Received September 1, 2019, accepted December 5, 2019, date of publication December 16, 2019,

date of current version December 23, 2019.

Digital Object Identifier 10.1109/ACCESS.2019.2959790

Research on the Switching Arc Loss of on-Load

Tap Changer
DONGDONG SONG , JIWEI MA, YUQUAN MA, HONGJU LIN, AND SHENGTAO LIU
College of Mechanical Electrical and Engineering, Hebei Normal University of Science and Technology, Qinghuangdao 066004, China
Corresponding author: Dongdong Song (mapex_dongdong@126.com)
This work was supported by the Hebei Colleges and Universities Science and Technology Research of Youth Foundation under Grant
QN2018211. Project name: power electronic on-load tap changer modeling research.

ABSTRACT The on-load tap changer (OLTC) switching arc will affect the life of the contact, so it
is necessary to test its working condition. Based on DQ coordinate transformation, fundamental wave
component extraction, and wavelet algorithm, proposed an OLTC arc detection method based on energy
conservation theory in this paper. Used the fundamental wave component extraction algorithm extracted the
transformer excitation parameters, and established the OLTC equivalent circuit by combining the Habedank
arc model. Compared the differences between the follow effect and frequency calculation of simulated
and measured waveforms, which were caused by the five orthogonal transformation algorithms, including
Transport-Delay, Hilbert Transformation, Inverse Park Transformation, SOGI and Improved SOGI. The DQ
coordinate transformation based on the Transport-Delay orthogonal converter is determined to calculate the
active power of each part of the circuit. According to the principle of energy conservation, obtained the data
of arcing loss under nine simulated conditions. Based on the simulation circuit, built the OLTC prototype
and dSPACE hardware in loop simulation platform, and measured three load conditions. The sampling signal
was filtered by the wavelet algorithm. Considering the measurement and calculation errors, the simulation
and measured results show that the actual arcing power has a consistent change with the calculated value
between different working conditions, and the maximum difference between simulation results and measured
results is thirty times and ten times. It can distinguish the difference of arcing energy of diverter switch under
different working conditions and has the measurability. Meanwhile, it verifies the effectiveness of this study,
hoping to provide a theoretical basis for developing OLTC on-line contact detection system.

INDEX TERMS On-load tap changer, arcing power, condition monitoring, orthogonal transformation, DQ
coordinate transformation.

I. INTRODUCTION rent waveform by using empirical modal analysis and Hilbert-

The power quality problem of distribution network becomes
more and more obviously with the diversity of power supply
structure and load. In the 10∼35kV level, the regulating
transformer with the on-load tap changer (OLTC) becomes
the basic configuration. As an accessory equip-ment, OLTC
is required to have high reliability, so it needs to be off-grid
maintenance and overhaul. OLTC tester is used to measure
the equipment parameters, sacrificing the reliability of power
supply. Therefore, the research of OLTC online status moni-
toring has become a hot spot.
References [1] aimed at OLTC used in aerospace system
power grid, carried out the spectral analysis of switching cur-
The associate editor coordinating the review of this manuscript and
approving it for publication was Xue Zhou . by synchronizing vibration signals, Fourier transform, etc.,
Huang transform, and obtained the abnormal components in
current signal to realize equipment detection. The detection
process is qualitative judgment without quantitative calcu-
lation and analysis [1]. Reference [2], proposed an OLTC
mechanical fault detection method based on hidden Markov
model. Through the time series analysis of vibration signals,
established a fault feature extraction method, and completed
the training based on HMM and the establishment of the
expert database, which could realize the accurate identifi-
cation of OLTC faults. The average success rate of failure
classification was about 95%, but the quantitative analysis
of switching arc loss was not reflected [2]. Reference [3]
used the method of dynamic time warping to analyze the
mechanical state of OLTC, and forms an energy matrix

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see http://creativecommons.org/licenses/by/4.0/
VOLUME 7, 2019 180793

