Curve Fitting Analysis of Expulsion Fuse Links

through the Cross-Validation Technique

Guilherme Braga da Costa, Augusto Zanin Bertoletti Adriano Peres de Morais, Ghendy Cardoso Junior
Universidade Federal de Santa Maria (UFSM) Universidade Federal de Santa Maria (UFSM)
Santa Maria – RS – Brasil Santa Maria – RS – Brasil
Email: {guilhermebragadacosta, augustozb1}@gmail.com Email: adriano@ctism.ufsm.br, ghendy@ufsm.br

Abstract—The most common protective device in distribution

systems is the expulsion fuse link. This work aims to assist the
engineer in carrying out expulsion fuse links modeling using curve
fitting. The proposed approach used the fuse time-current
characteristic to obtain the polynomial function that represent the
fuse link model. Preferred K and H fuses were used. The
polynomial coefficients was obtained through the MATLAB
software. The polynomial order that best represents the fuse link
was decided using the cross-validation technique. The technique
consists of iteratively partitioning the data set, separating test
points and training points on each iteration.

Keywords—expulsion fuse link, time-current characteristic,

curve fitting, cross-validation.

I. INTRODUCTION
Several protective devices, such as breakers, overcurrent
relays, reclosers and expulsion fuse links, compose distribution
systems. The fuses are low-cost and easily replaced devices.
Hence, it is the most common protective device in distribution
systems. The concept of expulsion fuse is very simple: a fusible
element made of tin or silver melts under high current [1]. The
melting time is inversely proportional to the current value. The
fuses are mainly used to protect branches, distribution
transformers, capacitor bank, voltage regulators and potential
transformers.
The expulsion type cutouts are classified as open-fuse
cutouts, enclosed-fuse cutouts, and open-link-fuse cutouts. The
classification is according to the external appearance and
operation method [2]. The fuse link is divided in preferred and Fig. 1 – Example of MM Curves for K fuse links [3].
nonpreferred current ratings. The current ratings for preferred
sizes are given as 6, 10, 15, 25, 40, 65, 100, 140 and 200 A and data used are available from the manufacturers on a graphs using
for nonpreferred sizes as 8, 12, 20, 30, 50 and 80 A. Industry logarithmic scale as shown in Fig. 1. Theses graphs contain the
standards specify three types of expulsion fuses: “K”, “T” and fuse time-current characteristics (TCC).
“H”. As stated in [3], the type “K” link is consider fast fuse, and
the “T” is somewhat slower. The TCC are represented by two curves: the minimum
melting curve (MM) and the total clearing curve (TC). The MM
Type H fuses are manufactured in ratings of 0.5, 1, 2, 3, 5, curve informs the minimum time versus current required to melt
and 8A. According to [3], the primary purpose of a transformer the fuse. The TC curve informs the maximum time versus
fuse is to disconnect the transformer from the circuit if it fails. current required to melt the fuse and extinguish the arc [2].
The fuse should avoid misoperation from inrush current and
cold-load pickup. To assist the engineer in carrying out fuse links modeling,
several methodologies were proposed to mathematically model
Several studies focus on distribution systems protection were the curves of expulsion fuse links. In [4] is presented an
published in the recent years. According to [4], when the evaluation of the methodologies that use analytical equations to
computational simulations uses fuse links, it is necessary to represent this behavior in computational simulations. The main
represent the fuse element using digital models. Normally, the functions used were polynomial, exponential and linear function
by parts. The present work used the polynomial function to
The authors are thankful to Brazilian government through Capes
(Coordenação de Aperfeiçoamento de Pessoal de Nível Superior). This work
was supported by Capes.

978-1-5386-5844-4/18/$31.00 ©2018 IEEE

model the behavior of the expulsion fuse link type K and H in
computational simulations. The function was selected because
presented the best results according to the criterion of Residual
Sum of Squares (RSS) [4].
The polynomial order was selected through the Cross-
Validation (CV) technique. According to [5], CV is a technique
used to evaluate the predictive capacity of a mathematical
model, from a set of data, where some samples are removed as
part of the analysis. This technique consists of partitioning the
data into two sets: one is used to construct the model, and the
other is used to test it, i.e., to verify its generalization capacity.
In each partition, the metric used is the Mean Square Error
(MSE). In this paper, only the preferred current ratings were
used.
This paper is organized in the following manner. Section II
introduces the fuse links modeling. Section III describes the
methodology approach. Section IV presents the obtained results
and the conclusion is drawn in Section VI.

