You are on page 1of 11

This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIA.2017.2688408, IEEE
Transactions on Industry Applications
2016-METC-0636 1

Life Assessment of Electric Arc Furnace


Transformers
E. A. Cano-Plata, Senior Member, IEEE, O. J. Soto-Marin, Student Member, IEEE, and
A. F. Ustariz-Farfan, Member, IEEE

furnace parameters such as voltage and current, real and


AbstractElectric arc furnace (EAF) transformers are very reactive power, and resistance and reactance exhibit
important in the steel production industry. Currently, it is very significant variations especially during fusion stage [2][3];
common for companies to buy used transformers considering the subjecting the transformer to strong electrical and magnetic
huge investment that represents the acquisition of a new
stresses which can be above nominal values.
transformer and also the delay in the purchase transaction.
Before accepting a commercial agreement and putting the
transformer into operation, it is necessary to evaluate the level of
wear out due to the electrical and magnetic stresses to which the
transformer is subjected. Electrical, mechanical and insulation
tests, that follow procedures stipulated in guides and standards,
are performed on used transformers during the purchase process
in order to identify incipient faults and to determine their general
condition. A methodology is proposed to carry out these tests. In
addition, a new model based on fuzzy logic is implemented to
assess the level of wear of the transformer, its remaining life and
its feasibility of reuse. The fuzzy input variables of this new
model are the results of the tests performed on each element of
the transformer.

Index TermsAnalysis of dissolved gases, Arc resistance, Arc


stability, Electric arc furnace, Fuzzy model, Transformer.

I. INTRODUCTION

T HE main objective in a steel manufacturing industry is


to maximize annual production. A secondary objective is
to minimize the costs related to energy consumption and to the
Fig. 1. Schematic of the EAF system.

Consequently, the application of electric and magnetic


wear out of electrodes. These will be directly affected by the
stresses above nominal values leads to premature deterioration
Electric Arc Furnace (EAF) operating strategy [1]. In addition,
or aging of the transformer, and possibly causes anticipated
there should be proper electromagnetic coupling between the
failure [4][5].
furnace and the feeding system. Thus, the determination of the
Due to the large investment in the acquisition of a new EAF
maximum power transfer condition of the feeding system at
transformer, and the delay in the purchase process (which can
the time of scrap delivery, and the operating condition that
take up to one year), some companies decide to buy used
will keep the arc stable, are both points of great interest in the
transformers when the installed transformer fails or to
proper operation of the system. Fig. 1 shows a schematic
restructure the system by power increase. This alternative is
representation of the elements that make up the EAF electrical
very common but it becomes necessary that in the receiving
system.
process the level of wear out caused by electrical stresses is
Due to the non-linear and stochastic nature of the EAF,
evaluated before accepting a commercial agreement and
putting the transformer into operation [6].
This paper was submitted for review on July 26, 2016. The authors wish to
thank Universidad Nacional de Colombia, Manizales Branch, the Group of Currently, the processes of reception of used transformers
Distribution and Power Networks, GREDyP and the Laboratory of Power are based on electrical, mechanical and insulation tests [7].
Quality and Power Electronics, LACEP. Off-line and on-line methods are used to assess the
E.A. Cano-Plata is a full professor at Universidad Nacional de Colombia,
Manizales Branch. (e-mail: eacanopl@unal.edu.co).
condition of the transformers and to detect incipient faults.
O. J. Soto-Marin is currently a Ph.D. student in the Electrical Engineering Fig. 2 lists the main tests used.
Program at Universidad Nacional de Colombia, Manizales Branch. (e-mail: The off-line methods are usually conventional tests that are
ojsotom@unal.edu.co).
A.J. Ustariz-Farfan is associated professor in the Electrical, Electronic and
regulated by standards or norms.
Computer Engineering Department, at Universidad Nacional de Colombia, The on-line methods can be invasive and non-invasive. The
Manizales Branch, (e-mail: ajustarizf@unal.edu.co).

0093-9994 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIA.2017.2688408, IEEE
Transactions on Industry Applications
2016-METC-0636 2

latter are mainly oriented to the acquisition and processing of The automatic control system controls the position of the
signals, such as electromagnetic, acoustic waves, and heat or electrodes and makes them to fall and get in contact with the
infrared waves. scrap, thus creating an electric arc at that moment, the control
tries to maintain a constant and balanced current, which is
quite difficult due to the irregular shape of the scrap.
Consequently, the arc has a stochastic behavior throughout the
fusion process where current fluctuations vary between short
circuit values (electrodes are in direct contact with the scrap),
to open circuit (the electrodes are separated from the welding
material and the electric arc becomes extinguished).Currents
measured in a tap-to-tap cycle on the primary transformer side
are shown in Fig. 3.
TABLE I
VOLTAGES AND CURRENTS OF THE EAF TRANSFORMER
Tap position LV side HV side LV side EAF
OLTC Voltage Current current load
(v) (A) (kA) (MVA)
1 700 1255 25 30
2 660 1255 26 30
3 620 1255 28 30
4 580 1171 28 28
5 540 1088 28 26
6 500 1004 28 24
7 460 933 28 22
8 420 853 28 20
9 380 770 28 18