and identifies the differences of different vibration signals
through characteristic parameters, to complete fault identi-
fication. The fault identification effect could reach 100%,
which also did not involve the calculation process of arcing
loss. Similar vibration signal analysis can be found in [4],
[5]. Reference [6] combined vibration signal with an acoustic
signal, and proposed the analyzing method of OLTC vibra-
tion acoustic signal by the Savitzky-Golay filter, to obtain
OLTC working status and arcing information. This method
could only qualitatively calculate the arc position informa-
tion and did not calculate the arcing energy quantitatively.
Similar in [7], [8] analyzed the acoustic signals of OLTC
working process by discrete wavelet method, obtained the
frequency and energy spectrum of different switching points,
and qualitatively judges the working condition of the contact.
Reference [9] calculated the instantaneous differential power
of transformer terminal and obtained the OLTC arcing power,
to realize the detection of contact state. The calculation pro-
cess needs to use the voltage across the switch, and the actual
measurement process is difficult for the actual equipment.
The arc model used in the simulation is difficult to keep
consistent with the actual switching arc, and the contact
wear threshold has a large fluctuation, so the parameters are
not easy to determine. Also, the difference of the algorithm
results under different load conditions is not considered in the
measured process.
There are other detection methods, including measuring
the oil temperature and dissolved gases, reflecting abnormal
arcing, etc. [10], [11]. However, the numerical calculation of
switching arc energy was not targeted.
The existing OLTC online monitoring used electri-
cal or non-electrical sensing elements to obtain state signals
such as acoustic, optical and vibration of OLTC in the work-
ing process indirectly. Most of them belonged to the qualita-
tive judgment of the equipment state, which was not capable
of quantitative calculation of the contact arcing power. More-
over, the system structure was complex relatively, and there
were larger interference factors in the measured signal, which
affected the detection effect. It needs to be optimized.
Base on [9], this paper further studies the monitoring
method of switching arc energy based on the principle of
energy conservation and increases the actual measurement
calculation under different load conditions. By comparing
arcing information before and after switching, the severity of
loss caused by arcing can be understood indirectly, which can
be used as an auxiliary basis to judge the working condition
of diverter switch contact, and realized online monitoring
of OLTC switching arc and mechanical contact working
condition.
II. PRINCIPLE AND STRUCTURE
The basic principle structure of the research is shown in Fig.1,
where U1 , I1 , Uload , Iload is the transformer input voltage,
current and load voltage, current respectively. P1 is input
power of transformer, P2 is load consumption power, Pm is
transformer excitation resistance power, PRσ is transformer
180794
D. Song et al.: Research on the Switching Arc Loss of OLTC
FIGURE 1. OLTC diverter switch arcing power monitoring principle
structure diagram.
FIGURE 2. Topology structure and working sequence diagram OLTC.
dc resistance consumption power, PR1R2 is OLTC transition
resistance loss.
The whole structure includes the transformer system,
acquisition storage unit, algorithm unit, computing unit and
expert database. The acquisition storage unit is responsible
for the transformer input and output current and voltage signal
sampling and storage. Algorithm unit complete the coordi-
nate transformation and getting active component parameters,
using switching key point identification algorithm to get a
half-bridge and bridge range information. Computing unit
can realize the active power distribution transformer system
calculation, to get the diverter switch arcing power loss data,
which can be used as the data source or comparison data of the
subsequent expert system database, to indirectly determine
the working conditions of mechanical contacts. This paper
mainly completes the acquisition storage, algorithm and cal-
culation, the expert system not mentioned for the time being.
In the calculation process, regarding the voltage regulating
transformer and load as a unified system with the same
reference voltage. On this basis, it is meaningful to carry
out DQ coordinate transformation, which can be used for
mathematical calculation of the d-axis component, meeting
the theoretical requirements.
III. EQUIPMENT SYSTEM STRUCTURE AND MODE-LING
A. OLTC TOPOLOGY OF THE PROTOTYPE
The equipment prototype of this study, referring to a domestic
OLTC product, is shown in Fig.2a.
S1 ∼S3 are transformer winding taps, K is the main switch-
ing contact, K11 and K12 are transition contacts, and R1 , R2
are transition resistance. Under normal conditions, K, K12 are
connected with S2 , and K11 is suspended. According to the
up or down timing sequence, the tap switching is completed.
The arcing positions of K and K12 are marked in the Fig.2b.
To study the applicability of energy conservation in the OLTC
switching process, a simulation was carried out in the early
VOLUME 7, 2019
D. Song et al.: Research on the Switching Arc Loss of OLTC
FIGURE 3. Diagram of simulation equivalent circuit.
stage which could analyze the energy distribution of the
system during the switching process.
B. MODELLING AND SIMULATION ANALYSIS
The arcing of OLTC equipment is the same as other
switchgear with uncertainty, which makes the arc modelling
difficult. This study did not focus on the accuracy of the
arc model, but only on the measurability of switching arc
power. Therefore, setting the arc model as a fixed parameter
tentatively, the measurability of arcing power value under
this parameter was investigated, and ignored the deviations
between simulation results and subsequent measurements.
The equivalent circuit of the simulation model was estab-
lished according to the prototype structure of the equipment,
as shown in Fig.3a. Due to the limitations of the conditions,
the model used five switches K1 ∼K5 to replace the original
mechanical contacts of K, K11 and K12 . The model used
autotransformer, w1 and w2 are series winding and common
winding, respectively. A and B are two taps, respectively,
and U1 is the power supply voltage. V1 and V2 respectively
measured transformer winding voltage and load voltage, A1
and A2 respectively measured system injection transformer
current and load current.
To ensure the equivalence with the real equipment, set the
working sequence of the model as shown in Fig.3b. K1 ∼K5
represents the action sequence of five switches, with the high
level being closed and low level opened. According to the
original equipment parameters, the timing sequence of five
switches was set as t1 = 0.0199s, t2 = 0.015s, t3 = 0.0301s,
t4 = 0.0259s, t5 = 0.026s, t6 = 0.017s, and t7 = 0.020s.
C. MODEL WORK EQUIVALENT CIRCUIT
The power variables in Fig.1 are distributed during the
switching process. To obtain the values of the power loss
distribution in each stage, the equivalent circuit of switching
should be analyzed, added the values of each stage together
to get the power loss in the whole switching process, and then
the arcing loss was gotten. According to the timing sequence,
the switching process is equivalent to six circuit models. See
Fig.4.
Where, R1//2 is the transition resistance parallel value;
Zm , Zw1 , Zw2 are respectively the transformer excitation
impedance, w1 leakage impedance, w2 reduced leakage
impedance; I1 , I0 , Iw1 , Iw2 , Iw , Ir1 , Ir2 , I1//2 , Iload are respec-
tively the total current excitation current, w1 current, w2
VOLUME 7, 2019
FIGURE 4. Equivalent circuit of model.
current, total winding current, R1 current, R2 current, parallel
transition resistance current and load current.
The output power of the power supply is:
S1 = P1 +jQ1 = ud1 id1 + uq1 iq1 +j(−ud1 iq1 +uq1 id1 ) (1)
The power consumed by the load is:
S2 = P2 +jQ2 = ud2 id2 +uq2 iq2 + j(−ud2 iq2 +uq2 id2 ) (2)
where P1 , Q1 represents the total active and reactive power of
the regulating transformer; ud1 , id1 is the d-axis component of
u1 and i1 ; uq1 , iq1 is the q-axis component of u1 and i1 ; P2 ,
Q2 is the active and reactive power of the load; ud2 , id2 is
the d-axis component of uload and iload ; uq2, iq2 is the q-axis
component of uload and iload . The voltage of transformer
step winding is ustep , the dc resistance of the transformer
is Rσ , and the excitation impedance of the transformer is
Rm and Xm , respectively, taping winding voltage is uw1 , uw2
respectively, The transition resistance of OLTC is R1 , R2 , and
the corresponding power loss is PR1 and PR2 respectively.
Among the above parameters, u1 , i1 , uload , iload is the direct
measurement parameter, which can be obtained through PT
and CT, get ustep from measured data or manufacturer data,
and the rest are indirect calculation parameters.
The equivalent circuit (1) corresponds to the time series t1
interval, K1 and K2 closed, K2 branch is short connected by
K1 branch. The winding voltage of the transformer is u1 , the
difference between P1 and P2 should be the excitation loss:
Pm = P1 − P2 = u2d1 /Rm (3)
The equivalent circuit (2) corresponds to the time series
t2 interval, K1 ∼K3 are closed, R1 is short connected by K1
branch. The losses in the loop include Rm , R2 and Rσ :
(
Pm = u2d1 /Rm
(4)
PR2Rσ = PR2 + PRσ = u2d1 /(R2 + Rσ )
where, PR2Rσ is the sum losses of R2 and Rσ .
The equivalent circuit (3) corresponds to the time series t3
interval, K2 , K3 are closed, and OLTC is in bridge state. The
losses in the loop include Rm , Rσ and R1 , R2 . Because the
leakage reactance of the transformer is small, its influence is
ignored :
(
Pm = u2d1 /Rm
(5)
PR1R2Rσ = PR1 + PR2 + PRσ = u2d1 /Rx − P2
180795
 D. Song et al.: Research on the Switching Arc Loss of OLTC