II. EXPULSION FUSE LINK MODELING

Several articles approach the fuse link modeling. The linear
interpolation between two consecutive points was used in [6]
and [7]. According [8] and [9], the curve fitting method was
used to approximate the segment of interest by linear equation.
The linear equation is shown in (1).
log(t ) = a. log(I ) + b (1)
where t and I for the fuse denote operating time and current Fig. 2 – Polynomial Regression routine.
respectively, and coefficients a and b can be calculated by a
curve fitting method. A. Data Acquisition
The exponential equation also was used to model the fuse The points used were collected from the TCC curves, which
links. The authors from [10] and [11], used a curve fitting are supplied by manufacturers of fuse link. The WebPlotDigitzer
method to determine a, b, c, e d coefficients in (2) and an software [14] was used to collect the points. The software was
coefficient in (3) respectively. selected because it allows the user setting both axis in
logarithmic scale. The present work collected approximately 60
t = a.e bxI + c.e dxI points for each fuse link curve. Five important fuse
(2)
manufacturers have been considered in this paper, i.e.: S&C
4 Positrol® [15], Cooper Power Systems Kearney™ [16],
t = exp(  a n .(ln( I )) n ) (3 ) Eletrofusi Power Systems [17], Delmar Hubbell Power Systems
n =0
[18] and Indel Bauru [19].
where t is the fuse operating time and I is the fault current seen B. Curve Fitting
by the fuse.
Through Curve Fitting Toolbox™ present in MATLAB®
In [12] and [13], the authors developed an algorithm to [20], the polynomial function was implemented to obtain the
evaluate and identify which of the equations selected provides coefficients. Using (4), an evaluation of the adjustments from
the smallest adjustment of errors. the first to the fifteenth order was made using the CV as a
The polynomial function was mentioned by [8] and [9]. The comparative parameter. The order with the lowest CV value is
authors affirm that a second-order polynomial function can be selected to represent the fuse link mathematically.
used to model the fuse link. The equation coefficients was
determined through a curve fitting method. C. Polynomial Function
There are several approaches proposed in the literature for
III. METHODOLOGY model the expulsion fuse links. In [21] and [22], the authors used
the polynomial function to model the fuse link. The curve fitting
The Polynomial regression routine used to determine the
method was used to determine the p coefficients in (4) for each
best polynomial order to fit the curves of each fuse link is shown
fuse link.
in Fig. 2. The methodology used follows the steps.
log(t ) = p0 + p1. log(I ) + p2 . log 2 ( I ) + ... + pn log(I ) n (4)
where t is the fuse link operating time; I is the fault current curve of 6K, 10K, 25K, 40K, 65K and 100K fuse link were
flowing through the fuse; n is the polynomial order; and p are plotted for all manufacturers. The graphs are shown in Fig. 5,
the fuse coefficients. Fig. 6 and Fig. 7.
0,0025
D. Cross-Validation (CV) Cooper Kearney
Eletrofusi Power System
0,0020
In [4], the authors used the Least Square Method and the RSS
as a comparative parameter to select the polynomial order. The (a) 0,0015

present work used the CV technique to evaluate the accuracy of

MSE
0,0010
the polynomial order.
0,0005
The K-fold procedure was used in the CV method. The basic
concept is to split the data into K partitions of approximately 0,0000
6K 10K 15K 25K 40K 65K 100K 140K 200K

equal sizes - one partition is used for testing and the K-1 0,008
Cooper Kearney
partitions are used for parameter estimation and the quadratic Eletrofusi Power System

error calculation. This process is performed K times, so that all 0,006

(b)
data is allocated as a test partition at least once. At the end of the

MSE
0,004
K iterations, the MSE is calculated considering the K partitions
[5]. 0,002

