A Review of Ferroresonance

John Horak, Basler Electric Company

There is a little confusion in literature as to what constitutes a ferroresonant circuit. In

most, ferroresonance refers to electric circuit conditions where an inductance is oscillat-
ing into and out of its magnetic saturation region and where the oscillations involve an
interaction with local capacitance. Some also use the term to refer to a system with
transformers approaching saturation where there is voltage rise that occurs when XL
and XC are in series and due to the effective canceling of impedance, current rises and
voltages across the elements rise, even though the transformer never enters saturation.

The concerns that are most commonly associated with ferroresonance are high peak
voltage relative to other phases and relative to ground and abnormal voltage wave-
forms (magnitude, phase angle, and harmonic content).

The Ferroresonant Electric Circuit

Before entering into the theory of ferroresonance, it may help to be able to recognize
some typical circuit configurations that have a ferroresonance risk. The listing below is
not exhaustive, but gives an overview. The ferroresonant circuit involves a capacitance
and a transformer core. A very lightly loaded network also is involved. Beyond this
basic data, there are a multitude of possible circuits that may be involved. Basically,
one is looking for series connections of capacitance and a saturable inductance with
minimal burdening resistive elements. As soon as a transformer is involved, XL can be
in two states, either XM,SAT or XM,UNSAT or some value in between leading to multiple pos-
sible system states. In the figures following, a very simplified model of the transformer
is used for brevity, showing the transformer with an unloaded secondary with a core
simply modeled as XM. A more complete model will be covered later.

Single Phase Switching of Delta and Ungrounded Wye Windings

A circuit involving either an ungrounded 3 phase transformer (ungrounded wye, delta,
or single phase transformer connected phase-phase) is a subject to creation of a series
L-C network. The system voltage is set up to look through the magnetizing branch of
the transformer into the line-ground capacitance of other phases. In each case, note
that the equivalent circuit is a series connection of an inductance and capacitance.

In these circuits, the power system defines the voltage across the total circuit, but does
not define the voltage at the midpoints in the circuit, and hence voltage of the open
phase (i.e., across XC) may be high or the voltage across the transformer may be high,
which will create high voltage relative to ground and which may be a risk to surge
arresters and any connected load.

1 Revision Date 02-04-04

Figure 1a: Xfmr Delta Winding, One Phase Energized

Figure 1b: Equivalent Network

Figure 2a: Xfmr Delta Winding, Two Phases Energized

Figure 2b: Equivalent Circuit

2
 Figure 3a: Xfmr Wye-Ungrounded Winding, One Phase Energized

Figure 3b: Equivalent Network

Single Phase Switching of Grounded Wye Transformers with Ungrounded

Capacitor Banks
This configuration is very similar to the previous examples, except the system voltage
looks through an ungrounded bulk power factor correction capacitor bank into the
excitation branch of a grounded transformer. The XC values here are, generally speak-
ing, much lower in magnitude than the previous examples.

Figure 4a: Wye Grounded Winding, with Ungrounded System Capacitance, One Phase Energized

3
 Figure 4b: Equivalent Network

Figure 5a: Wye Grounded Winding, with Ungrounded System Capacitance, Two Phases Energized

Figure 5b: Equivalent Network

Mutual Coupling onto De-Energized Line that has a Connected Transformer

In most power system engineers' experience, the coupling between two lines is dis-
cussed only in terms of zero sequence coupling; i.e., zero sequence current on one
line can induce a zero sequence voltage on an adjacent line. There is a similar but
weaker coupling between two lines that exists when only normal positive sequence
voltages and currents are present. The coupling can be both capacitive and inductive.
Assume lines X (phase A, B, C) and line Y (phase A', B' and C') being run on a common
tower. The coupling may be simply the effect of phase A and A' being closer together
than A-B' and A-C' for an extended distance, hence an open phase A is capacitively
coupled more closely to A' than B' and C' and hence A is partially energized to the

4
potential of A'. Similar single phase weak magnetic coupling can occur as well. In most
cases the coupling is too minor to be considered in power system studies, but it may
be sufficient to create a ferroresonant condition between line to ground capacitance
and a de-energized transformer on an isolated section of a line.

Consider the circuit shown below. The only load on the supposedly de-energized line is
the magnetizing impedance of the transformer, which could be a VT or a power trans-
former. The energized line couples, capacitively and inductively, to the de-energized
line. As the transformer and lines are low loss devices, once the circuit starts to oscil-
late (possibly initiated by a through fault on the energized line that induces a large zero
sequence voltage on the parallel line), a sustained resonance can be set up, limited
mainly by the transformer and line losses.

Figure 6: Mutual Coupling to De-energized line

Reference [17] is a report of an occurrence of this condition.

Breaker Contact Capacitance Feeding a Bus with a Voltage Transformer

Breakers can partially energize a dead bus via the capacitance across open contacts,
especially breakers with multiple breaks surrounded by grading capacitors. Multiple
break circuit breakers are becoming less common, but there are still some being pro-
duced and there is an installed base of these devices. The circuit has the elements for
ferroresonance: a series L-C network created between the power system, the breaker
(and bus) capacitance, and the VTs on the bus. In breakers without grading capacitors,
the capacitance across open contacts is on the order of 50pf, which is on the order of
50megohms at 60Hz, so a ferroresonant condition will likely need to involve several
breakers in parallel before enough current can be pulled to support ferroresonance. For
multiple break breakers the capacitance is on the order of 1000pf.

Figure 7: Breaker with Grading Capacitor Feeding a VT

5
Reference [18] is a report of an occurrence of this condition.

Delta Source Feeding Wye-G/Wye-G Transformers with Ongoing Ph-Gnd Fault

This case is different than the other examples. In this application, a nominal to high
voltage is applied to a parallel L-C connection so that the voltages across the L and C
elements is defined by the source, and now the current in the elements is free to vary
from that seen at the source, rather than the voltage across the elements. One way this
application arises is when a load entity, fed by a Delta/Wye-G transformer (utility on
Delta winding), obtains generation and becomes a Distributed Generator (DG). Upon a
ground fault in the utility system, the utility separates. The DG is unable to feed the
ground fault with this transformer configuration. One protection scheme allowed by
some utilities is for the DG to install a Wye G/Wye G connected VT with a 3Vo calculat-
ing relay, or a Wye G/Broken Delta connected VT with a 59N element monitoring the
broken delta voltage. During a line to ground fault, one VT sees 0 volts, and the remain-
ing two see voltage that is 1.732 per unit of nominal line to ground duty. At this elevated
voltage it is easier to see the saturation of the VT and a subsequent resonance. If the VT
had inadvertently been ordered based upon only line-ground voltage ratings, rather
than line to line voltage ratings, the VT would virtually be assured of saturation on every
ground fault, and hence the risk of ferroresonance would an order of magnitude higher.
A similar saturation and subsequent resonance can be set up between any line-to-
neutral connected power transformers on the line.

