Các bài báo khoa học về Cao áp và vật liệu điện

eBook for You
I HC BCH KHOA H NI
B MN H THNG IN
Tng hp cc bi bo khoa hc giai on 2007-2012
PHN I
eBook for You
CAO P V VT LIU IN
B mn H thng in - i hc Bch Khoa H Ni
NGHIN CU CNG NGH PHN TCH CC PHN T C TNH

IN DN KHC NHAU BNG K THUT CAO P TNH IN
STUDYING THE TECHNOLOGY FOR SEPARATING ELEMENTS WITH DIFFERENT
ELECTRICAL CONDUCTIVITITES USING HV ELECTROSTATIC TECHNIQUE
Nguyn nh Thng, inh Quc Tr

Trng i hc Bch Khoa H Ni
TM TT
Vic tch cc phn t c tr s in dn khc nhau c ng dng rt rng ri
trong cc lnh vc cng nghip, nng nghip, x l cht thi in t Bi bo trnh
by m hnh thit b v kt qu nghin cu s dng cng ngh phn tch cc ht
bng k thut cao p tnh in. Cng ngh ng dng k thut cao p tnh in
c nghin cu nhiu nm ti cc nc trn th gii do c rt nhiu u im nh
tiu tn t nng lng, c hiu sut cao v khng gy nhim mi trng. Tuy nhin
ti Vit Nam, vic nghin cu su cng ngh v nh hng ca cc yu t khc
nhau nh in trng, mi trng ti hiu qu thit b cn cha c u t tha
ng.
ABSTRACT
eBook for You
The separation of elements with different conductivities is widely applied in

industry, agriculture, e-waste processing technology This paper presents the
developed device and the research results, implementing the technology of highvoltage electrostatic technique in particles separation. Application of the technology
has been studied over the world for many years due to its advantages, such as low
power consumption, high separation performance and environment-friendly. However
in Vietnam, the deep research of technology and impact of different factors, such as
the electric field and the environment, has not been invested sufficiently.
T kha: phn tch ht, x l cht thi in t, cao p tnh in.
1. MT S C TNH CA CC
PHN T CN PHN TCH
1.1 in dn v trng lng ring
phc v cho vic nghin cu
cng ngh v th nghim, cc mu c
chn cha phn t cn phn tch l sa
khong ly t m Cm ha (Cm xuyn,

H Tnh). C s ly sa khong lm th
nghim da trn s khc bit v tr s in
tr sut ca cc ht phn t cn phn tch
trong sa khong c th hin trong bng 1
nh sau (theo [1]):
Bng 1. c tnh ca mt s ht khong cht c phn tch

TT
Sa khong
Trng lng ring, g/cm3
in tr sut, .cm
Phn loi
Ilmenite
4,7
1-10-3
Dn in
Zircon
4,6-4,7
1013-1015
in mi
Rutile
4,2-5,2
1-102
Dn in
Thch anh
2,5-2,8
1012-1017
in mi
Pirit
4,9-5,2
10-5-10-1
Dn in
ni. Vic la chn m hnh thit b kiu
1.2 Kch thc quy i ca cc ht
ny da trn mt s c s sau:
Trong thc t cc ht khong sn c
tnh ton v m phng ngi ta thng
M hnh thit b ny hin cn cha

c ng dng ti Vit Nam.
quy v hnh cu, hnh elip hoc bn elip
Thit b c nghin cu c nhiu u
[2,3]. Bng thit b chuyn dng v
im so vi cc thit b ang c s
phng php quy hnh dng cc ht v
dng ti Vit Nam: c cng sut v
dng hnh cu c th tnh c bn knh
hiu sut phn tch cao, d dng lp
tng ng ca cc ht theo cng thc
t vn hnh hiu chnh, tiu hao t
sau [2,3,4]:
nng lng.
rtd
0, 62 3 abc
(1)
Trong a, b v c tng ng l
chiu di, rng v cao ca ht.
Cc ht o c c ly ngu
nhin vi s lng mu l 400 ht.
T kt qu o c v tnh ton c th
Vic s dng m hnh ny cho php

tin hnh th nghim theo cc ch
eBook for You
hnh dng rt a dng, thun tin cho
vn hnh thc t, nhng vi u im

cu to n gin nn d dng thay i
cc thng s k thut (chiu di, hnh
dng, v tr in cc).
S m hnh thit b trn hnh 1:
rt ra kt lun s b: bn knh tng

ng ca cc ht khong sn ti m Cm
ha dao ng trong khong t 70 n
230m. Kch thc ny rt ph hp vi
vic dng thit b kiu mng nghing v
cho hiu sut tch cao [3,5].
2. M HNH CA THIT B
M hnh thit b th nghim do tc
gi cng cc ng nghip thit k, ch to
v lp t ti phng th nghim cao p v
Hnh 1. S ca m hnh thit b phn
vt liu ca Trng i hc Bch khoa H
tch kiu mng nghing

2
1 - Phu cha nguyn liu; 2 - Mng
Tin hnh nhiu th nghim vi vic
nghing; 3 - in cc trn; 4 - in cc
thay i tr s gc nghing ca mng ta
di (hnh tr); 5 - Khay hng sn phm;
cng d dng nhn thy thay i ca s
thun tin cho vic phn tch cc

kt qu th nghim y tc gi chn s
lng khay hng sn phm l 20, cho php
thu hi sn phm phn tch c ti nhiu
phn b lng ht trong cc khay thu hi

sn phm (hnh 2). Nguyn nhn ch yu
l do nh hng ca tc chuyn ng
ban u ca cc ht. C th kt lun trong
trng hp ny s phn b cc ht trong
v tr khc nhau.
cc khay thu hi ph thuc vo kch thc

Cc kt qu th nghim cho php
nh gi nhng c tnh ca cc phn t
cn tch v phn tch nhng nh hng ca
v t trng ring, nhng hiu sut phn

tch khng cao (hm lng Ilmenite nhn
c l 50%, i vi Zircon l 45%).
in cao p tnh in cng nh mt s nh

hng khc n hiu sut ca m hnh
thit b, trn c s c th hiu chnh
eBook for You
nhm hon thin m hnh thit b.

M t qu trnh hot ng ca thit b:
. Trng hp cha c in trng:
Di tc dng ca thit b rung cc
ht s chuyn ng t phu xung mng
nghing. Lc ny cc ht chuyn ng
hon ton di tc dng ca trng lc v
bay vo cc khay hng sn phm.
Hnh 2. Phn b khi lng cc ht trong

cc khay theo gc nghing ca mng.
Qu o chuyn ng ca cc ht
b. Trng hp c in trng: (in cc
ch chu tc ng ca trng lc, lc ny
pha trn c cp in p cc tnh m,
c tnh theo cng thc (2) nh sau:
cn in cc pha di cc tnh dng).
Fg
mug
4
.S.rtd3 .J. g.cos D
3
(2)
in trng s xut hin gia cc

in cc vi mng nghing. Khc vi
Trong : m - khi lng ca ht; rtd - bn
trng hp u, cc ht chuyn ng theo
knh tng ng ca ht; J - t trng
mng nghing sau chuyn ng trong
ring ca ht; g - gia tc ri t do vi tr
in trng. in trng tc ng ln cc
s g = 9,8m/s2; D - gc nghing ca mng.
ht ny v lm thay i qu o chuyn
3
ng ca chng ty theo tnh cht ca ht.
Fc
0,5.cx .J.v 2 .S
Thay i tr s in p t ln in cc tc
l thay i cng in trng, khi c
th thy rt r v ghi li s thay i qu
Vi cx - H s kh ng hc ca mi
trng, v - vn tc chuyn ng ca ht, S
- tit din ca ht.
o ca cc ht.
Phng trnh chuyn ng ca cc
ht sa khong theo nh lut 2 Newton
c bit khi ta tng in p cc

trn vt qu mt tr s no s xut
hin hin tng c mt s ht khi chuyn
nh sau:
ng trong in trng mnh c xu hng
JG
F
(3)
m- khi lng ca ht sa khong,

- vecto vn tc chuyn ng ca ht,
bay ln v va p vo in cc trn sau

quay ngc tr li vo mng nghing.
ng thi mt s ht c xu hng b ht
v pha cc di (cc tnh dng). iu
tng ca cc lc tc ng ln ht, t- thi
ny chng t xy ra hin tng tch
gian chuyn ng.
in vi cng cao ca cc ht. So snh
Cc lc tc ng ln ht sa khong
bao gm nhng thnh phn sau:
vi trng hp trc (khi cha c in

trng), cc ht s phn b trn s lng
eBook for You
G
dv
m.
dt
cc khay thu hi nhiu hn (xem hnh 3).
- Trng lc:
Fg
4
.S.rtd3 .J. g.cos D
3
mug
- Lc tc ng ca in trng:
Fe
E .q
Vi E - cng in trng, q - in tch

ca cc ht.
- Lc do s phn b in trng khng
Hnh 3. Phn b khi lng cc ht trong

cc khay theo in p t ln in cc
u:
Fp
H 1
4.S.H0 .r
.E . gradE
H2
3. KT LUN:
Quan st khay thu hi sn phm bng
Vi H0 - hng s in mi tuyt i v bng
mt thng d dng nhn thy pha
8,854.10-12 F/m, H - hng s in mi.
nhng khay nm pha xa in cc cha

cc ht c mu sm hn (cc ht Ilmenite)
- Lc cn ca mi trng:
cn cc khay gn pha di in cc c
mu sng hn (Zircon v thch anh). Cc
mu sn phm nhn c trong khay thu
hi qung c tin hnh nh gi
nhm hon thin thit b.
Mu tinh qung Zircon nhn c cha

99,0% Zircon.
Nh vy c th thy vic s dng m
hnh thit b v cng ngh cao p tnh in
vi vic la chn ng cc thng s k
Cn c vo kt qu th nghim trn m
thut lin quan (cng in trng, gc
hnh thit b vi vic thay i nhiu thng
nghing ca in cc, v tr cc in
s c th nhn c kt qu tch vi hiu
cc) cho php nng cao ng k hiu
sut cao. Khi la chn c thng s ti
sut lm giu qung.
u nh cng in trng, kch thc,

v tr v gc nghing ca in cc, kt
qu phn tch sa khong t m Cm Ha
c th t c nhng t l nh sau:
Mu tinh qung Ilmenite nhn c c
Cc kt qu nghin cu trn khng

nh li mt ln na li th ca dng m
hnh thit b ny khi p dng tch cc
loi khong sn khc c tnh cht gn
ging vi sa khong ca m Cm Ha.
eBook for You
cha 99,6% Ilmenite;
TI LIU THAM KHO:
1. Mesenhiashin .I. , ., , 1978.

2. Vereshagin I.P., Levitov V.I., Mirdabekian G.D., Pashin ..
. ., , 1974.
3. inh Quc Tr, S.. rivov.

(). /. . XII -
. .3- , 2006 , 440-441.
4. Fraas F. Electrostatic Separation of granular Materials. Bull. U.S Mines, 603,1962.
5. Berlinxki .I. , ., . 1988.
a ch lin lc:
inh Quc Tr, T. (04)38692009

B mn H thng in, trng i hc Bch Khoa H Ni
eBook for You
S 1, i C Vit, H Ni.
Streamer inception in mineral oil under ac voltage

Influence of electrode geometry, liquid conditioning,
and consequences for test methods of liquids
O. Lesaint
T.V. Top
Grenoble Electrical Engineering Laboratory (G2E lab)

CNRS, Grenoble-INP and Joseph Fourier University
Grenoble, France
olivier.lesaint@grenoble.cnrs.fr
now at: Hanoi Polytechnic Institute (IPH)