where, PR1R2Rσ is the sum of power consumed by R1 , R2

and Rσ . In this stage, R1 , R2 and Rσ constitute the resistance
network, and its equivalent resistance is:
Rc (Rb + Rw2σ )
Rx = Rload + Ra + (6)
Ra + Rb + Rw2σ
R1 Rw1σ R2 Rw1σ
where, Ra = R1 +R 2 +Rw1σ
, Rb = R1 +R 2 +Rw1σ
, Rc =
R1 R2
R1 +R2 +Rw1σ . Rw1σ , Rw2σ are dc resistors of two windings,
respectively, and Rw1σ : Rw2σ = 1:1. FIGURE 5. Transformer no-load current test signal and filter signal.
The equivalent circuit (4) corresponds to the time series t4
interval, with only K3 is closed and OLTC in R2 half-bridge
state. The losses in the loop include Rm , Rσ and R2 : on the feasibility of arcing energy calculation. Based on sim-
 ulation, used the dSPACE platform to complete the analysis
Pm = ud1 /Rm
2
 and calculation of the sampled signal and realize the online
PR2Rσ = PR2 + PRσ = (7) detection research.
ud1 /(R2 + Rσ + Rload ) − P2
 2

D. ELEMENT MODEL
The equivalent circuit (5) corresponds to the time series t5 (1) Transformer model establishment. It was necessary to
interval, with the closure of K3 and K5 . The losses in the loop determine the transformer parameters by measuring the
include Rm , Rσ and R1//2 : experimental equipment and used the no-load test and short
 circuit test to obtain the excitation parameters and leakage
Pm = ud1 /Rm
 2
resistance impedance parameters. To simplify the calculation
PR1//R2Rσ = PR1//R2 + PRσ = (8) process and facilitate the modelling, extracted the resistive
ud1 /(R1//2 + Rσ + Rload ) − P2
 2

active component parameters and inductive reactive compo-
nent parameters to determine the excitation parameters. There
The equivalent circuit (6) corresponds to time series t6
were many harmonic components in the test waveform, which
interval, and K3∼K5 are closed. The losses in the loop include
filtered by the fundamental wave component extraction algo-
Rm and Rσ , which is consistent with the interval loss of
rithm. As shown in Fig.5, is the no-load current original signal
sequence t7 :
( and fundamental wave component signals.
Pm = u2d1 /Rm The mean square relative error was used to calculate the
(9) extraction accuracy [12], [13].
PRσ = u2d1 /(Rσ + Rload ) − P2
L
According to energy conservation, the transformer input 1 X |i(tn ) − ic (tn )|2
εcurr
2
= (12)
power should be consistent with the sum of load loss, trans- L i(tn )2
n=1
former body loss and OLTC power consumption, that is:
where, L is the signal length, i(tn ) and iC (tn ) are the original
Ploss = P1 − P2 = Pm + PRσ + PR1R2 + PdeltRarc (10) signals and calculated signal respectively, tn = nTs (n =
1,2,3... L), Ts is the sampling period. This equation can
where, Ploss is the total power loss of transformer system, Pm , describe the relative deviation between the fundamental wave
PRσ , PR1R2 are equivalent to the total loss during the switch- component and the original signal, which is 2.82E-1. On this
ing, and PdeltRarc is the arcing power of OLTC diverter switch, basis, DQ decomposition of fundamental wave data was per-
which can be obtained by calculating the power difference of formed to calculate excitation impedance parameters of the
the transformer system: transformer under no load.
PdeltRarc = P1 − P2 − Pm − PRσ − PR1R2 (11) ud1 + juq1
= id1 + jiq1 (13)
Rm + jXm
where, P1 and P2 could be obtained by calculating the sam- Rm = BXm (14)
pled signal, and the power losses of the other three are the
u +u A u +u B
results after adding up the calculated values at each stage, where, A = iid1q1
, B = uq1 d1
q1 A−ud1
, Xm = i q1(B2 +1)
d1
, ud1 , uq1 , id1 ,
d1
which can be calculated by the formula (1∼9). So far, realized iq1 represent the active and reactive components of system
the calculation of arcing loss based on power conservation by voltage and current. By calculating the Rm = 1210, Xm =
analyzing the equivalent circuit of switching. All parameters 61.89, Rσ ≈ 1 ∼ 2.
were obtained through sampling and actual measurement (2) Diverter switch model establishment. The description
calculation. Since the parameters to be tested have low real- model of arc diverter switch needs to be established. Accord-
time requirements, the requirements can be met through ‘ex ing to reference [14]–[16], the black-box arc model was
post calculation’. Therefore, the early simulation verification suitable for circuit simulation and easy to solve, but the model
was completed under the upper computer, mainly focusing parameters were uncertain. According to the above analysis,

