1

Voltage sag propagation: Case study based on

measurements
Roberto Chouhy Leborgne, Member, IEEE, Jose Maria Carvalho Filho, Eder G. C. Novaes, and José
Policarpo G. Abreu

propagation of the voltage sags [6][7].

Abstract—This paper presents an analysis of the propagation However, the previous works are based on simulations and
of voltage sags caused by faults in a transmission grid. The generic grids. Here we intend to extend the validation of the
investigation is performed considering a set of measured voltage previous results to a real system assessment. We also intend to
sags obtained during a six month monitoring period. The analysis
apply the proposed sag propagation index to analyze the
includes vertical and horizontal propagation and its
quantification through a sag propagation index. The results show relation of this index and the network characteristics such as
how the propagation of the disturbance is affected by the network generation buses, transformer windings connections, and
topology, the transformers winding connection, and the load loads.
characteristics.
II. VOLTAGE SAG PROPAGATION
Index Terms—Power quality, voltage sag (dip), propagation.
Voltage sags are caused by faults, transformers
I. INTRODUCTION energization, large loads switching, and induction motors
starting. In general, the deep sags are caused by faults on the
T HE voltage sag is the most important disturbance of power
quality regarding the effect on sensitive industrial
processes. A voltage sag of 0.7 per-unit magnitude and 80 ms
transmission and distribution grid, whereas transformers
energization, large loads switching, and induction motors
starting are responsible for shallow sags [3].
duration can have the same effect of a 3 minutes interruption
Most of the sensitive loads are connected at low and
for some industrial customers [1]. Additionally, a feeder
medium voltage levels. They are affected by voltage sags
interruption affects only the customers connected to this
generated by faults at the nearby distribution level and by fault
feeder, whereas a voltage sag is able to cause problems to
at the transmission level. Hence, it is relevant to analyze the
customers connected a hundred kilometers away from the fault
propagation of the sags in the same voltage level (horizontal
location [2][3].
propagation) and the propagation through different voltage
A voltage sags can be characterized in terms of magnitude
levels (vertical propagation).
and duration. The magnitude can be defined as the minimum
Voltage sags caused by faults have been intuitively
rms voltage of the three phases. The duration can be defined as
classified in 7 types through the so-called ABC classification
the period of time during which at least one of the three phases
method [3]. Four of the sags types are presented in Table I.
experiences a rms voltage below a certain threshold.
This classification is a useful tool to differentiate events that
Additionally, three-phase unbalanced voltage sags are
seems to be equal when only the magnitude and duration is
characterized by the ABC classification that describes the
known.
layout of the three voltage phasors [3][4].
Previous investigations have addressed the issue of voltage TABLE I
sag propagation from a theoretical perspective. In one of these PARTIAL CLASSIFICATION OF THREE-PHASE VOLTAGE SAGS
works an index to assess the sag propagation in systems with Type B Type C
embedded generation and large induction machines was
proposed [5]. Other investigations analyzed in detail the
influence of the transformers winding connection on the

This work was supported in part by CAPES Grant BEX1640-02-0 Type E Type F
(Ministry o Education of Brazil).
R.C. Leborgne is with Chalmers University, Göteborg, SE-412 96 Sweden
(e-mail: roberto.leborgne@chalmers.se).
J.M. Carvalho Filho is with Itajuba Federal University, 37500-903 Brazil,
(e-mail: jmaria@unifei.edu.br).
E.G.C. Novaes is with Grupo Rede, Sao Paulo, Brazil (e-mail:
eder.novaes@gruporede.com.br)
J.P.G. de Abreu is with the Itajuba Federal University, 37500-903 Brazil,
(e-mail: polica@unifei.edu.br).
 2