If the VT saturates and possibly enters ferroresonance, the secondary voltage will be
distorted, the fundamental voltage output may be severely reduced and phase shifted,
and hence reducing measured 3Vo. A subsequent failure of the relaying to trip may
ensue.

Figure 8: Delta Source Feeding a Ground Fault with Wye-G Transformers

It is common practice in this configuration to connect the VTs secondary in broken delta
and connect a resistor across the broken delta. The intent is to burden the system
down; burdening a ferroresonant circuit is a good way to dampen ferroresonance. This
topic is covered further later in the paper.

6
Capacitive Voltage Transformers
A CVT is a capacitive voltage divider network with a transformer tapped midway into the
capacitance string. Further, inductances are added in series with the transformer and
capacitance to improve regulation and to compensate for phase shifting that the ca-
pacitances introduce. The circuit is an invitation for ferroresonance.

Figure 9: CVT with Possible Ferroresonant Suppression Circuits

All CVT manufacturers are aware of the ferroresonance characteristics of their devices,
and they install appropriate ferroresonance suppression circuits. One suppression
concept is to always keep some burden on the CVT. A second approach is to install a
load in series with a circuit tuned to fundamental frequency. The circuit blocks funda-
ment frequency from the load and passes other frequencies to the load. A third ap-
proach is to add voltage sensitive loading; if voltage increases above normal, saturable
reactors effectively add more load. A fourth approach is to add surge suppressors that
clamp any abnormally high voltage spikes that might be associated with
ferroresonance. A combination of these approaches may be used in a given CVT.

These suppression techniques built into the CVTs removes the need of the user to be
concerned about ferroresonance in the CVT, but the user needs to be aware that these
circuits can adversely affect the performance of the CVT, especially when rapid
changes in voltage magnitude and phase angle are involved. Analysis of the transient
performance of CVTs is the topic of a variety of papers and is not covered here.

Magnetically Coupled Phases in 3 Phase Transformers

One approach to preventing ferroresonant circuits from arising is to treat the three
phases as independent systems to avoid interphase coupling associated with
ferroresonance. This means all loads are connected phase to neutral, rather than
phase to phase. On first pass it may appear that a three phase Wye-G/Wye-G bank
accomplishes this, but that may not be the case. When the phases of a transformer
share a common core, one energized phase can couple into the other disconnected
phases and excite them. If there is a capacitance on those phases, a resonant condi-
tion can be set up.

7
Consider the 4 core/5 legged transformer design below. Assume only phase AP is
excited, that the secondary is totally disconnected, and that BP and CP have some level
of connected capacitance but no connected load. The AP coil sees the flux in cores 1
and 2. Initially, as long as no current flows in BP, about half the flux needed for AP back
emf (i.e., excitation flux) flows in cores 1 and 2. The flux in core 2 however induces
about 0.5 per unit voltage on BP. This VB causes current to be driven into the attached
capacitance on BP. The current on BP decreases net flux in core 2, so an equal amount
of increased excitation current is required on phase A. Hence, an effective 1:1 current
transformer has been set up between phase AP and Bp. When the current in AP in-
creases, this increases the flux in core 1. If the total current on AP is more than normal
excitation current, then core 1 is driven into saturation. Hence, more excitation current
flows in AP, which increases the flux in BP directly proportionately, and the voltage VB
increases. Now there is a 1:1 voltage transformation between AP and BP. A similar trans-
fer of current passes between phase C and phase A, and core 4 operates in saturation
as well. Even core 3, depending on the relative magnitudes of XC on phase B and C,
may be operating partially into saturation. The saturation of cores occurs because the
transformer is designed to transfer current from primary to secondary, on a common
core leg, not from primary to primary. An inspection of the equivalent circuit shows
many combinations of L-C networks that could be subject to resonance.

Figure 10a: 4 core 5 legged transformer design

Figure 10b: Equivalent circuit: Excite Phase A and load B and C with XC

Single core/3 legged transformer (Delta/Wye-G applications) designs and single core/4
legged transformer (Wye-G/Wye-G applications) designs have a similar ability to mag-
netically couple an energized phase to other phases, though a more complex flux
balance analysis is involved.

8
 Figure 11a: 3 Legged Transformer Design, for Delta/Wye-G Applications

Figure 11b: 4 Legged Transformer Design, for Wye-G/Wye-G Applications

References [8] and [9] both showed that the 4 core/5 legged design transformer easily
can go into ferroresonance and [8] showed in tests of actual transformers that the
single core/4 legged transformer has just as much capability to enter ferroresonance as
the 4 core/5 legged transformer. The peak voltages are typically relatively low, on the
order of 1.2 - 1.8 pu, however.

Analysis of a Basic Ferroresonant Circuit

A basic series ferroresonant circuit is shown below. It includes a series connection of a
voltage source, a relatively small circuit resistance, R, a saturable magnetic core induc-
tor comprising the unsaturated LM,US (which becomes LM,S above saturation) and the
leakage inductance LL, and a capacitance C. The inductor in the circuit typically repre-
sents either a power transformer or a voltage transformer rated for continuous voltage
5% to maybe 15% (or higher with some VT applications) above nominal operating
voltage, depending on how it is bought and applied, so the inductor is somewhat close
to entering saturation. Also, consider that there is no transformer secondary load or
primary shunt loading. If only fundamental frequency voltages and currents are in-
volved, the circuit can be converted over to the basic impedance concepts of XL,Total
(referred to as XM for brevity) and XC. The resultant highly basic circuit follows.