E. Comparative Parameter 0,000

It is important to highlight that the procedure described 6K 10K 15K 25K 40K
Fuse Link
65K 100K 140K 200K

above evaluates the predictive capacity of a given model. As the

Fig. 3 - Type K Fuse Links x MSE (a) MM curve (b) TC curve.
objective is to evaluate the best polynomial order, the K-fold
procedure is repeated n times, where n is the highest polynomial Through the figures, it is possible to verify the accuracy of
order tested, until all MSE are obtained. The polynomial order the CV technique. The technique allows us to use a small
that presents the lowest MSE is selected. The smaller the MSE, polynomial order if we used another parameter such as RSS for
the closer the fit is to the original data. The MSE can be comparison. Fig. 3 shows the MSE results for type K fuse links
calculated by (5). for Cooper Kearney and Eletrofusi. It is possible to note that all
the results are lower than 0.025.
MSE = ( − ) (5)
where T denotes an estimate and θ denote the parameter being 2. Type H Fuse Link
estimated. TABLE II shows the polynomial order of MM and TC curves
In this way, the authors of this work expect that this study for H fuse link, selected by the CV technique for the three
can contribute to the development of studies in distribution manufacturers. As stated for the K fuse link, it can be observed
system protection, therefore facilitating the modeling of that it was not possible to achieve the same polynomial order
expulsion fuse links. for the fuse link models. Fig. 8, Fig. 9 and Fig. 10 shows the
MM and TC curve of 3H and 5H fuse link. The figures confirms
IV. RESULTS the capability of the technique.
This section presents the results of the type K and H fuse Fig. 4 shows the MSE results for type H fuse links for
links curve fitting using the polynomial function given in (4). It Delmar and Eletrofusi. It is possible to note that all the values
is important to note that approximately 60 points were collected are lower than 0.006.
for each curve and polynomials order from 1 up to 15 were
tested. 0,00020
Delmar
Eletrofusi Power System
Tests have identified that the number of minimum points for 0,00015

a satisfactory fit of the curve are 60 points. When the number (a)
of points is less than 60, the curve adjustment is compromised, 0,00010
MSE

and it is necessary to increase the degree of the polynomial. An

amount greater than 60 points modifies the coefficients of the 0,00005

polynomials, but does not change the quality of the curve fit. 0,00000
0.5H 1H 2H 3H 5H
1. Type K Fuse Link 0,0005
Delmar
Eletrofusi Power System

TABLE I shows the polynomial order of MM and TC curves 0,0004

for K fuse link, selected by the CV technique for the three (b)
manufacturers. From TABLE I, it can be noticed that it was not 0,0003
MSE

possible to obtain the same polynomial order for the different 0,0002

fuse link models; therefore, a previous definition of the order 0,0001

can lead to a less reliable predictive capacity of the
mathematical model. 0,0000
0.5H 1H 2H 3H 5H
Fuse Link
To prove the efficiency of the technique, the MM and TC
Fig. 4 - Type H Fuse Links x MSE (a) MM curve (b) TC curve.
 3H MM Collected Data
6K MM Collected Data 3H MM Polynomial Function
6K MM Polynomial Function 3H TC Collected Data
100 6K TC Collected Data 100 3H TC Polynomial Function
6K TC Polynomial Function 5H MM Collected Data
10K MM Collected Data 5H MM Polynomial Function
10K MM Polynomial Function 5H TC Collected Data
5H TC Polynomial Function
10K TC Collected Data
10K TC Polynomial Function
10 10

Time (s)
Time (s)

1
1

0,1
0,1

0,01
0,01
10 100 1000 10 100

Current (A) Current (A)

Fig. 5 – 6K and 10K MM and TC curve for Cooper Power Systems Fig. 8 – 3H and 5H MM and TC curve for Delmar Hubbell Power Systems.
Kearney™.

