By Tamer Abdelazim Mellik, While preparing a modern data center for startup, the commission-

Thomas J. Dionise, and ing process involved primary circuit switching that resulted in two voltage trans-
Robert Yanniello former (VT) failures. As a result, we conducted a comprehensive investigation
of the VT failures. As the investigation proceeded, VT ferroresonance on circuit
opening and high-frequency switching transients on closing emerged as possible
root causes of the failures. After incorporating extensive transient simulations
and three rounds of field transient measurements, we designed and implemented
a complete solution that included the sizing of snubbers to overcome excessive
switching transients and the development of a saturable reactor to protect VTs
against the effects of ferroresonance. This article describes the root causes, simu-
lations, field measurements, recommended solutions, and solution implementation

A Case Study of
Voltage Transformer Failures

©istockphoto.com/shironosov

SOLUTION IMPLEMENTATION IN A MODERN DATA CENTER

Digital Object Identifier 10.1109/MIAS.2016.2600682

Date of publication: 6 September 2017

98 IEEE Industry Applications Magazine œ J ANUARY/F EBRUARY 2018 1077-2618/18©2018IEEE

for this event. The correlation between field measure- design is a primary selective system for the unit substa-
ments and simulation results shows the effectiveness of tions [10]–[12]. Figure 4 provides a simplified one-line
modeling the implemented solutions. diagram illustrating this design and shows the line-end
and bus-connected VTs of interest. The UPM devices are
Circuit Switching manually operated, fused, pad-mounted switches. This one
A commonly performed operation during commission- line is used for reference in reporting the results of the
ing or normal facility operation is circuit switching. transient study, ­ferroresonance analysis, and onsite mea-
The switching operation causes transients [1], which surements in this article.
can result in equipment damage and danger to personnel We conducted a comprehensive investigation of the VT
who are near the damaged or affected equipment [2], [3]. failures by modeling the data center power distribution
With the widespread application of vacuum breakers
for transformer switching, this phenomenon is receiving
renewed attention [4], [5]. The severity of the switching
transient voltage (i.e., high magnitude, high frequency, VT CPT Power
and the damage caused by the transient voltage) are deter- Source Breaker Cable Transformer
mined by the following circuit characteristics [6], [7]:
●● short bus or cable distance between the circuit breaker
(a)
and transformer Open
●● basic insulation level of the transformer (cast-coil, dry-
+ + dc + + VT
type, and some oil filled) – – Trapped – – Nonlinear
●● inductance of the load being switched (transformer) Charge Inductance
●● circuit-breaker switching characteristics: nonsimultane- (b)
ous operation of poles, current chop (vacuum or SF6),
and restrike (vacuum). FIGURE 1. A conceptual explanation of VT ferroresonance: (a) a one-
An equally important problem is ferroresonance. line diagram and (b) circuit parameters influencing ferroresonance.
ANSI/IEEE Standard 100-1984 [8] defines ferroresonance VT: voltage transformer; CPT: control power transformer.
as “a phenomenon usually characterized by overvol-
tages and very irregular wave shapes and associated
with the excitation of one or more saturable induc-
tors through a capacitance in series with the inductor.”
The key elements are saturable inductors in series with
capacitance, as shown in Figure 1. In the ferro­resonant
circuit, the capacitance (XC) can be the capacitance of
cable, overhead lines, or stray capacitance of transform-
er windings or bushings. Under normal operation, XC
is smaller than the inductive reactance (XL). However,
if some switching event causes the voltage to increase,
then the transformer core may be pushed into saturation
and the XL lowered. It is possible at some higher voltage
that this lower saturated value of XL may equal XC, form-
ing the series resonant circuit called ferroresonance [9].
The ferroresonance phenomenon, as it relates to a VT, is FIGURE 2. A line-end VT failure.
explained in detail in [3].
On 18 November 2013, this data
center experienced the failure of two
line-end VTs at 24.9 kV in newly
installed switchgear. The damage to
the line-end VT is shown in ­Figures 2
and 3. During the investigation, the
bus VTs for the same gear failed on
28 December 2013. The 24.9-kV elec-
trical distribution system for the data
center is extensive, with some circuits
as long as 800 m, which is partially (a) (b)
illustrated in Figure 4. A significant FIGURE 3. A cross section of the line-end VT failure: (a) the VT high-voltage side and (b) the VT
characteristic of the 24.8-kV system windings.

J ANUARY /F E BRUARY 2018 œ IEEE Industry Applications Magazine 99

 Using the Alternative Transients Program version of
the Electromagnetic Transients Program (EMTP) [13], the
UTS-A UTS-B
24.9 kV 24.9 kV medium-voltage distribution system was modeled for
FB-A FB-B switching transient and ferroresonance analysis. Open-
ing of the vacuum circuit breaker (VCB), stray capacitance
754 m 800 m of the cable, and nonlinear inductance of the VTs being
switched (i.e., VT saturation [14], [15]) were represented
with sufficient detail.
UPM-A UPM-B
PD-A PD-B The study indicated the system was subjected to two
different switching phenomena—high-frequency over-
voltages and VT ferroresonance. Each issue required its
19 m 22 m own unique mitigation technique. The high-frequency
overvoltages were addressed through implementing a
NM VT NM VT
custom resistor–capacitor snubber circuit at each VCB. VT
ferroresonance problems are traditionally addressed by
NM-A NM-B the application of a loading resistor across the secondary
27 kV of the VT to dampen the effects of switching transients.
SWGR Because the system utilized over 100 VTs, resistor losses
Arrester
Surge