Fig.tests,
In addition to the 2. Testing in transformers.
the service of an expert is important
to assess the condition of each component of the system [7].
However, there are no available tools to integrate the results of
these tests into a single value to determine the interrelations
and contributions of each element to the reliability of the
entire system. Thus, the calculation of the level of the
remaining useful life of the transformer becomes inaccurate.
This paper proposes the implementation of a new model
Fig. 3. Behavior of the current seen from the primary,
based on fuzzy logic [20][21][22] that considers as input 6 loads of scrap
variables the results of the tests applied to each component in
Due to the low density of the scrap, the furnace receives
the reception stage. Using the results of this model, it is
several loads until the desired weight is reached
possible to assess the wear out condition of the transformer
(approximately 5 loads of scrap for the 30 tons in the EAF
and to determine the possibility of its reuse [23][24].
under study).
Since the voltage and current characteristics in an EAF are
II. CONCEPTUAL FRAMEWORK
nonlinear, all the involved electrical variables become also
This section presents an introduction to the operation and nonlinear [2]. The processes that take place in this circuit, as
specifications of the EAF installation under study and includes well as the dependencies between the electrical parameters of
details of the requirements of energy transfer in the the system are quite complex [3].
transformer. Afterwards, an analysis of the reliability of the Fig. 4 shows the variations of the EAF parameters as a
transformers feeding the EAF is performed. function of the current. Here: PARC is the arc power, Q is the
A. EAF system benchmark reactive power, P the active power, VARC is the arc voltage, pf
is the power factor in the secondary of the transformer and
The transformer that feeds the EAF is of 30 MVA. The
LARC is the length of the electrical arc.
secondary voltage varies from 380 to 700 V depending on the
The electric arc starts with priming and lasts until the end of
position of the on-load tap changer (OLTC). Table I shows
the melting stage. During this stage, the arc is very unstable
voltage and current values of operation in different positions.
and its length varies considerably due to scrap movement.
The secondary circuit of the transformer is connected to the
Initially, PARC and P increase to a certain maximum value and
electrodes of the furnace through refrigerated flexible cables,
then decrease rapidly, as the current increases, until the arc
constantly fed by a flow of water.

0093-9994 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIA.2017.2688408, IEEE
Transactions on Industry Applications
2016-METC-0636 3

length reaches zero [3]. 2640kWh (6)


trefining 396s
On the contrary, Q increases to a maximum value and it is at 400 kWh
this point where the greatest electrical and magnetic stresses 60s
take place in the transformer.
Thus, the tap-to-tap operation time of the system is given by:

ttaptotap t fusion tloading tcasting t prefinig (7)

ttaptotap (1700s 1800s 600s 396s)taptotap 4496s

After defining the load cycle in which the transformer must


operate, the thermal stress to be applied to the transformer is
evaluated in order to estimate the level of wear out for the load
cycle.
The loadability analysis is performed according to the
procedure established in the IEEE Std C57.91 standard [24].
Fig. 4. Parameter variation of EAF Fig. 5 shows the transient behavior of the hot spot temperature
in the winding according to load fluctuations. The maximum
The EAF of the installation under study has a capacity of 30
hot spot temperature is 108.3 C, which is below the nominal
tons of liquid steel. The total weight of the steel is 33 tons
temperature of 110C according to [25].
taking into account an operating point that allows a power
It is possible to approach the thermal analysis of the
factor of 0.8, energy transfer of 28 MW/3600s=466 kWh/60s,
and the calculation of the remaining steel weight left in the
furnace in 3 tons. The fusion requires 400 kWh/ton, therefore,
the estimated total energy is:

Etotal 400 kWh 33ton 13200kWh (1)


ton

In this way, the time used in the fusion is:

13200kWh (2)
t fusion 1700s
466 kWh Fig. 5. Transformer hot spot temperature
60s
transformer considering an equivalent load for the tap-to-tap
The crane must transport the scrap and it is necessary to load cycle, which will allow a better understanding of the
consider the way the furnace is filled. Estimating an average 6 thermal requirements in the transformer. For this, an
loads at a rate of 300 s, gives: equivalent constant load that generates the same total losses as
a fluctuating load is estimated and an equivalent load from the
tloading 6 * 300s 1800s (3) point of view of temperature is assumed [25].
Due to the random behavior of the requested power in the
Additionally, the time estimated for the casting is 600 s. fusion process observed in Fig. 3, the load cycle for the tap-to-
Times for loading and casting are required only in the fusion tap time was estimated according to the transformer operation
stage. data previously shown. The estimated load cycle for the tap-
The refining process requires 80 kWh/ton and the available to-tap time is shown in Fig. 6.
power is of 24 MW (30 MVA), therefore:

