APPLICATION
Evaluation of a new 3-phase dip
definition
by M Schilder and RG Koch, Eskom
Experience has shown that a jump in the phase angle of the voltages at the start of a dip can cause some line-commutated variable-speed
drives to trip, even if the dip magnitude is relatively small.
This is a nuisence in mining and other
industries and has led to the proposal for a
new dip definition. Most dip definitions are
based on the duration and magnitude of the
dip event only (e.g. as specified in IEC-610004-30). A new definition has been proposed
that considers the unbalance characteristics
during a dip and the phase angle jump on
initiation of the dip [1].
This paper evaluates the application of the
two proposed algorithms in the classification
of dips measured in South Africa. Although the
new definition provides a better assessment
of the severity of the dip based on both
voltage magnitudes and phase angles, it
was found that several types of dips are not
classified correctly. This is particularly true for
the case of measurements where the phase
angle of one of the phases is set to a fixed
reference (i.e. 0°). In this case the respective
algorithms classify only 50% to 80% of
measured dips correctly.
The new definition provides better insight
into complex dip characteristics, but further
work is required on the classification algorithms
as well as specifications for instruments
to measure dips when applying this
new definition.
In contrast with the current
Drop in Characteristic
6-phase:
Type
T
PN factor
dip definition that only
phases
voltage
lowest phase
uses voltage magnitude
Ca
0
bc
V=V1-V2
F=V1+V2
VBC
(disregarding phase
Dc
1
c
VC’
V=V1+αV2
F=V1-αV2
angle changes) and dip
Cb
2
ac
VCA
V=V1-α²V2
F=V1+α²V2
duration, the proposed
Da
3
a
V=V1+V2
F=V1-V2
VA’
dip definition attempts
Cc
4
ab
VAB
V=V1-αV2
F=V1+αV2
better classification of
unbalanced three-phase
Db
5
b
VB’
V=V1+α²V2
F=V1-α²V2
voltage dips in terms
Table 1: Dip analysis characteristics.
of both the voltage
magnitude and the
phase angle changes
on the difference between the positive and
that take place. It is based on a sound
negative sequence remaining voltages.
theoretical analysis deriving from simplified
Depending on the phases that are involved
in the fault, the negative sequence voltage
symmetrical components using a voltage
will be multiplied by a factor αx as shown in
divider model [1, 2, 3]. According to the
Table 1 where:
proposed method, a dip is classified by a
characteristic voltage and a PN factor and
attempts to classify the dip in one of two
main categories, C or D. The dip is further
identified by a subscript that indicates the
symmetrical phase, i.e. the least dipped
phase for two-phase (C-type) dips, or the
most dipped phase for single-phase (D-type)
dips, as shown in Fig. 1.
The main feature is the characteristic voltage
or equivalent dip magnitude that is based
α = e j120° and x=0, 1, 2
A few examples of measured dips are shown
in Fig. 2. It is immediately clear that all dips do
not fit into the definitions of Fig. 1; either the
relative phase angles differ from the definition
or the magnitudes of all three phases are
different. The classifications shown in Fig. 2 are
based on the relative magnitudes of the three
phase voltages only. However, the severity of
the dip (characteristic voltage) will depend
on the relative phase
angles between the
phases.
Fig. 1: D (single-phase) and C (two-phase) dip types.
Fig. 2: Example vector diagrams of practically measured dips from Eskom dip database, which
do not directly fit the definitions shown in Fig. 1.
energize - June 2006 - Page 64
Two algorithms were
previously proposed
to classify the dips
according to the new
definition into either
C - o r D -t y p e d i p s
[1, 2]. This paper reports
on the results of a study
that was undertaken
to evaluate whether
the new definition
is appropriate for
use within the South
African context – and
if so, which algorithm
should be used for
the classification. The
evaluation comprised
APPLICATION
Description
# dips
T=Tsym
T=T6
Tsym=T6
T=Tsym
T=T6
Transmission
520
222
459
220
42,69%
88,27%
Distribution
913
352
846
365
38,55%
92,66%
Table 2: Dip classification results for dips during 2003.
several steps:
Various dip types were simulated using Matlab, from this, the types and
numbers of dips that may be classified incorrectly were identified, i.e.
where the classified dip type does not correspond to the definition of
Fig. 1 based on the phase magnitudes only, these results were then
compared to measured dip data from the Eskom National Power
Quality Database to determine how many of these poorly-classified
dips occur in reality.
Testing of classification algorithms
Fig. 3: Characteristic voltage and PN factor for a single-phase dip, as
calculated by the symmetrical components method.
The following steps were implemented to test the success of the two
algorithms in classifying simulated and practical dips into the C and D
classes shown in Fig.1:
Six-phase algorithm
In this case, six voltages are calculated, namely the three line-to-line
voltages as well as the three modified phase voltages. These voltage
vectors exclude the zero-sequence component, which is first subtracted.
(This is because the algorithms aim at providing an assessment of dip
impacts on customer plant – and zero-sequence voltage components
do not propagate far in most power systems).
