Lightning impulse wave-shapes: defining the true origin and it’s impact on parameter

evaluation
D.J.Swaffield1*, P.L.Lewin1, N.L.Dao1 and J.K.Hallstrom2
1
University of Southampton, Southampton, Hampshire, SO17 1BJ, UK
2
Helsinki University of Technology, 02015 TKK, Finland
*Email: [email protected]

Abstract: High-voltage testing of equipment is The generation of impulse wave-shapes for testing
conducted in accordance with IEC 60060-1 which generates disturbances in the measured signal including
defines the parameters of full lightning impulse wave- oscillations near the origin and overshoot. If the wave-
shape and methods for generating and recording the test shape is measured digitally it is desirable to remove the
wave-shape. In the next revision of the standard it is disturbances to the wave-shape where these do not have
proposed that the signal be digitally recorded and after significant effect on the test result. Thus it is planned
suitable signal processing the parameter evaluation be that within the next revision of the IEC60060-1 to
implemented by a standardized algorithm. This paper include a suitable digital filtering technique for removal
examines two methods for parameter evaluation; firstly of disturbances to the recorded signal and method for
using a linear rising edge approximation and secondly a parameter evaluation. This will offer an improvement
3 standard deviations method. Obtained parameters are over the loosely defined existing method and ensure a
benchmarked against the IEC 61083-2 Test Data uniform approach is adopted worldwide.
Generator and experimental results generated at the In the next revision of IEC60060-1 it is proposed to
Tony Davies High Voltage Laboratory, University of filter either the whole of the recorded wave-form or just
Southampton. The sensitivity of the results to the the oscillations and overshoot, known as the residual
method of implementation whilst using a double data. This residual is found by subtracting a curve of
exponential function for the curve fitting is highlighted. the double exponential (DE) form given in (4) and fitted
The use of a separated double exponential function for using a least mean squares approach.
curve fitting is proposed and shown to overcome this
sensitivity.  − (t −t 0 ) − (t −t 0 ) 
 
u (t ) = α  e τ 2 − e τ 1  (4)
 
1 INTRODUCTION  

Standard high-voltage impulse testing requires the The choice of a suitable digital filter, termed the
application of a specified wave-shape. IEC60060-1 and K-factor, has been discussed elsewhere [2-7]. However
-2 are international standards that specify the high- it is important to also consider the method of parameter
voltage measuring techniques and measuring equipment evaluation and specifically the method for defining the
to be used for testing [1]. With reference to Fig. 1, a true origin as this will influence the recorded front and
full lightning impulse wave-shape is specified as having tail times (T1 and T2) using the parameter evaluation
a front time (T1) of 1.2us ± 30% and a tail time (T2) of method currently specified by IEC60060-1. It has been
50us ± 20%. Where proposed that defining the true origin using (3) be
replaced by finding the time at which recorded data
T1 = 1.67(t90 − t30 ) 0.84 µs ≤ T1 ≤ 1.56 µs (1)
UP
and U90

T2 = t50 − t0 40 µs ≤ T2 ≤ 60 µs (2)
Voltage

Where t0, t30 and t90 are the true origin of the U50
impulse, the time taken to reach U30, 30% of the peak
voltage, and the time taken to reach U90 that is 90% of U30
the peak voltage. Within the current standard the true
origin can be calculated using

U 30
t 0 = t 30 − (t 90 − t 30 ) (3) t0 t30 t90 Time t50
U 90 − U 30
Fig. 1. Definitions for calculation of lightning impulse voltage
waveform parameters.
rises above 3 standard deviations from the baseline data 5
x 10
12
at the start of a recorded wave, where it is assumed this
is time at which the wave-shape signal is greater than 10
the level of background noise [5].
In this paper a series of experimental wave-shapes 8
have been recorded and are filtered digitally using a

Voltage (V)
6
non-causal zero-phase filter that matches the K-factor
filter proposed for the next edition of IEC60060-1 [8]. 4
Using this experimental data and waveforms obtained
from the IEC 61083 -2 Test Data Generator; the 2
influence of method used to define the true origin and
the resultant impact on the front time and tail time 0

recorded from the test wave-shapes is examined.