Etotal 33ton 80 kWh 2640kWh (4)


ton

Pavailable 24MW

24MVA
Prefinig 400 kWh (5)
3600s 60s
Fig. 6. Cycle of average load
Consequently, the fusion time is: Therefore, the equivalent load for the tap to tap load cycle
can be calculated by equation [25]:

0093-9994 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIA.2017.2688408, IEEE
Transactions on Industry Applications
2016-METC-0636 4

it is important to consider the high frequency of these


L t1 L t 2 L t3 .. L t N
2 2 2 2 overcurrents, which can lead to deterioration in the insulation
Loadequivalent 1 2 3 N (8)
paper. Higher current values might result in permanent
t1 t 2 t3 .. t N
deformations in the winding structure that might lead to
mechanical failures or insulation perforations.
Were LN means the load in MVA introduced in the time tN.
Replacing the load cycle values, gives an equivalent load of B. Reliability of EAF transformers
Loadequivalent=20.26MVA, corresponding to 72.4% of the The reliability of EAF transformers depends mainly on the
nominal load. It is concluded that the thermal requirements, degree of continuous monitoring of the most important
for the transformer under study, do not suppose a wear or functional parameters that define their technical condition.
deterioration of the insulating material. Systems are subjected to continuous deterioration under
However, as observed in the measurements presented in Fig. normal operation and transformers are no exception to this
3, overcurrent values (values up to 200% of nominal value) rule.
take place very frequently in the transformer. The effects of It is possible to track the natural or accidental deterioration
the overcurrent on the transformer are divided into two of transformers using modern monitoring equipment and
categories [4]. diagnostic techniques. [27].
Thermal effects Preventive and corrective maintenance helps to prolong the
Magnetic forces estimated life of the transformer, allows the calculation of the
remaining life and avoids costly losses.
Regarding the thermal effect, the IEEE Std C57.12.10 The application of electric and magnetic stresses above the
standard [26] establishes 250C as the maximum limit of plate nominal values implies an accelerated wear out. The
temperature increase in copper conductors. An approximate wear out can occur immediately or may be the result of
expression for the calculation of temperature increase in accumulated deterioration over the years. Fig. 7 shows the
conductors after t seconds is [4]: behavior of the failure rate.

2
e 2 1
t 1 D
C
100 eS eZ (9)
dh

where C is the temperature rise in degrees centigrade


e: is the winding eddy current loss, per cent
D: is the current density in windings, A/mm2
: is the resistivity of the conductor material
d: is the density of the conductor material
h: is the specific heat of the conductor material
eZ: is the per unit impedance of transformer
eS: is the per unit impedance of supply.
Fig. 7. Failure rate in time.

The estimated maximum time of the overcurrent in the first An analysis of causes and failure modes was performed
moments of the fusion process is 5 s. Table II shows other using the fuzzy fault tree method on the laboratory records of
parameters to calculate temperature increase in the conductor. forensic analysis of 600 transformers in Colombia.
TABLE II The traditional method of fault trees allows the evaluation
PARAMETERS OF THE WINDINGS of reliability based on Boolean algebra. The method allows the
Parameter Value development of complex events, studied as a function of the
Overcurrent time 5s basic faults of the elements, to calculate its probability.
Winding eddy current loss 20% To build the fault tree, a complex event called the maximum
Current density in windings 6 A/mm2 event must be systematically decomposed into intermediate
Resistivity of the conductor material 0.0259 mm2/m
Density of the conductor material 8.89 g/cm3
events connected by logical gates AND and OR until reaching
Specific heat of the conductor material 0.397 J/g C the basic event or the failure cause [28].
Per unit impedance of transformer 6,2% Although the calculations of this model are accurate, it must
Impedance of supply 10,27%
be taken into account that there is an inherent uncertainty in
the degree of influence of each of the causes of failure in the
failure mode.
Replacing these values in equation (9) gives, C=56 C. Fuzzy logic provides an intuitive and natural way of
Considering a hot spot temperature of 125 C before the handling this uncertainty by regarding the probability of
overcurrent, the temperature at the end of the conductor is 181 failure as a fuzzy number. Such a number is a function of two
C. This value is below the allowed limit of 250 C However,

0093-9994 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIA.2017.2688408, IEEE
Transactions on Industry Applications
2016-METC-0636 5

values: the variable of event occurrence and a probability of


currencies of that variable (the fuzzy number corresponding to
a confidence interval value that takes values between 0 and 1).
The confidence interval value represents the belief of the
expert in a selected probability for the fault. These values are
assigned by the expert at each level of the fault tree (basic,
intermediate and maximum levels).
The operators, triangular norms, t-norm and t-conorm are
used to calculate the values of the fuzzy number and the
probability value of the next level according to the associated
logic gate, as shown in the following equations [27].

