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/TPWRD.2019.2922505, IEEE
Transactions on Power Delivery
Design and Testing of an Arc Resistant Power
Transformer Tank
Samuel Brodeur and Jean-Bernard Dastous, Senior Member, IEEE
Abstract—This paper presents the design and testing of a full
scale 330-kV, 210-MVA transformer tank to withstand the pres-
sure effects of a 20 MJ, three-cycle low-impedance fault in oil,
and fail at its cover in a safe and controlled manner above that
amount of energy. The transformer tank, equipped with a
mockup active part, was designed for arc resistance using nonlin-
ear static finite-element analysis and a specific rupture criterion.
It was then tested using an experimental methodology based on
injecting pressurized air into the water-filled tank. To adequately
reproduce the effects of an arc in oil, the test parameters were
determined using explicit dynamic simulations. The transformer
tank performed as predicted by the finite-element and explicit
dynamic simulations. The energy levels used in the tests here are,
to our knowledge, the largest ever reported. This study demon-
strates that it is feasible to design a power transformer to safely
withstand high-energy low-impedance faults of short duration.
Index Terms—explosions, explosion protection, fault protec-
tion, finite-element analysis, fires, internal arc, power transform-
er, pressure effects, risk mitigation, tank rupture.
I. INTRODUCTION
D ue to their explosive nature, low-impedance arcing faults
in power transformers and reactors are an important area
of concern for utilities and equipment manufacturers. Possible
consequences of such faults include oil spills and environmen-
tal pollution, projection of parts and flammable material, dam-
age to adjacent equipment, fires and safety threats.
A summary of a worldwide survey concluded that the
probability of a transformer fire is in the order of 0.1% per
service year [1]. The accumulated probability of a fire result-
ing from an internal arc is therefore 4% over a typical service
life of 40 years. In a survey of transformer and reactor failures
on its 120-kV to 735-kV system, from 1965 to 1986, Hydro-
Québec observed that the rate of explosion resulting in fire
was on average 0.14% per year, and that the rate of fire on its
735-kV system (0.22%) was about 10 times higher than on its
120-kV system (0.02%) [2]. In view of the preceding, Hydro-
Québec has implemented arc resistance requirements in order
to obtain equipment tanks that can withstand specified levels
of arc energy according to their voltage classes, while failing
safely beyond, by requiring that failure happen at the cover
level in order to minimize oil spills, propagation of fires, dam-
age to adjacent equipment, and risk to workers in the sur-
S. Brodeur is with ABB, Varennes, QC, J3X 1S1, Canada (e-mail: samu-
el.s.brodeur@ca.abb.com).
J.-B. Dastous is with the Institut de recherche d’Hydro-Québec (IREQ),
Varennes, QC, J3X 1S1, Canada (e-mail: dastous.jean-bernard@ireq.ca).
0885-8977 (c) 2019 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.
1
roundings [2], [3]. Accordingly, ABB has been considerably
improving its arc resistant tank design to meet these require-
ments.
This paper presents the design and testing of the transformer
tank of a full scale three-phase power transformer (330 kV,
210 MVA) to ensure that it meets Hydro-Québec arc re-
sistance requirements. This transformer was designed for arc
resistance using nonlinear static finite-element analysis and a
specific rupture criterion. It was then tested using an experi-
mental methodology based on injecting pressurized air. The
test parameters were determined using explicit dynamic simu-
lations, to adequately reproduce the effects of an arc in oil.
II. TANK DESIGN
A. Design requirements
The main effect of a low-impedance internal arc is a rapid
pressure rise within the tank, of the order of 100 ms for a
50 ms (3 cycle) fault duration [3], [4]. In its requirements,
Hydro-Québec specifies that this pressure rise is uniform and
depends on the volume of gas generated by the arc energy, and
the tank expansion characteristics. A simplified expression
relating these parameters was first proposed by [5]; it is based
on the conservative assumption of an isothermal expansion of
the gas bubble generated in the oil by the arc. Equation (1) and
Fig. 1 present a modified version [3] which includes different
arc parameters and a more elaborate dynamic amplification
factor than the single value suggested in [6].
é k×E ù
Pd = F ê100 × 0.25 + - 50ú + Ph (1)
ë 100 × c û
where Pd is the design pressure (kPa), F is the dynamic ampli-
fication factor (Fig. 1), k is the arc energy conversion factor
(=5.8 x 10-4 m3/kJ), E is the arc energy to withstand (kJ), c is
the tank expansion coefficient (m3/kPa), and Ph is the hydro-
static pressure (kPa) evaluated at mid height of the tank from
mid height of the conservator.
Using Eq. (1), the design pressure is determined using an
incremental procedure as detailed in [7], according to the tank
expansion coefficient c which varies nonlinearly with the tank
plastic deformation. The arc energy used for the tank under
consideration is 20 MJ, corresponding to the value specified
for the 330-kV voltage class in Hydro-Québec requirements,
for a 50 ms fault duration [3].
It is important to underline that (1) is a design formula
aimed at covering tanks of different flexibilities. Therefore, as
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/TPWRD.2019.2922505, IEEE
Transactions on Power Delivery
2

