© 2001 IMAGE STATE

An improved method for

arc-flash hazard analysis

BY R. WILKINS, HE 2002 EDITION OF THE NEC REQUIRES EQUIPMENT, ON

M. ALLISON,
& M. LANG
T which work may be required to be done when energized, to be labeled,
IEEE INDUSTRY APPLICATIONS MAGAZINE • MAY|JUNE 2005 • WWW.IEEE.ORG/IAS

warning of arc-flash hazard [1]–[3]. The 2004 edition of NFPA 70E

requires a flash-hazard analysis to be done before a person can work near to
energized equipment and to determine the type of protective clothing needed [2].
There are several different methods in use at present to calculate the flash-boundary dis-
tance and incident energy upon a worker [2], [4]–[7], and IEEE 1584 contains formulas
based on a statistical fit to test data obtained in several high-power test laboratories in North
America [8].
In this article, an improved method that uses time-domain analysis is presented. It can be
used as an arc-flash calculator, but it also allows current limitation by fuses and other effects
to be studied.
Figure 1 shows a schematic representation of the test set-up, which was used in most of
the IEEE tests to model an arc-flash hazard incident, the arcing being initiated by fine trig-
ger fuse wires. High-current arcs that are not restricted move, due to magnetic forces, away
from the source of supply. For the geometry shown, this causes the arcs to be driven down-
wards and burn from the electrode (busbar) tips. However the behavior of the three-phase,
free-burning arcing fault in equipment is chaotic, involving rapid and irregular changes in
arc geometry due to thermal buoyancy (convection) and electromagnetic forces, arc extinc-
tion, plasma jets, sudden shortening due to restriking and reconnection across electrodes or
plasma parts, and many other effects.
These effects have been vividly illustrated by Stokes and Sweeting [9] using high-speed
40 video photography. For tests in the open, they showed the formation of an

1077-2618/05/$20.00©2005 IEEE

Authorized licensed use limited to: UNIVERSITY OF SOUTHAMPTON. Downloaded on October 14,2021 at 08:43:09 UTC from IEEE Xplore. Restrictions apply.
expanding plasma cloud, fed by jets from the electrodes, values. Other anomalies occur at higher voltages. For
and forced away from the electrodes (downwards in the example, if V > 0.783 kV and g = 32 mm, the arcing
case of Figure 1). current exceeds IBF at IBF = 100 kA.
The calorimeters shown in Figure 1, which represents For higher voltage systems the IEEE 1584 equation is
the IEEE tests, are not directly located in the path of the
plasma cloud and receive heat energy from the arcing zone log10 Iarc = 0.00402 + 0.983 log10 IBF (2)
principally in the form of radiation.
The time-domain model has been developed as a
research tool, and it gives a better representation of the arc- This gives arcing currents higher than the bolted-fault
flash phenomena than the current IEEE 1584 equations. It current for IBF < 1.724 kA.
can also contribute to future revisions of IEEE 1584, which The second stage of the IEEE 1584 method requires
are likely to be based on physical models. the calculation of a normalized incident-energy density En
using
The IEEE 1584 Formulas
There are two principal stages in arc-flash calculations:
a) calculation of the rms arcing current Iarc so that the log10 En = K1 + K2 + 1.081 log10 Iarc + 0.0011 g,
operating time of protective devices can be found (3)
b) calculation of the incident-energy density E at a dis-
tance d so that a safe working distance or the where
required personal protective equipment can be K1 = −0.792 or −0.555 (open or box configuration)
determined. K2 = 0 or −0.113 (grounded or ungrounded system).
In IEEE 1548, the following equation is given for This is then adjusted to the actual fault duration (lin-
the calculation of Iarc (originally for system voltages early) and for the distance d using a power-law with a “dis-
under 1 kV). tance exponent” X, which depends on the equipment
type. This procedure can also give anomalous results. The
test results with the electrodes arranged as in Figure 1
log10 Iarc = KA + 0.662 log10 IBF + 0.0966 V show that the incident energy is significantly higher for
+ 0.000526 g + 0.5588Vlog10 IBF tests in a box, because of the “focusing” effect, but use of a
distance exponent can produce the opposite result, as
− 0.00304 g log10 IBF , (1) shown in Figure 3.