R over the life of the facility would have been excessive. As

Snubber

an alternative, the problem was addressed through the

C
application of a custom saturable reactor across the sec-
Bus Bus SUB
CPT VT TRAN
ondary of each VT. These solutions are described in more
detail in the “Solutions” section.
Baseline testing, without remediation, was performed
FIGURE 4. A simplified one-line diagram of a typical cell at the data
center. R: resistor; C: capacitance; SWGR: switchgear; SUB TRAN: on 25 February 2014. These results were compared to
substation transformer; PD: pad-mounted switch. the snubber/saturable reactor performance testing that
was conducted on 3 September 2014 (solution testing).
Both tests were conducted by a third party at the facility.
Recorded Volts/Amperes/Hertz Zoomed Detail: The baseline testing was used to characterize the system
25 February 2014 17:50:36–25 February 2014 17:50:36 as it was found and validate the model and resulting
transient overvoltages (TOVs) at the line-end and bus
20.0
VTs, i.e., no additional transient mitigation was installed.
15.0 The solution testing characterized the system after the
installation of custom saturable reactors on the VT sec-
Primary VT Voltage (kV)

10.0
ondaries and resulting TOVs at the line-end and bus VTs,
5.0 i.e., after mitigation was implemented.

0.0
Existing System Conditions (Baseline)
–5.0 A group of sequence operations was performed to cover
all operational scenarios of concern. These scenarios
–10.0
were opening and closing the UTS feeder breakers,
V Waveform Average AG switching between NM-A and NM-B, and UPM switching
–15.0
V Waveform Average BG
V Waveform Average CG
activities during various levels of power loading. A set of
–20.0
high-end portable power quality meters were connected to
1,000 : 1 voltage dividers to monitor three medium-volt-
0

0
64

66

68

70

72

74
.

age points of interest: NM-A line VTs, NM-B line VTs, and
35

35

35

35

35

35
0:

0:

0:

0:

0:

0:
:5

:5

:5

:5

:5

:5

the common bus VTs. Another set of identical meters was

17

17

17

17

17

17

Hours:Minutes:Seconds:Milliseconds installed on the secondary side of the NM-A line VTs and
the NM-B line VTs. The resultant waveforms were com-
FIGURE 5. The bus VT primary average. AG: phase A to ground; BG: pared for both baseline and solution testing. The wave-
phase B to ground; CG: phase C to ground. forms of interest for selected transients are d
­ iscussed in
the following sections.
system, including the switchgear and VTs online and bus-
side of the switchgear lineups, and ­performing a switching UTS Breaker Opening During Baseline Testing
transient study to quantify the transient ­overvoltage pro- During this scenario, a 5-kV square wave at 5–20 Hz was
duced by the 27-kV breakers at the UTS unit substations. measured during baseline testing. This is very typical

100 IEEE Industry Applications Magazine œ J ANUARY/F EBRUARY 2018

 25.00 Recorded Volts/Amperes/Hertz Zoomed Detail:
25 February 2014 17:30:05–25 February 2014 17:30:05
18.75 100
Primary VT Voltage (kV)

12.50 80

6.25 60

Secondary VT Voltage (V)

0.00 40
20
–6.25
0
–12.50
–20
–18.75
–40
–25.00
–60 V Waveform Average AG
0.00 0.05 0.10 0.15 0.20 0.25 0.30
V Waveform Average BG
Time (s) –80
V Waveform Average CG
(file Case3.pl4; x-var t ) –100
V Waveform Average AN

50

00

50
.0

.1

.1
V Waveform Average BN

05

05

05
0:

0:

0:
:3

:3

:3
V Waveform Average CN

17

17

17
Hours:Minutes:Seconds:Milliseconds
FIGURE 6. The primary line-to-ground voltage after the breaker is
opened, illustrating ferroresonance on phase B (from simulation). FIGURE 7. The bus VT secondary. AG: phase A to ground; BG: phase B
AN: phase A to neutral; BN: phase B to neutral; CN: phase C to neutral. to ground; CG: phase C to ground.

of most breaker opening cases, as seen on the bus VT

primary shown in Figure 5. Similar results occur when 100
any upstream 27-kV breaker is opened. This is the low-
75
frequency form of ferroresonance associated with VT
Secondary VT Voltage (V)