-2
-1 0 1 2 3 4 5
Time (s) -6
x 10
2 DEFINING THE TRUE ORIGIN
(i)
5
x 10
10
2.1. Method One – linear approximation
In the current version of IEC 60060-1 the true origin 8
is defined using a linear approximation of the rising
slope, this line is extended back to the point where it Voltage (V) 6
crosses the time axis to define the true origin, (3). If
this technique is applied before a filter is applied to the 4
raw data of a waveform noise, near the origin or on the
rising slope, can cause the line to be plotted with an 2
erroneous gradient. This may arise because the noise
may cause a second value at which the signal is at 30 % 0
of the peak voltage, for an example of this see Fig. 2(ii).
-2
2.2. Method Two – 3 standard deviations -1 0 1 2 3 4 5
Time (s) -6
x 10
With reference to Fig 1. it is clear that even using a (ii)
filtered signal a linear rising edge approximation may 5
x 10
2
lead to a discrepancy between the true origin, if it is to
be defined as the point at which the signal rises above 0
the background noise compared to an estimate using (3).
This has lead to the proposal of defining the true origin -2
as the point when the signal rises above the noise using
Voltage (V)

the point when the signal first rises above 3 standard -4

deviations and the mean of the next five data points is

-6
also above the 5 standard deviation level.
The difficulty with this measure is that again -8
electrical noise near the start of the impulse signal
collected may contain oscillations greater than the -10
background noise due for example to the firing of spark
gaps, for an example see the experimental wave-shapes -12
-1 0 1 2 3 4 5
shown in Fig. 3. This may again lead to an incorrect Time (s) -6
x 10
estimation of the true origin. (iii)

3 TEST WAVE-SHAPES Fig. 2. IEC 601083-2 test data generator impulse waveforms:
(i) Case 08, (ii) Case 11, (iii) Case 14
Three cases from the IEC 61083-2 Test Data
Generator have been used to test parameter evaluations, 4 RESULTS
these are shown in Fig. 2. In addition to these the three
evaluated experimentally generated wave-shapes with It has been established from other studies [5,6,7] that
oscillations deliberately generated on the signal shown the residual filtering method is more appropriate than
in Fig. 3 were also assessed. global filtering for parameter evaluation of the
measured impulse waveform and consequently only the
residual filtering approach has been implemented to 4.1. Double exponential function
assess the performance of the different methods for
Currently within IEC 60060-1 the formula used to
parameter evaluation.
x 10
4 define the curve of best fit is the double exponential
12
(DE) of the form of (4), the best fit of which is found by
employing a least mean squares curve fitting algorithm.
10
Table 1 shows the resulting parameter evaluations using
the DE form. What is immediately clear from this table
8
is that the use of (3) the 30%-90% linear rising edge
approximation to fit the curve and evaluate parameters
6
leads to the greatest least mean square (LMS) errors.
Voltage (V)

The reason for this error between the true data and the
4
curve fit is down to a poor estimate of t0.
As an example Fig. 4 shows the original data, fitted
2 curve and final curve after residual filtering has been
applied for case 8 of the IEC 61083-2 Test Data
0 Generator using both methods for defining t0. With
reference to Fig. 4 it is clear that when using the DE
-2
-1 0 1 2 3 4 5 curve form, selection of (3) or 3 standard deviations
Time (s) x 10
-6
makes a difference to both the gradient of the rising
(i) edge of the fitted curve and also the shape near the
4

2
x 10 origin. Considering the peaks shows that selecting (3)
over the 3 standard deviations method leads to a sharper
0 gradient (Fig. 4 (i) and (iii)), which in turn has the effect
-2
of causing a negative overshoot near the origin in
Fig. 4(ii). A better curve fit, i.e. minimised error, results
-4 from selecting the 3 standard deviations method for this
example. Table 1 shows that the error is reduced
Voltage (V)

-6
considerably and no significant change is seen in the
-8 error or parameters from selecting either method for
parameter evaluation on the resultant wave-form, indeed
-10
these results fall within the IEC reference limits.
-12 However considering a case where there is noise near
the origin further complicates the task of correctly
-14 selecting t0. As an example Fig. 5 shows the resultant
-16
plots of curves fitted to the experimentally generated
-1 0 1 2 3 4 5
Time (s) -6
waveform of Fig. 3 (ii). Fig. 5(ii) and (iv) show the
x 10
region near the origin for curves fitted using the DE
(ii)
x 10
4 form by using (3) and the 3 standard deviations method
12
respectively. This shows that whilst selecting the
3 standard deviations reduces the error significantly it is
10
not as good as results later found from the separated
double exponential (SDE) method. This is because
8
without an exact definition of t0 there will be greater
residual near the origin additional to the noise already
6 present in this region. Thus after the residual is filtered
Voltage (V)

and added back onto the fitted curve there is again a

4 resultant negative overshoot near the origin.