The ABC classification can be seen as an intuitive way to propagation of the voltage sags [5]. One of the proposals is the
relate the fault with the generated sag. For example, a SLG use of a sag propagation index (SPI)
fault generates a voltage sag type B and a LL fault generates a N
voltage sag type C as shown in Table II. SPI = Bi (1)
In order to understand the vertical propagation of voltage %V N
Bt
sags it is necessary to analyze how the voltage sag type
where %V is a voltage threshold (50%, 70%, 90%), NBi is the
changes from the primary to the secondary side of a
number of buses experiencing voltage sag with magnitude less
transformer. When a voltage sag propagates through a
than %V, and NBt is the total number of monitored buses.
transformer, the relation of the phasors of the secondary side is
The SPI is similar to the well-known SARFI (System
not the same as the relation of the phasors of the primary side.
Average Rms variation Index)
For example, in a delta/wye transformer, a phasor at the
secondary side is the subtraction of two phasors of the primary N
SARFI = i
side. The consequences are that the zero sequence component (2)
X% N
is filtered and the positive and negative sequences are rotated T
by a multiple of +/-30 degrees respectively. where Ni is the number of customers experiencing sags with
Table II shows the voltage sag classification for different magnitude below X% and NT the total number of customers.
fault types and how the sag type changes when propagates However, the SARFIX represents the average number of
through a Dy transformer. For example, when a SLF fault specified voltage sags that occurred over the assessment period
occurs at bus I the sag seen at this bus is type B. However, per customer as it was proposed in [8], where the specified
after a Dy transformer, at bus II the sag seen is type C. After disturbances are those with a magnitude less than X%.
another Dy transformer, the sag seen at bus III is type D. It is A new interpretation of the SARFIX consists in the number
interesting to remark that the sag type alternates from C to D of voltage sags with magnitude less than X% in a given
type and from D to C type each time that the sag propagates location during a period of time (normally one year) [4]. This
trough a Dy transformer. A detailed analysis of the change in interpretation of SARFI gives it a quite different meaning
voltage sag type due to other transformer winding connections compared with the propagation index SPI.
can be found in the references [6][7].
III. MONITORED SYSTEM
TABLE II The system used in the case study is shown in Figure 1. The
THE CHANGE ON VOLTAGE SAG CLASSIFICATION THROUGH DELTA-WYE
TRANSFORMERS network contains 67 transmission lines (138 and 230 kV) with
a total length of 6619 km. There are 93 substations with a
Location / Voltage sag type
transformer-installed capacity of 2076 MVA. The generation
Fault
Bus I Bus II Bus III capacity is larger than the present demand. The excess of
type
generated power is exported to another regional grid through
LLL A A A the RND substation.
UHE
210MW SNP
SLG B C D 138kV
P3
RND SIN Y Y
230kV COX NOB 230kV ∆
P2
LL C D C P1
188 Km
P6 ∆ Y ∆ Y

LLG E F G 188 Km 104 Km 346 Km

P9 RND COX NOB ITA

138kV
In a meshed transmission grid, the horizontal propagation Y Y
Y Y
140 Km 215 Km

of a voltage sag is larger than its vertical propagation. Y Y Y Y

P5 UHE
However, since the sag is generated until it reaches a sensitive ∆
P7
∆
P4 84 MW

load it travels through the network on a combined process of Y ∆ P10

∆
horizontal and vertical propagation. Therefore, both the P8
Y

247 Km
IVM
254 Km
P11
horizontal and vertical propagation have to be addressed. 77 Km
P12
The analysis of the horizontal propagation is more complex Y ∆ UTE
150 MW
and requires computer simulation. It can be assumed that the UTE
230,0 kV
138,0 kV

voltage sag type is not changing. However, its magnitude and 480 MW 34,5 kV
13,8 kV

duration may be affected by the proximity of generation plants
and large induction machines that behave as generators at the
beginning of the sag. The topology of the network and the
impedance of the line also play an important role on the
voltage sag dumping as shown in section IV.
Few attempts have been done in order to quantify the
Figure 1. Diagram of the monitored power system.
A total of 12 buses located at 5 substations were selected
for voltage sag monitoring. The measured buses (P1…P12) are
indicated in Figure 1. The criteria for the bus choice included:
network topology, load concentration, sensitive-loads location,
 3

main generation plants, and transformer connections. 60%

The monitoring system consists in 12 PQ-monitors that 49%
50%
work with the same time-stamp due to GPS synchronization.
Once a day, all information is uploaded to a centralized server 40%