9
 Figure 12: Basic Circuit for Analysis

Numerical analysis using specific purpose software or more generic ATP/EMTP type
software is the method required to truly model ferroresonance. However there may be
too many unknowns in a partially saturated multi-phase core transformer (e.g., steel
B-H curve response over large non-uniform core area, with multiple windings, and a
oddly distributed capacitance) to exactly model what occurs during ferroresonance.
The non-linearity of minor loads that may be present further complicate analysis. See
[13] for details on the extensive difficulties involved.

This paper will not broach the task of numeric modeling. This paper will cover two
simplified intuitive approaches to viewing ferroresonant circuits:
• Fundamental Frequency Linear Circuit Analysis. This method is more classical and
has been used by engineers for most of the history of power systems. The ap-
proach sees XM in terms of an effective fundamental frequency value that changes
with the fundamental frequency voltage magnitude.
• Dual State XM Analysis. There are few (if any) papers that actually define this ap-
proach, but it is used or obliquely referred to by various papers that delve deeply
into ferroresonance analysis, such as references [9], [10], [14], and [16]. In this
approach, the transformer is either unsaturated so XM is very high, which allows a
trapped VDC charge to exist in the system capacitances and eventually leads to
transformer saturation; or the transformer is saturated so XM is very low and the
capacitance is rapidly discharging, then recharging with opposite polarity, with a
subsequent return to the unsaturated transformer state.

Some basic concepts of elements of the ferroresonant circuit will be discussed below
before discussing the two basic approaches referred to above.

Transformer Modeling
A common measure of excitation is flux density vs. magnetic field strength (B vs. H).
However, to make the data useful for most power system studies, one needs to have
excitation in terms of V and I, which then is reduced further to V and I at the system

10
fundamental frequency. An interesting method of converting between B-H curves and
the V vs. I is shown. The graph is fairly involved but self explanatory. Start at a given
applied voltage, v(t). Trace upward to find the corresponding flux level, φ(t). Note the
voltage is the derivative of flux (v=df/dt), and if there is no voltage or flux offset from 0,
and the input voltage is a clean sine wave, the flux is a 900 phase shifted reproduction
of the voltage sine wave. Thereafter, trace across to find the corresponding B, then H,
and then the associated current. Finally plot the current against the voltage curve to
see I vs. V. Note the non-linear B-H curve and its hyseresis strongly distort the excita-
tion current.

Figure 13: Steel DC / AC Excitation Curve; B vs. H

It can be seen by this example above that a notable source of error is being introduced
by simply modelling the core as an equivalent XM. The core effective impedance is fairly
complex and has a high harmonic content. However, to further simplify the matter, and
develop a representative transformer voltage vs. excitation current curve, an approxi-
mate AC excitation curve is developed, shown below.

Figure 14: Transformer AC Excitation Curve, VRMS vs. IRMS

11
The RMS excitation curve loses some details of excitation current:
• Since the graph is in terms of RMS quantities, and since voltage is a derivative of
flux rather than a direct measurement of flux, the harmonic quantity, phase relation-
ship, and instantaneous magnitude of current vs. voltage is lost.
• The hysteresis effect is lost.

It may be a hair-splitting effort to carry transformer modeling to a high degree, but to be

complete, a model will need to account for several other aspects of a transformer:

• Eddy current loading effects. It may be very difficult to separate eddy current losses
from hysteresis losses.
• Capacitive current flow including winding to winding and winding to ground.
• Magnetic coupling between phases and the multiple legs of the transformer, each
possibly saturating at a different current level (possibly due to current in neighbor-
ing windings).

In most published analysis of ferroresonance, transformer phase to ground and wind-

ing to winding capacitance is ignored, but as discussed in [9], there are cases where
the transformer capacitance may be important in the analysis. Typical transformer
capacitance may be hard to find. Transformer manufacturers, when testing excitation
current, typically make no attempt to separate the capacitive current flow into the ex-
cited winding from the inductive excitation current. The best likely source of data is
typical transformer insulation testing.

A model of a single phase transformer, including the capacitance, may take the form
below:

Figure 15: Transformer Model

Line Capacitance
One can refer to [22] and [23] for calculation of line sequence component capaci-
tances. However, one should be aware of a couple aspects of sequence component
capacitances.

12
The system positive sequence shunt capacitance is the balanced capacitive loading
each phase sees. The positive sequence shunt capacitive loading includes both phase
to ground capacitance and phase to phase capacitance (XC1=XC,LG+XC,LL, WYE-EQ). The
phase to phase delta impedance, if completely balanced, can be represented as an
equivalent wye load where XC,LN = (1/3)XC,LL and where the neutral of this equivalent wye
will be at ground potential. Hence, to determine the phase to phase capacitance in a
balanced impedance system, one needs to perform the calculation: XC,LL = 3(XC1 - XC0).
In cables XC1 = XC0 so this calculation will show XC,LL = 0.

For real world applications, the phase to phase and phase to ground capacitances are
not the same for each phase. Note sequence component impedances assume sym-
metrical phase impedances; e.g., Zan = Zbn = Zcn. Most sequence component calcula-
tions such as those in [22] and [23] find a mean distance between the conductors and
ground using an RMS-like approach, so one can only approximately recreate the indi-
vidual phase and ground capacitances from XC1 and XC0.

Figure 16: Line Capacitances

Fundamental Frequency Linear Circuit Analysis

There have been numerous technical articles written over the years (see many of the
references) that have used a fundamental frequency linear circuit approach to analyzing
the ferroresonant circuit. The material below cannot give the entire content of these
resources but will serve to give the conceptual frame they work in and summarize some
of their important points. The papers tend toward finding a maximum XC that can be
allowed for a given XM,US that will limit VXfmr and VC to some specified maximum. As long
as the transformer XM remains in the linear XM,US region and system resistive elements are
small, the voltage VXfmr and VC is a simple calculation using linear circuit theory. In most
of these approaches, a minimum ratio for XC/XM is chosen as the guideline. The typical
guidelines for minimum XC/XM ranges greatly depending on the system and the trans-

13
former configuration. For example gathering together the suggestions of Ralph
Hopkinson in his various works:
Winding Connection (primary/secondary) vs. Minimum XC/XM:
Delta / Wye-g 40 for if 3x1 ph, 30 if 5 legged core
Delta/Delta 30
Wye-ug/Delta 30
Wye-g/Wye-g 0 for 3x1 ph; 0.1 for 3x1ph, overhead feeder
1 for 5 legged core

Many of the papers back up their analysis with tests of either actual transformers or
small scale reproductions on transient network analyzer (TNA) systems, including
Hopkinson. These reports generally work by taking actual transformers and connecting
a variety of capacitances or cable lengths to the open phases. They generally do not
report specific waveforms that are achieved but simply report the maximum peak volt-
age measured. In many of the cases they are driving the transformer into a
ferroresonant condition. The limitations on XM/XC they provide tend to verify the linear
circuit analysis, but do not depend upon it, and use little numerical modelling.

