Tracking number: OE 12.

09
Award No: DE-OE0000072
Identification Number: DE-0000072
CFDA No. 81.122

Final Report

Development of Toroidal Core Transformers

By

Francisco de Leon
Associate Professor
Department of Electrical and Computer Engineering
New York University
Six Metrotech Center
Brooklyn, NY 11201
Tel (718) 260 3961
fdeleon@poly.edu
fdeleon@nyu.edu
http://www.poly.edu/power

August 2014


Acknowledgment: This material is based upon work supported by the Department of Energy
under Award Number DE-OE0000072.

Disclaimer: This report was prepared as an account of work sponsored by an agency of the
United States Government. Neither the United States Government nor any agency thereof, nor
any of their employees, makes any warranty, express or implied, or assumes any legal liability or
responsibility for the accuracy, completeness, or usefulness of any information, apparatus,
product, or process disclosed, or represents that its use would not infringe privately owned rights.
Reference herein to any specific commercial product, process, or service by trade name,
trademark, manufacturer, or otherwise does not necessarily constitute or imply its endorsement,
recommendation, or favoring by the United States Government or any agency thereof. The
views and opinions of authors expressed herein do not necessarily state or reflect those of the
United States Government or any agency thereof.

Page 2 of 37
Executive Summary

The original objective of this project was to design, build and test a few prototypes of single-
phase dry-type distribution transformers of 25 kVA, 2.4 kV primary to 120 V transformers using
cores made of a continuous steel strip shaped like a doughnut (toroid). At different points during
the development of the project, the scope was enhanced to include the more practical case of a 25
kVA transformer for a 13.8 kV primary system voltage. Later, the scope was further expanded to
design and build a 50 kVA unit to transformer voltage from 7.62 kV to 2x120 V. This is a
common transformer used by Con Edison of New York and they are willing to test it in the field.

The project officially started in September 2009 and ended in May 2014. The progress was
reported periodically to DOE in eighteen quarterly reports. A Continuation Application was
submitted to DOE in June 2010. In May 2011 we have requested a non-cost extension of the
project. In December 2011, the Statement of Project Objectives (SOPO) was updated to reflect
the real conditions and situation of the project as of 2011. A second Continuation Application
was made and funding was approved in 2013 by DOE and the end date was extended to May
2014.

The technical challenges that were overcome in this project include: the development of the
technology to pass the impulse tests, derive a model for the thermal performance, produce a
sound mechanical design, and estimate the inrush current. However, the greatest challenge that
we faced during the development of the project was the complications of procuring the necessary
parts and materials to build the transformers. The actual manufacturing process is relatively fast,
but getting all parts together is a very lengthy process.

The main products of this project are two prototypes of toroidal distribution transformers of
7.62 kV (to be used in a 13.8 kV system) to 2x120 V secondary (standard utilization voltage);
one is rated at 25 kVA and the other at 50 kVA. The 25 kVA transformer passed the impulse test
in KEMA high-voltage laboratories. Additional products include: nine papers published in the
IEEE Transactions on Power Delivery, one patent has been filed, three PhD students were

Page 3 of 37
supported from beginning to graduation, five postdoctoral fellows, and three MSc students were
partially supported.

The electrical characteristics of our dry-type toroidal transformers are similar to those of the
oil-immersed pole mounted transformers currently in use by many utilities, but toroids have
higher efficiency. The no-load losses of the 50 kVA prototype are only 45 W. A standard
transformer has no-load losses between 90 and 240 W. Thus, even the finest transformer built
today with standard technology has double the amount of no-load losses than the prototype
toroidal transformer.

When the manufacturing process is prepared for mass production, the cost of a dry-type
toroidal transformer would be similar to the price of an oil-filed standard design. However,
because of the greatly reduced losses, the total ownership cost of a toroidal transformer could be
about half of a traditional design.

We got a grant from Power Bridge NY in the amount of $149,985 from June 2014 to May
2015 to continue developing the transformer with commercialization objectives. We are
considering the possibility to incorporate a company to manufacture the transformers and have
contacted investors.

The current status of the real life testing is as follows: after several months of silence, Con
Edison has re-started conversations and has shown willingness to test the transformer. Other
companies, PSE&G and National Grid have recently also shown interest and we will present our
product to them soon.

Page 4 of 37
Contents
Document History ..........................................................................Error! Bookmark not defined. 

1.  Introduction ............................................................................................................................. 7 

2.  Project Development ............................................................................................................. 10 

2.1 Original Objective ............................................................................................................... 10 

2.2 Enhanced Objective............................................................................................................. 10 

2.3 Design Challenges ............................................................................................................... 11 

2.3.1 Electromagnetic Design................................................................................................ 11 

2.3.2 Insulation Design .......................................................................................................... 12 

2.3.3 Thermal Design ............................................................................................................ 13 

2.3.4 Mechanical Design ....................................................................................................... 14 

2.3.5 Inrush Currents ............................................................................................................. 15 

3.  Construction of Prototypes .................................................................................................... 16 

3.1 Design Specifications .......................................................................................................... 16 

3.2 Insulation ............................................................................................................................. 18 

3.3 Core and Electrostatic Shielding ......................................................................................... 18 

3.4 High-Voltage Winding ........................................................................................................ 19 

3.5 Low-Voltage Winding......................................................................................................... 19 

3.6 Selection of Bushings and Surge Arresters ......................................................................... 21 

3.7 Transformer Tank................................................................................................................ 22 

3.8 Installation of the Active Element in the Tank ................................................................... 23 

3.9 Connections ......................................................................................................................... 25 

3.10 Nameplate Information ..................................................................................................... 25 

Page 5 of 37
 3.11 Serial Number ................................................................................................................... 27 

4.  Value Propositions ................................................................................................................. 28 

5.  Products ................................................................................................................................. 30 

5.1 Prototypes ............................................................................................................................ 30 

5.2 Papers .................................................................................................................................. 31 

5.3 Patent ................................................................................................................................... 31 

5.4 List of Students Supported from the Grant ......................................................................... 32 

6.  Conclusions ........................................................................................................................... 33 

7.  References ............................................................................................................................. 36 

8.  Appendices ............................................................................................................................ 37 

Page 6 of 37
 1. Introduction

The U.S. Environmental Protection Agency estimates losses of 60 to 80 billion kWh

attributable to distribution transformer inefficiencies, which rob U.S. business and American
consumers of approximately $4 billion per year. The American Council for an Energy Efficiency
Economy has identified (in 2006) distribution transformers as the second devices with the largest
potential to save energy (only after residential furnaces and boilers).

Currently, there are two basic arrangements for the iron-cores used to build distribution
transformers: (1) The Core-Type, cores are assembled by stacking laminations and the
transformer is completed by sliding pre-made windings; (2) Shell-Type, a continuously wound
core is cut and wrapped around the windings a few laminations at a time. As a consequence, both
arrangements of the finished core are left with air gaps that increase the magnetizing current and
the no-load loss. Figure 1.1 illustrates the internal construction of both types.

Core type Shell type

Figure 1.1. Arrangement of core and windings (active element) of core-type and shell-type distribution
transformers.

Page 7 of 37
 The alternative construction proposed in this project, currently used in low-voltage and low-
power applications, is to use a core made of a continuous steel strip that is wound into a
doughnut shape (toroid) and then wrapped entirely in coils (see Figure 1.2). This gapless
construction allows for smaller, more efficient, lighter, and cooler transformers with reduced
electromagnetic interference and lower acoustic noise. The main technical advantage is that the
no-load loss is substantially reduced. There are also savings to be found in the load losses
because the windings have fewer (and shorter) turns. These advantages have a greater impact for
transformers that operate in lightly loaded (suburban and rural) areas because the no-load loss is
very small. Since toroidal transformers can be made smaller than standard transformers, it is
possible to replace oil immersed overhead transformers with dry toroidal units, reducing the
potential for violent faults in addition to the environmental benefits of avoiding the use of oil.

Figure 1.2. Arrangement of core and windings (active element) of the new toroidal distribution
transformers.

Toroidal transformers are not currently in use in distribution systems. Given the lack of
experience with toroidal design at medium and high voltages, efforts have been made to develop
the technology to pass the impulse tests, study the thermal performance and produce a sound
mechanical design. All the design has been done at the School of Engineering of New York

Page 8 of 37
University, by graduate students (M.S. and Ph.D.) and post-doctoral fellows under the guidance
of Prof. Francisco de Leon.

The development of toroidal core transformers for medium-voltage applications was a

project supported by the US Department of Energy under Grant DEOE0000072. The underlying
idea is to benefit from the virtues of the toroidal construction to manufacture, test and install
toroidal transformers suitable for power distribution applications. At this time (August 2014) we
have built two working prototypes, one of 25 kVA and another one of 50 kVA. The 25 kVA
transformer has been successfully tested and passed the impulse tests at Kema high-voltage
laboratory. The electrical characteristics of the toroidal dry-type transformers are similar to those
of oil-immersed pole mounted transformers currently in use by many utilities, but with higher
efficiency. The performance of toroidal transformers is not typical; for example the 50 kVA
transformer has no-load losses of only 45 W. A standard transformer has no-load losses between
90 and 240 W. Thus, even the finest transformer built today with standard technology has double
the amount of no-load loss than the prototype toroidal transformer.

The 50 kVA transformer was designed and built according to the Con Edison purchase
requirements. It is intended to be installed on an overhead system to allow for the observation of
the unit under real life operation.

It is estimated that, in the mass production phase, the cost of a dry-type toroidal transformer
will be very close to that of a standard oil-immersed transformer. However, because of the higher
efficiency, the operating cost will be almost half. Its dry-type construction inherently makes it
environmental friendly and it is not subjected to explosions. There are 100 million pole mounted
transformers in the world. There are 40 million in the US alone with a life expectancy is 40
years. Therefore, about one million every year are substituted even with no growth.

Page 9 of 37
 2. Project Development

2.1 Original Objective

The objective of this project is to design, build and test a few prototypes of single-phase dry-
type distribution transformers using cores made of a continuous steel strip shaped like a
doughnut (toroid). The scope is to minimally build a toroidal transformer with the following
specifications: 25 kVA, 2.4 kV primary to 120 V secondary, 95 kV BIL, operating at 60 Hz, with
a minimum efficiency of 98.91 %.

2.2 Enhanced Objective

During the development of the project, the scope was enhanced to include more practical
distribution transformer applications. The medium voltage of 2.4 kV is a legacy voltage level.
Modern distribution systems use higher voltage, for example 13.2 kV or 25 kV. Therefore, a 25
kVA transformer for a 13.2 kV primary system voltage was proposed as the new objective.

Later, the scope was further expanded to design and build a 50 kVA unit to transform voltage
from 7.62 kV to 2x120 V. The reason is that although 25 kVA transformers are still in use, the
local utilities (Con Edison and Long Island Power Authority) substitute their 25 kVA for 37.5
kVA or 50 kVA. The 50 kVA was preferred because this is a very common transformer for Con
Edison and they are willing to test it in the field.

Although toroidal distribution transformers will be dry-type, their efficiency will be even
better than the efficiency of the corresponding oil-filled transformers. Then for a 25 kVA
transformer the minimum efficiency will be 98.91% and for the 50 kVA the minimum efficiency
will be 99.08%.

Page 10 of 37
 2.3 Design Challenges

Since there is no industry experience designing and building distribution transformers in

toroidal cores, the initial challenge was to design the transformer from top to bottom. All
electromagnetic and mechanical design processes necessary to produce a working transformer
had to be analyzed. In this section all aspects of toroidal transformer distribution design are
reported. Next chapter describes the construction challenges.

2.3.1 Electromagnetic Design

A design program was developed in Matlab to provide preliminary design parameters from
the transformer specifications. The input data are the power, frequency, winding voltages, basic
insulation level (BIL), and magnetic flux density. Using Faraday’s law and the dimensions of the
components (core, wires and insulation) the number of turns and the general dimensions of the
transformer are estimated. Designs were eventually validated using finite elements simulations.
We realized that there were no formulae available to estimate the leakage inductance;
therefore we needed to derive such formulae. The leakage inductance is a very important
performance parameter of a distribution transformer since it determines the voltage drop and the
short-circuit currents. Three different techniques were proposed to control the leakage
inductance: (1) add spaces between the windings; (2) insert a second core between primary and
secondary; (3) leave unwound sectors around the core. The details have been published in the
following two papers:

I. Hernández, F. de León, and P. Gómez, “Design Formulas for the Leakage Inductance of
Toroidal Distribution Transformers”, IEEE Transactions on Power Delivery, Vol. 26, No. 4,
October 2011, pp. 2197-2204.

F. de León, S. Purushothaman, and L. Qaseer, “Leakage Inductance Design of Toroidal

Transformers by Sector Winding”, IEEE Transactions on Power Electronics, Vol. 29, No. 1,
January 2014, pp. 473-480.

Page 11 of 37
 The design program directly provides the drawing model of the toroidal transformer ready to
be simulated in COMSOL Multiphysics. Then we can perform all magnetic, thermal, and
mechanical calculations necessary to verify the design.

2.3.2 Insulation Design

The design of the insulation was the most important concern that we had at the beginning.
The reason is that there is no experience in the industry. After substantial numerical
experimentation (finite elements simulations) and based on the experience the team members had
with large power transformer, an electrostatic shielding was proposed to control the impulse
response of toroidal transformers. An important difficulty is the fact that the turns touch in the
internal part and are far in the outside. This produces a highly nonlinear distribution of the
impulse stresses. The first and second turns see a tremendous stress. The results of the study are
published in the following paper:

P. Gómez, F. de León, and I. Hernández, “Impulse Response Analysis of Toroidal Core

Distribution Transformers for Dielectric Design”, IEEE Transactions on Power Delivery,
Vol. 26, No. 2, April 2011, pp. 1231-1238.

The solution to the impulse response problem came in the form of an electrostatic shield.
This is standard in the field of large power transformers. However, we proposed a novel solution,
only applicable to the toroidal geometry, which is to make the magnetic core to double as the
electrostatic shield by electrically connecting it to the high voltage terminal; see Figure 2.1. This
is a patented solution that in addition of controlling the impulse distribution, allows for the use of
very thin insulation between layers, therefore providing an excellent thermal performance. The
following patent has been filed:

Page 12 of 37
 F. de Leon, “Electrostatic Shielding for Transformers”, United States Patent Application,
Serial No. 61/857,581; Provisional patent filed on July 23, 2013; Utility patent application
filed on July 23, 2014.

Figure 2.1. Toroidal transformer winding and core showing the connection of the core to the high voltage
terminal.

2.3.3 Thermal Design

One of the most important challenges to overcome was the thermal design. Since a dry-type
transformer design was sought and air is substantially less efficient than oil at removing heat, it
was important to minimize losses. Additionally, since there is no need to cut the core to set the
windings on it, we can take full advantage of the excellent properties amorphous materials to
produce an extremely efficient transformer. As a result of extensive finite elements simulations,
we realized that the thermal performance will greatly improve when we set the high voltage
winding inside and the low-voltage winding outside (the opposite to current technology). This
construction prevents heat from the low-voltage winding to be trapped in the multi-layer
insulation necessary for the high-voltage winding. Moreover, this construction allows setting the

Page 13 of 37
common ground between windings in the middle. All the details, including a model and
experimental validation, can be found in the following paper:

S. Purushothaman and F. de León, “Heat Transfer Model for Toroidal Transformers”, IEEE
Transactions on Power Delivery, Vol. 27, No. 2, April 2012, pp. 813-820.

2.3.4 Mechanical Design

The mechanical design includes two major parts: the short circuit stresses and the mechanical
integrity of the tank. Because of its closed construction, it is expected that the mechanical
strength of toroidal transformers to be better than that of traditional designs. A set of finite
elements simulations where performed to corroborate the above hypothesis. Figure 2.4 show an
arrow plot of the mechanical forces in a toroidal transformer during a high current short-circuit.
Different to traditional designs, the stresses are almost only in the direction tending to elongate
or contract the conductors and negligible compression/expansion forces are exerted on
insulation. Therefore, no modifications to the design are necessary to deal with short-circuit
stresses.

Figure 2.4. Arrow plot of force density for a toroidal transformer under short circuit.

Page 14 of 37
 The solution that we proposed for the design of the tank was to use exactly the same tanks
that utilities use. In fact, we have gotten several tanks from Con Edison. In this way, we are sure
that tanks comply with the specification and all clamping ironwork is at the right location for
ease of installation.

2.3.5 Inrush Currents

The sole disadvantage that toroidal transformers have over traditional designs is the larger
inrush currents. This is caused by the lack of gaps in the core, which is precisely the feature that
gives toroidal transformers their technical advantage over traditional constructions. Inrush
currents are produced by the saturation of the core during the energization of the transformer.
Since our prototypes have not yet been installed in a real application, we do not have
experimental evidence that the larger inrush would be a problem. However, we have continued
the research in this area and have published/submitted a number of papers offering solutions to
reduce the inrush currents. At this time none of the solution has been implemented because they
will inevitably reduce the efficiency of the transformer. The details of this research have been
reported in the following papers:

F. de León, A. Farazmand, and P. Joseph, “Comparing the T and  Equivalent Circuits for
the Calculation of Transformer Inrush Currents”, IEEE Transactions on Power Delivery,
Vol. 27, No. 4, October 2012, pp. 2390-2398.

S. Jazebi, A. Farazmand, B. Murali, and F. de León, “A Comparative Study on pi and T

Equivalent Circuits for the Analysis of Transformer Ferroresonance”, IEEE Transactions on
Power Delivery, Vol. 28, No. 1, January 2013, pp. 526-528.

A. Farazmand, F. de León, K. Zhang, and S. Jazebi, “Analysis, “Modeling and Simulation of

the Phase-Hop Condition in Transformers: The Largest Inrush Currents”, IEEE Transaction
on Power Delivery, Vol. 29, No. 4, August 2014, pp. 1918-1926.

S. Jazebi and F. de León, “Experimentally Validated Reversible Multi-Winding Transformer

Model for the Accurate Calculation of Low-Frequency Transients”, accepted for publication in
the IEEE Transactions on Power Delivery (early access available).

Page 15 of 37
 3. Construction of Prototypes

In this section, after describing the design process, the manufacturing steps are illustrated. As
mentioned above, a design program was developed to consider all features described above. IN
fact, the program goes through an exhaustive optimization process that complies with all
specifications and gives the lowest cost.

3.1 Design Specifications

It is required that the design complies with the DOE final rule on efficiency requirements as
presented in Table 3.1. Although the toroidal transformers developed here are dry-type, the
transformers are designed to substitute oil-immersed transformers with similar electrical
characteristics. Therefore, the design is carried out with the higher efficiency of oil-immersed
transformers.

The hottest-spot temperature rise above ambient temperature1 shall not exceed the values
given in Table 2.2 [2]. Note that higher winding average temperature rises may apply if the
manufacturer provides thermal design test data sustaining that the temperature limits of the
insulation are not exceeded [2]. Also temperature of external parts accessible to operators (tank)
shall not exceed the temperature rises over ambient temperature at maximum rated load shown in
Table 3.2. Hence, for the toroidal pole-mounted transformer, the tank temperature shall not
exceed 80°C. The design program computes the temperature of the different parts of the
transformer with the electrical model presented in [3].

1
Based on an average daily ambient temperature of 30°C, with a maximum ambient temperature of 40°C

Page 16 of 37
 Table 3.1. Energy efficiency standards for single-phase low-voltage distribution transformers [7]
kVA Dry-Type (%) Oil-immersed (%)
15 97.7 98.76
25 98.0 98.91
37.5 98.2 99.01
50 98.3 99.08
75 98.5 99.17
100 98.6 99.23
167 98.7 99.25

Table 3.2. Limits of temperature rise for continuously rated dry-type transformer windings [2]
Insulation temperature class Winding hottest-spot temperature rise Average winding-temperature rise
(°C) (°C) by resistance (°C)
130 90 75
150 110 90
180 140 115
200 160 130
220 180 150

Table 3.3. Allowable temperature rise of external parts over ambient [2]
Readily accessible 65°C
Not readily accessible 80°C
Note: Not readily accessible is considered to apply to equipment parts located at heights greater than 2.0 m above
floor level or otherwise located to make accidental contact unlikely

Page 17 of 37
3.2 Insulation

A toroidal winding machine is loaded with insulation paper. The insulation used in the
toroidal transformer is 6 mil DMD-F composite paper with thermal class F, 155° C. The same
insulation material is used to insulate the HV windings of all neighboring layers. The last
insulation layer covers the entire HV winding. This layer insulates the HV winding from the LV
windings. Note that the LV windings are made with insulated welding cable. Therefore, due to
the low continuous operating voltage, there is no need to insulate between layers of the LV.

3.3 Core and Electrostatic Shielding

The material of the core is amorphous iron. The purpose for using amorphous cores is to
reduce the energy losses. As the result, the transformer could be designed smaller, lighter and
cooler. The nominal flux density is 1.4 [Tesla] for the transformer design.

The core is electrically connected to the high voltage (HV) terminal to use less insulation
between the core and the winding. Therefore, the inner winding is the HV and the low voltage
(LV) winding is wound on top of the HV winding. This technique creates an electrostatic shield
between the core and the HV winding. The function of the electrostatic shield is to produce a
more uniform distribution of the electrical stresses that the inter-turn and inter-layer insulation
undergo during the impulse test [4], [5].

For this purpose the core is drilled (perpendicular to laminations). An insulated wire (with
the same gauge as the HV winding) is screwed to the core with a cable lug; see Figure 3.1. The
core connection is then held in place with adhesive tape to the insulation to avoid the
replacement of the wire; see Figure 3.1(b).

Page 18 of 37
 (a) (b)

Figure 3.1. (a) Connection of the core to the HV after drilling the core perpendicular to laminations; (b)
the insulated core and the electrical connection.

3.4 High‐Voltage Winding

The transformer contains a single high voltage winding which is wound on 340° of the core.
Note that, 20° are left unwound to pass two wires; the high voltage winding and the electrostatic
shield lead to the core. For the high voltage winding, magnet wire with the electrical
characteristics presented in [6] and in the Appendix are used.

3.5 Low‐Voltage Winding

The low voltage winding is manufactured with welding cables [7]. On the low-voltage side,
the transformer has two windings and series/parallel connection capabilities. Therefore, it can
supply 120/240 V loads. The two low-voltage windings are similar and each of them is wound
on 160° of the core. This method completely avoids the use of additional insulation between the

Page 19 of 37
two low-voltage windings. Also, this approach increases the impedance between the two LV
windings which limits the short circuit current [8], [9].

The low voltage winding strategy is shown in Figure 3.2. Note that, on each side, there is a
20° gap between the two windings to avoid the impulse test failure. It is necessary that the LV
windings are wound in a way that the terminals are geometrically located at 180° from the HV
winding. However, the other HV lead (connection to the neutral) is geometrically in phase with
the LV leads.

(a) (b)

Figure 3.2. Low voltage winding strategy with two layers.

Both windings start with the first turn located near the HV winding connection to the neutral.
In order to get the additive polarity, one of the LV windings is wounded clockwise and the other
is wounded counter clockwise. The first turn of the second layer starts on top of the last turn of
the first layer. This procedure is followed until the last layer is completed. Note that, to establish

Page 20 of 37
proper connections to the LV bushings, the last layers of the two windings are wound in a way
that the end leads locate close to the start leads.

3.6 Selection of Bushings and Surge Arresters

The surge arrester was selected from Ultrasil Polymer-Housed evolution (10 kA) surge
arresters manufactured by cooper power systems [10] and with respect to the recommended
arrester rating per IEEE Std. C62.22 standard [11]. For example, for the 13.2 Y/7.62 and solidly
grounded neutral, the voltage rating of the surge arrester is selected 15 kV rms, where
MCOV=12.7 kV rms. The part number is URT1507-0A1C-1D1C.

Transformer mounting bracket is used to install the surge arrester as shown in Figure 3.3(a)
and Figure 3.4(c). The bracket dimensions are shown in Figure 3.3(b). A surge arrester bracket is
provided adjacent to the HV bushing. This bracket allows mounting of a surge arrester with
enough space from the transformer tank to prevent the tank from interfering with the operation of
the surge arrester. Bushings should comply with the requirements of IEEE C57.19.00 and IEEE
C57.19.01 [12], [13].
 

Page 21 of 37
 (a) (b)

Figure 3. 3. Evolution surge arrester with isolator, insulated hanger and transformer mounting bracket.

3.7 Transformer Tank

Transformer tanks need to be sealed and need to be rigid. The structure of the tank is shown
in Figure 3.4. One earth terminal is located on the primary side of the tank, vertically below the
center LV bushing (X2). The other earth connection (tank ground connection) is through a
mounting nut welded to the lower part of the tank vertically below the HV bushing. The
transformer needs welded lifting lugs and hanger brackets for direct-to-pole mounting. The lead
(cover) needs to be electrically connected to the tank; see Figure 3.4(c).

The interior and exterior surfaces of the tank shall either be abrasive blast-cleaned or be
pickled, rinsed and dried to avoid corrosion. Before being coated, both surfaces shall be free
from rust, grease, oil and moisture. Electrostatically applied polyester powder paint system
could be used for superior corrosion protection.

Page 22 of 37
 (a) (b)

(c) (d)

Figure 3.4. Different parts and accessories of the tank.

3.8 Installation of the Active Element in the Tank

To shield the LV winding from the bottom of the tank, alumina plates are utilized. The
thermal resistivity of alumina is very low while the electrical resistivity of this material is very

Page 23 of 37
high. Therefore, it is suitable for this application. First, six 2"×2"×0.412" plates are placed at the
bottom part of the tank with the arrangement presented in Figure 3.5. Then, the active part of the
transformer is moved with the crane into the tank and placed on top of the alumina plates. Then,
the transformer tank is filled with the epoxy resin (see Figure 3.6). Note that, to increase the heat
transfer efficiency and reduce the weight of the epoxy, a PVC pipe with the same diameter as the
ID of the active part could be utilized.

Figure 3.5. Arrangements of ceramic plates at the bottom of the tank.

Figure 3.6. Filling the tank with the epoxy resin.

Page 24 of 37
3.9 Connections

Closed ring (O-type) cable lugs are used for the connection of the low voltage cables to the
low voltage bushings. One end of a cable lug is crimped to the cable and the connection end of
the lug is fastened to the bushing terminal by means of a nut (see Fig. 3.7).

The IEEE standards [14], applicable to single-phase distribution transformers 200 kVA and
smaller, having high voltage windings rated 8660 volts or less requires transformers with
additive polarity. All other single-phase transformers must have subtractive polarity. Therefore,
to have the additive polarity, windings are connected according to IEEE standard with series
connection [15]. The connection diagram is presented together with the nameplate in Figure. 3.8.

Figure 3.7. Connection of the LV leads to the bushing terminals.

3.10 Nameplate Information

The name plate information and connection diagram are provided according to IEEE
Standards [2], [15]. The nameplate of a 50 kVA transformer is shown in Fig3.8 as an example.
The nameplate information is engraved and filled black on a stainless still plate with 0.03"
thickness. The plate is 6"×3-3/8" with round corners. Four corner holes (1/16") are needed for
mounting.

Page 25 of 37
 Serial No. 20140422-001-00000003

kVA 50 Hz 60 Ph. 1 %Z 0.61

H.V. 7.62 kV L.V. 120/240 V H.V. Amps 6.56 L.V. Amps 208

H.V. BIL 95 kV L.V. BIL 30 kV H.V. °C Rise 79 L.V. °C Rise 75

Total Wt. 426 kg Conductor Copper Type Dry-type Class GA

H1
H2 Voltage Terminals

120 V X1-X2, X3-X2

A D
C B 240 V X1-X3

X3 X2 X1

(a)

(b)
Figure 3.8. Nameplate for the 50 kVA prototype transformer; (a) designed (b) picture.

Page 26 of 37
3.11 Serial Number

The serial number of the product consists of three parts with 8, 3, and 8 digits. The first part
indicates the manufacturing date (YYYY/MM/DD), the second part indicates the product type
identification number, and the third part indicates the product number for the specific type.
Figure 3.9 illustrates the meaning of the different digits in the serial number.
 

 
Figure 3.9. Serial number generation.

Page 27 of 37
 4. Value Propositions
 
Nowadays, the standard technology is the oil-immersed transformers. This is so because oil
has a better thermal conductivity and insulation properties than air. Therefore, transformers with
oil are built with higher efficiency than dry-type transformers. These transformers have several
disadvantages. During an overload, the windings produce heat and break down the oil into
nitrogen, hydrogen, and oxygen. This also happens in faulty transformers and may produce
serious explosions. The oil is usually mineral oil, and may also include fire resistance substances
such as PCBs and silicone which are toxic for human.

Today dry-type transformers are expensive. Utilities such as Con Edison should pay 7 times
the price of oil-immersed transformers for a dry-type transformer with similar specifications. The
insulating material (air/paper) is not usually a good thermal conductor. Therefore dry-type
transformers are not as efficient as oil-immersed transformers.

The designed and manufactured transformers object of this project are made with toroidal
cores with continuous still strips that are wound like a donut. The core does not have any air-
gaps, therefore iron core losses decrease to minimum. Also, because of the special geometry, the
resistance of the windings is reduced. This factors makes it very high efficient.

The final product is an ultra-high efficiency dry-type transformer. The efficiency satisfies
standards even with 150% overloads. The toroidal geometry makes the transformer smaller and
lighter. This transformer has almost the same price as standard transformers. However, the
operational costs are almost half of the oil-immersed transformers over the life time. This is
shown with comparison of the Capex and Opex in Figure 4.1. For example, for a typical medium
sized utility such as Orange and Rockland, with approximately 40,000 oil-filled transformers, if
only 10% of the current transformers are replaced, this utility will save $24M/20 years, which
means average savings of $1.2M/year in losses.
 

Page 28 of 37
 Current Standard $13571

Our Tech

$6862 $6900
$5530 $5167
$3444 $3489
$1373 $698 $1751

Acquisiton Cost Opex 2 years Opex 5 years Opex 10 years Opex 20 years
 
 
Figure 4.1. Capital and operational costs comparison between the standard transformers and the toroidal
transformer.
 
   

Page 29 of 37
 5. Products

The list of tangible products of this project is given in this section. We have produced two
full-size prototypes, published nine papers, applied for one patent, graduated three PhD students
and three MSc students, and supported five postdoctoral fellows. We obtained a grant from
Power Bridge NY in the amount of $149,985 from June 2014 to May 2015 to continue
developing the transformer with commercialization objectives.

5.1 Prototypes

25 kVA toroidal transformer 50 kVA toroidal transformer

Page 30 of 37
5.2 Papers

The following nine papers have been published in the IEEE Transactions:

1) P. Gómez, F. de León, and I. Hernández, “Impulse Response Analysis of Toroidal Core

Distribution Transformers for Dielectric Design”, IEEE Transactions on Power Delivery, Vol.
26, No. 2, April 2011, pp. 1231-1238.

2) I. Hernández, F. de León, and P. Gómez, “Design Formulas for the Leakage Inductance of
Toroidal Distribution Transformers”, IEEE Transactions on Power Delivery, Vol. 26, No. 4,
October 2011, pp. 2197-2204.

3) S. Purushothaman and F. de León, “Heat Transfer Model for Toroidal Transformers”, IEEE
Transactions on Power Delivery, Vol. 27, No. 2, April 2012, pp. 813-820.

4) F. de León, A. Farazmand, and P. Joseph, “Comparing the T and  Equivalent Circuits for the
Calculation of Transformer Inrush Currents”, IEEE Transactions on Power Delivery, Vol. 27,
No. 4, October 2012, pp. 2390-2398.

5) S. Jazebi, A. Farazmand, B. Murali, and F. de León, “A Comparative Study on pi and T

Equivalent Circuits for the Analysis of Transformer Ferroresonance”, IEEE Transactions on
Power Delivery, Vol. 28, No. 1, January 2013, pp. 526-528.

6) F. de León, S. Purushothaman, and L. Qaseer, “Leakage Inductance Design of Toroidal

Transformers by Sector Winding”, IEEE Transactions on Power Electronics, Vol. 29, No. 1,
January 2014, pp. 473-480.

7) F. de León, S. Jazebi, and A. Farazmand, “Accurate Measurement of the Air-Core Inductance of

Iron-Core Transformers with a Non-Ideal Low-Power Rectifier”, IEEE Transaction on Power
Delivery, Vol. 29, No. 1, February 2014, pp. 294-296.

8) A. Farazmand, F. de León, K. Zhang, and S. Jazebi, “Analysis, “Modeling and Simulation of the
Phase-Hop Condition in Transformers: The Largest Inrush Currents”, IEEE Transaction on
Power Delivery, Vol. 29, No. 4, August 2014, pp. 1918-1926.

9) S. Jazebi and F. de León, “Experimentally Validated Reversible Multi-Winding Transformer Model

for the Accurate Calculation of Low-Frequency Transients”, accepted for publication in the IEEE
Transactions on Power Delivery (available in early access).

5.3 Patent

F. de Leon, “Electrostatic Shielding for Transformers”, United States Patent Application, Serial No.
61/857,581; Provisional patent filed on July 23, 2013; Utility patent application filed on July 23, 2014.

Page 31 of 37
5.4 List of Students Supported from the Grant

M.Sc. Students:
Githanjali Venkataramani
Matthew Terracciano
Noel Augustine

Ph.D. Students:
Sujit Purushothaman
Ashkan Farazmand
Saeed Jazebi

Post Doctors:
Pablo Gomez
Layth Qaseer
Ashkan Farazmand
Ivan Hernandez
Saeed Jazebi
   

Page 32 of 37
 6. Conclusions

The research performed for this project has demonstrated that it is possible to design and
build utility-grade distribution transformers in toroidal cores. The gapless construction of
toroidal transformers brings important advantages over the traditional designs. It has been shown
that the higher efficiency of the toroidal construction makes possible to substitute oil-filled
transformers by dry-type transformers for pole mounted applications.

The technical challenges that were overcome in this project include: the development of the
technology to pass the impulse tests, derive a model for the thermal performance, produce a
sound mechanical design, and estimate the inrush current. However, the greatest challenges that
we faced during the development of the project were the complications of procuring the
necessary parts and materials to build the transformers. The actual manufacturing process is
relatively fast, but getting all parts together is a very lengthy process.

The following key technologies were developed in the course of this project: (1) to take full
advantage of the gapless core construction, an amorphous material was selected for the core.
These cores produce very little losses yielding an improved transformer thermal behavior; (2) the
magnetic core is electrically connected to the high-voltage terminal to produce an electrostatic
shield. Because of this, the insulation layers can be made very thin, which again improves the
heat dissipation characteristics of the transformer; (3) the high-voltage winding is wound inside
and thus the low-voltage winding is outside. Thus, very little heat is trapped in the multi-layer
insulation system needed for the high-voltage winding. With exception of technology (3), these
technologies are not currently used in standard transformers. In fact, technologies (1) and (2) are
not applicable in standard designs.