IEEE INDUSTRY APPLICATIONS MAGAZINE • MAY|JUNE 2005 • WWW.IEEE.ORG/IAS

where Location of
KA =−0.153 or −0.097 (open or box configura- IARC Worker
tion) Vac g (or Calorimeter in
IBF = bolted three-phase symmetrical fault current, Laboratory Test)
kA
V = system voltage, kV Three-Phase
Power
g = gap between arcing electrodes in millimeters. System
In order to obtain a good fit to the test data, (1) was
derived using a least-squares method. However, the
grouping of the variables on the right-hand-side of (1) is d0 d
not based on physical phenomena and can produce anom- 1
alous results. Arc-flash in an open box.
The resistance of the arcing fault produces an arcing
current that must always be lower than the bolted-fault 1.4
current. Furthermore, if the arcing gap distance is
increased, the resistance increases (though by a 1.2
Arcing Current, Per Unit

relatively small amount [9]), and the arcing current

1
should fall.
Figure 2 shows the ratio of arcing current to bolted- 0.8
fault current for the “box” case, with V = 0.48 kV, and
arcing gaps from 32–152 mm. The ratio should always be 0.6 32
less than 1.0, but Figure 2 shows that it can exceed 1.0 for
0.4
low bolted-fault currents.
Furthermore, for bolted-fault currents less than 0.2
1.489 kA, irrespective of system voltage, the effect of the 152
gap length is reversed (incorrectly), giving higher arcing 0
0.1 1 10 100
currents for longer gaps. Although at 0.48 kV the situa- Bolted-Fault Current (kA)
tion improves for bolted-fault currents above 1.489 kA, 2
41
the effect of the anomaly is still significant for much higher Iarc /IBF calculated from (1).

Authorized licensed use limited to: UNIVERSITY OF SOUTHAMPTON. Downloaded on October 14,2021 at 08:43:09 UTC from IEEE Xplore. Restrictions apply.
 The curves in Figure 3 were calculated using the The circuit model in Figure 4 includes a set of three
“switchgear” distance exponents given in [8] 1.473 for current-limiting fuses in series with the arcing fault. The
low voltage (LV) and 0.973 for high voltage (HV), and arcing fault is initially modeled as a set of fine trigger
the results shown are independent of the actual voltage, fuse wires with a fixed melting I 2 t, and then the subse-
gap length and bolted-fault current. Using the LV equa- quent three-phase arcs are modeled as a Y-connected set of
tions the “open” incident-energy density exceeds the “box” nonlinear resistances. The transient circuit current can
value if d is less than 166 mm, while for the HV equa- then be found by numerical solution of the circuit differ-
tions, this occurs at 358 mm, but the effect of the anom- ential equations:
aly remains significant for higher values of d.
The reason for the anomaly lies in the use of different diA e A − RiA − vfuse1 − varc1 − v P
distance exponents. The normalized incident energy =
dt L
En (box) is always higher than En (open) at the standard diB e B − RiB − vfuse2 − varc2 − v P
distance of 610 mm [8], but when this is corrected for =
lower values of d, the situation can be reversed because dt L
the correction curve for En (open) with X = 2 rises more diC e C − RiC − vfuse3 − varc3 − v P
= . (4)
steeply than for En (box). dt L

Time-Domain Model If the fictitious star point is grounded v P = 0 and the

The anomaly in the calculation of Iarc can be avoided by computed phase currents do not interact. In this case com-
the use of a time-domain model such as that shown in putation continues until all fuses have cleared or until a
Figure 4. The circuit parameters are derived from the preset time is reached, corresponding to the opening of a
system data (voltage, bolted-fault current, frequency, the backup breaker. However, an ungrounded model is
X/R ratio and closing angle). more realistic. For this case the sum of the phase currents
Although the fault arc behavior is difficult to model, is zero, which enables the instantaneous potential v P to be
the behavior of the electrical circuit can be computed pre- calculated as follows.
cisely, reducing the area of uncertainty to that of the fault a) if no fuses have cleared:
arc model. Whatever fault arc model is used, the calculated v P = −(vfuse1 + vfuse2 + vfuse3
arcing current will always be lower than the bolted-fault
current with a time-domain model of this type. + varc1 + varc2 + varc3 )/3

b) after the first fuse has opened, say in phase a:

IEEE INDUSTRY APPLICATIONS MAGAZINE • MAY|JUNE 2005 • WWW.IEEE.ORG/IAS

10.0
v P = (e B + e C − vfuse2
− vfuse3 − varc2 − varc3 )/2.
> 1kV
E(Open)/E(Box)