180796 VOLUME 7, 2019

D. Song et al.: Research on the Switching Arc Loss of OLTC

the purpose of the simulation was to verify the measurability

of switching arc power, and only highlight the existence of
arc. Based on qualitative research, then the quantitative prob-
lem was considered. Therefore, the deviation of the black-
box arc model parameters was ignored for the time being. FIGURE 6. Inverse park structure.
In Fig.2b, set the K1 , K2 corresponding to the down switch-
ing and K4 , K5 corresponding to the up switching as arc
switch models, and the other switches were arc free switches.
To be consistent with the follow-up measurement process,
the arcing medium is air. Considering the characteristics of
air medium arc, and meet the requirements of current peak
and zero-crossing arc characteristics, Habedank model with
the characteristics of Cassie and Mayr models was selected
FIGURE 7. Second order generalized integrator schematic diagram.
as the switching arc model, [17]. This model was described
as:
1 u2 g2

dgc signal, was expressed as:
= ( 2 − gc )


τc Uc gc

 dt

1 ∞ x(t − τ )
 Z

_ 1
dgm 1 u2 g2 (15) x(t) = dτ = x(t) ∗ (16)
= ( − gm ) π −∞ τ πt

 dt τm P0
1 1 1

 _
where, x(t) is continues signal, x(t) was transformed signal, It

 =
 +
g gc gm could be regarded as a filtered signal with impulse response of
where, gc , gm , g is Cassie model arc conductance, Mayr h(t) = 1/π. When the input signal is us = uα = um sinωt, and
model arc conductance and total arc conductance respec- the output uβ = −um cos(ωt), the orthogonal transformation
tively, P0 is the arc dissipation power constant of Mayr model, was achieved.
τc , τm is Cassie and Mayr model time constant respectively, (3) Inverse Park Transformation. This method utilizes the
Uc is arcing voltage constant of Cassie model, and u is arcing principle that two coordinate transformation algorithms work
voltage. complementary. The structure is shown Fig.6, us is a single-
phase signal, which, as an α phase, enters the Park trans-
IV. ANALYSIS OF ALGORITHM formation unit together with the phase-locked θ and the β
After obtaining voltage and current measurement parame- phase output of the algorithm. The output DQ phase, after
ters, Used the DQ coordinate transformation algorithm to first-order filtering, enters the inverse park transformation
get active component parameters of d-axis. To improve the unit, the αβ phase output, and the uq enters the phase-locked
accuracy of the transformation results, it is necessary to coop- loop, generating phase-locked θ. After dynamic closed-loop
erate with the fundamental wave component extraction and regulation, the output signal tends to be stable.
filtering algorithm. Because the harmonic component will appear in the signal
transformation process, it is necessary to add the filtering
A. DQ COORDINATE TRANSFORMATION algorithm. The expression is 1/(ps + 1), which can be ex-
According to the research idea, completed the active power pressed as p =0.002 during the test.
calculation of the system. Therefore, DQ coordinate trans- (4) SOGI. This algorithm is a second-order generalized
formation method was used. In a single-phase system, the integral algorithm. Fig.7a is a schematic diagram.
Transport-Delay, Hilbert Transformation, SOGI and other The integral realizes Clark transformation of single-phase
algorithms could be used to construct the two orthogonal signal, and each parameter is defined as us is the original
quantities of uα , uβ [18]–[20]. Compared the differences of signal, uα , uβ are the original signal corresponding to the
algorithms, five methods, namely Transport-Delay, Hilbert αβ shaft signal, And uα is synchronized with us , ahead of
Transformation, Inverse Park Transformation, SOGI, and uβ 90◦ , ω0 is the angular frequency of the network side, k is
Improved SOGI were respectively tried to construct orthogo- the bandwidth coefficient, which can be adjusted to calculate
nal vectors. signals of different frequencies. The transfer function of each
(1) Transport-Delay. Assuming the original signal was link is:
us = um sinωt, take it as uα , then uβ = um sin(ωt − π/2) can ω0 s
SOGI (s) = (17)
be obtained through a backward phase shift. This algorithm s + ω02
2
had a simple structure, but no filtering function, and needs to kω0 s
assist other filtering algorithms. Hα (s) = (18)
s + kω0 s + ω02
2
(2) Hilbert Transformation. The frequency amplitude char-
acteristic of Hilbert Transformation was one, which could kω02
Hβ (s) = (19)
realize the isamplitude orthogonal transformation of the s2 + kω0 s + ω02