The main products of this project are two prototypes of toroidal distribution transformers of
7.62 kV (to be used in a 13.2 kV system) to 2x120 V secondary (standard utilization voltage);
one is rated at 25 kVA and the other at 50 kVA. The 25 kVA transformer passed the impulse test
in KEMA high-voltage laboratories. Additional products include: nine papers published in the

Page 33 of 37
IEEE Transactions on Power Delivery, one patent has been filed, three PhD students were
supported from beginning to graduation, five postdoctoral fellows, and three MSc students were
partially supported.

When the manufacturing process is prepared for mass production, the cost of a dry-type
toroidal transformer would be similar to the price of an oil-filed standard design. However,
because of the greatly reduced losses, the total ownership cost of a toroidal transformer could be
about half of a traditional design.

We obtained a grant from Power Bridge NY in the amount of $149,985 from June 2014 to
May 2015 to continue developing the transformer with commercialization objectives.

Page 34 of 37
   

Page 35 of 37
 7. References

[1] Energy Conservation Program for Commercial Equipment: Distribution Transformers Energy Conservation
Standards, Department of Energy Final Rule, 10 CFR Part 431.
[2] IEEE Standard General Requirements for Dry-Type Distribution and Power Transformers, Including Those
with Solid-Cast and/or Resin Encapsulated Windings, IEEE Std. C57.12.01, May 2006.
[3] S. Purushothaman and F. de León, “Heat Transfer Model for Toroidal Transformers”, Transactions on Power
Delivery, vol. 27, no. 2, April 2012, pp. 813-820.
[4] P. Gómez, F. de León, and I. Hernández, “Impulse Response Analysis of Toroidal Core Distribution
Transformers for Dielectric Design”, IEEE Transactions on Power Delivery, vol. 26, no. 2, April 2011, pp.
1231-1238.
[5] F. de Leon, “Electrostatic Shielding for Transformers”, U.S. Patents, 2013.
[6] Machinery's Handbook, 21st Edition, 1982, Industrial Press Inc.
[7] Industrial and Mining Grade Cables, TFcables.
[8] F. de León, S. Purushothaman, and L. Qaseer, “Leakage Inductance Design of Toroidal Transformers by
Sector Winding”, IEEE Transactions on Power Electronics, vol. 29, no. 1, January 2014, pp. 473-480.
[9] Hernández, F. de León, and P. Gómez, “Design Formulas for the Leakage Inductance of Toroidal Distribution
Transformers”, IEEE Transactions on Power Delivery, vol. 26, no. 4, October 2011, pp. 2197-2204.
[10] UltraSIL polymer-housed Evolution (10 kA) IEEE surge arresters for MV systems to 36 kV, Technical Data
235-99, Cooper Power Systems.
[11] IEEE Guide for the Application of Metal-Oxide Surge Arresters for Alternating-Current Systems, IEEE Std.
C62.22, July 2009.
[12] IEEE Standard General Requirements and Test Procedure for Power Apparatus Bushings, C57.19.00, Jun
2005.
[13] Characteristics and dimensions for outdoor apparatus bushings, IEEE C57.19.01, Jan 2000.
[14] IEEE Standard for Overhead-Type Distribution Transformers 500 kVA and Smaller: High Voltage, 34 500 V
and Below; Low Voltage, 7970/13 800Y V and Below, IEEE Std. C57.12.20, Sep. 2011.
[15] IEEE Standard for Standard Terminal Markings and Connections for Distribution and Power Transformers,
IEEE Std. C57.12.70, Feb. 2012.
   

Page 36 of 37
 

8. Appendices

The manuscript versions of the nine papers published as part of this US Department of Energy
award DE-OE0000072 follow as an appendix to this report.

Page 37 of 37
 1

Impulse Response Analysis of Toroidal Core

Distribution Transformers for Dielectric Design
Pablo Gómez, Member, IEEE, Francisco de León, Senior Member, IEEE, Iván A. Hernández, Student
Member, IEEE


Abstract—Toroidal transformers are currently used only in
low voltage applications. There is no published experience for
toroidal transformer design at distribution level voltages. This
paper explores the lightning impulse response of toroidal
distribution transformers in order to obtain a dielectric design
able to withstand standardized impulse tests. Three-dimensional
finite element simulations are performed to determine the
capacitance matrix on a turn-to-turn basis. Then, a lumped
parameter RLC model is applied to predict the transient
response of the winding, as well as to obtain the potential
distribution along the winding and corresponding dielectric
stresses. The model computes the impulse potential distribution
and the dynamic (inter-turn and inter-layer) dielectric stresses.
Different insulation design strategies are proposed by means of
electrostatic shielding and variation of the inter-layer insulation.

Index Terms— Distribution Transformers, Electrostatic

Analysis, Finite Element Method, Impulse Test, Insulation
Design, Toroidal Transformers, Transient Analysis.
Fig. 1. Toroidal transformer (only a few turns of one winding are shown)
I. INTRODUCTION

T HERE are two basic arrangements for the iron-cores used

at present to build distribution transformers: (1) Core-
type, where the cores are assembled by stacking laminations
savings in raw materials (iron and copper) for the same losses
than a standard design and even the tank is smaller.
This work is part of a project supported by the US
and sliding pre-made coils; and (2) Shell-type, where a Department of Energy aimed to benefit from the toroidal
continuously wound core is cut and wrapped around the coils construction virtues to construct and install toroidal
a few laminations at a time [1], [2]. In both arrangements the transformers suitable for power distribution application. Given
finished core has air gaps that increase the magnetizing current the lack of experience with this type of design at medium and
and the no-load losses. high voltages, studies including electromagnetic, thermal and
An alternative construction, currently used for low voltage mechanical analysis are required to understand its particular
applications and explored in this paper for distribution level physical behavior. This paper is part of a series describing
voltages, is to use a core made of a continuous steel strip such studies via computational design, optimization and
shaped like a doughnut (toroid) with the coils wound around verification, building prototypes, performance verification and
[3]; see Fig. 1. This gapless construction allows for the observation of prototypes installed on a utility distribution
construction of smaller, more efficient, lighter, and cooler system.
transformers [4], [5]. The no-load losses are substantially This paper is focused on analyzing the lightning impulse
reduced. There are also savings in the load losses because the response of a toroidal distribution transformer in order to
windings have fewer turns since these transformers can be obtain a dielectric design able to withstand standardized
designed with a larger flux density. Therefore, there are impulse tests. This is done by means of three-dimensional
finite element simulations, as well as electromagnetic transient
This work is supported by the U.S. Department of Energy under Grant
DEOE0000072
simulations considering a lumped parameter RLC (turn-by-
P. Gómez is with the Electrical Engineering Department of SEPI-ESIME turn) model of the transformer winding. Such computational
Zacatenco, Instituto Politécnico Nacional (IPN), U. P. "Adolfo López tools, which have been extensively used for electromagnetic
Mateos", Edificio Z-4 Primer piso C. P. 07738, Mexico, D. F. MEX (e-mail:
pgomezz@ipn.mx).
transient analysis of conventional transformer arrangements
F. de León is with the Department of Electrical & Computer Engineering, (see for instance [6]-[11]), are applied in this paper for toroidal
Polytechnic Institute of New York University, Six MetroTech Center, distribution transformers for the first time.
Brooklyn, NY 11201, New York, USA (e-mail: fdeleon@poly.edu).
I. Hernandez is with CINVESTAV Guadalajara, 45015 Jalisco, México
(email: ihernand@gdl.cinvestav.mx
 2

Another contribution of this paper is the application of

Internal
electrostatic shielding in the design of the insulation system of winding
toroidal transformers. Two insulation design strategies are
proposed in the paper and their effectiveness in reducing the
transient voltage and dielectric stress in the winding is
demonstrated. The first one is the addition of an electrostatic
shield uniformly spaced with respect to the winding. The
second one is the use of an electrostatic shield that has a
varying distance to the winding, by means of a gradual
increase of insulation thickness between winding and shield
(without affecting the winding positions). The two strategies
are equally successful to properly distribute the impulse surge.
The selection between them depends on manufacturer
efficiencies and preferences
The dynamic performance of the toroidal transformer Fig.2. Geometry and meshing for FEM simulations (distances between
insulation system for lightning impulse is studied by means of layers were exaggerated for illustration purposes)
two examples, one transformer of 25 kVA and another one of
50 kVA. Both transformers have the same ratings in terms of Mutual capacitance Cij is computed from the electrostatic
voltage ratio (13.8/0.120 kV) and BIL (95 kV). energy Wij obtained when applying voltage at both turns i and
j:
II. ELECTROSTATIC ANALYSIS
Given the complex geometry of the windings in a toroidal Wij 
1
2
1

Cij ViV j  CiiVi  C jjV j
2
 (2)
transformer, a 3D arrangement is required for the electrostatic
analysis, as shown in Fig. 2. In this paper the internal (low- Self capacitances must be calculated first from (1), in order
voltage winding, which is grounded) is represented by a solid to obtain the mutual elements from (2). Mutual capacitances
toroidal shape since its detailed representation is not needed. between non-adjacent turns or layers are not considered since
Note that the transformer core is not visible. For the purposes FEM simulations have shown that, for the arrangements under
of this paper each turn of the high-voltage winding is modeled study, their values are at least one order of magnitude smaller
as a closed loop, then the mutual capacitances can be obtained than the values between adjacent turns. Transient simulations
from the energy method. in which capacitive values for all turns (including non-
Assuming that the high voltage winding has N layers and n adjacent) were included confirmed that they have no effect on
turns per layer, the following capacitive values need to be the results for the geometrical configuration under analysis.
computed: An important issue when finding the solution of such a
Cs,o Self capacitance of any turn at the outer layer (N) detailed geometry lies in the finite element meshing.
Cs,i Self capacitance of any turn at the inner layer (1) Considering the thin insulation between turns produces very
Cs,m Self capacitance of any turn at any interior layer (2, narrow regions. This is particularly true at the internal part of
… N-1) the winding. Therefore, a very large number of elements (in
Cit,o Mutual capacitance between any two adjacent turns at the order of millions) is required to obtain an accurate
the outer layer (N) solution.
Cit,i Mutual capacitance between any two adjacent turns at Taking advantage of the toroidal symmetry to speed up the
the inner layer (i) simulations and consume less memory, the geometry can be
Cit,m Mutual capacitance between any two adjacent turns at simplified by considering only a section of the actual number
any interior layer (2, … N-1) of turns and layers. For the example shown in Fig.2, three
CiL,o Mutual capacitance between the i-th turn at the outer layers and nine turns per layer are found sufficient to
layer and the i-th turn at the following interior layer approximate the capacitance values of a real arrangement of
CiL,m Mutual capacitance between the i-th turns of any 2 11 layers with 214 turns per layer. This has been validated by
interior layers initial simulations in which the results from the complete
These elements are computed by means of FEM geometry are compared to those of the simplified one.
simulations using the electrostatic energy method [12]. Self Each electrostatic simulation for the calculation of the
capacitances are computed from the electrostatic energy Wi capacitive matrix takes about 12 minutes in a powerful
obtained when applying a voltage Vi to the ith turn of the computer (two Xeon multi-core processors running at
winding: 2.27 GHz with 72 GB RAM).
It can be observed in Fig. 2 that, in contrast to shell or core
1 type transformers, the distance between turns in a toroidal
Wi  Cii Vi2 (1)
2 configuration is not constant. While the distance between turns
at the internal part of the toroid is kept at the minimum
required to avoid dielectric breakdown, the distance at the
 3

external part is several times larger resulting in small

turns
capacitive coupling between turns (series capacitance). Thus,
the well-known distribution constant   Cground / Cseries is
several times larger for toroidal transformers than that for
conventional constructions. This particularity of toroidal
transformers produces highly non-uniform initial potential
distribution (at the wave front), giving rise to large dielectric
stresses as well as increased transient overvoltages. This
makes necessary the use of electrostatic shielding.

layers
III. TRANSIENT ANALYSIS
Fast and very fast front transients in transformers are
commonly analyzed using internal models, which can take
into account the distribution of the incident surge along the
windings. These models are described either by distributed
parameters, using the transmission line theory [13], [14], or as
a ladder connection of lumped parameter segments [6], [15]. Fig.3. Circuital representation of the winding. Mutual inductances between
The latter models can be solved by network analysis or by turns and between layers, as well as ground capacitances of outer layers,
are omitted in the figure for the sake of simplicity.
integrating the corresponding state-space equations.
Additionally, an admittance matrix model (black box
model) based on terminal measurements has been presented in V() and I() correspond to the vectors of nodal voltages and
[16] and [17]. This model can be implemented in time domain currents; Y() is the nodal admittance matrix, which is
simulation programs by means of a rational approximation defined as follows
procedure.
Y( )  G  j C  Γ  G con (7)
For the size of a distribution toroidal transformer and the
frequency range involved in the lightning waveform, a turn of Matrix Gcon contains the conductance elements required for
the transformer can be considered as electrically short. the topological connection of layers, as well as the source and
Therefore, a lumped parameter model considering a winding ground connections (if needed);  is the nodal matrix of
turn as the basic element is chosen in the present work. inverse impedance, computed from Z = R + jL and the
This Section describes the lumped parameter model used in incidence matrix K (since Z is a branch matrix):
this paper to obtain the transient response of the winding. It is
based in [6], and considers a lossy and frequency dependent Γ  K Z 1 K t (8)
multilayer winding.
where
After computing the winding capacitance matrix C, the
geometric inductance matrix is obtained as 1 0 0  00
 1 1 0  0
0
1
L  0  C (3) 
 0 1 1  0 0
In (3),  is the permittivity of the surrounding medium. K 
Conductor losses due to skin and proximity effects can be      
0 0 0  1 0
computed from the following expression [18]:  
 0 0 0   1 1
(9)
1 2
R L (4)
d  c c
Finally, the time domain response of the winding is
In (4), d is the distance between layers,  is the angular obtained by solving (6) for V and applying the inverse
frequency, c is the conductivity of the winding conductor and numerical Laplace transform [19], [20].
µc is its permeability. On the other hand, dielectric losses can Maximum dielectric stresses (DS) between turns and
be included in the form of a shunt conductance matrix given between layers can be obtained from the elements of the nodal
by voltages vector V and the minimum distance between
corresponding turns as:
G   tan   C (5)
Vi  V j
where tan is the loss tangent of the winding insulation. From 
max DS ij    
min d ij
(10)
matrices R, L and C and G a nodal system can be defined to
describe the winding (Fig. 3):
I( )  Y( )V( ) (6)
 4

IV. ELECTROSTATIC SHIELDING

There are three essential methods to improve the impulse Cseries
response of power transformers: electrostatic shielding,
addition of dummy strands, and interleaving of turns [1]. The
Cground
latter method is in general preferred for transformers working
at high-voltage transmission levels. However, for a toroidal
transformer working at distribution level voltage with a large
(a)
turns ratio (e.g. 13.8/0.120kV), the winding arrangement (by
layers) and the small cross sectional area of the winding shield
conductors makes it cumbersome and ineffective to attempt
any interleaving or addition of dummy strands. Cseries Cshield
Hence, electrostatic shielding is chosen for toroidal
distribution transformers. Its basic idea is to improve the
initial potential distribution by compensating the current Cground
drained by the ground capacitances with currents injected to
the series capacitances [20]. This is illustrated in Fig. 4. The
shield is connected to the winding terminal and therefore it (b)
needs to be isolated from the turns and the tank along its Fig. 4. Initial current distribution along the winding: (a) original, (b) with
electrostatic shield.
length. Also, the shield should not form a closed path; a gap
between the shield ends is necessary.
An electrostatic shield, inverted C-shaped, is proposed for
the toroidal transformer constructed by means of a thin
conductor material covered by an insulation layer and partially
wrapped around the winding. The internal part of the winding
remains unshielded (unwrapped) since the turns are close
enough to each other in this region; see Fig. 5. Additionally,
note that the size (and therefore the cost) of the toroidal
transformer is very much dependent of the minimum internal
diameter needed for the winding machine. Therefore, not
shielding the center is convenient.
The distance between the shield and the winding is of Fig. 5. Axisymmetric view of the toroidal transformer with an inverted C-
particular importance. The shield has to be close enough to the shape electrostatic shield.
winding to be effective and far enough from the winding to
avoid dielectric breakdown. This is analyzed for the test case 1st layer is wound in the clockwise direction, then the 2nd layer
presented the next Section. is wound in the counterclockwise direction and so forth. This
winding strategy yields reduced dielectric stresses when
V. TEST CASES compared with continuous (same direction) windings.
Two toroidal transformers with rating of 25 and 50 kVA The transient response of the transformers is analyzed by
are considered. Voltage ratio and BIL rating are the same for means of the injection of a standard 1.2/50µs lightning
both: 13.8/0.120 kV and 95 kV. The main geometrical data of impulse (full wave) at the initial terminal of the winding,
the high-voltage windings of these two transformers is listed which is located at the outermost layer of the winding. The
in Table I. The following assumptions are made for simulation lumped parameter model shown in Fig. 3 is constructed and
purposes: solved as described in Section III.
a) The number of turns is considered equal for all
TABLE I
layers; in an actual transformer each outer layer has MAIN GEOMETRICAL DATA OF THE TRANSFORMERS UNDER STUDY
fewer turns than the previous one. Rating [kVA] 25 50
b) Due to the previous assumption, turns from each External diameter of the core [mm] 510 600
layer are considered completely aligned, as shown
Internal diameter of the core [mm] 250 250
in Fig 2.
Conductor gauge [AWG] 11 7
c) The minimum distance between turns is given by
Conductor diameter [mm] 2.3048 3.6648
the typical thickness of the varnish film for the
corresponding conductor diameter [22]. Distance between layers [mm] 1.0762 1.0940
d) The distance between layers is initially assumed to Distance between windings [mm] 1.0000 1.0000
be 1 mm (plus the conductor varnish). Distance between winding and core [mm] 1.0000 1.0000
The set of capacitive values obtained from FEM for both Minimum distance between turns [mm] 0.0762 0.0940
transformers is listed on the Appendix. An alternating Number of layers 11 12
direction of the winding between layers is proposed, i.e., if the Number of turns per layer 214 108
 5

The performance of the shield is improved by a winding. It can be seen in the plots how the stress is reduced
configuration equivalent to gradually removing the shield by applying the different shields. The maximum value of
from the winding, which helps to approximate a uniform inter-turn stress in the 25kVA and the 50kVA transformers is
potential distribution. This is possible by: (a) decreasing the reduced by 57.2% and 56.1%, respectively, with the uniform
shield surface or (b) increasing the distance between the shield located 1 mm from the winding. On the other hand,
winding and the shield. However, option (a) implies a constant these stresses are reduced by 65.4% and 55.6% with the
distance between the shield and the winding, which could varying shield. It can also be noticed that, even without any
result in dielectric breakdown given that the initial potential shield, the stress is kept at an acceptable level. The maximum
along the winding drops rapidly while the potential in the value obtained for both transformers is well below the
shield remains almost constant. dielectric strength of any high performance varnish [17].
After substantial simulation tests, three alternatives of Therefore, no extra insulation needs to be added between
electrostatic shielding are deemed to be practical: two shields turns.
with constant distances of 1 and 2 mm to the outer layer of the The inter-layer stress is plotted in Fig. 10. The inter-layer
winding, as well as a shield with a varying distance to the stresses are several times larger than the inter-turn stresses.
outer layer, from 0.1mm to 1 mm. The latter shield is included The potential difference between turns of consecutive layers
by means of a gradually increasing the insulation thickness can be very large, particularly at the layers’ ends
between winding and shield. (corresponding to the peaks in Fig. 9). The stress is especially
Fig. 6 shows the initial potential distribution along the large between the first two layers for both transformers under
windings. As expected, the potential distribution without analysis. However, the values obtained with or without the
shield (continuous line) is highly non-uniform for both shield are below the dielectric strength of a varnish included
transformers. Additionally, some spikes can be seen, which as reference (56 MV/m) [23].
are a consequence of the capacitive coupling between layers at One can see from Fig. 10 that the shields produce reduced
the layers’ ends. This distribution can be improved by inter-layer stresses when compared to the unshielded case. The
including an electrostatic shield in the transformer design. reduction (in percent) of the stress at each inter-layer when
The way in which the different shields affect the initial applying the shields is shown in Table II. It can be noticed that
potential distribution is shown in Fig. 6. By producing a more the reduction is slightly larger when applying the varying
uniform distribution, the voltage drop between consecutive shield. Furthermore, the shields produce an increase (by a
turns along the winding is reduced. small percentage) in the stress between layers 1 and 2 for the
Fig. 7 shows the transient response of the winding at turn 50kVA transformer. This does not present a problem since the
107 for the 25 kVA transformer and at turn 52 for the 50 kVA stress is still below the dielectric strength of the varnish
transformer, corresponding to the regions of maximum voltage considered.
stress. One can appreciate that the shield is able to damp the From Figs. 8, 9 and 10, it seems that the best two options
transient oscillations reducing the maximum transient are: (a) use a uniform shield spaced 1 mm from the winding
voltages. Additionally, as expected, the closer the shield is to or, (b) use a shield with a varying distance to the winding,
the winding, the larger the mitigation of the overvoltage. from 0.1 to 1 mm. Both strategies keep the transient voltage
However, this distance is limited by the dielectric strength of below the BIL, while the inter-turn and inter-layer stresses
the insulation between winding and shield. The results for the have acceptable levels.
uniform shield distanced 1 mm to the winding and the varying
shield are almost identical for both transformers. TABLE II
REDUCTION OF THE INTER-LAYER STRESS WITH APPLICATION OF
Fig. 8 illustrates the distribution of the maximum voltage ELECTROSTATIC SHIELDING
obtained along the winding for the whole transient period, Dielectric stress reduction (%)
hereafter called impulse potential distribution. The voltage Inter-layer Uniform shield Varying shield
distribution along the whole winding of the different shielded 25 kVA 50 kVA 25 kVA 50 kVA
transformers is more uniform compared to the unshielded 1-2 12.0 -3.9* 17.0 -5.1*
transformers. The performance of the varying shield in the 2-3, 3-4 22.3 9.2 23.9 11.2
context of mitigating the transient voltage is very similar to 4-5, 5-6 21.5 25.7 25.1 28.4
that of the uniform shield separated 1 mm from the winding. 6-7, 7-8 16.3 16.3 19.3 18.3
With these two shielding strategies, the maximum value of 8-9, 9-10 13.5 13.7 16.0 15.8
transient voltage is reduced by 21.8% for the 25 kVA 10-11, 11-12 14.6 14.1 17.0 15.9
transformer, and by 11.3% for the 50 kVA transformer, with HV-LV 14.5 10.2 17.4 16.6
*Negative values correspond to increase in stress
respect to the unshielded case.
The dielectric performance of the winding is analyzed
considering three main variables:
a) Inter-turn dielectric stress
b) Inter-layer dielectric stress
c) Winding-to-shield dielectric stress
Fig. 9 shows the inter-turn stress along the complete
 6

(a)

(b)
(b)
Fig.8. Impulse potential distribution: (a) 25 kVA transformer, (b) 50 kVA
Fig.6. Initial potential distribution: (a) 25 kVA transformer, (b) 50 kVA
transformer
transformer

(a)

(b)
Fig.7. Transient response at the turn of max. voltage stress: (a) 25 kVA
transformer, turn 107, (b) 50 kVA transformer, turn 52 Fig.9. Inter-turn dielectric stress: (a) 25 kVA transformer, (b) 50 kVA
transformer
 7

The performance of the shields in terms of the dielectric

stress between the shield itself and the winding is shown in
Fig. 11. While the uniform shield presents a growing behavior
of the stress along the outer layer of the winding, this stress
tends to be constant for the varying shield. This means that, if
the insulation between the winding and the shield is too thin,
there is a possibility of dielectric breakdown at the end of the
layer when a uniform shield is applied. However, the
manufacturing process to include the varying shield is more
complicated. Consequently, the uniform shield placed at the
correct distance (1 mm for the cases analyzed) can be a better
option. All transient voltages and stresses (between turns,
(a) layers and to the shield) are kept at acceptable levels without
requiring of the cumbersome manufacturing of a varying
distance of shield to the winding.

VI. CONCLUSIONS
The dynamic impulse response of a toroidal distribution
transformer has been presented in this paper. By means of
electrostatic 3D-FEM simulations the turn-by-turn capacitance
matrix of the winding has been computed. Transient
simulations on a lumped parameter model of the winding are
used to design the insulation. In contrast to conventional
transformers, the distance between turns in a toroidal core
transformer is not constant. The larger distance between turns
(b) at the external region of the toroidal core yields a smaller
Fig.10. Inter-layer dielectric stress: (a) 25 kVA transformer, (b) 50 kVA series capacitance compared with traditional designs
transformer producing a very non-uniform initial potential distribution.
This posses stringent design constraints since the non-uniform
potential distribution gives rise to large transient voltages and
dielectric stresses. To overcome this issue, three electrostatic
shielding configurations have been proposed: two uniform
shields with different distance to the winding and a shield with
a linearly increasing distance to the winding. From the results
of the simulations performed, the following conclusions are
obtained:
1. Inter-turn stress is low for the whole winding. A
typical insulation film corresponding to its AWG
size and a dielectric strength above 12 MV/m is
shown to be adequate for the tested cases.
2. Inter-layer stress is the critical factor for this type
(a) of transformers. The distance between layers has to
be carefully selected to avoid inter-layer
breakdown.
3. The inclusion of a shield at 1 mm from the winding
or a shield with a varying distance to the winding
(from 0.1 to 1 mm) results in lower inter-turn and
inter-layer stress, as well as damped transient
voltages.
4. When a uniform shield is considered, the distance
between shield and winding has to be carefully
selected in order to achieve the largest possible
reduction in dielectric stress and transient voltage
while avoiding dielectric breakdown between
(b) shield and winding.
Fig.11. Winding-to-shield dielectric stress: (a) 25 kVA transformer, (b) 5. Proposed in this paper is a shield with a varying
50 kVA transformer
distance to the winding, which prevents dielectric
 8

breakdown between winding and shield. electromagnetic transients,” IEEE Trans. Power Delivery, vol. 13, pp.
834–842, July 1998.
[17] B. Gustavsen, “Wide band modeling of power transformers,” IEEE
VII. APPENDIX Trans. Power Delivery, vol. 19, no. 1, pp. 414–422, January 2004.
[18] K. J. Cornick, B. Filliat, C. Kieny, and W. Muller, “Distribution of
CAPACITIVE VALUES FOR THE 25 KVA AND 50 KVA very fast transient overvoltages in transformer windings,” CIGRE
TRANSFORMERS WITHOUT SHIELDING Report 1992, pp. 12–204.
[19] P. Gómez and F. A. Uribe, “The numerical Laplace transform: an
Value (pF) accurate tool for analyzing electromagnetic transients on power system
Capacitance* devices,” Int. Journal of Electrical Power & Energy Systems, Vol. 31,
25 kVA 50 kVA No. 2-3, pp. 116-123, Feb.-Mar. 2009.
Cs,o 71.71 104.32 [20] J. Wilcox, “Numerical Laplace Transformation and Inversion,” Int. J.
Cs,i 56.67 84.23 Elect. Enging. Educ., Vol 15, pp. 247-265, 1978.
[21] P. Chowdhuri, Electromagnetic Transients in Power Systems. Research
Cs,m 63.20 88.70 Studies Press Limited/ John Wiley & Sons, Inc, 1996, pp. 348-351.
Cit,o 25.78 35.23 [22] ANSI/NEMA MW1000-2003 Standard for Magnet Wire.
[23] B. Górnicka and L. Górecki, “Method of assessment of varnishes
Cit,m 15.48 16.44 modified with nanofillers,” Materials Science-Poland, Vol. 27, No.
CiL,o 13.43 24.76 4/2, 2009.
CiL,m 12.74 23.24
Cit,i 10.45 10.90 IX. BIOGRAPHIES

*Refer to Section II for the corresponding nomenclature Pablo Gómez (S’01, M’07) was born in Zapopan, México, in 1978. He
received the B.Sc. degree in Mechanical and Electrical Engineering from
Universidad Autonoma de Coahuila, Mexico, in 1999 . He received the M.Sc.
VIII. REFERENCES and Ph.D. degrees in Electrical Engineering from CINVESTAV, Guadalajara,
[1] S. V. Kulkarni and S. A. Khaparde, Transformer Engineering: design Mexico in 2002 and 2005, respectively. Since 2005, he is a full-time professor
and practice. CRC Press 2004, pp. 36-38. with the Electrical Engineering Department of SEPI-ESIME Zacatenco,
[2] M. Heathcote, J & P Transformer Book. Newnes, 2007, pp. 13-15. National Polytechnic Institute, Mexico City, Mexico. from 2008 to 2010, he
[3] I. M. Gottlieb, Practical Transformer Handbook. Newnes, 1998, pp. was on a postdoctoral leave at Polytechnic Institute of New York University,
12-15. Brooklyn, New York, USA. His research interests are in the modeling and
[4] G. J. Taggart and J. D. Goff, “Optimizing linear power supply simulation for electromagnetic transient analysis and electromagnetic
performance with line frequency toroidal transformers,” NorthCon ’94, compatibility.
Seattle, WA , USA, 11-13 Oct. 1994, pp. 67-71.
[5] F. de León, B. Gladstone, and M. van der Veen, “Transformer Based
Solutions to Power Quality Problems,” Power Systems World Francisco de León (S’86–M’92–SM’02) received the B.Sc. and the M.Sc.
Conference and Exhibit, Rosemont IL, USA, September 2001, pp 303- (Hons.) degrees in electrical engineering from the National Polytechnic
314. Institute, Mexico City, Mexico, in 1983 and 1986, respectively, and the Ph.D.
[6] P. I. Fergestad and T. Henriksen, “Transient oscillations in degree from the University of Toronto, Toronto, ON, Canada, in 1992. He has
multiwinding transformers,” IEEE Trans. Power App. Syst., vol. PAS- held several academic positions in Mexico and has worked for the Canadian
93, pp. 500–509, Mar./Apr. 1974. electric industry. Currently, he is an Associate Professor at the Polytechnic
[7] Z. Azzouz, A. Foggia, L. Pierrat, and G. Meunier, “3D Finite Element Institute of NYU, Brooklyn, NY. His research interests include the analysis of
Computation of the High Frequency Parameters of Power Transformer power definitions under nonsinusoidal conditions, the transient and steady-
Windings,” IEEE Trans. on Magnetics, vol. 29, no. 2, pp. 1407-1410, state analyses of power systems, the thermal rating of cables and transformers,
March 1993 and the calculation of electromagnetic fields applied to machine design and
[8] E. Bjerkan, H. K. Høidalen, “High Frequency FEM-based Power modeling.
Transformer Modeling: Investigation of Internal Stresses due to
Network-Initiated Overvoltages”, International Conference on Power
Systems Transients (IPST’05), Montreal, Canada, June 19-23, 2005.
Iván Hernández (S’06) was born in Salamanca, Guanajuato, Mexico in 1979.
[9] G. Liang, H. Sun, X. Zhang, and X. Cui, “Modeling of transformer
He received the B.Sc. in electrical Engineering from the University of
windings under very fast transient overvoltages,” IEEE Trans. on
Guanajuato (Mexico) in 2002, and the M.Sc. degree in Electrical Engineering
Electromagnetic Compatibility, 48(4), November 2006.
from the CINVESTAV Guadalajara (Mexico) in 2005. He is a Ph.D. student
[10] F. D. Torre, A. P. Morando, and G. Todeschini, “Three-Phase
in CINVESTAV Guadalajara. From 2008 to 2010 he was a visiting researcher
Distributed Model of High-Voltage Windings to Study Internal Steep-
at Polytechnic Institute of New York University. He was an electrical engineer
Fronted Surge Propagation in a Straightforward Transformer,” IEEE
designer for two years in FMS Ingenieria, Guadalajara, Mexico. His research
Trans. on Power Delivery, Vol. 23, No. 4, pp. 2050-2057, Oct. 2008.
interests are the numerical analysis applied to machine design and software
[11] X. M. Lopez-Fernandez and C. Alvarez-Mariño, “Computation Method
simulation tools applied to electromagnetic fields.
for Transients in Power Transformers With Lossy Windings,” IEEE
Trans. on Magnetics, Vol. 45, No. 3, pp. 1863-1866, March 2009.
[12] Comsol Multiphysics, AC/DC User’s Guide, Comsol AB Group, 2006,
pp. 1-156.
[13] Y. Shibuya, S. Fujita, and N. Hosokawa, “Analysis of very fast
transient overvoltage in transformer winding”, IEE Proc. Generation
Transmission and Distribution, Vol. 144, No. 5, September 1997.
[14] M. Popov, L. V. Sluis, and G. C. Paap, “Computation of very fast
transient overvoltages in transformer windings,” IEEE Trans. Power
Delivery, vol. 18, no. 4, pp. 1268–1274, Oct. 2003.
[15] R.C. Degeneff, W.J. McNutt, W. Neugebauer, J. Panek, M.E.
McCallum, C.C. Honey: “Transformer Response to System Switching
Voltage”, IEEE Transactions on Power Apparatus and Systems,
Volume PAS-101, No.6, June 1982, pp. 1457-1470.
[16] B. Gustavsen and A. Semlyen, “Application of vector fitting to the
state equation representation of transformers for simulation of
 1

Design Formulas for the Leakage Inductance of

Toroidal Distribution Transformers
Iván Hernández, Student Member, IEEE, Francisco de León, Senior Member, IEEE, and Pablo Gómez,
Member, IEEE

Abstract—In this paper design formulas for the calculation of

the leakage inductance of toroidal transformers are presented.
The formulas are obtained from the analytical integration of the
stored energy. The formulas are sufficiently simple and accurate
to be introduced in the loop of a design program avoiding
expensive finite element simulations. It is found that toroidal
transformers naturally produce the minimum leakage inductance
possible for medium-voltage power transformers. To limit the
short-circuit currents in power and distribution systems, a larger
than the minimum leakage inductance is often required. This
paper presents two methodologies to increase the leakage
inductance of toroidal distribution transformers: selectively Fig. 1. Photo of Faraday’s original transformer [1]
enlarging the inter-winding spacing and inserting a piece of
ferromagnetic material in the leakage flux region between the
windings. Extensive validation with 2D and 3D finite element
simulations is performed. Additionally, experimental verification
of both formulas and numerical simulations was carried out
comparing the calculations against measurements on prototypes.

Index Terms — Toroidal Transformers, Leakage Inductance,

Finite Element Method.