we shall see, it does not predict accurately the peak pressure Fig. 4.
for the fault and tank considered, but rather provides an upper
bound value for design purposes.

Fig. 1. Dynamic amplification factor.

B. Tank design governing principles

The tank under consideration is a 330-kV, three-phase
power transformer rated at 210 MVA (Fig. 2); the tank inner
dimensions are 4.9 m length x 2.4 m width x 4.2 m height. Its Fig. 2. Tank model.
thickness is for the bottom 32 mm, walls 10 mm and cover 25 D. Modeling parameters
mm. This tank was designed using static nonlinear finite ele-
The global model mesh dimension required for conver-
ment analysis. Several design principles were used to prevent
gence of the tank displacements, and therefore to reproduce
failure at locations most likely to rupture, such as weld joints
adequately the change in tank volume under applied pressure
at wall corners, bottom, cover, and ends of stiffeners [8]:
has been identified as 40 mm of edge. The large majority of
1) Make tank corners with a single bent plate instead of a
the mesh is made of 3D eight-node solid shell elements close
welded butt joint between two walls at 90o [8], [9];
to this size, for a total of over 200,000 elements.
2) End tank stiffeners before tank wall joints between tank
cover and bottom [10];
3) Use numerous smaller stiffeners shaped and positioned
adequately [9], [10], [11];
4) Increase tank flexibility while providing sufficient
strength [9], [12].
This last point is of significant importance as it can be in-
ferred from (1) that the tank design pressure will decrease with
an increase of flexibility. Our approach was therefore largely
based on this principle, and several design iterations were
performed using corresponding finite element models, until a
satisfactory design was obtained.
C. Analysis methodology
Due to large plastic deformations at some locations, the
tank needed to be analyzed using static nonlinear finite ele-
ment software that can take material and geometrical nonline-
arities into account.
The analysis is done in two steps. First the design pressure
is determined using a global model of the tank (emptied of its
content), by applying a uniform pressure over all its inner
Fig. 3. Global finite element model.
surfaces; this pressure is increased linearly until the design The submodel meshes are made of solid elements of ap-
pressure is determined using the procedure described in [7]. proximately 1 mm of edge in and around the weld area. Such a
The global finite element model used is shown in Fig. 3. small dimension is required to capture adequately the large
Next, submodeling is used to analyze specific locations deformation gradients that can occur in small parts of the tank
where the deformations are the highest under the design pres- that are subjected to highly localized deformations [7], [15],
sure, in order to verify that the tank can withstand it. With the especially in the welds. Such a mesh size can result in large
same submodels, the pressure is then increased until the rup- submodels, of the order of 1 million elements.
ture pressure is determined. The submodeling procedure is Two materials are present in the tank: CSA G40.21 50W
detailed in [7] and an example of a submodel used is shown in steel is used everywhere, except for the welds where E70C

0885-8977 (c) 2019 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/TPWRD.2019.2922505, IEEE
Transactions on Power Delivery
3

wire material is used. The corresponding material properties E. Results

used in modeling were determined from average values based Analysis of the global model yielded the parameters and re-
on mill tests of each material. The materials were modeled sults summarized in Table II.
using true stress-true strain curves in the elastic and plastic TABLE II
deformation domains, up to the corresponding ultimate stress. Determination of the design pressure
The main properties used are given in Table I. Parameter/result Symbol Value
Arc energy to contain for 330 kV (kJ) [3] E 20,000
Volume of oil inside the tank (m3) V 27.5
Hydrostatic pressure at tank mid height (kPa) PH 20.8
Volume change at design pressure (m3) ∆V 4.02
Pressure change at design pressure (kPa) ∆P 535
Flexibility at design pressure (m3/kPa) c = ∆V/∆P 0.0075
Dynamic amplification factor as a function of
F 1.545
c/V (Fig. 1)
Design pressure (kPa) Pd 556