Hanoi, Vietnam
I.
INTRODUCTION
It has been known for a long time that initiation of

breakdown in liquids under ac voltage is a very complex
process. Even in the simplest situation (a liquid between two
metallic electrodes), many parameters are able to influence the
initiation of breakdown:
- the liquid chemical nature;
- scale effects. Breakdown voltage in quasi-uniform field
decreases when the stressed liquid volume and/or electrode
surface area are increased [1,2]. "Volume" or "electrode
surface area" effects are usually interpreted in terms of "weak
points able to trigger breakdown (solid particles in the liquid
volume, or electrode surface defects). The probability to get
large weak points increases with stressed liquid volume, and
electrode surface area;
- pollution: breakdown voltage in quasi-uniform field under
ac decreases when solid particles (metallic, hydrated cellulose
fibers, ) and water are present [3, 4];
- time: breakdown initiation shows a large statistical
variation, and is strongly affected by the time duration of
voltage application (the longer the time, the lower the initiation
voltage);
- injected space charge may also have an influence: field

reduction by homocharges, or increase with heterocharges.
It remains very difficult to model and predict the initiation
of breakdown in practical situations. This study is devoted to
obtain a more comprehensive and quantitative description of
these phenomena under ac voltage. Since breakdown results
from the initiation and propagation of streamers, the study of
initiation can be done with two main types of experiments:
- under moderately divergent field, the average field in the
gap is very high, and all initiated streamers propagate to
breakdown. The measured breakdown voltage is equal to the
voltage required to initiate a streamer;
- under divergent field (point-plane or rod-plane at large
gaps), streamers can appear due to a high local field, but are
unable to propagate to breakdown. It is thus necessary to detect
streamer inception with more sophisticated techniques. The
main practical advantage of these experiments is the absence of
breakdown (no destruction of electrodes, limited degradation of
the oil). On the other hand, experiments with large rod
electrodes impose us to use high voltage and very large gaps to
avoid breakdown.
A previous study of streamer initiation was carried out with
impulse voltage [5] over a wide range of electrode shapes
(from sharp points of m size to large rounded rods, and with
fixed metallic particles on a flat electrode). This study mainly
highlighted the effect of electrode surface under impulse
voltage. Under ac, the problem becomes much more complex,
since the liquid pollution, time, and injected space charges will
have a large influence. With short impulses these parameters
have a small influence: the time duration is too small to allow
motion of particles from the liquid volume up to electrodes,
and space charge development is also very limited. The
experiments carried out here are done with the objective to
separate (as far as possible) the effects of electrode shape, time,
and pollution by solid particles and water.
II.
EXPERIMENTAL TECHNIQUES
The test cell used was a 150 liter transparent PMMA

container (Fig. 1). The high voltage electrode is facing a
grounded aluminum plane, 50x50 cm in size. Steel points (tip
radius of curvature rp = 10 to 100 m), and rods with a
eBook for You
AbstractThis paper describes an experimental study of

streamer inception in mineral oil under ac voltage, with rod and
point electrodes. Positive and negative streamer inception
frequencies versus voltage are investigated in gaps up to 40 cm
with different electrode shapes and different conditioning of the
oil (filtered oil, addition of cellulose particles, water). Streamer
inception probability increases exponentially versus field, and it
is not possible to simply define an inception voltage. A voltage
(or field) value correlated to an inception probability must be
used to properly compare different experiments (comparison
between liquids, influence of pollution, etc.). Under ac, several
effects are superposed to reduce dielectric strength: scale
effects, influence of pollution, long time duration. With sharp
points, injected space charges considerably influence
experiments, and the results obtained cannot be extrapolated to
practical applications in which the effect of space charge is
mostly absent.
hemispherical end (radii rp = 0.5 to 20 mm) were used. Gap

distances d were investigated up to 40 cm. Both the high
voltage electrode and the distance could be changed without
opening the test cell, and without changing or circulating the
oil, in order to keep exactly the same oil condition. The high
voltage supply was a 300 kV test transformer. To get very
clean oil, the test cell was included in a closed loop including
an oil processing system (1m filter, degassing and drying). An
insulating tube allowed us to take oil samples in the test cell to
measure particle and water contents. To avoid particle
sedimentation and get an homogeneous and stable particle
content, the oil was continuously stirred with two polyethylene
propellers. Cellulose particles, metallic perticles and water can
be added to clean oil. Table 1 shows typical results of particle
counting when different amounts of cellulose were added to
filtered oil.
a few pC, duration <1s), and very large ones at high voltage
with large rods (charge > 1C, duration > 100s). Detection
must be very sensitive in order to detect all streamers. During
propagation of a long streamer, a large number a fast current
pulses is detected. Conventional PD measurement systems
based on pulse detection are unable to record properly
streamers: counting of all current pulses leads to considerably
overestimate the streamer number actually generated. The
detection system must also have a very low level of spurious
noise, typically less than 1 shot per hour (in some experiments,
a very low number of streamers can appear, typically 1 per
hour). The system must be able to count properly such rare
events. In this study, inception was detected by the streamer
light emission using a photomultiplier (PM). This provided a
very sensitive detection. A dead time of 200 s was fixed
after each detection to avoid overcounting streamers. The
detection threshold was fixed above the background noise of
the PM, and the test cell was placed in a totally dark room.
Pollution inlet
HV
electrode
III.
STREAMER INCEPTION FREQUENCY
Fig. 2 shows a typical result obtained with a rod electrode

of medium size (rp = 0.5 mm, distance d = 40 cm), in oil of
technical quality without filtration (figures 2 to 4 correspond to
the same oil sample, i.e. without opening the test cell or
circulating the oil).
PM
-1
Streamer inception frequency F (minute )

3
eBook for You
10
stirrers
Oil
sampling
10
Drying and
filtration unit
Negative Streamer
-1
10
Positive Streamer
Figure 1.
Test cell
-3
10
TABLE I. Typical particle countings.
100
Particle
Filtered
+ 0,7 mg/l
+ 18 mg/l
+ 70 mg/l
size (m)
oil
cellulose
cellulose
cellulose
2 -5
252
748
9690
16316
5-10
20
34
6060
806
10-15
16
56
76
15-25
18
18
25-50
50-100
100-150
>150
Streamer detection has to fulfill several main requirements.

The size of streamers vary considerably in the experiments:
very small streamers with sharp points at low voltage (charge:
150
200
250
Crest voltage (kV)

Figure 2. Inception frequency of positive and negative streamers in oil
without filtration, 20ppm water content. Rod radius rp = 0.5mm, gap distance
d = 40 cm.
The average inception frequency F increases exponentially

versus voltage, up to a value 103 streamers / minute
(corresponding to about 1 streamer initiated every half-wave).
The increase of F at higher voltage is then much slower. At
very low voltage, the exponential variation is still observed
down to very low discharge rates (< 10-2 discharge / minute,
i.e. less than 1 streamer per hour). To obtain significant
measurements at very low rates, total durations up to 2 days
were used in some experiments. In all experiments, no
indication of a threshold minimum voltage for streamer
inception could be obtained. The number of positive and
From figure 2, it is quite clear that it is impossible to define

an inception voltage. In order to make proper comparisons
between experiments (for instance if the liquid nature or
conditioning is changed), only the voltage value corresponding
to an arbitrary discharge frequency (for instance 1
streamer/minute) can be used. Most experiments were carried
out in the frequency range 10-1< F< 102 in order to limit both
the duration of experiments, and the degradation of oil at very
high discharge rates.
-1
10
10

r =10m
r =0.5mm
r =1mm
r =2.5mm
10
10
10
r =5mm
p
is changed. All plots show an exponential increase of inception

frequency. With large rods, the number of positive streamers
becomes slightly higher than negatives, whereas the opposite is
seen with sharp points.
Figure 4 shows the same results plotted versus the
maximum field calculated at the extremity of rods (or points)
by finite elements method. It is very interesting to observe that
all measurements carried out with a fixed radius rp at different
distances d (5 to 40cm) group together to form a unique plot.
This shows that the maximum field is a good parameter to
describe streamer initiation in such geometry. This figure also
shows an exponential increase with the same slope whatever
the radius rp, and this tends to prove that the initiation process
is the same in all cases. However, the plots corresponding to
different radii rp do not group together, and this shows that a
single field value does not exist to describe streamer initiation
in all cases. Conversely, plots are widely shifted: at a fixed
inception frequency, calculated fields are nearly x100 higher
with a 10m point compared to 8mm rod.
IV.
-1
r =10mm
10
-2
10
50
150
250
350
450
Crest voltage (kV)
Figure 3: Streamer inception frequency versus voltage with different electrode

radius rp (40 cm gap distance, open dots: negative streamers, full dots: positive
streamers).
INFLUENCE OF PARTICLES AND WATER
The influence of cellulose particles was studied by adding

increasing amounts of a concentrated solution to well filtered
oil. This concentrated solution was prepared with particles
obtained by de-structuring transformer pressboard. Fig. 5
shows the measured inception frequency measured with either
a sharp point (rp = 40 m) or large rod (rp = 10 mm), versus
cellulose concentration in oil with 35 ppm water.
-1
eBook for You
negative streamers was quite close and varied in a similar way

versus voltage.
10
-1

4
10
10m
0.3mm
40m
10
1mm
10
2.5mm
2
10
10
8mm
1
10
-1
10
Filtered oil
2mg/l cellulose
13mg/l cellulose
25mg/l cellulose
10
40cm 30cm 20cm 10cm 5cm -
-1
10
-2
10
100
200
300
400
500
Crest voltage (kV)
-2
10
-1
10
10
10
calculated tip field (MV/cm)
Figure 4: Streamer inception frequency versus calculated maximum field, with

different electrode radius rp and distances d (open dots: negative streamers, full
dots: positive streamers).
Figure 3 obtained with the same oil sample shows the

variation of discharge frequency when the rod (or point) radius
Figure 5: Streamer inception frequency versus voltage with two different

electrodes (rp = 40 m and 10 mm), and different cellulose quantity added to
filtered oil (40 cm gap distance, 35 ppm water).
This figure shows a quite different behavior with both

electrodes. With the large rod, a large increase of streamer
inception frequency is seen. The voltage corresponding to a
fixed frequency (for instance 1 streamer/minute) is nearly
divided by two between filtered oil, and oil with 25mg/liter
10
cellulose. The effect is very important in this experiment, since

both the cellulose and water content are large. On the other
hand, nearly no variation is seen with the sharp point. Similar
results are seen on figure 6, were the voltage corresponding to
F = 1 streamer /minute shows a very small variation with rp =
10 m, and much larger with rp = 8mm. Figure 6 also shows
the combined effects of cellulose and water. These experiments
show that solid particles (hydrated cellulose) are of primary
importance for the triggering of streamers when the field is low
(large rods), whereas the very high field created by sharp tips is
sufficient to directly induce streamers.
certainly due to the large influence of injected space charges

under ac, when the calculated tip field exceeds 1MV/cm. In
similar conditions, measurements under impulses are not
affected by space charges. Since field calculations are carried
out without space charges in figure 7, this means that the
calculated values with ac are certainly strongly overestimated
compared to the field actually present at the electrode
extremity.
Initiation field @ 10 streamer/minute (MV/cm)
1
10
voltage (kV) @ 1 streamer/minute
500
impulse voltage
400
II
10
17 ppm
300
35 ppm
200
-1
10
100
Filtered oil (17 ppm)

+ 25 mg/l cellulose
r =8mm
p
-2
g (
150
10
15
20
25
10
30
-3
10
)@
17 ppm
100
r =10m
p
0
0
b)
10
15
20
25
30
cellulose content (mg/l)
Figure 6. Voltage @ 1 streamer/minute versus cellulose and water content,

with two different electrodes: a) rp = 8 mm , b) rp = 10 m (40 cm gap).
The results are summarized on figure 7, showing the

calculated field at F =10 streamer/minute versus electrode size.
On this figure, we have also plotted initiation fields measured
under impulse voltage in the same mineral oil [5]. The same
overall tendency is observed with ac and impulses: decrease of
initiation field when the electrode size is increased. Two main
zones can be seen on figure 7.
In zone I (large rods, rp > 0.5mm), the initiation field under
ac is about half the value measured with impulses. This is quite
logical since the duration of voltage application is much longer
with ac. Adding particles further decrease the value under ac.
When the electrode size is increased, the initiation field under
ac decreases in a similar way as with impulses. With impulses,
this effect was mainly attributed to a surface effect, since
particles have a negligible influence. This shows that in this
zone, all mechanisms able to degrade the liquid properties
superpose under ac: time, pollution, electrode size. This is
consistent with observations made for a long time in practical
applications of liquids. In zone II (points, rp < 0.5mm), the
slope of the plot changes, and the calculated initiation field
becomes much higher than with impulses. This effect is
-1
10
10
10
Electrode radius of curvature (mm)

Figure 7. Calculated tip field @ 10 streamer/minute versus electrode radius,
for filtered oil and 25mg/l cellulose (17 ppm water).
V.
35 ppm
50
-2
10
CONCLUSIONS
The experiments presented here show the stochastic

character of streamer inception under ac, influenced by the
presence of particles and water. The inception probability
increases exponentially versus voltage, and no inception
threshold can be observed. Injected space charges
considerably influence experiments with divergent fields, when
the local field exceeds 1MV/cm. The results obtained in such
conditions are not relevant of practical applications such as
transformers, in which the effect of space charge is mostly
absent. If liquids are compared by measuring partial discharges
with sharp points under ac, it is impossible to know which
property of the liquid (discharge inception properties or ability
to inject space charges) is revealed by the measurement.
REFERENCES
[1]
[2]
[3]
[4]
[5]
W. R. Wilson, A. L. Streater and E. J. Tuohy, "Application of Volume