Before proceeding, it should be acknowledged that linear circuit analysis has its limita-
tions. On one hand it is backed up by empirical comparison to field test data, employs
a basic intuitive concept, and appears to have been applied successfully by at least a
few utilities. On the other hand the approach is possibly an excessive simplification of a
complex non-linear region of operation. Also, there have been changes in transformer
design over the last couple of decades since the original tests of the concepts were
done. Modern transformers are lower loss and higher capacitance than older designs.
See further discussion under "Preventing Ferroresonance" following and reference [9].

Refer back to the AC excitation curve in Figure 14. Assume that the curve is not a chart
of RMS currents and voltages, but of the fundamental frequency currents and voltages.
If this is approximately true, and we can ignore the high frequency components and
true non-linear nature of what is occurring, then the approximate value of XM is simply
V/I for a given point on the graph. If a transformer can be simply modeled as XM even
into saturation, and we can ignore system resistances and subtle issues such as hys-
teresis flux lag, hysteresis load effects, cross phase coupling in the core, cross phase
coupling via external capacitances, et cetera, then the voltage drop in the simple circuit
shown in Figure 12 is simply:
E= ( jX M − jX C ) ⋅ I
= VM − VC I lagging E by - j
Eq. 1
= VC − VM I leading E by j

If XC remains very high relative to XM, then I and VM are low, and VC is approximately
equal to E. The difficulty is that if XC is sufficiently low, and if the transformer enters

14
saturation so that XM goes very low, there are alternate states that the circuit can oper-
ate in where I is high, VM is low, and VC is high. This alternate operating mode is har-
monically rich and actually falls outside the realm of linear and phasor circuit analysis,
but a basic concept of ferroresonance can be derived using linear analysis of the situa-
tion. The various operating modes are seen in Figures 17-20.

In Figure 17, XC is greater than XM,US. In this condition, no matter what E is applied, even
if the core is sent into saturation, there is one operating point for IEXC where (jXM - jXC)·I =
E. This is the basic safe design to prevent ferroresonance. Note, however, that opera-
tion in this region does not prevent a voltage rise in the circuit that is inherent in all L-C
networks. The voltage rise is limited, however, by selecting an XC that is high (in most
papers this is done in terms of a high XC/XM ratio), which limits current flow, and hence
limits voltage rise.

Figure 17: XC > XM,US

Figure 17 would be a 3 dimensional graph if any resistances were involved. In Figure

17, I either leads or lags by 90°. A resistive component would require the voltages and
currents to have a vector that rises out of the plane of the paper to handle leading and
lagging over any degree.

In Figure 18, XM and XC are approximately equal. To develop sufficient current so that
(jXM - jXC)·I = E, the transformer has to approach saturation. The voltage across XC
relative to ground has risen above nominal voltages. A problem has started, but what is
commonly referred to as a ferroresonant condition has not quite arisen.

15
 Figure 18: XC = XM,US

In Figure 19, the ferroresonant issue begins to be seen. Now there are three points on
the curve where (jXM - jXC)·I = E. Point 2 can be seen, more intuitively than mathemati-
cally, to be unstable because at point 2, if E rises, current must fall to reach the new
operating point. Hence, points 1 and 3 are the possible stable operating conditions.
Currents at points 1 and 3 are 1800 out of phase with one another and much different in
magnitude. The voltages relative to ground are much higher at point 3. An interesting
aspect of point 3 is that once the situation has started, voltage E can be reduced
greatly, and the resonant situation maintained. As voltage reduces, this simply moves
the operating point at 3 to the right, closer to the intersection of the VC and VM,S curves,
but there is no reason for the operation to ever transition beyond the intersection to-
ward the origin, even as E approaches 0, in this simplified view of the circuit, due to the
lack of any resistive burdening elements.

Figure 19: XM,S < XC < XM,US

One last operating region to consider is what occurs as XC continues to decrease, until
the impedance angles meet the condition of ∠θM>∠θC. This might represent the condi-

16
tions where the capacitance involved is a bulk power capacitor. In this condition, it is
hard to build VC and all system voltage is developed across VM. There is only one inter-
section of VC with the VM line, at the origin. There is one valid operating point, shown in
Figure 20.

Figure 20 : XM,S > XC

A second reason to bring Figure 20 to the picture is it gives a hint to the impedances
that are seen by the system to harmonic current flow. Ferroresonance involves a high
degree of harmonic current flow. The system presents a decreasing XC, and an increas-
ing XM,S as harmonic frequency increases.

Including Resistive Elements

A more thorough model is needed to account for the dampening effect of resistive
elements. For instance, consider the circuit below, which may reflect a version of the
circuit in Figures 1-5, or 12. The line impedance contains resistance RL. The transformer
secondary load may not be negligible, so RLoad,SEC is shown referred to the primary and
in parallel with XM. Any load remaining on the open circuited phase(s) is represented by
RLoad,Open Phase, in parallel with the line to ground capacitance.

Resistive elements detune the circuit and change the simple voltage analysis given
previously. The analysis requires complex number math and the due to the three di-
mensional nature introduced by quadrature voltages and currents, the easily grasped
graphical nature is substantially lost.

17
 Figure 21: Circuit Model with Resistive Elements

Resistance will act to remove the voltage rise across the inductances and capacitance
associated with an L-C network. Series line and transformer resistance reduces the
voltage across the reactive elements, and resistive loads shunt current around the
capacitor or transformer. It intuitively can be seen that if RLoad in parallel with XM or XC
approaches respectively XM or XC, then the resonance capability of a circuit goes
very low. This agrees with commonly reported analysis that if transformer load is only a
few % of transformer MVA, ferroresonance commonly is eliminated.

