Power Acceptability and Voltage Sag Indices in the Three Phase Sense

R. S. Thallam
Salt River Project
Phoenix, AZ
rsthalla@srpnet.com

G. T. Heydt
Arizona State University
Tempe, AZ
heydt@asu.edu

Abstract
The electric power acceptability curves are an
empirical set of curves that represent the intensity
and duration of bus voltage disturbances. These
curves are discussed with regard to the energy
delivered to the load, and alternatives for the
assessment and measurement of bus voltage sags.
Special attention is given to the three phase case.

The fact is that few real world loads are suitable for
the use of Equation (1): most load currents are
aperiodic and harmonics are only applicable to
periodic i (t). Further, which phase current (in a three
phase circuit) should be used in (1); where can the
sum in the numerator be truncated; the failure of
THD to capture the generally more harmful effects of
high frequency is problematic; and what should one
do when the fundamental term is missing entirely?
Yet one may generally readily calculate Equation (1):
many instruments do the calculation automatically;
any time series of measurements can be passed
through a fast Fourier transform to find the harmonic
components and subsequently find the THD. Most
instruments that calculate THD will produce a
numerical value even when the basic elements of
Equation (1) are invalid. These few remarks indicate
the tradeoff between simplicity of calculation and
mathematical validity. There is also a tradeoff
between simplicity of calculation and ability to
capture complex phenomena. Figure (1) illustrates
these remarks.

1. Introduction
There are a number of electric power quality
indices that have been applied (and misapplied) to
reflect flicker, momentary outages, the impact of
harmonics, high voltage conditions, power factor,
losses, electromagnetic interference, and other
phenomena. Many of these indices have evolved
over the years as measures that capture a given
phenomenon. There is always a tradeoff between the
complex and the simple: when the physical world is
highly complex, engineers often appeal to indices as
a simple quantity that more-or-less captures a given
phenomenon.
As a quick example, consider
harmonic currents in a nonlinear load. In this case,
the load may be a very complex physical process; it
may be dynamic; it may be chaotic; and it may have
time dependence. In many cases, load currents are
dependent on non-electrical phenomena such as
temperature, heat flow, pressure in a process,
chemical composition, and other factors.
The
invention of the total harmonic distortion,

THD

I
i2

2
i

Mathematical
Validity

(1)
Simplicity Of
Calculation

I1

(Ii are the harmonic components of load current i(t)),

gives a simple measure that is fairly easily calculated.
When the THD is high, various problems can be
assumed to occur. When the THD is low, all is well.
Or that is the assumption.

Capture Of Complex
Phenomena

Figure 1. Tradeoffs of power quality indices

In this panel presentation, there is a discussion of
the power quality indices, especially applicable to
voltage sag events, to reflect three phase phenomena.
The power acceptability curves, like THD above, are
simply applied, may be automated, and give some
information on high voltages (swells or surges)
and low voltage conditions (sags). Information on
momentary events including outages is captured.
But what are the limitations of the calculation? And
can the curves be redesigned or reinterpreted to give

For presentation at the Panel Session on Power

Quality Voltage Sag Indices IEEE PES Summer
Meeting, July 2000, Seattle, WA

additional information especially in the three phase

sense? Each of these areas is discussed in general
and some suggestions and recommendations are
made.

supply. That is, when disturbances occur, one may

plot the disturbance on the curve, and readily identify
whether loss of load is expected. This information
may be used to identify whether the distribution
system design is adequate, whether power quality
enhancement equipment is needed, and whether load
vulnerability needs to be lessened. Unfortunately, the
power acceptability curves may not capture all the
complex phenomena of low and high voltage
conditions, and blind use of the curve may give false
conclusions. As a quick example, the CBEMA curve
was originally designed to identify mainframe
computer (vintage 1970s) vulnerability to power
supply disturbances. But the curve has been applied
to adjustable speed drives, fluorescent lighting,
general (unspecified) loads, and modern solid state
(and microprocessor based) computer loads. One of
the most visible power quality problem found in the
United States at this time is the undervoltage event.
This type of event occurs when a fault occurs in the
transmission, subtransmission, or distribution system.
The flow of fault current generally causes a
depression of bus voltages throughout the system
until the fault is cleared. The bus voltage sag is
plotted on the power acceptability curve: the intent is
a rapid way to determine whether the sag is tolerable.