AND(t norm) : T ( x, y) xy (10)

OR(t conorm) : S ( x, y) 1 (1 x) (1 y) (11)

An additional operation, called implication, must be


performed to find the fuzzy number of the level being
developed. Implication is the term used in fuzzy logic to
describe a logical connector between propositions (usually
input and output).
The extended classical implication operation to allow fuzzy
Fig. 8. Windings fault tree.
membership functions is [28]:
1 (1 0.75) (1 0.60) (1 0.60) 0.96
R ( x, y) max{min[ A ( x), B ( y)],1 A ( x)} (12) 1 (1 0.091) * (1 0.036) * (1 0.015) 13.69%

The following equation describes the operations required to Using equation (12) for the implication operation, the
calculate probabilities among the fault tree levels. calculated fuzzy number for event 1 is:

R ( x, y) max{min[ 0.85,0.96],1 0.85} 0.85


(13)
Therefore, the fuzzy values of fuzzy number and probability
Bk: Matrix resulting from 2x1 of the fuzzy number and the for event 1 are:
probability value after norm and implication operations.
k: Successive level of the fault tree. 0.85
Event 1 :
A: Input matrix of 2xn of basic events of fuzzy number and 13.69%
probability.
BK-1: Calculated matrix of 2x1 of basic events of fuzzy The same procedure is implemented in order to find the
number and probability level (k-1). values of event 2 (cellulose degradation) and also the
T: Calculated fuzzy number of level k. maximum event or failure mode.
P: Calculated probability value of level k This same methodology was implemented for fuzzy fault
T*: Calculated fuzzy number of the norm operation tree construction of: On-load tap changer (OLTC), bushings,
P*: Calculated probability value of the norm operation core, tank and Auxiliary elements. Fig. 9 shows the results of
the statistical evaluation.
The development of the fuzzy fault tree for winding failure The results showed that the main failure modes are due to
is presented below. Fig. 8 shows the construction of the fault the windings and to OLTC and that the main cause of failure
tree in which the basic events are the failure cause. is the deterioration of the insulating material. This
deterioration is manifested in the decrease of the basic
Input values for event 1(mechanical damage, see fig. 8): insulation level (BIL) and in the reduction of the mechanical
resistance of the paper, which protects the integrity of the
Fuzzy Number : 0.75 0.6 0.6 windings.
Probaility : 5.6% 1.4% 0.57% The phrase adopted by engineers the life of the transformer
is the life of its solid insulation [5] comes from the fact that
the restoration of the paper requires the change of the
Applying the norm operation S(W1,W2,W3) gives:
transformer windings.

0093-9994 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIA.2017.2688408, IEEE
Transactions on Industry Applications
2016-METC-0636 6

transformer. This will allow the decision making of


acceptance or rejection.
It is necessary to carry out electrical tests to establish the
operation condition of the transformer according to design
parameters. Table III shows the electrical, mechanical and
insulation tests that must be taken into account in the reception
and maintenance program of EAF transformers. The first
column corresponds to the part on which the test is to be
performed, the second defines specific tasks or clarifications
about the test, the third has the numbering of the fuzzy
variables for the construction of the model; and columns 4 and
5 show the periods in which the tests are to be performed.
The tests performed on the "oil-paper" insulation system are
some of the most important controls that can be done to the
Fig. 9. Statistics of transformer failure. transformer in order to establish the condition of the different
parameters involved in insulation deterioration. Pyrolysis,
Many researchers have recognized that the behavior of hydrolysis and oxidation [30] are the most widely accepted
thermal deterioration of insulation materials might be best mechanisms of paper insulation deterioration.
represented by the Svante Arrheniuss equation of 1880 Dissolved gases, polymerization degree (PD), moisture,
[28][29]. acidity, dielectric strength and power factor are the main
variables that can be evaluated to determine the condition of
15000

H 273 the insulating material [11]. The diagnosis of these variables is
L 9.8 10 18 e (14)
focused on determining the condition of the oil-paper
insulation or the level of aging. This approach can provide
Fig. 10 shows the graphical representation of this equation.
valuable information on the condition of the transformer [30].
It is assumed that for a temperature of 110 C, which is the
Table III shows the conventional electrical tests that define
nominal temperature of hot spot [24], there will be a loss of
the operating condition of the transformer. The results of these
life in the insulation of 1 in per unit (no wear). Consequently,
tests determine whether the machine is operating properly
lower or higher values of temperature will cause a gain or loss
without any internal problems, especially in the windings.
of life respectively.
In addition, the transformer should be evaluated in the
This model will be used in the output space of the fuzzy
reception process by an expert to determine the operation and
model implemented in this paper to calculate the remaining
condition of all auxiliary elements, such as temperature
life of the transformer.
indicators, pressure indicators, valves, oil leakage or any other
anomaly outside the parameters of normal operation. See
Table III.
B. Phase 2: Fuzzy Model Development
This section presents a new method for the estimation of the
parameters and the introduction of the fuzzy model.
The construction of the fuzzy model is based on the Takagi-
Sugeno system [20] where the fuzzy inference system (FIS) is
composed of rules, See equation (15). Each rule has input
variables associated with fuzzy sets which are related to a
linear output equation.

Fig. 10. Transformer life as a function of temperature.