Analysis of the submodels identified that the highest de-

formations were in the weld between the cover and the wall
flange, at mid length on the longest side of the transformer
(Fig. 4). This area withstands the design pressure correspond-
ing to a fault of 20 MJ, and a further increase of the pressure
yielded a rupture pressure of 946 kPa retrieve from the simula-
tion result of Fig. 5, with the opening of the cover at this point,
as required in our fail-safe approach. At this failure state, it is
important to have good safety margin on the tank walls such
Fig. 4. Submodel of the weld at tank cover. as observed with the strain results of Fig. 6.
TABLE I
Material properties
Parameter Symbol 50W E70C
7,860 7,860
Density r
kg/m3 kg/m3
Young’s modulus E 203 GPa 203 GPa
Shear modulus G 78 GPa 78 GPa
Yield stress Sy 409 MPa 666 MPa
True ultimate plastic stress Su 608 MPa 789 MPa
True ultimate plastic strain 0.14 0.08

A strain-based rupture criterion is used to predict failure

when the equivalent plastic strain exceeds the ultimate
plastic strain throughout the thickness of the part under
consideration, rather than in a single finite element. An exam-
ple is shown in Fig. 5 where all areas in magenta represent
elements where the equivalent plastic strain exceeds the ulti-
mate plastic strain; it can be observed that the magenta area is
continuous throughout the thickness of the weld. This criterion
has been validated by simplified experiments on pressurized
tank walls, as detailed in [15]. The criterion used is therefore:
e eP ³ e uP throughout the thickness
Fig. 5. Example of the attainment of the rupture criterion in a weld.
Fig. 6. Tank strain comparison at design and rupture pressures.
III. EXPERIMENTS
A. Review of past work
Experiments are essential for validating calculation meth-
ods and increasing the level of confidence in their accuracy.
(2) To our knowledge, few experiments aimed at reproducing
low-impedance fault conditions have been done on full-scale
power transformers or reactors.
To our knowledge, the first study ever done demonstrated
the importance of tank flexibility and fault location in an oil-
filled rectangular transformer, subjected to sudden discharges
of nitrogen to reproduce the effects of an arc in oil [16]. Alt-
hough not reported, the level of equivalent fault energy for
these tests was obviously low, since the maximum recorded
pressure was of the order of only 40 kPa.
Another investigation, using a special powder combustion
vessel to replicate the effects of an arc in two 275-kV trans-
formers (one shell form and one core form), demonstrated that
such tanks, with appropriate reinforcements, could withstand