65K MM Collected Data 3H MM Collected Data

65K MM 13th Polynomial Order 3H MM Polynomial Function
100 65K TC Collected Data 3H TC Collected Data
65K TC 14th Polynomial Order
100 3H TC Polynomial Function
5H MM Collected Data
100K MM Collected Data 5H MM Polynomial Function
100K MM 13th Polynomial Order 5H TC Collected Data
100K TC Collected Data 5H TC Polynomial Function
10 100K TC 15th Polynomial Order
10
Time (s)
Time (s)

1
1

0,1
0,1

0,01

0,01
100 1000 10 100
Current (A) Current (A)

Fig. 6 – 25K and 40K MM and TC curve for S&C Positrol®. Fig. 9 – 3H and 5H MM and TC curve for Eletrofusi Power Systems.

25K MM Collected Data 3H MM Collected Data

25K MM 12th Polynomial Order 3H MM Polynomial Function
25K TC Collected Data 3H TC Collected Data
100 100 3H TC Polynomial Function
25K TC 13th Polynomial Order
5H MM Collected Data
40K MM Collected Data 5H MM Polynomial Function
40K MM 12th Polynomial Order 5H TC Collected Data
40K TC Collected Data 5H TC Polynomial Function
40K TC 14th Polynomial Order
10
10
Time (s)
Time (s)

1
1

0,1
0,1

0,01

0,01
100 1000 10 100
Current (A) Current (A)

Fig. 7 – 65K and 100K MM and TC curve for Eletrofusi Power Systems. Fig. 10 – 3H and 5H MM and TC curve for Indel Bauru.
TABLE I – POLYNOMIAL ORDER OBTAINED WITH THE CROSS-VALIDATION TABLE II – POLYNOMIAL ORDER OBTAINED WITH THE CROSS-VALIDATION
TECHNIQUE FOR TYPE K FUSE LINK. TECHNIQUE FOR TYPE H FUSE LINK.

Cooper Power Eletrofusi Power Delmar Hubbell Eletrofusi Power

S&C Positrol® Indel Bauru
Systems Kearney™ Systems Power Systems Systems
Fuse Fuse
MM TC MM TC MM TC MM TC MM TC MM TC
Link Link
6K 12 15 15 15 14 13 0.5H 15 13 15 15 - -
10K 12 15 11 12 14 12 1H 15 13 15 15 14 12
15K 13 12 15 11 14 12 2H 15 12 13 15 11 10
25K 12 13 14 11 14 11 3H 15 14 14 15 12 12
40K 12 14 13 10 14 15 5H 14 12 14 15 15 14
65K 13 14 11 14 13 14
100K 12 13 11 15 13 15
140K 14 14 11 12 - -
200K 11 11 12 11 - -