Frequency (%)
located at the utility control and operation room. The 30%
communication between the substations PQ-monitors and the 24%
20%
centralized server is established using the mobile telephone 13%
network. This communication was adopted because some of 10% 7%
4%
the substations were not remote-controlled or there were not 2%
0%
free communication channels available in the existing 20 - 100 101 - 200 201 - 300 301 - 500 501 - 1000 1001 - 3000
communication network. A second server located at the Sag duration (ms)
university was updated daily from the utility server. The
Figure 3. Histogram of voltage sag durations.
communication was established using standard internet
connection.
The measurements included: rms voltage, voltage phasors, B. SPI (Sag Propagation Index)
and sag duration at each phase. The time resolution of the rms In order to characterize the severity of the events
values and phasors were 8.33 ms (half cycle). The selected sag considering the propagation of the voltage sags the SPI was
threshold was 0.85 per unit in order to neglect the most estimated. The SPI was calculated considering the number of
shallows sags and not to overload the communication system. PQ-monitors that triggered, normalized by the total number of
installed PQ-monitors. The index for the 30 faults is shown in
IV. RESULTS Figure 4.
1.00
A. Voltage sag indices 0.90
After six months of measurements, 30 events were selected 0.80
for the propagation analysis. The 30 fault-events generated a 0.70
total of 89 voltage sags on the 12 monitored buses. The 0.60
SPI-85%

histogram and the cumulative distribution of the voltage sags 0.50

magnitudes are presented in Figure 2. There is a high 0.40

concentration of sags with magnitudes between 0.7 and 0.85 0.30

pu. However, 48% of the sags have a magnitude below 0.70 0.20
0.10
per unit. These sags may cause malfunction on many sensitive
0.00
electronic loads, such as ASDs and PLCs, as related in
previous studies [1].
1

11

13

15

17

19

21

23

25

27

29
Event

100% Figure 4. SPI-85% for the 30 analyzed events.

90% It’s important to mention that most of the faults do not
80% affect a great part of the system. In average, each fault triggers
70%
25% of the monitors (the average SPI is 0.25). However, some
Frequency (%)

60%
of the faults were considerable more severe. For example, one
50% 48%
40%
of the events triggered 11 PQ-monitors (SPI=0.91).
30% 27% 25%
C. Fault location
20% 13%
11%
10% 3% 7%
9% The analysis of the SPI permits the localization of the fault
0% positions that cause more damage to the system from the
<0.1 0.1-0.2 0.2-0.3 0.3-0.4 0.4-0.5 0.5-0.6 0.6-0.7 0.7-0.8 0.8-0.85 voltage sag point of view. For instance, it could be interesting
Sag magnitude (pu)
to investigate the events 10, 14, 24, and 30 that produced
Figure 2. Distribution of voltage sag magnitudes. voltage sags in at least half of the monitored buses, as shown
in Table III.
The histogram of the voltage sag durations is shown in The four events with the largest SPI (E10, E14, E24, and
Figure 3. Approximate 50% of the sags have a duration of less E30) are located in Figure 5. It is observed that faults located
than 100 ms. These sag durations agree with the protection near generation units produced voltage sags at a large number
setting used by the utility, where the fault-clearing time for the of buses. Here, three of the four highlighted events are located
first zone is about 100 ms. in lines that are connected to the bus COX, one of the main
generation buses.
 4