Cyclically similar expressions may be written if phase b

or phase c clears first [11]. If fuses are not used, the fuse
1.0 voltages are all set to zero.
LV
Fault Arc Characteristics
The single-phase, high-current arc in air has a rising V-I
characteristic which can be represented as
0.1 Varc = VE + kIarc X Y
g (5)
100 1,000
Distance to Calorimeters (mm) Measurements by Fisher [12] using currents up to
3
Incident-energy ratio using IEEE 1584 equations.
41.6 kA and arcing gaps g from 25–100 mm found that
X ≈ 0.15 and Y ≈ 0.5. Ignatko [13] studied arcs from
5–150 kA with gaps from 5–200 mm. He measured
the electrode-fall voltage (VE ) with Langmuir probes
Vfuse1 Varc1
(23.5V for copper electrodes), and the actual arc length
R L (which is greater than the gap distance) was measured
iA
photographically, to obtain the column gradient.
eA Ignatko’s data also fits the form of (5), with similar X
VP
and Y to Fisher’s.
iB Stokes and Oppenlander [14] found X ≈ 0.12 and
Y ≈ 1.0 for horizontal and vertical gaps of 5–500 mm
iC
with currents up to 20 kA. Their photographs revealed the
complex variations in arc geometry in detail. Paukert [15]
4 reviewed data from seven different laboratories and found
42
Circuit model. approximate average values of X ≈ 0.2 and Y ≈ 0.47.

Authorized licensed use limited to: UNIVERSITY OF SOUTHAMPTON. Downloaded on October 14,2021 at 08:43:09 UTC from IEEE Xplore. Restrictions apply.
 Given the very variable nature of the fault arc, the
data in the literature shows a remarkable agreement. 40
The arc voltage shows a weakly rising dependence on 30
current, with X ≈ 0.12 − 0.2. In some cases it is not
20
clear whether published data refers to instantaneous cur-

Current (kA)
rent or true rms current, but the trend is the same. The 10
dependence on g is more variable, probably as a result of 0
the use of differing electrode geometries. For the three- –10
phase case with horizontal electrodes, Stokes and Sweet- –20
ing [9] found X ≈ 0.12 and a weak dependence upon
–30
gap distance.
As a first step (and as originally suggested by Fisher) –40
0 0.01 0.02 0.03 0.04 0.05
the three-phase case can be represented as three separate Y-
connected single-phase arcs (see Figure 4), each of which Time (s)
5
can be modeled by an equation of the same form as (5). Computed current transients.
For use with the three-phase, time-domain model, the
unknown values of X and Y were determined using the
following procedure. First, the value of a constant arc is quasi-static and possibly that the effects of arc extinc-
voltage was found, which gave a true rms arcing current tion and restriking around voltage zero were not modeled.
that agreed exactly with the values measured in the IEEE There was also a box effect; K must be multiplied by
tests. This was done by repeatedly solving (4) for each test 0.797 for tests in a box.
shot, computing the true rms current over the last cycle Using this model together with (4), the circuit cur-
before the circuit opened, and iteratively adjusting Varc to rents, voltages, power, and energy can be computed. Typi-
obtain agreement. Then X and Y were determined by a cal results are shown in Figures 5–8 for an ungrounded
multiple regression fit to equation (5). (The analysis used arcing fault.
304 test shots, excluding those with series current-limit- The waveshapes are similar to published data [5], [16].
ing fuses.) This gave X = 0.173 and Y = 0.222, values The current stabilizes quite quickly because of the damp-
which are consistent with the literature. ing effect of the fault arc resistance. The delay in appear-
Then it was assumed that the same X and Y can be ance of the arc voltage is the fusion time of the fine trigger
used to relate the instantaneous arc voltage and current (v wires in each phase. Figure 7 shows the instantaneous
and i), giving a nonlinear transient arc model of the form power as a fraction of the bolted-fault VA. The build-up of

IEEE INDUSTRY APPLICATIONS MAGAZINE • MAY|JUNE 2005 • WWW.IEEE.ORG/IAS

arc energy in Figure 8 is almost linear, but with a delay of
a few milliseconds after the fault begins.
varc = VE + Kiarc 0.173 0.222
g . (6) These solutions were obtained using fourth-order
Runge-Kutta integration of the equations, with auto-
matic adjustment of the time step to achieve a preset
accuracy. However the resistance of the arc model (6)
Finally the value of K was found by a second iterative tends to infinity as the current nears zero, giving a
fitting to the measured arcing current. However K was not very low circuit time-constant, which causes the time
constant, but a relatively strong function of the line-to- step to be reduced to a very small value, and the solu-
line test voltage VLL . (K = 1.827VLL 0.377 with VLL in V). tion “grinds to a halt.” A numerical procedure for solv-
This dependency is not easy to explain, but it is also ing this problem, ensuring a smooth and rapid
implied in Schau and Schade [16] and the IEEE formula. progression of the solution through the current zeros,
It is probably connected with the assumption that the arc is given in [10].