This also gives some insight to why [9] promoted transformer losses as a key (though
empirical) indication of how much capacitance can be added to the 5 legged wye-wye
wound transformers in their studies without risk of the unit entering ferroresonance
during single phasing. The core losses tend to be more than proportionate with trans-
former primary voltages and flux levels, tending to detune the inductive part of the
circuit more linearly with increasing voltage than the unclear effect of secondary loads
on a saturated transformer.

A linear circuit analysis of the above circuit is found in reference [27]. This MS Excel
spreadsheet has an approximate excitation impedance model similar to that shown in
Figure 14, but mapped in terms of fundamental frequency components. In the spread-
sheet, given a set of impedances from the user for the elements in the figure above, the
voltages and currents in the network for a range of ITransformer are calculated. The spread-
sheet determines the fundamental voltages that will result from the design, which the
user then compares to actual system voltages to see if more than one ITransformer will
correspond to the same system voltage E, indicating a circuit that is capable of
ferroresonance.

18
 Figure 22: Linear Circuit Analysis

Dual-State XM Analysis
The fundamental frequency linear circuit analysis above aids in a conceptual under-
standing of some aspects of ferroresonance, but is severely lacking in details of what
occurs during a ferroresonant condition, and in fact, once one has an improved under-
standing of what occurs in ferroresonance, one should have a suspicious discomfort
with projecting linear circuit analysis into the saturated region of a transformer.

A ferroresonant condition might be better seen as a circuit that is repeatedly switching

between two alternate states (two values for XM: XM,S and XM,US), rather than as a circuit
that is an extrapolation of the unsaturated linear state with a single valued XM. In state 1,
the capacitor is charged and impressing a voltage across the transformer. The capaci-
tor voltage adds to the voltage presented by the system. The impedance of XM of an
unsaturated transformer is high enough to be considered an open circuit to charges
trapped in a series capacitance. The capacitor is effectively a DC value that eventually
will saturate the transformer. The time that it takes to saturate the transformer under the
application of a given voltage wave is dependent upon the area under the voltage wave
form. In state 2, the transformer has reached saturation and the capacitor is discharg-
ing and then recharging to an opposite voltage state, at which time the transformer
returns to state 1.

Before walking through an example ferroresonant condition, a short review of the equa-
tion relating current and voltage in pure capacitances are provided below. The equa-
tions that relate flux, current, and voltage, for a short review are:

d d
v(t) = ϕ( t ) = L i(t)
dt dt
ϕ( t ) = ∫ v ( t ) dt + ϕ 0 = Li ( t )
Eqs 2-4
1
i( t ) =
L
∫ v ( t ) dt + i 0

A negative may show up in some forms of the above equations when one wishes to

19
indicate that induced current will oppose flux buildup (Lenz's law). A similar collection
of equations for a pure capacitance, for completeness:

d d
i( t ) = QC (t) = C v(t)
dt dt
Q C ( t ) = ∫ i ( t ) dt + Q C , 0 = Cv ( t )
Eqs 5-7
1
v(t) =
C
∫ i ( t ) dt + v 0
When the transformer saturates, the trapped charges will discharge and generate high
frequency currents. The discharge waveform is dependent upon the resonant fre-
quency of the circuit during the discharge process:

1
ω = 2 πf X L = ωL XC =
ωC

1 X C , 60 Hz Eqs 8-11
f Resonance, (XL = Xc) = = 60
2 π LC X L , 60 Hz

When a manufacturer designs a transformer, the integration of voltage more than 1/2 of
a cycle (equation 3 above), with 0 initial flux and some degree of permissible continu-
ous overvoltage, defines the peak flux that the core must support. Then the designer
selects a steel with a defined B−H characteristic. With B nearing saturation, the core
must be large enough to carry the required flux as defined by the following equations,
which in turn define the excitation current requirements of the transformer:
ϕ = N Turns ⋅ B ⋅ Area Cross section
B = µH ( µ at steel peak flux)
Eqs 12-14
H = I Exc ⋅ N Turns / core length
The transformer manufacturer hence has only a few parameters to work with in trans-
former design.

One does not need to know the specifics of a transformer's design implied by Eqs. 12-
14 in order to determine if an applied voltage will cause saturation. Note in equation 3
that flux is an integration of voltage. The transformer has to be designed to accept the
normal sine wave voltage applied to it. For any voltage wave that is applied to a trans-
former, along with an assumption of some corresponding initial flux φ0, one can deter-
mine if the wave is capable of driving a core into saturation by comparing the area
under this curve to the sine wave voltage the transformer is rated for. This can be done
visually in simple applications for an intuitive feel of the situation. Once this saturation
level is reached, even transiently, there is risk that a ferroresonant condition may begin
and continue indefinitely.

20
The details of what may occur in a dual state XMferroresonant circuit may be found by
reviewing a few sample wave forms. The examples below were derived from a study of
a number of resources, but much of the credit should be given to [10], [14], and [16] in
their concepts.

Assume the circuit shown in Figure 12, and assume a simple flux vs. excitation current
relationship shown in Figure 17. A possible trace of system voltage, capacitor voltage,
flux, and current ((e(t), vC(t), φ(t), and i(t)) is shown in Figure 23. Assume that at time t0
there is no flux in the transformer, and before time t1 that XM is very high relative to XC.
Virtually all of the system voltage is across the transformer. Excitation current and
hence transformer flux rises until at time t1 the transformer enters saturation. At this
point the excitation current becomes relatively large, XM becomes low, and hence the
capacitor starts to charge rapidly. Midway between time t1 and t2 the capacitor voltage
reaches the system voltage, so that the voltage across the transformer is now 0, but
leakage inductance in the lines and transformer force current to continue and then
gradually decay until time t2, when current drops low enough for the transformer to
leave excitation. Hence, the first current spike has been formed and at time t2 the ca-
pacitor voltage is possibly, for this example, 0.75 times the peak system voltage.

When the transformer enters saturation there will be an L-C-R circuit in operation that
will have a very large impact on the ferroresonant waveform. Note the half sine wave
that is associated with the discharge and subsequent over-swing and recharge of the
capacitance. One can intuitively see a notably different waveform will result if fResonance for
the saturated condition is very low, giving a slow discharge of the capacitance, com-
pared to what the waveform will look like if fResonance for the saturated condition is very
high, giving a fast discharge of the capacitance. Further, if resistance is high, then the
capacitively held energy will be less on each transition, dampening the resonance.