I. INTRODUCTION

F ARADAY in 1831 built the first transformer in a toroidal

core [1]; see Fig. 1. The first industrial grade transformer,
the one of the Ganz factory in Budapest of 1885, was also
Fig. 2. Drawing of the Ganz factory transformer [2]
wound on a toroidal core [2]; see Fig. 2. Currently, however,
toroidal transformers are not widely used for transmission and
distribution of bulk power. There are two basic arrangements applications; see [9]-[11]. There are no papers published
used to build the iron-cores of medium and large transformers related to mid- or high-voltage toroidal transformer intended
[3]-[6]: (1) Core-type where the cores are assembled by for use at utility voltages. Transformers wound on non-
stacking laminations and sliding pre-made windings, and; (2) gapped toroidal cores using grain oriented silicon steel are
Shell-type where a continuously wound core is cut and more efficient, smaller, cooler, and emit reduced acoustic and
wrapped around the windings a few laminations at a time. In electromagnetic noise when compared with standard
both arrangements the finished core has air gaps that increase transformer constructions. To extrapolate these advantages to
the magnetizing current and the no-load losses. distribution transformers, an effort is being made now, as part
Toroidal transformers have found modern applications in of a US Department of Energy funded project, to produce
the low-voltage low-power of many power supplies for toroidal transformers suitable for power distribution system
electronic equipment, avionics, and audio systems [7], [8]. A applications. Although toroidal transformers have many
very limited amount of published material exists in the IEEE advantages over the traditional constructions, there are also a
related to toroidal transformers for power conversion few disadvantages that need to be overcome before
widespread adoption of toroidal transformers is possible. Most
This work was supported by the U.S. Department of Energy under Grant importantly, there is no published experience in the industry
DEOE0000072. designing and building toroidal transformers suitable to
I. Hernández is with CINVESTAV Guadalajara, 45015 Jalisco, México
(email: ihernand@gdl.cinvestav.mx
operate at medium and high voltage. Unresolved issues with
F. de León is with the Department of Electrical and Computer toroidal transformer design and manufacturing include:
Engineering of Polytechnic Institute of New York University, Six Metrotech matching the leakage impedance specification, limiting inrush
Center, Brooklyn, NY, 11201 (email: fdeleon@poly.edu). currents, designing and constructing to withstand short-circuit
P. Gómez is with the Electrical Engineering Department of SEPI-ESIME
Zacatenco, Instituto Politécnico Nacional (IPN), U. P. "Adolfo López currents, the study of electromagnetic transients (impulse test),
Mateos", Edificio Z-4 Primer piso C. P. 07738, Mexico, D. F. MEX (e-mail: design for cost optimization, and the ability to pass industry
pgomezz@ipn.mx). standard acceptance tests. This paper is part of a series
 2

describing the solutions to those issues via electromagnetic

design, design verification, building prototypes, performance
verification and observation of prototypes installed on a utility
distribution system. In low-voltage, low-power applications
the leakage inductance can be minimized using planar
transformers or highly interleaved windings. For high-power,
medium-voltage transformers, the leakage inductance of
toroids is the minimum achievable. The reason for this is the
closed concentric geometry. The first winding completely
covers the core and subsequent windings cover the internal
windings. There are no yokes where the flux could escape to
the air. Therefore, the electromagnetic coupling is maximized,
while the leakage and stray fields are minimized. The small (a)
regulation characteristic that can be obtained with toroidal
transformers by minimizing the leakage impedance is
desirable for many applications. However, in a power system
the transformers’ leakage impedance is one of the important
components limiting the short-circuit currents. Consequently,
a larger than natural leakage inductance may be required for a
toroidal transformer.
A contribution of this paper is to propose two methods to
increase the leakage inductance of toroidal transformers: (1)
Enlarging the spacing between primary and secondary
windings, and; (2) Inserting high permeability materials
between primary and secondary windings.
Another contribution of this paper is the derivation of (b)
equations suitable for implementation in a design program for
the calculation of the leakage inductance of toroidal
transformers. The final expressions are numerically very
efficient and sufficiently accurate for practical design work.
Validation against a large number of finite element
simulations in 2D and 3D covering distribution transformers
of 25, 37.5, 50 and 75 kVA was performed.

II. DISTRIBUTION OF THE LEAKAGE FIELD

Coherent with the standardized method to measure the
leakage inductance, for its computation one must simulate the
short circuit test. In other words, force N1I1 = N2I2, eliminating
the magnetizing current. Fig. 3(a) shows an axisymmetric (c)
view of the distribution of the magnetic field strength in a
toroidal transformer during a short circuit test. Five distinct
sections having different field distribution characteristics can
be identified:
1. Vertical internal part of the windings
2. Vertical external part of the windings
3. Top and bottom horizontal parts
4. Internal corners
5. External corners
One can distinguish three sub-regions: two corresponding
to the two windings and one for the insulation between them
in each of the five regions. Fig. 3(b) shows the magnetic field
strength on the vertical part of the windings along the line (d)
A-A’. One can see that the magnetic flux in the vertical Fig. 3. Distribution of the magnetic field strength in the toroidal transformer:
direction almost follows the trapezoidal distribution (a) Axisymmetric view; (b) Radial distribution of the magnetic field on the
vertical sections; (c) Magnetic field strength on the horizontal sections at
characteristic of traditional transformer designs. Additionally, three positions; (d) Radial variation of the field at the insulation of the
note that the magnetic field strength is independent of the horizontal parts.
vertical position.
 3

The top and bottom sections, regions 3 of Fig. 3(a), have

identical magnetic field distributions as shown in Fig. 3(c).
Note, however, that while the vertical variation of the field
follows the trapezoidal distribution, the field strength reduces
in inverse proportion with distance to the axis; see Fig. 3(d).
The leakage inductance of the toroidal transformer can be
obtained through closed form volumetric integration of the
distribution of the magnetic energy stored as follows:
2 W stored 2 H B 0 2
Lleak 
I
2

I V 2 2  I V
dV  dV
2 
(1) H

It is noticed that the different components of the leakage

inductance can be obtained by analyzing the distribution of the
magnetic field strength H at each section. Two main
assumptions are made regarding the distribution of the
magnetic field strength:
 The radial distribution (around the toroidal circumference)
is considered constant (axisymmetric model).
 The distribution of H transversal to the windings is
considered as follows: it rises linearly in one winding,
varies inversely with x in the insulation between windings
and decays linearly in the opposite winding. This type of
distribution can be described by the following expression:
 x
 H peak   , 0 xa
 a Fig. 4. Main geometrical data of a toroidal distribution transformer.
H   H ( x), a  x ag (2)
   NI NI NI
 H peak 1  x a g 
, a  g  x  a  g  b H i ,1  ; H gi ( x )  ; H i ,2  ; (4a)
  b  2 R2 2 ( R2  x ) 2 R3
where Hpeak is the maximum value of the magnetic field NI NI NI
H e,1  ; H ge ( x)  ; H e,2  ; (4b)
strength; in this paper Hpeak is identified in five ways 2 R6 2 ( R6  x) 2 R7
depending on the section being considered: Hi, He (internal
and external vertical sections of the winding, respectively); where N is the number of turns of the exciting winding, I is the
Hgi(x), Hge(x) correspond to the internal and external spaces current; R2 and R6 are the internal radii of the insulation for the
between the windings (i.e. insulation); and Hh(x) (horizontal vertical regions 1 and 2, respectively; and R3 and R7 are the
sections of the winding); while a, b and g correspond to the internal radii of the external winding for regions 1 and 2. The
thickness of the high-voltage (HV) winding, low-voltage (LV) reduction of the magnetic field strength between the windings,
winding and inter-winding insulation, respectively (as from Hgi to Hge as 1/x is considered. When the insulation
indicated in Fig. 4). between windings is small, we can assume that H has a
trapezoidal distribution. In [12] we have computed that 1 mm
III. DESIGN FORMULAS FOR THE LEAKAGE INDUCTANCE of insulation between windings is enough to produce
From the identification of the five different sections, the transformers class 95 kV BIL.
total leakage inductance of the winding can be computed as: Combining (1) and (4), the leakage inductance of section 1
is computed from:
2  0 h 
2 ag
Lleak  Lleak,1  Lleak,2  2 Lleak,3  2 Lleak,4  2 Lleak,5
a
(3) x 2
Lleak ,1  2
 Rm1  H i ,1   dx  Rm 2  H gi dx
I  0
a
  a

where Lleak,i correspond to the leakage inductance component (5)

a  g b
 xag 
2

of the i-th section of the winding (for i = 1, 2, 3, 4, 5).  Rm 3  H i ,2  1 
b  dx 
Expressions for each section are obtained below (using the ag   
Cartesian coordinate system). where h is the height of the toroid, Rm1, Rm2 and Rm3
correspond to mean radii of the HV winding, insulation and
A. Vertical Parts (Sections 1 and 2) LV winding, respectively, and computed in general as:
In sections 1 and 2 (internal and external vertical parts of 
Rm j  Rj  Rj 1 / 2  (6)
the winding, respectively), the peak values of H (Hi, He, Hgi,
Substituting (4a) into (5) and performing the integral one gets:
Hge) are shown in Fig.4. These peaks can be computed from
Ampere’s Law as follows:
 4

N 2 0 h  Rm1  a  Rm 2  g  Rm 3  b  t3  3b  4s (15c)
Lleak ,1         (7) Similarly, the leakage inductance for the external corners is
2  R22  3  R2 R3  2  R32  3 
computed as:
The leakage inductance for section 2 is computed in a
N 2 0  Rm 5t3b Rm 7 t1a 
similar manner as: Lleak ,5    Rm 6 t2 g   (16)
4 R62  6 6 
N 2 0 h  Rm 5  a  Rm 6  g  Rm 7  b 
Lleak ,2         (8)
2  R62  3  R6 R7  2  R72  3  D. Generalized Expression
One can appreciate that (7), (8), (12), (14) and (16) have a
B. Horizontal Parts (Sections 3) similar form. Therefore, a generalized expression for the
The top and bottom parts have the same field distribution; calculation of the contribution to the leakage inductance of
see Fig 3(c). The value of H at the inter-winding insulation is each section can be obtained as follows:
computed from Ampere’s Law as follows:
N 2 0
NI Lleak ,i  i i a  i g  i b  (17)
H h ( x)  (9) 2
2 x
The coefficients for the different sections are given in Table
The radial distance on x-axis can take values from Ri  x  Re, I. The total leakage inductance is computed from (3).
where Ri and Re are the internal and external radii of the
toroid, respectively. Thus, the leakage inductance of the TABLE I
horizontal sections is obtained from: COEFFICIENTS FOR THE DIFFERENT COMPONENTS OF THE LEAKAGE
INDUCTANCE FORMULA (17)
2 0 Rmh  e  b
R 2 b g
 y Coefficient
R  0 h  b  b H h dx
2
Lleak ,3    H dx 
I2 Section ηi i i i
 i 
(10) Rm 1 Rm 2 Rm 3
b g a
 y b g 
2
  1 h
  H h 1   dx  dy 
3R22 2 R2 R3 3R32
 a 
b g   Rm 7 Rm 6 Rm 5
Rmh is the mean radius of the horizontal sections, given by: 2 h
3R72 2 R6 R7 3R62
Rmh   Ri  Re  / 2 (11)
Re2  Ri2 1 1
Substituting (9) in (10), performing the integral, and using 3 1
Re Ri 3 3
(11) we get:

Lleak ,3 
 
N 2  0 Re2  Ri2  b a
 g  (12) 4
1
2 R22
Rm1t1
Rm 2 t 2
Rm 3 t 3

2 Re Ri 6 6
3 3
1 Rm 7 t1 Rm 5 t 3
5 Rm 6 t 2
2 R62 6 6
C. Corners (Sections 4 and 5)
For the corners, the same peak values for the magnetic field IV. TEST CASES
defined for the internal Hi and external He vertical parts are
considered as given by (4a) and (4b). The trapezoidal Table II shows the design parameters of a set of toroidal
distribution of H is around the corner, so it was necessary to distribution transformers used to demonstrate the applicability
of the methods and the accuracy of the formulas. We have
perform the integral around its periphery denoted by φ (from 0
selected the standardized sizes for distribution transformers as
to π/2); the leakage inductance for the internal corners is
per [13]. The leakage inductance reference values have been
obtained from the following expression:
computed with 3-D finite element simulations using the
2 0  
 /2 2 ag
a
x 2 commercially available software (COMSOL Multiphysics)
Lleak ,4  2
I 0 
   Rm1  H i ,1 
a
 xdx  R m 2  H gi xdx [14].
  0 a
The FEM simulations performed solve for the magneto-
(13)
a  g b 2
  static formulation. All materials are considered as being
 xa g 
 Rm 3  H i ,2  1   xdx  d 
isotropic; we used copper windings and electrical steel M4
ag  b    (0.28 mm) for the main core considering its B-H curve as
provided by the manufacturer.
Solving (13), it follows that:
In the simulations the toroid was enclosed by a tank
N 2 0  Rm1t1a Rm 3t3b  represented by a rectangle in the axisymmetric 2-D case and
Lleak ,4  2 
 Rm 2 t2 g   (14)
4 R2  6 6  by a cylinder in the 3-D case. Magnetic insulation was applied
where: to the boundaries of the tank walls. For the 2-D simulations
about 40,000 triangular elements were necessary consuming
t1  3a  4  s  b  g  (15a)
about 2 GB on RAM. For the 3-D simulations about 400,000
t2  g  2  s  b  (15b) tetrahedrons were employed consuming 9 GB on RAM. The
 5

axisymmetric 2-D and 3-D simulation results were almost V. METHODOLOGIES FOR INCREASING THE LEAKAGE
identical. Therefore, we conclude, as expected from a INDUCTANCE OF TOROIDAL TRANSFORMERS
symmetrical construction, that to compute the leakage A. Increasing Inter-Winding Spacing
inductance 2-D axisymmetric modeling is sufficient.
Table III shows the values of leakage inductances and One can perceive from Tables III and IV that the leakage
reactances in percent that can be achieved with toroidal inductance of a 25 kVA toroidal transformer may be as small
transformers. The inductive values are referred to the HV as half of what is specified in the standard [12].
winding. From Table III one can appreciate that the results are From the expressions obtained in Section III, and their
in good agreement, with maximum differences of 3%. analogy with the technology of traditional transformer
Table IV shows the leakage impedance values constructions, it can be inferred that increasing the spacing
recommended by the IEEE Standard 242-1986 [13] for the between windings will increase the leakage inductance. This is
calculation of short-circuit currents. It can be noticed that the a technique known to designers and manufacturers of
reactance in percent of toroidal transformers may be traditional transformer constructions. It is possible to identify
substantially smaller than that of conventional transformers. in (7), (8), (12), (14) and (16) the middle term as the
Therefore, larger short-circuit currents can be expected. inductance corresponding to leakage flux in the insulation (or
Although small regulation is in general a desirable air). To build toroidal transformers the internal space at the
characteristic for a transformer, for some applications the center of the toroid must be large enough for the winding
larger short-circuit currents that occur may not be acceptable. machine to pass. Therefore, only the top, bottom and external
In the next section two methods to increase the leakage regions can be used in practice to increase the leakage path.
inductance are proposed. Furthermore, when considering manufacturing aspects the
most suitable region to increase the inter-winding space is the
external part (region 2 of Fig, 3(a)). Therefore, in this paper,
TABLE II only the external inter-winding space of the toroidal
DESIGN PARAMETERS FOR SINGLE PHASE TOROIDAL TRANSFORMERS transformer is used to increase the leakage inductance; see
25 kVA 37.5 kVA 50 kVA 75 kVA Fig. 5. Taking this into consideration, the leakage inductance
HV-(kV) 13.80 13.80 13.80 13.80 for the vertical external component of the winding (region 2),
LV-(kV) 0.12 0.12 0.12 0.12 given by (8), is modified as follows:
Bm (T) 1.70 1.70 1.70 1.70
f (Hz) 60.00 60.00 60.00 60.00
Np 4715 4370 4370 3335  Rm5  b 
Ri (mm) 100.00 101.50 110.50 121.50  2   
Re (mm) 180.00 185.50 194.50 217.50  R6  3 
h (mm) 80.00 84.00 84.00 96.00 
N 0 h 
2 
1  g 
a (mm) 10.24 12.64 17.39 17.90 
L leak ,2   Rm6   g   g  
2  2R6 (R7  g)   2 
b (mm) 10.41 10.41 11.68 20.81
(18)
g (mm) 1.00 1.00 1.00 1.00  
s (mm) 0.50 0.50 0.50 0.50  Rm7  g  a 
Bm is the magnetic flux density average in the core, f is the operation  R2  3 
frequency, Np is the number of turns of the HV side.  7  

where g is the increased space in the inter-winding region.

TABLE III The leakage inductance corresponding to the horizontal
PARAMETERS COMPUTED FOR SINGLE-PHASE TOROIDAL TRANSFORMERS components of the winding (regions 3 and 4), given by (12), is
Leakage Leakage also modified, resulting in the following expression:
Inductance (H) Reactance (%) %
FEM Formula FEM Formula Diff
N 2 0  Re  g   Ri2   b
2
25 kVA 0.1050 0.1079 0.5198 0.5339 2.76
  a
37.5 kVA 0.1011 0.1041 0.7508 0.7724 2.99 L  leak ,3   g  (19)
50 kVA 0.1200 0.1236 1.1879 1.2234 3.01 2  Re  g  Ri 3 3
75 kVA 0.1086 0.1114 1.6121 1.6544 2.86
The computed values were referred to the HV winding.
Fig. 6(a) shows the variation of the leakage inductance with
the inter-winding space for the four transformer ratings under
study. One can appreciate that increasing the inter-winding
TABLE IV spacing increases the leakage inductance by a relative modest
IMPEDANCE DATA FOR SINGLE PHASE TRANSFORMERS FROM [13] amount. The values have been normalized with respect to the
kVA Suggested Normal Range minimum inter-winding space needed for insulation purposes
1-phase X/R Ratio for of Percent
Calculation Impedance (% Z)
(1 mm).
25.0 1.1 1.2-6.0 The results from the formulas of this paper against FEM are
37.5 1.4 1.2-6.5 compared in Fig. 6(b) for the transformer 25 kVA. One can
50.0 1.6 1.2-6.4 appreciate a very good match between the formulas and FEM
75.0 1.8 1.2-6.6
(differences of about 4%).
 6

Fig. 5. Enlarging the external vertical inter-winding space to augment the

leakage inductance.
(a)
As a conclusion of this section one can observe that the
technique of increasing inter-winding spacing is effective
when relatively small increments of the leakage inductance are
needed. However, when large increments are sought, a
different technique is necessary. Furthermore, adding larger
spaces than required for insulation purposes adds cost and
weight to the transformer. The most significant negative
consequence is that the external winding has a longer mean
length (adding production cost and operation losses).

B. Ferromagnetic Inserts
The second technique proposed in this paper to increase the
leakage inductance is to augment the permeability of the
material in the leakage region. By inserting a ferromagnetic
material between the windings we can dramatically magnify (b)
Fig. 6. Variation of the leakage inductance: (a) Calculated for 4 different
the leakage inductance without a noticeable increase in the
ratings of toroidal distribution transformers, (b) Comparison of the analytical
transformer size. results with FEM for a 25 kVA toroidal transformer.
The underlying idea is to install a thin core in the inter-
winding region on the external face; see Fig. 7. This produces
an enlargement of the leakage inductance component
corresponding to such region. Equation (8) is modified as:

 Rm5  Rm5 b 
 2   
 R6  3 
N 20 h  1   gc 

L leak,2   m6R   g   g  
2  2R6 (R7  gc )   2 
(20)
r c

 
 Rm7  gc  a 
 R2  3 
 7  
where gc is the thickness of the region occupied by the
ferromagnetic material and µr is its relative permeability. The
leakage inductance for the horizontal components of the
winding is modified in a similar fashion as (12), yielding:

N 2 0  Re  g c   Ri2   b
2

  a Fig. 7. Illustration of adding ferromagnetic inserts between windings to

Lleak ,3  3g 3 (21)
2  Re  g c  Ri   increase the leakage inductance.
 7

By adding a material with high relative permeability (r) VI. EXPERIMENTAL VERIFICATION
the value of the leakage inductance can be magnified by a With the purpose of validating the formulas proposed in
large factor. When using this technique care must be taken to this paper and the FEM simulations, a set of prototypes were
avoid saturation of the thin core placed between the windings. built with ratings of 150 VA, 300 VA, 1 kVA, 2 kVA, and 4
Different ferromagnetic materials [15] were considered for kVA. The leakage inductance was measured applying two
the simulations performed to validate this technique. Fig. 8(a) methods: using the standardized short circuit (SC) test and
shows the variation of leakage inductance with thickness for using an RLC meter (7600 Precision LCR meter) available in
materials with different permeability. The plot is given in per the lab. This meter uses an ac signal of 2 V at 60 Hz and it
unit (p.u.) normalized to the minimum insulation space and gives the equivalent series R-L circuit of the transformer
permeability of air 0. A comparison between the results of the directly. In all cases the secondary windings of the
formulation and FEM is shown in Fig. 8(b). One can notice transformers are shorted and the primary windings are
that the differences are very small. connected to the source.
Table V shows the comparison of the measurements on the
five prototypes against finite elements simulations and the
formulas of this paper. One can appreciate that, for most cases,
the results are very close between the four different methods
(SC, RLC meter, FEM, and formulas). The differences are in
general under than 3%. The sole exception is the SC
measurement of the 300 VA double-core transformer with
8.47% difference. This transformer was opened and unwound.
We found that the external (powder) core was fractured.
Therefore, the effective permeability of this core was reduced
by the irregular (unintended) air-gap explaining why the
measurements gave a slightly smaller leakage inductance
when compared with FEM and the formulas.
These experiments not only corroborate the accuracy of the
calculation method proposed in the paper, but also confirm the
applicability of ferromagnetic inserts to increase the leakage
inductance when large leakage is necessary.
(a)

TABLE V
LEAKAGE INDUCTANCE MEASURED AND COMPUTED FOR SINGLE-PHASE
TOROIDAL TRANSFORMERS
Leakage Leakage %
Inductance Inductance Diff
Transformer Measured (mH) Computed (mH) SC versus
SC RLC FEM Formula Formula
150 VA 0.7200 0.7350 0.6890 0.7095 1.46
1 kVA 0.2150 0.2300 0.2100 0.2092 2.68
2 kVA 0.0493 0.0491 0.0490 0.0503 2.03
4 kVA 0.0209 0.0220 0.0205 0.0205 1.53
*300 VA 13.644 ** 15.100 14.800 8.47
* Transformer with double core (as in Fig. 7)
** Not possible to measure with the RLC meter

VII. CONCLUSIONS
Formulas suitable for a design program for the calculation
(b)
of the leakage inductance of toroidal transformers have been
Fig. 8. Increase of the leakage inductance. (a) Inserting 4 different developed. From the observation of the distribution of the
ferromagnetic materials between the windings; (b) Comparison of results magnetic flux in the leakage region, precise expressions have
between formulas and FEM for the 25 kVA transformer.
been derived for the magnetic field strength. The leakage
inductance is obtained by the analytical integration of the total
energy stored in the magnetic field. The formulas have been
compared against 2D and 3D finite element simulations
yielding very good results; differences of under 4%.
Two methodologies to augment the leakage inductance of
toroidal transformers have been proposed. We have
 8

investigated increasing the inter-winding spacing and the X. BIOGRAPHIES

addition of a ferromagnetic core in the leakage region. Iván Hernández (S’06) was born in Salamanca, Guanajuato, Mexico in 1979.
Increasing the inter-winding spacing is effective for up to 1.5 He received the B.Sc. in electrical Engineering from the University of
Guanajuato (Mexico) in 2002, and the M.Sc. degree in Electrical Engineering
p.u. increment of the leakage inductance at the cost of from the CINVESTAV Guadalajara (Mexico) in 2005. He is a Ph.D. student
increasing the mean length of the external winding. The in CINVESTAV Guadalajara. From November 2008 to August 2010 he was
addition of a ferromagnetic core between the windings offers on a study leave at the Polytechnic Institute of New York University.
an inexpensive alternative to augment the leakage inductance. Previously, he was an electrical engineer designer for two years in FMS
Ingeniería, Guadalajara, Mexico. His research interests are the numerical
This technique can be used conveniently to increase the analysis applied to machine design and software simulation tools particularly
leakage inductance several orders of magnitude. for electromagnetic fields.
The accuracy of the formulas and the applicability of the
Francisco de León (S’86–M’92–SM’02) was born in Mexico City in 1959.
methods to increase the leakage inductance have been He received the B.Sc. and the M.Sc. degrees in Electrical Engineering from
corroborated experimentally for a set of prototypes of various the National Polytechnic Institute (Mexico), in 1983 and 1986 respectively, and
sizes. obtained his Ph.D. degree from the University of Toronto (Canada) in 1992.
He has held several academic positions in Mexico and has worked for the
Canadian electric industry. Currently he is an Associate Professor at the
VIII. ACKNOWLEDGEMENTS Polytechnic Institute of New York University, Brooklyn, NY. His research
interests include the analysis of power phenomena under nonsinusoidal
The authors would like to thank Mr. Ulrik Poulsen of conditions, the transient and steady-state analyses of power systems, the
Bridgeport Magnetics for his fast response and expertise thermal rating of cables, and the calculation of electromagnetic fields applied
building the prototypes. Also, we would like to thank Messrs. to machine design and modeling.
Chandira Prabhu and Noel Augustine, M.Sc. students of
Pablo Gómez (S’01, M’07) was born in Zapopan, México, in 1978. He
Polytechnic Institute of New York University, for performing received the B.Sc. degree in Mechanical and Electrical Engineering from
the leakage inductance tests to the prototypes. Additionally, Universidad Autonoma de Coahuila, Mexico, in 1999. He received the M.Sc.
we would like to thank the reviewers for their sharp comments and Ph.D. degrees in Electrical Engineering from CINVESTAV, Guadalajara,
Mexico in 2002 and 2005, respectively. Since 2005, he is a full-time professor
that have added value to the paper. with the Electrical Engineering Department of SEPI-ESIME Zacatenco,
National Polytechnic Institute, Mexico City, Mexico. From 2008 to 2010, he
was on a postdoctoral leave at the Polytechnic Institute of New York
IX. REFERENCES University, Brooklyn, New York, USA. His research interests are in the
modeling and simulation for electromagnetic transient analysis and
[1] F. A. Furfari and J. W. Coltman, “The Transformer”, IEEE Industry electromagnetic compatibility.
Applications Magazine, Jan./Feb. 2002, pp. 11-12.
[2] S. Jeszensky, “History of Transformers”, IEEE Power Engineering
Review, December 1996, pp. 9-12.
[3] J. H. Harlow, “Electric Power Transformer Engineering”, 2nd Ed., New
York, CRC Press, 2007.
[4] S. V. Kulkarni, S. A. Khaparde, “Transformer Engineering Design and
Practice”, New York, Marcel Dekker, Inc., 2004.
[5] R. M. Del Vecchio, B. Poulin, P. T. Feghali, D. M. Shah, R. Ahuja,
“Transformer Design Principles - With Application to Core-Form Power
Transformers”, Canada, Gordon and Breach Science Publishers, 2001.
[6] M. Heathcote, “J & P Transformer Book”, 12th Edn, Great Britain,
Butterworth-Heinemann Ltd., 1998.
[7] M. van der Veen, “Modern High-end Valve Amplifiers: Based on
Toroidal Output Transformers”, Elektor Electronics Publishing, 1999.
[8] A. A. Halacsy, “Reactance and Eddy Current Loss in Toroidal
Transformatoric Devices-II”, AIEE Transactions on Power App. and
Sys., vol. 81, No. 3, April 1962, pp. 1017 – 1019.
[9] R. Prieto, J. A. Cobos, V. Bataller, O. Garcia, and J. Uceda, “Study of
toroidal transformers by means of 2D approaches”, IEEE 28th Annual
Power Electronics Specialists Conference, St. Louis, MO, Jun. 22-27,
1997.
[10] R. Prieto, V. Bataller, J. A. Cobos, and J. Uceda, “Influence of the
winding strategy in toroidal transformers”, Proceedings of the 24th
Annual Conference of the IEEE Industrial Electronics Society, 1998,
IECON '98, 31 Aug-4 Sep 1998, Vol. 1, pp. 359 – 364.
[11] J. P. Myers, K. A. Weaver, W. R. Wieserman, and U. Poulsen, “O cores
- a new approach”, Proceedings of the Electrical Insulation Conference
and Electrical Manufacturing & Coil Winding Technology Conference,
2003, 23-25 Sept. 2003, pp. 193 – 198.
[12] P. Gómez, F. de León, and I. Hernández, “Impulse Response Analysis of
Toroidal Core Distribution Transformers for Dielectric Design”, paper
accepted for publication in the IEEE Transactions on Power Delivery.
[13] IEEE Recommended Practice for Protection and Coordination of
Industrial and Commercial Power Systems, IEEE Standard 242-1986,
Feb. 1986.
[14] Comsol Multiphysics, AC/DC User’s Guide, Comsol AB Group, 2006,
pp. 1-156.
[15] A. Goldman, Handbook of Modern Ferromagnetic Materials, vol. I.
Kluwer Academic Publishers Group, 1999, pp. 64-135.
 1

Heat Transfer Model for Toroidal Transformers

Sujit Purushothaman, Student Member, IEEE, and Francisco De León, Senior Member, IEEE

 els have also been proposed to determine the top oil tempera-
Abstract—Toroidal transformers provide increased design ture (TOT) and the hottest spot temperature (HST) [5], [6].
flexibility, efficiency and compact design when compared to tra- However, the application of toroidal transformers in power
ditional shell or core type transformers. In this paper the steady transmission and distribution at medium voltage is stunted.
state thermal analysis for toroidal transformers is conducted
This so not only because its construction could be more ex-
using a lumped parameter model which can be applied to small
power and distribution grade toroidal transformers as well. Two pensive than traditional designs, but perhaps due to the lack of
cases are considered: (1) when the transformer is kept in open air previously published research work.
and (2) when it is installed in sealed enclosures. The detailed The toroidal construction has many advantages over stand-
model includes the effects of number of turns of windings, num- ard power transformers, for example: The lack of an air-gap in
ber of layers, insulation properties and geometric properties of the toroidal core allows for a higher design flux density. The
the transformer. The model is capable of finding the hot-spots
closed geometry (where the second winding completely covers
that are of paramount importance for the designer. The model
parameters are calculated from the design (geometrical) infor- the first) produces a transformer with a smaller leakage in-
mation, therefore it is suitable to be included in the design loop of ductance than that of traditional designs producing transform-
transformer design software. Results are compared with finite ers with small regulation. Additionally, the acoustic noise and
element simulations and lab tests on prototypes of various power electromagnetic emissions are smaller. In an effort to forward
ratings fitted with thermocouples to record internal tempera- the advantages of toroidal transformers to distribution sys-
tures. The model can also be used with varied external media and
tems, the US Department of Energy has funded a project to
encapsulation, such as: air, oil, and epoxy.
design and develop toroidal transformers for medium voltage
Index Terms—Toroidal Transformers, Thermal rating, Heat distribution systems. This paper is part of a series of papers
Transfer, Equivalent thermal circuit, Finite element method. describing solutions to problems related to the design and con-
struction of these utility grade transformers. Equations to ac-
I. NOMENCLATURE curately compute the leakage impedance have been obtained
HV: High voltage and verified experimentally in [7]. The insulation design based
LV: Low voltage on the propagation of the impulse wave in windings on toroi-
HST: Hottest Spot Temperature dal core has been presented in [8].
Qloss: Total Ohmic loss in transformer [W] The power rating of a transformer is limited by the tempera-
h: Heat transfer coefficient [W/m2.K] ture of the hot-spots. This paper describes a procedure to set-
k: Thermal conductivity [W/m.K] up a thermal model of toroidal transformers. The model can
Nu: Nusslet number predict accurately the temperature of each winding layer along
Gr: Grashoff’s number four directions. This model can be used to study temperature
Pr: Prandtl number distribution for transformers used in medium voltage distribu-
Ra: Rayleigh number tion systems.
Results from the model are compared with finite element
simulations yielding a good match. The proposed model was
II. INTRODUCTION also validated with a set of prototypes (of various power rat-
ings) especially built with thermocouples placed at strategic
T HE first transformer was built by Faraday in 1831 on a
toroidal core [1]. Nowadays, toroidal transformers are
mostly being used in power supplies for avionics, audio sys-
locations within the transformer. The model has proven to be
sufficiently accurate and efficient for practical implementation
in a design program.
tems and electronic equipment rated for low voltages and rela-
tively low power [2]-[3]. Transformers used in bulk power
III. GEOMETRIC ARRANGEMENT
transmission are of core type or shell type construction. Over
the years considerable research has been done on thermal Traditional core type or shell type transformers consist of
modeling of oil immersed transformers. Equivalent electrical uniform windings around the core. This makes it easy to per-
circuits with non-linear resistors have been used to model the form thermal studies using lumped parameter circuits [4-6].
air or oil convection currents [4] in transformers. Many mod- Fig. 1 shows the uneven winding distribution due to the geom-
etry of the toroidal core. The core has unequal surface area on
This work was supported in part by the U.S. Department of Energy under the inside and the outside surface because of the smaller radius
Grant DEOE0000072. (Perimeter = 2 r). Hence the conductor spacing is more on
S. Purushothaman and F. de León and are with Polytechnic Institute of the outside than on the inside. Since the cross sectional area of
NYU, Brooklyn, NY 11201 (e-mail: sujitp@ieee.org, fdeleon@poly.edu).
 2

Fig. 1. Top view of toroidal transformer showing winding distribution.

the conductor remains the same everywhere, the conductors’

bundle is thicker on the inside and thinner on the outside. This
Fig. 2. 2D axial symmetric geometry of toroidal transformer
is also applicable to any insulation wrapped between layers.
Since a large temperature gradient exists within an insulator,
this non-uniform distribution of insulation thickness on the thermal problem whereas that in the top and bottom regions is
inside and outside is of critical importance to this study. Cartesian.
The unequal surface areas and the non-uniform distribution The resistance of the top and bottom insulation layers can
of windings on toroidal core leads to a complex analysis ex- be computed as [12]
plained below in Section V. t
The transformer under study consists of a toriodal core, Rinsul  (1)
covered by a layer of insulation. The LV windings with n lay- kA
ers are wound first on the insulated core followed by the m where t is the thickness and A is the surface area of the insula-
layers of the HV windings. The insulation class requirements tion layer. The resistance of the inner and outer insulation (cy-
may cause insulation layers to be added between layers of the lindrical) layers can be computed as

 
HV windings. This reduces thermal performance and hence an 1 OD
accurate thermal modeling is a crucial step in the design a Rinsul  ln (2)
distribution grade toroidal transformer. 2  k H insul ID
where OD and ID are the outer and inner diameters of the in-
IV. EQUIVALENT THERMAL CIRCUIT sulation layer respectively. Hinsul is the height of the layer.
The thermal-electric analogy for the analysis of heat trans- The heat loss in the nth layer, Qln is equal to total heat loss
fer phenomena is well known and a good explanation can be in the winding, Q times the ratio of the number of turns in the
found in [12], and [14]. The core and both the windings are present layer Nln to the total number of turns in the winding,
metallic materials (steel and copper) and hence offer high N.
thermal conductivity, k. The windings carry current and pro- Nl
Qln  Q n (3)
duce heat due to Ohmic losses. The eddy current losses and N
hysteresis losses constitute the core losses. Therefore, in the Assuming the heat dissipated in the conductor per unit length
electrical equivalent circuit the windings and the core are to be constant, Qln times the ratio of fraction of the length of a
modeled as current sources. The insulation is essentially made turn lx in the respective direction x to the length a turn in the
of several layers of thin Mylar wound tape, having low electri- nth layer Ln gives the current sources in each directional cir-
cal and thermal conductivity and so the insulation layers are cuit.
modeled as thermal resistors in the circuit. (Qln ) x  Qln x
l
(4)
Fig. 2 shows the thermal equivalent circuit super-imposed Ln
on the axial slice geometry of a typical toroidal transformer. The temperature of the surrounding medium is modeled as
The uneven distribution of windings causes uneven tempera- an ideal voltage source since it is assumed that the ambient
ture field around the core. Therefore an equivalent thermal temperature would not be affected by the presence of the
circuit is proposed for each of the four directions; namely, top, transformer under consideration. The thermal resistance of the
outer, bottom and inner directions. The detailed equivalent surrounding medium is highly dependent on the physical state
thermal circuit in the outer direction for the toroidal trans- of the medium (solid (epoxy), liquid (oil) or gas (air)). This
former is as shown in Fig. 3. The circuits for the other three resistance can be nonlinear and its computation is complex as
directions are similar with different parameter values. All cir- described in the next section.
cuits are connected at the core (center) producing a cross- Since the model deals with steady state calculations only,
shaped equivalent circuit. no capacitances are needed in the thermal equivalent circuit.
The heat flow in the inner and outer regions is a cylindrical The algorithm for computing the results is given in Appendix
II.
 3

Fig. 3. 2D The detailed thermal circuit in the outer direction for toroidal transformer.