The characteristic voltage is, according to this algorithm, defined as
the magnitude of the lowest of the six voltages, while the phase with
the largest magnitude of the six voltages is the PN factor. The dip type
is determined from Table 1 (last column) based on the lowest of the
six-phase voltages.
Symmetrical component algorithm
In this case, the positive (V1) and negative (V2) sequence voltages are
calculated using Fortesque’s decomposition matrix.
Fig. 4: Dip type (six-phase method) for a SLGF (‘A’-phase) when
varying both magnitude and phase jump (Dip type = 3).
The dip type is classified according to the factor T given in Table 1, which
is calculated by Eqn. 1. (It may be noted that the addition of the 20o
before evaluating the argument in Eqn. 1 was determined empirically
for practical cases [2]).
(1)
The characteristic voltage (V) and PN factor (F) can be found by
using Eqn. 2:
(2)
Reference dip classification
The above classification methods for C and D type events require voltage
vector measurements. A recent paper [4] discusses classification of
voltage dips when only RMS values are available. The dips are classified
into the different dip types and an estimate of the voltage vectors is then
made. A seven-dip type definition is used in this paper [4].
Fig. 5: Dip type (symmetrical component method) for a SLGF
(‘A’-phase) when varying both magnitude and phase jump
(‘A’-phase phase referenced to 0°).
In order to study the two algorithms proposed by Bollen [1, 2, 3], a similar
technique to that in [4] was implemented to determine the dip type
based only on the magnitudes of the phase voltages. Only C and D
type dips, as shown in Fig. 1, were studied. The most difficult dip types
to classify are those where the difference between the maximum and
middle, and between the middle and minimum voltage magnitudes
energize - June 2006 - Page 66
APPLICATION
an indication of the severity
of a dip, based on either
the voltage drop or the
phase jump of the faulted
phase. The characteristic
voltage is constant at
0,33 pu when the
faulted phase drops to
0 V (horizontal cur ve),
correlating with practical
results where it was found
that a SLGF should not
drop the voltage on
the customer side of a
delta-wye or wye-delta
Fig. 6: Dip type (six-phase method) for a LLF (‘CA’-phases) when
transformer to less than
varying both magnitude and phase jump (Dip type = 2).
0,3 pu [5,6]. It also shows
are the same. In classifying simulated and
that a 180° phase shift of
measured data with these characteristics; if
the faulted phase will result in classification as
the voltage on two of the phases was above
a very severe dip event, even if the voltage
0,9 pu (classical dip threshold), the dip was
magnitude does not drop significantly. Such
classified as a single-phase event (D-type)
an indication of severity, based on either a
otherwise it was considered a two-phase
drop in magnitude or a phase jump during
event (C-type).
the dip, may be advantageous to describe
dips for plant sensitive to such phase angle
Theoretical comparisons
jumps, once the impact of phase jump on
Bollen provides proof, using a set of synthetic
these sensitive items has been quantified.
dip events, to explain that a perfect algorithm
Phase shifts measured by Eskom’s power
does not exist to classify dip characteristics [2].
quality monitors range up to 127°. Dip
Since the purpose of this project is to test the
simulations using Digsilent show that phase
two proposed characterisations for local dips,
jumps of up to 180° can be expected for
various dip types were generated in Matlab
single-phase dips and up to 120° for twoto test which dips will be classified incorrectly,
phase dips for any of the phase voltages,
before analysing measured dips.
when the ‘A’-phase angle is normalised to 0°.
An example of the calculated characteristic
The phase angle normalisation is required for
voltage (using the symmetrical component
comparison with the measurements, since
method) for a single-phase dip (Da) is shown
the instruments that were used normalise
in Fig. 3. Each curve represents a constant
the ‘A’-phase angle to 0° when assessing
remaining voltage for the dipped phase
the event.
for various phase angle jumps, shown on
Simulated dips
the abscissa.
and 7, where the symmetrical component
Generic dips were simulated, varying both
the magnitudes and phase angles of the
three phase voltages symmetrically, i.e.
only one phase is varied or
the two phases are varied
equally and symmetrically
around the healthy phase.
This method may include
dips not generally measured
in practice, but provides
a good overview of the
types of dips that may be
classified incorrectly.
the magnitude and phase values (in contrast
Graphs showing the dip type
vs. phase magnitude and
angle for Da and Cb dip
types are shown in Figs. 4
to 7. The impact of rotating
the dip for a 0° ‘A’-phase
reference is clear in Figs. 5
phase voltage magnitudes, while “sym” and
The remaining voltage curves start at 1 pu,
varied in steps of 0,125 pu to 0 pu. The graph
shows that the characteristic voltage provides
Fig. 7: Dip type (symmetrical component method) for a LLF (‘CA’phases) when varying both magnitude and phase jump (‘A’-phase
phase referenced to 0°).
energize - June 2006 - Page 68
method changes from all practical dips
classified correctly, to very few dips classified
correctly. The graph for dip type Cc is similar
to the Cb graph.