In the next half cycle, the voltage across the transformer is not just e(t), but e(t) + vC(t).
Hence, excitation current and transformer flux builds at a much higher rate, but the
excitation current has to be totally reversed. At time t3 the transformer enters saturation
again, but with flux in the opposite direction compared to t1. Because the voltage
across the transformer includes the capacitor voltage, a larger current spike occurs
than at t1. The charge/discharge/recharge routine continues one more time, then at t4
an interesting point is reached where the capacitor voltage vC(t) is pulled in phase with
e(t). Then the condition becomes even more interesting. If the system voltage were
reduced at this point, the voltage across the transformer would actually increase (note
vXfmr(t) = e(t) - vC(t) ). Inspecting the waveform, it can be seen this means the resonant
condition could continue indefinitely even if e(t) were reduced to a magnitude well
below a value that would not have driven the transformer into the initial saturation
condition.

21
 Figure 23: Example Square Wave Capacitive Voltage

The previous example does not show vXfmr(t), but it can be seen by the reader by sub-
tracting vC(t) from e(t). The next example shows another case where vXfmr(t) in a steady
state condition and shows how a sub-harmonic voltage is developed. The integration of
the voltage across the transformer determines when the core will enter saturation. It is
possible for this time to exceed one system frequency cycle. (It should be noted that
hysteresis also has a role in causing sub harmonic oscillations.) If this time exceeds
one cycle, then the ferroresonant circuit will oscillate at the fundamental frequency and
the given sub-multiple of the system frequency. The figure below shows the voltage
across the transformer for two symmetrical cases of the transformer entering saturation
at approximately the same point in the voltage waveform on each cycle. See [10] for
additional analysis of this waveform type.

Figure 24: Example Transformer Voltages

22
The winding resistance, eddy current losses, and hysteresis losses of the transformer
tend to dampen out the LC oscillations. This effect is discussed in references [9] and
[10], and is the particular factor that [9] argues is the critical factor in determining how to
prevent ferroresonance. See “Preventing Ferroresonance,” following, for additional
discussions.

Ferroresonance involves multiple time constants, transformers entering saturation at

various points on the applied voltage wave form, different capacitive discharge time
constants, and hysteresis effects that allow some current reversal without a change in
core flux. These compound to create waveforms much messier than shown in these
examples. There are both higher frequency components and a random nature to the
events. There may be components randomly moving between the in phase and 1800 out
of phase form seen in Figure 23. Voltage peaks may be close to nominal voltage peaks,
or may be 2-4 times nominal. There is commonly at least a partial chaotic look to the
waveforms, such as seen in Figure 25.

Figure 25: Chaotic Waveform

Preventing Ferroresonance
The most common recommended means to prevent ferroresonance is to always switch
transformers using three phase devices, so as to prevent a series inductive/capacitive
(SLC) circuit. Directly related to this is a recommendation to energize transformers using
switches directly adjacent to the transformer, so a core is never energized from one
terminal with an open circuited long un-energized capacitive circuit (esp. a cable) con-
nected on another terminal.

When transformers are energized via ungrounded windings (delta, ungrounded wye,
open delta, Scott-T, etc.) special care should be made to implement ferroresonance
prevention techniques because such windings have the greatest ability to create a series
LC network.

23
If a circuit must be switched single phase remote from the transformer, then limit the
capacitance on the open phase, which translates to limiting the length of the cable or
overhead conductor on the circuit. Some references that give guidelines on the length
of cable that would be used include [1] - [4]. A summary of the suggested XC/XM from
one source was given earlier in this paper.

Reference [9] and members of the DSTAR organization (see below) have come to the
conclusion that allowed attached capacitance on the open phase during single phase
switching should be based upon transformer loss levels rather than XM but DSTAR does
not publish specifics for the general public. Reading between the lines a bit more than
possibly clearly stated, reference [9], in tests of three phase 5 legged/4 core and 4
legged/single core transformers, promotes limiting ferroresonance risks by limiting the
ration of BC/PCore Loss (=XC/RLoss @ V=1pu) and appears to suggest XC/RLoss be limited to
0.5 or less. It should be noted that reference [9] is a summary of some investigations
by DSTAR into ferroresonance, so the paper is not a full release of their results.

If a circuit must be switched single phase and the circuit has the possibility of setting
up an SLC circuit, have a load on the transformer secondary or in parallel with the
capacitance to detune the resonant circuit. Most papers show that a load on the trans-
former secondary only needs to be a few percent of the transformer rating to block
ferroresonance though in some references cases are shown where required load
exceeded 10%. This is approximately equivalent to detuning the XM branch so that it
has an apparent impedance angle of around 450 or lower (i.e., RLOAD>XM,US). Since
during transformer saturation the effective value of XM becomes lower, transformer
action may not be complete, the secondary may not see the same voltage as seen on
the primary, and loads may not respond to the voltage waveforms seen during
ferroresonance, a load resistance closer to |XM,S| may be needed to detune the trans-
former reactance during ferroresonance. Similarly, an RLoad in parallel with XC to create a
net phase angle of 450 should be sufficient, to block ferroresonance.

The load values discussed above are not tested concepts; the references should be
consulted for a more complete discussion of how much load is required to block
ferroresonance.

Transformers may be bought with a low flux density design. Such transformers will of
course be larger. A low flux density design will be harder to drive into saturation. Once
in saturation, they still have the ability to be a non-linear circuit through which a system
capacitance is charged and discharged. A larger core will have higher core loss associ-
ated with hysteresis and eddy currents, so such a design may serve to cause the
transformer to be self loading if it enters into a ferroresonant condition.

Surge arresters have some minor ability to block low level ferroresonant conditions.
See [19] and [20] for some in-depth analysis.

24
Associated Issues
Peak Voltages and Currents
One of the large concerns of ferroresonance is the peak voltage that will be generated
and the resultant damage that can occur. In delta-wye and other ungrounded primary
transformers, fed by a grounded source, it is commonly mentioned in papers on the
topic and by those performing tests of sample transformers, that peak voltages of 2 per
unit (relative to normal sine wave peaks), with occasional spikes to even 4 or 5 per unit,
are easily obtained. See results found in many of the references.