2. The Power Acceptability Curves

The power acceptability curves are loci drawn in
the bus voltage - duration time plane. These loci
quantify the acceptability of supply power as a
function of duration versus magnitude of bus voltage
disturbances. Figure (2) shows a well cited power
acceptability curve known as the CBEMA (i.e.,
Computer Business Equipment Manufacturers
Association) curve. Examination of the CBEMA
curve reveals that there are two loci: the overvoltage
locus above the | V | 0 axis and the
undervoltage locus below this axis. The ordinates of
these loci represent the intensity of a bus voltage
amplitude disturbance. This is measured as a
deviation in voltage amplitude from the rated value.
For the CBEMA curve depicted in Figure (2), the
ordinate is shown in percent -- a percent deviation
from rated voltage. Thus the | V | 0 axis
corresponds to operation at rated voltage. The
abscissa represents the duration of the event being
studied. The disturbance time duration is usually
expressed in either cycles or seconds (Figure (2)
shows values of time in seconds on a logarithmic
scale). Steady state is at t and short term
events occur to the left on the time axis.
Overvoltages of extremely short duration are usually
tolerable if the event occurs below the upper limb of
the power acceptability curve -- that is, in the
acceptable power region. Examples of overvoltage
events that may occur are: lightning impulses, line
switching surges, and capacitor switching surges.

There is not a single power acceptability curve,

but several that are described in Table (1). The
CBEMA curve was redesigned in early 1996 and
renamed for its supporting organization, the
Information Technology Industry Council (ITIC).
The main difference in the ITIC curve and the old
CBEMA curve is that the newer curve describes an
acceptable operating region in steps rather than as a
smooth curve. Instrumentation to check compliance
with the ITIC curve is more easily designed. The
new ITIC curve, depicted in Figure (3), has an
expanded (or enhanced) acceptability region for
portions of the |V | t plane. Like the CBEMA
curve, the ITIC curve is recommended as a design
target for manufacturers of computer equipment. It is
conjectured that the ITIC curve will be applied to a
wide range of unintended applications as the
CBEMA curve has been.

250

200

OVERVOLTAGE CONDITIONS

0 .5 C Y C L E

100

50

RATED

ACCEPTABLE
POWER

3. Three Phase Issues

As indicated above, there are numerous complex
issues of load vulnerability to power supply
disturbances, and many phenomena in the three phase
primary distribution supply that are not captured by a
simple power acceptability curve. Perhaps the most
evident issue is that the load is usually a dynamic
process, and one may question how a simple static
curve can represent load vulnerability under all

VOLTAGE

8 .3 3 m s

P E R C E N T C H A N G E IN B U S V O L T A G E

150

-50

UNDERVOLTAGE CONDITIONS

-100
0.0001

0.001

0.01

0.1

10

100

1000

TIME IN SECONDS

Figure 2. The CBEMA power acceptability curve

The general application of the CBEMA curve

as well as other power acceptability curves described
below is to identify the acceptability of the service

operating regimes. This and other factors are listed in

Table (2) as frailties of power acceptability curves.

sequence supply do not result in energy supplied to

the load.

250

Table 2. Frailties of static power acceptability curves

200

OVERVOLTAGE CONDITIONS

Main difficulty
The load is a dynamic
process, and a static
curve may be unable to
identify
load
vulnerability

Three
phase

The
primary
distribution system is
usually three phase.
The
power
acceptability
curves
seem to be single line
(or
single
phase)
representations.
The
power
acceptability
curves
show |V| versus time.
But
voltage
disturbances depend
on where in the cycle
the disturbance occurs
(for short duration
disturbances).
No
phase information is
plotted on the power
acceptability curves.
The
power
acceptability curves no
not consider harmonics
or deviations in supply
frequency.
Repeated disturbances
(occurring near in time
to each other) are not
considered.