0.6
300

200
Power, p.u.

100
Voltage (V)

0 0.3

−100

−200

−300 0
0 0.01 0.02 0.03 0.04 0.05 0 0.01 0.02 0.03 0.04 0.05
Time (s) Time (s)
6 7
43
Computed fault arc voltage transients. Instantaneous three-phase power.

Authorized licensed use limited to: UNIVERSITY OF SOUTHAMPTON. Downloaded on October 14,2021 at 08:43:09 UTC from IEEE Xplore. Restrictions apply.
 Gammon and Matthews [17] calcu- before circuit opening) was computed
lated arcing currents for single-phase THE TIME and compared with the measured values
arcing faults using a similar time- given in [8]. The results are shown in
domain method (Runge-Kutta integra- DOMAIN Figure 9, and the values predicted by (1)
tion, using both Fisher’s and Stokes or (2) are shown in Figure 10.
and Oppenlander’s model). They APPROACH Fuses were not used for the test
assumed that the arc extinguishes at data of Figures 9 and 10, the circuit
each current zero and then reignites in CAN BE USED being cleared by a back-up breaker.
the next half-cycle when the gap volt- The data covers voltages from 208 V
age reaches a fixed breakdown level TO INVESTIGATE to 13.8 kV, bolted-fault currents from
(dielectric reignition), whereas the INTERACTIONS 700 A to 106 kA, arcing gaps from
model described here shows a continu- 7.1 mm to 152 mm, and various box
ous variation of current through the BETWEEN THE dimensions, as well as tests in the open
zero-crossing period. Dielectric reigni- (304 tests in all). The time-domain
tion can be seen to occur for a single- CIRCUIT, THE model gives a slightly better correla-
phase arc where the power input to the tion (r 2 = 0.989) than the IEEE equa-
plasma drops to zero when the current ARCING FAULT, tion (r 2 = 0.978). However, this
reaches zero. However, for a three- small improvement is not its main
phase arcing fault the situation is less AND CURRENT- advantage. The time-domain model
clear. Although the current in one always predicts arcing currents that are
phase may reach zero, the power input LIMITING FUSES. lower than the bolted-fault current and
to the plasma continues via the other fall as the gap length increases.
two phases [9].
Using the time-domain model, the rms arcing current Calculation of
(geometric mean value for the three phases over the last cycle Incident-Energy Density
The electrical energy input to the high-current arc plasma
is transferred to the surroundings by conduction, convec-
0.7 tion, and radiation, and is also consumed in melting and
0.6
vaporizing the electrode metal at the arc roots. For
enclosed equipment, a substantial part of the energy is
0.5 also converted to pressure rise. The overall energy balance
IEEE INDUSTRY APPLICATIONS MAGAZINE • MAY|JUNE 2005 • WWW.IEEE.ORG/IAS

Energy (MJ)

0.4
is discussed in [16].
For the geometry of Figure 1, the calorimeters princi-
0.3 pally intercept radiant heat from the arcing zone. For tests
in the open, with a total energy Warc the direct radiated
0.2
energy density at a large distance d is β Warc /(4π d 2 )
0.1 where β is the fraction of the total arc energy that is emit-
ted as radiant heat, assuming spherical symmetry. In this
0
0 0.01 0.02 0.03 0.04 0.05 case, the distance exponent is two. (The inverse-square law
Time (s) does not apply close to the arcing zone.)
8 For tests in an open box, the focusing effect of the box
Total arc energy. changes the situation, as shown in Figure 11.

100 100
Arcing Current, Model (kA)

Arcing Current, IEEE (kA)

10 10

1 1

0.1 0.1
0.1 1 10 100 0.1 1 10 100
Arcing Current, Test (kA) Arcing Current, Test (kA)
9 10
Comparison of predicted and measured arcing current Comparison of predicted and measured arcing current
44
(time-domain model). (IEEE 1584 model).