2
4.2. Separated double exponential function
It is clear therefore from the proceeding results that
0 the chief difficulty with employing the double
exponential function is choosing the true origin. To
-2
-1 0 1 2 3 4 5 overcome this apparent difficulty an alternative
Time (s) x 10
-6
approach has been sought and the separated double
(iii) exponential function has been proposed. Again this can
be used to describe a wave-shape having two
Fig. 3. Experimentally generated impulse voltages exponential components as is the case for the circuit
used to generate the impulse wave-shape. The form of of the reference data for both curve fits. As has been
the separated double exponential function is discussed elsewhere [8] this is due a change in the
treatment of oscillations superimposed onto the high-
 − (t − t 0 )   − (t − t 0 )  voltage test wave-form and their reduction by the
u (t ) = A1 exp  − A2 exp  (5) application of digital filtering. The match in the results
 τ2   τ1  arises in this case because this wave-form is without
large oscillations or noise at the origin and both DE and
Where A1 and A2 replace the single constant α used in SDE curves provide a good fit when using the
(4) and introduce an extra degree of freedom when 3 standard deviation method to define an initial estimate
fitting the curve. This makes the task of finding the best of t0. The most notable change is in tail time between
fit curve more difficult and requires careful choice of the different methods for front time evaluation; this is
the curve fitting algorithm to ensure solver stability. To because tail time is sensitive to the selection of t0 as is
fit curves of both DE and SDE forms the same seen from Fig. 1.
Levenberg-Marquardt least mean squares curve fitting
algorithm has been applied. Despite the extra variable 11
x 10
5
1
x 10
5

data
in the SDE form, there is no noticeable difference in 10.5 fit curve data
final curve fit curve
computation time when running demonstration 10
0.5
final curve

voltage (V)

voltage (V)
algorithms on a desktop computer to find solutions for 9.5 0
each of the two curve forms. Both the DE and SDE 9
methods take 2-3 seconds per curve. Additionally -0.5
8.5
computation times maybe extended up to three times
8 -1
longer if a poor choice for origin is selected for the DE 0 0.1 0.2 0.3 0.4 0.5 0.6
time (s)
0.7 0.8 0.9
x 10
1
-5
-6 -4 -2 0
time (s)
2 4 6
x 10
-6

method. (i) (ii)

Table 1 shows that the results when using the SDE 11
x 10
5
1
x 10
5

data
function produces a lower least mean square (LMS) 10.5 fit curve
final curve
data
fit curve
error for all the example wave-shapes examined. 10
0.5
final curve
voltage (V)

voltage (V)
Investigation of the variation of the parameters 9.5 0

evaluated, peak voltage, front and tail time with respect 9

to a change in the initial estimate of the origin time is -0.5
8.5
shown for DE and SDE curves, Fig. 7 an Fig. 8
8 -1
respectively. From Fig. 7 there is a clear minimum in 0 0.1 0.2 0.3 0.4 0.5 0.6
time (s)
0.7 0.8 0.9
x 10
1
-5
-6 -4 -2 0
time (s)
2 4 6
x 10
-6

the error results, signifying the sensitivity of a good fit (iii) (iv)
of the DE function to a good choice of t0. Conversely 11
x 10
5
1
x 10
5

data
Fig. 8 shows that there is very little change in any of the 10.5 fit curve
final curve
data
0.5 fit curve
evaluated parameters across a range twice as large as 10
final curve
voltage (V)

voltage (V)

that of Fig. 7. What is in evidence at this scale is a 9.5 0

small instability in the estimation of the parameters, 9

which arises due to instability in solving for the 8.5
-0.5

variables of the SDE.