0.5 CYCLE

100

50

+-- 10%
0

RATED

ACCEPTABLE
POWER

VOLTAGE
8.33 ms

PERCENT CHANGE IN BUS VOLTAGE

150

Factor
Load
dynami
cs

-50

UNDERVOLTAGE CONDITIONS

-100
0.0001

0.001

0.01

0.1

10

100

1000

TIME IN SECONDS

Figure 3. The ITIC power acceptability curve

Short
term
phenom
ena
phase
informa
tion

Table 1. Alternative power acceptability curves

Curve
FIPS power
acceptability
curve

Year
1978

Application
Automatic data
processing
(ADP)
equipment
Computer
business
equipment

Source
U.S. federal
government

CBEMA
curve

1978

ITIC curve

1996

Information
technology
equipment

Failure rate
curves for
industrial
loads
AC line
voltage
tolerances
IEEE
Emerald
Book

1972

Industrial loads

1974

Mainframe
computers

IEEE Standard
446

1992

Sensitive
electronic
equipment

IEEE Standard
1100

Computer
Business
Equipment
Manufacturers
Association
Information
Technology
Industry
Council
IEEE Standard
493

Freque
ncy

Multipl
e events

Long
term
events
Ground
ing

For the case of loads that behave like AC to DC

converters, most energy transfer may be viewed from
the point of view of a frequency converter. That is,
the switching in the converter is done synchronously
in such a way that positive sequence 60 Hz is
converted to DC. By this reasoning, negative
sequence supply voltage will be converted to 120 Hz
in the DC link, and zero sequence 60 Hz in the
supply voltage converts to 60 Hz in the DC link.
Assuming that the ultimate load can only accept DC
from the DC link, the only energy transfer occurs
when the supply voltage is in positive sequence 60
Hz.
Therefore it is suggested that power
acceptability should be applied to the positive
sequence supply voltage.
Negative and zero

Wrong
energy
model

Are long term events

the province of the
power
acceptability
curves?
Neutral to ground
voltage not modeled.

The
given
power
acceptability
curve
does not properly
model the vulnerability
of a selected load.

Potential problem
The power acceptability
curve may falsely dismiss
a problematic power
supply condition or may
falsely
identify
an
acceptable power supply
condition
Inability to identify cases
of unbalance, excessive
negative
or
zero
sequence.

It is expected that
disturbances that occur
near voltage zeroes will
have less effect that those
that occur near voltage
peaks.
But
all
disturbances are treated
equally with respect to
phase in the application
of the power acceptability
curves.
Deviations of frequency
are not accounted.

Events that are deemed

acceptable may not be
acceptable if they are
repeated close in time to
each other.
These events may be
reliability issues.
Neutral voltages may
present a hazard and
create
unacceptable
operating
conditions.
Similarly, neutral currents
may be unacceptable.
A single static power
acceptability curve may
not be able to model all
load types.

The difficulty with the foregoing analysis is that

there may be some loads for which negative and zero
sequence supply voltages will have some impact on
the load operation. Also, for transients (e.g., less than
a few cycles in duration) the symmetrical component
transformation is invalid. This is a consequence of
the fact that the symmetrical component

transformation is a phasor transformation, and as

such the sinusoidal steady state is assumed.

undervoltage event. The impact depends on how

much excess energy is delivered or how much was
not delivered. Many of the equipment installed in
industrial, commercial and residential loads are
sensitive to voltage sag events. During a voltage sag,
the voltage is below normal for some period of time
which reduces the power and energy delivered to load
by the system.

4. Voltage Sag Index Using Lost Energy

Different types of equipment behave in different
ways during voltage sag events. Some of the
equipment that are sensitive to voltage sags are:
adjustable speed drives (ASDs), programmable logic
controllers, starters, contactors, high pressure sodium
lamps, computers and microprocessors, digital
clocks, and certain other devices.