V. THERMAL RESISTANCE OF SURROUNDING MEDIUM

 = kinematic viscosity
C p = specific heat capacity at constant pressure
Fluid medium (air, oil) dissipate heat by convection and ra-
diation. Convection is the phenomenon of heat transfer by  = dynamic viscosity.
conduction in moving media. Radiation is medium independ- The convective thermal resistance Rconv is given by
ent and accounts for 20-30% of the total heat flux, hence 1
should not be neglected. Rconv  (12)
This section explains the computation of heat transfer coef- hconv A
ficient hconv for natural convection in the laminar regime. hconv The radiative heat transfer coefficient hrad is calculated as
is a function of the geometric arrangement, temperature and [12], [13]
properties of convective medium of the surface under consid-
hrad   (Ts  T )(Ts  T )
2 2
(13)
eration. hconv is given by [12], [13]
Nu k where  is the emissivity of the surface with area A, dissipating
hconv  (5) radiative heat flux. Ts and T are the surface temperature and
L
where k is the thermal conductivity of the surrounding medi- temperature of the ambient surroundings respectively and  is
um; L is the characteristic length and Nu is the Nusselt num- the Stefan-Boltzmann constant.
ber. In general the relationship between Nu and the Rayleigh The radiative thermal resistance Rrad is given by
number RaL is given by (6) and depends on the orientation of 1
the surface under consideration [11] Rrad  (14)
hrad A
Nu  C Ra L (6)
m

Since convection and radiation occur simultaneously at the

For the top surface [11] surface, the total thermal resistance of the medium is the paral-
Ra L  10  10 C  0.54, m  0.25
4 7
lel combination of Rconv and Rrad.
(7) In the case of encapsulation of the transformer, the materi-
Ra L  10  10 C  0.15, m  0.334
7 9
al (epoxy resin) may be treated as a solid insulation medium.
and L is the ratio of surface area to perimeter. For bottom sur- Hence Rconv would be substituted by (1) and (2) since now the
face, C = 0.27, m = 0.25 and L is the length. For vertical faces, heat transfer is done by conduction.
L is the length and
0.25 VI. FINITE ELEMENT MODEL
 Pr RaL 
Nu  0.671    (8) A 2-D axisymmetric model of the prototype was built in
 Pr  0.492  (0.986  Pr )  COMSOL Multiphysics [10] simulating open air conditions.
The expression for RaL is given as follows [12], [13] The model shown in Fig. 4 simulates a toroidal transformer
RaL  GrL  Pr (9) suspended in air to ensure all surfaces contribute to heat dissi-
where GrL is the Grashoff’s number and Pr is the Prandtl pation.
number, given by: The following set of nonlinear equations are solved simul-
taneously with the finite elements method (FEM): The Navier-
g  (Ts  T ) L
3

GrL  (10) Stokes equation [12]


2
u
   (u  )u  p   2 u   g (15)
Cp t
Pr  (11)
k the continuity equation
g = acceleration due to gravity 
   (u)  0 (16)
 = volumetric thermal expansion coefficient t
 4

Fig. 5. Schematic for lab tests performed on the prototypes

Fig. 4. FEM model of toroidal transformer under test.

and the energy equation

T
C p   C p u   T    ( k T )  Q (17)
t
T is temperature at a point, u is the velocity field,  is the den-
sity, p is the pressure, and Q is the heat generated.
Fig. 4 shows the FEM surface plot for temperature, T along
with an arrow plot showing fluid velocity u. The lower bound-
ary of the model is fixed at 1 atm pressure (p) and temperature
(T) at 22⁰C. The pressure at the top boundary also fixed at 1
atm pressure. It is also defined as a heat sink. The vertical sur-
face is defined as heat insulation (zero temperature gradient).
The vertical surface and transformer surfaces are defined as
no-slip (u = 0). Internal boundaries representing inter-layer
insulation are defined as thermal resistive boundaries with Fig. 6. Placement of thermocouples inside the prototype.
thickness calculated as described in Section III.
It must be noted that the solution of this highly non-linear depicted in this figure are used later in section VIII to compare
problem of natural convection is time consuming even in the results. The transformer was fed from a constant voltage
most powerful PC computers (dual Intel Xeon processors run- source while the load power was kept constant. The input
ning at 3.33 GHz with 96 gigabyte RAM) available today power, output power and the temperatures were recorded at
(2011). The FEM model has 4099 2nd degree elements and can intervals of 30 minutes until steady state was achieved as per
take a couple of hours for a solution. Such a solution cannot be IEEE C57 [10]: i.e. the temperature does not change more
included in a design program which may take several itera- than one degree in one hour. The difference in input and out-
tions to obtain the final design. Hence the need for a simple put power is the power dissipated in the transformer and is
and accurate model is required which can yield the tempera- responsible for the temperature rise.
ture distribution within the transformer without being compu- Estimation of power loss in individual windings and layers
tationally intensive. accurately is essential for this model. The mean length per turn
times the total number of turns per layer would yield total re-
VII. LABORATORY TESTS sistance per layer. From this the I2R loss can be calculated
precisely.
Lab tests were conducted on toroidal core prototypes of
various power ratings to verify the proposed model. Fig. 5
VIII. MODEL RESULTS
presents the test schematic. The prototype (detailed in Appen-
dix I) was suspended in air to simulate the conditions as given This section summarizes the results from the equivalent cir-
in section VI. A Yokogawa PZ-4000 Power Analyzer was cuit, the finite element simulations, and the lab test on the pro-
used to measure input and output power of the prototype. A totype. As shown in Fig. 6 horizontal cross section AA’ is
National Instruments, NI-USB 9213 Thermocouple measure- defined as a line starting from the inner surface to the external
surface. Section AA’ is used to plot temperatures in the inner
ment unit was used to record temperatures on a computer. The
windings, core and outer windings. Fig. 7 presents this com-
prototype has six thermocouples placed at locations internal to
parison. Similarly, cross-section BB’ is defined from the bot-
the transformer as per Fig. 6. Cross-sections AA’ and BB’
tom to the upper surface and is used to plot for bottom wind-
 5

Fig. 7. Comparison of temperature distribution along section AA’

Fig. 9. FEM model of toroidal transformer enclosed in sealed enclosure

under test.

The effective tank resistance, Rtank can be calculated as

1 1 1 1
   (1.1)
Rtank Rtop Rvertical Rbottom
where Rtop, Rbottom, Rvertical are the thermal resistances of the
Fig. 8. Comparison of temperature distribution along section BB’.
top, bottom and vertical surfaces of the tank to the external
medium. They can be evaluated from the equations (6)-(8).
ings, core and top windings. Fig. 8 presents the results along All equations presented in Section V are applicable for the
this section. computation of temperature distribution within the transformer
It is observed that the equivalent circuit behaves relatively kept in an enclosure with the exception of the ambient temper-
accurately with the FEM model for the inner, outer and bottom ature. This must be equated to the temperature of enclosure
windings. The differences in the outer winding are larger. The Ttank, since the transformer is not exposed to the surroundings
prototype test results are also within practical error limits of at ambient temperature.
the model (max. differences of 4%). The hottest spot (HST) in The algorithm to compute Ttank is presented in Appendix III.
this transformer is located at the inner LV winding. The probe A. Lab Tests
in the prototype being outside this region was unable to catch
the HST, which location we did not know when the prototype Various lab tests were conducted on prototypes (used in ear-
was built. lier sections) kept in sealed enclosures, to study the de-rating
effect that the tank would have on the transformer. A dry-type
IX. TRANSFORMER IN ENCLOSURE arrangement (air inside enclosure) was used for this setup. To
avoid exceeding the temperature limitations of the prototypes,
Distribution grade transformers are required to be enclosed a 75% of rated load was used for these tests. The same proce-
in an air-tight metallic enclosure for safety reasons. The enclo- dure as explained in Section VII was followed for these tests.
sure may be solid (epoxy), liquid (oil) or air filled. The tank Fig. 9 shows the transformer in the enclosure under test.
provides additional, although small, thermal resistance to the
B. Finite Element Simulations
heat flux and hence raises the operating temperature of the
transformer. The tank can be modeled as a single lumped re- A FEM simulation with the enclosure modeled was used to
sistance in series with the ambient temperature ‘voltage’ compare results between the proposed model and the lab tests.
source (Fig. 2), effectively raising the ambient temperature to Fig. 9 shows the FEM surface plot for temperature and an ar-
Ttank given by row plot for fluid velocity. The boundaries of the enclosure are
closed since the fluid medium inside the enclosure is not al-
Ttank  T  Qloss  Rtank (10)
lowed to escape. The equations solved and external boundary
where Ttank is the enclosure temperature and Rtank is the effec- conditions are the same as explained in Section VI.
tive tank resistance.
 6

C. Results TABLE I
Results comparing steady state temperature between the RESULT COMPARISON FOR VARIOUS PROTOTYPES AND
tests, FEM model and proposed model are given in Figs. 10 CONFIGURATIONS
and 11. The results are plotted along section AA’ for the hori- HST for Free Air Tests
zontal and BB’ for the vertical temperature distribution within Rating Load HST (Test) HST (Model) HST (FEM)
the transformer. 1 kVA 1 kW 83.9 85 85.5
It is observed that the model predicts the temperature varia- 2 kVA 2 kW 78.5 78.8 80.4
tion very well. The HST occurs at the same location (inner LV 4 kVA 4 kW 82.8 83.5 84.7
winding) as the transformer exposed to ambient. The maxi- HST for potted Transformer Tests
mum error is less than 4% and is observed around the HST. 1 kVA 1 kW 86.4 88.5 88.8
4 kVA 3.7 kW 78.5 80.6 80.2
X. CONSOLIDATED RESULTS HST for Transformer in Enclosure Tests
A concise report of all the tests conducted on the prototypes 2 kVA 1.25 kW 81.7 78.7 80.4
is presented in Table 1. Five prototypes (3 standard and 2 en- Ttank = 43 Ttank = 46.2 Ttank = 45.4
capsulated in epoxy) of various power ratings with thermo- 89.1 86.6 86.9
4 kVA 3 kW
couples installed as described in Fig. 6 were tested until steady Ttank = 48.9 Ttank = 44.8 Ttank = 43.9
state was achieved as per IEEE C57 [10]. The load column
gives the constant electrical load connected at the transformer
terminals. It is observed that encapsulation or enclosing the
transformer leads to a de-rating. The comparison of the HST XIII. APPENDIX II
from the tests, model and FEM are provided. It must be noted The algorithm to compute temperature distribution within
that all the results are within practical engineering error limits transformer is given in Fig. 12
(less than 4%).

XI. CONCLUSIONS XIV. APPENDIX III

The paper has presented a model that provides detailed in- The algorithm to compute Ttank is given in Fig. 13.
sight into the variation of temperature within a toroidal trans-
former. The equivalent electrical circuit accurately models the
non-linear effects of convection and radiation and takes sub-
stantially less computational effort when compared to FEM. XV. ACKNOWLEDGMENT
The new model takes into account all geometric and electric
The authors gratefully acknowledge the contributions of
parameters such as: physical dimensions, insulation thickness,
number of turns, number of layers and conductor gauge. The Iván Hernández and Noel Augustine for their continuous sup-
model accurately predicts the hot-spots which knowledge is port to the research work. The authors are also thankful to
essential for the transformer designer. The model has been Chandira Prabhu for conducting all the tests and providing
validated against transient FEM simulations and measure- results presented in the paper. Finally, the authors would like
ments on actual transformers. The model if this paper is suita- to thank Mr. Ulrik Poulsen of Bridgeport Magnetics for his
ble to be used as an important component in a transformer fast response and expertise building the prototypes.
design computer program.

XII. APPENDIX I – PROTOTYPE TRANSFORMER DETAILS

120 / 120 Volt, 4 kVA, Rated current 33.33 A
Core ID: 12.7 cm
Core OD: 22.86 cm
Core height: 7.62 cm
No. of layers in primary and secondary: 2 each
Turn distribution for each layer: [118, 46, 105, 59]
Conductor gauge: AWG 9
Insulation (Mylar) thermal conductivity (k): 0.2 [W/(m.K)]
Rated core loss: 14.9 [W]
Full load primary winding loss: 40.8 [W]
Full load secondary winding loss: 48.9 [W]
 7

Fig. 10. Comparison of temperature distribution along section AA’ for

transformer in sealed enclosure

Fig. 13. Algorithm to compute temperature of the transformer tank

XVI. REFERENCES
[1] F. A. Furfari and J. W. Coltman, “The Transformer”, IEEE Industry
Applications Magazine, Jan./Feb. 2002, pp. 11-12.
[2] Colonel Wm. T. McLyman, “Transformer and Inductor Design Hand-
book”, 3rd Ed, New York: Marcel Dekker, Inc., 2004.
[3] M. van der Veen, “Modern High-end Valve Amplifiers: Based on To-
roidal Output Transformers”, Elektor Electronics Publishing, 1999.
[4] G. Swift, T. S. Molinski and W. Lehn, “A Fundamental Approach to
Transformer Thermal Modeling- Part I: Theory and Equivalent Circuit,”
Fig. 11. Comparison of temperature distribution along section BB’ for IEEE Transactions on Power Delivery, vol. 16, no. 2, pp.171-175, April
transformer in sealed enclosure 2001.
[5] O. A. Amoda, D. J. Tylavsky, G. A. McCulla and W. A. Knuth, “A New
Model for Predicting Hottest-Spot Temperatures in Transformers,” The
40th North American Power Symposium, pp.1-8, 2008
[6] D. Susa, M. Lehtonen and H. Nordman, “Dynamic Thermal Modeling of
Power Transformers,” IEEE Transactions on Power Delivery, vol. 20,
no. 1, pp.197-204, Jan 2005.
[7] I. Hernández, F. de León, and P. Gómez, “Design Formulas for the
Leakage Inductance of Toroidal Distribution Transformers”, paper sub-
mitted to the IEEE Transactions on Power Delivery.
[8] P. Gómez, F. de León, and I. Hernández, “Impulse Response Analysis of
Toroidal Core Distribution Transformers for Dielectric Design”, IEEE
Transactions on Power Delivery, Vol. 26, No. 2, April 2011, pp. 1231-
1238.
[9] Comsol Multiphysics, Heat Transfer Module User’s Guide, Comsol AB
Group, 2006, pp. 1-222.
[10] IEEE Standard Test Code for Dry-Type Distribution and Power Trans-
formers, IEEE Standard C57.12.91-1995, June. 1995.
[11] O. G. Martynenko, P. P. Khramtsov, Free convective Heat Transfer,
Springer-Verlag Berlin Heidelberg, 2005.
[12] R. B. Bird, W. E. Stewart and E. N. Lightfoot, Transport Phenomena,
2nd ed., John Wiley and Sons. Inc., 2007.
[13] S. Kakac and Y. Yener, Convective Heat Transfer, 2nd ed. CRC Press,
1995.
[14] M. N. Ozisik, Heat Conduction, 2nd ed., John Wiley and Sons. Inc.,
1993.

Fig. 12. Algorithm to compute temperature distribution in a transformer

 8

XVII. BIOGRAPHIES
Francisco de León (S’86–M’92–SM’02)
received the B.Sc. and the M.Sc. (Hons.) degrees
Sujit Purushothaman (S’09) received his B.E. in electrical engineering from the National
degree in electrical engineering from Mumbai Polytechnic Institute, Mexico City, Mexico, in
University (Sardar Patel College of Engineering), 1983 and 1986, respectively, and the Ph.D.
India in 2005. His work experience includes degree from the University of Toronto, Toronto,
testing and development of medium voltage ON, Canada, in 1992. He has held several
switchgear for Siemens India. He received his academic positions in Mexico and has worked
Master’s degree in 2009 and is currently pursu- for the Canadian electric industry. Currently, he
ing his Ph.D. at Polytechnic Institute of NYU. is an Associate Professor at the Polytechnic
His research interest includes power system Institute of NYU, Brooklyn, NY. His research
transients, subsynchronous resonance damping, interests include the analysis of power definitions
machine design and modeling and thermal mod- under nonsinusoidal conditions, the transient and steady-state analyses of
eling of electrical machines. power systems, the thermal rating of cables and transformers, and the
calculation of electromagnetic fields applied to machine design and modeling.
 1

Comparing the T and  Equivalent Circuits for

the Calculation of Transformer Inrush Currents
Francisco de León, Senior Member, IEEE, Ashkan Farazmand, and Pekir Joseph, Student Member,
IEEE

 core and shell types), the obtained  model has only one
Abstract— The most commonly used equivalent circuit for leakage inductance branch in series and two shunt
transformers is the traditional (Steinmetz) T-equivalent proposed
towards the end of the 19th century. This model has two leakage
magnetizing branches (see Fig 2). In 1951 Boyajian [4]
impedance branches and one magnetizing branch. The T model discussed the benefits of the  equivalent circuit emphasizing
properly represents the terminal behavior of the transformer for the unity of the leakage reactance between a pair of windings.
most low-frequency operating conditions. There exists another
model derived from the principle of duality between magnetic
and electric circuits, the  equivalent circuit, which has two
magnetizing branches and one leakage branch. This paper shows
that while the two equivalent circuits provide the same accuracy
in steady state, better accuracy for the calculation of inrush
currents is obtained with the -equivalent circuit. Laboratory
tests performed on three transformers with different
characteristics demonstrate that inrush current simulations with
the T equivalent circuit can have errors of up to 73%, while the 
equivalent predicts the measurements in every case within a few Fig. 1. T equivalent circuit
percent.

Index Terms— Duality, inrush currents, transformer

equivalent circuits, transformer modeling.

I. INTRODUCTION

F OR longer than a century the generally accepted

equivalent circuit for a two-winding transformer has been
the T equivalent. This model has the leakage inductance (Ls) Fig. 2.  equivalent circuit
divided into two branches, one is associated with the primary
(Ls1) and another one associated with the secondary winding In 1953, Slemon [5] generalized the theory of duality and
(Ls2). The model is completed with a shunt magnetizing showed how nonlinearities can be introduced into the circuit
branch (composed by the parallel Rm, Lm); see Fig 1. The elements of the  equivalent since they have a one-to-one
originator of the equivalent circuit seems to be Steinmetz in relationship with the transformer flux paths. Duality-derived
1897 [1]. A detailed discussion of the physical meaning of the models have long been used for the calculation of
elements of the T equivalent circuit is also given in [1]. electromagnetic transients [6]-[8]. However, they have not
As early as 1925, Boyajian [2] demonstrated the made the transition to steady state. The reason is perhaps that
impossibility of a physically meaningful resolution of the for steady state studies (not involving heavy saturation), the T
leakage inductance as belonging partially to the primary model gives almost perfect results. Moreover, for the most
winding and partially to the secondary winding (as it is done common power system studies such as: load flow, short
in Fig. 1). The leakage inductance can only be defined (or circuit, and stability, the (shunt) magnetizing branch, whose
measured) for a pair of windings. Therefore, the T equivalent impedance is normally very large when compared to the
circuit should be seen only as a terminal equivalent circuit (series) leakage impedance, is often neglected. This renders
since its elements do not have any physical relationship with the two circuits identical.
the building components of a transformer (core and windings). This paper shows, experimentally and analytically, that
Cherry [3] in 1949 showed that equivalent circuits for there are conditions where the T equivalent circuit is not
transformers could be conveniently obtained from the capable of properly representing the transformer under heavy
principle of duality between magnetic and electric circuits. saturation conditions. For example, when a transformer has
When duality is applied to a single-phase transformer (both large leakage inductance and the core saturates, the T
equivalent circuit fails to reproduce the terminal behavior.
F. de León, A. Farazmand, and P. Joseph are with the Department of Errors in the order of 73% were measured with the T
Electrical and Computer Engineering, Polytechnic University, Six Metrotech
Center, Brooklyn, NY, 11201. (e-mails: fdeleon@poly.edu,
equivalent circuit in the inrush currents for transformers with
afaraz01@students.poly.edu, pekirj@yahoo.com). relatively large leakage inductance.
 2

To explain the reasons why the  model performs better 4% to 20% of the rated voltage to get the rated current in the
than the T model three existent transformers with different low-voltage winding (see Table I).
parameters were selected for the experimental study: (1) A
standard transformer (TS), which is characterized by typical
TABLE I
leakage and magnetizing inductance values. (2) A standard RESULTS OF THE STANDARDIZED TESTS ON THE FOUR TRANSFORMERS
torodial transformer (TT), which is characterized by having
large magnetizing inductance and very small leakage Transformer TS TT TL
Standard Reduced Enlarged
inductance. (3) A second toroidal transformer (TL) designed Leakage Leakage Leakage
with no overlapping sectored windings of 180˚. This produces 120.18 120.04 120.19
Voc [V]
a very large leakage inductance. The geometrical information 5.3297 0.30886 0.254976
Ioc [A]
is given in Appendix A. 39.08 10.18 13.44
Poc [W]
In the next Section, the parameters of the equivalent circuit
Ratio 1:1 1:1 1:1
for each transformer are measured and compared. In Section
Rating [kVA] 1 1 1
III the air-core inductance, essential for the proper calculation
Vsc [V] 5.15 5.08 24.73
of inrush currents, is computed with 3D finite elements
Isc [A] 8.38 8.73 8.75
simulations. In Section IV the inrush performance of the two
Psc [W] 40.351 43.9859 46.871
models is compared. In Section V a parametric analysis on
how the division of the leakage and magnetizing inductances
affect the transformer inrush current is presented. Finally, in
Section VI, the large errors obtained with the T model are C. Calculation of Circuit Parameters
explained by analyzing the variation of the open circuit Table I shows the results of the standardized open circuit
impedance as the core saturates and the leakage inductance (oc) and short-circuit (sc) tests (at 60 Hz) for the three
increases. transformers under study. The parameters of the equivalent
circuits are computed with the following expressions:
II. PARAMETER MEASUREMENT
Accurate determination of the transformer magnetizing and Psc
leakage parameters is of paramount importance to produce a R1  R2  (4a)
I sc2
correct comparison of model performance. To determine the
parameters the procedures of the IEEE Standard C57.12.91- 2
1  Vsc 
1995 [9] for open circuit and short-circuit tests were followed.
 
2

The measurements of instantaneous voltage and current are

L s    R1  R2 (4b)
2 f  I sc 
obtained very precisely using a YOKOGAWA 2 MHz power
analyzer (PZ4000), with a sampling rate of 20 µs. From the (Voc  R1 I oc ) 2 (V  R1 I oc ) 2
measuring system, 833 samples per cycle of voltage and R m ; L m  oc (5)
current are obtained. The rms values for voltage and current Poc 2 f Qoc
are computed from basic principles as follows: where Psc and Poc are the active powers computed from the
T T short circuit and open circuit tests, respectively. Qoc is the
1 2 1 2
 v  t  dt ; irms  i  t  dt (1)
T 0
vrms  open circuit reactive power. Vsc and Isc, are the rms values of
T 0 short circuit voltages and currents, respectively. Voc and Ioc are
the rms values of open circuit voltages and currents,
The active power is computed from the average of the respectively. Ls is the total (series) leakage inductance. Rm is
instantaneous power as: the magnetizing resistance, Lm is the magnetizing inductance,
T R1 and R2 are the primary and the secondary ac resistances,
1
v  t  i  t  dt (2)
T 0
P  respectively, and f=60 Hz.
The total series ac resistance R1 + R’2 is computed from
The reactive power is calculated with the following formula: (4a). Individual break down of the resistances is needed for the
equivalent models. Additionally, primary and secondary
Q  (vrms irms ) 2  P 2 (3) leakage inductances are also needed for the T model. When no
information is given on the value of the individual dc
A. Open Circuit Test resistances, it is accepted to divide the leakage (and ac
The low-voltage winding of the transformer is energized resistance) equally into the two windings [10]. For this paper,
with rated voltage keeping the high-voltage side in open- measurement of the dc resistance was performed. Therefore,
circuit. The terminal voltage of the high-voltage (open) side the leakage inductance and ac resistance are divided into two
and current of the low-voltage (connected) side are captured. as it is traditionally done in proportion to the dc resistances
[11]; see Table II. In Section V, this division of the leakage
B. Short-Circuit Test impedance is varied over a wide range to gauge the effect of
The high-voltage winding is energized with the low-voltage having more or less leakage to each side. The magnetizing
winding short-circuited. The voltage applied is varied from parameters of the T model are obtained directly from (5).
 3

TABLE II
CIRCUIT PARAMETERS FOR T AND  MODELS

Transformer TS TT TL
Standard Reduced Enlarged
Leakage Leakage Leakage
R1 [] 0.251 0.277 0.306
R’2 [] 0.324 0.300 0.305
Ls1 [mH] 0.302 0.111 4.393
L’s2 [mH] 0.390 0.121 4.385
Rm [] 369.53 1,415.97 1,074.71
Lm [mH] 71.91 1,284.73 1,669.60
Ls [mH] 0.692 0.232 8.778 Fig. 4. Hysteresis cycles of the toroidal transformers (TT and TL)
Rm1 [] 739.06 2,831.94 2,149.42
Rm2 [] 739.06 2,831.94 2,149.42 III. AIR-CORE INDUCTANCES
Lm1 [mH] 143.82 2,569.46 3,339.21
It was not possible to measure the deep saturation section
Lm2 [mH] 143.82 2,569.46 3,339.21
region of the hysteresis loops in the lab due to the large power
Lm/Ls 103.89 5,537.63 190.20
requirements (high-voltage and high-current). Yet this region
is of paramount importance to compute properly the inrush
For the  model the leakage inductance is obtained directly currents. 3-D finite element (FEM) simulations were
from (4b) and the magnetizing parameters are the double of performed to determine the air-core inductance. The
those obtained from (5) [8]. Therefore, Rm1 = Rm2 = 2 Rm and commercial program COMSOL Multyphisics was used for
Lm1 = Lm2 = 2 Lm. Also in Section V, this division of the this purpose [13]; see Appendix B.
magnetizing impedance is varied over a wide range to The dimensions of the low voltage winding were used for
determine the effect of assigning more or less magnetizing to the FEM simulations using air cores. The volume magnetic
each side. The parameters computed from rated measurements energy is extracted directly from COMSOL and then the
are shown in Table II. inductance is calculated using the following formula:
D. Hysteresis Cycles 2W
Lair  core  2 (6)
A family of hysteresis curves were obtained for each of the I
three transformers under test. These hysteresis curves are Table III gives the air-core inductances of the three
acquired from the measurement of the instantaneous values of transformers studied in this paper. Note, however that the
voltage and current. Faraday’s Law is then used to convert the construction details of the standard transformer TS are not
induced voltage into flux. The hysteresis cycles of known. Therefore, an estimation was obtained from the inrush
transformers TS, TT, and TL are shown in Fig. 3. In Appendix tests. The hysteresis curve is extended using the air-core
C the numerical values of the upper part of the cycles are inductances as the slope from the last measured point to
given (as required by the EMTP-RV [12]). infinity. These values are included in the tables of Appendix
One can appreciate from Fig. 3 that the standard C.
transformer (TS) shows a traditional hysteresis cycles. The
toroidal transformers (TT and TL) have a flat and narrow TABLE III
hysteresis cycles. This is so because there are no gaps in the AIR CORE INDUCTANCE FOR THE TRANSFORMERS
core. Fig. 4 shows a zoom on the hysteresis cycles of the TS TT TL
toroidal transformers. Note that the transformer with enlarge Transformer Standard Reduced Enlarged
leakage TL has a slightly wider cycle, but the saturation flux is Leakage Leakage Leakage
the same. Air-core
1000 [µH] 316 [µH] 463 [µH]
Inductance

IV. MODEL COMPARISON

A. Description of the Inrush Current Experiments
Starting with the transformer core demagnetized, the
worst conditions (maximum inrush currents) occur when the
energization coincides with the voltage wave zero-crossing
[14]. This situation can be reproduced in the laboratory by
connecting the transformer through a zero-crossing detecting
switch as shown in Fig. 5. To obtain accurate and consistent
inrush current measurements, any remanence in the
transformer from the previous energization must be removed
[15]. The remanence removal process was done by gradually
Fig. 3. Hysteresis cycles of the transformers TS (standard design), TT (reduced
leakage), and TL (enlarged leakage). reducing the voltage to zero before de-energizing the
transformer from the source.
 4

Fig. 5. Model of experimental set up for measuring inrush currents.

A zero-crossing switch consisting of voltage regulators,

optic isolators, digital logic control circuits, and MOSFETs is
used to switch-on the transformer. When the source voltage
Fig. 6. Inrush current comparison: experimental versus simulated using T and
crosses zero, the switch is closed and the waveforms of inrush π models for transformer TS (standard design).
current are captured with the power analyzer. The results are
compared with simulations in the next section).
B. Simulations versus Measurements
The EMTP-RV [12] was used for the simulation of the tests
described in Fig. 5 using the two equivalent circuits (Fig. 1
and Fig. 2). The non-linear characteristics of all inductors
representing iron-core components are modeled with the built-
in hysteresis fitter (using the data computed in the previous
section). The short-circuit impedance of the source was
measured, which is almost purely resistive with a value of
Zsource = 0.1444 . Figs. 6, 7, and 8 compare the results
obtained by simulation using the T and  models against the Fig.7. Inrush current comparison: experimental versus simulated using T and π
measurements for the three transformers under study. models for transformer TT (very small leakage inductance).
Fig. 6 shows the results for the standard transformer TS. The
peak value of the inrush current using the  model is 121.1 A,
which is very close to the experiment result (123.7 A). In this
case the difference is only 2%, while the T model gives 111.7
A, which corresponds to 10% difference with respect to the
experiment result. The current shape of both models follows
the same path for low currents and only towards the peak they
separate. The peak of the measured inrush current is about 10
times larger than the rated 11.78 A peak (8.33 A rms).
Fig. 7 shows the results for the toroidal transformer TT,
whose main characteristics are to have very small leakage
inductance and very large magnetizing inductance (see Table
Fig.8. Inrush current comparison: experimental versus simulated using T and π
II). The  model and experiment give the same value for the models for transformer TL (very large leakage inductance).
first peak (254.5 A), while the T model shows 241.4 A. In this
case for the T model the difference is about 5% at the peak For all three transformers the  model gives more precise
with the experiment result. For this transformer the measured results than the T model. When the leakage inductance is small
inrush current is about 22 times larger than the rated current. (transformers TT and Ts) the T model results are also
Fig. 8 shows the results for the toroidal sector wound acceptable and relatively close to the experiment, but when the
transformer TL, whose main features are to have a very large
leakage inductance is large (transformer TL), the T and 
leakage inductance and also a very large magnetizing
models behave quite differently: the T model shows a very
inductance (see Table II). In this case, the measured peak of
large error, while the  model is close to the experimental
the inrush current is 201.4 A peak (about 17 times larger than
results.
the rated current).
From the results of the simulations and measurements of
The  model gives 179.4 A at the first peak, which
this section, one can conclude that model selection plays an
represents a difference of 10.9% with respect to the
important role in the calculation of inrush currents when the
experiment result, while the T model yields 54.9 A, which
leakage inductance is large. We make the statement that the 
corresponds to a very large error of 72.7% at the peak.
model should be always used, not only because its elements
Note that, the measured peak inrush current of transformer
have a clear physical meaning, but also because large errors
TT is about 26% larger than the peak inrush current of
may appear when using the traditional T model. Both circuits
transformer TL. This is because of the larger leakage
inductance value of transformer TL in comparison to TT are very simple, the  model has only one more circuit
(almost 38 times) which limits the inrush current considerably. element (7 versus 6) than the T model, but the  equivalent
may provide better results under heavy saturation conditions.
 5

V. PARAMETRIC STUDIES Section IV). Fig. 10 compares the inrush current waveforms
In this section, a parametric analysis of how the division of for the different cases with the experiment result. Analyzing
the leakage and magnetizing inductances affect the calculated Figs. 9 and 10 one can see that the calculations are less
inrush currents using the T and  models is presented. sensitive to the division factor in the  model than in the T
Transformer TL has been selected to illustrate the model.
parametric simulations because this is the one that presents
TABLE IV
larger variations; see Table IV. The first column presents a PARAMETRIC STUDY OF T MODEL (LEAKAGE INDUCTANCE DIVISION)
division factor a used to split the total leakage inductance (Ls)
into primary and secondary sides of the T model. For example, a (%) Ls1 [mH] Ls2 [mH] Peak Current [A] Error [%]
the first row (corresponding to a = 0) presents the case when 0 0.000 8.778 230.36 14.38
10 0.878 7.900 139.75 -30.61
all leakage inductance is entirely on the secondary side of the
20 1.756 7.022 100.72 -49.99
T model. In the next row (a = 10%), 10% of the leakage is
30 2.633 6.145 78.83 -60.86
placed on the primary side and 90% on the secondary side. In 40 3.511 5.267 64.78 -67.84
the last case (a = 100%), all the leakage inductance is on the 50 4.389 4.389 54.99 -72.70
primary side of the transformer. 60 5.267 3.511 47.78 -76.28
The last column of Table IV presents the errors in the 70 6.145 2.633 42.24 -79.03
calculated peak currents between the T model and the 80 7.022 1.756 37.86 -81.20
experimental results. From the results, it is obvious that 90 7.900 0.878 34.3 -82.97
increasing the primary side leakage (Ls1) limits the inrush 100 8.778 0.000 31.35 -84.43
current considerably, which causes large errors. For the case in Case with zero error:
which the division factor is 50% [10], the error is 71.97%; 2.2 0.192 8.586 201.4 0.00
when the leakage inductance is divided based on the dc
resistances (as recommended in [11]), the error is 72.7%; and TABLE V
the error is zero when only 2.7% of the total leakage PARAMETRIC STUDY OF  MODEL (MAGNETIZING INDUCTANCE DIVISION)
inductance is on the primary side of the transformer. Fig. 9 a (%) Peak Current [A] Error [%]
compares the inrush current waveforms for five different cases 0 230.3 14.38
using a 25% division factor. From the figure, one can observe 25 211.3 4.92
that the inrush current computed with the T model show a 50 179.4 -10.92
large sensitivity especially at the beginning. 75 128.2 -36.35
To study the splitting of the magnetizing impedance in the  100 31.34 -84.43
model, a division factor of 25% has been selected. The total Case with zero error:
magnetizing current between the two magnetizing branches is 32.5 201.4 0.00
divided proportionally. Remember that the magnetizing model
is nonlinear since it includes saturation and hysteresis. As it is VI. ANALYSIS OF THE TERMINAL IMPEDANCE
shown in Table V, the first and last cases (with division In this section, the large errors obtained with the T model
factors of 0 and 100%, respectively) are equal to the cases are explained by analyzing the variation of the open circuit
with one magnetizing branch. Therefore, for these two cases impedance as the core saturates (Lm reduces). In addition, the
the results are exactly the same as the first and last cases of T effect of increasing the leakage inductance (Ls) is presented.
model (see Table IV). The (open circuit) equivalent impedance for the T and π
In all other cases, the error is smaller than for the T model. equivalent circuits can be computed from series-parallel
The error is zero when the division factor is about 32.5%, and simplifications of the circuits of Fig. 1 and Fig. 2,
for a 50% division factor, the error is 10.9% (as presented in respectively, as follows:

(a) (b)

Fig.9. Inrush current comparison using T model for transformer TL under different division factors for the leakage inductance. (a) Error with respect to the
experiment; (b) Inrush currents waveforms.
 6

(a)
(b)
Fig.10. Inrush current comparison using  model for transformer TL under different division factors for the magnetizing impedance. (a) Error with respect
to the experiment; (b) Inrush currents waveforms.
much faster than the impedance of the  model. Under heavy
1 (7) saturation conditions Lm is small. This explains why the T
ZT  R1  j L s1  model under predicts the inrush currents for transformers with
1 1
 large leakage inductance (see Fig. 8).
R m j L m
To study the effect on the open circuit impedance of the
reduction of the magnetizing inductance due to saturation, Lm
1 together with Lm1=Lm2= 2 Lm, were decreased in small steps to
Z  R1 
1 1 1 a value one thousand times smaller. The terminal impedance
 
Rm1 j Lm1 j L  1 calculations (shown in Fig. 14) indicate that the saturation of
s
1 1 the core by itself is not responsible for the large terminal
 (8)
Rm 2 j Lm 2 impedance differences between the T and  models. When the
leakage inductance of the transformer is increased 10 times
from 0.692 mH to 6.92 mH the impedance variation of Fig. 15
Using the values for the standard transformer (TS) from the
is obtained. Small differences exist when the magnetizing
first column of Table II, we find the equivalent circuits of Fig.
inductance is large (not saturated), but larger differences can
11 and Fig. 12.
be observed when the magnetizing inductance is small
(saturated).
The results of this section explain why both models give
about the same inrush current for transformers with small
leakage inductance; see Figs. 6 and 7. Looking at the topology
of the two circuits (Figs. 11 and 12) one can observe that in
the T model the primary winding leakage inductance (LS1)
Fig. 11. T equivalent circuit for the TS transformer at nominal voltage. limits the circulation of current to the magnetizing branch.
This prevents large currents (especially inrush when the core
saturates) to be drawn by the transformer. In the  model the
path of the inrush current is open to the one of the magnetizing
branches. Therefore, in this case the  model is more precise
than the T model (see Fig. 8).