Authorized licensed use limited to: UNIVERSITY OF SOUTHAMPTON. Downloaded on October 14,2021 at 08:43:09 UTC from IEEE Xplore. Restrictions apply.
 Reflections of radiant heat from the back and sides of dent-energy density is always increased when the arcing
the box can make the arc and the box appear as one much fault is enclosed by a box.
larger heat source, reducing the effective distance exponent.
Effect of Current-Limiting Fuses
Tests in the Open A further advantage of the time domain approach is that
Fitting to the IEEE test data using multiple regression it can be used to investigate interactions between the cir-
gave the following model: cuit, the arcing fault, and current-limiting fuses. The cur-
rent-limiting fuse models used in this work were based
on those described in [19], with some enhancements.
Emax = 84.61ES 0.958 g0.284 VLL −0.532 (7)

Emax = mean maximum energy density at a

distance d, (cal/cm2 )
ES = spherical component of energy density,
(J/mm2 )
= WARC /(4π d 2 )
Warc = total arc energy computed using the
time-domain model, J.
Figure 12 shows a good correlation (r 2 = 0.949) d
between the predictions of (7) and the test values. Open Tests Give Focused Radiation Emitted from
Spherical Radiation (X∼2). Box Is Less Divergent (X < 2).
Tests in a Box with One Side Open
To avoid the anomaly caused by the use of distance expo- 11
nents less than 2, it is possible to calculate the focusing Box-focusing effect.
effect of the box directly, using radiative view factors
[10], [18]. The view factor Fij between 2 surfaces i and j
is defined as the fraction of the radiated energy leaving
surface i which strikes surface j. 100
Radiated energy from the arc strikes the back and sides

IEEE INDUSTRY APPLICATIONS MAGAZINE • MAY|JUNE 2005 • WWW.IEEE.ORG/IAS

Emax, Computed (cal/cm2)

of the box and is then reflected out towards the calorime-

ters. It is necessary to take multiple reflections into
account, as these are not negligible. The inner surfaces 10
behave as diffuse absorbers and reflectors with a reflectivi-
ty α. Incident radiation is reflected equally in all direc-
tions, as illustrated in Figure 13. 1
It is shown in [10] that the presence of the box can be
taken into account by modifying (7) to
0.1
0.1 1 10 100
−0.532
Emax = 84.61{ES + FR(α)Warc } 0.958 0.284
g VLL , (8) Emax, Test (cal/cm2)
12
Predicted incident-energy density, all open tests.
where the term FR(α)Warc is an additional energy term
due to single and multiple reflections from the back
and sides of the box, and can be computed using radia-
tive view factors that are calculated directly from the
dimensions of the box. (The units of FR(α) are mm−2 .) 1 = Back of Box
2 = Top
The only unknown is the reflectivity α. By varying α 3 = Bottom
and computing the correlation between the predictions 4 = Far Side
of (8) and the test data, the optimum value of α was 5 = Near Side
found to be 0.56. In a typical case, direct spherical radi-
ation accounts for about 50% of the incident radiant Arcs
energy. Figure 14 shows a comparison between the inci-
dent-energy density predicted by (8) and the measured
mean maximum values for the entire IEEE data set.
Figure 15 gives a similar comparison for the IEEE
formula. In this case, the time-domain model gives a sig- d0 d
nificantly better correlation (r 2 = 0.856) than the IEEE 13
45
formula (r 2 = 0.775). Use of (8) also ensures that the inci- Box geometry for calculation of reflections.

Authorized licensed use limited to: UNIVERSITY OF SOUTHAMPTON. Downloaded on October 14,2021 at 08:43:09 UTC from IEEE Xplore. Restrictions apply.
 Prearcing Model
During the prearcing time, the fuse voltage is assumed
100 to be zero up to the time when the fuse melts. The
Emax, Computed (cal/cm2) instant of melting can be found by computing the evolu-
tion of the true rms current in each phase, and switching
10 to the arcing state when the fuse’s melting-time/current
characteristic is crossed, as illustrated in Figure 16.
The fuse time-current characteristic is stored as a table
1
which is dynamically fitted with cubic spline functions,
and interpolation is used (as with all the models), to find
0.1
the exact crossing point. For times shorter than the lowest
tabulated value, the adiabatic melt I 2 t is used.
The true rms (virtual) current in each phase is computed as
0.01
0.01 0.1 1 10 100


Emax, Test (cal/cm2)
i 2d t
14 IV = . (9)
t
Time domain model prediction compared with test.