saturation. The simulation illustrated the worst condition 50

occurred by opening the 27-kV breaker FB-B in switchgear 25
UTS-B with a cable length of 800 m. This case, shown in
Figure 6, represents the existing system condition and is 0
the most severe, with the breaker opening time chosen –25
at the phase B maximum voltage. Figures 5 and 6 show
–50
a saturation form of ferroresonance, with 5–10-peak kV
(kVpk) square waves and an estimated frequency of 17 Hz. –75
The magnitude is well below the normal peak value of
–100
20 kVpk. However, it appears the square wave persists. 0.00 0.05 0.10 0.15 0.20 0.25 0.30
Another example of the saturation form of ferro- Time (s)
resonance of 5–10 Hz is shown for the UTS breaker FB-A (file Case3.pl4; x-var t )
opening during baseline testing, as seen on the bus VT V Waveform Average AN
secondary in Figure 7. The simulation illustrates that the V Waveform Average BN
worst condition occurs by opening the 27-kV-breaker V Waveform Average CN
FB-B in switchgear UTS-B with a cable length of 800 m.
This case, shown in Figure 8, represents the existing sys- FIGURE 8. The VT secondary voltage after the breaker opens; the
tem condition and shows the VT secondary voltage, the ferroresonance is demonstrated from simulation. AN: phase A to
more severe case with the breaker opening time chosen neutral; BN: phase B to neutral; CN: phase C to neutral.
at the phase B maximum voltage. Figures 7 and 8 show
a saturation form of ferroresonance, with a 10–30-Vpk a saturation 5-kV square wave followed by a sawtooth
square wave and an estimated frequency of 5–10 Hz. with six peaks. The figure shows a saturation form of
The magnitude is well below the normal peak value of ferroresonance with a 5-kVpk square wave. The sawtooth
100 Vpk. indicates the line-to-ground voltage coupled through the
Another example of the saturation form of ferroreso- voltage dividers. The magnitude is well below the normal
nance, during UTS breaker opening, is shown in Figure  9, peak of 20 kVpk.

J ANUARY /F E BRUARY 2018 œ IEEE Industry Applications Magazine 101

 Recorded Volts/Amperes/Hertz Zoomed Detail: Recorded Volts/Amperes/Hertz Zoomed Detail:
26 February 2014 09:49:28–26 February 2014 09:49:29 27 February 2014 11:26:18–27 February 2014 11:26:18
20,000 450,000
400,000
15,000 350,000
300,000
Primary VT Voltage (kV)

Primary VT Voltage (kV)

10,000 250,000
200,000
5,000 150,000
0 100,000
50,000
–5,000 0
–50,000
–10,000 –100,000
–150,000 V Waveform Average AN
V Waveform Average AN
–15,000 –200,000 V Waveform Average BN
V Waveform Average BN
–250,000 V Waveform Average CN
–20,000 V Waveform Average CN –300,000

0
0

1
5

.6

.6

.6

.7

.7

.7
.4

.5

.5

.6

17

17

17

17

17

17
28

28

28

28

6:

6:

6:

6:

6:

6:
9:

9:

9:

9:

:2

:2

:2

:2

:2

:2
:4

:4

:4

:4

11

11

11

11

11

11
09

09

09

09

Hours:Minutes:Seconds:Milliseconds Hours:Minutes:Seconds:Milliseconds

FIGURE 9. The NM-A VT primary, from transients. AN: phase A to FIGURE 11. The bus VT primary, from baseline testing. AN: phase A to
neutral; BN: phase B to neutral; CN: phase C to neutral. neutral; BN: phase B to neutral; CN: phase C to neutral.

During UPM switch closing, a prestrike was captured

Recorded Volts/Amperes/Hertz Zoomed Detail: as seen on the bus VT primary transients (Figure 11).
27 February 2014 11:39:33–27 February 2014 11:39:33 The prestrike occurred during the breaker closing on all
300,000 three phases; the highest magnitude was 435 kVpk. The
V Waveform Average AN prestrike at phase B had a magnitude of 305 kVpk, and
250,000 V Waveform Average BN
the one at phase C had a magnitude of 190 kVpk.
200,000 V Waveform Average CN
During the UPM switch closing, a prestrike was cap-
Primary VT Voltage (V)

150,000
tured, as seen on the NM-A VT secondary (Figure 12).
100,000 The prestrike occurred during the switch closing on all
50,000 three phases, with the highest magnitude being 500 Vpk
0 at phase A. The prestrike at phase C had a magnitude
–50,000
of 135 Vpk, and the one at phase B had a ­magnitude of
260 Vpk. This is the line-to-ground voltage capacitively
–100,000
coupled from the primary to the ­secondary of the VT.
–150,000 The switch closing case was simulated by closing
–200,000 the UPM switch PD-B upstream of the NM-A VT, with a
cable length of 19 m and a pole disagreement of 0.4 ms.
85

90

95

00

05
.8

.8

.8

.9

.9

Similar to the UPM transient testing, no load was con-

32

32

32

32

32
:

:
39

39

39

39

39

nected at the VT secondary (existing system configura-

:

:
11

11

11

11

11

Hours:Minutes:Seconds:Milliseconds tion). Figure 13 shows the line-to-ground TOV is 22.91 kV,

with a frequency of 33 kHz. A similar waveform was
FIGURE 10. The NM-B VT primary, from UPM transients. AN: phase A to captured during the energization of a feeder with the
neutral; BN: phase B to neutral; CN: phase C to neutral. NM breaker open. This was typical of all transients
observed during the testing, as shown in Figures 14
UPM Switch Closing During Baseline Testing and 15.
During the UPM switch closing, a prestrike was captured
on the NM-B VT primary transients, as shown in Fig- Solutions
ure 10. The prestrike shown during breaker closing on all The data center’s medium-voltage distribution system was
three phases exhibited the highest magnitude, 300 kVpk, at simulated. The following main parameters were modeled
phase C. The prestrike at phase B was a magnitude of for the switching transient and the ferroresonance analysis:
440 kVpk. This is the line-to-ground voltage coupled ●● utility system maximum and minimum fault values

through the voltage dividers. ●● utility transformers

102 IEEE Industry Applications Magazine œ J ANUARY/F EBRUARY 2018

 Recorded Volts/Amperes/Hertz Zoomed Detail:
27 February 2014 11:27:04–27 February 2014 11:27:04 25.00
*103
500
18.75
450
400
12.50