0885-8977 (c) 2019 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/TPWRD.2019.2922505, IEEE
Transactions on Power Delivery
an equivalent fault energy estimated at 11.2 MJ [12]. The
pressure rises within these tests were compared to a simplified
analytical model, and a good agreement was observed between
the two.
The study reported in [17] describes several arcing tests in
three different power transformers from 6 to 20 MVA that
examine the spatial distribution of pressure within a tank, as
well as the efficiency of a protective device against overpres-
sures. The maximum arc energy in these tests was 2.4 MJ,
with the majority being around 1 MJ or less.
From the preceding review, it can be inferred that the max-
imum energy yet used in a test has been about 11 MJ, while
the majority of the tests reported have been at much lower
levels of energy. It is also noteworthy that no tests have been
reported describing the failure, under overpressure, of a full-
scale power transformer.
B. Test plan and objectives
In the experimental part of our study, the general test plan
consisted of two tests:
1) Design test: to demonstrate that a tank designed with
the finite element method can withstand the effects of a
20 MJ fault in 50 ms.
2) Rupture test: to demonstrate that the tank would fail
safely in a controlled manner by opening at the tank
cover, as predicted by the finite element simulation, for
a fault above 20 MJ of energy.
These tests were done on a single tank specimen. The rup-
ture test was done following the design test, on the already
deformed tank.
C. Methodology
The effects of an arc in oil were reproduced in our study by
injecting pressurized air into the tank, in the same way that an
arc generates a gas bubble by vaporizing the surrounding oil.
The physics, assumptions and modeling of the mechanical
effects of arcing in oil have been described in [18] and [19],
among others, which are based on several studies that show
the relationship between arc energy and the quantity of gas
generated by oil decomposition [19]. In our study, the quantity
of generated gas is assumed to be 85 cm3 by kJ of arc energy
at standard temperature and pressure, which value is in good
agreement with several studies as reported in [19]. In the past,
this value led to an excellent correlation between a numerical
model based on it and experiments by different investigators
on distribution transformers [19], [21]. With this value, it can
be shown that 24% of the total electrical arc energy is actually
transformed in the tank into mechanical energy, which gener-
ates the pressure rise. Therefore, the mechanical effects of an
arc can be reproduced by introducing this fraction of energy as
mechanical energy within a tank, which is done here by the
injection of pressurized air. This method is based on the com-
pressible flow of air from a pressurized pressure vessel. Such a
flow is “choked” at the smallest cross section area in the pip-
ing system used; that is of constant velocity equal to the speed
of sound in air. For such a flow, the instantaneous mass flow
rate of air is given by [20]:
0885-8977 (c) 2019 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.
4
.
̇= (3)
where ̇is the mass inflow rate (kg/s), Po is the absolute pres-
sure in the pressure vessel (Pa), A is the smallest cross section
area (m2), and To is the temperature in the pressure vessel (oK)
Considering the tank as a closed volume, neglecting any
external heat transfer during the injection process (given its
short duration), neglecting the inflow of kinetic energy from
the gas (since the mass of the gas is low), considering air as a
perfect gas during the process, and neglecting any losses in the
piping system used, it can be demonstrated from the first prin-
ciple of thermodynamics that the rate of transfer of energy
from the pressure vessel to the tank is given by:
̇= ̇ (4)
̇
where is the rate of transfer of energy, γ is the gas heat ca-
pacity ratio, and μ its internal energy, which for a perfect gas
is a function of the temperature only.
The quantity of mechanical energy introduced by a flow of
a given duration can be determined using (3) and (4) in a nu-
merical time integration scheme, taking account of the varia-
tion of Po and To, as the pressure vessel empties itself. The
main advantages of such a methodology are that it does not
involve a high-power source, and that the amount of energy
injected is controlled and repetitive, contrary to an actual arc,
which has varying resistance from test to test [22]
D. Test set-up
The general test set-up is shown in Fig. 7 and Fig. 8. The
active part of the transformer has been replaced by three grav-
el-filled hollow steel cylinders of similar dimensions. To avoid
projection of parts, most of the external accessories are re-
moved except for the pressure relief device (PRD), rated for
opening at 70 kPa, and chains are installed to prevent large
motions of the transformer during the tests. The tank is filled
with water instead of oil to avoid soil contamination in case of
a large spill, and since it is easier to handle.
A 312-litre pressure vessel, with a maximum working pres-
sure of 3,000 psi, was used to inject the air, and the duration of
injection was controlled by two electro-pneumatic valves in
parallel, each mounted on 2-inch inner diameter pipe from the
pressure vessel, converging together into a series of 3-inch
pipes connected to the tank at the injection location. The fault
location where the gas is injected was chosen strategically
near the bottom of the transformer, centered on the longest
wall, to demonstrate that, even for a fault farthest from the
cover, the failure point would remain at the cover level as
required. It has also been demonstrated through explicit dy-
namic simulations (next section), that a fault located at the
center of the longest wall yields the most severe stresses on
the tank.
The test instrumentation consisted of nine pressure sensors
distributed on the walls and cover of the transformer, and one
pressure sensor within the pressure vessel injecting the gas.
3D scans were performed before and after the tests in order to
measure tank deformations.
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/TPWRD.2019.2922505, IEEE
Transactions on Power Delivery
Fig. 7. Tank set-up.
Fig. 8. Schematic of gas injection system with respect to the tank.
E. Determination of test parameters using explicit dynamic
simulations
Explicit dynamic simulation is a numerical methodology
fully suited to the investigation of short duration phenomena,
such as faults within transformers. We have used it in the past
to obtain representative results [4], [19], [21] of actual fault
situations and tests; the reader is referred to these latter refer-
ences for a deeper description of its use. The test parameters
were determined using this methodology, as described below.
1) Theoretical test parameters versus test installation ca-
pabilities
The design test required that the effects of a
20 MJ arc/50 ms duration be reproduced. As discussed, this
corresponds to injecting 24% of 20 MJ in the form of mechan-
ical energy, or 4.8 MJ in 50 ms. In order to inject this amount
of energy, the initial pressure in the pressure vessel must be
determined, as well as the duration during which the valves
must be opened.
Considering the 3,000-psi maximum working pressure of
the pressure vessel and pipe sizes (2 x 2 inch diameter pipes
corresponding to a throat area A of 6.28 in2) available to us to
inject the gas, the shortest possible theoretical duration to
inject 4.8 MJ, assuming instant opening and closing of the
valves, is determined using (3) and (4) to be about 100 ms.
However, due to the restrictions introduced by the valves and
piping system, it was determined experimentally that the ef-
fective area of the injection system at hand corresponded to
two pipes of 1.25 inch diameter, for an effective throat area A
0885-8977 (c) 2019 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.
5
of 2.83 in2; much less than the physical size of the actual
pipes. With this smaller area, and considering again instant
opening and closing of the valves, the shortest duration for
injecting 4.8 MJ with an initial pressure of 3,000 psi would be
about 250 ms. However, since the valves have finite opening
and closing times, we found that at 3,000 psi, the duration
required to inject 4.8 MJ with our system would be longer. For
this reason, the impact of a longer duration of injection in
relation to an actual arc of 50 ms was investigated. Since the
injection would take place within water as opposed to oil, this
effect was also investigated as described below.
2) Differences between an arc in oil and injection of air
into a water-filled tank
To study whether injecting air into a water-filled tank pro-
duces effects similar to an actual arc in oil, we used explicit
dynamic analysis to simulate both situations for a 50-ms 20-
MJ fault. Fig. 9 first presents a comparison of the average
pressure within the tank under consideration for the two meth-
ods. Similar results were obtained for the maximum average
pressure: 394 kPa for arc in oil versus 409 kPa for air injec-
tion, a 3% difference. The design pressure of 556 kPa from (1)
is also shown. It is noteworthy that the latter is about 20%
higher than the pressure calculated by dynamic simulation;
this is not surprising as (1) was determined to cover a whole
range of tanks of different flexibilities and is therefore gener-
ally conservative. Several simulations outside the scope of the
present paper, on many different tanks, also confirmed this.
Fig. 9 also presents a comparison between the plastic work
accumulated in the tank through plastic deformations. It is
observed that the air injection method leads to greater plastic
work by about 13% over the arc in oil, with 904 kJ for air
injection and 798 kJ for the arc in oil. This is explained by the
difference in rigidity between the liquids used inside the tank;
water is more rigid than oil, with a bulk modulus of 2.20 GPa
for water versus 1.47 GPa for oil. Since water is more rigid, it
stores less energy from compression than oil. Thus, water
transfers more energy to the tank (there is also a variation
between the energy stored in the gas bubble between oil de-
composition and air injection since their specific heat ratios
are different: 1.4 for air versus 1.352 for oil [19]).
These comparisons show that the injection of air in a water-
filled tank generates overpressure similar to an arc in oil for
equivalent energy and duration, but produces slightly more
tank plastic deformation. Therefore, injecting air in a water-
filled tank produces conservative, yet representative results.
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/TPWRD.2019.2922505, IEEE
Transactions on Power Delivery
6