Theory of Dielectric Strength to Oil Circuit Breakers" AIEE, Trans. on
Power App., 1955, pp 677-688.
N. Giao Trinh, C. Vincent and J. Rgis, Statistical Dielectric
Degradation of Large-Volume Oil-Insulation, IEEE Trans. PAS,
Vol.101, n10, 1982, pp 3712-3721.
T.V. Oommen, E.M. Petrie, Particle Contamination Levels in Oil-Filled
Large Power Transformers IEEE Trans. PAS, vol. 102, 1983, pp 14591465.
K. Miners, Particle and Moisture Effect on Dielectric Strength of
Transformer Oil Using VDE Electrodes, IEEE Trans.PAS, vol.101,
1982, pp 751-756.
O. Lesaint and T.V. Top, "Streamer initiation in mineral oil. Part I:
Electrode surface effect under impulse voltage", IEEE Trans. on DEI,
Vol.9, pp.84-91, 2002. Part II: Influence of a metallic protrusion on a
flat electrode", IEEE Trans. on DEI, Vol.9, pp.92-96, 2002.
11
eBook for You
a)
eBook for You
12
EMTP Simulation of Induced Overvoltage in Low

Voltage System
Thinh Pham
Nhung Pham and Top V. Tran
Institute of Material Science

University of Connecticut
Storrs, CT 06269, USA
Email: thinh.pham@ims.uconn.edu
Department of Power Systems

Hanoi University of Technology
1- Dai Co Viet Street, Hanoi, Vietnam
Email: tvtop-htd@mail.hut.edu.vn
In this work, the induced overvoltage in a typical low voltage

system in rural areas of Vietnam will be investigated by
EMTP/ATP simulation. Rusk model is used to simulate the
current source affecting the low voltage system. The influence of
grounding resistance in consumer side and the size of load will be
analyzed. The discussion and results may provide useful
information in insulation coordination of low voltage system.
Keywords-low voltage system; induced lightning; Rusks theory,
ATP/EMTP simulation
I.
INTRODUCTION
The limited height of low voltage distribution system

makes it more prone to nearby lightning than direct lightning.
Induced lightning causes overvoltage on insulation, which is
usually designed with low BIL, and harms electrical and
electronic devices of such a system. In Vietnam, most of
electricity consumers locate in rural areas where distribution
network mainly uses overhead line. Furthermore, the distance
between distribution transformer and consumer in those areas
may range from several hundred meters up to kilometers. As a
result, low voltage system in the areas is especially threatened
by overvoltage due to induced lightning.
Among other theories involved in calculating induced
voltage [1-4], Rusk model [5] is widely used for its easy
handling by analytical formulations [6,7]. In this paper,
induced overvoltage in a typical TN low voltage system in rural
areas of Vietnam was simulated in the ATP/EMTP transient
program using Rusk method. The effects of grounding
resistance and load size were also analyzed and discussed.
II.
RUSK MODEL
Lightning induced voltage on the transmission line

proposed by Rusk is based on the following assumptions:
The return stroke current has the shape of step-function

with the maximum value I0, which propagates along
the lightning channel with a constant velocity .
978-1-4244-6301-5/10/$26.00 @2010 IEEE

This return stroke generates an electric field which is

given by:
e (r , z, t ) = (r , z, t )

A(r , z, t )
t
(1)
where: is the scalar potential, is the vector potential, t

is time, r and z are calculated points in cylinder coordinate.
This electric field couples with the transmission

line and generates a total induced voltage u(x,t):
h A ( x, z , t )
dz
u ( x, t ) = u ( x, t ) + z
t
0
(2)
where: h is the height of the conductor, u (x,t) is the

induced voltage in the transmission line due to the scalar
potential, Az is the vertical component of the vector
potential. Those parameters are derived from the
transmission line equations:
u ( x , t )
i ( x, t )
+ i ( x, t ).R + L
=0
t
x
(3)
i ( x, t )
u ( x, t )
( x, h, t )
+C
=C
x
t
t
(4)
where R, L and C are the corresponding resistance,

inductance and capacitance per unit length of the
transmission line; i(x,t) is the current which goes through
the line.
The induced voltage in the line could be considered as the
injecting of two current sources, the first one Ie(x,t) is due
induced voltage due to scalar potential u and the second one
Iv(x,t) is due to vector potential A. These current sources are
defined as [5,7,9]:
1 ( x, t )
I ( x, t ) =
x
e
vZ
t
c
I ( x, t ) = z
v
(5)
A( x, t )
t
(6)
where Z is the surge impedance of the line, x is the line

section to be divided for the computation.
13
eBook for You
AbstractShielded by high structure surrounding and with short

length of power line, low voltage system is seldom suffered from
direct strokes. However, this system is especially threatened by
induced voltage due to nearby strokes. The effects of induced
overvoltage may be very harmful to power quality and to low
BIL peculiar to electrical equipment in low voltage system.
As the scalar potential and vector potential are derived from

e(r,z,t) (equation 1), which is originated from the velocity of
return stroke current and the charge distribution q0 of the
lightning channel, after some mathematical manipulations
equation 5 and 6 became:
1
.x
2
r
(vt + z)2 +

P
22/0,4
Load
(7)
RL
y
280m
z0 I0
1
1
I (x, t) =
+
v
4
2
2
r
r
(vt z)2 +
(vt + z)2 +

700m
(8)
N
1,05m
P
100m
Where r is the distance between the stricken point and the

conductor, 0 and 0 are dielectric constant and magnetic
constant of the air,

with c is the speed of light.
The injected current sources are connected to the line as shown

in figure 1 [7].
Flashing point
Figure 3. A phase in low voltage system to be simulated and the

position of the flashing point (right: distribution transformer, left:
load)
A distribution transformer delta-grounded wye 22/0.4kV

was used. The neutral conductor is commonly grounded with
the neutral point of distribution transformer through a
resistance of 2. In this TN system, the load is typically
grounded through a resistance of 50 (figure 3).
Figure 1. Injected current sources in the EMTP simulation for 1 of

two conductors (phase or neutral) [7]
III.
SIMULATION
A. System configuration
1.05m
0.35m
A
B
h = 6.5m
C
N
Figure 2.
Configuration
of adistribution
typical low voltage
A typical
section
of 0.4kV
line in system
rural area of
Vietnam as shown in figure 2 was chosen to investigate. The
system consists of 3 phase conductors and 1 neutral conductor
B. Modeling method
Distribution line: A length of 700m of single
phase of the line was simulated. In order to
simulate the maximum induced overvoltage
across the load (between phase and neutral), the
latter is powered by phase A and N (figure 3).
Coupling effect from other phase conductor was
neglected for the sake of simplicity. The line is
assumed to be lossless for the worst case.
Lighting flashes to a point on the ground in the
vicinity of phase A. The line was divided into 10
sections of 70m.
Load: An inductance was used to model the
induced-lightning response of the load, as

recommended in [8] for TN configuration. The
value of this representative inductance depends on
the load size, which varies from 2H to 10H.
Distribution transformer: As the neutral of low
voltage winding of the transformer is directly
grounded, the transformer is represented by a
small inductance which is empirically determined
by [8]:

where S and U are the rated power and rated
voltage of the distribution transformer. In this
case, S=160kVA, U=380V and L=17.89H.
14
eBook for You
q
1
0
Ie (x,t ) =
+
4 Z
2
0 c
r
(vt z)2 +
which are horizontally held at 6.5m above the ideal conducting

ground plane. The velocity of return stroke current is 100m/s.
Current sources: Integrated simulation language

MODEL was used to simulate the current sources
Ie and Iv as described in section II.
C. Simulation result
1) Voltage profile along the line
Figure 6. Dependency of induced voltage across the load on

grounding resistance of the load
Figure 4. Induced voltage in phase and neutral conductors at

transformer side and load side
Figure 5. Induced voltage across the load and the transformer
Figure 7. Dependency of induced voltage across the transformer on

grounding resistance of the load
It was observed that the more the value of grounding

impedance is, the less the induced voltage stresses across the
load (figure 6). The voltage behavior is similar to the case of
the transformer, as the low grounding impedance decreased
the induced voltage on neutral phase but gave rise to the
potential difference between phase and neutral conductors.
However, the grounding resistance of the load did not have
any influence on the induced voltage across the transformer
(figure 7)
3) Influence of the size of the load
In order to investigate the influence of the load on the
induced voltage, the load size was changed from small size
(L=10H) to large size (2H) according to [8]. The
computation was performed with grounding resistance of the
load R=40 and plotted in figure 8.
However, the low grounding impedance had a serious

influence on the induced voltage across the transformer. With
I0=10kA, the peak induced voltage across the transformer
(v:TRANSP-TRANSN) is about 35kV, nearly three times
higher than that across the load (v:LOADP-LOADN) as shown
in figure 5.
2) Influence of grounding resistance of the load
The simulation was performed on the load of small size
(L=10H) with three values of grounding resistance of the
load: R=20, R=40 and R=60, which correspond to
different value of soil resistivity of rural areas.
Figure 8. Dependency of induced voltage across the load on the size
of load
15
eBook for You
A very high value of induced voltage (~50kV) was

observed at the position of the load, on the phase conductor and
neutral conductor. Low grounding impedance of the
transformer substantially decreased the induced voltage on
phase and neutral conductor. Although the transformer locates
closer to the flashing point than the load, the induced voltage
on phase and neutral conductors at transformer position
(TRANSP and TRANSN) is much lower than that at the load
position.
lightning on transformer. Overvoltage due to induced lighting

on the load depends on its size and the value of grounding
resistance. Co-ordination between these parameters may fulfill
both requirements of safety and of protection against induced
lightning.
ACKNOWLEDGMENT
This article was funded in part by a grant from the Vietnam
Education Foundation (VEF). The opinions, findings, and
conclusions stated herein are those of the authors and do not
necessarily reflect those of VEF.
The same voltage behavior across the transformer was

observed as the previous case in which different grounding
resistances of load were accounted (figure 8). The induced
voltage across the transformer was independent of the load
size, and remained at very high value due to low grounding
impedance of the transformer. Larger size of load decreased
induced voltage on it as increasing the grounding resistance of
the load (figure 9). The effect of load size, which is
comparable to that of grounding resistance of the load,
suggested that co-ordination of those two factors may provide
an optimal protection against hazard and induced lightning.
IV.
CONCLUSION
Lighting induced voltage results in harmful effects low

voltage system, on both sides of distribution transformer and of
load. The induced voltage on phase conductor is estimated to
be about 40kV for a typical value of lightning current 10kA,
this voltage value is well above the BIL of any equipment in
low voltage system. The induced voltage on neutral conductor
greatly depends on the value of grounding resistance.
Therefore, induced voltage across the equipment in question
greatly depends on the value of resistance that it is grounded
through. As the regulation of electric utilities, the grounding
impedance of distribution transformer is maintained at low
value (typically from 2 to 5) for the purpose of correct
operation, this value increases the harmful effect of induced
REFERENCES
[1]
[2]
[3]
[4]
[5]
C. Taylor, R. Satterwhite and C. Jr. Harrison, The response of a

terminated two-wire transmission line excited by a nonuniform
electromagnetic field, IEEE Trans. on Antennas and Propagation,
vol.13, no.6, November 1965.
A. Agrawal, A, H. Price and S. Gurbaxani, Transient response of
multiconductor transmission lines excited by a nonuniform
electromagnetic field, Antennas and Propagation Society International
Symposium, vol.18, June 1980.
F. Rachidi, Formulation of the field-to-transmission line coupling
equations in terms of magnetic excitation field, IEEE Transaction of
Electromagnetic Compatibility, Vol. 35.,no. 3, August 1993.
P. Chowduri and E.T.B. Gross, Voltage surges induced on overhead
lines by lightning strokes, Proc. IEE, Vol. 114, no.12, December 1967.
S. Rusk, Induced lightning overvoltages on power transmission lines
with special reference to the over-voltage protection of low voltage

networks, Royal Institute of Technology, PhD Thesis, Stockhom 1957.
[6]
[7]
[8]
[9]
H. K. Hoidalen, Calculation of lightning overvoltages using

MODELS, International Conference on Power Systems Transients
(IPST), Budapest, June 20-24, 1999.
A. E. A. Araujo, J. O. S. Paulino, J. P. Silva, H. W. Dommel,
Calculation of lightning induced voltages with Rusks method in
EMTP. Part I: Comparison with measurements and Agrawals coupling
model, Electrical Power System Research , vol. 60, 2001.
H. K. Hoidalen, Lightning induced voltages systems and its
dependency on overhead line termination, Internationl Conference on
Lightning Protection (ICLP), Birmingham, September 14-18, 1998.
J. G. Anderson and T. A. Short, Algorithms for calculation of lightning
induced voltages on distribution lines, IEEE Trans. on Power Delivery,
vol. 8, no. 3, July 1993.
16
eBook for You
Figure 9. Dependency of induced voltage across the transformer on

the size of load
Proceedings of 2008 International Symposium on Electrical Insulating

Materials, September 7-11, 2008, Yokkaichi, Mie, Japan
EB-5
Electrical Field Behavior of Transmission Line Insulators in Polluted Area

