
obtained from the constructed model as well as the AC Breakdown Voltage in Inhomogeneous
statistical study referred that the ability of the Field", IEEE MELECON 2006, May 16-19,
constructed model for assigning the relation Benalmádena (Málaga), Spain, pp. 1206-1208.
between the oil voltage breakdown and some of [8] Zitouni, M.,Guerbas F., Boukezzi, L., et.al.:
barrier parameters such as the gap space between 'Modeling by Design of Experiments Method of
the two plates (d), the position of the barrier in the the AC Breakdown Voltage of Transformer Oil
gap (a/d)% and finally the barrier diameter (D). the Point–Plane Gaps with Insulating Barrier', IET
average error for all experiments runs is computed Gen. Trans.& Dist., Vol. 10, Iss. 1, pp. 232–
as 2.82 % which indicated a good relation between 239, 2016,.
the actual and the constructed model. The [9] M. Cavazzuti, Optimization Methods: From
statistical results refer that the barrier diameter has Theory to Design, Springer-Verlag Berlin
not a significant effect on the breakdown voltage Heidelberg 2013, pp. 13.
since the value of tstat, p-value at this parameter is [10] http://www.itl.nist.gov/div898/handbook/pri/sect
the worst. ion3/pri3362.htm
ACKNOWLEDGMENTS
Authors may wish to acknowledge Laboratories&
Researches and Tests, Egyptian Electricity Holding
Company, Egypt, for giving us an opportunity to do
our works.
REFERENCES
[1] F. Guerbas, M. Zitouni, A. Boubakeur, and A.
Beroual, "Barrier effect on breakdown of point–
plane oil gaps under alternating current
voltage," IET Generation, Transmission &
Distribution, Vol. 4, Iss. 11, pp. 1245–1250,
2010.
[2] Ekram Husain, Kamaluddm Imtiaz Ashraf, and
Atif lqbal, "Study of Breakdown of Transformer
Oil in the Presence of Barrier", Proceedings of
the 4th International Conference on Roperties
and Applications of Dielectric Materials July 3-
8, 1994, Brisbane Australia, pp. 56-59.
[3] A. Beroual and A. Zouaghi, "Barrier Effect on
the Pre-breakdown and Breakdown
phenomena in Long Oil Gaps", Conference
Record of the ICDL '96 12th International
Conference on Conduction and Breakdown in
Dielectric Liquids ,Roma, Italy, July 15 - 19,
1996, pp. 300-303.
[4] A. Zouaghi and A. Beroual, "Discharge
Structure and Dielectric Strength of Long Oil
Gaps in the Presenceof an Insulating Barrier",
1997 IEEE Annual Report - Conference on