Primary VT Voltage (V)

350
300 6.25
250
200
Volts (V)

0.00
150
100 –6.25
50
0 –12.50
–50
–100 –18.75
–150
–200 –25.00
–250 0 4 8 12 16 *10–3 20
Time (s)
0

0
8

0
.8

.8

.8

.8

.9

(file NM-1-Close-PD-NoSNUB.pl4; x-var t )

3

3
:0

:0

:0

:0

:0
7

7
:2

:2

:2

:2

:2

V Waveform Average AN
11

11

11

11

11

Hours:Minutes:Seconds:Milliseconds V Waveform Average BN

V Waveform Average AN V Waveform Average CN
V Waveform Average BN
V Waveform Average CN
FIGURE 13. The primary line-to-ground voltage after the switch is
FIGURE 12. The UPM VT secondary, from UPM transients. AN: phase closed (from simulation). AN: phase A to neutral; BN: phase B to
A to neutral; BN: phase B to neutral; CN: phase C to neutral. neutral; CN: phase C to neutral.

●● cable circuits
●● equivalent stray capacitance at the switchgear and VTs Recorded Volts/Amperes/Hertz Zoomed Detail:
27 February 2014 11:39:10–27 February 2014 11:39:10
●● VTs (parameters that include saturation characteristics)

●● equivalent stray capacitance across the circuit breaker 200

contacts
150
●● existing surge arrestors

●● VCB switching characteristics, including chop current 100

and pole disagreement.
Volts (V)

50
These are common circuit parameters ­represented in the
0
EMTP model and are usually available from ­manufacturers,
although equivalent stray capacitance across circuit vacuum –50
breaker contacts is difficult to obtain. When the vacuum
–100
interrupter contacts are fully parted inside the vacuum bot-
tle, a capacitance can be measured across the open contacts. –150
Although very small, it will have an effect on the transient –200
response following the opening of the breaker; this was
0

50

55

60

65

70
34

34

.3

.3

.3

.3

.3
0.

0.

10

10

10

10

represented in the simulations.

:1

:1

:1

9:

9:

9:

9:
39

39

39

:3

:3

:3

:3

The simulation results in the “Existing System Condi-

:

:
11

11

11

11

11

11

11

tions (Baseline)” section show that the ferroresonance Hours:Minutes:Seconds:Milliseconds

­phenomena occurred for the VTs on the line side of the
circuit breakers within the worst ferroresonance ­locations.
A proven solution to ferroresonance is to apply damp-
ing resistors to the secondary of the PT. Damping resis-
tors have many advantages, as the resistors are off the
shelf, easily available, and provide reasonable damping
times. However, there are disadvantages to the damp-
ing resistors: Their burden affects VT accuracy, and they
will consume power during the steady state. To counter
these disadvantages, switching circuits can be designed
to insert the resistor only during ferroresonance. For
V Waveform Average AN
V Waveform Average BN
V Waveform Average CN
FIGURE 14. The bus VT secondary, from UPM transients. AN: phase A
to neutral; BN: phase B to neutral; CN: phase C to neutral.
this data center application, due to the associated power
losses and the concern that adding switching circuits may
complicate the design, an alternative solution to damping
resistors was desired.

J ANUARY /F E BRUARY 2018 œ IEEE Industry Applications Magazine 103

 A custom-design saturable reactor was proposed and
Recorded Volts/Amperes/Hertz Zoomed Detail: became the subject of this analysis, which forms the
27 February 2014 11:39:10–27 February 2014 11:39:10 basis of this article. This alternative solution requires
500,000
450,000 extensive simulations to custom design the saturation
400,000 curve for the reactor. It does not affect VT accuracy,
350,000

11:39:10.288
11:39:10.266
Primary VT Voltage (V)

300,000 has small steady-state losses, and no additional switch-

250,000

–2,338 V
200,000 ing circuit is needed, all of which were preferred by the

–170 V
150,000 owner of this data center. The simulations show that,
100,000
50,000 with small-reactor resistance, high-speed ferroresonance
0 damping is achieved much faster than the conventional
–50,000
–100,000 solution of damping resistors.
–150,000
–200,000 The proposed and effective solution for mitigating
–250,000 the ferroresonance is a saturable reactor, as shown in
–300,000
–350,000 Figure 16. After the upstream circuit breaker opens, there
will be a dc trapped voltage on the open line. With the VT
0

0
6

9
.2

.2

.2

.2 online side of the downstream breaker, the trapped voltage

0

0
:1

:1

:1

:1
9

can be large enough to saturate the VT magnetizing imped-

:3

:3

:3

:3
11

11

11

11

Hours:Minutes:Seconds:Milliseconds ance, setting up a resonance condition, which causes the