T. Pham Hong * and Tran Van Top
Department of Power System, Faculty of Electrical Engineering, Hanoi University of Technology (HUT)
1, Dai Co Viet Street, Hanoi, Vietnam
* E-mail : thinhph-htd@mail.hut.edu.vn
INTRODUCTION
Although the use of silicone rubber composite
insulators has been increased significantly in recent
years, porcelain and glass insulators are still
manufactured and remain predominant in distribution
and transmission lines of Vietnam. Except for
hydrophilic property, porcelain and glass insulators are
widely used because they offer many advantages: low
cost, flexible maintenance and high strength. When
energized in polluted area such as coal industry zone
and coastal area, the insulators are easily contaminated
[1], dry band will be formed and leading to flashover
[2].
Other
3%
Pollution
26%
Other
49%
Lightning
71%
Pollution
33%
Lightning
18%
The results will be a good indication for designing

insulator, especially for polluted areas.
INSULATOR TO BE MODELED
Porcelain insulator type -425 in this study is widely
used 35kV distribution networks, 110kV and 220kV
transmission lines in Vietnam. The number of unit per
insulator string depends on rating voltages and
operating environment, e.g: the number of units per
string of 35kV distribution network is varied from 2 to
4 while 7 or more are usually used in 110kV
transmission networks. Figure 2 shows the detailed
geometry dimension of a porcelain insulator. The cap
and pin are made of steel and they are embedded in
mortar layers in order to fix with porcelain shell. The
shell is made porcelain with a relative permittivity of 6
and a conductivity of 2.10-13S.m-1. The creepage length
of porcelain shell is 280mm. Commercially available FE
software ANSYS are used for calculation. The
modeling is carried out with a typical insulator string
used in a typical 110kV transmission line, which
consists of 7 insulators without corona ring. The unit is
numbered from the line to ground with unit 1
corresponding to that close to live-line-end and the unit
7 is close to ground-fitting-end. Static analysis are
performed at steady state condition at f=50Hz. An AC
voltage of 110kV is applied to the whole string.
eBook for You
Abstract: Finite element method (FEM) was used to

study the electric field distribution along creepage path
of a cap and pin porcelain insulator string of
transmission line. The effect of pollution layer
conductivity and dry band width is considered in order
to investigate electric field behaviors of insulator using
in polluted areas.
Figure 1: Service interruption of some 110 kV transmission

lines in Vietnam due to pollution flashover in 2004: in Quang
Ninh province (left side), in 12 coastal provinces of the
central Vietnam (right side)
Exploring 7 million tones of coal per year, Quang Ninha northern province of Vietnam faces not only
environment problem arising from coal dust but also the
outages of power distribution and transmission lines.
Field data has recorded about 20% outages in 110kV
transmission line of Quang Ninh province is due to
flashover [3]. Power Company 3 which manages
transmission network of 12 coastal provinces in the
central Vietnam has reported that 33% of service
interruption is also due to flashover [4] (Figure 1).
The paper presents the results of finite element (FE)
calculation of the electrical field distribution along a
string of cap and pin porcelain insulator using in current
transmission lines in Vietnam. Pollution level and dry
band width are varied in order to investigate their
effects on field distribution along the creepage path.
Figure 2: Geometry dimension of a typical transmission

line insulator used in the calculation
The conductivity of pollution layers are selected in
accordance with IEC 60815. In this study, the
calculation is performed with pollution conductivity
varying from =8S to =20S which correspond to
medium and heavy pollution levels. Morever, the effect
of mortar layers is neglected in the model.
RESULTS AND DISCUSSION
Potential and electric field distribution in clean
530
17
insulator string
Figure 5: Electric field stress along creepage path of unit 1 in

clean string
eBook for You
Figure 3: Equipotential distribution along a clean string
Figure 4: Electric field distribution along a clean string

The calculation is firstly performed with a clean
insulator string. Figure 3 shows the equipotential lines
for the clean string. Potential is distributed along the
string in accordance with the capacitance of each unit
and the stray capacitance to line and to ground. The
potential distribution is found to be maximum in the
unit 1 (near the live-end-fitting) with 18% of total
voltage applied on the string, whereas 13.5% is found in
the unit 7 (near ground-end-fitting) and minimum
potential distribution is found in the unit 5 (12%). These
behaviours are well correlated with the results of
practical field measurements on a test string using a
horizontal sphere gap in the Laboratory for High
Voltage Engineering and Electrotechnical Material at
Department of Power System.

clean string
values toward live-line-fitting. Figure 5 depicted the

field strength magnitude along the creepage path of the
unit 1. In reality, field strength along the creepage path
of each unit follows the same trend, high values appear
in the triple junction region (air-cap-porcelain),
live-end-fitting, and near the sections with small radius
of curvature. In the unit 1, the stress in the triple
junction and live-line-fitting take the same value of
2.2kV/mm. However, the highest stress of the string is
reached in the air-cap-porcelain region of the unit 2
(Figure 6). It is observed that the stress in this region is
slitghtly higher that of unit 1 with 2.5kV/mm, but this
value is still beyond that needed for corona discharge.
As shown in the figure 4, electric field intensifies as far

from the ground-fitting-end and reaches the highest
531
18
end and pin-fitting regions. From the point of view of

field distribution, the presence of pollution layer plays a
positive role in linearizing the field along the insulator
string.
Influence of pollution layer
Figure 7: Modeling of electric field in presence of pollution

layers next to metal cap
Figure 9: Electric field strength magnitude along creepage

path of unit 2 in presence of pollution layer of 14.5PS
Influence of pollution layer conductivity
eBook for You
A pollution layer with 14.5PS and 25% of upper surface

length (27.5mm) is deposited on each unit (Figure 7). In
this case, the stress distribution is predomenantly
controlled by the conductivity of pollution layer instead
of capacitance distribution like in clean surface case [6].
In this case the electrical stress are put from triple
junction region to the pollution layer ends in every unit,
the highest stress is transferred from the triple
junction of unit 2 to unit 1 (Figure 8) and remains the
same value as in the clean case (|2.5kV/mm). However
the region in which the highest stress appears is
displayed toward the pollution layer end and this value
is still well beyond that can lead to partial discharge in
the air.
Because the presence of pollution layer had negligeble

effect on field distribution in others unit, only the unit 1
is analysed to study the influence of pollution layer
conductivity. Using the same geometry of pollution
layer as the previous case (with 25% of upper surface
length), the influence of pollution conductivity is
studied by performing calculation with different
pollution levels: =8S, =14.5 S and =20S. The
electric field magnitudes at air-cap-porcelain (cap),
pollution layer end (PL end) and live-end-fitting (pin)
are plotted in the figure 10. It is clear that the stress in
cap region decreases with the conductivity, while the
electric field in pollution layer ends and live-end-fitting
regions slightly increases versus the conductivity.
However, the stress in these regions is still inferior to
3kV/mm. As a result, the pollution flashover could not
occur even in presence of heavy pollution level.
3000
cap
For other units, the presence of pollution layer reduced

the maximum stress near the cap and tends to linearize
the field distribution along the creepage path. As an
exemple, the magnitude of electric field strength along
creepage path in the unit 2 is depicted in the figure 9. It
is observed that the electric field is more evenly
distributed along the creepage path in comparison with
the clean case (Figure 6). The highest stress in the unit 2
is reduced from 2.5kV/mm to 1.2kV/mm and is
transferred from air-cap-porcelain to the pollution layer
PL end
pin
2250
E(V/mm)

presence of pollution layer of 14.5PS
1500
750
0
0
14.5
20
conductivity (S)
Figure 10: Influence of pollution layer conductivity on

electric field stress
Influence of dry band width
532
19
As mentionned in introduction paragraph, the flashover

in polluted insulator are initiated by the formation of
one (or more) dry band. In order to study the influence
of dry band on electric field distribution, the calculation
has been performed with a dry band inside pollution
layer of 8PS. A pollution layer with 50% of upper
surface length is deposited on each unit, a dry band
width ranging from 0 to 1mm is created in the middle of
the pollution layer. The stress distribution along
creepage path of unit 1 is depicted in the figure 11 with
a dry band of 0.25mm. It is clear that the maximum
strength of 6.5kV/mm appears near the dry band and
increases threefold in comparison with clean case. The
highest stress value decreases with the dry band length,
but the stress at dry band are always dominant in
comparison with other regions such as pin-fitting or
pollution layer ends (Figure 12). This indicates that
when a dry band forms inside the pollution layer, partial
discharge could begin from these points and leads to
flashover.
curvature regions. These behaviors are well correlated

with the results predicted by the simulation in the
previous paragraph.
Figure 13: Flashover process from live-end-fitting and

small curvature of a polluted insulator (from left to right)
CONCLUSION
eBook for You
The presence of a pollution layer on upper surface of

insulator strongly modified the field distribution along
creepage path. With a homogenous pollution layer
deposited on each unit, the live-fitting-end region of
unit 1 submits the highest stress, but the magnitude of
electric field is similar to that in clean case. In presence
of a dry band in the middle of pollution layer, the stress
reaches maximum value in the dry band and exceeds the
breakdown strength of air. Flashover can occur from
this points and were observed by field measurements.
ACKNOWLEDGMENT
Center for Development and Application of Software
for Industry (DASI) at HUT is gratefully acknowledged
for its help during this study.
Figure 11: Electric field stress along creepage path of unit 1

in presence of a dry band of 0.25mm in the middle of
pollution layer
REFERENCES
8000
cap
bandgap
PL end
pin
E(V/mm)
6000
4000
2000
0
0
0.25
0.5
bandgap length (mm)
Figure 12: Electric field stress in different position of

creepage path versus dry band length.
Some pictures taken during performing measurement in

the laboratory are shown in the figure 13. Increasing the
applied voltage will lead to flashover which initiates
from the dry band, live-end-fitting and the small
[1] J. S. T. Looms, Insulators for high voltages, Peter

Peregrinus Ltd, 1988
[2] David D. Jolly, "Contamination Flashover Theory and
Insulator Design", Journal of The Franklin Institute,Vol. 294,
No.6, December 1972.
[3] Do Khanh Ninh, Influence of polluted environment on
the performance of glass insulators using in 110kV network of
Quang Ninh province, Master thesis of Hanoi University of
Technology, 2006
[4] Le Thanh Giang and Nguyen Quoc Viet, Modeling of
field distribution along a set of insulators, Conference on
student research, Hanoi University of Technology, 2007
[5] Vosloo W. L. and Holtzhausen J. P., The electric field of
polluted insulators, Africon
533
20
eBook for You
21
eBook for You
22
eBook for You
23
eBook for You
24
eBook for You
25
eBook for You
26
eBook for You
27
eBook for You
28
eBook for You
29
eBook for You
30
eBook for You
31
eBook for You
32
eBook for You
33
eBook for You
34
+RDQJ 1JRF 1KDQ
3KDP +RQJ 7KLQK DQG 1JX\HQ 7KL /DQ +XRQJ
)DFXOW\ RI $XWRPDWLRQ 7HFKQRORJ\ (OHFWULF 3RZHU

8QLYHUVLW\
+RDQJ 4XRF 9LHW+DQRL9LHWQDP
(PDLO QKDQKQ#HSXHGXYQ
)DFXOW\ RI (OHFWULFDO (QJLQHHULQJ +DQRL 8QLYHUVLW\ RI

7HFKQRORJ\
'DL &R 9LHW+DQRL9LHWQDP
(PDLO WKLQKSKKWG#PDLOKXWHGXYQ
$EVWUDFW1RZDGD\V WKH HOHFWURPDJQHWLF LQWHUIHUHQFH SUREOHP

EHFRPHV PRUH LPSRUWDQW WR SURWHFW WKH HOHFWULF HOHFWURQLF
GHYLFHV ,Q WKLV SDSHU WKH HOHFWURPDJQHWLF FKDUDFWHULVDWLRQ RI
3DQL38 LQ PXOWLOD\HUHG VWUXFWXUH ZDV VWXGLHG LQ RUGHU WR
RSWLPL]H WKH VKLHOGLQJ HIIHFWLYHQHVV DW WKH PLFURZDYH IUHTXHQF\
,Q WKH WKUHHOD\HUHG VWUXFWXUH WKH LQVXODWLQJ OD\HU LQFUHDVHV QRW
RQO\ WKH PHFKDQLFDO SURSHUWLHV EXW DOVR LPSURYH WKH VKLHOGLQJ
HIIHFWLYHQHVV
.H\ZRUGV &RQGXFWLQJ FRPSRVLWH VKLHOGLQJ HIIHFWLYHQHVV
PLFURZDYH
,
,1752'8&7,21
1RZDGD\V WKH HOHFWURPDJQHWLF LQWHUIHUHQFH (0, EHFRPHV