There is a need to develop a parameter to

indicate the severity of voltage sag events at a
location. CBEMA curve (and the recent ITIC curve)
is a magnitude-duration curve, indicating that
severity of a sag event is a function of both
magnitude and duration. CBEMA curve for the
undervoltage events is basically a constant energy
characteristic. The curve defines a threshold energy
level for a load, and if the energy delivered during the
sag event is less than the specified, it could impact
the load. For example, the CBEMA curve for the
voltage sag events can be defined by the constant
energy characteristic as,

Adjustable speed drives are the most critical

equipment in industrial processes. They may affect
the processes directly in some cases, but in some
cases they may affect the processes indirectly. For
example, stopping of chillers and then restarting with
time delay may affect the precise temperature control
required in a processing plant, affecting the quality of
the product produced. In some cases, as in case of an
extrusion plant, or wire drawing process, the process
may come to a complete stop if an ASD responds to a
sag by slowing down the process, or worse stopping
the process completely.

(V)3.14 t = 12.45

Voltage sag events may be qualified for the

purpose of calculating the index. At least in one case
[1], voltage sag events were qualified as those events
where at least one phase voltage goes below 75% of
nominal. The events with voltage below 10% are
considered as interruptions and counted for reliability
indices.

where V will be in per unit of nominal voltage and t

in milliseconds.
Some of the methods for calculating voltage sag
indices are reviewed here.
5. Review of Methods Proposed by Others
The Detroit Edison sag score method [1] is
probably the first used in a contract by a utility. The
sag score is defined as

In this paper, all sag events with voltages

between 85% and 10% are considered for sag index
calculation.

Sag Score = 1 Several methods have been proposed for

developing a composite index of all voltage sag
events during a defined period, say in a month or a
year. A simple average value of maximum voltage
sags experienced in the three phases is one method
[1]. An elaborate set of indices (as many as 20),
similar to the reliability indices used in the utility
industry is another possibility [2]. Even with many
indices for one location, the method in two does not
consider three-phase sag events. In this paper, it is
recognized that both voltage magnitude and duration
of the sag in all three phases should be considered to
derive a meaningful voltage sag index.

VA VB VC
3

VA, VB and VC are rms values in per unit. The

advantage of this method is simplicity. However, this
method does not consider the duration of the sag
which is an important parameter to indicate the
impact on loads. Some of the criteria in calculating
the annual index in this method are:
1.
2.
3.

Lost Energy During A Voltage Sag Event

An overvoltage or an undervoltage event at the
load terminals will have impact, because either
excess energy is delivered for an overvoltage event or
some energy is not delivered to the load for an

Voltage sag data is aggregated for 15-minute

interval at each location.
If one or two phases are greater than 1.0 per unit
(because of neutral shift), they will be reset to
1.0 p.u.
Sags will be qualified. A qualifying sag has at
least one phase equal to or below 0.75 p.u. That
is, the sags with minimum voltage above 0.75
per unit are not counted.

Other methods suggest a set of indices similar to

the SAIDI (System Average Duration Interruption
Index) used for reliability that considers
interruptions. One of them is proposed in a paper by
Sabin et al., [2]. The main disadvantage is that the
proposal requires about 20 indices to define voltage
severity of voltage sags at a location. Even with 20
indices, the method still does not consider threephase voltage sags. The method suggests considering
only the phase with deepest voltage sag.

In most cases, t1, t2 and t3 are equal.

This method is used for data obtained at three
monitor locations at two substations. There was one
monitor at substation A, and two monitors at
substation B (in bays 1 and 2). Data was collected
during the period May 98 through December 1998,
for a period of eight months. Data are shown in
Tables (3 through 5).
Table 3. Voltage sags at Substation A

6. Concepts based on Voltage Sag Lost Energy

Index (VLSEI)
In this paper, a method is proposed that takes
into consideration voltage sag in all three phases and
their duration. This is based on the energy that was
not delivered by the system to the load during a
voltage sag event.