Fig. 12.  equivalent circuit for the TS transformer at nominal voltage.

The effect of increasing the leakage inductance on the

terminal impedance is studied varying the parameters of the
equivalent circuits of Figs. 11 and 12. The leakage
inductances, Ls together with Ls1 and Ls2, were increased (one
thousand times) in small steps. Fig. 13 shows the terminal
(open circuit) impedance for the T and  models against the
leakage inductance (normalized with the magnetizing
inductance Lm). One can see that for small Ls / Lm ratios both
Fig.13. Variation of the terminal (open circuit) impedance with respect the
circuits give the same terminal impedance. This is the normal ratio of leakage versus magnetizing inductances increasing the leakage
region because Lm>>Ls for most transformers. However, as the inductance.
ratio Ls / Lm increases the impedance of the T model increases
 7

IX. APPENDIX A – CONSTRUCTION INFORMATION OF THE

TOROIDAL TRANSFORMERS UNDER TEST
Core Dimensions [inches] Winding Characteristics
Internal Outer Height Primary winding Secondary winding
Diameter Diameter Number Wire Number Wire
of Turns gauge of Turns gauge
3.375 5.875 2 196 13 196 13

X. APPENDIX B – AIR-CORE INDUCTANCE

Since it was not possible to measure the air-core inductance
in the lab because of the high power requirements, 3D finite
element simulations were performed. Figs. 16(a) and 16(b)
Fig.14. Variation of the terminal (open circuit) impedance with respect the show the top view of the distribution of the magnetic flux
ratio of leakage versus magnetizing inductances reducing the magnetizing
inductance. density for the TT and TL transformers respectively.
Transformer TT has the winding distributed over the entire
360˚, while transformer TL occupies only 180˚. From Fig.
16(a), one can see that the field is mostly contained inside the
coil with higher flux densities towards the inner diameter. Fig.
16(b) shows that for the TL transformer the flux density is
concentrated inside the wound semi-circle, but the return
through the air is quite scattered.

Fig.15. Variation of the terminal (open circuit) impedance with respect the
ratio of leakage versus magnetizing inductances for an increased leakage
inductance transformer.

VII. CONCLUSIONS
This paper has shown experimentally that the traditional T
model of transformers may yield large errors when computing
inrush currents. This is especially true when the transformers
have large leakage inductance. Better accuracy for the (a)
calculation of inrush currents has been obtained with the 
equivalent circuit. Laboratory tests performed on several
transformers demonstrate that inrush current simulations with
the T equivalent circuit could under predict the inrush currents
by as much as 72.7%, while the equivalent circuit predicts
the measurements with a small percent error.
Physical, numerical, and analytical explanations on the
performance difference of the two models were given. The
topology of the T model, having the primary winding leakage
inductance element before a magnetizing branch, is the cause
for the model inaccuracies since it (incorrectly) limits the
circulation of current to the magnetizing branch when the core
saturates.

VIII. ACKNOWLEDGEMENTS
(b)
The authors would like to express their gratitude to Mr.
Rajesh Kumar and Mrs. Xiaojing Xu, both ex-graduate Fig.16. (a) Magnetic flux density for the TT transformer, (b) Magnetic flux
density for the TL transformer.
students of NYU-Poly, for their help during the initial
experimental stages of this project. We also would like to
thank Mr. Baris Kovan, current graduate student of NYU- XI. APPENDIX C – VALUES OF THE HYSTERESIS CYCLES
Poly, for performing the finite element simulations to compute Tables VI and VII present the numerical values of the
the air-core inductance. The efforts of Mr. Noel Augustine hysteresis cycles used for the T and  models for the three
with the finite elements simulations are also recognized. transformers. Note that because the  model has two shunt
inductors the value of the current is half for the same flux.
 8

TABLE VI XII. REFERENCES

NUMERICAL VALUES OF THE HYSTERESIS CYCLE OF TRANSFORMERS
(T-MODEL) [1] C. P. Steinmetz, Theory and Calculation of Alternating Current
TS (Standard Leakage) TT (Reduced Leakage) TL (Enlarged Leakage) Phenomena, McGraw Publishing Company, 1897. First Edition
Available through Open Library http://openlibrary.org.
I [A] Flux [Wb] I [A] Flux [Wb] I [A] Flux [Wb]
[2] A. Boyajian, “Resolution of Transformer Reactances into Primary and
0.1873 0 0.0598 0 0.0992 0
Secondary Reactances”, AIEE Transactions, Vol xx, No. yy, .pp. 805-
0.2025 0.0351 0.0714 0.0619 0.1024 0.0323
810, June 1925.
0.2739 0.1030 0.0793 0.1823 0.1119 0.0955 [3] E. C. Cherry, “The Duality Between Interlinked Electric and Magnetic
0.3148 0.1366 0.0897 0.2115 0.1224 0.1269 Circuits and the Formation of Transformer Equivalent Circuits”, Proc.
0.4082 0.1700 0.1031 0.2683 0.1258 0.1579 Of the Physical Society, Vol (B) 62, Feb. 1949, pp. 101-111.
0.5248 0.2030 0.1117 0.2958 0.1334 0.1886 [4] L. F. Blume, A. Boyajian, G. Camilli, T.C. Lenox, S. Minneci, and V.M.
0.7024 0.2354 0.1312 0.3226 0.1463 0.2188 Montsinger, Transformer Engineering: A Treatise on the Theory,
0.9172 0.2673 0.1520 0.3485 0.1647 0.2485 Operation, and Application of Transformers, p. 70, John Wily & Sons,
1.1522 0.2986 0.1593 0.3736 0.1781 0.2775 1951.
1.4849 0.3291 0.1978 0.3978 0.2030 0.3059 [5] G. R. Slemon, “Equivalent circuits for transformers and machines
1.9048 0.3588 0.2283 0.4210 0.2301 0.3335 including nonlinear effects,” Proc. Inst. Elect. Eng., pt. IV, vol. 100, pp.
3.3269 0.4153 0.3082 0.4641 0.3009 0.3860 129–143, 1953.
4.4774 0.4420 0.3906 0.4840 0.3492 0.4108 [6] F. de León and A. Semlyen, "Complete Transformer Model for
6.0106 0.4675 0.4938 0.5026 0.4059 0.4346 Electromagnetic Transients," IEEE Transactions on Power Delivery,
8.0034 0.4918 0.6799 0.5201 0.4767 0.4573 Vol. 9, No. 1, January 1994, pp. 231-239.
10.2941 0.5147 0.9827 0.5362 0.5560 0.4787 [7] J. A. Martinez and B. Mork, “Transformer Modeling for Low- and Mid-
12.7635 0.5362 1.5833 0.5509 0.6588 0.4989 Frequency Transients – A Review”, IEEE Transactions on Power
15.2251 0.5562 2.8455 0.5641 0.7828 0.5177 Delivery, Vol. 20, No. 2, April 2005, pp. 1525-1632.
17.7922 0.5749 5.1459 0.5758 0.9642 0.5351 [8] F. de León, P. Gómez, J. A. Martinez-Velasco, and M. Rioual, Chapter
20.3166 0.5920 9.6796 0.5859 1.2713 0.5506
4, “Transformers” in Power System Transients: Parameter
Determination (Edited by J.A. Martinez-Velasco), CRC Press, Boca
22.7446 0.6074 17.9407 0.5941 1.8412 0.5636
Raton FL, 2009, pp. 177-250.
25.0224 0.6213 29.6356 0.6002 2.8446 0.5739
[9] IEEE Std C57.12.91-1995, “IEEE Standard Test Code for Dry-Type
27.2454 0.6335 40.0726 0.6042 4.6666 0.5816
Distribution and Power Transformers”, Recognized as an American
29.3236 0.6440 45.7422 0.6062 8.2994 0.5872 National Standard (ANSI).
30.6528 31.2725 46.8378 0.6066 15.2306 0.5912 [10] V. Del Toro, "Principles of Electrical Engineering", Second Edition,
33.0559 0.6600 600 0.7814 23.6656 0.5937 Prentice-Hall, Englewood Cliffs, New Jersey, 1972.
34.5354 0.6653 30.3695 0.5952 [11] Staff of the Department of Electrical Engineering of Massachusetts
35.6353 0.6690 34.4554 0.5960 Institute of Technology, “Magnetic Circuits and Transformers”, John
36.1479 0.6709 35.8700 0.5962 Wiley and Sons, New York, 1943.
400 1.0348 600 0.8574 [12] DCG-EMTP (Development coordination group of EMTP) Version
EMTP-RV, Electromagnetic Transients Program. [Online]. Available:
TABLE VII http://www.emtp.com.
NUMERICAL VALUES OF THE HYSTERESIS CYCLE OF TRANSFORMERS [13] Comsol Multiphysics 4.2a User’s Guide, Burlington, MA, 2011.
(PI-MODEL) [14] A. Greenwood, “Electrical Transients in Power Systems”, Second
TS (Standard Leakage) TT (Reduced Leakage) TL (Enlarged Leakage) Edition, John Wiley & Sons, New York 1991, pp. 113-116.
I [A] Flux [Wb] I [A] Flux [Wb] I [A] Flux [Wb] [15] B. Kovan, F. de León, D. Czarkowski, Z. Zabar, and L. Birenbaum,
0.0936 0 0.0299 0 0.0496 0 “Mitigation of Inrush Currents in Network Transformers by Reducing
0.1013 0.0351 0.0357 0.0619 0.0512 0.0323 the Residual Flux with an Ultra-Low-Frequency Power Source”, IEEE
0.1370 0.1030 0.0397 0.1823 0.0560 0.0955 Transactions on Power Delivery, Vol. 26, No. 3, July 2011, pp. 1563-
0.1574 0.1366 0.0449 0.2115 0.0612 0.1269 1570.
0.2041 0.1700 0.0516 0.2683 0.0629 0.1579
0.2624 0.2030 0.0558 0.2958 0.0667 0.1886 XIII. BIOGRAPHIES
0.3512 0.2354 0.0656 0.3226 0.0732 0.2188
Francisco de León (S’86–M’92–SM’02) received the B.Sc. and the M.Sc.
0.4586 0.2673 0.0760 0.3485 0.0823 0.2485
(Hons.) degrees in electrical engineering from the National Polytechnic
0.5761 0.2986 0.0797 0.3736 0.0890 0.2775
Institute, Mexico City, Mexico, in 1983 and 1986, respectively, and the Ph.D.
0.7424 0.3291 0.0989 0.3978 0.1015 0.3059
degree from the University of Toronto, Toronto, ON, Canada, in 1992. He has
0.9524 0.3588 0.1141 0.4210 0.1150 0.3335 held several academic positions in Mexico and has worked for the Canadian
1.6635 0.4153 0.1541 0.4641 0.1504 0.3860 electric industry. Currently, he is an Associate Professor at Polytechnic
2.2387 0.4420 0.1953 0.4840 0.1746 0.4108 Institute of NYU, New York. His research interests include the analysis of
3.0053 0.4675 0.2469 0.5026 0.2029 0.4346 power phenomena under nonsinusoidal conditions, the transient and steady
4.0017 0.4918 0.3400 0.5201 0.2383 0.4573 state analyses of power systems, the thermal rating of cables, and the
5.1470 0.5147 0.4913 0.5362 0.2780 0.4787 calculation of electromagnetic fields applied to machine design and modeling.
6.3818 0.5362 0.7916 0.5509 0.3294 0.4989
7.6126 0.5562 1.4227 0.5641 0.3914 0.5177 Ashkan Farazmand was born in Tehran, Iran, in 1983. He received his M.Sc.
8.8961 0.5749 2.5729 0.5758 0.4821 0.5351 (Hons.) from the University of Tehran in 2009, and is currently pursuing a
10.1583 0.5920 4.8398 0.5859 0.6357 0.5506 Ph.D. in Electrical Engineering from the Polytechnic Institute of NYU,
11.3723 0.6074 8.9703 0.5941 0.9206 0.5636 Brooklyn, New York. His research interests are design and analysis of
12.5112 0.6213 14.8178 0.6002 1.4223 0.5739 transformers, electrical transients, derating of electrical machines under non-
13.6227 0.6335 20.0363 0.6042 2.3333 0.5816 sinusoidal and unbalanced conditions, and power quality.
14.6618 0.6440 22.8711 0.6062 4.1497 0.5872
15.6362 0.6528 23.4189 0.6066 7.6153 0.5912 Pekir Joseph (S’09) was born in Brooklyn, New York in 1987. He received
16.5280 0.6600 300 0.7814 11.8328 0.5937 his Bachelor degree in Electrical Engineering in 2009 and his Master’s degree
17.2677 0.6653 15.1848 0.5952
in Power System Engineering in 2011, both, at the Polytechnic Institute of
New York University. His research interests are in electrical transients,
17.8176 0.6690 17.2277 0.5960
switching surges, relay protection and coordination, transmission and
18.0740 0.6709 17.9350 0.5962
distribution, and inrush currents in transformers.
200 1.0348 300 0.8574
 1

A Comparative Study on π and T Equivalent Models for the

Analysis of Transformer Ferroresonance
Saeed Jazebi, Student Member, IEEE, Ashkan Farazmand, Brahadeesh Perinkolam Murali, and
Francisco de León, Senior Member, IEEE

 20 µF capacitance was connected in series with the terminals.

Abstract—The performance of the T and the  equivalent
models used to represent transformers are tested under ferrores-
Both models show the occurrence of ferroresonance with volt-
onance. Comparisons between simulations and laboratory exper- ages within a few percent error when compared to the experi-
iments show the superiority of the  equivalent circuit. ments (details are not presented).
Index Terms—Ferroresonance, transformer modeling. TABLE I
ELECTRICAL PARAMETERS OF TRANSFORMERS
I. INTRODUCTION Code
R1 R’2 Rm Ls Lm-lin Lm-sat
() () () (mH) (mH) (mH)

F ERRORESONANCE may cause severe temporary over-

voltages and damage the internal or external insulation of
transformers. To predict possible over-voltages, proper model-
T1
T2
0.277
0.306
0.300
0.305
1,415.9
1,074.7
0.23
8.78
1,284.7
1,669.6
316
463

ing of ferroresonance is required for computer simulation. The R1+Rs Cs Ls/2 Vn L's/2 R'2 R1+Rs Cs Vn1 Ls Vn2 R'2
T equivalent circuit is the most common representation of a
two winding transformer [1]; see Fig. 1(a). An alternative, the Vs Lm Rm Vs Lm1
2Rm Lm2 2Rm
 model [2], is a duality derived representation for a trans-
former that has advantages over the T model; see Fig. 1(b). In (a) (b)
this paper, the T and  models are compared using time- Fig. 1. Equivalent circuit of the experimental setup; a) π model; b) T model.
domain simulations against laboratory experiments. The re-
In the second experiment, ferroresonance occurs on T2 with
sults show that the T model may produce large errors while
the series 20 µF capacitance (see Figs. 2 and 3). Note howev-
the  model predicts properly the occurrence of ferroreso-
er, that the T model exhibits a completely different behavior
nance. All simulations in the paper are carried out with the
than the measurements. The mismatch is evident in both volt-
EMTP considering detailed representation of the hysteresis
age and current; and even the frequency of oscillation is dif-
curves (except when noted) including non-linear magnetiza-
ferent. The computed overvoltage is 44 % higher than the ex-
tion and losses.
perimental result. On the other hand, the current and voltage
II. SIMULATIONS VERSUS LABORATORY EXPERIMENTS of the  model are visibly correct with a relative difference of
Two 1 kVA, 120:120 V transformers (T1 and T2) with maximum about 5% with respect to the experimental results.
electrical parameters presented in Table I are selected. T1 has The third experiment presents ferroresonance between T1
typical impedance parameters for a small power transformer, and a 60 µF capacitance. The voltage waveforms are present-
while T2 has been selected because it has a substantially larg- ed in Fig. 4. One can note that the experiments and the π mod-
er leakage inductance and serves to accentuate the differences el show a normal operating condition (no ferroresonance), but
between the two circuits. The equivalent circuits for the exper- the T model predicts ferroresonance.
imental setup are depicted in Fig. 1. The parameters are ob- III. DISCUSSION
tained from the standard impedance and open circuit tests ac-
During ferroresonance, transformers have transitions be-
cording to the IEEE Std C57.12.91-1995.
tween the linear and the nonlinear regions of the hysteresis
A large number of experiments have been carried out with
curve. In this section, to study the performance of the trans-
the secondary of the transformers open circuited and applying
formers, the nonlinearities are represented by piecewise-linear
rated voltage. Ferroresonance is chaotic and depends on initial
models with only two sections (see Fig. 5). Parameters Lm-lin
conditions. To get consistent results, the core was demagnet-
and Lm-sat are the slopes of the linear and deep saturation parts
ized and the series capacitor was discharged before each ex-
of the magnetizing curve respectively.
periment. We made sure that the results were consistent, and
In the π model, there exist two shunt magnetizing branches
not affected by the chaotic nature of ferroresonance. Only
with internal nodal voltages denoted Vn1 and Vn2. The (inter-
three cases are discussed here. The first test is on T1 when a
nal) voltage of the T model’s magnetizing branch is Vn (see
Fig. 1). The relations between the internal node voltages and
S. Jazebi, A. Farazmand, B. Murali, and F. de León are with the depart- the source voltage, neglecting all damping components are:
ment of Electrical and Computer Engineering, Polytechnic Institute of New
York University, Six Metrotech Center, Brooklyn, NY, 11201 (e-mails: Vn 2Cs Lm 2
jazebi@ieee.org, afaraz01@students.poly.edu, bra-  1
Vs Cs ( Ls  2 Lm ) 2  2
hadeesh.murali@gmail.com, fdeleon@poly.edu).
 2

Vn1 Cs Lm1 ( Ls  Lm 2 ) 2 ences become more noticeable for transformers with large
  2 leakage inductance. Thus, it is possible that Lm1 goes into satu-
Vs Cs Lm1 ( Ls  Lm 2 ) 2  Lm1  Lm 2  Ls
ration while Lm2 is still working in its linear part; this situation
corresponds to Case 3. For transformer T2, the terminal volt-
T Model  Model Experiment age versus the value of the series capacitance is presented in
300
Fig. 6. The figure shows that the resonance behavior of T and
200
π models is quite different at various operating conditions.
Terminal Voltage [V]

100 This can also be observed from the capacitance values that
0 would produce resonance:
-100 Lm1  Lm 2  Ls 2
C  CT   4
-200 Lm1 ( Ls  Lm 2 ) 2 ( Ls  2 Lm ) 2
-300 where Lm, Lm1 and Lm2 can be substituted by Lm-lin or Lm-sat de-
0.02 0.03 0.04 0.05 0.06 0.07 0.08
pending on the values of the instantaneous voltages Vn, Vn1
Time (Sec) and Vn2. Note that, the differences between Cπ and CT become
Fig. 2. Terminal voltage of T2 with 20 µF series capacitance.
larger for transformers with higher leakage inductance. For
transformer T1, the resonance response of the models is much
T Model closer than for T2 (results not shown). However, sometimes
30
 Model Peak=15.4 A
20 Peak=34.1 A Experiment the T model fails; Fig. 4 shows a case when the T model pre-
Terminal Current [A]

Peak=32.9 A
dicts ferroresonance when it does not occur in reality.
10
A comprehensive sensitivity analysis on transformer pa-
0
rameters (Ls/Lm) with respect to terminal behavior of both
T Model
-10
Peak=23.2 A models for the calculation of inrush currents is presented in
-20 [3].
-30

0.02 0.03 0.04 0.05 0.06 0.07 0.08 Lm-sat

Time (Sec)
Fig.3. Terminal current of T2 with 20 µF series capacitance.
Lm-lin

T Model  Model Experiment

200
Current [Amp.]
Terminal Voltage [V]

100 Fig. 5. Simplified magnetizing curve for T2 used for analysis purposes.

0 Vn (Case 3)
200
-100 0
Vn1, Vn2 (Case 1)
-200 -200
Voltage (Pu)

Vn (Case 2)
-400
0.02 0.03 0.04 0.05 0.06 0.07 0.08
Time (Sec) -600

-800 Vn1, Vn2 (Case 2)

Fig. 4. Terminal voltage of T1 with 60 µF series capacitance. Vn (Case 1)
-1000
Vn1, Vn2 (Case 3)
Vn 2 Cs Lm1 Lm 2 2
  3 -1200
Vs Cs Lm1 ( Ls  Lm 2 ) 2  Lm1  Lm 2  Ls 4 6 8 10
Cs (F)
12 14 16
-6
x 10
To highlight the difference in the resonance behavior of the
Fig. 6. Non-linear branch voltages by varying series capacitance for T2.
two equivalent circuits, three cases are investigated:
1) Lm=0.5Lm1=0.5Lm2=Lm-lin (non-saturated conditions) IV. CONCLUSIONS
2) Lm=0.5Lm1=0.5Lm2= Lm-sat (saturated conditions)
This letter has shown that the T model may fail to reproduce
3) Lm=0.5Lm1=Lm-sat, 0.5Lm2=Lm-lin (Lm and Lm1 saturated
ferroresonance measurements, while the  model predicts the
and Lm2 non-saturated)
measurements adequately in all tested cases.
The saturation status (instantaneous flux) depends on the
instantaneous voltages applied to the non-linear inductances; V. REFERENCES
see (1) to (3). In the first case, it is assumed that both models [1] Slow Transients Task Force of the IEEE. Modeling and Analysis of
are working in the linear part of the magnetizing curve. The System Transients Using Digital Programs Working Group “Modeling
second case is when both T and π models are saturated. Due to and Analysis Guidelines for Slow Transients--Part III: The Study of Fer-
the leakage inductance between the magnetizing branches in roresonance,” IEEE Trans. Power Del., vol. 15, pp. 255–265, Jan. 2000.
the π model, there are differences between Vn1 and Vn2. Differ-
 3

[2] F. de León, P. Gómez, J. A. Martinez-Velasco, and M. Rioual, Chapter

4, “Transformers” in Power System Transients: Parameter Determina-
tion, CRC Press, Boca Raton FL, 2009, pp. 177-250.
[3] F. de León, A. Farazmand, and P. Joseph, “Comparing the T and 
Equivalent Circuits for the Calculation of Transformer Inrush Currents”,
paper accepted for publication, IEEE Trans. on Power Del, July 2012.
 1

Leakage Inductance Design of Toroidal

Transformers by Sector Winding
Francisco de León, Senior Member, IEEE, Sujit Purushothaman, Member, IEEE, and Layth Qaseer


Abstract— Toroidal transformers are commonly used in power electronics applications when the volume or weight of a component
is at a premium. There are many applications that require toroidal transformers with a specific leakage inductance value. A trans-
former with a large (or tuned) leakage inductance can be used to eliminate a (series) filter inductor. In this paper a procedure to con-
trol the leakage inductance of toroidal transformers by leaving unwound sectors in the winding is presented. Also a simple formula is
obtained in this paper that can be used to design transformers with a specific leakage inductance value. The leakage inductance for-
mula is expressed as a function of the number of turns, the geometrical dimensions of the toroidal transformer, such as: core internal
diameter, external diameter and height, and the angle of the unwound sector. The formula of this paper has been obtained and vali-
dated from laboratory experiments and hundreds of three-dimensional finite element simulations. The techniques described in this
paper will find applications in the design of transformers that in addition of providing voltage boosting need to double as filters.

Index Terms— Leakage inductance, toroidal transformers, sector winding.

I. INTRODUCTION

TOROIDAL transformers with enlarged leakage inductance find applications in several power electronics devices that require a
transformer with a specified leakage inductance value. For example, a transformer with a large leakage inductance can be used
to eliminate a series inductor for filtering or tuning. Among the applications we can find a number of converters [1]-[4] and elec-
tromagnetic noise reduction transformers [5]-[10]. Particular leakage inductance values for transformers are used to distribute
the power flow of parallel paths and to limit the short circuit currents [11].
Tape wound toroidal transformers made with grain oriented silicon steel are more efficient, smaller, cooler, and emit reduced
acoustic and electromagnetic noise when compared with standard transformers built on staked laminations [12]. Toroidal trans-
formers are commonly used in the power supply of audio, video, telecommunications, and medical equipment. These transform-
ers are finding new applications in small to medium size UPS systems and in the lighting industry (especially in halogen light-
ing). Aircrafts have also benefited from the advantages of toroidal transformers [13].
The equations for computing the leakage inductance of E-I transformers at 60 Hz are readily available [14], [15]. Also, avail-
able are analytical expressions for computing winding losses and leakage inductance for high frequencies [16], [17].
The theory for toroidal transformers is not nearly as advanced as the theory for E-I transformers. This may be because at this
moment toroids are restricted to small powers (tens of kVA) and low voltages (possibly up to a few kilovolts). In references [18]
and [19] an analytical study of the losses at high frequency was presented for toroidal inductors, but the leakage inductance was
not considered.

This work was supported by the U.S. Department of Energy under Grant DEOE0000072.
F. de León is with the Department of Electrical and Computer Engineering of Polytechnic Institute of New York University, Six Metrotech Center, Brook-
lyn, NY, 11201 (email: fdeleon@poly.edu).
S. Purushothaman is with FM Global Research, 1151 Boston Providence Turnpike, Norwood, MA 02062 (email: sujitp@ieee.org)
L. Qaseer is with Al-Khwarizmi College of Engineering, University of Baghdad, Baghdad, Iraq (email: laythqaseer@yahoo.com).
 2

Perhaps, due to the complexity of the winding, researchers have preferred numerical solutions such as finite elements [20],
[21]. There exists a semi-empiric formula for computing the leakage inductance of small toroidal common mode chokes [22].
However, in all our cases the formula in [22] predicted erroneous values. We should mention that there is a substantial differ-
ence in the sizes of our transformers and those of [22].
In [23] an analytical formulation for computing the leakage inductance of toroidal transformers with circular cross-sectional
area is derived elegantly from the solution of Maxwell Equations. In [23] the toroidal core is opened and elongated to form a
linear rod with circular cross-sectional area and Fourier techniques are applied (this is possible because the rod is terminated
with magnetic end planes, which are replaced by images on the infinite rod). This works well in [23] because the toroids are
very small and the windings, which never overlap, cover only a small portion of the core perimeter. The transformer cores of
this paper are much larger and the windings overlap. Additionally the cores here do not have circular cross-sections.
A technique to enlarge the leakage inductance using inter-winding spacing and magnetic insets is given in [24]. The tech-
nique is highly controllable and can achieve large increases in leakage inductance; however, the transformer becomes larger,
heavier and more expensive. Sector winding, as advanced in this paper, produces very large increases in the leakage inductance
at virtually no added cost or weight. The method of [24] is applicable for relatively small leakage inductance gains, say for a
target increase of up to 5 times the natural (or minimum) leakage inductance (L0). The method promoted in this paper will find
applications when the desired leakage inductance is several orders of magnitude larger than the natural value.
Recently, in [25] a turn-by-turn formulation to compute the leakage inductance in common mode chokes was presented. A
rectangular turn is broken into four straight line conductors and approximate solutions on infinitely long geometries are used for
each region. Thus, the inner conductor is modeled as an eccentric conductor inside of a ferromagnetic cylinder. Similarly, the
outer conductor is represented as being outside the ferromagnetic cylinder. The lateral conductors are considered as filamentary
currents on top of an infinite ferromagnetic plane with the method of images. The method of [25] is applicable to toroids with a
few thick turns that can be wound in only one layer (for example common mode chokes), but it is not applicable to multi-layer
transformers. The frequency dependency is considered by including the resistances and the capacitances producing a wideband
circuital model. Previously, in [26] a method to measure the leakage inductance of multi-winding chokes was presented. A mod-
el to describe the terminal behavior is given, but there are no equations to compute the parameters from dimensions.
The objectives of this paper are two: First is to present a methodology to increase the leakage inductance of toroidal trans-
formers by leaving unwound sectors in the windings (see Fig. 1). Second is to propose an equation for the calculation of the
leakage inductance suitable for a design program.
Although toroidal transformer manufacturers know that leaving unwound sectors increases the leakage inductance, the de-
sired leakage inductance value is obtained by trial and error. In this paper the transformer leakage inductance is expressed as a
function of the number of turns (N), the geometrical dimensions of the toroidal transformer, internal diameter (ID), external di-
ameter (OD) and height (HT), and the angle of the unwound sector (θ).
This paper deals with a wide range of power transformer sizes of rectangular cross-sectional area. The core dimensions cover
the following range: height from 1 to 6 inches; external diameter between 4 and 13 inches; and internal diameter from 1 to 10
inches. These combinations cover most of the power conditioning application today from one kVA to perhaps one hundred kVA
(depending on the switching frequency). We have only experimented with unwound sector angles from 30º to 180º. It is quite
possible, however, that the equations of this paper are applicable to much larger transformers with larger unwound angles. A few
numerical experiments shown below indicate this, but more research is needed to make stronger claims.
The formula proposed in this paper is obtained from the observation of the behavior of the leakage inductance when the con-
struction parameters of toroidal transformers are varied. More than 400 3D FEM (Finite Elements) simulations have been per-
 3

formed to cover a very wide range of applications. Over 20 prototypes were built to validate the FEM simulations and proposed
formula.

D O
Core
Fig. 1. Toroidal transformer with sectored windings.

II. THEORETICAL CONSIDERATIONS

To make the presentation accessible to wider audiences and to establish the nomenclature, we start by presenting the basic
concepts of leakage flux for toroidal transformers. Two geometrical arrangements are discussed: toroidal transformers that are
wound around the 360º and toroids with sectored windings. Leakage flux is defined for a pair of windings as the flux that links
only one winding and does not link the other winding. The corresponding leakage inductance is obtained in the laboratory
through the short circuit test, which consists of feeding a winding with rated current when the other winding has its terminals
short-circuited. The test can be simulated with FEM to obtain the leakage inductance. Additionally, with simulations one can
fully eliminate any influence from the magnetizing current, while the short circuit-test does not fully cancel the magnetizing
flux.

A. Toroidal Transformers with 360º Windings

The leakage flux in a toroidal transformer, whose windings are one on top of the other for the entire 360º, is produced by the
current in the windings that are equal in magnitude (i.e. N1I1 = N2I2), but opposite in direction. By forcing N1I1 = N2I2 there is no
(magnetizing or leakage) flux in the core. As shown in Fig. 2 the leakage flux does not start nor it ends in the core, but closes in
itself. The left-hand quadrant shows the surface plot of the distribution of the magnetic flux density while the right-hand quad-
rant shows the direction of the streamlines (concentric circles). Note that most of the leakage flux is in the insulation between
the windings; some flux is also present in the windings, but there is no flux outside the region occupied by the windings. The
leakage inductance for such geometry is computed in [24] from the energy stored yielding:

N 2 0 5
L0  i i a  i g  i b 
2 i 1
(1)

where variables i, i, i, and i are computed from the radii of the windings and include the factors of partial linkage fluxes in
the windings; a, b and g are the thicknesses of the inner winding, the outer winding and the insulation layers, respectively; all
the details are given in the Appendix.
 4

core
insulation

inner outer
winding winding
Fig. 2. Axial view of a toroidal transformer with windings covering 360. The left-hand quadrant shows the surface plot of the distribution of the magnetic flux
density while the right-hand quadrant shows the direction of the streamlines (concentric circles).

B. Sectored Wound Toroidal Transformers

In sectored wound transformers, i.e. when the windings do not cover the entire 360º, the leakage flux follows a completely
different path. Fig. 3 shows the top view of the leakage flux distribution. One can see that in this case the path of the leakage
flux includes a section of the core. The amount of leakage flux that a winding links depends on the sector that is not wound.
From Fig. 3, it is possible to see that many lines of flux only link partially the winding. We make the remark that the shape of
the leakage flux does not change significantly as the angle of the wound sector varies. However, the intensity of the leakage
flux increases substantially as the unwound angle increases. It should be mentioned that the flux in the core contributes very
little to the leakage inductance since the energy stored depends on the square of the magnetic field strength (H), which is very
small in the core due to its high permeability.