VOLUME 7, 2019 180797


FIGURE 8. Flow chart of solving active and reactive components of the
system.
When ω0 is the power frequency angular velocity,
the dynamic characteristics of the algorithm can be realized
by adjusting the coefficient k. Assuming us = um sinωt, then
in the time domain:
uα = Um Hα (t)
Um
q
= Um sin ωt − p sin(ω 1−(k/2)2 t)e−kωt/2
1−(k/2)2
(20)
uβ = Um Hβ (t)
Um
= −Um cos ωt − p
1−(k/2)2
q algorithm.
× cos(ω 1−(k/2)2 t −ϕ)e−kωt/2 (21)
k/2
where, ϕ = arctan( √ ). The value of bandwidth
1−(k/2)2
coefficient k affects the response speed of the system, and
the empirical value of 1.5 could be taken.
(5) Improved SOGI. After the difference between the origi-
nal signal and the output α phase of SOGI algorithm, the har-
monic component in the original signal is obtained, which
is removed by filtering to get the dc offset, and then it is
subtracted from the β phase parameters to realize the filtering
effect on the β phase parameters. The structure is shown
in Fig.7b. The filter design can refer to the filter parameters
in the Inverse Park Transform algorithm.
The αβ parameters obtained by the above method can real-
ize Park transformation in combination with phase-locked θ,
and the phase-locked loop can adopt PI adjustment algorithm,
with the structure shown in the Fig.8. Phase lock angle is
captured by PI adjustment, in the case that axis d is ahead
of the coordinate system of axis q, θ is the phase lock angle,
and when the phase lock is complete, ωt = θ, or uq = 0. uq is
connected to the PI regulator, and the change in uq will lead
to the same change in ω, and then the same change in θ after
integration,
The above five orthogonal converter algorithms were sim-
ulated and analyzed. According to Fig.3a, established OLTC
simulation model and experimental circuit, and obtain the
system current signal and corresponding dc component. The
simulation and measurement data were shown in the Fig.9.
Each switching point can be identified by the algorithm,
which is consistent with Fig. 3b.
All the algorithms could follow the amplitude of the sam-
pling signal with good effect. The d-axis component obtained
by the Transport-Delay and Hilbert Transformation has good
180798
D. Song et al.: Research on the Switching Arc Loss of OLTC
FIGURE 9. Calculation results of system current amplitude by each
algorithm.
FIGURE 10. Calculation results of system voltage and frequency by each
dynamic adjustment performance. Other algorithms show the
obvious transition process for the current mutation informa-
tion at each switching point, and the Inverse Park Transforma-
tion is the most serious one. Only the Transport-Delay could
lock the signal amplitude of each switching interval, and all
other algorithms have overshoot.
The Fig. 10 shows the frequency calculation results of the
simulation and measured of each algorithm, respectively. The
fundamental frequency is 50Hz, and the measured waveform
contains certain harmonic components.
The relative errors between the d-axis component and
the actual amplitude of the original signal obtained by the
Transport-Delay, Hilbert Transformation and the Inverse Park
Transformation were calculated, which are 4.40%, 4.76% and
4.71% respectively. Numerically, the three algorithms had
similar computing errors. In this study, the fundamental com-
ponent was extracted before DQ transformation to eliminate
the pulsation component of d-axis component. Therefore, the
Transport-Delay with relatively good amplitude following
effect was selected for the follow-up study.
B. FUNDAMENTAL WAVE COMPONENT EXTRACTION
When there was (inter) harmonic in the original sam-
pling signal, d and q axis components will contain pul-
sating components, generating deviations in the calculation
and requiring filtering processing, used Fourier algorithm,
Kalman filtering algorithm and adaptive filtering, etc. The
Space lest mean square algorithm was used to extract (inter)
harmonic parameters, and compared with Prony, DFT, win-
dowed DFT and other algorithms, [12]. The waveforms of
arc furnace and frequency converter were analyzed by Prony
VOLUME 7, 2019
D. Song et al.: Research on the Switching Arc Loss of OLTC

TABLE 1. The setting of simulation condition.

FIGURE 11. D axis simulation waveform of voltage and current in the

transformer system.

algorithm, [21]. Wavelet and Prony algorithms were used

to reconstruct the signal and evaluated their advantages and
disadvantages, [22].
Considering the rapidity and universality of the algorithm,
the improved 3-point Fourier algorithm was adopted for FIGURE 12. Simulation waveform of system power difference.
transformer parameter calculation. Based on a discrete signal
sequence, the signal spectrum distribution was constructed as
follows: V. SIMULATION VERIFICATION
1 Three kinds of arc model parameters were set in the sim-
X1 (n) = Xa (n) − (Xa (n − 1) + Xa (n + 1))
2 ulation, which could produce the arcing process. Each arc
sin(γ π)
= AN e(j(ϕ+γ π)) parameter corresponds to three loads, a total of nine condi-
2π(n − k1 − γ )((n − k1 − γ )2 − 1) tions, as shown in Table 1. Take the down switching simula-
(22) tion as an example, and set the step size to 50µs, the transition
where, Xa (n) is a fundamental spectrum sequence, n is the resistance as 10 , uw1 = uw2 = 155 V.
ordinal number of sampling points, N is the sampling points,
ϕ is the signal phase, k1 is the adjacent spectrum line of A. SIMULATION RESULT
fundamental frequency k, k = k1 + γ (γ ∈ (0, 1)). |n − k1 − (1) R load. Taking the arc parameters in working condition
γ ||(n − k1 − γ )2 − 1| in the reconstructed spectral line can two as an example, the active components of the system and
accelerate the frequency attenuation rate. When the amplitude the load voltage and current d-axis are obtained, as shown
of k1 and k1 + 1 is assumed to be the largest, there is: in Fig.12.
  X1 (k1 )  2 − γ
  The interval time information of switch action can be easily
a = 
  = obtained by d-axis component waveform. Moreover, due to
X1 (k1 + 1)  1 + γ (23) the large parameter setting of the arc model, both K1 and
γ = 2 − a

1+a K2 switch action positions have a serious transition process,
which can reflect the arcing process indirectly. Since the
So:
equivalent circuits of t6 and t7 are the same, the switching
 AN sin(γ π)e(j(ϕ+γ π)) 
  

 A=   points of the two intervals cannot be determined, which has
N sin(γ π)


no impact on the research.
 
 ϕ = angle(X1 (k1 )) − γ π (24)
f = (kmax + γ − 1) fs


 P2 , Pm , PRσ , PR1R2 and PdeltRarc obtained by calculation are
N drawn as shown in Fig.12a.
where, f is the sampling frequency, and kmax is the maximum
spectral line number.
C. WAVELET FILTERING
In order to reduce the interference of sampling noise to the
original signal, required noise suppression before and after
DQ transform of the original signal. By comparing the filter-
ing effect of different wavelet basis functions, the commonly
used wavelet threshold denoising method was adopted. Used
Db6 wavelet basis function to decompose the original signal
in ten layers, and adjusted the threshold value according to the
decomposed noise at each layer by using optimized predictive
variables Heursure rules.
According to equations (1) ∼ (11), the waveforms of P1 -
PdeltRarc is the difference between P1 -P2 and Pm + PRσ +
PR1R2 , which is enlarged as shown in Fig.12b. Except for
the K1 and K2 , the other four power differences should be
zero. However, due to the suddenly change of voltage and
current signals at various stages, the DQ transformation algo-
rithm needs to rephase lock the voltage component, and the
transient component caused by the delay generated in this
process produces deviations. Therefore, when the all switches
in the model are arc-free switches, there is still a power spike
deviation at positions 1∼6.
To determine the relationship between peak and arcing
power at the K1 and K2 breaking points concerned. The actual
arcing power Pk1arc and Pk2arc are compared with Fig.12b,