TABLE III approximate half of the SPI obtained for LLL faults.
CHARACTERISTICS OF THE EVENTS WITH LARGE SPI
It is interesting to see that there is a compensation between
Event # PQ-Monitors triggered Fault Characteristics the SPI value and the number of faults. For example, the
LL at radial line from number of SLG faults is about double the number of LLL
E10 P4/P5/P6/P7/P8/P9/P10 faults, whereas the SPI is about half. Therefore, for the
COX 138V
monitored period the number of monitored buses affected by
E14 P6/P7/P8/P9/P10/P11 LLG RND-COX 138 kV
sags caused by SLG faults and LLL faults were similar.
E24 P1/P2/P3/P4/P5/P6/P7/P8/P9/P10 LLL NOB-SIN 230 kV E. State of generation
E30 P1/P6/P7/P9/P10/P11 LL RND-COX 138 kV The generation unit dispatch also influences the sag
propagation. In the monitoring period the thermal-generation
was off during the occurrence of five events: E20, E22, E23,
UHE
210MW SNP
138kV
E24, and E25. When the average SPI is estimated considering
P3
RND
230kV COX
E10 NOB
SIN
230kV
Y
∆
Y
the generation dispatch a considerable difference is found, as
P2
E14
P1
shown in Table V.
188 Km
P6 ∆ Y ∆ Y TABLE V
188 Km 104 Km E24 346 Km AVERAGE SPI CALCULATED FOR EACH GENERATION STATE
P9 RND COX Generation state Average SPI
NOB ITA
138kV
Y Y
Y Y
140 Km 215 Km Thermal-generation ON 0.23

Y Y
Y Y Thermal-generation OFF 0.35
P5 ∆ UHE
∆ P4 84 MW
P7
E10
∆ P10
It is interesting to analyze more in detail the event E24
Y
Y ∆ IVM
P11 P8 247 Km 254 Km because it has caused sags at almost all monitored buses. The
77 Km
E14 E30
P12 fault was located on the 230 kV system that supplies the 138
Y ∆
UTE
150 MW 230,0 kV kV network when the thermal-generation is not dispatched.
Therefore, a fault in the 230 kV system during these
UTE 138,0 kV
480 MW 34,5 kV
13,8 kV
generation conditions was expected to cause voltage sags in
Figure 5. Location of the events with large SPI. The influence region for the
events E10 and E14 is represented by the circles. most of the 138 kV buses, as confirmed by the event E24.
F. Vertical and horizontal propagation
Figure 5 shows the region of influence for the events E10
Considering the meshed characteristic of the system, the
and E14. This is a very illustrative way to visualize the
horizontal propagation of sags is expected to be more relevant
propagation of the sags. We can clearly see that the region of
than the vertical one. Table VI shows the average SPI
influence of the event E10 is larger than the region of
considering the voltage level where the faults happen and the
influence of the event E14. The regions of influence that are
voltage level of the measured buses. The results agree with the
shown in Figure 5 are due to a sag threshold of 0.85 per unit.
predictions. The horizontal propagation of the faults at 230 kV
A lower sag threshold would define smaller regions of
reached a SPI of 0.5 whereas the vertical propagation to the
influence.
138 kV and 13.8 kV levels reached a SPI of 0.167. The
D. Fault type dumping for the vertical propagation can be explained by the
It is also relevant to investigate the variation of the SPI for combined effects of the transformers winding connection, the
the different types of faults that happened during the rotational loads, and the distributed generation.
monitoring period. Hence, the SPI is averaged for each type of
fault. Table IV shows the number of faults and the average SPI TABLE VI
SPI CALCULATED FOR VERTICAL AND HORIZONTAL PROPAGATION
for each type of fault.
Average SPI at

TABLE IV Faults at 13.8 kV 138 kV 230kV

AVERAGE SPI CALCULATED FOR EACH TYPE OF FAULT 138 kV 0.198 0.325 0.194
Type of fault 230 kV 0.167 0.167 0.500
SLG LL LLG LLL
# faults 13 8 3 6 The vertical and horizontal propagation of voltage sags can
SPI 0.17 0.28 0.28 0.35 also be studied through the detailed analysis of a single event,
for example the event E14. This event was a LLG fault at the
As expected, the voltage sags caused by LLL faults line RND-COX (138 kV). The magnitude and duration of the
propagates in average more than voltage sags caused by other measured sags generated by the event E14 are shown in Table
types of faults. On the other hand, the sags caused by SLG VII.
faults are the ones with the lowest propagation index,
 5