Fig. 3. Top view of the distribution of the leakage flux in a sectored wound transformer.
 5

III. INITIAL EXPERIMENTATION

A first set of prototypes were built consisting of 7.25 kVA transformers V1 = 215 V, V2 = 1,928 V. These transformers are
used in a PWM application to drive a sonar amplifier. A standard toroidal transformer design for the specified power and volt-
age levels has a leakage inductance of under 10 H. For those conditions an external series inductor of around 800 H is needed
to help filtering the input of the amplifier at 450 Hz. Alternatively, we designed a transformer with increased leakage induct-
ance. The transformer parameters are N1 = 97 turns, N2 = 870 turns. The core dimensions are: OD = 175 mm, ID = 100 mm,
HT = 45 mm.
Table I shows the total leakage inductance, referred to the low voltage side (N=97), of a set of prototypes built with equal
unwound sectors in both windings, but displaced 180; see Fig. 4. As a reference, note that the magnetizing inductance of these
toroidal transformers is about 1 H, which is much larger than the natural inductance of L0 = 9.3 H (for θ =0º) and even substan-
tially larger than the largest leakage we measured of 2.6 mH resulting from sectored windings (for θ =180º) .
Fig. 5 shows the variation of the leakage inductance with respect to the unwound angle, which seems to be perfectly quadrat-
ic. Therefore, added to the plot of Fig. 5 there is a fitted quadratic equation of the form:

L  K2 (2)

For this example, K = 7.203102 when the unwound angle θ is given in degrees and L in μH.

TABLE I
MEASURED LEAKAGE INDUCTANCE VERSUS UNWOUND ANGLE
Point Angle 1 = 2 L (measured)
[Degrees] [H] N=97
0 0 L0 = 9.3
1 15 17.6
2 30 56.7
3 45 151
4 65 320
5 80 499
6 100 777
7 120 1032
8 180 2600

Fig. 4. Toroidal transformer with 90 sectored windings displaced 180.

 6

Fig. 5. Fitting a quadratic function to the experimental data.

It is difficult to control the inter-turn spacing with high speed winding machines and overlapping frequently occurs. Howev-
er, “messy” windings when are elements of a sector winding strategy have relatively little effect in the leakage inductance (pro-
vided that they cover certain angle). A few experiments using “bank winding”, which consists in purposely producing overlap-
ping by changing the rotation direction around the rollers, show very little increase in the leakage inductance. However, to ob-
tain consistent leakage inductance values, it is important to precisely control the unwound angle. For this, a physical barrier be-
yond which the winding cannot pass is used.

IV. SYSTEMATIC EXPERIMENTATION

A set of eleven prototypes was built with the purpose of shedding light on the parameters influencing the value of K. This set,
in addition to varying the unwound sector, also included variation of other geometric parameters of the core i.e. ID, OD and HT.
Table II gives the geometric details of the prototypes along with the leakage inductance obtained in the laboratory with short
circuit (SC) test. Measurements with an LCR meter (at 60 Hz) confirmed the results of the SC tests. In Table II the value of L0
has been added as reference. One can appreciate that L0 is negligible for unwound angles of 90º and larger. Prototypes 1, 2 and 3
have all parameters but ID constant. These three prototypes can be used to study the effect of ID on the leakage reactance. Simi-
larly, prototypes 3, 4 and 5 can be used to study effect of the variation of OD on the transformer leakage. Height variations can
be studied with prototypes 5, 6 and 7. All prototypes have 400 turns on each winding.

TABLE II
LEAKAGE INDUCTANCE COMPARISON BETWEEN FEM AND SHORT CIRCUIT (SC) TESTS ON PROTOTYPES WITH N=400
Core Dimensions [inch] Angle θ L0 FEM L SC test L %
No
OD ID HT [deg] [mH] [mH] [mH] difference
1 10 8 2 180 0.322 42.71 45.27 -5.64
2 10 6 2 180 0.287 47.25 52.6 -10.17
3 10 4 2 180 0.252 51.40 53.14 -3.27
4 8 4 2 180 0.215 42.18 42.27 -0.21
5 6 4 2 180 0.179 32.72 36.65 -10.72
6 6 4 3 180 0.115 38.85 39.28 -1.10
7 6 4 1 180 0.363 26.24 28.11 -6.64
8 6 4 1 150 0.363 17.34 18.6 -6.78
9 6 4 1 110 0.363 8.80 9.14 -3.72
10 6 4 1 90 0.363 5.75 5.67 1.48
11 6 4 1 40 0.363 1.15 1.08 6.88
 7

Although we found very little effect on the core losses at 60 Hz due to sector winding, it has been found in [27] that the core
losses increase considerably due to the orthogonal flux in cut tape-wound cores at high frequencies. Therefore, although the
techniques of this paper are directly applicable to ferrite cores over a wide frequency range, further investigation is needed to
gauge the effect on losses for uncut tape-wound cores at high frequencies.
Measurements with the LCR meter at 1 kHz show an average reduction in the leakage inductance of about 12% from the val-
ue at 60 Hz; the larger the transformer, the larger the reduction. Further research will be carried out to model the frequency de-
pendency of the leakage inductance in sectored winding toroidal transformers.

V. FEM SIMULATIONS

Three-dimensional (3D) finite element simulations are performed to generate additional cases needed for the derivation of a
mathematical model. The leakage inductance is computed from the total energy stored in the magnetostatic field when one
winding is fed with unity current in one direction and the second is fed with unity current in the opposite direction. This effec-
tively eliminates any effect of the magnetizing current since N1I1 = N2I2 is strictly enforced. A total of 420 different transformer
configurations were analyzed with 3D FEM simulations.
Even though the toroidal core is symmetric around the central axis, the windings are not. Each winding exists for 360º– θ
degrees around the central axis as shown in Fig. 1; moreover the core height is not infinite in depth. Hence an axisymmetric or a
two-dimensional (2D) model cannot be used to represent a sector wound toroidal transformer.
The windings are modeled as thin sheets carrying currents in opposite direction to simulate the conditions of the short-circuit
test needed to measure the leakage inductance. The windings were initially modeled as volume regions with finite thickness hav-
ing an impressed current density J, but it was found from many experiments that the coil thickness played only a minor role in
the leakage inductance when there is an unwound sector of at least 30º. Hence the optimum FEM simulations use a current sheet
to represent the windings. A cross section of the FEM model is presented in Fig. 6. It must be noted that such a 3D model con-
sists of 100,000 to 200,000 second order finite elements and takes 30 minutes to solve using a server that has 24 cores in its
CPU running at 3.33 GHz each as well as 96 GB of DDR3 RAM.
Table II shows the comparison between the experimental results and the corresponding 3D FEM simulations. One can ap-
preciate that the simulations yield very good results when compared to the experiments. The small differences are attributed to
manufacturing tolerances in the prototypes. Fig. 7 shows cuts of the front and top views of the distribution of the magnetic flux
density.
The surface current densities Kvertical and Khorizontal are chosen such that the total current is the same (N1I1=N2I2). While Khorizontal
is a function of spatial coordinates, Kvertical is constant in magnitude and is not a function of spatial coordinates. In a completely
wound (θ = 0º) transformer, the leakage flux flows through the inter-winding gap, g and hence is a critical factor contributing in
the leakage inductance; see [24]. In a sectored wound transformer, the leakage flux is dictated by θ, ID, OD and HT.
 8

Fig. 6. Cross section of the FEM model.

Fig. 7. FEM flux density streamline plot. (a) Front view; (b) Top view.

VI. MATHEMATICAL MODEL

The validation of the FEM simulations against experimental results, as shown in Table II, enables the derivation of a mathe-
matical formula for the calculation of the leakage inductance of sectored winding toroidal transformers based on the results of
FEM simulations. In this section a double regression method is employed to obtain a simple formula for the leakage inductance.
From numerous tests on transformers having the same number of primary and secondary turns of 400, it was found that there
exists a linear relationship between the leakage inductance (L) and outer diameter of the core (OD). This can be written in the
following form:

L  b10  1 OD (3)

where L is the leakage inductance and β1 is the contribution factor for outer diameter. Fig. 8 shows the relationship between the
leakage inductance and the outer diameter, keeping all other parameters fixed (ID = 4", HT = 2"). Results are plotted for (un-
wound) sector angles of 60o, 70o and 80o.
 9

Fig. 9 shows the relationship between the leakage inductance and the inner diameter (ID) when keeping the other parameters
fixed (OD = 10", HT = 2"). Results are plotted for sector angles of 60o, 70o and 80o. It can be seen from Fig. 7 that the leakage
inductance varies linearly in inverse proportion to the inner core diameter (ID), which can be expressed as

L  b20   2 ID (4)

where β2 is the contribution factor for inner diameter.

We have also observed that the leakage inductance varies linearly with core height (HT) as:

L  b30 3 HT (5)

where β3 is the contribution factor for core height. Fig. 10 shows the relationship between the leakage inductance and the core
height when keeping all other parameters fixed (OD = 6", ID = 4"), for sector angles of 90o, 100o and 110o.
Finally, we note that the leakage inductance varies with the unwound sector angle as a quadratic function. This is given as:

L  b40  c1 2 (6)

where θ is the sector angle in radians. Fig. 11 shows the relationship between the leakage inductance and the unwound sector
angle for a transformer keeping other parameters fixed (OD = 6", ID = 4"). Results are plotted for three cases with core
height = 2", 3" and 4". The sector angle is varied from 10º to 350º to cover the entire spectrum. The slope m = 2 of the lines in
the log-log plot confirms the quadratic variation.
Consistent with theory, all analytical formulas for the calculation of inductances reveal that they depend of the square of the
number of turns. Combing the above four equations (3)–(6) into a single equation gives an inductance formula as a function of
the inner diameter, outer diameter, core height, sector winding angle and the number of turns as follows:

L  L0  0 N 2  1 OD   2 ID  3 HT  (7)

where L0 is the leakage inductance for a transformer with complete 360º windings or sector angle of 0º. The procedure to evalu-
ate L0 has been given in [24].
Exhaustive analysis of the numerical results has yielded that the contribution factors, βi are quadratic functions of the sector
angle θ as follows:
1  k1 2
 2  k2 2 (8)
3  k3 2

1, 2 and 3 correspond to the slopes of the lines of Figs. 8, 9, and 10, respectively. Hence (7) simplifies to:

L  L0  0 N 2  k1 OD  k2 ID  k3 HT   2 (9)

OD, ID and HT are given in inches, θ in degrees and L is in milli-Henry. The thickness of the windings a and b in equation (1)
only affects L0.
 10

9 L = 0.8092 OD + 0.7495
Theta 60 R² = 0.9998
8 Theta 70
Theta 80

Leakage Inductance [mH]

7 L = 0.6175 OD + 0.5747
R² = 0.9998
6

5 L = 0.4542 OD + 0.4289
R² = 0.9998
4

2
4 5 6 7 8 9 10
Core Outer Diameter [inch]

Fig. 8. Variation of leakage inductance with core outer diameter. The dots correspond to the simulated values. The trend lines and their equations are also pre-
sented.
10
Theta 60
L = -0.3954 ID + 10.423 Theta 70
9 R² = 0.9917 Theta 80
Leakage Inductance [mH]

7 L = -0.3081 ID + 7.9823
R² = 0.9927

6
L = -0.2332 ID + 5.9053
5 R² = 0.9939

3
0 2 4 6 8
Core Inner Diameter [inch]
Fig. 9. Variation of leakage inductance with core inner diameter. The dots correspond to the simulated values. The trend lines and their equations are also pre-
sented.

20 theta 90
L = 1.9907 HT + 6.8926
18 theta 100 R² = 0.9995

theta 110
Leakage Inductance [mH]

16
L = 1.633 HT + 5.6145
R² = 0.9995
14

12 L = 1.3183 HT + 4.4889
R² = 0.9995
10

4
0 1 2 3 4 5 6
Core Height [inch]

Fig. 10. Variation of leakage inductance with core height. The dots correspond to the simulated values. The trend lines and their equations are also presented.
 11

1000
core height = 2
L = 0.0015 1.9801

Leakage inductance [mH]

core height = 3
100 R² = 0.9963
core height = 4

10

0
10 100
Unwound andgle of windings [deg]

Fig. 11. Variation of leakage inductance with unwound sector winding angle. The dots correspond to the simulated values. The trend line for core height = 3
inch and its equation is also presented.

6
Beta 1
β3 = 0.0002 2 - 0.0517
5 Beta 2 R² = 0.9979

4 Beta 3 β1 = 0.0001 2 - 0.0542

R² = 0.9988
3

0
0 5000 10000 15000 20000 25000 30000
-1
β2 = -0.00005 2 - 0.0248
-2 R² = 0.9976
Square of unwound Sector Angle [Deg2]

Fig. 12. Variation of βi with unwound sector winding angle. The dots correspond to the simulated values. The trend lines and their equations are also presented.

The values of k1, k2 and k3 in (9) are evaluated by a two step regression readily available in Excel. Multiple cases are generat-
ed using the FEM model described in Section V. The geometric parameters ID, OD and HT are varied in steps for every sector
angle θ.
A linear regression is first performed on data with constant θ and the values of β1, β2 and β3 are evaluated. This is repeated for
various values of sector angle θ, yielding different values of βi. 210 of the 420 cases generated by the FEM model were used for
this regression. It must be noted that the coefficient of determination, R2 is larger than 99% for all the cases indicating an excel-
lent fit. Fig 12 shows the variation of βi with respect to the square of the sector angle.
Fig. 12 confirms the quadratic variation of βi with sector angle given in (8). The second regression is performed on the data
plotted in Fig. 12 to satisfy (8) yielding the values of ki as follows:

Units for OD, ID, HT k1 k2 k3

-5 -5
mm 2.6444×10 -1.104×10 3.178×10-5
inch 6.7168×10-4 -2.8043×10-4 8.0723×10-4
 12

k1, k2 and k3 are the contributing factors of the geometric parameters of the core OD, ID and HT, respectively. When the geo-
metric dimensions of the transformer are given in mm, k1, k2 and k3 have units of degree–2.
The double regression used in this paper (to model the leakage inductance for sectored winding transformers) gives a good
balance between simplicity and accuracy. Equation (9) is very simple, yet sufficiently accurate for engineering design. We have
investigated the effects of fitting curves for other variables and ratios. For example, it was experimented with ratios of OD/ID,
differences (OD-ID), compound ratios as (OD-ID)/(OD+ID), and their powers (squares and roots). Slightly more accurate re-
sults can be obtained with some combinations, however the resulting equations are substantially more intricate than (9).

VII. RESULTS

Table III presents the comparison between 24 of the close to 250 cases used for validation of the proposed model against
FEM simulations. These cases are provided to cover a wide range of core dimensions. For each set of OD, ID and HT, the un-
wound sector angle θ, is set to 60º, 120º and 240º. Most of the results match very well. It was observed that transformers with
cross sectional area close to a square aspect ratio (HT  (OD–ID)/2) have errors smaller than 5%. Tall and flat looking trans-
formers have errors smaller than 10%. If there are no special constraints on dimensions, square aspect ratio is preferred because
the turn length is shorter. The value of L0 has been included in Table III. One can appreciate that L0 in all these cases is negligi-
ble, but this is not always the case. For smaller angles, say up to 60º, L0 may play a role. Note that the values presented in Table
III do not match with the same degree of accuracy as the cases used to fit the equations (as presented in Figs. 8 to 12). This is so
because Table III gives the extremes used to validate the model; these cases represent the worst case scenarios.
To gauge the validity of the formulas for large transformers, a set of FEM simulations for unrealistically large transformers
were performed. We used: OD = 6 m, ID = 4 m and HT = 1 m for unwound angles of 60º, 120º, 240º, and 270º. We found dif-
ferences between FEM and (9) of only -6.80%, -1.56%, -2.08%, and -6.58%, respectively.

VIII. CONCLUSIONS

A methodological technique to increase considerably the leakage inductance of power toroidal transformers by leaving un-
wound sectors has been developed. Additionally, a formula to compute the leakage inductance for sectored wound transformers
has been derived from the observation of its behavior through hundreds of 3D FEM simulations. The leakage inductance is
computed with a simple formula from the physical quantities of the transformer: number of turns and core dimensions: internal
and external diameter, height, and the angle of the unwound sector. Therefore, the equation is suitable for implementation in
transformer design programs or even hand calculations. The FEM simulations and the formula have been corroborated experi-
mentally with over 20 prototypes of varied sizes and winding conditions.

ACKNOWLEDGMENTS

The first author would like to thank the people (co-workers and students) who have worked with him in this project over the
past 12 years; in chronological order: Valeriu Tatu, Salvador Magdaleno, Ivan Hernandez, Noel Augustine, and Chandira Prab-
hu.
 13

TABLE III
COMPARISON BETWEEN FEM AND PROPOSED MODEL (9)

Core Dimensions [inch] Unwound L[mH]

%
No Angle θ L0 Sector Model Total
OD ID HT FEM difference
[deg] (1) L - L0 in (9) (9)
1 60 2.48 0.049753 2.33 2.380 6.05
2 1 120 9.11 0.049753 9.30 9.350 -2.09
3 240 35.19 0.049753 37.22 37.270 -5.77
1
4 60 4.57 0.003843 4.08 4.084 10.72
5 4 120 16.01 0.003843 16.32 16.324 -1.94
6 240 61.30 0.003843 65.26 65.264 -6.46
4
7 60 1.88 0.068701 1.92 1.989 -2.13
8 1 120 7.59 0.068701 7.68 7.749 -1.19
9 240 29.91 0.068701 30.72 30.789 -2.71
3
10 60 3.76 0.011310 3.67 3.681 2.39
11 4 120 14.63 0.011310 14.69 14.701 -0.41
12 240 57.43 0.011310 58.77 58.781 -2.33
13 60 5.91 0.079918 6.19 6.270 -4.74
14 2 120 24.30 0.079918 24.76 24.840 -1.89
15 240 94.87 0.079918 99.05 99.130 -4.41
4
16 60 8.46 0.023534 8.53 8.554 -0.83
17 6 120 34.28 0.023534 34.11 34.134 0.50
18 240 132.32 0.023534 136.45 136.474 -3.12
12
19 60 4.78 0.101876 5.18 5.282 -8.37
20 2 120 20.20 0.101876 20.70 20.802 -2.48
21 240 79.03 0.101876 82.81 82.912 -4.78
9
22 60 7.32 0.032882 7.51 7.543 -2.60
23 6 120 30.37 0.032882 30.05 30.083 1.05
24 240 117.49 0.032882 120.21 120.243 -2.32

APPENDIX – CALCULATION OF L0

In this section, the necessary information to compute L0 using (1) is reproduced from [24]. The geometrical arrangement and
the definition of all variables are given in Fig. 13. The coefficients i, i, i, and i for the different sections are given in Table
V. The following relationships are needed to complete the information:


R m j  Rj  Rj 1 / 2  (11a)
t1  3a  4  s  b  g  (11b)
t2  g  2  s  b  (11c)
t3  3b  4s (11d)
 14

TABLE V
COEFFICIENTS FOR THE DIFFERENT COMPONENTS OF THE LEAKAGE INDUCTANCE FORMULA GIVEN IN (1)

Coefficient
Section ηi i i i
Rm 1 Rm 2 Rm 3
1 h
3R22 2 R2 R3 3R32
Rm 7 Rm 6 Rm 5
2 h
3R72 2 R6 R7 3R62
Re2  Ri2 1 1
3 1
Re Ri 3 3
1 Rm1t1 Rm 3t3
4 Rm2 t2
2R22 6 6
1 Rm 7 t1 Rm 5t3
5 Rm6 t2
2R62 6 6

REFERENCES
[1] J. O. Krah and J. Holtz, “Total Compensation of Line-Side Switching Harmonics in Converter-Fed AC Locomotives”, IEEE Transactions on Industry
Application, Vol. 31, No.6, November/December 1995, pp. 1264-1273.
[2] S. Saetieo, R. Devaraj, and D. Torrey, “The Design and Implementation of a Three-Phase Active Power Filter Based on Sliding Mode Control”, IEEE
Transactions on Industry Application, Vol. 31, No.5, September/October 1995, pp. 993-1000.
[3] N. Kutkut, D. Divan, and R. Gascoigne, “An Improved Full-Bridge Zero-Voltage Switching PWM Converter Using a Two-Inductor Rectifier”, IEEE
Transactions on Industry Application, Vol. 31, No.1, January/February 1995, pp. 119-126.
[4] M. Kheraluwala, R. Gascoigne, D. Divan, and E. Baumann, “Performance Characterization of a High-Power Dual Active Bridge dc-to-dc Converter”,
IEEE Transactions on Industry Application, Vol. 28, No.6, November/December 1992, pp. 1294-1301.
[5] A. Yagasaki, “Highly Improved Performance of a Noise Isolation Transformer by a Thin-Film Short-Circuit Ring”, IEEE Transactions in Electromagnetic
Compatibility, Vol. 41, No.3, August 1999, pp. 246-250.
[6] S. Ogasawara, H. Ayano, and H. Akagi, “An Active Circuit for Cancellation of Common-Mode Voltage Generated by PWM Inverter”, IEEE Transactions
on Power Electronics, Vol. 13, No. 5, September 1998, pp. 835-841.
[7] T. Yanada, S. Minowa, O. Ichinokura, and S. Kikuchi, “Design and Analysis of Noise-Reduction Transformer Based on Equivalent Circuit”, IEEE Trans-
actions on Magnetics, Vol. 34, No.4, July 1998, pp. 1351-1353.
[8] T. Yanada and O. Ichinokura, “High Performance Step-Down Noise-Reduction Transformer Constructed with C-Core”, IEEE Transactions on Magnetics,
Vol. 33, No.5, September 1997, pp. 3331-3333.
[9] S. Ogasawara, and H. Akagi, “Modeling and Damping of High-Frequency leakage Currents in PWM inverted-Fed AC Motor Drive Systems”, IEEE
Transactions on Industry Applications, Vol. 32, No. 5, September/October 1996, pp. 1105-1114.
[10] F. Forest, E. Laboure, T. A. Meynard, and V. Smet, “Design and Comparison of Inductors and Intercell Transformers for Filtering of PWM Inverter Out-
put”, IEEE Transactions on Power Electronics, Vol. 24, No. 3, March 2009, pp. 812-821.
[11] IEEE Recommended Practice for Protection and Coordination of Industrial and Commercial Power Systems, IEEE Standard 242-1986, Feb. 1986.
[12] Plitron Manufacturing Inc, “Toroidal Transformer Catalog”, www.plitron.com.
[13] W. T. McLymann, “Transformer and Inductor Design Handbook”, Marcel Dekker, 1998.
[14] K. Karsai, D. Kerenyi, and L. Kiss, “Large Power Transformers”, Elsevier, 1987, p 95.
[15] G. Slemon, “Electric Machines and Drives”, Addison-Wesley Publishing Company, Inc., 1992, pp. 252-255.
[16] R. Stoll, “The Analysis of Eddy Currents”, Clarendon Press, Oxford, 1974.
[17] F. de León and A. Semlyen, "Time Domain Modeling of Eddy Current Effects for Transformer Transients," IEEE Transactions on Power Delivery, Vol. 8,
No. 1, January 1993, pp. 271-280.
 15

[18] K. W. E. Cheng and P.D. Evans, “Calculation of Winding Losses in High-Frequency Toroidal Inductors Using Single Strand Conductors”, IEE Proceed-
ings on Electric Power Applications, Vol. 141, No. 2, March 1994, pp. 52-62.
[19] K. W. E. Cheng and P. D. Evans, “Calculation of Winding Losses in High Frequency Toroidal Inductors Using Multistrand Conductors”, IEE Proceedings
on Electric Power Applications, Vol. 141, No. 5, September 1995, pp. 313-322.
[20] R. Prieto, J. A. Cobos, V. Bataller, O. Garcia, and J. Uceda, “Study of Toroidal Transformers by Means of 2D Approaches”, 28th Annual IEEE Power
Electronics Specialist Conference PESC’97, 22-27 June, 1997, pp.621-626.
[21] R. Prieto, V. Bataller, J. A. Cobos, and J. Uceda, “Influence of Winding Strategy in Toroidal Transformers”, 24th Annual IEEE Conference of the Indus-
trial Electronics Society IECON’98, August 31 – September 4, 1998, pp. 359-364.
[22] M. Nave, “On Modeling the Common Mode Inductor”, 1991 IEEE International Symposium on Electromagnetic Compatibility, August 1991, pp. 452-
457.
[23] W.G. Hurley and D. J. Wilcox, “Calculation of Leakage Inductance in Transformer Windings”, IEEE Transactions on Power Electronics, Vol. 9, No. 1,
January 1994, pp. 121-126.
[24] I. Hernández, F. de León, and P. Gómez, “Design Formulas for the Leakage Inductance of Toroidal Distribution Transformers”, IEEE Transactions on
Power Delivery, Vol. 26, No. 4, October 2011, pp. 2197-2204.
[25] M. Kovacic, Z. Hanic, S. Stipetic, S. Krishnamurthy, and D. Zarko, "Analytical Wideband Model of a Common-Mode Choke", IEEE Transactions on
Power Electronics, vol.27, no.7, pp.3173-3185, July 2012.
[26] A. J. Binnie and T. R. Foord, "Leakage Inductance and Interwinding Capacitance in Toroidal Ratio Transformers", IEEE Transactions on Instrumentation
and Measurement, vol.16, no.4, pp.307-314, Dec. 1967.
[27] B. Cougo, A. Tuysüz, J. Muhlethaler, and J. W. Kolar, "Increase of tape wound core losses due to interlamination short circuits and orthogonal flux com-
ponents," IECON 2011 - 37th Annual Conference on IEEE Industrial Electronics Society , pp.1372-1377, 7-10 Nov. 2011.

BIOGRAPHIES
Francisco de León (S’86–M’92–SM’02) received the B.Sc. and the M.Sc. (Hons.) degrees in electrical engineering from the National Polytechnic Institute,
Mexico City, Mexico, in 1983 and 1986, respectively, and the Ph.D. degree in electrical engineering from the University of Toronto, Toronto, ON, Canada, in
1992. He has held several academic positions in Mexico and has worked for the Canadian electric industry. Currently, he is an Associate Professor at the
Polytechnic Institute of NYU, Brooklyn, NY. His research interests include the analysis of power definitions under nonsinusoidal conditions, the transient and
steady-state analyses of power systems, the thermal rating of cables and transformers, and the calculation of electromagnetic fields applied to machine design
and modeling.

Sujit Purushothaman (S’09–M’13) received his B.E. degree in electrical engineering from Mumbai University (Sardar Patel College of Engineering), Mumbai,
India in 2005. His work experience includes testing and development of medium voltage switchgear for Siemens in Mumbai, India. He received his Master’s
and his Ph.D. degrees in electrical engineering from Polytechnic Institute of NYU in 2009 and 2011, respectively. He is currently a research engineer at FM
Global Research in Norwood, MA. His research interests include power system transients, damping of subsynchronous resonance, machine design and model-
ing, and thermal modeling of electrical devices.

Layth Qaseer received his B.Sc., M. Sc. and Ph.D. degrees from the University of Baghdad, Baghdad, Iraq in 1979, 1993 and 2004 respectively, all in electrical
engineering. Between 1979 and 2001 he has worked at national scientific research center and the ministry of industry. In 2005 he joined Al-Khwarizmi College
of Engineering, University of Baghdad, Baghdad, Iraq. He was a visiting professor in the department of electrical & computer engineering at the Polytechnic
Institute of New York University during the 2010-11 academic year. His areas of research interest include linear electric machines, electromagnetic systems,
electrical machine design, permanent magnet synchronous motors and modeling.
 1

Analysis, Modeling and Simulation of the

Phase-Hop Condition in Transformers:
The Largest Inrush Currents
Ashkan Farazmand, Francisco de León, Senior Member, IEEE, Kuang Zhang, Student Member, IEEE,
and Saeed Jazebi, Student Member, IEEE


Abstract—Inrush currents in transformers can have very dis- 200
ruptive effects such as: voltage sags, false tripping of the protec-
tive devices, and mechanical stresses in the transformer wind- 100

Voltage (V)
ings. This paper shows that there are operating situations that
may cause a transformer to draw abnormally high inrush cur- 0
rents. Examples include the normal operation of off-line UPS
systems, interruptions, voltage sags, and notching. These condi-
-100
tions may produce inrush-like currents of more than twice the
value of the “normal” maximum inrush caused by energizing at
voltage zero-crossing. For this condition, the term “phase-hop” is -200
10 20 30 40 50 60 70 80
used in this paper. Laboratory experiments performed on four Time (s)
different transformers (1 kVA) with varied characteristics show Fig. 1. Voltage wave shape of the phase-hop condition.
the impact of phase-hop in the magnitude of inrush currents. The
experiments are also used to validate the EMTP model used for As it will be explained below, the operation of off-line UPS
analysis of multiple cases. In addition, the behavior of the mag- systems, interruptions, voltage sags, and notching in power
netic flux in a transformer under phase-hop is investigated and systems can lead to a condition called “phase-hop” coined for
compared with different operating conditions using finite ele- the shape of the voltage wave shown in Fig. 1. When this
ments. The results of this paper have implications in transformer
condition occurs, there are two positive (or negative) semi-
design and in the operation and design of UPS systems to prevent
the damaging effects of phase-hop. cycles applied consecutively to the transformer. The maxi-
mum phase-hop current has been reported as an important
Index Terms—inrush currents, interruptions, phase-hop, design parameter by engineers of the leading manufacturers of
transformer modeling, UPS systems, voltage sags. UPS systems for transformers rated at 25 kVA [4].
Phase-hop causes the transformer core to go into a deep
I. INTRODUCTION saturation level and draws very large inrush-like currents.

P OWER quality problems are critical issues nowadays be-

cause of the increased use of power electronics loads. In-
terruptions and blackouts are the worst forms of power quality
Transformers and protections need to be designed to prevent
false tripping or damages during phase-hop. The large cur-
rents could also damage the UPS systems, or cause problems
problems. Blackout is a complete loss of supply voltage or in the switching operation of the rectifiers, since they may not
load current for longer than a minute [1]. Harmonics, inter- be designed for these abnormally large currents.
harmonics, power frequency variations, voltage unbalances, This paper introduces and investigates for the first time the
interruptions, notching, undervoltages, overvoltages, swells, effect of phase-hop on transformers. The study is performed
noise, dc offset, voltage fluctuations, and voltage sags are both experimentally and with validated computer simulations.
common power system operation phenomena which cause It is found that phase-hop currents can be over twice as large
power quality problems [2]. as the “normal” maximum inrush currents caused by switch-
In order to solve the aforementioned problems, Uninter- ing at zero crossing.
ruptible Power Supplies (UPS) are often used [3]. UPS are False tripping during phase-hop is more probable than dur-
designed to automatically provide emergency electricity to ing transformer energization because of the unpredictable tim-
critical loads in case of supply voltage failure. Some UPS sys- ing of this phenomenon. In practice, a common technique
tems also regulate or filter the utility power [1]. used to prevent false tripping of the protective devices during
transformer energization is to add a time delay. However, the
This work was supported by the U.S. Department of Energy under Grant occurrence of phase-hop is not predictable and a delay cannot
No. DEOE0000072. be applied.
A. Farazmand, F. de León, K. Zhang, and S. Jazebi are with the The correct estimation of phase-hop currents is important
Department of Electrical and Computer Engineering, Polytechnic University,
Six Metrotech Center, Brooklyn, NY, 11201. (e-mails: farazmand@nyu.edu,
for power system design. Inasmuch as their quantification is
fdeleon@poly.edu, kzhang02@students.poly.edu, jazebi@ieee.org). vital for UPS operation and design since UPS systems are
 2

precisely used to provide backup power, therefore false trip- controls, and inverters. In this condition, a brief short circuit
ping of vital loads could be disastrous. between two phases occurs [3], [6].