VOLUME 7, 2019 180799


FIGURE 13. Simulation waveform of arcing power and difference at K1 ,
K2 (R).
TABLE 2. Simulation data (R).
FIGURE 14. Simulation waveform of arcing power and difference at K1 ,
K2 (RC).
TABLE 3. Simulation data (RC).
which is expressed by the product of the voltage at the switch
end and the conduction current. See Fig.13.
At the breaking point of K1 , the waveform of PdeltRarc and
Pk1arc rises at the same time, and the process coincides. The
calculated power difference at K2 breaking point is consistent
with the actual arcing power. The values of arcing power and
energy in the two places are shown in Table 2. Arcing energy
is the integral value of arcing power in the arcing interval.
(2) RC load. Based on the d-axis components of each
parameter, the power difference calculated at the position K1 ,
K2 and the actual arcing power curve are obtained, as shown
in Fig.14.
Similar to the results in the upper working condition,
the calculated arcing power at the K1 breaking point is large,
and the calculated value at the K2 breaking point is close to
the actual value. The power data is shown in Table 3.
(3) RL load. The curve of calculated and actual arcing
power of K1 and K2 underRL load is shown in Fig.15.
180800
D. Song et al.: Research on the Switching Arc Loss of OLTC
FIGURE 15. Simulation waveform of arcing power and difference at K1 ,
K2 (RL).
TABLE 4. Simulation data (RL).
FIGURE 16. Comparison waveform of arcing power difference and actual
arcing power (condition 2).
Under this condition, the calculated values of arcing power
at K1 and K2 are in good agreement with the actual values.
The data is shown in Table 4.
B. SIMULATED ANALYSIS
Simulation results of each working condition are drawn in
Fig.16. There is the actual and calculated arcing energy of
the switch at K1 and K2 respectively.
Due to the larger switching current and change gradient of
K1 , the transient transition process of DQ transformation is
more serious than that of K2 , and the calculation deviation
increases. According to the data, the position deviation of
K1 is related to the nature of the load. Under inductive load,
the calculated value of arcing power is close to the actual
value. The reason is that the actual arcing power is large due
to the inductive load, which accounts for the main component
in PdeltRarc .
Fig.17 and 18 respectively show the curves of PdeltRarc ,
Pk1arc and Pk2arc with different loads under the other two arc
parameters.
Three kinds of arc parameters determine the difference of
arcing energy, as the load current is small, the actual arcing
power at K1 and K2 breaking points under RC load is small,
and the actual arcing power under RL load is larger. Under
VOLUME 7, 2019
D. Song et al.: Research on the Switching Arc Loss of OLTC
FIGURE 17. Comparison waveform of arcing power difference and actual
arcing power (condition 1).
FIGURE 18. Comparison waveform of arcing power difference and actual
arcing power (condition 3).
FIGURE 19. Experimental test schematic diagram.
different loads, the power distribution at the K2 breaking
point has a good consistency.
VI. EXPERIMENT VERIFICATION
Based on the simulation, a mechanical OLTC prototype was
set up and the experimental circuit was shown in Fig.3. The
transformer model was TSGC2J, with source input and output
range of 0∼430V, SN = 20kVA, w2 /w1 = 1. The transition
resistance and load parameters were same as the simulation,
and the sampling rate was 20kHz. Due to the limitations of the
conditions, ac contactor (ACC-JKC1R-32A) was used as the
switching device in the experiment. As ACC has decentral-
ized action, and there is a delay of 10∼20ms in the general
on-off process, but this study has not been impacted. Since
this study mainly focuses on the energy loss in the switching
process, the time requirement is not strict. Took the driving
board of the main control unit of TMS320F28335 as the ACC
control circuit, which outputs the action sequence based on
the model-based design method and controls the K1 ∼K5
actions through five relays on the interface board. The control
timing is the same as that in Fig.3b.
The actual platform is based on dSPACE hardware in loop
simulation system, and its structure is shown in Fig.19.
VOLUME 7, 2019
FIGURE 20. D-axis test waveform of voltage and current in the
transformer system.
FIGURE 21. Test waveform of system power difference.
The experimental system includes power and measure-
ment and calculation. The power part completes the switch-
ing action of OLTC to the transformer. The signal is intro-
duced into the ADC interface of DS1103 board through
the sampling device. The equipment used: auxiliary volt-
age probe (PINTECH DP100/100MHz/100:1), current clamp
(Tektronix A622/100kHz/10mV/A) and oscilloscope (Agi-
lent Technologies InfiniiVision DSO7104 A).
A. EXPERIMENTAL RESULT
According to the time sequence, a down switching was com-
pleted with a sampling time of 0.2s. The voltage and cur-
rent signals of the system and load contain certain (inter)
harmonic components. The waveform contains large pulsat-
ing components after the DQ transformation. By using the
above wavelet threshold denoising algorithm, signal process-
ing is carried out for the 4-way d-axis component signals.
Fig.20 shows the RC load waveform. Compared with Fig.8,
the measured data have a larger pulsation component, which
is related to the measured interference and calculation devi-
ation. In the d-axis component signal, the position of each
switch could still be identified, and the simulation model is
reliable.
By calculating the sampling data algorithm, the system
power difference and loss waveform as shown in Fig.21. The
transformer excitation resistance still takes the Rm = 1210.
P1 , P2 , Pm +PRσ +PR1R2 , P1 −P2 and PdeltRarc are identified
in the figure.
With the range of deviation considered, P1 − P2 and Pm +
PRσ + PR1R2 have a good coincidence, and the difference
between them is the arcing loss, which is amplified into
Fig.21b. There is also a large peak value at the closure of
180801