PRUH VLJQLILFDQW GXH WR WKH SUROLIHUDWLRQ RI FRPPHUFLDO
PLOLWDU\ VFLHQWLILF HOHFWULFDO GHYLFHV DQG HTXLSPHQWV LQ KLJK
IUHTXHQFLHV 7R SURWHFW DJDLQVW WKH LQFRPLQJ DQG SRWHQWLDOO\
GLVWXUELQJ UDGLDWLRQ SHQHWUDWLQJ LQWR WKH HTXLSPHQW HOHFWURQLF
GHYLFHV PXVW EH VKLHOGHG
7KH LQWHUDFWLRQ PHFKDQLVP RI DQ HOHFWURPDJQHWLF ZDYH ZLWK
D PDWHULDO FDQ EH GLYLGHG LQ WKUHH PHFKDQLVPV UHIOHFWLRQ
DEVRUSWLRQ DQG PXOWLSOH UHIOHFWLRQV LQVLGH WKH PDWHULDO >@ 7KH
ILUVW RQH UHIOHFWLRQ GHSHQGV RQ WKH SHUPLWWLYLW\ DQG
FRQGXFWLYLW\ RI WKH PDWHULDO 7KH HIIHFW RI WKLV PHFKDQLVP
LQFUHDVHV ZLWK SHUPLWWLYLW\ DQG FRQGXFWLYLW\ RI VKLHOG 7KH
VHFRQG RQH RI (0, VKLHOGLQJ DEVRUSWLRQ UHTXLUHV WKH H[LVWHQFH
RI PRELOH FKDUJH FDUULHUV HOHFWURQV RU KROHV ZKLFK LQWHUDFW
ZLWK WKH HOHFWURPDJQHWLF UDGLDWLRQ $QG WKH WKLUG RQH RI (0, LV
WKH ZDYH DWWHQXDWLRQ GXH WR PXOWLSOH UHIOHFWLRQV 7KLV
PHFKDQLVP GHSHQGV RQ SK\VLFDO SURSHUWLHV DQG JHRPHWULF
VWUXFWXUH RI VKLHOGLQJ PDWHULDO
+DYLQJ KLJK FRQGXFWLYLW\ PHWDO LV XVXDOO\ XVHG DV WKH
VKLHOGLQJ PDWHULDO ,Q WKH UHFHQW \HDUV QHZ PDWHULDOV VXFK DV
LQWULQVLFDOO\ FRQGXFWLQJ SRO\PHUV ,&3 ZLWK KLJK
FRQGXFWLYLWLHV DQG OLJKWZHLJKW VWUXFWXUHV DSSHDU DV DQ
DOWHUQDWLYH WR PHWDO $PRQJ WKH ,&3 3RO\DQLOLQH 3$QL KDV
ORWV RI SRWHQWLDO DSSOLFDWLRQV LQ HOHFWURQLF DQG PLFURHOHFWURQLF
LQGXVWU\ VXFK DV RSWRHOHFWURQLF GHYLFHV VHQVRUV HOHFWURVWDWLF
GLVFKDUJH OD\HUV SULQWHG FLUFXLW ERDUG >@ DQG
HOHFWURPDJQHWLF LQWHUIHUHQFH VKLHOGLQJ >@WKDQNV WR WKHLU
JRRG HOHFWULF SURSHUWLHV DQG HQYLURQPHQWDO VWDELOLW\
1-4244-2426-9/08/$20.00 2008 IEEE
/HH HW DO >@ SUHVHQWHG WKH (0, VKLHOGLQJ HIIHFWLYHQHVV RI

WKH PL[WXUHV RI 3$QL DQG FRQGXFWLQJ SRZGHUV OLNH VLOYHU
JUDSKLWH DQG FDUERQ EODFN LQ WKH IUHTXHQF\ UDQJH IURP 0+]
WR *+] $V DQ H[HPSOH WKH VKLHOGLQJ HIIHFWLYHQHVV RI WKH
PL[WXUH EHWZHHQ GRSHG 3$QL DQG $J SRZGHU LV DERXW G% DW
URRP WHPSHUDWXUH 7KH DXWKRUV VKRZHG WKHRUHWLFDOO\ WKDW WKH
VKLHOGLQJ HIIHFWLYHQHVV FRXOG EH LPSURYHG LQ WKUHHOD\HUHG
VWUXFWXUH 0DNHOD HW DO >@ PHDVXUHG WKH VKLHOGLQJ
HIIHFWLYHQHVV RI WZROD\HU VWUXFWXUH FRPSRVHG RI 3$QL&6$
DQG IRXQG D 6( RI G% DW *+] LQ WKH IDUILHOG UHJLPH ,Q
WKLV SDSHU WKH VKLHOGLQJ WKHRU\ RI PXOWLOD\HU VWUXFWXUH LV
VWXGLHG DQG WKHQ WKH UROH RI HDFK OD\HU LQ WKUHH OD\HUHG LV DOVR
LQYHVWLJDWHG LQ RUGHU WR FRQFHSW DQ RSWLPL]HG (0, VKLHOGLQJ
PDWHULDO
,,
(;3(5,0(17$/
$ &RQGXFWLQJ FRPSRVLWH
$ )UHHVWDQGLQJ 3DQL38 FRPSRVLWH
7KH IUHHVWDQGLQJ 3DQL38 ILOPV ZHUH SUHSDUHG DV GHVFULEHG
LQ >@ 7KH SHUPLWWLYLW\ DQG FRQGXFWLYLW\ RI WKH VDPSOHV ZHUH
PHDVXUHG E\ XVLQJ WKH RSHQ HQGHG FRD[LDO SUREH IRU GLIIHUHQW
ZHLJKW UDWLR RI 3DQL 3DQL38 3DQL38 DQG 3DQL38
ILOPV >@ (OHFWULFDO SURSHUWLHV DW *+] RI WKH FRPSRVLWHV ZHUH
SUHVHQWHG LQ WKH WDEOH ,
7$%/( ,
&+$5$&7(5,67,&6 2) 7+( )5((67$1',1* ),/06 2) 3$1,38

$7 *+=
0DWHULDO
3DQL38
3DQL38
3DQL38
3DQL38
3DQL38
3DQL38
3DQL38
GP

6P

$ 7KUHHOD\HUHG 3DQL38 FRPSRVLWH

7KH WKUHHOD\HUHG FRQGXFWLQJ FRPSRVLWH ZDV VWUXFWXUL]HG
E\ 3DQL38 DV WKH ILUVW DQG WKH WKLUG OD\HU DQG WKH .DSWRQ ILOP
DV WKH PLGGOH OD\HU ILJXUH 7KH FRQGXFWLYLW\ RI 3DQL38
366
35
eBook for You
(OHFWURPDJQHWLF FKDUDFWHULVDWLRQ RI 3DQL38 LQ PXOWL

OD\HUHG VWUXFWXUH DSSOLFDWLRQ IRU (0, SURWHFWLRQ DW
PLFURZDYH IUHTXHQF\
YDULHV IURP LQVXODWLQJ VWDWH 6P WR FRQGXFWLQJ VWDWH

6P 7KH .DSWRQ VOLGH LV LQVXODWLQJ PDWHULDO ZLWK RI
UHODWLYH SHUPLWWLYLW\
.DSWRQ
3$QL38
ILOP
[0 ] = >0 @>0 @> 0 1 @ =
+
(
3
6( = ORJ W = ORJ W = ORJ W
+L
(L
3L
:KHUH (L+L DQG (W+W DUH WKH LQFLGHQW DQG WUDQVPLWWHG

ILHOG VWUHQJWK
$ WKHRUHWLFDO PRGHO EDVHG RQ SODQZDYH DQG WUDQVPLWWHG
ZDYH PDWUL[ ILHOG LV SURSRVHG WR VWXG\ WKH IDUILHOG RI
PXOWLOD\HU PDWHULDOV 7KH UHIOHFWLRQ DQG WKH WUDQVPLVVLRQ RI D
QRUPDO (0 ZDYH RQ WKH 1OD\HU VWUXFWXUH ZHUH SUHVHQWHG LQ WKH
ILJ &RQVLGHU WKDW HDFK OD\HU LV KRPRJHQHRXV DQG LVRWURSLF
WKH HOHFWURPDJQHWLF SDUDPHWHUV RI LOD\HU DUH QRWHG L
SHUPHDELOLW\ L FRQGXFWLYLW\ L SHUPLWWLYLW\ DQG GL
WKLFNQHVV 7KH LQWULQVLF LPSHGDQFH RI LWKOD\HU =L GHSHQGV RQ
WKH IUHTXHQF\ RI LQFLGHQW ZDYHDQG JLYHQ E\
[
\

Q
Q
(7
(5
]
G
G
G1
)LJXUH 7KH UHIOHFWLRQ DQG WKH WUDQVPLVVLRQ RI D QRUPDO (0 ZDYH RQ WKH

1OD\HU VWUXFWXUH
7KH FKDUDFWHULVWLF PDWUL[ SUHVHQWHG E\ .RQJ >@ ZDV XVHG

WR FDOFXODWH WKH FRHIILFLHQWV RI UHIOHFWLRQ DQG WUDQVPLVVLRQ RI DQ
HOHFWURPDJQHWLF ZDYH WKURXJK D PXOWLOD\HUV VWUXFWXUH
7KHUHIRUH IRU LWKOD\HU ZH KDYH
0
0
1OD\HUV VWUXFWXUH LV LQ FRQWDFW ZLWK WZR VHPLLQILQLWH DLU

PHGLD VR = =1
% 6KLHOGLQJ HIIHFWLYHQHVV WKHRU\

7KH HOHFWURPDJQHWLF VKLHOGLQJ HIIHFWLYHQHVV 6( RI D
PDWHULDO LV GHILQHG E\ WKH UDWLR RI WKH WUDQVPLWWHG SRZHU 3W
WKURXJK WKH PDWHULDO WR WKH LQFLGHQW SRZHU 3L RI DQ
HOHFWURPDJQHWLF ZDYH >@ ,Q JHQHUDO 6( LV JLYHQ LQ GHFLEHO
G%

0
0
)LJXUH 7KUHHOD\HUHG FRQGXFWLQJ FRPSRVLWH
(,
7KH FKDUDFWHULVWLF PDWUL[ RI WKH ZKROH VWUXFWXUH SUHVHQWHG
G
L
=L =
M
+

L
L
NL =
7KH FRHIILFLHQWV RI UHIOHFWLRQ 5 DQG WUDQVPLVVLRQ 7 ZHUH

FDOFXODWHG DFFRUGLQJ WR 1DLVKDGKDP >@
5=
(0 = 0 ) = (0 0 = )
(0 = 0 ) + = (0 0 = )
7=
[0 (0= 0 ) + 0 (0 0= )]

(0= 0 ) + =(0 0= )
)URP HTXDWLRQ FRQVLGHU WKDW WKH LQFLGHQW ILHOG HTXDO RQH

XQLW VR WKH 6( LV JLYHQ E\

6( = ORJ 7
& 6KLHOGLQJ HIIHFWLYHQHVV PHDVXUHPHQW LQ IDUILHOG

7KH WHFKQLTXH PHDVXUHPHQW RI WKH (0, VKLHOGLQJ
HIIHFWLYHQHVV LQ WKH IDUILHOG RI D FRQGXFWLQJ PDWHULDO FRQVLVWV
RI SODFLQJ WKH WHVW PDWHULDO EHWZHHQ D VRXUFH RI SODQH ZDYHV
DQG D GHWHFWRU 7KH VKLHOGLQJ HIIHFWLYHQHVV FDQ EH PHDVXUHG E\
XVLQJ WKH 7(0 FHOO 0+] WR *+] WKH UHFWDQJXODU
ZDYHJXLGHV *+] WR *+] DQG DQWHQQD WR *+]>@
7ZR VHFWLRQV RI WKH PHDVXULQJ FHOO ZHUH FRQQHFWHG WR WKH
YHFWRU QHWZRUN DQDO\VHU 91$ YLD WZR FRD[LDOWRZDYHJXLGH
DGDSWHUV %HIRUH SHUIRUPLQJ PHDVXUHPHQWV WKH 91$ ZDV
FDOLEUDWHG E\ XVLQJ WKH FDOLEUDWLRQ NLWV ZLWK WKH IXOO WZRSRUW
FDOLEUDWLRQ PHWKRG SURYLGHG E\ +HZOHWW 3DFNDUG &RUSRUDWLRQV
7KH VKLHOGLQJ HIIHFWLYHQHVV LV H[WUDFWHG IURP PRGXOHV RI WKH
WUDQVPLWWHG FRHIILFLHQWV 6
' 5HVXOWV DQG GLVFXVVLRQV
,Q WKH SUHYLRXV ZRUN >@ WKH VKLHOGLQJ HIIHFWLYHQHVV PRGHO
ZDV YDOLGDWHG E\ PRGHOLQJ WKH VKLHOGLQJ HIIHFWLYHQHVV RI
3DQL38 IUHHVWDQGLQJ DQG WKUHHOD\HUHG ILOPV ,Q WKLV SDSHU
VKLHOGLQJ HIIHFWLYHQHVV EHKDYLRXUV RI WKH PDWHULDO YHUVXV
WKLFNQHVV DQG FRQGXFWLYLW\ DUH LQYHVWLJDWHG LQ RUGHU WR ZRUN RXW
DQ RSWLPDO VKLHOGLQJ PDWHULDO LQ IDUILHOG
7KH HIIHFW RI WKLFNQHVV RQ VKLHOGLQJ HIIHFWLYHQHVV RI
3DQL38 3DQL38 DQG 3DQL38 DW *+] ZDV VKRZQ LQ
ILJXUH ,W LV REYLRXV WKDW DW WKLV IUHTXHQF\ WKH VKLHOGLQJ
HIIHFWLYHQHVV LQFUHDVHV ZLWK WKH FRQGXFWLYLW\ :LWK WKH VPDOO
DPRXQWV RI 3DQL FRUUHVSRQGLQJ WR DQG RI ZHLJKW UDWLR
D UHVRQDQFH RI VKLHOGLQJ HIIHFWLYHQHVV LV YLVXDOL]HG DW WKH
36
eBook for You
G
(3)
[( L
+ L M ) F ] DQG LV