[8] S. Chaitusaney and A. Yokoyama, “Prevention of Reliability

Degradation from Recloser–Fuse Miscoordination Due To
V. CONCLUSION Distributed Generation,” IEEE Trans. Power Deliv., vol. 23, no. 4,
The expulsion fuse links are widely used in the protection pp. 2545–2554, 2008.
of distribution systems, so it is important to represent the [9] A. F. Naiem, Y. Hegazy, A. Y. Abdelaziz, and M. A. Elsharkawy,
computational model of the MM and TC curves with good “A Classification Technique for Recloser-Fuse Coordination in
Distribution Systems With Distributed Generation,” IEEE Trans.
precision. Power Deliv., vol. 27, no. 1, pp. 176–185, Jan. 2012.
The present work used the polynomial function to model [10] H. A. Abdel-ghany, A. M. Azmy, N. I. Elkalashy, and E. M.
the TCC fuse links. Preferred type K and H fuse links were Rashad, “Optimizing DG penetration in distribution networks
concerning protection schemes and technical impact,” Electr.
used. Nonetheless, the methodology can be apply to any kind Power Syst. Res., vol. 128, pp. 113–122, 2015.
of fuse.
[11] S. Santoso and T. A. Short, “Identification of fuse and recloser
The MATLAB software was used to determine the operations in a radial distribution system,” IEEE Trans. Power
Deliv., vol. 22, no. 4, pp. 2370–2377, 2007.
polynomial coefficients. The polynomial order that presented
[12] A. C. Enríquez, E. Vázquez-Martínez, and H. J. Altuve-Ferrer,
the smallest deviation for modeling the fuse link through the
“Time overcurrent adaptive relay,” Int. J. Electr. Power Energy
CV technique was selected. The technique was unable to Syst., vol. 25, no. 10, pp. 841–847, 2003.
obtain a single polynomial order for the fuse link models. [13] A. C. Enríquez and E. V. Martínez, “Enhanced time overcurrent
coordination,” Electr. Power Syst. Res., vol. 76, no. 6–7, pp. 457–
In the tests performed, it was observed that the CV
465, 2006.
technique presents as a general solution a polynomial order
[14] A. Rohatgi, “WebPlotDigitizer,” 2017. [Online]. Available:
different than that polynomial with degree determined only http://arohatgi.info/WebPlotDigitizer/. [Accessed: 02-Feb-2017].
observing the criterion of the smallest RSS errors, quite used [15] “S&C ELECTRIC COMPANY Positrol Fuse links,” 2017.
in curve fitting literature. However, this does not compromise [Online]. Available: http://www.sandc.com/en/products--
the overall quality of the solution. services/products/positrol-fuse-links/. [Accessed: 02-Jan-2017].
[16] “EATON COOPER POWER KearneyTM,” 2017. [Online].
REFERENCES Available:
http://www.cooperindustries.com/content/public/en/power_
[1] Leonard L. Grigsby, Electric Power Generation, Transmission systems/products/fusing/fuse_links/kearney.html. [Accessed: 02-
and Distribution, Third Edition, 3rd ed. Boca Raton: CRC Press, Jan-2017].
2012.
[17] “ELETROFUSI POWER SYSTEMS Fuse links,” 2017. [Online].
[2] Turan Gönen, Electric Power Distribution Engineering, Third Available: http://www.eletrofusi.com.br/eng/fuse_eng.htm#linkb.
Edition, 3rd ed. Boca Raton: CRC Press, 2014. [Accessed: 02-Jan-2017].
[3] T. A. Short, Electric Power Distribution Handbook, Electric P. [18] “Delmar Hubbell Power Systems, Inc.,” 2017. [Online].
CRC Press, 2004. Available: http://hubbellpowersystems.com.br/PDF/ELOS.PDF.
[4] G. B. da Costa, A. C. Marchesan, A. P. de Morais, G. C. Junior, [Accessed: 02-Jan-2017].
and M. Gallas, “Curve Fitting Analysis of Time-Current [19] “Indel Bauru,” 2017. [Online]. Available:
Characteristic of Expulsion Fuse Links,” 17th Int. Conf. Environ. http://www.indelbauru.com.br/indel/en/produto/visualizar/codpro
Electr. Eng., p. 6, 2017. duto/35/type-h-nema.html. [Accessed: 02-Jan-2017].
[5] W. L. Martinez and A. R. Martinez, Computational Statistics [20] Mathworks, “MATrix LABoratory (MATLAB®).” .
Handbook with MATLAB, 2nd ed. Boca Raton: CRC Press, 2007.
[21] Y. Tang and R. Ayyanar, “Verification of Protective Device
[6] J. J. G. Ledesma, L. R. de Araujo, D. R. R. Penido, and J. L. R. Coordination in Distribution Systems with Photovoltaic
Pereira, “Algorithm for the Analysis of Coordination of Protection Generation,” IEEE 40th Photovolt. Spec. Conf., pp. 2100–2105,
for Distribution Systems,” IEEE Lat. Am. Trans., vol. 13, no. 7, 2014.
pp. 2168–2175, Jul. 2015.
[22] W. Tian, Chengwei Lei, Y. Zhang, D. Li, R. Fu, and R. Winter,
[7] E. Campitelli, L. W. Da Silva, R. F. Pereira, and J. S. Mantovani, “Data analysis and optimal specification of fuse model for fault
“Interactive System for Placement and Coordination of study in power systems,” in IEEE Power and Energy Society
Overcurrent Protective Devices,” in IEEE/PES Transmission & General Meeting (PESGM), 2016, pp. 1–5.
Distribution Conference and Exposition: Latin America, 2006, pp.
1–6.