8 -1
An investigation of the variation of the LMS error 0 0.1 0.2 0.3 0.4 0.5 0.6
time (s)
0.7 0.8 0.9
x 10
1
-5
-6 -4 -2 0
time (s)
2 4 6
x 10
-6

against a poor estimate of t0, Fig. 6, demonstrates how (v) (vi)

5 5
insensitive a good fit of the SDE is to the initial 11
x 10
1
x 10
data
estimate. Indeed providing the estimate is within the 10.5 fit curve
final curve
data
0.5 fit curve
flat region of the curve in this example from 10 final curve
voltage (V)

voltage (V)

approximately -1.5 µs to +3 µs the error varies by only 9.5 0

0.028% this can be compared to 100% in the same 9

region for the DE function. The variation of τ1 and τ2 8.5
-0.5

when curve fitting a SDE is also small compared to the

8 -1
equivalent parameters for the DE; indicating that the 0 0.1 0.2 0.3 0.4 0.5 0.6
time (s)
0.7 0.8 0.9
x 10
1
-5
-6 -4 -2 0
time (s)
2 4 6 8
x 10
-6

front and tail shape and consequently front time (vii) (viii)
parameter will also be less sensitive to a poor initial
estimate of t0. Fig. 4. Plots of resultant curves for different curve fitting
Disregarding the Tl-Tl result already discussed, for methods and for DE and SDE forms applied to case 8 of IEC
case 8 of the Test Data Generator (Fig. 2(i)) there is no 60183-2 Test Data Generator; (i) and (ii) show the peak and
difference in the Ts-Ts and Tl-Tl results of the DE when origin using (3) and DE, (iii) and (iv) show the peak and origin
using 3 standard deviations and DE, (v) and (vi) show the
compared to the SDE curve fit (Table 1). The front and peak and origin using (3) and SDE, (vii) and (viii) show the
tail times fall within the reference limits provided by peak and origin using 3 standard deviations and SDE
IEC 60183-2. However the peak is found to be just low
 5 4 5
x 10
-1.35 1.2 x 10 x 10
data data 12 10.04
fit curve 1 fit curve
final curve 0.8 final curve 10 10.02
-1.4 0.6 10
8

peak(V)
0.4
9.98

error
voltage (V)

voltage (V)
t0s
0.2 6 t0l
-1.45
0
9.96
4
-0.2 9.94
-1.5 -0.4 2 t0l 9.92
-0.6 t0s
-0.8
0 9.9
-1 -0.8 -0.6 -0.4 -0.2 0 -1 -0.8 -0.6 -0.4 -0.2 0
-1.55
-2 0 2 4 6 8 10 12 14 16
-1
-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1
time(s) x 10
-6 time(s) x 10
-6
time (s) -6 time (s) -5 -6 -5
x 10 x 10
(i) x 10 (ii) x 10
(i) (ii) x 10
-6
x 10
-5
5 4 1.4 4.9
-1.35 x 10 1.2 x 10
data data 1.3
fit curve
1 fit curve t0l
1.2 t0s

front time (s)

final curve 0.8 final curve

tail time (s)

-1.4 1.1 4.85 t0l
0.6 t0s
0.4 1
voltage (V)
voltage (V)

0.2 0.9 4.8

-1.45
0 0.8
-0.2
0.7
-1.5 -0.4
0.6 4.75
-0.6 -1 -0.8 -0.6 -0.4 -0.2 0 -1 -0.8 -0.6 -0.4 -0.2 0
-0.8 time(s) x 10
-6 time(s) x 10
-6
-1.55 -1
0 2 4 6 8
time (s)
10 12 14 16 18
x 10
-6
-1 -0.8 -0.6 -0.4 -0.2 0
time (s)
0.2 0.4 0.6 0.8
x 10
1
-5
(iii) (iv)
(iii) (iv) Fig.7. DE variables against time, where t0l is the origin using
-1.35
x 10
5
1.2 x 10
4
(3), t0s is the origin using 3 standard deviations; (i) least mean
data data
fit curve 1 fit curve square error against time, (ii) peak voltage against time, (iii)
final curve 0.8 final curve
-1.4 0.6 front time against time, (iv) tail time against time.
0.4
voltage (V)

voltage (V)