If x1 is A11 and ...and xn is A1n then y1 f ( x1 ,..., xn )
If x1 is A21 and ...and xn is A2 n then y2 f ( x1 ,..., xn ) (15)
III. RECEPTION PROPOSAL OF EAF TRANSFORMERS
A description of phases proposed for acceptance or If x1 is Am1 and ...and xmn is Amn then ym f ( x1 ,..., xn )
rejection of used transformers is presented below. The first
phase corresponds to the identification of the fuzzy variables The input variables x1,,xn of each rule correspond to the
used in the proposed model. Table III shows a total of 32 set of values shown in Table III; A11,,Am1 are the fuzzy sets
variables. that comprise all the possible values of the input variable x1.
A. Phase 1: Transformer Reception The linear output equations of each rule define the trajectory
of a nonlinear equation, in this particular case, the Arrhenius
This procedure aims to describe the instructions for the
equation.
reception and evaluation of the remaining life of the

0093-9994 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIA.2017.2688408, IEEE
Transactions on Industry Applications
2016-METC-0636 7

TABLE III
TESTS FOR THE RECEPTION AND MAINTENANCE OF EAF TRANSFORMERS
Fuzzy Maintenance
Element Task
Variable Monthly Annual
Oil tests:
Total Dissolved Combustible Gases Before and after any test x1 6 months
(TDCG)
Polymerization degree x2 6 months
Moisture x3 6 months
Acidity x4 6 months
Dielectric strength x5 6 months
Interfacial tension x6 6 months
Power factor (PF) x7 6 months
Oil screen tests Main tank and tap change x8 6 months
Electrical tests:
Impulse test Before buying x9
Applied potential test Before buying x10
Induction potential test Before buying x11
Ratio test Before buying x12
Regulation test Before buying x13
Efficiency test, full load Before buying x14
Copper loss Before buying x15
Core loss Before buying x16
Data Reading sheet: X
Record of discharge pulses Maximum oil temperature x17 X
Temperature indicators Maximum winding temperature x18 X
Thermometers Internal and external of oil and water x19 X
flow
Switch meter Record of number of operations x20
Operating tests Testing of protection devices x21 X
Protection system:
Overvoltage arresters Oil leaks and overheating x22 X
Discharge meters Record of discharge meters x23 X
Conservator: Oil leakage control X
Buchholz relay Indicator reading testing x24 X
Oil level indicator Change of dehydrating breather if pink x25 X
Breathers x26
Switch: Inspection and testing of the engine X
control
Motor driving unit Inspection of the drive shaft coupling x27 X
Transmission shafts Control of the limit switch x28 X
Upper position indicator Check position of top tap x29 X
Record of number of operations X
Off-load tap changer Check electrical and mechanical x30 X
interlocks
Control cabinet Collection of gas and oil x31 X
Relay RS Filter change, manometer reading x32 X
Oil leakage inspection X

The contribution value of each rule, that is ym, is considered


in the complete solution of the system by means of the
following equation [20]:

n
wi yi
y0 (16)
i 1 wi

Therefore, Fig. 11 shows the transformer model


implemented in the Takagi-Sugeno inference system. In this
way, it is very easy to see the model to evaluate the condition
of a transformer of EAF defined by the fuzzy logic model
proposed in this paper.
The optimal number of fuzzy sets of the model is
determined by applying "fuzzy cluster" methods. To do this, a Fig. 11. Fuzzy model to assess the EAF transformer.
partition of the universe of the output variable is done, in this
case the Arrhenius equation, and then the input space is

0093-9994 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIA.2017.2688408, IEEE
Transactions on Industry Applications
2016-METC-0636 8

projected, which are the sets of input variables. See Fig. 12. The heuristic method was one of the first methods of
The criterion used to obtain optimal fuzzy sets of the model identification for fuzzy linguistic models. This method is
is to minimize the distance between the output data and the based on the experience and knowledge of the expert.
center of each cluster. The case study proposed in this paper is carried out
applying the knowledge acquired over several years of
maintaining transformers for the steel industry. It considers
the application of a nonlinear equation that uses two sets, B11
and B21, to represent the behavior of the Arrhenius model. See
equation (18).

3
y (1 B11 B21) (18)

The benefit of using equation (18) lies in a smaller number


of clusters to represent the behavior of the Arrhenius model,
which translates into lower computational cost. Fig. 13
Fig. 12. Relation of sets in the Takagi-Sugeno system.
compares the results of the three models: Arrhenius, Complete
After an iterative procedure of optimization of distances, the fuzzy model and the nonlinear equation model.
number of fuzzy sets or clusters is obtained as well as the
degree of belonging of the output data to each set for an error
index given. This technique seeks to find subsets with a
certain degree of similarity within a data collection, which
correspond to the space of the output variable, and then
projecting them to the input space.
The criterion used in the optimization process is defined by
the regularity criterion (RC), which is defined by the
following equation.