TABLE VII
VOLTAGE SAG CHARACTERISTICS FOR THE EVENT E14 TABLE VIII
VOLTAGE PHASORS RECORDED FOR THE EVENT E14 (VOLTAGES IN PU)
PQ-monitor Magnitude (pu) Duration (ms)
P6 – 230 kV – COX P9 – 230 kV – RND
P6 – 230 kV - COX 0.85 72 VA=0.86 VB=0.99 VC=0.85 VA=0.82 VB=1.0 VC=0.83
P7 – 138 kV – COX 0.81 82
P8 – 13.8 kV – COX 0.80 83
P9 – 230 kV – RND 0.82 65
P10 – 138 kV - RND 0.74 79
P11 – 13.8 kV - RND 0.72 86

The analysis of the vertical propagation is intended to

explain the changes on the magnitude, duration, and sag type
at different voltage levels. The event E14 occurred at 138 kV.
Therefore, the voltage sag magnitude at this level (P7 and P10) P7 – 138 kV – COX P10 – 138 kV – RND
is lower than the sag magnitude at the bulk 230 kV system (P6 VA=0.81 VB=0.97 VC=0.81 VA=0.74 VB=0.94 VC=0.74
and P9).
The duration of the sags increases at lower voltage levels as
a consequence of the post-fault voltage sag caused by the high
currents due to induction motors re-starting. For example, the
sag duration at P10 (RND 138 kV) is 79 ms whereas the sag
duration at P11 (RND 13.8 kV) is 86 ms.
The sag type can be extracted from the phasor diagrams,
shown in Table VIII. The sags were generated by a LLG fault
between phases A and C, hence the sag is type E with phase B
being the symmetrical phase, at the buses were the faulted line
is connected (P7 ad P10). The vertical propagation from 138 P8 – 13.8 kV – COX P11 – 13.8 kV – RND
kV to 230 kV should not affect the sag type due to the winding VA=0.94 VB=0.94 VC=0.80 VA=0.89 VB=0.90 VC=0.72
connection of the 230/138 kV transformers. These
transformers are Yy grounded at both sides. The phasor
diagram at P6 and P9 show the same type of sag observed at
P7 and P10.
On the other hand, the vertical propagation to 13.8 kV
changes the sag type as a consequence of the transformer
winding connections. The 138/13.8 kV transformers are Dy,
therefore the sag type changes from type E at P7 and P10 to
type F with phase C being the symmetrical phase at P8 and
P11, respectively.
Additionally, the sag type changes during the propagation
to the distribution level due to the combined effect of the V. CONCLUSIONS
induction machines and the pre-fault system unbalance. This paper presented a detailed analysis of voltage sag
Induction machines behave as a generator during the sag propagation on a transmission network based on measurements
keeping the voltage magnitude higher and affecting the sag results. The 30 analyzed events generated 89 voltage sags on
type. The voltage unbalance affects the sag classification the 12 monitored buses. It was noticed that 48 % of the sags
because this characterization method was proposed for a pre- were deep enough (V<0.70 pu) to cause malfunction on
fault balanced system. critical sensitive loads. The duration of the sags agreed with
For the horizontal propagation the sag magnitude at a the typical fault clearing time of the distance protection in the
certain voltage level is evaluated. For the analyzed event the analyzed system. About 50 % of the sags had a duration lower
voltage sag magnitude at 138 kV is higher at COX than at than 100 ms.
RND. One of the explanations is the proximity of the The sag propagation index (SPI) was calculated for the 30
generation units to the bus COX. This phenomenon is repeated faults considering a total of 12 monitored buses. The average
at the 13.8 kV and 230 kV levels when the horizontal value of SPI was 0.25; meaning that in average each fault
propagation is evaluated. The concentration of generation near generated a voltage sag at 3 of the monitored buses. However,
COX keeps the voltage at this substation at higher level than at some faults were very critical affecting a great part of the grid
RND substation. (SPI=0.91). These events were identified and their influence
region was delimited. It was observed that these faults are