0.2 5
-1.45 x 10
0 0.7052 9.9932
-0.2 9.9931 t0s
-0.4 0.7052 9.993
-1.5 t0l
-0.6 9.9929

peak(V)
0.7051 9.9928
error

-0.8
-1.55 -1 9.9927
-2 0 2 4 6 8 10 12 14 16 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 0.7051
time (s) x 10
-6 time (s) x 10
-5
9.9926
(v) (vi) 0.705 9.9925
9.9924
4
5
1.2 x 10
t0l
-1.35 x 10 0.705 t0s 9.9923
data data -2 -1.5 -1 -0.5 0 -2 -1.5 -1 -0.5 0
fit curve 1 fit curve
final curve 0.8 final curve time(s) x 10
-6 time(s) x 10
-6

-1.4 -6 -5
0.6
0.4
(i) x 10 (ii) x 10
-6 -5
voltage (V)
voltage (V)

0.2 x 10 x 10
-1.45
1.1523 4.8235 t0l
0
1.1523 4.8234
-0.2
1.1523
front time (s)

-0.4
tail time (s)

-1.5 4.8233
-0.6 1.1523
-0.8 4.8232
1.1523 t0lt0s
-1.55 -1
-2 0 2 4 6 8 10 12 14 16 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 4.8231
time (s) -6 time (s) -5 1.1523
x 10 x 10
1.1523 4.823
(vii) (viii) t0s
1.1523 4.8229
Fig.5. Plots of resultant curves for different curve fitting -2 -1.5 -1 -0.5 0 -2 -1.5 -1 -0.5 0
time(s) x 10
-6 time(s) x 10
-6
methods and for DE and SDE forms applied to the -150 kV
(iii) (iv)
experimental example (Fig. 3(ii)); (i) and (ii) show the peak
Fig.8. SDE variables against time, where t0l is the origin
and origin using (3) and DE, (iii) and (iv) show the peak and
using (3), t0s is the origin using 3 standard deviations; (i) least
origin using 3 standard deviations and DE, (v) and (vi) show
mean square error against time, (ii) peak voltage against time,
the peak and origin using (3) and SDE, (vii) and (viii) show
(iii) front time against time, (iv) tail time against time.
the peak and origin using 3 standard deviations and SDE

5 CONCLUSIONS
It is argued that the SDE provides a more robust
formula for curve fitting and negates the need to change
the method used for parameter evaluation from (3) to a
LMS error

3 standard deviation method. Although using the 3

standard deviations method to provide the initial
estimate of t0 offers significant improvement when used
for the DE curve, only a small gain may be found in
using it at the initial curve fitting stage when using the
SDE. However the DE curve fit never matches the
performance of fitting an SDE curve. It is suggested
time that the SDE may provide a method worthy of adoption
for the filtering of high-voltage full lightning impulse
Fig.6. Error against drift from true origin when fitting the wave-shapes.
SDE curve form
 Table 1. Parameter evaluations for DE and SDE*
Case08 DE curve fitting SDE curve fitting IEC reference
Tl-Tl Ts-Ts Ts-Tl Tl-Tl Ts-Ts Ts-Tl
Low High
LMS error 3.862 0.434 0.434 0.433 0.433 0.433
Peak(MV) 1.039 1.039 1.039 1.039 1.039 1.039 1.040 1.060
Front (µs) 1.737 1.603 1.603 1.603 1.603 1.603 1.600 1.700
Tail (µs) 47.56 47.16 47.49 47.49 47.25 47.49 45.00 49.00

Case11 DE curve fitting SDE curve fitting IEC reference

Tl-Tl Ts-Ts Ts-Tl Tl-Tl Ts-Ts Ts-Tl
Low High
LMS error 1.742 1.566 1.566 1.045 1.045 1.045
Peak(MV) 0.955 0.955 0.955 0.955 0.955 0.955 0.940 0.960
Front (µs) 1.353 1.319 1.319 1.219 1.169 1.169 1.070 1.190
Tail (µs) 87.45 87.27 87.43 87.36 87.64 87.36 82.00 91.00

Case14 DE curve fitting SDE curve fitting IEC reference

Tl-Tl Ts-Ts Ts-Tl Tl-Tl Ts-Ts Ts-Tl
Low High
LMS error 6.478 0.783 0.783 0.767 0.767 0.767
Peak(MV) -1.026 -1.028 -1.028 -1.028 -1.028 -1.028 -0.950 -0.970
Front (µs) 2.238 2.171 2.171 2.154 2.154 2.154 1.850 2.050
Tail (µs) 42.03 41.47 41.89 41.90 41.68 41.90 43.00 47.00