V Waveform Average AG
current to oscillate as a square wave. The oscillation can
V Waveform Average BG last for an extended period of time, until the eventual fail-
V Waveform Average CG ure of the VT. The failure may not necessarily occur based
upon just one switching event but can occur over time
FIGURE 15. A similar scenario to Figure 14, but with a higher voltage based upon a number of switching events and where the
transient. AG: phase A to ground; BG: phase B to ground; CG: phase circuit breaker opens on the voltage waveform [3].
C to ground. Modeling and site measurements have shown that
VCB-induced voltage transients can cause damage to the
bus and line VTs because of the short cable lengths. The
results of the study indicated that one snubber on the line
Rp Xp VT Rs Xs
side of each VT would dramatically reduce the magnitude
of the voltage transient as well as its dc offset. The snub-
Xsr ber was shown to reduce the frequency of the transient
Xm Rm oscillation below the recommended 1,000 Hz. The snub-
bers were sized per IEEE Standard C57.142-2010 [2], [15]–
Rsr [17], as shown in Figure 17.
For mitigating the ferroresonance, a custom saturable
reactor was created based on the VT saturation curve, as
shown in Figure 18. The saturable reactor should be
FIGURE 16. The proposed saturable reactor for mitigating VT designed to saturate above the VT it is protecting; other-
ferroresonance. wise it creates a high impedance load on the VT. When

1,000
VT Saturable Reactor
PT
Current (mA)

100
100 Ω
Noninductive
22-kV Peak SA
17.5 kJ 10
0.25 µF at 27 kV

1
0.01 0.1 1 10
Voltage (V)

FIGURE 17. The snubber solution for mitigating transient overvoltage FIGURE 18. A custom saturable reactor designed to mitigate
at the VT. SA: surge arrestor. ferroresonance.

104 IEEE Industry Applications Magazine œ J ANUARY/F EBRUARY 2018

the VT sees an overvoltage that goes beyond its thermal
rating, which was around 125% of the rated voltage,
then the reactor kicks in—but not before. During site test-
ing, the initial reactor design was saturating before the VT.
To shift the curve to the right of the VT’s saturation curve,
an air gap of about 0.020 cm/in was added. This tech-
nique is incorporated in many CTs used for transient per-
formance to reduce remnant (residual magnetism) effects.
The air gap shifted the curve about a decade to the
right, which is what was needed from the basic design. It
also changed the slope of the linear portion of the curve,
which has no bearing on its performance. By adjusting
the core size and area as well as the number of turns, no
UTS Breaker Opening During Solution Testing
As shown in Figure 19, the saturable reactor VT load quickly
dampens the ferroresonance to an acceptable level. The sim-
ulation results in Figure 20 show similarities to the measured
results in Figure 21, and although the simulation case was
performed on the most severe case (opening breaker time at
the phase B maximum voltage), both cases show damping
of ferroresonance in lower than 250 ms with the saturable
25.00
*103
18.75

Primary VT Voltage (V)

dc component can pass through the VT secondary circuit. 12.50
The coil resistance was increased by reducing the second- 6.25
ary winding copper area to yield 1 X. By adding a bifilar
0.00
winding, an added 5 X was achieved, but this had zero
net effect on flux, inductance, and voltage in the reactor –6.25
winding. This winding was in series with the main coil to –12.50
give a net resistance of 6 X. –18.75
–25.00
Solution Testing 0.00 0.05 0.10 0.15 0.20 0.25
This section shows selected waveforms that prove the Time (s)
effectiveness of the solution. In the “Existing System (file Case_H_DesignE-Final1.pl4; x-var t )
Conditions (Baseline)” section, tests were performed that
V Waveform Average AN
show the two problems in the present system configura-
V Waveform Average BN
tion. The same testing scenarios will be repeated but after
V Waveform Average CN
installing/simulating the proposed mitigation; i.e., the VTs
are loaded with the saturable reactors and custom snub- FIGURE 20. The NM-A incoming feeder UTS opening (from simulation).
bers are applied. AN: phase A to neutral; BN: phase B to neutral; CN: phase C to neutral.

Recorded Volts/Amperes/Hertz Zoomed Detail:

Recorded Volts/Amperes/Hertz Zoomed Detail: 3 September 2014 12:57:56–3 September 2014 12:57:56
3 September 2014 13:05:55–3 September 2014 13:05:55 20.0
20.0 V Waveform Average AG 15.0
Primary VT Voltage (kV)

V Waveform Average BG
15.0 V Waveform Average CG 10.0
Primary VT Voltage (kV)

10.0 5.0

5.0 0.0
–5.0
0.0
–10.0
–5.0
–15.0
–10.0
–20.0
–15.0
0

0
32

34

36

38

–20.0
.

.
56

56

56

56
7:

7:

7:

7:
:5

:5

:5

:5
50

00

12

12

12

12
:5

:6

Hours:Minutes:Seconds:Milliseconds
54

54
5:

5:
:0

:0

V Waveform Average AG
13

13

Hours:Minutes:Seconds:Milliseconds V Waveform Average BG

V Waveform Average CG

FIGURE 19. The NM-A incoming feeder UTS opening. AG: phase A to FIGURE 21. The NM-B primary switchgear voltage. AG: phase A to
ground; BG: phase B to ground; CG: phase C to ground. ground; BG: phase B to ground; CG: phase C to ground.