located near the main generation bus.
It was concluded that LLL faults propagates about two
times more than the SLG faults. However, the number of SLG
faults was about two times the number of LLL faults.
Consequently, LLL and SLG faults contributed similarly to the
number of measured sags.
The generation unit dispatch affects the average SPI. When
the system is dispatching the thermal units that are located near
the main loaded buses within the grid the SPI was 0.23,
whereas when the energy comes from the hydro plants that are
located far from the load centers the SPI increased to 0.35.
The vertical sag propagation through transformers affects
the voltage sag type as a consequence of the transformer
winding connections confirming the former theoretical
analysis. However, the changes in the sag type are also
influenced by the proximity of rotational loads and the pre-
fault unbalance level of the system.
The horizontal propagation of the faults at 230 kV had a
larger SPI than the vertical propagation to the 138 kV and 13.8
kV levels due to the meshed characteristic of the grid. The
dumping on the vertical propagation can be explained by the
effects of the transformers, the rotational loads, and the
distributed generation.
VI. REFERENCES
[1] R.C. Leborgne, J.M. Carvalho Filho, J.P.G. Abreu, T.C. Oliveira, A.A.
Postal, L.H. Zaparolli, “Alternative Methodology for Characterization of
Industrial Process Sensitivity to Voltage Sags”, Proceedings IEEE-PES
Power Tech, Bologna, Italy, June 2003.
[2] L. Conrad, K. Little, C. Grigg, “Predicting and preventing problems
associated with remote fault-clearing voltage dips”, IEEE Transactions
on Industry Applications, Vol.27, No.1, 1991.
[3] M.H.J. Bollen, “Understanding Power Quality Problems – Voltage Dips
and Interruptions”, New York: IEEE Press, 1999.
[4] IEEE Voltage Dips Indices, IEEE Working Group P1564, August 2003.
[5] R. Gnativ and J.V. Milanovic, “Voltage sag propagation in systems with
embedded generation and induction motors”, Proceedings Power
Engineering Society Summer Meeting, 2001.
[6] L. Zhang, “Three-phase unbalance of voltage dips”, Licentiate thesis,
Chalmers University of Technology, Sweden, 1999.
[7] M.T. Aung and J.V. Milanovic, “The influence of transformer winding
connections on the propagation of voltage sags”, IEEE Transactions on
Power Delivery, Vol.21, No.1, Jan. 2006.
[8] D.L. Brooks, R.C. Dugan, M. Waclawiak, and A. Sundaram, “Indices
for assessing utility distribution system RMS variation performance”,
IEEE Transactions on Power Delivery, Vol.13, No.1, Jan. 1998.
6
VII. BIOGRAPHIES
Roberto Chouhy Leborgne (M’2001) received his Electrical Engineering
Degree and M.Sc. in Electric Power Engineering from the Itajuba Federal
University, Brazil, in 1998 and 2003 respectively. He received his Licentiate
degree from Chalmers University of Technology (Sweden) in 2005.
His employment experience includes ABB-Daimler Benz Transportation
Brazil and Teyma Abengoa Uruguay. He is currently with the Department of
Electric Power Engineering at Chalmers University of Technology. His fields
of interest include power quality and power system planning and operation.
José Maria Carvalho Filho received his M.Sc. and D.Sc. degree in
Electrical Engineering from the Itajuba Federal University, Brazil, in 1996
and 2000, respectively.
At present he is Associate Professor at Itajuba Federal University and a
Power Quality Study Group Member. His fields of interest include voltage
sags and other Power Quality issues. He is also a specialized consultant in
industrial planning.
Eder G. C. Novaes received his Electrical Engineering Degree from the
Universidade Federal de Itajuba, Brazil, 2003.
He is currently with the Grupo Rede, Sao Paulo, Brazil, and he is working
towards his master degree at the Itajuba Federal University. His fields of
interest include voltage sags.
José Policarpo G. de Abreu was born on Madeira Island, Portugal, in 1952.
He received the B.S.E.E. and M.Sc. degrees from the Itajubá Federal
University, Itajubá, Brazil, and the D.Sc. degree in electrical engineering from
the University of Campinas, Brazil. From March 1999 to June 2000 he was
on leave for a Post-Doctoral program at Worcester Polytechnic Institute,
Worcester, MA-USA.
He is a Full Professor at the Itajubá Federal University, where he also
serves as the Power Quality Study Group Coordinator. His research interests
include Power Quality issues, such as Power Definitions, Harmonics,
Imbalance, and Voltage Sags. Induction Motors, Transformers, and Converter
Transformers are other of his interests.