Experiment DE curve fitting SDE curve fitting

100kV Tl-Tl Ts-Ts Ts-Tl Tl-Tl Ts-Ts Ts-Tl
LMS error 2.334 0.711 0.711 0.496 0.496 0.496
Peak(kV) 100.2 100.1 100.1 100 100 100
Front (µs) 1.436 1.336 1.336 1.269 1.269 1.269
Tail (µs) 45.87 45.64 45.86 45.85 46.29 45.85

Experiment DE curve fitting SDE curve fitting

-150kV Tl-Tl Ts-Ts Ts-Tl Tl-Tl Ts-Ts Ts-Tl
LMS error 2.390 0.668 0.668 0.423 0.423 0.423
Peak(kV) -149.3 -149.2 -149.2 -149.1 -149.1 -149.1
Front (µs) 1.419 1.336 1.336 1.253 1.253 1.253
Tail (µs) 45.87 45.66 45.88 45.87 46.00 45.87

Experiment DE curve fitting SDE curve fitting

110kV Tl-Tl Ts-Ts Ts-Tl Tl-Tl Ts-Ts Ts-Tl
LMS error 4.316 1.537 1.537 0.908 0.907 0.907
Peak(kV) 103.6 105 105 106.1 106.3 106.3
Front (µs) 0.751 0.601 0.601 0.434 0.401 0.401
Tail (µs) 39.21 38.38 38.48 37.78 38.23 37.66

*Notation: within Tab. 1. Tl-Tl signifies results obtained using the 30%-90% line approximation (3) to define t0 for the
curve fitting algorithms and the subsequent use of the same approximation to evaluate T1 and T2 post filtering. Ts-Ts
signifies the use of the 3 standard deviation method for an initial estimate of t0 for use by the curve fitting algorithms
and then for parameter evaluation. Ts-Tl signifies the use of the 3 standard deviation method for an initial estimate of t0
for use by the curve fitting algorithms and then (3) to define t0 for parameter evaluation.

REFERENCES [5] J Hällström, S Berlijn, M Gamlin, F Garnacho, E Gockenbach,

T Kato, Y Li and J Rungis “Applicability of Different
Implementations of K-factor Filtering Schemes for the
[1] IEC 60060-1:1989 “High Voltage Test Techniques” – Part 1 Revision of IEC60060-1 and -2” in Proceedings 14th
“General definitions and test requirement” published by the International Symposium on High Voltage Engineering,
International Electrotechnical Commission Beijing, China, paper B-3, 25-29th August 2005.
[2] J Rungis and Y Li “Precision Digital Filters for High Voltage [6] Y Li and J Rungis “Evaluation of Parameters of Lightning
Impulse Measurement Systems” IEEE Transactions on Power Impulses with Overshoot” in Proceedings 13th International
Delivery, vol.14, no.4, pp.1213-1220, 1999. Symposium on High Voltage Engineering, Delft, Netherlands,
[3] K Hackemack, P Werle, E Gockenbach and H Borsi “A New 25-29th August 2003.
Proposal for the Evaluation of Lightning Impulses” in [7] M Gamlin “Implementation of the K-factor for the Lightning
Proceedings 6th International Conference on Properties and Impulse Evaluation by means of Digital FIR Filtering” in
Applications of Dielectric Materials, Xi’an, China, pp.93-96, Proceedings 14th International Symposium on High Voltage
21-26th June 2000. Engineering, Beijing, China, paper B-79, 25-29th August 2005.
[4] P Simon, F Garnacho, Berlijn and E Gockenbach “Determining [8] P Lewin, T Tran, D Swaffield and J Hällström “Zero phase
the test voltage factor function for the evaluation of lightning Filtering for Lighting Impulse Evaluation: A K-factor Filter for
impulses with oscillations and/or an overshoot” IEEE the Revision of IEC60060-1 and -2” IEEE Transactions on
Transactions on Power Delivery, vol.21, no.2, pp. 560- 566, Power Delivery in press.
April 2006.