Arcing Models
100 Unlike free-burning arcs in air, the geometry of arcs in a
sand-filled fuse is closely controlled by the surrounding
quartz sand, and short-circuit faults can be modeled quite
Emax, IEEE (cal/cm2)

10 accurately. The models used here are fully described in

[19] and include the effects of arc ignition in the fuse ele-
1
ment notches, burnback of the elements, fusion of the
sand and expansion of the arc cross-section, arc merging,
and final arc extinction, each arc being modeled as a sim-
0.1 ple cylindrical channel. For each fuse design, details of
the element construction and materials are needed, and
IEEE INDUSTRY APPLICATIONS MAGAZINE • MAY|JUNE 2005 • WWW.IEEE.ORG/IAS

0.01
the resulting models give very good agreement with
0.01 0.1 1 10 100 oscillograms obtained from fuse type testing. The fuse
Emax, Test (cal/cm2) models produce a further set of differential equations
15 which have to be solved simultaneously with (4).
IEEE formula prediction compared with test.
Typical Results
Figure 17 shows the current transients computed using
the time-domain model for the interruption of a 43.8-kA
ungrounded arcing fault in a 600-V, 60-Hz, three-phase
Fuse TCC system by three 600-A class RK1 fuses with fault occur-
rence at 0◦ of phase a, and a source power factor of 0.1.
Initially, all three fuses are in the prearcing state, and
the phase currents are lower than the available values
t
because of the arcing fault voltages. In the case shown, the
fuse in phase b melts first (at 2.1 ms) and limits the cur-
rent, the fuse arc voltage acting in series with the arcing
fault voltage. The appearance of the phase b fuse arc voltage
True rms Value of changes the rates-of rise of current in the other two phases.
Circuit Current The phase b fuse finally clears at 4.6ms, leaving the phase a
and phase c fuses in circuit carrying equal and opposite cur-
tmelt rents, and the line-to-line fault current begins to increase.
The phase c fuse melts next (at 5.0 ms) followed shortly
afterwards by the phase a fuse (at 5.4 ms), and the arc volt-
ages of these then act in series to reduce the line-to-line
current rapidly to zero. After this, all three fuses are open.
Figure 18 shows the current waveforms taken from a
test in a high-power laboratory under the same conditions
rms Current as modeled in Figure 17. The sequence of fuse operation in
16 the three phases is the same as predicted, with similar
46
Computation of melting time. waveshapes and peak currents. The difference at the start

Authorized licensed use limited to: UNIVERSITY OF SOUTHAMPTON. Downloaded on October 14,2021 at 08:43:09 UTC from IEEE Xplore. Restrictions apply.
of Figure 18 is due to nonsimultaneous Arc-Flash Characteristic
pole closure in the circuit making For a given set of data (equipment
switch. However the results do indi- THE TIME DOMAIN type and circuit parameters) it is use-
cate that the time domain model gives ful to plot the arc-flash energy densi-
a realistic representation of the inter- MODEL GIVES ty as a function of available
ruption of arcing faults by fuses. bolted-fault current. The curves in
The possible sequences of events
WAVESHAPES Figure 19 show a typical theoretical
during clearing are very complicated, AND PEAK characteristic computed with a time
involving fuse melting and clearing in domain model for a set of three 600-
each phase, and interaction between the CURRENTS THAT A class RK1 fuses, a box size of 508
phases (if the fault is ungrounded). If a × 508 × 508 mm, g = 32 mm,
fuse just fails to melt within a particu- CORRESPOND d0 = 102 mm, and d = 457 mm.
lar half-cycle, the melting time jumps The upper curve is the maximum
to a subsequent half-cycle. Sometimes WELL WITH THOSE value (worst closing angle) and the
all three fuses open, but in many cases lower curve is the minimum value
only two fuses operate. OBSERVED IN (most favorable closing angle). The
HIGH-POWER knee of the characteristic corresponds
Point-on-Wave Effects roughly to the 0.01 s melt current on
The results are also affected by the TESTS. the fuse’s time-current characteristic.
point-on-wave at which the arcing Also shown in Figure 19 are a
fault begins. For three-phase sys- number of experimental points
tems, all possible outcomes are cov- obtained in a high-power test labora-
ered if the closing angle θ (with respect to the voltage tory using the IEEE 1584 test method. The square-block
of phase a ) is varied in the range 0 ≤ θ ≤ 60◦ . points are the median calorimeter readings from a test
A study of point-on-wave effects [10] has shown that
below the fuse’s current-limiting threshold current the
incident-energy density is not significantly affected by 30
the closing angle, but within and above the threshold a
region the closing angle has a significant influence. 20
After examining point-on-wave effects for several differ-
ent fuse designs, and considering the additional varia- 10