V3 (per

Energy

0.89
0.85
0.87
0.85
0.94
0.92
0.96
0.73
0.96
0.63
0.82
0.96
0.96
0.81
0.87
0.97
0.84
0.86
9.73
0.83
0.67
0.87
0.75
0.82
0.88
0.78
0.81
0.96
0.72
0.98
0.93
0.97
0.89
0.98

0.72
0.85
0.67
0.89
0.75
0.70
0.85
0.72
0.72
0.62
0.87
0.81
0.82
0.74
0.92
0.86
0.64
0.67
0.29
0.82
0.71
0.79
0.76
0.74
0.59
0.64
0.82
0.82
0.70
1.0
0.96
0.84
0.81
1.0

0.90
0.86
0.84
0.73
0.71
0.68
0.86
0.76
0.71
0.65
0.66
0.78
0.75
0.96
0.80
0.84
0.57
0.85
0.70
0.93
0.66
0.82
0.75
0.77
0.58
0.98
0.82
0.71
0.70
0.82
0.38
0.98
0.93
0.81

1.99
6.91
2.38
3.33
2.75
5.11
0.32
13.82
3.19
129.01
4.64
6.83
7.87
20.03
2.09
0.35
34.39
2.98
76.07
2.48
85.07
1.82
37.06
8.71
21.08
4.91
0.96
1.68
63.99
4.59
18.52
1.54
0.33
0.08

(1 Vpu)3.14 * t

0.2
0.4
0.5

0.43
2.81
4.73

Duration

Where,
Vpu is the phase voltage in per unit of nominal voltage
during a sag event.
t = sag duration in milliseconds.
The power of voltage, 3.14 is derived using curve
fitting method to the CBEMA curve. The method of
least squares was applied to a log plot of the CBEMA
curve for this purpose. The exponent becomes
simply the mean square slope for such a plot.
Example: Calculated energy values for some
hypothetical voltage sags. Only one phase voltage is
used in the example.

0.8
0.6
0.5

V2 (per

1 Vpu

W = (1 Vpu)3.14 * t

Time
(milliseco
nds)
66.7
50
41.7

V1 (per

0.099
0.952
0.067
0.167
0.082
0.1
0.067
0.3
0.082
1.417
0.116
0.485
0.45
2.67
0.249
0.067
0.301
0.084
0.2
0.286
1.064
0.133
3.983
0.3
0.165
0.1
0.066
0.067
2.008
0.033
0.083
0.484
0.05
0.015

The lost energy in a sag event, W will be

calculated as,

Voltage
(per unit)

Duration

Total sag energy: 576.9

Number of sag events:
34
Average sag energy index: 16.97
Table 4. Voltage sags at Substation B, Bay 1
0.117
0.115
0.434
0.083
0.100
0.083
0.033
0.099
0.10
0.067

Number of qualified voltage sags: 3

Average Voltage Sag
(0.43+2.81+4.73) = 2.32

Energy

Index:

1/3

For three-phase calculation, lost energy for all three

phases will be added,
Energy Lost = (1 V1pu)3.14 * t1 + (1 V2pu)3.14 * t2
+ (1 V3pu)3.14 * t3