FIGURE 22. Comparision waveform of Pk1arc , Pk2arc , Pk3arc and
PdeltRarc (R).
FIGURE 23. Test waveform of arcing power difference and actual arcing
power.
TABLE 5. Test data (R).
K3 , which is preliminarily considered to be caused by the
closure of K3 . To prove that the power peak at the breaking
points of K1 and K2 is consistent with the arcing power at the
two places of ACC, measured the actual arc parameters of the
switches at two positions, and the branch current and end volt-
age data of the switches are obtained. Meanwhile, the closed
transient process of the end of K3 is measured and compared
in the figure. As shown in Fig.22, the waveform Pk1arc , Pk2arc ,
Pk3arc and the calculated power difference waveform PdeltRarc
of K1 ∼K3 are shown.
It can be seen from the figure that the difference between
the measured arcing power and the calculated power differ-
ence of the energy consumption model in the three places
is partially consistent, but the phase and amplitude between
waveforms are greatly deviated due to sampling and other
factors. The arcing process does exist at the closure of K3 ,
confirming the previous hypothesis. The power data is shown
in Table 5.
B. CONTRASTIVE ANALYSIS
For the convenience of analysis and comparison, the calcula-
tion and actual arcing energy of K1 and K2 under three load
conditions are drawn in Fig.23.
180802
D. Song et al.: Research on the Switching Arc Loss of OLTC
Due to sampling, calculation and other factors, the actual
and calculated values of arcing energy at the K1 and K2 break-
ing points have certain deviations. However, the calculated
results are similar to the actual data distribution, and the cal-
culated results can be distinguished under different conditions
with the largest difference between the data is about ten times
the order of magnitude, and the smallest is about one time.
Through analysis of multiple measured, it is shown that the
data distribution of each switching calculation is relatively
concentrated under the same load, and the change of arcing
power and energy caused by load change or abnormal contact
can be determined through the change range of quantity,
to play the role of monitoring and measurement.
The arc model is a customized parameter, which is not an
accurate model of ACC, so there are differences between the
model and the measured results. Also, different calculation
step sizes also lead to the existence of differences. However,
the author believes that it does not affect the development of
this study. Both the simulation and the measured results can
be analyzed independently without forced comparison. The
actual results show that the difference of arcing power and
energy under different working conditions is distinguishable
and within the measured range, which can be used to iden-
tify switching conditions. Therefore, this research scheme is
meaningful.
VII. CONCLUSION
To realize the measurement of OLTC switching arc param-
eters, based on DQ coordinate transformation, fundamental
wave component extraction and wavelet algorithm, this paper
proposes an OLTC switching arc monitoring model based
on energy conservation theory, and obtains the following
conclusions:
(1) The simulation model was established, and the param-
eters of arc with different load and different arc parameters
at K1 and K2 breaking points were calculated. In different
working conditions, the calculated arcing power changes by
the actual value. Under inductive load, the calculated value
of arcing power was close to the actual value. The maximum
difference of calculated value between different loads was
thirty times of magnitude. Arcing power can be distinguished
and identified.
(2) According to the simulation model, a laboratory OLTC
prototype and dSPACE hardware in loop simulation plat-
form were built to complete the measurement and calculation
of three kinds of load switching. There were differences
between the arcing power calculation and the actual value,
but the variation trend of different working conditions was
consistent and the calculated value of differrent loads varied
by a maximum of ten times, which could be distinguished.
(3) The integration of measured sampling step and filtering
results in the difference between simulation and measured
data. However, independent analysis shows that both of them
met the measurable conclusion of arcing power, which could
reflect OLTC switching contact condition and transformer
VOLUME 7, 2019
D. Song et al.: Research on the Switching Arc Loss of OLTC
system power loss indirectly, and verify the effectiveness of
this method.
REFERENCES
[1] Z. Li, Q. Li, Z. Wu, J. Yu, and R. Zheng, ‘‘A fault diagnosis method for on
load tap changer of aerospace power grid based on the current detection,’’
IEEE Access, vol. 6, pp. 24148–24156, 2018.
[2] Q. Li, T. Zhao, L. Zhang, and J. Lou, ‘‘Mechanical fault diagnostics of
onload tap changer within power transformers based on hidden Markov
model,’’ IEEE Trans. Power Deli., vol. 27, no. 2, pp. 596–601, Apr. 2012.
[3] R. Yang, D. Zhang, Z. Li, K. Yang, S. Mo, and L. Li, ‘‘Mechanical fault
diagnostics of power transformer on-load tap changers using dynamic time
warping,’’ IEEE Trans. Instrum. Meas., vol. 68, no. 9, pp. 3119–3127,
Sep. 2019.
[4] E. F. Simas F, L. A. L. de Almeida, and A. C. de C. Lima, ‘‘Vibration
monitoring of on-load tap changers using a genetic algorithm,’’ in Proc.
IEEE Instrum. Meas. Technol. Conf., vol. 3, May 2005, pp. 2288–2293.
[5] E. Rivas, J. C. Burgos, and J. C. Garcia-Prada, ‘‘Condition assessment of