II. EFFECTS ON TRANSFORMER INRUSH CURRENTS OF POWER

SYSTEM ELECTROMAGNETIC PHENOMENA
The variation of the rms voltage from its nominal value is
described by two parameters: the magnitude of the voltage
change and its duration. Power system electromagnetic phe- Fig. 2. Off-line UPS performance when utility power is present (normal ac
nomena are classified in four main groups based on the dura- power mode).
tion of the disturbance: steady state variations, long duration
variations, short duration variations, and transients [2].
This section discusses how interruptions, voltage sags, and
notching in power systems can produce a phase-hop voltage to
be applied to transformers. In this section it is assumed that a
UPS system is not used to prevent these effects. Fig. 3. Off-line UPS performance when there is over/under-voltage or power
loss (inverter mode).
A. Interruptions
Interruptions are caused by transients that trigger utility
breakers or switches to open. A voltage interruption occurs
when the supply voltage decreases to less than 10% of its
nominal value in one or more phase conductors. The causes
for this phenomenon are: faults, component failure, switching, Fig. 4. On-line UPS system.
false breaker tripping, and malfunctions of control systems.
Depending on the duration of interruptions, they are classi- III. UPS SYSTEMS
fied in three types: momentary (0.5 cycle to 3 seconds), tem- UPS systems are intended to provide constant and regulat-
porary (3 seconds to 1 minute), and sustained (greater than 1 ed output voltage and power to critical loads regardless of
minute) [2]. The first two types are short duration variations power quality disturbances present in the mains. The objective
and the third is a long duration variation. The duration of the is to prevent voltage sags, power outages, impulses, noise,
interruption depends on the reclosing capability and speed of overvoltages or swells, harmonic distortions, frequency varia-
the protective device. Note that an interruption of exactly 0.5 tions, voltage fluctuations, and voltage surges [7], [8].
cycle produces the phase-hop voltage wave shape illustrated UPSs are classified into two groups: rotary and static. Ro-
in Fig. 1. tary UPSs normally use a diesel-fueled motor generator set
B. Voltage Sags and static UPSs use battery as the backup power source [1].
A voltage sag is a short duration decrease of the voltage Because there are several technical problems with rotary UPS
between 0.1 and 0.9 pu of the nominal voltage at the power systems, most of the modern UPSs are static [1]. There are
frequency for durations of 0.5 cycle to 1 min [2]. The IEC three kinds of UPSs: off-line, line interactive, and on-line.
word for this phenomenon is “dip” [5]. Sag durations are di- A. Off-line (Standby) UPS:
vided into three categories: instantaneous (0.5 to 30 cycles),
During the time when the utility power is present, off-line
momentary (30 cycles to 3 s), and temporary (3 s to 1 minute).
UPS systems pass the power directly to the load; the load is
The causes for this phenomenon are system faults, switching
not isolated from the mains. During this time, the battery
of large loads, and starting of large motors [2]. Voltage sags
backup is also charged and the inverter connected to the bat-
cause a partial phase-hop, but currents can be larger than the
tery is off (see Fig. 2).
“normal” inrush.
When the utility voltage is below a specified value or dur-
C. Notching ing a utility power outage, the UPS turns on its internal dc-ac
Notching is a repetitive steady state voltage disturbance inverter to produce ac power from the battery. In this case, the
lasting less than a half cycle. It represents a phenomenon that equipment is connected to the inverter output mechanically
is considered both a transient and a harmonic distortion since (see Fig. 3).
it occurs continuously and the frequency components related This method saves battery life by avoiding continuous
to it are high [2]. It can occur in opposite polarity to the main charging and discharging. However, as stated by most manu-
waveform. In this case, it is subtracted from the normal wave- facturers, there is a switch changeover time between 4 and 10
form. In an extreme case, notching may lead to a complete ms to engage the UPS during an interruption [1]. Practically,
loss of voltage for up to a half cycle [3] corresponding to the this delay can be as long as 25 ms depending on the time that
phase-hop wave of Fig. 1. it takes the UPS to detect the absence of utility voltage and
Notching can be produced during the commutating action transfer to the battery. Therefore, during the changeover time
from one phase to another in the normal operation of SCR- there is a voltage drop-out to the connected equipment and the
controlled equipment, such as three-phase converters, motor phase-hop condition is possible.
 3

B. Line-Interactive UPS: R1 and R′2 are the primary and the secondary ac resistances,
Line-interactive UPS is an off-line UPS connected with a respectively, and f=60 Hz. The applied voltage is 125 V rms.
tap-switching automatic voltage regulator (AVR). In this sys- The parameters computed from measurements are shown in
tem, when the power comes from the utility line, the AVR Table I. Data given for transformer Ta is for the inner-most
senses the UPS output voltage. When the utility voltage is low (first) winding. Hysteresis loops of the three transformers are
(utility brownout), the AVR automatically switches trans- obtained from Faraday’s Law integrating the induced voltage
former taps to increase the output voltage. When the utility to find the flux linkage as in [9].
voltage is large, the AVR reduces the output voltage. The set- The proper estimation of the air-core inductance is highly
up of this case is the same as the offline UPS (Figs. 2 and 3) important to compute the inrush current precisely. 3-D finite
element (FEM) simulations (using the commercial program
with the addition of a multi-tap variable voltage auto trans-
Maxwell 14) are carried out. The air-core inductance is calcu-
former after the utility block. In this case, the load is not com-
lated as follows [9], [11]:
pletely isolated from the mains power and therefore, phase-
hop can occur. 2W
Lair  core  (3)
C. On-Line UPS: I2
The on-line UPS system, as shown in Fig. 4, converts in- where W is the volume magnetic energy (computed from
coming ac power to rectified and regulated dc voltage and FEM), and I is the winding current. Table II presents the air-
then the inverter regenerates a regulated, clean, and sinusoidal core inductances of the four transformers. The air-core induct-
ac power from the dc voltage. Therefore, the load is isolated ances are used to complete the hysteresis loops. They are the
from the utility. This double conversion system leads to the slopes used to extend the hysteresis loops from the final
measured point (obtained from the open circuit test) to infini-
elimination of line noise, transients, harmonic distortion, and
ty. The model is implemented in the EMTP-RV [12].
voltage/frequency instability problems from the utility.
In this system, the load is always powered by the inverter V. MODEL VALIDATION AND WORST PHASE-HOPE CURRENTS
and the battery is connected to the dc bus. Therefore, this is a
To validate the model, laboratory experiments are per-
no-break system and there is no change-over time and phase-
formed on all four transformers under the worst possible
hop will not occur. This system provides a fully charged bat-
phase-hop conditions. In this case, the phase-hop condition
tery backup available at all times. It has the disadvantage of
occurs following the moment of energizing the transformer
shorter battery life because of the continuous charging and
using the zero-crossing switch. A zero-crossing and phase-hop
discharging of the battery. This UPS system is more expensive
switch is built and utilized in the laboratory to connect and
and less reliable than standby and line-interactive UPSs be-
disconnect transformers at specific time -instants. For the ze-
cause there are additional components connected in series.
ro-crossing condition, the switch energizes the transformer
IV. TRANSFORMER MODEL when the voltage of the ac power source crosses zero. For
phase-hop, the switch, in addition to energizing the transform-
In this paper, the π model is selected to represent the trans-
er at voltage zero-crossing, opens the connection between the
former [9]. Tests have been performed on four different trans-
second and third zero-crossings, thus re-establishing power at
formers (Ta, Tb, Tc, and Td) to obtain the parameters. Trans-
former Ta consists of four windings. In this paper, the inner- the third zero-crossing (see Appendix for more details).
most winding is called the first winding, the one after is called Fig. 5 shows the waveshapes and compares the results from
second winding, and so forth. The open circuit test is used to experiments and the model for the first winding of transformer
obtain the magnetizing parameters of the transformers as in Ta. The first peak values of inrush current (caused by the first
[9]. The leakage parameters of the transformers are obtained peak of the primary voltage) from experiment and simulation
accurately from the bucking test [10]. are 157.7 A and 162.5 A, respectively (difference of 3%). The
The total series ac resistance R1 + R′2 is computed from second peak (caused by the phase-hop voltage) from experi-
ment and simulation are 328.9 A and 330.2 A, respectively
PBK
R1  R2  2
(1) (difference of only 0.4%). Fig. 6 compares the results for
I BK transformer Td under the abovementioned condition. The dif-
Individual breakdown of the resistances is done based on ference between the peak currents of the model and the exper-
the dc resistance division between primary and secondary iment is 3.9% for the first peak and 3.3 % for the second peak.
windings obtained from the dc test. Total leakage inductance Note, however, that the inrush and phase-hop currents are
is computed from: much higher because transformer Td is a toroidal transformer.
2 Table III and IV compare the results of the first and second
1  VBK 
 
2
L s    R1  R2 (2) peaks of inrush current under the worst case of phase-hop for
2 f  I BK  all four transformers under study. Looking at Figs. 5 and 6
and Tables III and IV, one can observe a strong agreement
where PBK is the active power computed from the bucking test. between simulation and experimental results.
VBK and IBK, are the rms values of voltages and currents in the Note from Figs. 5 and 6 that the second positive peak of the
bucking test, respectively. Ls is the total leakage inductance. voltage in the phase-hop condition is smaller than the first
 4

peak. The reason for this is the existence of a large voltage higher than the first peak of inrush current, and under the
drop in the source resistance (Rsource = 0.1 ) caused by the worst case of phase-hop it is 2.41 times higher.
extremely large phase-hop currents. If the short-circuit power TABLE I
rating of the source were larger, higher inrush (and phase-hop) CIRCUIT PARAMETERS FOR DIFFERENT TRANSFORMERS
Transformer Ta Tb Tc Td
currents would occur.
Rating [kVA] 1 1 1 1
Primary voltage (V) & primary current (A)

Rated voltage 120 120 120 120

Primary current from model
300
Primary current
Primary current from experiment Turns ratio 1:1 1:1 1:1 1:1
Primary voltage from model
Primary voltage from experiment Construction Shell-type Shell-type Toroidal Toroidal
200 R1 [] 0.404 0.251 0.338 0.318
Primary Voltage
R’2 [] 0.345 0.324 0.288 0.271
100 Ls [mH] 0.254 0.692 8.843 0.232
Rm1 [] 2656 739 2149 2832
0
Rm2 [] 2656 739 2149 2832
-100 Lm1 [mH] 822 753 3339 2569
Lm2 [mH] 822 753 3339 2569
10 20 30 40 50 60 70 TABLE II
Time (ms) AIR CORE INDUCTANCES FOR THE FOUR TRANSFORMERS UNDER STUDY
Fig. 5. Comparison of model and experiment for the worst case of phase-
hop for the first winding of transformer Ta. One can see a perfect agree- Transformer Ta
Tb Tc Td
ment between simulation and experiment. Winding 1st 2nd 3rd 4th
Air-core inductance [µH] 645 850 1,069 1,300 1,000 463 316
Primary voltage (V) & primary current (A)

500 Primary current from model

Primary current from test
Primary current from model Primary current from experiment TABLE III
400 Primary voltage from model FIRST PEAK VALUES OF INRUSH CURRENT FOR DIFFERENT TRANSFORMERS
Primary voltage from experiment
300 UNDER WORST CASE OF PHASE-HOP (EXPERIMENT VERSUS SIMULATION)
Transformer Winding Test [A] Model [A] Difference [%]
200
Primary voltage 1st 157.7 162.5 3.0
100 2nd 130.4 134.5 3.1
Ta
3rd 125.2 122.3 -2.4
0
4th 123.1 111.1 -9.7
-100 Tb 149.9 153.1 2.1
Tc 208 213.2 2.5
10 20 30 40 50 60 70 Td 295.4 307 3.9
Time (ms)
Fig. 6. Comparison of model and experiment for the worst case of phase- TABLE IV
hop for transformer Td. One can see a very good match between simulation SECOND PEAK VALUES OF INRUSH CURRENT FOR DIFFERENT TRANSFORMERS
and experiment. UNDER WORST CASE OF PHASE-HOP (EXPERIMENT VERSUS SIMULATION)
Transformer Winding Test [A] Model [A] Difference [%]
VI. TRANSFORMERS UNDER PHASE-HOP CONDITION 1st 328.9 330.2 0.4
2nd 291.3 272.4 -6.5
In practice, it may not be common to have the phase-hop Ta 3rd 261.7 252.5 -3.5
condition right after the transformer energization (inrush). 4th 240.5 238.4 -0.9
That circumstance was used in Section IV to analyze this ex- Tb 353.3 368.5 4.3
treme, yet possible, case of phase-hop and validate the model. Tc 402.3 360.9 -10.3
The most practical and probable condition of the phase hop is Td 463.3 447.9 -3.3
when it happens during the normal operation in steady state TABLE V
COMPARISON OF MAXIMUM INRUSH CURRENT UNDER DIFFERENT CONDITIONS
(long after energizing the transformer for the first time). In FOR DIFFERENT TRANSFORMERS
order to simulate this condition in the EMTP and compare the Transformer First Peak Second Peak Second Peak
results, the transformer is energized at voltage zero-crossing Winding (inrush) [A] (phase-hop) [A] (worst case) [A]
and after reaching the steady state, phase-hop occurs. 1st 162.5 313.3 330.2
The results for the first winding of transformer Ta under 2nd 134.5 257.6 272.4
Ta
3rd 122.3 237.4 252.5
phase-hop are shown in Fig. 7a. The first peak of inrush cur- 4th 111.1 222.0 238.4
rent is 162.5 A while the one caused by phase-hop is 313.3 A. Tb 153.1 329.8 368.5
Fig. 7b shows a closer view of the phase-hop condition for Tc 213.2 359.3 360.9
this case. Figs. 8a and 8b presents the results for transformer Td 307 446.8 447.9
Td under typical phase-hop condition. The first peak of inrush
current is 307 A, and the second peak is 1.46 times higher at Note that for transformers Tc and Td, which are toroidal
446.8 A. Table V compares the results for all four transform- transformers, the difference between peak values of inrush
ers for the inrush currents caused by zero-crossing voltage, current in the normal and the worst cases of phase-hop is
typical condition of phase-hop, and worst case of phase-hop small (see Table V). This is so because in these transformers
through simulation. One can appreciate that the values of in- the hysteresis cycles are thinner and flatter than the ones of
rush current from the phase-hop condition are much higher standard transformers, because the cores have no gap. There-
than the first peak of inrush current caused by zero-crossing fore, under the worst case of phase-hop, the first spike of in-
voltage. As an example, for transformer Tb, the peak value of rush current reached zero at the start of the second peak, while
inrush current under normal phase-hop condition is 2.15 times for standard transformers (Ta and Tb) the second inrush cur-
 5

rent happens while the current is not yet zero; see Figs. 5 and sags with larger voltage magnitude, cause smaller inrush cur-
6 to compare the results for transformers Ta and Td. rents.

Primary current Primary voltage

300
Primary voltage(V) &

Primary current 500

primary current (A)

Primary current Primary voltage

Primary voltage
200

Primary voltage (V) &

400
Primary current

primary current (A)

100 300
Primary voltage
200
0
100
-100
0

100 200 300 400 500 600 -100

Time (ms)
(a) 0 100 200 300 400 500 600
Time (ms)
300 Primary current Primary current Primary voltage
(a)
500
Primary voltage (V) &

Primary voltage Primary current Primary voltage

200
primary current (A)

Primary current

Primary voltage (V) &

400

primary current (A)

100 300
Primary voltage

0
200
100
-100
0
-200 -100
510 520 530 540 550 560 570 580 590 600
Time (ms)
510 520 530 540 550 560 570 580 590 600
(b) Time (ms)
Fig. 7. Simulation of transformer Ta (first winding) under phase-hop condi- (b)
tion; (a) transient from the beginning of excitation (b) close view of the phase- Fig. 8. Simulation of transformer Td under phase-hop condition; (a) transient
hop part. from the beginning of excitation (b) close view of the phase-hop part.

Fig. 9 shows the effect of the duration of an interruption on

Primary voltage (V) & current (A)

300 I1 (0.5 Cycle)

phase-hop current. The primary currents and the applied volt- I1 (0.75 Cycle)
age (primary voltage) to the first winding of the transformer 200 V1 (0.75 Cycle)
Ta are presented for a zero volts interruption lasting 0.5, 0.75 V1 (0.5 Cycle)

and 1 cycle. One can see that the largest peak current is when 100
the duration of the interruption is half a cycle (313.3 A),
0
which is almost twice the normal zero-crossing inrush current
I1 (1 Cycle)
(162.5 A). Under this situation, a complete instance of phase-
-100
hop occurs. The case with no inrush current is when the dura-
V1 (1 Cycle)
tion of the interruption is one full cycle. This situation corre-
0.51 0.52 0.53 0.54 0.55 0.56
sponds to the normal sinusoidal condition since one complete Time (s)
cycle is eliminated. For an interruption of 0.75 cycle the peak Fig. 9. Primary voltage and caused inrush current of the first winding of the
current is 178.1 A. transformer Ta under 0% interruption.
Primary voltage (V) & current (A)

An example of a voltage sag is presented in Fig. 10. The 300

Sag Inrush
primary voltage and inrush currents under zero-crossing and Zero - crossing Inrush Primary voltage
200
10% voltage sag lasting for 10.5 cycle are shown, for the first
winding of transformer Ta. The value of inrush current caused 100
by the sag is 274.7 A (69% larger than the zero-crossing in-
rush current). 0
Table VI summarizes the inrush current results for 0% and 5%
interruptions, and for 10% and 50% voltage sags. The dura- -100

tion of the transient is between half a cycle (worst case) and -200
3600.5 cycles (around 1 minute). As shown in table VI, the 0 200 400 600 800 1000
worst cases of inrush-like currents occur when the fault dura- Time (ms)
Fig. 10. Primary voltage and caused inrush current of the first winding of the
tion is 0.5+n cycles; where n=0, 1, 2… This is so because transformer Ta under 10% voltage sag for 10.5 cycle duration.
there are two half cycles consecutively, which is the complete
phase-hop. In contrast, for sags lasting 0.5+n cycles, there is a In addition, a longer interruption or voltage sag causes a
small flux-cancellation effect which decreases magnitude of larger reduction in the built flux and as a result in the inrush
the inrush current. To illustrate this, the 20% sag with 2.5 cy- currents (see Table VI).
cle duration (n=2) is depicted in Fig. 11. Note that the integral To complete the study, EMTP simulations for various un-
of the voltage is the flux linkage. The areas A, B, C and D dervoltages were performed [2], [3]. Undervoltages lasting
cancel each other but the extra half cycle, E (highlighted in longer than 1 minute with magnitudes between 0.8 to 0.9 pu
Fig. 11) leads to a decrease in the built flux. This is the reason
 6

were analyzed. In no case, including undervoltages lasting rent, 0.6 A peak. At point c, the internal voltage is crossing
0.5+n cycles, is the phase-hop phenomenon observed. zero from positive to negative, at that moment the flux linkage
Form this study, it is concluded that under the phase-hop presents a first peak (0.81 Wb) and the “normal” peak of the
condition a very large current can be drawn by transformers inrush currents is reached (162.5 A). Then, the phase-hop oc-
due to heavy saturation of the iron core. Therefore, phase-hop curs and the negative semi-cycle of the voltage, between
should be considered in transformer and UPS design and op- points c and d disappears (see Figs. 13a and 13b). When the
eration to prevent its potential destructive effects. As it was terminal voltage reaches the next zero crossing at point d, the
shown, phase-hop can occur partially or fully depending on flux linkage has reduced a small amount, but it is still at a very
the magnitude and duration of electromagnetic phenomena high value (0.65 Wb) and the current has not reduced to zero
causing distorted input voltage to the transformer. (11.8 A). Because of the existence of a positive voltage be-
tween points d and e, the flux increases further until the tran-
TABLE VI
INRUSH CURRENTS UNDER DIFFERENT KINDS OF INTERRUPTIONS AND
sient reaches the maximum at point e with a flux linkage of
VOLTAGE SAGS FOR THE FIRST WINDING OF TRANSFORMER TA 0.95 Wb and a phase hop current of 330.2 A, which is almost
Current Peak Value (A) twice as large as the zero-crossing inrush current. At this time,
interruption interruption sag sag the internal voltage is crossing zero from positive to negative.
# of cycles 0% 5% 10% 50% From this point on, the peaks of flux and current reduce in
0.5 313.3 305.3 296.1 182
magnitude as the dc component damps. At point f voltage
0.6 296.4 287.4 277.4 162.9
0.7 232.3 222.7 212.3 105.3 reaches its first negative peak after phase-hop, with the value
0.8 109.1 101.7 93.6 20.8 of 0.3 Wb for the flux linkage and 0.31 A for the primary cur-
0.9 5.11 5 4.85 3.6 rent. The reversing points of the hysteresis cycle in the third
1 2.2 2.3 2.3 2.4 quadrant (points g, h and i) progressively decrease as the tran-
1.1 4.9 4.79 4.6 3.5 sient damps out and the flux becomes increasingly symmetric.
1.5 310.7 302.7 293.5 179.5
2.5 308.3 300.2 291 177.1
The magnitudes of the flux linkage are -0.2 Wb, -0.26 Wb,
10.5 293 283.93 274.7 160.5 and -0.28 Wb for these three points, respectively, which cor-
100.5 228.4 213.82 199.7 73.81 respond to the voltage zero crossings from negative to posi-
3600.5 201.5 184.3 166.9 37 tive.
VIII. MAGNETIC FIELD BEHAVIOR
To shed light into the internal behavior of the transformer,
in this section, the magnetic field of the transformer is investi-
gated for different operating conditions including: open cir-
cuit, normal operation (on-load), zero-crossing inrush, and
phase-hop. Simulations are performed using the FEM (Finite
Element Method) computer program, Maxwell 14. Magnetic
Fig. 11. 20% sag with 2.5 cycle duration flux lines are shown inside and outside of the core in Fig. 14.
Note that due to the geometrical symmetry of the transformer,
VII. PHYSICAL EXPLANATION OF THE PHASE-HOP CONDITION only a part of the core is shown.
Figs. 12 and 13 explain the phase-hop phenomenon physi- During open circuit, the situation presented in Fig. 14(a),
cally by illustrating the behavior of the primary voltage, inter- the magnetic field is concentrated inside the iron core (the
nal voltage, flux linkage, and current. The graphs correspond lines in the window are the boundaries of the windings). Dur-
to the first winding of the transformer Ta under the worst con- ing normal operation, when the transformer is supplying the
dition of phase-hop (phase-hop following transformer ener- nominal load, a part of magnetic flux “leaks” into the inter-
gization at zero crossing). An important component of the winding region (see Fig. 14(b)). This flux is what produces
explanation is the internal voltage (E1), which is computed as the leakage inductance. In Fig. 14(c) the magnetic flux for
follows: transformer energization at zero crossing is presented. One
can see that there is a considerable amount of flux in the air.
E1  V1  R1 I1 (4) In fact, the flux distribution resembles the behavior of an air-
core inductor. As shown in Fig. 14(d), the flux pattern during
where V1 is the primary terminal voltage, I1 is the primary phase-hop does not change significantly in comparison with
current, and R1 is the primary winding ac resistance. that of the normal inrush current. However, the amplitude
Nine points: a, b, c, d, e, f, g, h, and i are identified in Figs. (seen by the concentration of lines) of the magnetic field is
12 and 13 to highlight important performance stages of the larger.
transformer at different times during the inrush followed by a
phase-hop transient. IX. POSSIBLE SOLUTIONS FOR PHASE-HOP
The energization is done with zero residual flux (point a). The transient phenomenon known as transformer inrush
At that instant, voltage, current, and flux are all zero. When currents was first published by John Fleming in 1892 [13].
the voltage reaches its first peak (at point b) a quarter of a Since then, many publications have proposed techniques to
cycle later, flux linkage is building (0.4 Wb) and the current is limit inrush currents to prevent its destructive effects. Some of
still small at about the value of the normal magnetizing cur-
 7

Fig. 12. Core flux linkage vs. primary current for first winding of transformer Ta under the worst condition of phase-hop.

200 b 200 b

Primary internal voltage (V)

e
Primary voltage (V)

100 100
c
a d e h i
g h i
0 0 c
a d g

-100 -100

f f
-200 -200
0 10 20 30 40 50 60 70 80 0 10 20 30 40 50 60 70 80
Time (ms) Time (ms)
(a) (b)

1 e
c e
300
Flux linkage (Wbt)

Primary current (A)

0.5 d
b 200
f c
a
0 100
g h a b d f g h i
i
0
0 10 20 30 40 50 60 70 80 0 10 20 30 40 50 60 70 80
Time (ms) Time (ms)
(c) (d)

Fig. 13. (a) Primary voltage vs. time; (b) primary internal voltage vs. time; (c) core flux linkage vs. time; (d) primary current vs. time for the first winding
of the transformer Ta under worst condition of phase-hop.

(a) (b) (c) (d)

Fig. 14. Magnetic field behavior for saturated and non-saturated transformer iron core: (a) open circuit; (b) normal operating condition – transformer loaded;
(c) peak condition for inrush currents at zero-crossing switching; (d) peak condition for phase-hop currents.
 8

the methods are external (to the transformer) and others are opto-isolator will pass the sinusoidal waveform to a compara-
transformer-based solutions. External solutions consist of pre- tor, which checks for a zero value. As a result, a 50% duty
insertion impedances, negative temperature coefficient ther- ratio square wave, which rising and falling edges correspond
mistors (NTC) [14], transformer core demagnetizing [15], to the zero-crossing of power source, appears at the output of
phase-delayed switching [16], [17], and sequential phase en- comparator. The first rising edge triggers the digital logic con-
ergization [18], [19]. Transformer-based solutions consist of trol circuit, which turns-on the switch and finally energizes the
air gaps, virtual gaps [20], using low permeability materials transformer.
for the core, and special designs with larger values of air-core The phase-hop circuit of the switch is essentially the same
inductance. as the zero-crossing circuit except for the digital logic control
To some extent, each of the existing approaches diminishes circuit and an extra pair of MOSFETs. Three precise timers
inrush currents; however, there is a trade-off with each one of are utilized in the control circuit to generate the signals for
them. In addition, some methods are not applicable for phase switches 1 and 2 in Fig. 16. Switch 1 consists of two
hop. External demagnetizing techniques, for example, are not MOSFETs. It closes at the first zero-crossing and only opens
possible because there is not enough time to demagnetize the between the second and third zero-crossings. Switch 2 is add-
transformer core during the half a cycle between two consecu- ed to prevent cutting large inductive currents. When switch 2
tive peaks. Switching methods have some problems with the is closed, the inrush current inside the transformer will only
mutual effects with switches applied in the UPS system and
flow through switch 2. As a result, switch 1 and other circuit
also the reliability of the system. Implementations of pre-
elements are protected from the high voltages caused by large
insertion impedance methods are very complicated due to the
di/dt values.
difficulty in the detection of the phase-hop condition. Ther-
mistors do not work either because at the time of the phase-
hop the system is already on, therefore, thermistor resistances
are very small and cannot reduce the inrush current effective-
ly. In general, there are several problems with the addition of
series components with the transformer: (1) the reliability of
the system reduces, and (2) depending on the voltage level the
additional components need to comply with safety standards,
which makes them expensive.
It seems that the best solutions to prevent the destructive
effects of the phase-hop phenomena are transformer-based.
Application of these methods will be treated in a forthcoming
paper.

X. CONCLUSION
This paper has shown, for the first time, how the occurrence Fig. 15. Power and control circuits implemented in the zero-crossing and
of the phase-hop phenomenon in transformers can lead to ex- phase-hop switch.
tremely large currents. Phase-hop can occur at any time in a
power system because interruptions, voltage sags, and notch-
ing in the network are not predictable. In order to prevent
these phenomena, a UPS system can be used. However, the
action of off-line UPS systems may itself lead to large levels
of inrush currents for the transformers located between the
load and the UPS system as well.
The value of the phase-hop currents can be several times
higher than the magnitude of the “normal” inrush currents that
occur when a transformer is energized at voltage zero-
crossing. Fig. 16. Schematics of the phase-hop circuit.
The extremely large currents produced by the phase-hop
condition can lead to serious problems such as power quality XII. ACKNOWLEDGMENT
issues, mechanical stresses on transformer windings, and false The authors would like to thank Mr. Nazir Mahamedau, ex-
tripping of vital protections. The best techniques to prevent student of Polytechnic Institute of New York University, for
these serious effects seem to be transformer-based solutions. his collaboration in building the phase-hop switch and per-
forming the experiments during this project.
XI. APPENDIX
Fig. 15 presents a block diagram of the power and control XIII. REFERENCES
circuits implemented in the zero-crossing and phase-hop [1] B. W. Kennedy, "Power quality primer", McGraw-Hill Companies,
2000, pp. 122-125.
switch developed for this project. This switch consists of volt-
[2] IEEE Recommended Practice for Monitoring Electric Power Quality,
age regulators, opto-isolators, a digital logic control circuit IEEE Standard 1159-1995, Jun. 1995.
and MOSFET switches. When the ac power source is on, the
 9

[3] IEEE Recommended Practice for Powering and Grounding Electronic Ashkan Farazmand was born in Tehran, Iran, in
Equipment, IEEE Standard 1100-1999, Mar. 1999. 1983. He received the M.Sc. (Hons.) degree in
[4] Private communications between engineers of American Power Conver- electrical engineering from the University of Teh-
sion and Liebert with the second author, Toronto, Ontario, 1999. ran, Tehran, Iran, in 2009, and the Ph.D. degree
[5] IEC Standard 61000-4-11, Electromagnetic compatibility (EMC): Test- (Hons.) in electrical engineering from the Poly-
ing and measurement techniques- Voltage dips, short interruptions and technic Institute of New York University, Brook-
voltage variations immunity tests, Mar. 2004. lyn, NY, in 2013. Currently he is a Post-doctoral
[6] IEEE Recommended Practice and Requirements for Harmonic Control Fellow in the Electrical and Computer Engineering
in Electrical Power Systems, IEEE Standard 519-1992, Apr. 1993. Department of Polytechnic Institute of New York
[7] J. M. Guerrero, L. G. de Vicuna, and J. Uceda, "Uninterruptible power University. His research interests are design and
supply systems provide protection," Industrial Electronics Magazine, analysis of transformers, electrical transients,
IEEE , vol.1, no.1, pp.28-38, Spring 2007. derating of electrical machines under nonsinusoi-
[8] J. Seymour,” The seven types of power problems”, white paper 18, Re- dal and unbalanced conditions, and power quality.
vision1, Schneider Electric – Data Center, Science Center, 2011.
[9] F. de León, A. Farazmand, and P. Joseph, “Comparing the T
and pi equivalent circuits for the calculation of transformer inrush cur- Francisco de León (S’86–M’92–SM’02) re-
rents”, IEEE Trans. Power Delivery, vol. 27, no. 4, pp. 2390-2398, Oct. ceived the B.Sc. and the M.Sc. (Hons.) degrees in
2012. electrical engineering from the National Polytech-
[10] F. de León and A. Semlyen, “Efficient calculation of elementary parame- nic Institute, Mexico City, Mexico, in 1983 and
ters of transformers”, IEEE Trans. Power Delivery, vol. 7, no. 1 pp. 376- 1986, respectively, and the Ph.D. degree from the
383, Jan. 1992. University of Toronto, Toronto, ON, Canada, in
[11] F. de León, S. Jazebi, and A. Farazmand, “Accurate measurement of 1992. He has held several academic positions in
the air-core inductance of iron-core transformers with a non-ideal low Mexico and has worked for the Canadian electric
power rectifier,” accepted for publication in the IEEE Trans. Power Del. industry. Currently, he is an Associate Professor at
[12] J. Mahseredjian, S. Dennetière, L. Dubé, B. Khodabakhchian, and L. Polytechnic Institute of NYU, Brooklyn, New
Gérin-Lajoie, “On a new approach for the simulation of transients in York. His research interests include the analysis of
power systems,” Elect. Power Syst. Res., vol. 77, no. 11, pp. 1514–1520, power phenomena under nonsinusoidal conditions, the transient and steady
Sep. 2007. state analyses of power systems, the thermal rating of cables and transformers,
[13] J. A. Fleming, "Experimental researches on alternate-current transform- and the calculation of electromagnetic fields applied to machine design and
ers”, Journal of the Institution of Electrical Engineers, vol. 21, no. 101, modeling.
pp. 594-686, Nov. 1892.
[14] K. Billings and T. Morey, “Switchmode power supply handbook”, 2nd
ed., McGraw Hill, NY, pp. 1-6, 1989. Kuang Zhang (S’12) was born in Ziyang, China,
[15] B. Kovan, F. de León, D. Czarkowski, Z. Zabar, and L. Birenbaum, in 1990. He received the B.Eng. degree in Elec-
“Mitigation of inrush currents in network transformers by reducing the tronic Engineering from South China University
residual flux with an ultra-low-frequency power source”, IEEE Trans. of Technology, Guangzhou, China, in 2013, and
Power Delivery, vol. 26, no. 3, pp. 1563-1570, Jul. 2011. the B.Sc. degree in Electrical Engineering from
[16] J. H. Brunke and K. J. Frohlich, “Elimination of transformer inrush the Polytechnic Institute of New York University,
currents by controlled switching. I. Theoretical considerations”, IEEE Brooklyn, NY, in 2013. He is pursuing Ph.D.
Trans. Power Delivery, vol. 16, no. 2, pp. 276-280, Apr. 2001. degree in Electrical Engineering at the Polytech-
[17] J. H. Brunke and K. J. Frohlich, “Elimination of transformer inrush nic Institute of New York University, Brooklyn,
currents by controlled switching. II. Application and performance con- NY. His research interests include modeling,
siderations”, IEEE Trans. Power Delivery, vol. 16, no. 2, pp. 281-285, computation and measurement of transformer inrush currents. He is currently
Apr. 2001. working in the areas of power electronics and power quality.
[18] Y. Cui, S. G. Abdulsalam, S. Chen, and W. Xu, “A sequential phase
energization technique for transformer inrush current reduction- Part I:
simulation and experimental results”, IEEE Trans. Power Delivery, vol.
20, no. 2.pp. 943-949, Apr. 2005. Saeed Jazebi (S’10) was born in 1983, Kerman,
[19] W. Xu, S. G. Abdulsalam, Y. Cui, and X. Liu, “A sequential phase ener- Iran. He received his B.Sc. and M.Sc. degrees in
gization technique for transformer inrush current reduction - Part II: the- 2006 and 2008 in Electrical Engineering from
oretical analysis and design guide," IEEE Trans. Power Delivery, vol. Shahid Bahonar University, Kerman, Iran, and
20, no. 2, pp. 950- 957, Apr. 2005. Amirkabir University of Technology, Tehran, Iran,
[20] V. Molcrette, J. L. Kotny, J. P. Swan, and J. F. Brudny, “Reduction of respectively. He is currently a Ph.D. student at the
inrush current in single-phase transformer using virtual air gap tech- Polytechnic Institute of New York University,
nique”, IEEE Trans. Magnetics, vol. 34, no. 4, pp. 1192-1194, Jul. 1998. Brooklyn, NY. His field of interest includes: elec-
tromagnetic design, modeling and simulation of
electrical machines and power system components,
statistical pattern recognition applications in power
engineering, power system protection, and power quality.
 1

Experimentally Validated Reversible Single-Phase Multi-

Winding Transformer Model for the Accurate Calculation
of Low-Frequency Transients
Saeed Jazebi, Member, IEEE, and Francisco de León, Senior Member, IEEE

 Recently, an analytical solution for single-phase two-

Abstract—In this paper, a previously published model for the winding transformers was proposed based on equivalent reluc-
representation of the leakage inductance of multi-winding tance circuits [13]. The model accurately considers the deep
transformers is enhanced to support accurate calculations of saturation behavior of the two windings simultaneously, and
low-frequency transients, including: inrush currents, series therefore, the word reversible was coined for this model. How-
ferroresonance, and geomagnetic induced currents (GIC). The ever, the model of [13] cannot be built with circuit elements
new circuit is obtained from the principle of duality and there- available in EMTP-type programs. Thereafter, a dual magnetic-
fore is physically consistent. The unique characteristic of the electric model was proposed to overcome this drawback in [14].
improved model is that the very deep saturation behavior of the
The model of [14] is derived from terminal measurements and is
iron core is properly represented for each winding simultane-
easy to implement in any EMTP-type program since it uses
ously (reversible model) without changing parameters. The hys-
only standard circuit element.
teresis cycle and iron core losses are also included. In addition
to its reversible terminal behavior coupled with physical con- It is known that the terminal behavior of the duality derived
sistency, the proposed model can be built with circuit elements transformer models does not always match the terminal meas-
available in EMTP-type programs and all the parameters can be urements performed in the laboratory; see [8] and [14]. The
computed from terminal tests. The model is validated by com- main contribution of this paper is to enhance the terminal be-
paring computer simulations versus laboratory measurements havior of the multi-winding transformer model presented in [8]
for three- and four-winding transformers. and [9] to accurately represent the low-frequency behavior of
different windings involving very deep saturation.
Index Terms— Duality, electromagnetic transients, ferro- The equivalent magnetic-electrical model of this paper is de-
resonance, GIC, inrush currents, multi-winding transformers. rived from the principle of duality. Step-by-step guidelines to
compute the parameters of the model from measurements are
I. INT RODUCT ION presented. The model includes: leakage inductances, mutual

M ULTI-WINDING transformer models have broad applica-

tions in design and development of power system and
couplings, hysteresis loops, and iron core losses. The very
deep saturation regions of the magnetizing branches are calcu-
power electronic devices. Several multi-winding transformer lated from the solution of the equivalent circuit to match the
models exist in the literature. Among them, the models present- terminal measurements. The model is compatible with all circuit
ed in [1]-[10] need to be acknowledged. The saturation induct- simulators since only standard circuit elements are needed.
ance (frequently called “air-core” inductance) and winding The excellent agreement between simulations and laboratory
resistance, the dominant parameters for transients when the measurements demonstrate that the model is accurate and per-
core saturates, are different for each winding since the geome- fectly reversible.
try (at the very least the radius) is different; see [11] and [12]. This paper deals with single-phase multi-winding trans-
Therefore, inrush currents, geomagnetic induced currents formers. The same methodology will be applied to the multi-
(GIC), and ferroresonance occur at different levels of current phase multi-winding model presented in [10] in a forthcoming
and voltage for each winding. Laboratory measurements on a paper.
four-winding transformer are shown in Fig. 1. This figure
demonstrates the significant difference in the transient re- 130 st
1 winding
sponse of the windings during inrush currents. This attribute nd
2 winding rd
100 3 winding
is neither reported nor considered in publications dedicated to 4th winding
Current (A)

multi-winding transformer models.