V1

(per

0.88
0.84
0.76
0.97
0.77
0.95
0.85
0.97
0.68
0.97

V2

(per

0.88
0.85
0.75
0.85
0.84
0.77
0.98
0.70
0.49
0.86

V3

(per

0.74
0.86
0.75
0.80
0.79
0.70
0.84
0.60
0.57
0.84

Total sag energy: 54.8

Number of sag events:
10

Energy
2.00
0.90
16.08
0.75
2.05
2.72
0.19
7.83
21.93
0.35

Average sag energy index: 5.48

Table 5. Voltage sags at Substation B, Bay 2
Duration

V1 (per

V2 (per

V3 (per

Energy

0.117
0.115
0.434
0.083
0.100
0.083
0.033
0.099
0.10
0.067

0.88
0.84
0.76
0.97
0.77
0.95
0.85
0.97
0.68
0.97

0.88
0.85
0.75
0.85
0.84
0.77
0.98
0.70
0.49
0.86

0.74
0.86
0.75
0.80
0.79
0.70
0.84
0.60
0.57
0.84

2.00
0.90
16.08
0.75
2.05
2.72
0.19
7.83
21.93
0.35

explained by applying it to data obtained from the

power quality monitors. The indexes reflect the
severity of the voltage sags at each location. It is
recommended that when power acceptability curves
are used to assess the power supply quality, for cases
of loads that are effectively AC to DC converters, that
the positive sequence supply voltage be used. A
method to calculate positive sequence component of
voltage from the three-phase time domain data is
being studied.
8. References
[1] A. Dettloff, D. Sabin, F. Goodman Power
Quality Performance as a component for Special
Manufacturing Contracts between Power Provider
and Customer Proceedings of the Power Systems
World 99 pp. 283-291

Total sag energy: 43.8

Number of sag events:
10
Average sag energy index: 4.38
Summary of the Indices for the Monitor Locations
The data in Table (6) is the summary of number of
events, total sag energy for the monitored period, and
the average sag energy.

[2] D.L. Brooks, R.C. Dugan, M. Waclawiak, A.

Sundaram Indices for Assessing Utility Distribution
System RMS Variation performance Paper presented
at IEEE PES Distribution Subcommittee Meeting,
Las Vegas, 2/22/2000, Las Vegas, NV

Table 6. Average sag energy index

Total
Number
Average
sag
of events
voltage sag
energy
energy index
A
576.9
34
16.97
B - bay1
54.8
10
5.48
B - bay2
43.8
10
4.38
Substation

[3] G.T. Heydt, G.G. Karady, B. Cummings, J. Tang

Improved Application of Power Acceptability
Curves and Their Application to Three-Phase Loads
Electric Machines and Power Systems, 1999, vol. 27,
pp. 737-751.

Substation A had 34 qualified voltage sag events

during the monitored period of eight months,
compared to the substation B which had only 10 sag
events. Not only substation A had three times more
events, but the average lost energy per event was
about three times the average energy for substation B.
Within substation B, both bays 1 and 2 had the same
number of qualifying events, but bay 1 had the higher
average lost energy index than for bay 2. This is
because, actual voltage sag magnitude for different
bays can be slightly different because of differences
in transformer impedance characteristics and their
loading level.

Rao Thallam is Senior Principal Engineer with Salt River

Project, a public water and power utility serving central
Arizona. Dr. Thallam worked for General Electric
Company for ten years prior to joining SRP in 1985. He
published about 50 papers in HVDC Transmission,
Harmonics, Power Quality and Surge Arresters. He is also
the author of a chapter in Electrical Engineering
Handbook published by CRC Press, and another in the
soon to be published Electric Power Engineering. He was
awarded a US patent in 1982, and IEEE Standards Board
medallion for outstanding achievement through standards
in 1996.
Gerald Thomas Heydt spent about 25 years with the
faculty at Purdue University where he was a Professor of
Electrical Engineering. He also has worked with the United
Nations Development Program in various positions around
the world. In 1994, Dr. Heydt took a position as Center
Director and Professor of Electrical Engineering at Arizona
State University, Tempe, Arizona. His interests are in
electric power quality, distribution engineering, and
applications of computers in power system analysis. He is
the author of two books, one on electric power quality, and
the other on computer analysis of power systems. He is a
registered professional engineer, Fellow of the IEEE, and a
member of the National Academy of Engineering.

7. Conclusions
Voltage sag events are a serious concern to
industrial and commercial customers, and there is a
need to develop indexes to indicate the severity of
events at a location. The index should be based on
sag magnitudes of all three phases, and time duration
of the event. An index based on lost energy during
sag events is developed. The energy lost during a
voltage sag event is a function of missing voltage and
the time duration of the sag event. The power of
voltage to calculate energy lost is based on the
CBEMA curve. This method based on lost energy is