WKH ZDYHOHQJWK LQ WKH DLU DQG F LV WKH FXWRII ZDYHOHQJWK RI WKH
7( PRGH LQ WKH ZDYHJXLGH
:KHUH
3$QL38
ILOP
G
M= L VLQN L G L
FRV N L G L
M
VLQ

FRV NL G L
N
G
L L
=
L
[0 L ] =
WKLFNQHVV RI PP ZLWK WKH YDOXH RI G% 2WKHUZLVH WKH

VKLHOGLQJ HIIHFWLYHQHVV RI WKH IUHHVWDQGLQJ ILOP FRQWDLQQLQJ WKH
KLJKHVW UDWLR RI 3DQL LQFUHDVHV ZLWK WKH WKLFNQHVV LQ TXDVL
OLQHDU ZD\ LQ ORJDULWKLPLF VFDOH DQG WKLV EHKDYLRU LV IRXQG WR
EH VLPLODU WR WKDW RI FRQGXFWLQJ PDWHULDO
%\ FKDQJLQJ WKH FRQGXFWLYLW\ RI WKH VDPSOH RI PP

WKLFN DW *+] WKH VKLHOGLQJ HIIHFWLYHQHVV RI WKH PDWHULDO
YHUVXV WKH FRQGXFWLYLW\ LV SORWWHG LQ WKH ILJXUH 7KLV FXUYH
ZLOO EH XVHG WR HODERUDWH FRQGXFWLQJ PDWHULDO KDYLQJ GHVLUHG
VKLHOGLQJ HIIHFWLYHQHVV E\ NQRZLQJ WKHLU FRQGXFWLYLW\

$W ) *+]

3$QL38

HSV

6( G%
VLJ6P
WKH YDOXH RI SHUPLWWLYLW\ LV VR VPDOO WKDW LW FDQ EH QHJOHFWHG LQ

WKH LQWULQVLF LPSHGDQFH H[SUHVVLRQ 7KLV REVHUYDWLRQ LV LQ
DFFRUGDQFH ZLWK WKDW ZDV UHFRUGHG LQ >@ ZKHUHLQ WKH
SHUPLWWLYLW\ FDQ EH QHJOLJHDEOH LQ VKLHOGLQJ HIIHFWLYHQHVV
FDOFDODWLRQ LI WKH VWDWLF FRQGXFWLYLW\ RI PDWHULDO LV VXSHULRU WR
6P

PP

6 ( G%

G PP
)LJXUH 6KLHOGLQJ HIIHFWLYHQHVV RI 3DQL38 ILOPV YHUVXV WKLFNQHVV DW *+]

ZLWK GLIIHUHQW FRQFHQWUDWLRQV

G PP I *+]

6LJP D 6P

HSV
HSV

6 ( G %

HSV
0HDVXUHPHQW

0RGHOLQJ

G P
)LJXUH 6KLHOGLQJ HIIHFWLYHQHVV RI 3DQL38 YHUVXV WKLFNQHVV DW *+]

ZLWK GLIIHUHQW SHUPLWWLYLWLHV
)LJXUH GHSLFWHG WKH UHVXOWV RI PHDVXUHPHQWV ZKLFK ZHUH

SHUIRUPHG RQ 3DQL38 6P RI FRQGXFWLYLW\ ZLWK WKH
UHODWLYH SHUPLWWLYLW\ RI DQG UHVSHFWLYHO\ $V WKH
VKLHOGLQJ HIIHFWLYHQHVV NHHS DOPRVW FRQVWDQW LQ WKH PLFURZDYH
EDQG WKH GDWD DW *+] ZDV XVHG WR DQDO\VH 7KH PRGHOLQJ
FXUYHV DFFRUGLQJ WR HTXDWLRQ ZHUH DOVR LOOXVWUDWHG LQ WKH
ILJXUH DQG WKH\ GHVFULEHG YHU\ ZHOO WKH EHKDYLRXUV RI WKH
PDWHULDO ,W LV REVHUYHG WKDW WKH VKLHOGLQJ HIIHFWLYHQHVV RI VXFK
D KLJK FRQGXFWLYLW\ PDWHULDO LV LQGHSHQGHQW RI WKHLU SHUPLWLYLW\
WKH FXUYH FRUUHVSRQGLQJ WR WKH SHUPLWWLYLW\ RI VOLJKWO\
GHYLDWHV IURP WKH RWKHUV DW WKH WKLFNQHVV DERYH P 7KLV
EHKDYLRXU LV GXH WR WKH LQVLJQLILFDQW YDOXH RI SHUPLWWLYLW\ LQ
FRPSDULVRQ ZLWK WKH FRQGXFWLYLW\ ,Q RWKHU ZRUGV
7DEOH ,, VXPPDUL]HV PHFKDQLFDO DQG HOHFWULFDO SURSHUWLHV RI

WZROD\HU DQG WKUHHOD\HU PDWHULDO ZLWK WKH VDPH WRWDO WKLFNQHVV
RI FRQGXFWLQJ OD\HUV P 7KH PDWHULDO ZDV VWUXFWXUDOL]HG
DV GHVFULEHG LQ $ DQG $ 'DWD LQ WKH WDEOH ZHUH XVHG WR
VLPXODWH WKH VKLHOGLQJ HIIHFWLYHQHVV RI WKH PDWHULDO 7KH ILJXUH
FRPSDUHV WKH VKLHOGLQJ HIIHFWLYHQHVV EHWZHHQ WZROD\HUHG
DQG WKUHHOD\HUHG VWUXFWXUHV YHUVXV IUHTXHQF\ ,W LV REVHUYHG
WKDW WKH WKUHHOD\HUHG VWUXFWXUH VKRZV PRUH LQWHUHVWLQJ
EHKDYLRXU DW KLJK IUHTXHQF\ WKDQ WKH WZROD\HU RQH HVSHFLDOO\
DW YHU\ KLJK IUHTXHQF\ $OWKRXJK WKH VWUXFWXUH GLG QRW
LQIOXHQFH RQ WKH VKLHOGLQJ HIIHFWLYHQHVV DW *+] EXW G% RI
GLIIHUHQFH ZDV REVHUYHG EHWZHHQ WZR VWUXFWXUHV DW *+]
6LQFH .DSWRQ SRVVHVVHV YHU\ JRRG PHFKDQLFDO SURSHUWLHV WKH
WKUHHOD\HUHG VWUXFWXUH ZLOO ILQG PRUH DSSOLFDWLRQ LQ (0,
VKLHOGLQJ WKDQ WKH WZROD\HUHG RQH
7$%/( ,, 0(&+$1,&$/ $1' (/(&75,&$/ 3523(57,(6 2) 7:2/$<(5 $1'
7+5((/$<(5 0$7(5,$/
7ZR OD\HU
7KUHHOD\HU
VW OD\HU
6P
XP
6P
XP
QG OD\HU
6P
G XP
.DSWRQ
G XP
UG OD\HU
6P
G XP
37
eBook for You
)LJXUH 6KLHOGLQJ HIIHFWLYHQHVV YHUVXV FRQGXFWLYLW\ DW *+] RI WKH

WKLFNQHVV G PP

,9
46
9/,5+87
,QVWLWXWLRQDO
3URJUDP
LV
JUDWHIXOO\
DFNQRZOHGJHG IRU WKH ILQDQFLDO VXSSRUW SURYLGHG GXULQJ WKH
FRXUVH RI WKLV VWXG\
44
Two layer
43
Three layer
42
41
40
39
0
2E+09
4E+09
6E+09
8E+09
1E+10 1.2E+10
Frequency (Hz)
)LJXUH 6KLHOGLQJ HIIHFWLYHQHVV WZROD\HUHG DQG WKUHHOD\HUHG VWUXFWXUH

YHUVXV IUHTXHQF\
,,, &21&/86,21
,Q WKLV ZRUN QXPHURXV SDUDPHWHUV ZHUH FKDQJHG LQ RUGHU WR
REWDLQ DQ RSWLPDO VKLHOGLQJ PDWHULDO 7KH VKLHOGLQJ
HIIHFWLYHQHVV LQFUHDVHV LQ H[SRQHQWLDO VFDOH ZLWK WKH WKLFNQHVV
RI PDWHULDO GHVSLWH D UHVRQDQFH SHDN DSSHDUV DW ORZ
FRQFHQWUDWLRQ RI 3DQL 3HUPLWWLYLW\ KDV QHJOLJLEOH HIIHFW RQ
VKLHOGLQJ HIIHFWLYHQHVV LQ WKH ZKROH UDQJH RI WKLFNQHVV ,Q WKH
JRDO RI HODERUDWLQJ D FRQGXFWLQJ PDWHULDO ZLWK GHVLUHG
VKLHOGLQJ HIIHFWLYHQHVV EDVHG RQ 3DQL FRQFHQWUDWLRQ D FXUYH
ZKLFK LOOXVWUDWHV WKH GHSHQGHQFH RI VKLHOGLQJ WR WKH
FRQGXFWLYLW\ ZDV EXLOW 7KH WKUHHOD\HUHG FRQGXFWLQJ VWUXFWXUH
ZDV IRXQG WR KDYH EHWWHU HOHFWULFDO DQG PHFKDQLFDO SURSHUWLHV
WKDQ WKH WZR OD\HUHG RQH HVSHFLDOO\ DW KLJK IUHTXHQF\ %HFDXVH
RI YHU\ WKLQ GLPHQVLRQ WKH PDWHULDO FDQ EH FRPSDWLEOH ZLWK
DHURQDXWLF DSSOLFDWLRQV
9
5()(5(1&(
>@
&$ .OHLQ 0LFURZDYH VKLHOGLQJ HIIHFWLYHQHVV RI (&FRDWHG GLHOHFWULF

VODEV ,((( 7UDQVDFWLRQ RQ 0LFURZDYH 7KHRU\ DQG 7HFKQLTXHV 9RO
1R SS
>@ %DVXGDP $GKLNDUL DQG 6DUPLVKWKD 0DMXPGDU 3RO\PHUV LQ VHQVRU
DSSOLFDWLRQ 3URJ 3RO\P 6FL SS
>@ 0 $QJHORSRXORV &RQGXFWLQJ SRO\PHUV LQ PLFURHOHFWURQLFV ,0% -
5(6 '(9 9RO 12 -DQXDU\
>@ .ULVKQD 1DLVKDGKDP 6KLHOGLQJ HIIHFWLYHQHVV RI FRQGXFWLYH SRO\PHUV
,((( 7UDQVDFWLRQ RQ (OHFWURPDJQHWLF &RPSDWLELOLW\ 9RO 1R SS
)HEUXDU\
>@ +RDQJ 1JRF 1KDQ -/ :RMNLHZLF] -/ 0LDQH DQG 5 6 %LVFDUUR
/LJKWZHLJKW HOHFWURPDJQHWLF VKLHOGV XVLQJ RSWLPL]HG SRO\DQLOLQH
FRPSRVLWHV LQ WKH PLFURZDYH EDQG 3RO\PHU IRU $GYDQFHG
7HFKQRORJLHV 9ROXP ,VVXH SS
>@ & < /HH + * 6RQJ .6-DQJ ( - 2K $ - (SVWHLQ DQG - -RR
(OHFWURPDJQHWLF LQWHUIHUHQFH VKLHOGLQJ HIILFLHQF\ RI SRO\DQLOLQH
PL[WXUHV DQG PXOWLOD\HU ILOPV 6\QWKHWLF 0HWDOV SS