When the more common three phase 5 legged/4 core wye-g/wye-g transformer is
used, ferroresonance is still possible, but the peak voltages are much less. Peak volt-
ages are on the order of 1 to 1.5 per unit, with some short spikes to 2 per unit. The
waveforms have low enough energy that high energy surge arresters (especially types
that soak in the transformer oil) have the capability of absorbing the energy and
detuning the ferroresonant circuit without much risk of damage. See [19], [20], and
[21].

The current levels associated with ferroresonance are not typically very high, peaking
on the order of 1-20% of transformer rating. Note that are excitation level currents. Most
reports of ferroresonance do not record current levels, so it may be the case that some-
times current is high but not noted.

Protective Relaying
Relays commonly are not located at a point in the power system to detect ferroresonant
conditions on the distribution network. Further, a relay is typically designed to operate a
breaker, so if there is a relay, the single phase switching that is associated with
ferroresonance would likely not be occurring. However, some cases may arise where a
relay may see the condition. For example, a relay could sense ferroresonance on the
VT that it is using to monitor a circuit. Another issue is that protective relays are com-
monly fundamental frequency sensing devices. Ferroresonant circuits have a distorted
waveform that may have high peaks but have low fundamental components. Hence,
the typical relay may have trouble sensing the ferroresonant condition. One approach
would be to have the relay trip if it repeatedly senses a random single sample high
voltage spike.

Surge Arresters
Surge arresters are at risk of being damaged by ferroresonance conditions associated
with Figures 1-5. The voltage peaks and energy involved may be too high to be ab-
sorbed by surge arresters reliably. However, for the 4 and 5 legged grounded wye
transformer associated with Figure 6 and 7, the voltage peaks and energy involved in
ferroresonant conditions are not as large. References [19] and [20] indicate that for
these transformer configurations, a surge arrester with sufficient energy absorption
capability can withstand the condition and in a few cases help dampen the event.

25
Wye - Broken Delta Ground Fault Sensing
A classical method to sense ground faults on ungrounded systems is with a Wye-Gnd/
Broken Delta VT, and then monitor the broken delta voltage, which will be 3V0. VTs in
this application have the elements of ferroresonance. When VTs are purchased for this
application, they should be bought rated for full line to line potential across each Wye-
Gnd VT due to the high voltage that will be seen by the windings during line-ground
faults. The VT will by nature operate at a lower flux density under normal operating
condition.

It is typical for such VTs to have a resistor in the broken delta to dampen out
ferroresonant conditions. In [25], two approaches to selecting the resistance are ana-
lyzed:

A) Determine the rated current of the VT, then size the resistor such that rated current
flows in the delta during a line to ground fault.
B) Determine the 3 phase VA rating of the VT, and size the resistor so that full trans-
former VA is applied to the VT during a line to ground fault. This latter method will
run higher current than rated through the transformer, but this may be satisfactory
for short time events.

It is shown in tests in [10] that burdening down the broken delta is not a 100% reliable
method of preventing ferroresonance. The broken delta approach assumes that the
ferroresonance condition is creating zero sequence voltage. In [10] it is shown that
sometimes such transformer banks can enter into a three phase ferroresonance that
may produce negligible voltage at the broken delta resistor.

Issues with Distributed Generation

Distributed generation adds increased risk that the power system will be fed from
directions not normally intended, and hence an increased risk of single phasing of the
power system.

Induction generators where sufficient power factor correction capacitance has been
installed can result in an unexpected ferroresonance condition. Induction generators
and motors, when power factor correction approaches creating a unity power factor
condition, may not automatically shut down when the utility opens its breaker. In fact,
continuous high voltages may exist. This paper will not review this application; see
reference [24].

Transformer Damage
There are some reports of transformer damage has been associated with
ferroresonance, but this is likely more due to insulation damage than core heating
effects. If the voltages that are obtained do not damage the insulation, then immediate
transformer damage is not too likely. The current involved are fairly low. The transformer

26
is being driven deep into saturation, so eddy current losses and hysteresis losses are
high. However, ferroresonance occurs on unloaded transformers, cannot be maintained
well in a high loss circuit, and the current required for ferroresonance is low, so the
heating effect on the transformer is low. Informally, a DSTAR representative (see below
has stated to the author of this paper that they have seen ferroresonance continue for
matters of hours with no overly high heating.

DSTAR Ferroresonance Studies

Members of the DSTAR (Distribution Systems Testing, Application, and Research organi-
zation (www.dstar.org) performed some extensive ferroresonance studies in the 1990s
(www.dstar.org/Public/ferroresonance.htm). The organization is supported by members
that pay for research work, so DSTAR will not release information to the general public
(e.g., declined to share information with the author of this paper) but have published
some of their work in the form of references [9], [20], and a couple other papers, and
placed some information on their website as well.

Summary
The basics of a ferroresonant condition have been covered:
• The network circuits involved
• Two simple analysis processes:
• Linear Circuit Analysis Concept
• Dual XM Analysis Concept
• A review of how ferroresonance is prevented
• A variety of other issues were discussed

References
Databases such as IEEE Explore (http://ieeexplore.ieee.org) or Engineering Village
(http://www.engineeringvillage.org) list literally hundreds of documents on the subject
“ferroresonance.” These databases mainly cover papers since the mid 1980s and do not
contain the contents of many good resources such as George Tech, Texas A&M, and
Washington State Protective Relay Conferences, Doble Conferences, and Edison Elec-
tric Conferences, and the early GE and WH materials. Most good foundational and basic
explanatory papers predate these databases. The following list is a subset of the many
papers that are available with some stress on early foundational documents.