J ANUARY /F E BRUARY 2018 œ IEEE Industry Applications Magazine 105

 reactor installed. This shows the effectiveness of the satu-
rable reactor. Similar waveforms are shown in Figures 21 and
22 for the same switching event of opening the upstream
UTS breaker. As shown in Figures 20–22, the damping time
is lower than 100 ms, and the voltage magnitude is in the
normal peak value of 20 kV.
In Figures 23 and 24, after opening breaker FB-A,
secondary VT waveforms are shown. Figure 23 shows
the NM-A secondary voltage, and Figure 24 shows the
simulated VT secondary voltage. The simulated case
shows the worst case, but both show damping in fewer
than 250 ms.

100
Recorded Volts/Amperes/Hertz Zoomed Detail:
3 September 2014 12:58:09– 3 September 2014 12:58:09 75

Secondary VT Voltage (V)

20.0 V Waveform Average AG 50
V Waveform Average BG
15.0 25
V Waveform Average CG
Primary VT Voltage (kV)

10.0 0
5.0
–25
0.0
–50
–5.0
–75
–10.0
–15.0 –100
0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35
–20.0 Time (s)
(file Cell3_inductor_H_Design.pl4; x-var t )
00

50
:1

:1
09

09

V Waveform Average AN
8:

8:
:5

:5

V Waveform Average BN
12

12

Hours:Minutes:Seconds:Milliseconds V Waveform Average CN

FIGURE 24. The NM-A secondary voltage after UTS opening (from
FIGURE 22. The bus primary voltage after UTS opening. AG: phase A simulation). AN: phase A to neutral; BN: phase B to neutral; CN:
to ground; BG: phase B to ground; CG: phase C to ground. phase C to neutral.

Recorded Volts/Amperes/Hertz Zoomed Detail:

Recorded Volts/Amperes/Hertz Zoomed Detail: 3 September 2014 13:06:29–3 September 2014 13:06:29
3 September 2014 14:05:55–3 September 2014 14:05:55 80
100 60
80
40
Secondary VT Voltage (V)

60
Primary VT Voltage (kV)

20
40
0
20
0 –20

–20 –40
–40 –60
–60 V Waveform Average AG –80
–80 V Waveform Average BG V Waveform Average AG
–100
V Waveform Average CG V Waveform Average BG
–100 –120 V Waveform Average CG
–140
00

50

00
.3

.3

.4

0
55

55

55

35

36

37
5:

5:

5:

9.

9.

9.
:0

:0

:0

:2

:2

:2
14

14

14

06

06

06
:

Hours:Minutes:Seconds:Milliseconds
13

13

13

Hours:Minutes:Seconds:Milliseconds

FIGURE 23. The NM-A secondary voltage after UTS opening. AG: FIGURE 25. The NM-A primary VT voltage. AG: phase A to ground;
phase A to ground; BG: phase B to ground; CG: phase C to ground. BG: phase B to ground; CG: phase C to ground.

106 IEEE Industry Applications Magazine œ J ANUARY/F EBRUARY 2018

UPM Switch Closing During Solution Testing seen on NM-A and NM-B VT primaries and in Figures 25
This section discusses similar testing conditions as in the and 26; these figures also show that the TOV was damped
previous section, except custom snubbers were installed quickly after closing the upstream switch and that the volt-
at the primary of the VTs. During closing, as performed in age magnitude is in the normal peak value of 140 kV.
the “Existing System Conditions (Baseline)” section (closing This case was again simulated by closing the upstream
the UPM switch PD-A upstream), prestrike was captured as UPM switch PD-A. The TOV magnitude was 24.67 kV, with

Recorded Volts/Amperes/Hertz Zoomed Detail: 30

3 September 2014 13:40:56–3 September 2014 13:40:56
60 *103
19
40

Primary VT Voltage (V)

20
8
Primary VT Voltage (kV)

0
–20
–3
–40
–60 –14
–80
–100 –25
V Waveform Average AG 0 5 1015 20 25 30

*10–3
–120 V Waveform Average BG Time (s)
V Waveform Average CG (file NM-1-Close-PD.pl4; x-var t )
–140

V Waveform Average AN
90

95

00

05
.3

.3

.4

.4

V Waveform Average BN
56

56

56

56
0:

0:

0:

0:

V Waveform Average CN
:4

:4

:4

:4
13

13

13

13

Hours:Minutes:Seconds:Milliseconds

FIGURE 28. The NM-A primary VT voltage after closing the UPM
FIGURE 26. The NM-B primary voltage. AG: phase A to ground; BG: switch (from simulation). AN: phase A to neutral; BN: phase B to
phase B to ground; CG: phase C to ground. neutral; CN: phase C to neutral.

25.00
*103 Recorded Volts/Amperes/Hertz Zoomed Detail:
18.75 3 September 2014 14:37:35–3 September 2014 14:37:35
100
Primary VT Voltage (V)

12.50
6.25 50
Secondary VT Voltage (V)

0.00
0
–6.25
–12.50 –50

–18.75
–100
–25.00
0 5 10
15 20 25 30
*10–3

–150 V Waveform
Time (s)
(file NM-1-Close-PD.pl4; x-var t ) Average AB
–200
V Waveform Average AN
20

30

40

50
.6

.6

.6

.6

V Waveform Average BN
34

34

34

34
7:

7:

7:

7:

V Waveform Average CN
:3

:3

:3

:3
14

14

14

14

Hours:Minutes:Seconds:Milliseconds
FIGURE 27. The NM-A primary VT voltage after closing the UPM
switch using a snubber (from simulation). AN: phase A to neutral;
BN: phase B to neutral; CN: phase C to neutral. FIGURE 29. The bus VT secondary voltage after UPM switching.