Fig. 9. Comparisons between arc in oil and air injection for a 20 MJ/50 ms versus 394 kPa), yet both methods yield the same plastic work
fault.
as required. It is also observed that the pressure with the air
3) Investigation on the duration length
injection methodology still remains under the design pressure.
Since the duration at hand is much longer than an actual
Using injection tests directly in ambient air (unconnected to
20 MJ/50 ms fault, Fig. 10 presents the effects of using differ-
the tank), it was identified that this amount of energy could be
ent durations for the same energy, in the case of air injection
injected in about 550 ms.
in the same water-filled tank. It is observed that
· the maximum average pressure is nearly independent F. Test results and comparisons with simulations
on the duration, although the residual pressure (stabi- 1) Design test
lized pressure once dynamic effects have subsided) The tank under consideration was subjected to air injection
increases with the duration of injection; equivalent to a 30 MJ fault in approximately 550 ms, from an
· the amount of plastic work decreases with the dura- initial pressure of 3,000 psi (20.7 MPa) in the pressure vessel
tion; used. The tank withstood the test without any rupture, projec-
· the maximum average pressure and plastic work for tion of parts, or leak, despite significant (permanent) plastic
the 250 ms and the 500 ms injection durations are deformations. Fig. 12 shows the tank before and after the test.
nearly identical, indicating a convergence of results
from a duration of 250 ms and up.
It is therefore obvious that the duration has a very signifi-
cant effect on tank plastic deformation: significant dynamic
effects are present for shorter durations, while longer durations
lead to a more monotonous rise of pressure with fewer oscilla-
tions. The fact that the residual pressure is higher for longer
durations is explained by the fact that since the tank is less
deformed for these (less plastic work), the volume of the gas
bubble is smaller (this volume being almost the same as the
tank volume increase since the fluid is nearly incompressible),
which corresponds to a higher pressure within it and therefore
within the tank.
Fig. 11. Comparisons between a 20 MJ/50 ms arc in oil, and a 30 MJ/250 ms
air injection.