>@ 7 0DNHOD 63LHQLPDD 7 7DND 6-XVVLOD DQG +,VRWDOR 7KLQ
SRO\DQLOLQH ILOPV LQ (0, VKLHOGLQJ 6\QWKHWLF 0HWDOV SS

>@ +RDQJ 1JRF 1KDQ 0LDQH -/ DQG :RMNLHZLF] -/ 0RGHOLQJ RI
HOHFWURPDJQHWLF VKLHOGLQJ HIIHFWLYHQHVV RI PXOWLOD\HU FRQGXFWLQJ
FRPSRVLWHV LQ WKH PLFURZDYH EDQG )LUVW ,QWHUQDWLRQDO &RQIHUHQFH RQ
&RPPXQLFDWLRQ DQG (OHFWURQLFV +DQRL 2FW SS

>@ +RDQJ 1JRF 1KDQ 5PLOL + $]HGGLQH 0 :RMNLHZLF] -/ DQG
0LDQH -/ 1RQGHVWUXFWLYH &RPSOH[ 3HUPLWWLYLW\ 0HDVXUHPHQW RI
&RQGXFWLQJ 0DWHULDOV 8VLQJ 2SHQHQGHG &RD[LDO 3UREHV )LUVW
,QWHUQDWLRQDO &RQIHUHQFH RQ &RPPXQLFDWLRQ DQG (OHFWURQLFV +DQRL
2FW SS
>@ - $ .RQJ (OHFWURPDJQHWLF ZDYH WKHRU\ -RKQ :LOH\ DQG 6RQV ,QF
1HZ <RUN
>@ 6 )DXYHDX[ -/ :RMNLHZLF] DQG -/ 0LDQH %URDGEDQG
(OHFWURPDJQHWLF
6KLHOGV
8VLQJ
3RO\DQLOLQH
&RPSRVLWH
(OHFWURPDJQHWLF 9RO 1 'HF SS
38
eBook for You
45
SE (dB)
$&.12:/('*0(17
Proceedings of 2008 International Symposium on Electrical Insulating

Materials, September 7-11, 2008, Yokkaichi, Mie, Japan
MVP1-1-5
Conception and Realization of Multilayered Composite Electromagnetic Shielding Material

at Microwave Frequency by Using Genetic Algorithm
Pham Hong Thinh1,*
Hoang Ngoc Nhan2
Nguyen Thi Lan Huong1
Wojkiewicz J-L.3
1
Faculty of Electrical Engineering, Hanoi University of Technology (HUT)
1, Dai Co Viet Street, Hanoi, Vietnam
* E-mail : thinhph-htd@mail.hut.edu.vn
2
Faculty of Automation Technology- Electric Power University (EPU)
235, Hoang Quoc Viet-Hanoi-Vietnam
3
Ecole des Mines de Douai
Rue Charle Bourseul, Douai Cedex, France
Abstract: Genetic algorithm was used to optimize a

multilayered structure of conducting composite in
microwave frequency. The result of optimization showed a
good agreement with the measurements, and a 262m or
400m total thickness of three-layer Pani/PU composite
can be compatible to many industrial and military shielding
applications
general formula illustrated in figure 1 is converted into

conducting polyaniline by oxidation with camphor
sulphonic acid (CSA) as described in [5]. EB was fully
protonated by CSA with the 0.5 molecular ratio
(H+/PhN=0.5), and was dissolved in dichloroacetic acid
(DCAA) to concentration of 2.85%. Protonation reaction
lasted at least 3 days at room temperature. We eventually
obtain 1.2g Pani in 100g Pani-(CSA) 0.5/DCAA solutions.
INTRODUCTION
In this work, the polyaniline (Pani) doped by

camphosulphonic acid (CSA) was used to realize a
multilayered shielding electromagnetic material. The
electromagnetic properties of Pani/PU freestanding film
with different concentration of Pani in microwave band are
shown. The electromagnetic shielding performance of
three-layered Pani/PU composite are investigated in
microwave band in the goal of obtainning a suitable
attenuation for different applications. Genetic algorithm is
used to optimize the three-layer composite structure.
Measurements in far-field are performed and compare with
optimization results.
The non-conducting emeraldine base (EB) polyaniline with
1-y
Figure 1: Schematic representation of Emeraldine base, y=0.5
In this study, the Polyurethane (PU) thermoplastic material

was used to elaborate the conducting blend because of its
flexibility, good mechanical properties and solubility in
DCAA. First, PU was dissolved in the DCAA solution with
the mass ratio ranging from 1 to 10 %, and then they were
blended with Pani-(CSA)0.5 solution with different mass
fraction of Pani in the PU and CSA. The mixture was
carefully homogenized by a magnetic agitator for 10
minutes. Finally, the Pani-CSA05/PU solution was spread
out on a glass container to achieve the freestanding films,
or on a Kapton slide for the three-layered films. The
solvents were evaporated under an infra lamp at least 48h,
and then in a vacuum oven at 60C for 6 h. The obtained
Pani/PU freestanding films have a specific mass about
1g.cm-3.
Table 1: Static conductivity of the Pani/PU freestanding films
with different concentration of Pani
Material
Pani0.2/PU
Pani0.5/PU
Pani1/PU
Pani4.7/PU
Pani8.8/PU
Pani16/PU
Pani44/PU
EXPERIMENTAL
Shielding electromagnetic material
eBook for You
Nowadays, the electromagnetic interference (EMI)

becomes more significant due to the proliferation of
commercial, military, scientific electrical devices and
equipments in high frequencies. To protect against the
incoming and potentially disturbing radiation penetrating
into the equipment, electronic devices must be shielded.
Recently, intrinsically conducting polymers (ICPs) are
promising materials and an alternative to metal for EMI
shielding because of their high conductivity and dielectric
constant, lightweight structure and ease of controlling these
properties through chemical processing. Among the ICPs,
Polyaniline (PAni) has lots of potential applications in
electronic and micro-electronic industry such as
opto-electronic devices, sensors, electrostatic discharge
layers, printed circuit board [1,2] and electromagnetic
interference shielding [3,4]because of its good electric
properties and environmental stability.
d(m)
150
150
150
160
155
145
130
V(S/m)
0.2
0.3
4
235
792
2456
11500
Table 1 presents the static conductivity of Pani/PU with

different concentrations of Pani in the blends. The
measured conductivity of the PAni/PU film is typical
75
39
percolation phenomena and follows the scaling law of the

percolation:
(1)
V(p)= V0(p-pc)t
where 0, p, pc and t are respectively a constant, the
concentration of the Pani in the blends, the percolation
threshold, and the critical exponent. In our calculations, the
best fit is shown in figure 2 with pc = 0.188 wt% and t=2.3.
This very low percolation threshold shows that the
PAni/PU films present a good inter-chain interaction
between PAni, the dopant acid and the PU.
depending on the applications for the three-layered

material. The objective function is calculated according to:
f 1 ( x)
(10
SE Fix
20
T)
Minimum
(4)
or:
f1 ( x)
20 log( 10
SE Fix
20
T ) Maximum (5)
Kapton
10
PAni/P
PAni/P
d1
percolation law
d2
d3
Figure 3: Description of three-layer structure

-2
Measurment
Table 2: Range of physical parameters of three-layered

material
-6
Material
-10
PAni-PU
Kapton
PAni-PU
3.1
-
-5
-4
-3
-2
-1
0
ln(p-pc)
Figure 2: Static conductivity as function of Pani

concentration: experimental data and fit to percolation scaling
law
Conductivity
(S/m)
From 30 to 104
0
From 30 to 104
Thickness (Pm)
From 0 to 300
125
From 0 to 300
Genetic algorithm
The transmitted coefficient T was calculated from the

number of layer in the structure, electromagnetic properties
(conductivity, permittivity) and the thickness of each layer
[6]. The technique for measurement of shielding
effectiveness in far-field was also described in [6]
The genetic algorithm (AG) is a stochastic method of

research and optimization which mimic natural selection
and genetic [7]. The AG starts with the random creation of
a first population of individuals. Then the algorithm will
pass from a generation to others by applying the
mechanisms of evaluation, selection, crossing, and
mutation until obtaining a termination conditions (iteration
count N, better fitness). The genetic algorithm is
represented as follows [7]:
1. Initialize N random individuals of population.
2. Calculate the fitness of each individual.
3. Select N/2 pairs of parents from the current population
by using roulette wheel.
4. By performing crossover on the N/2 pairs of parents,
children are formed and replaces the parents in the
new population with probability Pc (in our work we
use Pc=0.9).
5. Mutation is performed on the new population with pm
(in this we use pm=0.01).
6. The new population becomes the current population.
7. Exit if the termination conditions are satisfied,
otherwise go to step 3 (number of generation, the
fitness or the time of calculation)
Objectif funtion
RESULTS AND DISCUSSIONS
To improve the shielding efficiency and the mechanical

properties of shielding material, the three-layer structure
was studied (figure 3), the range of physical parameters of
each layer are presented in table 2. The objective is to
obtain a desired attenuation (SEfix) in the microwave band
Shielding effectiveness of freestanding Pani/PU film in

X and Ku bands
Shielding effectiveness theory and measurement

The electromagnetic shielding effectiveness (SE) of a
material is defined by the ratio of the transmitted power
(Pt) through the material to the incident power (Pi) of an
electromagnetic wave [4]. In general SE is given in decibel
(dB):
P
E
H
SE 10log t 20. log t 20. log t (2)
P
E
i
i

Hi
Where Ei(Hi) and Et(Ht) are the incident and transmitted
field strength.
However, by choosing incident field equal one unit, the
shielding effectiveness can be calculated from the
transmitted coefficient T [5]:
(3)
SE 20. log T
eBook for You
ln Sigma (S/m)
Figure 4 depicts the SE evolution of freestanding Pani/PU

films in X and Ku bands. It is clear that the shielding
76
40
The interaction mechanism of an electromagnetic wave

with a material can be divided in three mechanisms:
reflection, absorption and multiple reflections inside the
material. This behavior depends on the skin depth G (table
3) of the material ( G
2 /( P 0VZ ) ). For the freestanding
Pani/PU films, if the thickness is smaller than its skin depth,
the reflection is dominant and the SE behaviour is plat in
this band (8-18GHz). Otherwise, the interaction
mechanism is due to absorption phenomena so the SE
behaviour increases with the frequency (for the Pani16/PU
and Pani44/PU).
(m)
2516
328
179
101
47
V(S/m)
4
235
792
2456
11500
Optimization results
The shielding effectiveness of three-layered material was
calculated from a combination of four parameters:
conductivity and thickness of the first and the third layers.
The optimization purpose is to find four parameters that
define a SE best suited for the specific application with the
least weight of PAni and the least total thickness. Knowing
that the required SE is usually greater than 40dB for many
commercial shielding applications, and the SE should be
significantly higher than 80dB for military applications.
For commercial shielding application (SE>40dB) from
50MHz
Table 4: The optimized parameters for commercial application
from 50MHz
Material
1st-layer: PAni /PU
2nd-layer: Kapton
3rd-layer: PAni /PU
H
3.1
-
V (S/m)
2215.3
|0
5888.7
d (m)
78.4
125
59.6
Table 3: Skin depth of PAni/PU films at 10GHz
Sample
Pani1/PU
Pani4.7/PU
Pani8.8/PU
Pani16/PU
Pani44/PU
By comparing with a film of Pani doped by

p-toluenesulphonic acid (PTSA) in 4-chloro-3-methyl
phenol (CMC) (d=300m, V=2200S/m, SE=39dB at
1GHz) [8], we found that the film Pani16/PU has the same
conductivity (2456 S/m), the SE is nearly the same (38dB
at 10GHz) but the thickness is half of the Pani-PTSA/CMC
film. In addition, the contribution of the polyurethane in
the conducting film provides a good mechanical property
with respect to the Pani-PTSA/CMC film. This advantage
shows that films Pani/PU may be more homogeneous than
films Pani-PTSA/CMC.
eBook for You
efficency depends on the conductivity and the thickness of

the material. For the Pani1/PU film, the attenuation is
1.75dB corresponding to 33% attenuation in incident field.
The SE increases with the conductivity following the
algorithm law: 21dB, 30dB, 38dB and 55dB for Pani5/PU,
Pani8.8/PU, Pani16/PU and Pani44/PU (Pani pure) films
respectively. With the SE of 33dB, the Pani16/PU provides
99.99% attenuation of the incident radiation. It indicates
that this composite material can be used for many EMI
commercial shielding applications (following the EMI
standard
proposed
by
Federal
Communications
Commission FCC in part 15, classe B). Here, the
modeling results are in good agreement with measurements
as we have 5dB of the measurements uncertainty induced
by the calibration procedure, the conductivity and the
thickness of each film. This modelling method permit to
calculate the shielding behaviour of material in other band,
and it can also be used in objective function for shielding
optimisation problem.
The GA started with the randomly generated initial

population. By using the objective function defined by
equation 5, the fitness evaluations were increased and
stabilized after 400 iterations with the transmitted
coefficient error between desired T and calculated T is
about 3.7 10-6. The best solution are found in the table 4.
d(m)
150
160
155
145
130
80
70
70
PAni pur d=130m