J ANUARY /F E BRUARY 2018 œ IEEE Industry Applications Magazine 107

 Recorded Volts/Amperes/Hertz Zoomed Detail:
3 September 2014 14:42:15–3 September 2014 14:42:15 150
200
V Waveform Average AG

Secondary VT Voltage (V)

V Waveform Average BG 100
150 V Waveform Average CG
Secondary VT Voltage (V)

50
100

50 0

0
–50
–50
–100
0 5 10 15 20 25 30

*10–3
–100
Time (s)
–150 (file NM-1-Close-PD.pl4; x-var t )

V Waveform Average AN
90

95

00

V Waveform Average BN
.2

.2

.3
15

15

15

V Waveform Average CN
2:

2:

2:
:4

:4

:4
14

14

14

Hours:Minutes:Seconds:Milliseconds
FIGURE 32. The NM-A secondary VT voltage after closing the UPM
FIGURE 30. The secondary bus voltage after closing the NM-B. AG: switch. AN: phase A to neutral; BN: phase B to neutral; CN: phase C
phase A to ground; BG: phase B to ground; CG: phase C to ground. to neutral.

phase A closes first, followed by a very small delay when

150 phase B closes, and then a much longer delay until phase
C closes. This pole separation or pole disagreement can
Secondary VT Voltage (V)

100 initiate transients in the other phases for which the pole
has not yet opened. In both phases A and B, there is a sig-
50 nificant TOV followed by a damped oscillation. The peak
as well as the rate of change of voltage can damage the VT
0 primary winding insulation.
This case was simulated by closing the UPM switch
–50 PD-B upstream. The TOV magnitude is 120.01 V with a
frequency of 3,300 Hz, as shown in Figure 31. Note that
–100 the case was initially simulated with a 30-X resistor and
*10–3

0 5
15 20 10 25 30
Time (s) a 0.125-µF capacitor, but repeated using a fine-tuned
(file NM-1-Close-PD.pl4; x-var t ) snubber (100 X and 0.25 µF). This analysis shows a
reduced TOV of 125.52 V, with a frequency of 2,500 Hz,
V Waveform Average AN
as shown in Figure 32. There is a reduction in the TOV
V Waveform Average BN
by using the fine-tuned snubber, but it still damped to
V Waveform Average CN
an acceptable level.

FIGURE 31. The NM-A secondary VT voltage after closing the UPM Conclusions and Observations
switch using an off-shelf snubber (from simulation). AN: phase A to
The “Existing System Conditions (Baseline)” and “Solution
neutral; BN: phase B to neutral; CN: phase C to neutral.
Testing” sections have shown the resultant waveforms dur-
ing the UTS breaker opening and closing with and without
a frequency of 3,300 Hz, as shown in Figure 27. Note that the saturable reactor as a solution to the ferroresonance
the case was initially simulated in EMTP with a 30-X resis- problem and with and without custom snubbers as a
tor and 0.125 µF capacitor. The case was repeated using ­solution to the switching transient problem. The effective-
a fine-tuned snubber (100 X and 0.25 µF). This showed a ness of the implemented solution has been ­verified through
reduced TOV of 26.14 kV, with a frequency of 2,500 Hz, as field testing and simulation. During baseline t­esting, the
shown in Figure 28. measured waves showed significant 5–20-Hz ferroresonance
Figures 29 and 30 show the VT secondary voltage when at the VT on opening and even worse prestrike/reignition
closing the UPM PD-B switch. In Figure 30, notice that transients on closing. After the installation of saturable reactors,