IEEE INDUSTRY APPLICATIONS MAGAZINE • MAY|JUNE 2005 • WWW.IEEE.ORG/IAS

tions that will be found in practice due to the chaotic
Current (kA)

fault arc behavior, it is concluded that it is not possible 0

to recommend a worst-case closing angle for arc-flash
testing, in a similar way to that which is used for type −10
testing of current-limiting fuses. The best method b
appears to be to use random closing, but with several −20
c
tests, to obtain a range of arc-flash energy values. The
results are also affected significantly by the size of the −30
trigger wires used to initiate the arcing fault. Variation 0 2 4 6 8 10
of the melting time of these wires produces effects simi- Time (ms)
17
lar to the variation of the point-on-wave.
Computed currents for arcing fault with fuses.

100
30
a
20

10 10
Current (kA)

Ei (cal/cm2)

−10 1
b
−20
c
−30
0 2 4 6 8 10 0.1
1 10 100
Time (ms) Bolted-Fault Current (kA)
18 19
47
Currents measured during high-power test. Arcing flash characteristic.

Authorized licensed use limited to: UNIVERSITY OF SOUTHAMPTON. Downloaded on October 14,2021 at 08:43:09 UTC from IEEE Xplore. Restrictions apply.
 series, while the diamonds are the maximum values. test requirements may use Stokes and Sweeting’s arrange-
Unfilled points are tests in which the fuses did not clear ment of the electrodes rather than that shown in Figure 1.
all phases before the backup breaker opened. The test A modification of the incident-energy model will be
results are similar to those that have been published previ- required in this case, as the heat transfer to the calorimeters
ously [20], but the availability of the time domain model will be increased due to the expanding plasma cloud.
gives a useful guide to the interpretation of the test data. However the beneficial effects of current-limiting protec-
The test and computed results show that this fuse can tion for high available currents will still apply.
limit the arc-flash energy density to a level well below the
critical value for a second-degree burn (1.2 cal/cm2 ), but References
only if the available bolted-fault current is high enough to [1] National Electrical Code, NFPA 70, 2002.
cause operation in the current-limiting mode. [2] Standard for Electrical Safety Requirements for Employee Workplaces, NFPA
70E, 2004.
For these calculations, it was assumed that the fault [3] R.A. Jones, D.P. Liggett, M. Capelli-Schellpfeffer, T. Macalady, L.F.
arcs could be represented by (6) with unchanged values of Saunders, R.E. Downey, L.B. McClung, A. Smith, S. Jamal, and V.J.
k, X and Y. However, some improvements are needed, Sporita, “Staged tests increase awareness of arc-flash hazards in electri-
because Stokes and Oppenlander [14] showed that for cal equipment,” in IEEE Petroleum and Chemical Industry Conf. Rec.,
Sept. 1997, pp. 313–332.
time durations of a few milliseconds, the arc does not [4] R.H. Lee, “The other electrical hazard: Electric arc blast burns,” IEEE
move far from its starting location. During the first few Trans. Ind. Applicat., vol. 18, pp. 246–251, May/June 1982.
cycles of arcing the arc length and voltage increase [9], so [5] T.E. Neal, A.H. Bingham, and R.L. Doughty, “Protective clothing
the fault arc geometry for very short times will be differ- guidelines for electric arc exposure,” IEEE Trans. Ind. Applicat., vol. 33,
ent from that which develops over a period of several pp. 1043–1054, July/Aug. 1997.
[6] R.L. Doughty, T.E. Neal, and H.L. Floyd II, “Predicting incident energy
cycles, giving a possibly significant change in fault arc to better manage the electric arc hazard on 600 V distribution
voltage and incident energy. systems,” in Proc. IEEE PCIC, Sept. 1998, pp. 329–346.
[7] R.L. Doughty, T.E. Neal, T.A. Dear, and A.H. Bingham, “Testing
Conclusions update on protective clothing and equipment for electric arc exposure,”
IEEE Ind. Appl. Mag., vol. 5, pp. 37–49, Jan.-Feb. 1999.
A time-domain model of arc-flash hazard has been devel- [8] Guide for Performing Arc-Flash Hazard Calculations, IEEE-1584, Sept.
oped. The ordinary differential equations for the three- 2002.
phase circuit and any current-limiting fuses are solved by [9] A.D. Stokes and D.K. Sweeting, “Electric arcing burn hazards,” in Proc.
fourth-order Runge Kutta integration with automatic 7th Int. Conf. on Electric Fuses and Their Applications, Gdansk University
control of the time step. of Technology, Poland, Sept. 2003, pp. 215–222.
[10] R. Wilkins, M. Allison, and M. Lang, “Time-domain model of 3-
The three-phase arcing fault is represented as a Y- phase arc flash hazard,” in Proc. 7th Int. Conf. on Electric Fuses and Their
connected set of nonlinear resistors, and the their charac- Applications, Gdansk University of Technology, Poland, Sept. 2003,
IEEE INDUSTRY APPLICATIONS MAGAZINE • MAY|JUNE 2005 • WWW.IEEE.ORG/IAS