y i y i y i y i
KA KB
A AB B AB

RC i 1 i 1 (17)
2

where KA and KB are the number of data of two sets from


the adjusted set, yA(i) and yB(i) are the outputs of sets A and B,
yAB(i) is the estimated output for the set A with the model
identified using a data set B, and yBA(i) is the estimated output
for the set B with the model identified using the date set A. Fig. 13. Comparison of models.
For example, for the variable TDCG that corresponds to the
fuzzy variable x1 four fuzzy sets are proposed (A11, A21, A31 y In this comparison, the values of Total Dissolved
A41) for the optimization model. Thus, it is possible to build Combustible Gases (TDCG) were used as the input space. As
15=(24-1) fuzzy models. First, four fuzzy models must be built observed in Fig. 13, the nonlinear equation has a better fit in a
with one input variable each. For each model the RC wider range of input variables taking into account only two
regularity criterion is calculated as shown in Table IV. The clusters.
model with lower RC is selected for 11 combinations.
TABLE IV IV. APPLICATION TO A CASE STUDY
VALUES OF REGULARITY CRITERION An installation of an arc furnace in a steel production
Model Input variables RC industry in Colombia is considered in this case study. The
installation did not achieve the objectives even after having
1 A11 0.623
2 A21 0.745 the heat transfer repaired, the system re-ordered using a foamy
3 A31 2.467 slag technique, and arrangements made to work at higher
4 A41 3.456 power and voltage levels. Thus, the decision was made to get
5 A11,A21 0.845
6 A11,A31 0.567 a new transformer.
7 A11,A41 0.496 The objectives of the new strategy were as follows:
8 A11,A21 A31 0.345 To reduce the tap-to-tap time, thereby increasing annual
9 A11,A21 A41 1.341
10 A11,A31 A41, 0.932
production.
11 A21,A31 A41, 2.016 To operate the EAF at a high power factor (PF) using
long arcs in order to reduce electrode consumption.

0093-9994 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIA.2017.2688408, IEEE
Transactions on Industry Applications
2016-METC-0636 9

The old transformer failed after launching this strategy. The TABLE VI
transformer of 14 taps, 13800/100-400V and 24/28MVA was RESULTS OF TDCG
urgently replaced by a transformer of 9 taps, 13800/380-700 V
Concentration of Gases ppm % -Fuel gases
and 27/30 MVA found in Pittsburgh, U.S.A. The replacement
was manufactured in 1969 and has more than 15 years of not H2 - Hydrogen 0.00 61.9%
being used. It is considered that this transformer has a wear CH4 - Methane 83.18 7.5%
CO- Carbon Monoxide 220.17 4.9%
out level and must be evaluated before setting it into C2H4 - Ethylene 6.09 0.7%
operation. This is the challenge of the technique previously C2H6 - Ethane 124.74 0.8%
presented. Table V shows the results of the tests carried out on C2H2 - Acetylene 0.00 0.0%
Total Dissolved Combustible Gases 434.18 100.0%
the transformer. Insulation and electrical tests were performed;
and the assessment of an expert in each of the auxiliary
components of the machine was considered.
According to the report results, the transformer is in good
condition and is able to operate normally. To evaluate the
remaining life using the conventional method, the laboratory
issued a diagnosis on the condition of the transformer.
Table VI shows the results of the gas chromatography
analysis performed on the assessed transformer. The values of
key gases are in parts per million (ppm). Fig. 14 shows the
graph of gas percentages. The report stated that the
concentration of TDCG was within normal limits, and based
on the DP= 953, the estimated remaining life of the
transformer was at 97%.
Fig. 14. Percentages of key gases.
TABLE V
RESULTS OF THE TRANSFORMER TEST REPORT
Fuzzy Fig. 15 shows a photograph of the transformer.
variable Fuzzy variable description Measure value

Oil tests:
x1 TDCG 1073.59
x2 Polymerization degree 953
x3 Moisture 8 mg/kg (ppm)
x4 Acidity < 0.02 (mg KOH)/(g of oil)
x5 Dielectric strength 37 kV
x6 Interfacial tension 36.3 mN/m
x7 Power factor (PF) 0.22%
x8 Oil screen tests Good condition
Electrical tests:
x9 Impulse test 100%
x10 Applied potential test 100%
x11 Induction potential test 100%
x12 Ratio test 0.04%
x13 Regulation test 1.3%
x14 Efficiency test, full load 99.7%
x15 Copper loss 209.4 kW
x16 Core loss 31.5 kW
Data Reading sheet:
x17 Record of discharge pulses Good condition
x18 Temperature indicators For cleaning
x19 Thermometers For cleaning Fig. 15. Assessed EAF transformer.
x20 Switch meter Good condition
x21 Operating test Good condition
The implementation of the methodology proposed in this
Protection test: paper shows that the remaining life of the transformer is 0.959
x22 Overvoltage arresters Good condition per unit. The transformer was installed after the evaluation
x23 Discharge meters Good condition and has been properly operating for the last 10 years.
Conservator:
x24 Buchholz relay Good condition In addition to this study, a 600 EAF transformer lot was
x25 Oil level indicator Good condition evaluated using this methodology. These transformers are
x26 Breathers For cleaning distributed in the territory of Colombia.
Switch:
x27 Motor driving unit Good condition
The results of gas analysis were considered as the fuzzy
x28 Transmission shafts Good condition input variables of the model in this study. Table VII shows the
x29 Upper position indicator Good condition technical characteristics of the transformer lot selected for the
x30 Control cabinet With oxidation
x31 Off-load tap changer Good condition
study.
x32 Relay RS Good condition