108 IEEE Industry Applications Magazine œ J ANUARY/F EBRUARY 2018

the system settles quickly for the same tests, i.e., in fewer
than 100 ms. It has been clearly shown that the saturable
reactors provide the necessary damping to mitigate the fer-
roresonance problem at this data center.
On breaker closing during both baseline and solution
testing, the transients on the primary of the VTs showed
significant magnitude, on the order of 500 kV, with pole
disagreement of about 1 ms. The corresponding transient
on the VT secondary was on the order of 500 V. The tran-
sients are capacitively coupled from the primary to second-
ary of the VT, not through the turns ratio, so the voltage
divider scaling/calibration is not a concern.
For the breaker closing during both baseline and solu-
tion testing, there were many prestrike events. Similar
prestrike events have been observed for a VCB closing
into an arc furnace transformer. In that case, the prestrike
was aggravated when the operators opened and closed
the VCB about five times within as many minutes. For the
testing at this data center, the iterations of breaker close/
open of up to six times may not have allowed the vacuum
interrupter to fully recover, resulting in prestrike on the
next close operation.
On breaker opening during baseline testing, there is
evidence of a saturation form of ferroresonance. After
the breaker opens, there is a pronounced square wave of
5 kV, with a frequency of 5–20 Hz as seen at the VT pri-
mary. This response is very typical of the breaker opening
cases. The square wave persists for a few cycles at 5–10 Hz
before damping out. During solution testing, the system
shows a fast damping of oscillations within 100 ms. How-
ever, in at least one case during baseline testing, the
saturation form of ferroresonance square wave had a
much higher magnitude of 75 V at 4 Hz, as seen at the
VT secondary. This higher magnitude was preceded by
reignition during breaker opening on all three phases. The
reignition may have trapped an additional charge, increas-
ing the voltage on the secondary of the VT.
During the opening of the NM-B breaker (baseline
testing), transients were noted at the NM-A VT on the
order of 475 kVpk. This is significant because the NM-A
breaker was already open by this point. The transient was
coupled from NM-B up through the 24.5-kV UTS feeder
and back down through the adjacent UTS feeder to NM-A.
The magnitude of the TOVs during baseline and solu-
tion testing on the order of 500 kVpk and rate of rise
reported on the primary of the VTs would be expect-
ed to cause the surge arrester to operate, but it did
not. It is possible that there was a scaling/calibration
issue between the voltage dividers and the power qual-
ity meter. The voltage dividers have a ratio of 1,000 : 1,
which gives 20-Vrms line-to-ground input to the power
quality meter. The power quality meter has a minimum
input voltage of 60 V. The transient response of the
power quality meter in the 20-V range is not known. Nor-
mally the power quality meter is applied at 120/208 V or
277/480 V, which results in input voltages well above the
minimum of 60 V.
Author Information
Tamer Abdelazim Mellik (abdelazim@ieee.org) is with
the Eaton Corporation Power Systems Engineering, Calgary,
Canada. Thomas J. Dionise is with Eaton Corporation
Power Systems Engineering, Warrendale, Pennsylvania.
Robert Yanniello is with Eaton Corporation, Arden, North
Carolina. Mellik is a Senior Member of the IEEE. This article
first appeared as “A Case Study of Voltage Transformer Fail-
ures in a Modern Data Center: Analysis, Mitigation, and
Solution Implementation” at the 2016 IEEE IAS 52nd Indus-
trial and Commercial Power Systems Technical Conference.
This article was reviewed by the IAS Power ­Systems Pro-
tection Committee.
References
[1] A Guide to Describe the Occurrence and Mitigation of Switching Tran-
sients Induced By Transformer And Switching Device Interaction, ANSI/
IEEE Standard C57.142-2010, Apr. 2011.
[2] D. Shipp, T. Dionise, V. Lorch, and B. MacFarlane, “Transformer failure
due to circuit breaker induced switching transients,” IEEE Trans. Ind.
Appl., vol. 47, no. 2, pp. 707–718, Apr./May 2011.
[3] D. Shipp, T. Dionise, V. Lorch and D. McDermit, “Medium voltage switch-
ing transient induced potential transformer failures; prediction, measure-
ment and practical solutions,” IEEE Trans. Ind. Appl., pp. 1726–1737,
July/Aug. 2013.
[4] D. Shipp and R. Hoerauf, “Characteristics and applications of various
arc interrupting methods,” IEEE Trans. Ind. Appl., vol. 27, pp. 849–861,
Sept./Oct 1991.
[5] IEEE Standard for AC High-Voltage Generator Circuit Breakers on
a Symmetrical Current Basis, ANSI/IEEE Standard C37.013-1997, Dec.
1997.
[6] IEEE Guide for the Application of Transient Recovery Voltage for AC
High-Voltage Circuit Breakers, ANSI/IEEE Standard C37.011-2011, Nov.
2011.
[7] D. Durocher, “Considerations in unit substation design to optimize
reliability and electrical workplace safety,” in Proc. Industrial and
­Commercial Power Systems Technical Conference (I&CPS), Tallahassee,
FL, May 2010, pp. 1–11.
[8] IEEE Standard Dictionary of Electrical and Electronics Terms, ANSI/
IEEE Standard 100-1972.
[9] R. H. Hopkinson, “Ferroresonant overvoltages due to open conduc-
tors,” General Electric Publication, 1967, pp. 3–6.
[10] Westinghouse Electric Corp., Distribution Transformer Division, “Fer-
roresonance,” in Westinghouse Distribution Transformer Guide. Athens,
GA: Westinghouse Electric Corporation, Apr. 1986, pp. 36–40.
[11] IEEE Guide for Application of Transformers, ANSI/IEEE Standard
C57.105-1978, 2008.
[12] Ontario Hydro, Distribution Technical Guide. Ontario, Canada:
Ontario Hydro, May 1999, pp. 72.1-1–72.1-10.
[13] Alternative Transients Program. (2015). Welcome to EEUG, users group
of ATP-EMTP. EEUG. [Online]. Available: http://eeug-test.hostingkunde.de/
[14] A. Greenwood, Electrical Transients in Power Systems. New York:
Wiley, 1971, pp. 91–93.
[15] L. Kojovic and A. Bonner, “Ferroresonance—Culprit and scapegoat,”
Cooper Power Systems, The Line, Dec. 1998.
[16] IEEE Guide to Describe the Occurrence and Mitigation of Switching
Transients Induced by Transformers, Switching Device, and System Inter-
action, ANSI/IEEE Standard C57.142-2010, Apr. 2011.
[17] C. S. Mardegan, D. D. Shipp, L. A. R. Melo, M. R. Santana, “The
experience acquired sizing snubbers to mitigate switching transients in
industrial power systems,” in Proc. IEEE IAS I&CPS Conference, May 2015,
pp. 1–10.

J ANUARY /F E BRUARY 2018 œ IEEE Industry Applications Magazine 109