Fig. 10. Effects of air injection duration (pressure: solid lines; plastic work: Fig. 12. Tank deformation before and after the design energy test.
dashed lines).
Fig 13 shows the pressure variation within the tank for the
4) Determination of the injection energy
nine sensors, as well as their average; all signals shown were
Since the tank design is based on its deformation under
filtered using a 20-point moving average. Also shown is the
pressure as discussed, it was decided to adjust the injected
average pressure within the tank from the simulation for a
amount of energy in order to obtain plastic work similar to a
500 ms injection duration. A good qualitative agreement is
20 MJ/50 ms arc in oil. To identify the corresponding amount
obtained between the simulation and the experimental results.
of required energy, several simulations were performed with
The maximum average pressure from the test is 430 kPa,
increased values of energy, using a 250 ms duration (since it
while the maximum average pressure from the simulation is
produces plastic work equivalent to longer durations but re-
488 kPa; a difference of 13%. Also, the design pressure at
quires less computational time). It was found that the mechan-
556 kPa of the nonlinear finite-element analysis is confirmed
ical energy corresponding to a 30 MJ fault was necessary in
as a safe design approach. Considering the phenomena in-
order to obtain about the same plastic work as a 20 MJ/50 ms
volved, such as large plastic deformations over a short dura-
fault. Fig. 11 presents a comparison between both sets of dura-
tion, as well as the idealizations inherent to the computer
tion and energy for the average pressure and plastic work. It is
model, the agreement between the test and simulation is very
observed that the peak average pressure is higher for the
good in our opinion. The observed differences may be ex-
30 MJ/250 ms air injection than for the arc in oil (513 kPa
plained by different factors. First, the computer simulation