power OLTC by vibration analysis using wavelet transform,’’ IEEE Trans.
Power Del., vol. 24, no. 2, pp. 687–694, Apr. 2009.
[6] J. Seo, M. Hui, and T. K. Saha, ‘‘On Savitzky–Golay filtering for online
condition monitoring of transformer on-load tap changer,’’ IEEE Trans.
Power Del., vol. 33, no. 4, pp. 1689–1698, Aug. 2018.
[7] T. B. Shanker, H. N. Nagamani, and G. S. Punekar, ‘‘Acoustic emission
signal analysis of on-load tap changer (OLTC),’’ in Proc. CACON, Kolkata,
India, 2013, pp. 347–352.
[8] R. Bhuyan, A. R. Mor, P. Morshuis, and G. C. Montanari, ‘‘Analysis
of the arcing process in on-load tap changers by measuring the acoustic
signature,’’ in Proc. EIC, Philadelphia, PA, USA, 2014, pp. 193–197.
[9] B. Feizifar and O. Usta, ‘‘A new arc-based model and condition monitoring
algorithm for on-load tap changers,’’ Electr. Power Syst. Res., vol. 167,
pp. 58–70, Oct. 2019.
[10] M. Wang, A. J. Vandermaar, and K. D. Srivastava, ‘‘Review of condition
assessment of power transformers in service,’’ IEEE Elect. Insul. Mag.,
vol. 18, no. 6, pp. 12–25, Nov./Dec. 2002.
[11] M. Duval, ‘‘A review of faults detectable by gas in oil analysis in transform-
ers,’’ IEEE Elect. Insul. Mag., vol. 18, no. 3, pp. 8–17, May/Jun. 2002.
[12] X. Hui and Z. Peng, ‘‘Subspace-Least mean square method for accurate
harmonic and interharmonic measurement in power systems,’’ IEEE Trans.
Power. Del., vol. 27, no. 3, pp. 1260–1267, Jul. 2017.
[13] A. Bracale, G. Carpinelli, R. Langella, and A. Testa, ‘‘Accurate methods
for signal processing of distorted waveforms in power systems,’’ EURASIP
J. Adv. Signal Process., vol. 2007, no. 19, pp. 1–14, 2007.
[14] K. Alireza, F. Sreffen, G. Sergey, U. Dirk, and R. F. Merhling, ‘‘Electrical
arc model based on physical parameters and power calculation,’’ IEEE
Trans. Plasma Sci, vol. 43, no. 8, pp. 2721–2729, Aug. 2015.
[15] G. Bizjak, P. Zunko, and D. Povh, ‘‘Circuit breaker model for digital
simulation based on Mayr’s and Cassie’s differential arc equations,’’ IEEE
Trans. Power Del., vol. 10, no. 3, pp. 1310–1315, Jul. 1995.
[16] K. Alireza, F. Steffen, G. Sergey, U. Dirk, M. Ralf, and W. Klaus-Dieter,
‘‘An improve arc model based on the arc diameter,’’ IEEE Trans. Power.
Del., vol. 31, no. 3, pp. 1335–1341, Jun. 2016.
[17] U. Habedank, ‘‘Application of a new arc model for the evaluation of
short-circuit breaking tests,’’ IEEE Trans. Power Del., vol. 8, no. 4,
pp. 1921–1925, Oct. 1993.
[18] V. Kaura and V. Blasko, ‘‘Operation of a phase locked loop system
under distorted utility conditions,’’ IEEE Trans. Ind. Appl., vol. 33, no. 1,
pp. 58–63, Jan./Feb. 1997.
[19] G. Hsieh and J. C. Hung, ‘‘Phase-locked loop techniques—A survey,’’
IEEE Trans. Ind. Electron., vol. 43, no. 6, pp. 609–615, Dec. 1996.
[20] F. Xiao, L. Dong, L. Li, and X. Liao, ‘‘A frequency-fixed SOGI-based PLL
for single-phase grid-connected converters,’’ IEEE Trans. Power. Electron.,
vol. 32, no. 3, pp. 1713–1719, Mar. 2017.
[21] Z. Leonowicz, T. Lobos, and J. Rezmer, ‘‘Advanced spectrum estimation
methods for signal analysis in power electronics,’’ IEEE Trans. Ind. Elec-
tron., vol. 50, no. 3, pp. 514–519, Jun. 2003.
[22] M. Meunier and F. Brouage, ‘‘Fourier transform, wavelets, prony analysis:
Tool for harmonics and quality of power,’’ in Proc. 8th Int. Conf. Harmon.
Qual. Power, Athens, Greece, 1998, pp. 71–76.
VOLUME 7, 2019
DONGDONG SONG received the B.Sc. and
master’s degrees in electric engineering from
Hebei Agriculture University, Baoding, China,
in 2005 and 2008, respectively. He is currently
pursuing the Ph.D. degree with China Agricul-
ture University, Beijing, China. He is a Lecturer
with the Hebei Normal University of Science and
Technology, Qinhuangdao, China. His research
interests include new energy power generation,
intelligent monitoring of electrical equipment, and
digital signal processing.
JIWEI MA received the B.Sc. and master’s degrees
in measuring and testing technologies and instru-
ments from Yanshan University, Qinhuangdao,
China, in 1990 and 1993, respectively. She is
currently an Associate Professor with the Hebei
Normal University of Science and Technology,
Qinhuangdao. Her research interests include mea-
surement and controlling systems.
YUQUAN MA received the B.Sc. degree in
electric engineering from Xi’an Jiaotong Uni-
versity, Xi’an, China, in 1993, and the master’s
degree in electric engineering from the Hebei
University of Technology, Tianjin, China, in 2009.
He is currently an Associate Professor with the
Hebei Normal University of Science and Technol-
ogy, Qinhuangdao, China. His research interests
include electric machinery and intelligent moni-
toring of electrical equipment.
HONGJU LIN received the B.Sc. degree in com-
puter and its application from Jilin University,
Changchun, China, in 2000, the master’s degree in
power electronics and electric drives from Xihua
University, Chengdu, China, in 2005, and the
Ph.D. degree in instruments science and tech-
nology from Yanshan University, Qinhuangdao,
China, in 2018. He is currently an Associate Pro-
fessor with the Hebei Normal University of Sci-
ence and Technology, Qinhuangdao. His research
interests include network control technology and system simulation.
SHENGTAO LIU received the B.Sc. and mas-
ter’s degrees in electric engineering from Harbin
Engineering University, China, in 2001 and 2004,
respectively. He is currently a Lecturer with the
Hebei Normal University of Science and Tech-
nology, Qinhuangdao, China. His research interest
includes control systems engineering.
180803