The inaccuracy of the symmetrical
component method when rotating the
measured dips to the 0° ‘A’-phase reference,
is due to the denominator in Eqn. 1, namely
the vector (1-V1). The positive and negative
sequence components’ arguments vary
symmetrically when the reference phase
angle is varied, but the vector (1-V1) behaves
completely differently – resulting in incorrect
dip type classification. Note that if the vector
(1-V1) is replaced by the positive sequence
vector, the results are similar to that of the
six-phase algorithm and the reference phase
angle becomes unimportant.
In theory, more than 65% of dips should
be classified correctly by the symmetrical
component method (based on the typical
distributions of SLGF and LLF). In practice,
phase shift on the ‘A’-phase is usually
expected, even for Db, Dc and Ca dip types,
resulting in many dips classified incorrectly.
Some of the expected problems with
the proposed algorithms have now been
highlighted. However, it is clear that these may
be useful, when looking at the characteristic
voltage for the different types of dips (as was
shown in Fig. 3).
Eskom national power quality database
analysis
The instruments used to monitor power quality
in transmission and distribution substations
calculate and store vector events once every
cycle. A Fourier transform is used to calculate
to half-cycle RMS values specified for dips in
IEC 61000-4-30). A fairly slow phase-lockedloop is implemented to synchronise the
measurements, which may lead to further
errors in the vector event data, especially if
the reference phase voltage drops to zero.
Various dip events measured on the
transmission and distribution systems were
analysed using the two algorithms. The results
for dip type classification are presented in
Table 2. No subscript to the T indicates the
reference type classification based on the
“6” subscripts indicate the result from the
symmetrical component or six-phase method
respectively. Note that all balanced dips
(type A) are classified correctly by both
algorithms and were therefore excluded from
the results shown in Table 2.
APPLICATION
As expected, the symmetrical component
algorithm yields very poor results, since most
dips experience some phase shift on the
reference ‘A’-phase. However, the results from
the six-phase algorithm, ranging from 88%
to 93%, are also unacceptable for practical
purposes. An attempt to compensate for the
loss of the ‘A’-phase angle information resulted
in the symmetrical component method
accuracy increasing to approximately 55%,
which is still unacceptable.
Conclusions and recommendations
The new dip definition indicates the severity of
a dip in terms of both the dip magnitude and
the phase angle jump. However, this severity
still needs to be linked to the sensitivity of motor
drives and other equipment. From the studies
based on simulated and measured dips,
various shortcomings of the new classification
algorithms and/or measurement methods
were found. These are, the known impact of
the phase jump during a dip on the accuracy
of the six-phase algorithm was confirmed.
Practical dips seem to be classified correctly
only 88% to 93% of the time. It was also found
that the symmetrical component method is
dependent on the true phase angle of the
References
reference voltage, i.e. the dip should not be
[1]
Bollen, M.H.J., Zhang, L.D., “Different
methods for classification of three-phase
unbalanced voltage dips due to faults”,
CIGRE, 10 pages.
[2]
Bollen, M.H.J., “Algorithms for characterizing
measured three-phase unbalanced
voltage dips”, IEEE Transactions on Power
Delivery, Vol. 18, No. 3, July 2003, pp. 937944.
[3]
Anderson, P.M., “Analysis of Faulted
Systems”, IEEE Press Power Systems
Engineering Series, 1995, ISBN 0-78031145-0.
[4]
Bollen, M.H.J., Goossens, P., Robert,
A., “Assessment of voltage dips in HVnetworks: Deduction of complex voltages
from the measured RMS voltages”, IEEE
Transactions on Power Delivery, Vol. 19,
No. 2, April 2004, pp. 783-790.
[5]
Zhang, L., Bollen, M.H.J., “Characteristic of
Voltage Dips (Sags) in Power Systems”, IEEE
Transactions on Power Delivery, Vol. 15, No.
2, April 2000, pp. 827-832.
[6]
cGranaghan, M.F., Mueller, D.R., Samotyj,
M.J., “Voltage sags in industrial systems”,
IEEE Transactions on Industry Applications,
rotated to have the reference phase at 0o
Particular attention in further work should
concentrate on the reference angle
dependency of the symmetrical components
method needs to be investigated further.
Due to the reservations about dip voltage
vector measurements, future specifications
for measurement of dip vectors should be
investigated and proposed. Equipment
sensitive to phase jumps during dips should
be tested in order to determine the real
impact of the phase jump.
Acknowledgements
The authors would like to thank Eskom
Distribution and Transmission for use of
data in the dip database, as well as
Mr. R Ragoonanthun for assistance in
extracting the required dip information.
This paper was first presented at Cigré’s
5th Southern Africa Regional Conference
in October 2005 and is reproduced
with permission.
energize - June 2006 - Page 69
Vol. 29, No. 2, April 1993, pp. 397-403.
Contact Melanie Schilder, Eskom,
melanie.schilder@eskom.co.za 