0885-8977 (c) 2019 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/TPWRD.2019.2922505, IEEE
Transactions on Power Delivery
idealized the way the gas is injected, by introducing it directly
within the tank [4], as opposed to the test, where the gas was
introduced through a series of pipes ending in a corrugated
hose attached to the tank (Fig. 8). Due to this, and the finite
opening time of the valves, a delay is observed in the pressure
rise during the test, as it is observed that the simulated pres-
sure rise has an earlier onset. It was also observed that as the
tank deformed during the test, the corrugated hose attached to
the tank was progressively kinked, and it is believed that this
introduced energy losses during the injection process. Also,
there are probably other unaccounted energy losses in the
piping system due to their construction and changes in pipe
area, as well as losses introduced by valve construction and
operation.
Fig. 13. Design energy level test pressure measurements.
Once the pressure in the transformer was relieved by PRD-
operation, a scan of the transformer was performed. The tank’s
permanent displacements were obtained and compared with
the pre-test scan, as shown in Fig. 14. It is observed that the
maximum displacement at the center of the wider wall is 108
mm. Fig. 14 also shows the permanent displacements from the
simulation results; the maximum observed is 112 mm. Com-
parisons of both figures and their maximum displacements
demonstrate a relatively good agreement between the test and
simulation.
Based on the comparison of pressure and displacements be-
tween test and simulation results, it can be concluded that the
computer model provided conservative, yet representative
results.
Fig. 14. Permanent displacements after the design energy level test (left:
simulation; right: scan measurement).
During the test, the PRD opening command was triggered
at 57 ms, corresponding to an average pressure of about
50 kPa within the tank at this time. Video recordings showed
0885-8977 (c) 2019 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.
7
that from that point on, water was forcefully ejected onto the
ground through the vertical pipe attached to the end of the
valve. After 1 s, the average pressure was 417 kPa, while it
was 220 kPa after 5 s. It is hence observed that the effect of
the PRD opening is rather slow in comparison to a 3-cycle
fault duration (50 ms). It is therefore inferred that while such
devices cannot be used effectively as a protection for low-
impedance faults of short durations, their effectiveness is
probably adequate for high-impedance faults of much longer
durations.
2) Rupture test
For this test, water was added to the tank until it was com-
pletely refilled, replacing the volume change obtained from
the design test. In order to generate rupture conditions within
the tank and verify that it would fail at the tank cover as de-
signed, the same test procedure was repeated without reclosing
the valves once opened, from the same initial pressure of
3,000 psi within the pressure vessel. This corresponded to a
potential release of energy amounting to a 66 MJ fault for the
complete emptying of the pressure vessel.
It was observed that the tank cover failed (Fig. 15) approx-
imately 450 ms after the start of the injection process, at an
average pressure of 486 kPa based on the peaks from the fil-
tered signals at the nine measurement locations; the unfiltered
signals revealed short time duration peaks over 1,000 kPa at
some locations but it is difficult to assess if these are real or
simply related to the dynamic response of the sensors.
Fig. 15. Rupture at the tank cover.
At the moment the cover failed, approximately 20 MJ of
additional energy was injected from the pressure vessel due to
the sudden pressure drop. Since the test started with an initial-
ly deformed tank which was refilled, and assuming that only a
small amount of energy is needed to recover the elastic defor-
mations in the tank to continue the permanent deformation
process, it can be hypothesized that this tank could withstand
at most 50 MJ of energy (30 MJ from the first test and 20 MJ
from the second test), injected in approximately 1 s. Assuming
the same duration effect as for the first test, where 1.5 times
the energy of a 50 ms fault was needed to obtain equivalence
of the tank plastic deformation state, it can again be hypothe-
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/TPWRD.2019.2922505, IEEE
Transactions on Power Delivery
sized that this tank could withstand a fault of about 30 MJ in
50 ms before rupturing (50 MJ/1.5). Additional dynamic simu-
lations following the same steps as the two experiments here
(first experiment, tank refilled, second experiment) supported
this hypothesis. A dynamic simulation of 30 MJ in 50 ms
yielded a peak pressure of about 1,000 kPa in the area close to
the cover, in good agreement with the 946 kPa rupture pres-
sure obtained from the tank design static calculations.
From the difference of approximately 10 MJ between the
energy level at rupture (30 MJ) and the design energy level
(20 MJ), it can be concluded that this particular design pos-
sesses a safe leeway before rupture at the cover.
IV. CONCLUSIONS
This study demonstrates that it is feasible to design a power
transformer to withstand high-energy low-impedance faults of
short duration, which fail safely at the cover at fault energies
above the design fault energy. The governing design principle
to achieve this was based on increasing the tank flexibility
while providing sufficient strength. The tank was designed
using nonlinear finite-element analysis, using a rupture criteri-
on that is safe from a design point of view and proved by the
experiments. The tests were designed using explicit dynamic
simulations in order to reproduce the effects of an arc in oil by