1 Ralph H. Hopkinson, “Ferroresonance During Single-Phase Switching of 3-Phase

Distribution Banks,” IEEE Transactions on Power Apparatus and Systems (PAS),
p289-293, PAS-84, April 1965, discussion June 1965, p514-517
2 Ralph H. Hopkinson, “Ferroresonant Overvoltage Control Based on TNA Tests on
Three Phase Delta-Wye Transformer Banks,” IEEE Trans. on PAS, p1258-1263, PAS-
86, Oct. 1967
3 Ralph H. Hopkinson, “Ferroresonant Overvoltage Control Based on TNA Tests on
Three-Phase Wye-Delta Transformer Banks,” IEEE Trans. on PAS, p352-361, PAS-87,
Feb. 1968

27
 (Note: Ralph Hopkinson wrote other material on Wye-g/Wye-g transformers, but as
best I can tell this material remained as internal GE documents. I have not found
where they were published for wide distribution.)
4 S. Prusty, M. Panda, “Predetermination of lateral length to prevent overvoltage
problems due to open conductors in three-phase systems,” IEE Proceedings, p49,
Vol 132, Pt. C, No.1 Jan. 1985
5 Philippe Ferracci, “Ferroresonance,” Group Schneider Cahier Technique Series, No
90, http://www.schneider-electric.com/en/expert/e2a.htm
6 George E. Kelley, “The Ferroresonant Circuit,” AIEE Transactions on PAS, p843-848,
PAS-78, Pt III, Jan 1959; discussion p1061
7 P. E. Hendrickson, I. B. Johnson, N. R. Schultz, “Abnormal Voltage Conditions Pro-
duced by Open Conductors on 3-Phase Circuits Using Shunt Capacitors,” AIEE
Transactions on PAS, p1183-1193, PAS-72, Pt III, Dec. 1953
8 D. R. Smith, S. R. Swanson, J. D. Borst, “Overvoltages with Remotely Switched
Cable Fed Grounded Wye-Wye Transformers,” IEEE Transactions on PAS, p1843-
1853, PAS-94, Sept. 1975
9 R. A. Walling, K. D, Parker, T. M. Compton, L. E. Zimmerman, “Ferroresonant Over-
voltages in Grounded Wye-Wye Padmount Transformers with Low-Loss Silicon Steel
Cores,” IEEE Transactions on Power Delivery, p1647-1660, Vol. 8 No. 3, July 1993
10 Elmo D. Price, “Voltage Transformer Ferroresonance in Transmission Substations,”
Texas A&M 30th Annual Conference for Protective Relay Engineers, April 25-27,
1977
11 Harold Peterson, book: “Transients in Power Systems,” Chapter 9: “Overvoltages
Caused by Open Conductors,” John Wiley and Sons, 1951, then republished with
slight corrections by Dover Publications for GE, 1966
12 Reinhold Rüdenberg, book: “Transient Performance of Electric Power Systems,”
Chapter 48, “Saturation of Iron in Oscillatory Circuits,” McGraw Hill, 1950, then
reprinted 169/1970 by MIT Press. ISBN 0 262 180367
13 M. R. Iravani, Chair, IEEE Working Group, Slow Transients Task Force of IEEE Work-
ing Group on Modeling and Analysis of System Transients Using Digital Programs,
“Modelling and Analysis Guidelines for Slow Transients - Part III: The Study of
Ferroresonance,” IEEE Transactions on Power Delivery, p255-265, Vol 15, No, 1,
Jan. 2000; Discussion in Trans. on Pwr Del., Vol 18, No 2, April 2003.
14 Frank S. Young, Ronald L. Schmid, Petter I Fergestad, "A Laboratory Investigation of
Ferroresonance in Cable Connected Transformers," IEEE Transactions on PAS,
p1240-1249, PAS 87, May 1968
15 Eugene C. Lister, “Ferroresonance on Rural Distribution Systems,” IEEE Transac-
tions on Industry Applications, p105-111, Vol IA-9, No. 1, Jan/Feb 1973.
16 R. H. Dennard, “Behaviour of the Ferroresonant Series Circuit Containing a Square-
Loop Reactor” AIEE Transactions, Communications and Electronics, p903-911, Jan
1959
17 E. J. Dolan, D. A. Gillies, E. W. Kimbark, “Ferroresonance in a Transformer Switched
with an EHV Line,” IEEE Transactions, p1273-1289, PAS-91, 1972

28
18 W. S. Vilcheck, M. V. Haddad, “Voltage Transformer Ferroresonance in Cogeneration
Substation,” Pulp and Paper Industry Technical Conference, p 148-158, Annual Mtg,
Jun 8-12, 1992, Available via IEEEXplore
19 Thomas A. Short, James J. Burke, Ramon T. Mancao, “Application of MOVs in the
Distribution Environment,” IEEE Transactions on Power Delivery, p 293-305, Vol. 9,
No. 1, Jan. 1994
20 R. A. Walling, R. K. Hartana, M. P. Sampat, T. R. Balgie, “Performance of Metal Oxide
Arresters Exposed to Ferroresonance in Padmount Transformers,” IEEE Transac-
tions on Power Delivery, p788-795, Vol. 9, No. 2, April, 1994
21 Leonard J. Bohmann, John McDaniel, E. Keith Stanek, “Lightning Arrester Failure
and Ferroresonance On A Distribution System,” IEEE Transactions on Industry
Applications, p1189-1195, Vol. 29, No. 6, Nov. 1993.
22 William D. Stevenson, text: “Elements of Power System Analysis,” McGraw Hill,
1982
23 Westinghouse (ABB) Central Station Engineers, “Electrical Transmission And Distri-
bution Reference Book,” Westinghouse/ABB 1964
24 Ferroresonance and Loading Relationships For DSG Installations,” W.B. Gish, W. E.
Feero, S. Greuel, IEEE Transactions on Power Deliver, p953-959, Vol PWRD-2, No 3,
July 1987
25 Application Note PC-59N01, "The 59N and Broken Delta Applications," available at
www.basler.com

The mathematically inclined should refer to the references listed in Reference 13. Also
refer to Reference 26; it has a listing of some early highly technical resources on the
topic:

26 Glen Swift, “An Analytical Approach to Ferroresonance,” IEEE Transactions on

Power Apparatus and Systems, p42-46, Vol PAS-88, No. 1, Jan 1969

As a last reference:

27 Excel Spreadsheet to Analyze the Circuit Shown in Figure 21. “Ferro_R1.xls” avail-
able at www.basler.com

AUTHOR'S BIOGRAPHY
John Horak received his BSEE degree from the University of Houston in 1988 and his
MSEE degree from the University of Colorado in 1995. He worked for nine years with
Stone and Webster Engineering and was on assignment for six years in the System
Protection Engineering offices of Public Service Company of Colorado. His work has
included extensive relay coordination studies and settings, as well as detailed control
design and equipment troubleshooting. He has presented technical papers at Texas
A&M Relay Conference and Georgia Tech Protective Relaying Conference. John is a
Senior Application Engineer for Basler Electric, based in Colorado, and is a member of
IEEE, IAS and PES.

29