0093-9994 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIA.2017.2688408, IEEE
Transactions on Industry Applications
2016-METC-0636 10

TABLE VII remaining life of these transformers. It can be observed that


CHARACTERISTICS OF THE EVALUATED TRANSFORMERS the graph has a behavior similar to the normal statistical
Characteristics of the machine Application range distribution.
Power range 15 100 MVA
Voltage range high side 13.2 34.5 kV ACKNOWLEDGMENT
Voltage range low side 200 700 kV
Chronological age range 5 -30 years The authors thank the Universidad Nacional de Colombia.
Measurement type On-line
REFERENCES
The proposed nonlinear model was used for this analysis [1] E. A. Cano Plata, A. J. Ustariz Farfan, and E. H. Taca, Hornos de arco
elctrico, Manizales, Colombia, Univ. Nacional Colombia, 2011.
and two clusters of the input variable were used. The figure [2] E. A. Cano Plata and H. E. Tacca, Arc furnace modeling in ATP-
shows that the expected life of the transformer lot is 26 years. EMTP, in Proc. 6th IPST, Montreal, QC, Canada 2005, pp. 19-23.
Fig. 16 shows the distribution of the expected life of these [3] E. A. Cano- Plata, A. J. Ustariz Farfan, O. J. Soto-Marin, Electric arc
furnace model in distribution system, IEEE Transaction on Industry
transformers. Applications, Vol. 51, No 5, Sep 2015.
[4] M. J. Heathcote, The J&P transformer book, Twelfth edition, Elsevier
Ltd, 1998.
[5] S. Chakravorti, D. Dey, B. Chatterjee, Recent Trends in the Condition
Monitoring of Transformers. [OnLine]. Available: www.springer.www
[6] E. A. Cano-Plata; O. J. Soto-Marn; A. J. Ustariz-Farfn, Validation
and reception of electric arc furnace Transformers, presented at the IAS
Annual Meeting, Portland, OR, USA, 2016, pp. 16.
[7] D. F. Phillips (2005, Nov), Mantenimiento preventivo de
transformadores de Hornos de Arco, ABB Kraft A/S National
Transformer. [OnLine]. Available: http://catedra.ing.unlp.edu.ar
[8] IEEE Standard Test Code for Liquid-Immersed Distribution, Power, and
Regulating Transformers, IEEE Standard C57.12.90, 2015.
[9] IEEE Guide for the Application and Interpretation of Frequency
Response Analysis for Oil-Immersed Transformers, IEEE Standard
C57.149, 2012
[10] IEEE Guide for Diagnostic Field Testing of Fluid-Filled Power
Transformers, Regulators, and Reactors, IEEE Standard C57.152, 2013.
Fig. 16. Statistical distribution of the expected life of the transformer lot in [11] IEEE Guide for the Interpretation of Gases Generated in Oil-Immersed
Colombia. Transformers, IEEE Standard C57.104, 2008.
[12] IEEE Recommended Practice for Partial Discharge Measurement in
Finally, the methodology proposed to estimate the Liquid-Filled Power Transformer and Shunt Reactors, IEEE Standard
C57.113, 2010.
remaining life of EAF transformers might be applied to [13] IEEE Standard General Requirements and Test Procedure for Power
reception programs of used transformers, maintenance Apparatus Bushings, IEEE Standard C57.19.00, 2004.
programs and even new transformers. [14] James L, et al., Model-based monitoring of transformers. Massachusetts
Institute of Technology, Laboratory for electromagnetic and electronic
systems, 1995.
V. CONCLUSIONS [15] Ming L, Jonsson B, Bengtsson T, Leijon M, Directivity of acoustic
signals from partial discharges in oil. IEE Proc Sci Meas Technol
The unstable behavior of the electric arc and the high values
142(1):8588, 1995.
of reactive power are the main causes of the electric and [16] Lazarevich AK, Partial discharge detection and localization in high
magnetic stress that bring about problems of quality in both voltage transformers using an optical acoustic sensor, The Virginia
the system and the transformer; thus a condition of premature Polytechnic Institute and State University, 2003.
[17] Gulski E, Kreuger FH, Krivda A, Classification of partial discharges.
aging in the transformer. IEEE Trans Electr Insul 28(6):917940, 1993.
By integrating the transformer test results and using the [18] IEEE Recommended Practice for Performing Temperature Rise Tests on
Takagi and Sugeno model, it is possible to assess the level of Oil-Immersed Power Transformers at Loads Beyond Nameplate Ratings,
IEEE Standard C57.119, 2001.
wear or the condition of the transformer in order to determine [19] N. Y. Utami; Y. Tamsir; A. Pharmatrisanti; H. Gumilang; B. Cahyono;
the feasibility of its reuse. R. Siregar, Evaluation condition of transformer based on infrared
The fuzzy model considers many variables. This case study thermography results, IEEE 9th International Conference on the
Properties and Applications of Dielectric Materials, Jul. 2009.
took into account the experience in maintenance of EAF [20] G. Chen, T. Tat, Introduction to Fuzzy Sets, Fuzzy Logic, and Fuzzy
transformers to make its relation with the fuzzy logic only in Control Systems, CRC Press LLC, 2001.
the most representative variables through the Pareto [21] D. Sez MSc, A Cipriano, A. W. Ordys, Optimisation of Industrial
Processes at Supervisory Level, Application to Control of Thermal
equilibrium. This has allowed the creation of a new model Power Plants, [OnLine]. Available: www.springer.www.
which uses fuzzy logic with fewer variables and in turn less [22] L. Chuen, Fuzzy logic in control systems: Fuzzy Logic ControllerPart
computational cost. See equation (18) 1, IEEE Transaction on systems, Man and Cybernetic, Vol. 20, No.2,
1990, p.p. 404-435.
It is possible to assess the remaining life of a transformer by
[23] S. B. Pandey, C. Lin Estimation for a live model of transformer
implementing a fuzzy model from the input variables shown in insulation under combined electrical and thermal stress, IEEE
Table III. The implementation of this new model allowed the Transaction on Reliability, Vol 41, No. 3, 1992, p.p.466-468
evaluation of the technical condition of 600 transformers that [24] D. M. Allan, Transformer life assessment aids system planning. Power
International (McGraw Hill InC.) Second Quarter 1991, pp 39-40.
are currently in operation. Figure 15 shows the trend of the