50
S. Jazebi and F. de León, are with the Department of Electrical and
Computer Engineering at the NYU Polytechnic School of Engineering,
Six Metrotech Center, Brooklyn, NY, 11201 (e-mails: jazebi@ieee.org, 0
13 15 20 25 28 44.9 51.5 58.5 65.2
fdeleon@nyu.edu). Time (ms) Winding mean radius (mm)
 2

Fig. 1. Inrush current amplitudes for concentric windings with different distribution of the magnetic energy completely changes, be-
mean radius. All windings have the same number of turns and conductor cause a saturated iron core becomes linear with incremental
cross sectional area. T he saturation inductance and the winding re-
sistance increase for windings with larger radii, which reduce the inrush
permeability similar to air. Therefore, the magnetic flux is no
currents. longer concentrated solely in the iron core, but is distributed in
the transformer window and air (see Fig. 3). Under these condi-
II. M ODELING PRINCIPLES tions, the flux between the windings and the core and the flux
The structure of the model is derived from the direct appli- outside of the transformer window become significant since
cation of the principle of duality to multi-winding transformers. they are comparable to the flux in the core and leakage flux
Fig. 2 illustrates half the window of a shell-type n-winding sin- between the windings.
gle-phase transformer. The equivalent electrical circuit is de-
picted on top of the transformer frame. The leakage inductanc-
es are modeled with linear inductors and mutual couplings . The
leakage inductances between the windings are represented by
mmf
Lij. The mutual inductances (Mij) compensate for the missing
mmf
linking flux between the windings [8], [9]. The iron core is mod-
eled using hysteretic inductors. Also, non-linear inductors are
considered for the contribution of the magnetic energy in the
air [14]. The methodology replicates the physical behavior of
the magnetic flux for different operating conditions. Therefore, Fig. 3. Magnetic field strength and the magnetic circuit for open circuit-
ed transformer (a) normal operating region; (b) deep saturation region
it is in full agreement with the principle of duality and modifies
(µr→1).
the equivalent circuit for high saturation conditions. This is so
because in the operating regions below the knee point (during
the short circuit, normal open circuit, and nominally loaded),
the value of the air inductances are negligible when compared
to the iron-core inductances. However, in deep saturation the

LA3
LA5 LA7
x=0
n1 Lm2 n2 Lm3 n3 Lm4 nn-1

LA2 LA4 LA6

M1n-1
M13 M2n-1
Lm1 LA1 M12 M23 M3n-1 LA2n-2 Lmn LA2n-1

L12 L23 L34 Ln-1n

w1 V1 w1 w2 w3 w4 wn
V2
V3
V4 n'1 n'2 n'3 n'n-1

Vn
Fig. 2. Direct application of the principle of duality on a multi-winding transformer. Note that, due to the symmetry of the equivalent electrical
circuit with respect to x=0 axis, only the right half side of the transformer window is illustrated.
M1n-1
M13 M2n-1
M12 M23 M3n-1
L12 L23 L34 Ln-1n
N1:N N2:N N3:N N4:N Nn:N
R1 R2 R3 R4 Rn
Rm1 Lm1 Rm2 Lm2 Rm3 Lm3 Rm4 Lm4 Rmn Lmn

Lair-1 Lair-2 Lair-3 Lair-4 Lair-n

Fig. 5. Reversible multi-winding transformer model including resistances and ideal transformers; Lair-1 , Lair-2 , … Lair-n are the saturation inductances
of the windings for the modification of the t erminal response of the model for the high saturation region.
 3

flux is concentrated in the iron core. Therefore, similar hystere-

/I=Ls1
Lm1 sis curve and magnetizing parameters (Rm and Lm) are obtained
0.5 from measurements on the different windings. This has been
 [Wb]

demonstrated experimentally by measuring almost the same

LA1
0 magnetizing current in all windings when excited with rated
voltage, within measuring accuracy.
Saturation In the normal operating region, the measured λm-im charac-
-0.5
0 2 10 25 50 75 100 teristic could be distributed between the n magnetizing
Current [A]
branches considering the leg/yoke geometrical proportions.
Fig. 4. Magnetizing characteristic of the branch Lm1 for a 1 kVA 120 V
shell-type transformer. Because the leakage inductances do not exist, the nonlinear
To accurately represent the above mentioned phenomenon, branches of Fig. 2 are effectively in parallel. Note that, it is
the non-linear air inductances (inductances LA1 to LA2n-1 in Fig. possible to estimate the design details, such as dimensions of
2) are represented with two slopes; zero in normal operating the iron core and windings with the method of [15]. Neverthe-
regions, and a constant slope in high saturation. Fig. 4 illus- less, for simplicity, it is assumed that the transformer window is
trates the magnetizing characteristics of the inductor Lm1 and square. Thus, the length of the legs is the same as the length
the air inductance LA1. In this figure, Ls1 represents the induct- of the yokes. Also, it is assumed that the distances between
ance of the linear part (in deep saturation) of the magnetizing the neighboring windings are the same. Hence, the leakage
curve. The series inductances LA1 and Lm1; LA2, LA3, and Lm2 …; fluxes leave the yoke at points located at 1/n, 2/n, …, (n-1)/n of
and LA2n-2, LA2n-1, and Lmn could be merged into single inductors. the length of the yoke (see nodes n 1, n 2, ..n n-1 in Fig. 2). There-
For example, according to the characteristics of LA1 shown in fore, the limbs are physically divided into n-2 regions which
Fig. 4, only the high saturation slope of Lm1 changes to L1= result in Lm2= Lm3=…=Lmn-1. Besides, Lm1=Lmn, because normally
LA1+ Ls1, where L1 is the modified slope of Lm1 in the high satu- the width of the center leg is twice the width of the side legs
ration region. Similarly, Ln represents the deep saturation slope and the length of the flux paths are the same (distance between
of the magnetizing inductor Lmn in the model. Then, the wind- nodes n 1, n ’1, and n n-1, n ’n-1 in Fig. 2). Finally, for the n-winding
ing resistances, core losses, and ideal transformers are added transformer, the method of [9] is extended as follows:
to the circuit of Fig. 2; see in Fig. 5 the final model. 4nLm
Lm1  Lmn  , Lm2  Lm3   Lmn1  2nLm (1)
The principal advantage of this model, which differentiates n2
it from the model of [8], is the computation of the deep satura- where Lm is the magnetizing inductance measured from any
tion inductances of the magnetizing inductors (Lmn). Note that winding. Hence, the λ-i characteristics of each branch are ob-
according to the equivalent circuits of Figs. 2 and 5, a hysteret- tained with the following expressions:
ic magnetizing inductor is connected in parallel with the termi- m1 ( k )  m2 ( k )   mn ( k )  m ( k )
nals of each winding. These n inductors are frequently called ( n  2) im ( k )
magnetizing or non-linear branches in this paper. The existence im1 ( k )  imn ( k )  (2)
4n
of the n branches provides adequate degrees of freedom to im ( k )
correctly characterize the different equivalent inductance val- im2 ( k )  im3 ( k )   imn 1 ( k ) 
2n
ues from the terminals in the saturation region.
where λmi-i mi is the magnetizing curve of the ith nonlinear
As noted above, the electromagnetic behavior of the core
branch and k is the k th point of the data. Trapezoidal rule of
and air is different under normal operating condition than in
integration is applied to compute λm from the measured terminal
deep saturation. Therefore, magnetizing branches are modeled
voltage obtained from the open circuit tests. Note that current
in two steps: First the non-linear behavior of the iron core in
im is measured in the primary (low voltage) winding and the
the non-saturated region, below the knee point, is considered
induced voltage is captured at the open circuited secondary
including hysteresis; second, the linear behavior of the iron
(high voltage) winding.
core in deep saturation is added. The guidelines for the calcu-
lation of the unknown parameters of the model are described in B. Magnetizing Branches: Deep Saturation Region
the following subsections. The magnetic circuit of an open circuited transformer con-
A. Magnetizing Branches: Hysteresis Curve sists of several parallel branches connected to a mmf (see Fig.
3). The circuit can be simplified (by series/parallel combina-
In an open circuited transformer excited with rated voltage,
tions) resulting in a single nonlinear reluctance with hysteretic
the iron core operates below the knee point, where the value of
characteristic for normal conditions and linear behavior in deep
the magnetizing inductance is substantially larger than the
saturation. The dual electrical representation of this model is a
leakage inductances. Hence, the leakage inductances together
single hysteretic branch. Therefore, the simplest dual represen-
with its mutual couplings are negligible in comparison with the
tation of a transformer model in the open circuit condition is a
magnetizing inductances. Note that, during the standardized
single hysteretic inductor in series with the terminal resistance
open circuit test with nominal voltage excitation, the magnetic
of the corresponding winding. This could be seen in Fig. 6 (a)
 4

and (b) for the first and the second windings of a two-winding Fig. 7. Equivalent circuits for the calculation of the saturation induct-
transformer. ances for a three-winding transformer; simplified circuit seen from (a)
the 1 st winding, (b) the 2 nd winding, (c) the 3 rd winding.
Note that, the leakage inductances do not have a physical
existence in open circuit conditions (because there is no leak- M13
age flux when only one winding is energized). However, to M12 M23 L34
L12 L23
simulate the normal operating conditions , a model requires of
the representation of the leakage flux between the two wind- L1
Vh1 i1 i5 L2 i6 L3 i7 L4
ings (when at least two windings are energized). To have a
unique model in both open and short circuit conditions, the
(a)
leakage components need to be added to the circuits according
M13
to Fig. 6 (c). However, the addition of the leakage inductance M12
L12 L23 M23 L34
affects the behavior of the open circuit condition of the model
of Fig. 6 especially in the deep saturation region. i8
Rs2 L1 Vh2 i2 L2 i9 L3 i10 L4
Rs1 Ls
N1:N N:N2

(b)
L1 R1 R2 L2 M13
M12 M23
L12 L23 L34
(a) (c) (b)

Fig. 6. Infrastructure of a duality-derived π model for a 2- winding trans- L1 i11 L2 i12 Vh3 i3 L3 L4
i13
former; (a) the simplest dual representation of the 1 st winding for all
open circuit conditions; (b) the simplest dual representation of the 2 nd
(c)
winding for all open circuit conditions; (c) addition of the leakage in-
M13
ductance to consider the contribution of the leakage flux during the M12
short circuit and normal loaded conditions. T he joint connection of Ls, L12 L23 M23 L34
L1 and L2 affects the open circuit behavior of the transformer esp ecially
in deep saturation. L1 i14 L2 i15 L3 i16 L4 i4 Vh4

M12 (d)
L12 L23
i1 Fig. 8. Equivalent circuits for the calculation of the saturation induct-
i4 i5
ances seen from the four terminals of the four-winding transformer;
Vh1 L1 L2 L3 simplified circuit seen from (a) the 1 st winding, (b) the 2 nd winding, (c)
the 3 rd winding, (d) the 4 th winding.
(a)
M12 M1n-1
L12 L23 M13
M12 M23 L34 Ln-1n
i6 iL712 L23
i2
L1 Vh2 L2 L3
Vh1 i1 L1 in+1 L2 in+2 L3 in+3 L4 Ln-1 i2n-1 Ln
(b)
L12 M12 L23
Mkn-1
i8 i9 i3
M1k-1 Mk-1k
L2 LL312 L Lkk+1 Ln-1n
Vh3 k-1k
L1
ikn
L1 (c)ikn-1 L2 Vhk ik Lk ikn+1 Lk+1 Ln-1 i(k+1)n-k Ln
Lk-1

M1n-1
M13
M12 M2n-1
L12 L23 M23 L34 Ln-1n

L1 in(n-1)+2 L2 in(n-1)+3 L3 in(n-1)+4 Ln

Ln-1 in2 in Vhn

Fig. 9. Simplified equivalent circuits seen from the n terminals of the n-winding transformer for the calculation of the saturation inductances.
 5

cuits are derived for the n-winding transformer as shown in

Fig. 9.
The first step is to obtain the equivalent inductances seen
The leakage inductance is negligible when transformers op- from the terminals of the model (Lair-1, Lair-2,…, Lair-n) with re-
erate below the knee point in the open circuit condition. This is spect to the variables L1, L2, …, Ln. The judicious selection of
so because the magnetizing inductances are much larger than
the meshes, including direction and numbering, as illustrated in
the leakage inductance. For higher excitations, however, the
Figs. 7 to 9, is essential to obtain simplified equations suitable
slope of the magnetizing curves decays to the values of the
for generalization. These mesh equations are written for the
deep saturation inductances L1, L2,…, and Ln, which are of the
same order of magnitude than the leakage inductances. Under fundamental components of the voltage and current, where Vi
these conditions, the effect of the leakage inductance in the is the fundamental component of the hybrid voltage source Vhi,
equivalent circuit cannot be neglected, considering the fact as follows:
that leakage inductances are requisites for normal operation  A B
j  K n2 n2  I n2 1  V n2 1 ,  K    T  (3)
and are a part of the model (see Figs. 2 and 5). This causes a
B D
mismatch between the terminal measurements and the behavior
of the model of [8] in the deep saturation region. To overcome The current and voltage vectors are as follows:
this drawback, general formulas are proposed to precis ely cal- T

culate L1, L2, …, and Ln.  I   i1 , i2 , i3 ,... in2  , V   v1 , v2 ,...vn , 0, 0,...0T (4)
The An×n matrix for the three-, four-, and n-winding trans-
1) Derivation of the System Equations: formers are written as follows:
To retrofit the terminal behavior of the model in deep satura-
tion, L1, L2, …, Ln need to be calculated correctly. Fig. 7 illus-  A33  diag  L1 L2 L3 
trates the equivalent circuits seen from different terminals of a
three winding transformer. Note that, in each case the second-  A44  diag  L1 L2 L3 L4 
ary and tertiary windings are open circuited and the damping  Ann  diag  L1 L2 ... Ln 1 Ln  (5)
components are removed for the analysis since only saturation
inductances are measured. The primary winding is excited with The elements of Bn×n(n-1) for the three-winding transformer
a hybrid ac/dc voltage source to drive the transformer into are zeros except for the following components:
deep saturation as recommended in [12]. The modeling tech- B11   L1 , B36  L3 , B23  L2 , B24   L2 (6)
nique is based on terminal measurements, therefore, the topol-
ogy of the model is consistent for transformers with or without For the four-winding transformer, the following elements of
tank, magnetic tank shunts, belts, etc. However, the deep satu- B are nonzero:
ration inductance measurements are affected in the presence of B11   L1 , B4,12  L4 , B24  L2 , B25   L2
these transformer parts, which change the values of L1, L2, …, (7)
B38  L3 , B39   L3
Ln. The term “air-core inductance” [12] can only be used for an
air coil or a transformer without core and tank. Therefore, in The nonzero elements of the B matrix for the n-winding
this paper, it is substituted with “saturation inductance” as in transformer are as follows:
[16]. As the result of the ac/dc excitation, all three magnetizing
B11   L1 , Bn,( n 1) n  Ln
branches operate in the linear saturated region. Therefore, (8)
these branches are represented with the corresponding co n- Bi ,(i 1) n  Li , Bi ,(i 1) n 1   Li , i  2,3, n 1
stant slope part of the deep saturation region (L1, L2, and L3).
The D matrix is written as follows:
Fig. 8 illustrates the equivalent circuits seen from the terminals
of a four-winding transformer in deep-saturation. Similar cir-

 L1  L2  L12 M 12  L2 M 13 ... M 1n 2 M 1n 1 
 M L L2  L3  L23 M 23  L3 M 24 M 2 n2 M 2 n 1 
 12 2  (12)
 M 13 M 23  L3 L3  L4  L34 M 34  L4 M 3n  2 M 3n 1 
 Da ( n1)( n1) 
M 24 M 34  L4

 
 M 1n  2 Ln  2  Ln 1  Ln  2,n 1 M n 2,n 1  Ln 1 
 
 M 1n 1 M 2 n 1 M 3n 1 ... M n 2,n 1  Ln 1 Ln 1  Ln  Ln 1,n 

f1 ( L1 , L2 , L3 )  L1   L21  L2  L3  L23  /  L1  L2  L12  L2  L3  L23    M12  L2    Lair 1  0

2
(15)
 
2   2
f 2 ( L1 , L2 , L3 )   L2  L1  L12   L3  L23   L2 M12  /  L1  L2  L12   L2  L3  L23    M12  L2    Lair 2  0 (16)

f 3 ( L1 , L2 , L3 )  L3   L23  L1  L2  L12  /  L1  L2  L12  L2  L3  L23    M12  L2    Lair 3  0

2
(17)
 
 6

 Da   C. Leakage Inductances
 
 Da  The leakage inductances are calculated as in reference [8]
 D ( n 1)2 ( n 1)2   

(9) using the standard short circuit tests performed on each pair of
  windings independently [22]. The self inductances are:
  Da  Li ,i 1  Lsi ,i 1 , i  1, 2, n  1 (19)
where, Da for the three-winding transformer is:
where Lsi,i+1 is the measured leakage inductance between wind-
 L1  L2  L12 M12  L2  ings i, and i+1, and Lsi,i =0 [8]. Consequently, the mutual in-
 Da 22   L2  L3  L23 
(10)
 M12  L2 ductances Mij are calculated with the following expression:
Lsi, j 1  Lsi 1, j  Lsi, j  Lsi 1, j 1
For the four-winding transformer Da is: M ij  (20)
2
 L1  L2  L12 M12  L2 M13  Expressions (19) and (20) have been validated experimental-
 Da 33   M12  L2 L2  L3  L23 M 23  L3  (11) ly in [10] for transformers of 96 and 360 MVA. The results are
 M13 M 23  L3 L3  L4  L34  identical to the BCTRAN model proposed in [1].
D. Core Losses
and for the n-winding transformer the Da matrix is given in (12)
Constant resistors Rm1, Rm2, …, Rmn are added to consider
shown at the bottom of the page.
the iron core losses [23]. The method applied in [9] is extended
2) Solution of the System Equations: for the n-winding transformers. The following equations are
The air core inductances (Lair-1, Lair-2,…Lair-n) seen from each obtained with the same assumptions presented in Section II. A
winding are calculated from the following expressions (see (above):
Figs. 7, 8 and 9): n2 Rm n2 Rm
v v v Rm1  Rmn  , Rm 2  Rm3   Rmn 1  (21)
Lair 1  1 , Lair 2  2 , ... Lair n  n (13) 2 n2
j i1 j i2 j in where Rm is the equivalent resistance computed from the
Equation (13) can be substituted into (3). Note that, un- standard open circuit measurements to represent the iron-core
knowns in+1, in+2, …, in2 in (3) do not need to be computed. losses.
Therefore, Kron reduction [17] is used to eliminate those varia-
bles and the system order decreases from n 2 to n: III. M ODEL VALIDAT ION
1
K new  A  B  D  B T Reversible models for three- and four-winding transformers
(14) are developed and validated in this section. The models are
I new  i1 , i2 ,... in  , Vnew  v1 , v2 ,...vn 
T T
implemented for a 1-kVA, 120 V, 4-winding isolation transform-
The result of this step is a system of n nonlinear equations er. The complete data, such as iron core dimensions, leakage
with n unknowns L1, L2, …, Ln. Equations (15) to (17), at the inductances between different windings , saturation inductanc-
bottom of the previous page, are the resultant equations for a es and resistances of different windings, etc. are available in
three-winding transformer. Note that, Lair-1, Lair-2, …, Lair-n are [14]. The reversible model is compared with the conventional
known parameters which are experimentally measured with a model (called nonreversible model) and measurements for vali-
hybrid ac/dc source method as proposed in [12]. Some alterna- dations in different transient conditions .
tive methods to measure high saturation behavior of trans- In the nonreversible model, all parameters are derived ac-
formers could be found in [18]-[21]. cording to the guidelines presented in previous sections ex-
To solve the system of non-linear algebraic equations, the cept the nonlinear branches. In this model, the magnetizing
trust-region-reflective algorithm is applied using the embedded characteristics are extended without the corrections provided
Matlab function ‘lsqnonlin’. This function minimizes the set of in Section II-B. The saturation inductance of the innermost
non-linear equations with least square data-fitting as follows: winding is used to adjust the model parameters which give the
 f1 ( L1 , L2 , L3 Ln ) 2  correct transient behavior of the innermost winding. Note that
  the same winding resistances are used in both models. There-
2   f 2 ( L1 , L2 , L3 Ln ) 2 
min f ( L1 , L2 , L3 Ln ) 2  min   (18) fore, the performance differences are only due to the use of the
  wrong saturation inductances in the traditional model.
  f (L , L , L 2
 n 1 2 3 Ln )  A. Three-Winding Model
The parameters are initialized with the saturation inductance The model for the three-winding transformer is obtained
values: L1(0)=Lair-1, L2(0)=Lair-2, L3(0)=Lair-3, …, Ln(0)=Lair-n. Final- from the first three windings of the four-winding transformer
ly, the characteristics of the n magnetizing branches computed under study. The 4th winding is left disconnected.
by (2) are extended from the last point to infinity using L1,
1) Inrush Currents:
L2,…, Ln as constant slopes.
The transformer is energized through a switch that closes
when the voltage of the sinusoidal source is crossing zero. The
 7

transformer is demagnetized before each experiment. Fig. 10 el; (a) Innermost winding, (b) Inner winding, (c) Outer winding.
illustrates the accuracy of the reversible model in comparison 3) Geomagnetic Induced Currents:
to the nonreversible model. In this case, the nonreversible Geomagnetic induced currents cause a dc potential on the
model overestimates the inrush currents by 21.5%, and 22.9%, surface of the earth. Therefore, during GIC, the transformer
for the second and the third windings, respectively. The com- neutral is biased with a dc voltage [24]. This condition is simu-
parison of the results for inrush currents are presented in Table lated in the laboratory and EMTP with a hybrid dc/ac excita-
I. The differences between the reversible model results and tion. The hybrid voltage source includes a dc generator in se-
measurements are about 5%. ries with an ac source. The dc voltage could be controlled with
2) Ferroresonance: the field excitation of the generator. The hybrid source is con-
For ferroresonance experiments, a 44 µF series capacitance nected to the primary terminal of the transformer while the sec-
is connected between the source and the transformer terminal. ondary terminal is open circuit. The schematic diagram of the
The transformer is completely demagnetized and the capacitor laboratory setup is presented in Fig. 11. The switch is always
is discharged before each measurement. The simulation results closed in this experiment.
are compared to measurements in Table II. One can observe Simulations results for the nonreversible and the reversible
that the simulation results are in good agreement with the la- models are compared versus measurements in Table III. One
boratory measurements. The same tests are performed for the can see a good agreement between the reversible model and
22 µF and 66 µF capacitors, and satisfactory results with errors measurements. The nonreversible model shows relative errors
less than 5% are achieved. EMTP simulations show that the of 8%, and 18.8% with respect to the measurements.
nonreversible model is also correct for the calculation of fer- 4) GIC + Energization
roresonance. The overall differences between the reversible One of the extreme cases of inrush currents could occur
and nonreversible models are less than 2%. when the transformer is energized on zero crossing of the volt-
140
Nonreversible Model age while the neutral of transformer is biased by geomagnetic
120 induced currents. This phenomenon is simulated with both the
Reversible Model
100 nonreversible and the reversible models (see Fig. 11 for the
Current [A]

80 simulated circuit). The results are compared for the three-

60 winding transformer model. The nonreversible model predicts
40 the inrush currents with 16.3%, and 20.5% errors for the inner,
Measurements
20 and outer windings, when compared to the revers ible model.
0 Fig. 12 shows the simulation results for the outer winding.
0 2.5 5 7.5 10 12.5 These results show the significance of the reversible model for
Time [ms]
extreme cases with higher degrees of saturation.
(a)
140 T ABLE I
Nonreversible Model COMP ARISON OF I NRUSH CURRENT P EAK VALUES FOR THE T HREE -
120 Reversible Model W INDING T RANSFORMER
100 Winding Meas. [A] Nonrev. [A] Diff. (%) Rev. [A] Diff. (%)
Current [A]

80 1 st 124.4 127.6 2.6 121.7 2.1

60 2 nd 110.6 134.4 21.5 114.5 3.5
40 3 rd 104.5 128.4 22.9 105.6 1.0
Measurements
20 T ABLE II
0 MAXIMUM T EMP ORARY OVERVOLTAGE MEASURED AND SIMULATED
0 2.5 5 7.5 10 12.5 FOR THE T HREE -W INDINGS T RANSFORMER AND A 44 µF CAP ACITOR [V]
Time [ms]
Winding Meas. [V] Nonrev. [V] Diff. (%) Rev. [V] Diff. (%)
(b) st
1 226.5 213.9 5.6 216.1 4.6
140
Nonreversible Model 2 nd 226.4 212.2 6.3 216.4 4.4
120 3 rd 223.5 214.7 3.9 216.5 3.1
Reversible Model
100
Current [A]

T ABLE III
80 COMP ARISON OF GEOMAGNETIC I NDUCED CURRENTS P EAK VALUES FOR
60 THE T HREE -W INDINGS T RANSFORMER

40 Measurements Meas. Nonrev.

Winding Vdc [V] Diff. (%) Rev. [A] Diff. (%)
[A] [A]
20
1 st 4.09 46.4 47.4 2.1 44 5.1
0
0 2.5 5 7.5 10 12.5 2 nd 3.59 39.9 43.4 8.0 37.7 5.5
Time [ms] 3 rd 4.74 44.8 51.3 18.7 44.5 0.7
(c)
Fig. 10. First peak of inrush currents to validate the three-winding mod-
 8

Zs
Current 4 th 4.72 42.6 47.3 11.0 43.8 2.8
Probe
Switch I1 N1:N2
Eac
IV. CONCLUSIONS
The model of references [8] and [9] has been retrofitted to

+
Edc produce a reversible model for multi-winding transformers.
3900W
Analytical formulae have been derived to calculate the required
DC Generator
parameters from terminal tests. The model can be easily imple-
Fig. 11. Laboratory test setup for the geomagnetic induced currents.
mented in EMTP-type programs for n-winding transformers
B. Four Winding Model since all components are available in their library. The model
The reversible model is validated for the four-winding trans- has been validated by comparing measurements and simula-
former under inrush current, ferroresonance, and GIC. The re- tions for three- and four-winding transformers for inrush cur-
sults are compared to the nonreversible model and measure- rents, ferroresonance, and geomagnetic induced currents.
ments in Tables IV to VI. The great agreement between the The results show the necessity of the proposed improve-
results of simulations for the reversible model and laboratory ment to compute transients involving deep saturation. The
measurements for all of the windings demonstrates the effec- model is physically sound and very simple to implement with-
tivity of the reversible model (all differences are under 5%). On out access to the construction geometry and material infor-
the other hand, the nonreversible model does not properly mation of the transformer. All model parameters can be com-
represent the behavior of all four windings simultaneously. puted from terminal tests.
The errors range from a few percent and up to 24%.
150 V. REFERENCES
Nonreversible Model
[1] V. Brandwajn, H. W. Dommel, and I. I. Dommel, “ Matrix Representation
Reversible Model of Three-Phase N-winding Transformers for Steady-State and Transient Stud-
ies,” IEEE Trans Power App. Syst., vol. PAS-101, no. 6, pp. 1369–1378,
100
Current [A]

Jun. 1982.
[2] V. A. Niemela, G. R. Skutt, A. M. Urling, Y. Chang, T. G. Wilson, H. A.
Owen, and R. C. Wong, “ Calculating the Short-Circuit Impedances of a
50 Multiwinding Transformer From its Geometry,” in Proc. IEEE Power Elec-
tron. Spec. Conf., 1989, pp. 607-617.
[3] H. Mohseni, “ Multi-Winding Multi-Phase Transformer Model with Satura-
ble Core,” IEEE Trans. Power Del., vol. 6, pp. 166–173, Jan. 1991.
0 [4] A. Schellmanns, K. Berrouche, and J. Keradec, “ Multiwinding Transform-
0 10 20 30 40 50
Time [ms] ers: A Successive Refinement Method to Characterize a General Equivalent
Circuit,” IEEE Trans. Inst. Meas., vol. 47, pp. 1316–1321, Oct. 1998.
Fig. 12. Inrush current when transformer coils are drawing geomagnetic [5] R. W. Erickson and D. Maksimovic, “ A Multi-Winding Magnetics Model
induced currents. Note that due to the lack of space only the behavior of Having Directly Measurable Parameters,” in Proc. IEEE Power Electron.
the outer winding is depicted. Spec. Conf., 1998, pp. 1472-1478.
[6] J. Wang, A. F. Witulski, J. L. Vollin, T. K. Phelps, and G. I. Gardwell,
T ABLE IV “ Derivation, Calculation and Measurement of Parameters for a Multi-
COMP ARISON OF I NRUSH CURRENT P EAK VALUES FOR THE FOUR- Winding Transformer Electrical Model,” in Proc. IEEE Applied Power Elec-
W INDING T RANSFORMER tron. Conf. and Exp., 1999, pp. 220-226.
[7] J. M. Lopera, M. J. Prieto, A. M. Pernia, and F. Nuno, “ A Multiwinding
Winding Meas. [A] Nonrev. [A] Diff. (%) Rev. [A] Diff. (%) Modeling Method for High Frequency Transformers and Inductors,” IEEE
1 st 124.4 123.1 1.0 121.2 2.6 Trans. Power Electron., vol. 18, pp. 896–906, May. 2003.
[8] C. Álvarez-Mariño, F. de León, and X. M. López-Fernández, “ Equivalent
2 nd 110.6 131.8 19.2 111.7 1.0
Circuit for the Leakage Inductance of Multi-Winding Transformers: Unifica-
3 rd 104.5 127.8 22.3 100.2 4.1 tion of Terminal and Duality Models,” IEEE Trans. Power Del., vol. 27,
4 th 94.6 117.4 24.1 93.8 0.8 pp. 353-361, Jan. 2012.
[9] F. de León and J. A. Martinez, “ Dual Three-winding Transformer Equivalent
T ABLE V Circuit Matching Leakage Measurements,” IEEE Trans. Power Del., vol. 24,
MAXIMUM T EMP ORARY OVERVOLTAGE MEASURED AND SIMULATED pp. 160–168, Jan. 2009.
FOR THE FOUR-W INDINGS T RANSFORMER AND A 44 µF CAP ACITOR [V] [10] M. Lambert, M. Martínez-Duró, J. Mahseredjian, F. de León, and F. Sirois,
Winding Meas. Nonrev. Diff. (%) Rev. Diff. (%) “ Transformer Leakage Flux Models for Electromagnetic Transients: Critical
Review and Validation of a New Model,” accepted for publication in the
1 st 226.5 214.3 5.4 216.9 4.2 IEEE Trans. Power Del.
2 nd 226.4 214.2 5.4 216.1 4.5 [11] C. M. Arturi, “ Transient Simulation and Analysis of a Three-Phase Five-
3 rd 223.5 214.3 4.1 215.7 3.5 Limb Step-Up Transformer Following and Out-of-Phase Synchronization,”
4 th 221.9 213.4 3.8 214.4 3.4 IEEE Trans. Power Del., vol. 6, pp. 196–207, Jan. 1991.
[12] F. de León, S. Jazebi, and A. Farazmand, “ Accurate Measurement of the
T ABLE VI Air-Core Inductance of Iron-Core Transformers with a Non-Ideal Low-Power
COMP ARISON OF GEOMAGNETIC I NDUCED CURRENTS P EAK VALUES FOR Rectifier,” IEEE Trans. Power Del., vol. 29, pp. 294-296, Feb. 2014.
THE FOUR-W INDINGS T RANSFORMER [13] S. E. Zirka, Y. I. Moroz, C. M. Arturi, N. Chiesa, and H. K. Høidalen,
“ Topology-Correct Reversible Transformer Model,” IEEE Trans. Power
Nonrev. Rev.
Winding Vdc [V] Meas. [A] Diff. (%) Diff. (%) Del., vol. 27, pp. 2037–2045, Oct. 2012.
[A] [A] [14] S. Jazebi, F. de León, A. Farazamand, and D. Deswal, “ Dual Reversible
st
1 4.09 46.4 46.2 0.4 47.3 1.9 Transformer Model for the Calculation of Low-Frequency Transients,” IEEE
2 nd 3.59 39.9 42.4 6.3 40.3 1.0 Trans. Power Del., vol. 28, pp. 2509-2517, Oct. 2013.
3 rd 4.74 44.8 50.2 12.0 46.1 2.9
 9

[15] S. D. Mitchell and J. S. Welsh, “ Initial Parameter Estimates and Con-

straints to Support Gray Box Modeling of Power Transformers,” IEEE
Trans. Power Del., vol. 28, pp. 2411-2418, Oct. 2013.
[16] M. Lambert, J. Mahseredjian, Electromagnetic Transient-Type Transformer
Models for Geomagnetically-Induced Current (GIC) Studies. EPRI, Palo Al-
to, CA: 2013, 3002000832.
[17] G. Kron, Tensor Analysis of Networks. Wiley, 1939.
[18] E. P. Dick and W. Watson, “ Transformer models for transient studies based
on field measurements,” IEEE Trans Power App. Syst., vol. PAS-100, no. 1,
pp. 409–418, Jan. 1981.
[19] C. G. A. Koreman, “ Determination of the magnetizing characteristic of three-
phase transformers in field tests”, IEEE Trans. Power Del., vol. 4, pp. 1779–
1785, Jul. 1989.
[20] S. Abdulsalam, W. Xu, W. L. A. Neves, and X. Liu, “ Estimation of trans-
former saturation characteristics from inrush current waveforms”, IEEE Trans.
Power Del., vol. 21, no. 1, pp. 170–177, Jan. 2006.
[21] S. Calabro, F. Coppadoro, and S. Crepaz, “ The measurement of the magnet-
izing characteristics of large power transformers and reactors through D.C. ex-
citation,” IEEE Trans. Power Del., vol. PWRD-1, no. 4, pp. 224–234, Oct.
1986.
[22] IEEE Standard Test Code for Dry-Type Distribution and Power Transform-
ers, Recognized as an American National Standard (ANSI), IEEE Standard
C57.12.91-1995, 1995.
[23] Slow Transients Task Force of the IEEE. Modeling and Analysis of System
Transients Using Digital Programs Working Group “ Modeling and Analysis
Guidelines for Slow Transients--Part III: The Study of Ferroresonance,” IEEE
Trans. Power Del., vol. 15, pp. 255–265, Jan. 2000.
[24] R. A. Walling and A. H. Khan, “ Characteristics of Transformer Exciting
Current During Geomagnetic Disturbances,” IEEE Trans. Power Del., vol.
6, pp. 1707-1714, Oct. 1991.