injecting pressurized air. It is noteworthy that the tank per-
formed with a relatively good agreement with the finite-
element and explicit dynamic simulations results. The energy
levels used in the tests reported are, to our knowledge, the
largest ever reported.
V. ACKNOWLEDGMENT
The authors gratefully acknowledge the contributions of
B. Lapointe, J. Hennessey and J.-P. Charest-Fournier for de-
signing the test set-up and performing the experiments report-
ed in this paper.
VI. REFERENCES
[1] Guide for Transformer Fire Safety Practices, CIGRÉ Technical Bro-
chure 537, Working Group A2.33, June 2013.
[2] M. Foata, “Power Transformer Fire Risk Assessment,” Presented at
CIGRÉ Transformer Technology Conference, Sydney, Australia, 2008.
[3] M. Foata and J.-B. Dastous, “Power Transformer Tank Rupture Preven-
tion,” Presented at CIGRÉ Transformer Technology Conference, Paris,
France, 2010.
[4] J.-B. Dastous, J. Lanteigne and M. Foata, “Numerical Method for the
Investigation of Fault Containment and Tank Rupture of Power Trans-
formers,” IEEE Trans. Power Delivery, vol. 25, no. 3, pp. 1657-1665,
July 2010.
[5] K. Goto and Y. Miura, "The Pressure Rise During the Internal Fault in
an Oil Filled Transformer," in CIGRÉ colloquim, Florence, 1987.
[6] T. Kawamura, M. Ueda, K. Ando, T. Maeda, Y. Abiru, M. Watanabe, K.
Moritsu, “Prevention of Tank Rupture Due to Internal Fault of Oil-Filled
Transformer”, CIGRE, 12-02, 1988.
[7] J.-B. Dastous, E. Taschler, S. Bélanger, and M. Sari, “A Comparison of
numerical methods for modeling overpressure effects from low imped-
ance faults in power transformers,” Procedia Engineering, vol. 202, pp.
202-223, 2017.
[8] N. Abi-Samra, J. Arteaga and B. Darovny, "Power Transformer Tank
Rupture and Mitigation – A summary of Current State of Practice and
Knowledge by Task Force of IEEE Power Transformer Subcommittee,"
IEEE Trans. on Power Delivery., vol. 24, pp. 1959-1967, Sept. 2009.
[9] S. Brodeur, Y. Salmi and A. Collier, "A Flexible Friend," ABB Review,
vol. 1, pp. 18-22, 2016.
0885-8977 (c) 2019 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.
8
[10] M. Foata and G. Khouzam, "Power Transformer Tank Rupture," Cana-
dian Electrical Association, Engineering and Operating Division, To-
ronto, ON, March 1994.
[11] R. E. Kothmann, and D. G. Thompson, "Power Transformer Tank
Rupture: Risk Assessment and Mitigation," EPRI, Technical Report TR-
104994, April 1995.
[12] H. Murakami, T. Kobayashi, Y. Shirasaka and Y. Ebisawa, “Tank
Design for Low Rupture Risk – Japanese Practice,” Presented at CIGRÉ
Transformer Technology Conference, Sydney, Australia, 2008.
[13] Atlas of stress-strain curves, Second ed., ASM International, 2002.
[14] G. E. Dieter and L. C. Schmidt, “Engineering Design,” 5th ed., Ed. New
York: McGraw-Hill, 2013, pp. 628-679.
[15] S. Bélanger, S. Brodeur, J.B. Dastous and N. Soucy, "Experimental and
Numerical Studies of Large Steel Plates Subjected to High Pressure
Loading," in Proc. World Congress on Computational Mechanics Ses-
sion, Barcelona, Spain, 2014.
[16] R. J. Ringlee and N. W. Roberts, "Tank Pressures Resulting from Inter-
nal Explosions," AIEE Trans., vol. 78, pt. III-B, pp. 1705-1710, 1959.
[17] S. Muller, M. Petrovan-Boiarciuc and M. Périgaud, “Pressure wave
propagation induced by short circuits inside power transformers: Devel-
opment of a simulation tool, comparisons with experiments and applica-
tions,” The International Journal of Multiphysics, vol. 3, no. 4, pp. 361-
386, 2009.
[18] M. Foata, M. Iordanescu, C. Hardy, “Computational methods for the
analysis of explosions in oil-insulated equipment,” IEEE Transactions
on Power Systems, Vol. 3, No.1, February 1988.
[19] J.-B. Dastous and M. Foata, "Analysis of Faults in Distribution Trans-
formers with MSC/PISCES-2DELK," presented at the MSC World Us-
ers Conf., Los Angeles, CA, 1991.
[20] I. G. Currie, Fundamental Mechanics of Fluids. McGraw-Hill Publish-
ing Company, 1974, pp. 356-383.
[21] J.-B. Dastous, M. Foata and A. Hamel, “Estimating Overpressures in
Pole-Type Distribution Transformers Part II: Prediction Tools,” IEEE
Trans. Power Delivery, vol. 18, no. 1, pp. 120-127, Jan. 2003.
[22] A. Hamel, J.-B. Dastous and M. Foata, “Estimating Overpressures in
Pole-Type Distribution Transformers Part I: Tank Withstand Evalua-
tion,” IEEE Trans. Power Delivery, vol. 18, no. 1, pp. 113-119, Jan.
2003.
VII. BIOGRAPHIES
Samuel Brodeur is a member of the IEEE working group developing Std
C57.156 (transformer tank rupture mitigation). He obtained his Bachelor of
Science in mechanical engineering in 2006, and Master of Science in 2014
from École de Technologie Supérieure, Montréal, Canada.
In 2008, he joined ABB Inc. in Varennes, Québec, as a mechanical engi-
neer and has worked in the field of power transformer structural analysis since
then. His work focuses on power transformer seismic design and tank rupture
mitigation under internal fault conditions.
Jean-Bernard Dastous is Chair of the IEEE working groups developing Std
605 (bus design in air insulated substations) and Std 1527 (seismic design of
buswork between substation equipment). He is also a member of the IEEE
working groups developing Std 693 (seismic design of substation equipment),
and Std C57.156 (transformer tank rupture mitigation). He obtained his
Bachelor of Science in mechanical engineering from Université de
Sherbrooke, Sherbrooke, Canada, in 1989, and Master of Engineering from
McGill University, Montréal, Canada, in 1993.
In 1989, he joined IREQ (Hydro-Québec’s research institute), in Varennes,
Québec, as a research engineer and has worked in the field of structural
analysis of substation equipment and structures since then. His areas of re-
search interest include seismic design, design of bus structures, and arc con-
tainment in oil-insulated equipment.