0093-9994 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIA.2017.2688408, IEEE
Transactions on Industry Applications
2016-METC-0636 11

[25] IEEE Guide for Loading Mineral- Oil-Immersed Transformers and Step-
Voltage Regulators, IEEE Standard C57.91, 2011.
[26] IEEE Requirements for Arc Furnace Transformers, IEEE Standard
C57.12.10, 2012.
[27] A. Shivhare. (2015, June). Online insulation monitoring of electric arc
furnace. [Online]. Available: www.linkedin.com/pulse/online-insulation-
monitoring-electric-arc-furnace-ashutosh-shivhare.
[28] T. J. Ross, J. M. Booker, W. Jerry Parkinson, Fuzzy Logic and
Probability Applications Bridging the Gap, Society for Industrial and
Applied Math. and American Statistical Association, 2002.
[29] R. R. Dixon, "Thermal aging predictions from an Arrhenius plot with
only one data point," IEEE Transactions on Electrical Insulation, vol.
E1-15, no. 4, 1980,p. 331.
[30] W. Flores, E. Mombello, G. Ratta, J. A. Jardini, Life of Power
Transformers Immerserd in Oil. -State-of-the-Art- Part I. Correlation
between Life and Temperature, EEE Latin America Transactions, Vol.
5, No 1, mar 2007.

Eduardo A. Cano-Plata (S96, M98,


SM15) was born in Neiva, Colombia, in
1967. He received the B.Sc. and
Specialist Engineering degree in 1990
and 1994 from Universidad Nacional de
Colombia, Manizales, both in Electrical
Engineering. He received the Doctorate
degree in Engineering in 2006 from the
Universidad de Buenos Aires. Since
1994, he has been a full professor at the Universidad Nacional
de Colombia, Manizales branch. His research interests include
Power Quality, Power Electronics, Power Systems Grounding
and Smart Grids. He is the current director of the
Transmission and Distribution Network Group, GREDyP.

Oscar J. Soto-Marin (S15) was born in


El Cairo, Colombia, in 1977. He
received the B.Sc. and M.Sc. (Laureate
thesis) degrees in electrical engineering
from the Universidad Nacional de
Colombia, Manizales, Colombia, in 2005
and 2015, respectively, where he is
currently working toward the Ph.D.
degree in electrical engineering. His
research interests include power quality and power systems
grounding.

Armando J. Ustariz-Farfan (S08,


M12) was born in Urumita (Colombia)
in 1973. He received a bachelors degree
in Electrical Engineer in 1997, and a
Masters in Electric Power in 2000 from
the Universidad Industrial de Santander.
He received the Ph.D. degree in
Electrical Engineering at the Universidad
Nacional de Colombia, in 2011. He is a
researcher and associated professor with the Electrical,
Electronic and Computer Engineering Department, at
Universidad Nacional de Colombia, Manizales Branch. His
research interests include Power Definitions under Non-
sinusoidal Conditions, Power Quality, and Power Electronics.

0093-9994 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

You might also like