teristics have been determined by least-squares fitting to pp. 223–231.

the published IEEE dataset. The resulting arc characteris- [11] R. Wilkins, “3-phase operation of current-limiting power fuses,” in
Proc 3rd Int. Conf. on Electric Fuses and Their Applications, Eindhoven,
tics are similar to those which have previously been mea- 1987, pp. 137–141.
sured for high-current single- and three-phase arcs in air. [12] L.E. Fisher, “Resistance of low-voltage arcs,” IEEE Trans. Ind. General
For arcing faults in an open box, the focusing effect of Applicat., vol. IGA-6, pp. 607–616, Nov./Dec. 1970.
the box is taken into account using radiation view factors [13] V.P. Ignatko, “Electric characteristics of AC open heavy-current arcs,”
to allow for multiple reflections of radiant heat. in Proc. 3rd Int. Symp. on Switching Arc Phenomena, TU Lodz, Poland,
1977, pp. 98–102.
The final model calculates the incident-energy density [14] A.D. Stokes and W.T. Oppenlander, “Electric arcs in open air,”
due to the arc flash at a distance d for tests in the open or J. Phys. D: Appl. Phys. vol. 24, pp. 26–35, 1991.
in a box of arbitrary dimensions, with a given electrical [15] J. Paukert, “The arc voltage and the resistance of LV fault arcs,” in
power system and interbus electrode gap. It gives good Proc. 7th Int. Symp. on Switching Arc Phenomena, TU Lodz, Poland,
1993, pp. 49–51.
correlation with the IEEE 1584 test data and can also be [16] H. Schau and D. Stade, “Requirements to be met by protection and
used to study point-on-wave and other effects. switching devices from the arcing protection point of view,” in Proc.
While there is considerable scope for improving the 5th Int. Conf.on Electric Fuses and their Applications, Technical Universi-
arcing fault model used here, its interaction with the cir- ty of Ilmenau, Germany, Sept. 1995, pp. 15–22.
cuit is correctly represented, so that the anomalies of the [17] T. Gammon and J. Matthews, “Instantaneous arcing-fault models
developed for building system analysis,” IEEE Trans. Ind. Applicat.,
IEEE 1584 equations are avoided. vol. 37, pp. 197–203, Jan./Feb. 2001.
Improvements that can be made to the model include [18] M.N. Ozisik, Heat Transfer. New York: McGraw-Hill, 1985.
a better representation of the reignition processes at cur- [19] R. Wilkins, “Standard fuse model for system short-circuit studies,” in
rent zero crossings and the dynamic growth of the fault 8th Int. Symp. on Switching Arc Phenomena, TU Lodz, Poland, 1997, pp.
163–166.
arc lengths. [20] R.L. Doughty, T.E. Neal, T.L. Macalady, and V. Saporita, “The use of
When used in conjunction with dynamic models of low-voltage current-limiting fuses to reduce arc flash energy,” IEEE
current-limiting fuses, the time domain model gives Trans. Ind. Applicat., vol. 36, pp. 1741–1749, Nov./Dec. 2000.
waveshapes and peak currents that correspond well with
those observed in high-power tests. The model also illus- R. Wilkins ([email protected]), M. Allison, and M.
trates the significant reduction in arc-flash hazard that can Lang are with Ferraz Shawmut, Inc. in Newburyport, Mass-
be achieved if current-limiting fuse protection is used and achusetts. This article first appeared in its original form at the
the available fault current is high enough to cause the 2004 IEEE Industrial and Chemical Power Systems Techni-
fuses to operate in their current-limiting region. Future cal Conference.
48

Authorized licensed use limited to: UNIVERSITY OF SOUTHAMPTON. Downloaded on October 14,2021 at 08:43:09 UTC from IEEE Xplore. Restrictions apply.