60
60
50
SE (dB)
SE(dB)
PAni 16%/PU d=145m
40
40
30
PAni 8,8%/PU d=155m
30
PAni 4,7%/PU d=160m
20
20
Measurement
10
10
0
8.0E+09
Modeling
50
0
5.0E+06
PAni 1%/PU d=150m
1.0E+10
1.2E+10
1.4E+10
1.6E+10
1.0E+10
F( Hz)
1.5E+10
2.0E+10
Figure 5: Shielding effectiveness of optimized multilayered

structure (calculation for commercial application)
1.8E+10
F(Hz)
Figure 4: Shielding effectiveness of Pani/PU films in X and
Ku bands (Dots are experimental data and solid lines are
modeling results)
5.0E+09
In order to verify the model, we calculated the SE from

equation 3 and compared it with the measurement. From
77
41
120
Measurement
100
70
d (m)
196
125
180.1
60
8.0E+09
1.0E+10
1.2E+10
1.4E+10
1.6E+10
1.8E+10
2.0
F(Hz)
After 200 generations, the best solution is found in the table

6 with the transmitted coefficient error between T designed
and T calculated is about 3.10-7. A three-layered film with
optimized parameters in the table 6 was realized. The first and
the third layer are the 30% of PAni/PU film, the second one is
Kapton. The thickness of each layer is respectively 210, 125 and
180m. A better precision of conductivity and thickness
between the optimized parameters and the real parameters show
a close agreement between measured SE and the calculated one
(figure 6).
The SE of the freestanding Pani30/PU film with the 390m of
thickness, which is equal the total thickness of the first and the
third layer in optimal structure, was modelized versus frequency
(X and Ku bands). It is observed that the optimal structure
shows more 6dB in shielding effectiveness in comparison with
freestanding film in X and Ku bands.
CONCLUSION
A high level of electromagnetic shielding effectiveness was
achieved by using the Pani/PU blends and the composites
with Kapton film. The physical properties of the Pani/PU
blends were varied from insulating (10-11S/m) to
conducting states (11500 S/m), and showed a very low
percolation threshold. The PAni16/PU film with only 145
m of thickness have an attenuation of 38dB in X and Ku
bands. The structure optimization was performed by using
the genetic algorithm to realize multilayered composites
with a desired SE level. The SE higher than 40dB from
50MHz and 80dB from 8GHz were found with a total
thickness below 1mm, which are compatible with many
commercial and military shielding applications.
Figure 6: Shielding effectiveness of optimized film and

comparison with freestanding film (modelling for military
application)
ACKNOWLEDGMENT
VLIR-HUT
Institutional
Program
is
gratefully
acknowledged for the financial support provided during the
course of this study
eBook for You
V (S/m)
5803.7
|0
5972.7
6dB
modeling of freestanding film
Table 5: The optimized parameters for military application

from 8GHz.
H
3.1
-
90
80
For military shielding application (SE>80dB) from

8GHz.
Material
1st-layer: PAni /PU
2nd-layer: Kapton
3rd-layer: PAni /PU
Thee-layered SE modeling
partition of X and Ku band
110
SE (dB)
the calculated conductivities, the concentration of Pani in

the blends was deduced by using the percolation model
presented in equation 1. A multilayered structure has been
realized in which the first layer is Pani16/PU with 80m in
thickness, the second one is Kapton layer with 125m of
thickness, and the third layer is Pani30/PU with 55m of
thickness. Figure 5 compares the calculated SE and the
measured one of this film in X and Ku band. It is clear that
the measured SE is in good agreement with the calculated
SE using GA.
REFERENCES
[1] Basudam Adhikari and Sarmishtha Majumdar, Polymers in
sensor application, Prog. Polym. Sci. 29-2004, pp. 699-766.
[2] Angelopoulos M., Conducting polymers in
microelectronics, IMB. J. RES. DEV. Vol. 45, No. 1,
January 2001.
[3] Colaneri Nick F. and Shacklette Laurence W., EMI
shielding measurements of conductive polymer blends,
IEEE Transaction on Instrumentation and Measurement,
Vol.4, No.2, pp. 291-297, 1992.
[4] Krishna Naishadham, "Shielding effectiveness of
conductive
polymers",
IEEE
Transaction
on
Electromagnetic Compatibility, Vol. 34, No. 1, pp. 47-50,
1992.
[5] Cao Y., Qiu J. and Smith P., Effect of Solvents and
Co-solvents on the Processibility of Polyaniline: I.
Solubility and Conductivity studies Synthetic Metals 69,
pp. 187-190, 1995.
[6] Hoang Ngoc Nhan, Pham Hong Thinh and Nguyen Thi
Lan Huong, Electromagnetic characterization of Pani/PU
in multilayered structure, application for EMI protection at
microwave frequency, Proceeding of 2th International
Conference on Communication and Electronics, pp.
366-369, June 4-6, 2008
[7] Holland J., Adaptation in Natural and Artificial Systems,
The University of Michigan Press, Ann Arbor, 1975.
[8] Satheesh Kumar KK., Geetha S. and Trivedi D.C.,
Freestanding conducting polyaniline film for the control
of electromagnetic radiations, Current Applied Physics,
Volume 5, Issue 6, pp. 603-608. September 2005.
78
42
Tng hp cc bi bo2009
khoa hc
giai on
2007-2012
Annual
Report
Conference
on Electrical Insulation and Dielectric Phenomena
Leakage Current Analysis for Predicting

Flashover in Distribution Network
Thinh Pham Hong, Doan Ngo Trong and Dao Vu Hoang
Department of Power Systems-Hanoi University of Technology (HUT)
Abstract- Flashover process on outdoor insulator is believed to
be the interaction of the non-linear arc characteristic with the
linear resistance of the pollution layer. Such a complex problem
results in great difficulty of obtaining reliable performance of the
insulator. Flashover mechanism of outdoor insulator is
commonly referred to Obenaus theory in which dry band is
firstly formed in the contaminated surface. The voltage drops in
such band will create partial discharge in surrounding air and
finally lead to ultimate flashover. Just prior to flashover, the
leakage current increases in non-linear way, especially with the
severity of contamination layer. In this work, a leakage current
acquisition system is specifically developed for polluted insulator
using in distribution network (35kV). Leakage current
waveforms are measured with different amount of contamination
on two types of insulator: glass and silicone rubber insulator.
Partial discharge phenomena prior to flashover are well observed
by the non-linearity of the leakage current. The results could
provide a valuable tool for maintenance planning of the
insulators in distribution network, especially in industrially
polluted or coastal areas.
Fig. 1. Glass and silicone rubber insulators using in 35kV distribution network
I. INTRODUCTION
With the low height of towers, the insulator of distribution
network located in industrial and populated areas including the
coastal one is especially polluted by metallic oxide and metal
in form of dusts or powders, and also by soluble ionic and
inert layers. When energizing, the conductivity layer could be
formed with or without water on the insulator and leads to
flashover. The development of the leakage current (LC) just
prior to flashover leads to the damage of insulator surface [1],
and it was usually used as a tool for predicting the
performance of the insulator [2,3]. However, each type of
insulator has the particularity on their LC waveform
depending on the material in use, the rated voltage, and also
the condition of environment.
In this work, a system for LC measurement is specifically
developed for insulator in distribution network with the rated
voltage of 35kV. Frequency analysis of the LC has been
carried out for two types of insulator in glass and in silicone
rubber. The solid layer method has been used to simulate four
degree of pollution severity. The relationship between
different harmonic content of the LC and the pollution degree,
the electric stress and the resting time has been also discussed.
TABLE I
TECHNICAL DATA FOR GLASS AND SILICONE RUBBER INSULATORS
Shed diameter
Unit spacing
Leakage distance
Number of unit or shed
per string
TABLE II
THE SALINITY AND EQUIVALENT SDD OF THE TESTING PROCEDURE
Severity
A. Insulator parameters
Different type of insulator is currently used in Vietnam
distribution network. Standard cap & pin glass and porcelain
insulators are widely used in most of the old system. The
silicone rubber insulator is recently put into operation in new
Silicone rubber
95mm
580mm
880mm
8
B. Testing procedure
The solid layer method (clean fog method) [4] has been used
to simulate the pollution severity of the testing insulator.
While using the tap water, kaolin and a suitable amount of
NaCl, the desired volume conductivity of the suspension on
the insulator under test could be obtained. From the relation
between the volume conductivity of the suspension and the
amount of NaCl and the salt deposit density (SDD), the
surface conductivity of the insulator under test could be
inferred as described in [4]. In this work, the amount of NaCl
dissolved in 1000ml of tap water were 14g, 25g, 55 and 90g,
which correspond to 4 degrees of pollution as defined in [5]
(Table II)
II. EXPERIMENTAL SET-UP AND PROCEDURE
978-1-4244-4559-2/09/$25.00 2009 IEEE
Glass
225mm
146mm
303mm/unit
3
Light
Medium
Heavy
Very heavy
462
Salinity
(g/l)
14
25
55
90
Surface conductivity
(S/m)
2.2
3.8
7.6
10.9
SDD
(mg/cm2)
0.05
0.09
0.20
0.32
43
eBook for You
distribution system due to their hydrophilic properties. In this

work, glass insulator and silicone rubber insulator have been
selected for leakage current analysis. Fig. 1 and table I show
picture and some technical data of these insulators.
C. Leakage current measurement system

A LC data acquisition system has been specifically
developed during this work. The test voltage is applied to the
insulator from a testing transformer (0.2/100kV, 5kVA) fed
from a variable autotransformer (0.22/0-0.2 kV, 5kVA). The
leakage current is monitored by converting the current signal
into voltage through the use of shunt connected in series with
the test insulator. Depending on the level of the leakage
current, different shunts could be selected. The shunt voltage
across the selected resistor is fed to the analog input of a data
acquisition system by a coaxial cable. The voltage is then
amplified by an IC TL072CP with the gain of 33. An optical
amplifier ISO122JP was used as a buffer circuit to protect the
instrument from over-currents. The current was sampled at a
rate of 2000 sample/s, transferred to a data buffer by a
microcontroller ATMEGA 8L-8PU. In order to access the
reliability of the data acquisition system, a 20MHz Agilent HP
oscilloscope also provides visual display of the leakage
current (Fig. 2, 3 and 4).
distortion (THD) of the waveform is calculated from the

frequency spectrum using the following equation [6]:
%THD
100 AI 2 2 AI 3 2 ... AI N 2
(1)
AI1
where, A(I1) is the amplitude of the fundamental component,
A(IN) is the amplitude of the Nth harmonic and N is the
number of harmonics.
Fig. 4. Schematic diagram of experimental set-up with the LC monitoring

system
eBook for You
III. RESULTS AND DISCUSSION
Fig.2. Schema of leakage current measurement system
To PC
Fig.5. Harmonics of leakage current and THD of glass insulator versus the
salinity at 35kV
Data
acquisition
board
To OSC
Transformer 220V/12V
Voltage signal
from shunts
Shielded
box
Fig. 3. Hardware layout of leakage current measurement system
D. Signal analysis
The interface communication and the analysis were
performed by means of LabView package [6]. For the purpose
of storing and analyzing LC data, two main programs have
been written in this package. The first one is used to transfer
voltage waveforms from the microcontroller to the PC and
save them as data files. The second one displays the data files
in time domain and analyzes them in frequency domain using
the fast Fourier transformation (FFT). The total harmonic
Fig.6. Harmonics of leakage current and THD of silicone rubber insulator

versus the salinity at 35kV
463
44
Fig. 5 and 6 show the dependence of the LC harmonics and

THD on the degree of pollution ranging from light to very
heavy condition, for glass and silicone rubber insulators at
rated voltage (35kV). Due to hydrophobic properties and very
low surface conductivity, silicone rubber shows a very low LC
levels in the order of few A even at very heavy pollution.
The root mean square value (RMS), as well as the first
harmonic, of the LC on glass insulator remains constant with
the pollution degree below heavy level, but it substantially
increases when the surface becomes very heavily polluted.
However, an opposite trend was observed on the LC of
silicone rubber in which the first harmonic of the LC is tripled
when the insulator changes from clean condition to the
heavy one. Those behaviors suggest that although a
conductivity layer is easily formed on the surface of silicone
rubber, but small surface area of silicone rubber has a large
effect on retaining the conductivity layer deposited on the
insulator in comparison with that of glass insulator. For both
insulators, the amplitude of the third and fifth harmonics
seems to be insensitive to the pollution level, and the THD at
the rated voltage (35kV) is the same order for both insulators
(~20%).
A. Influence of electric stress

At very heavy polluted condition, the effect of electric stress
was studied on glass insulator with the applied voltages are
5kV, 15kV, 25kV and 35kV (fig 9a, 9b, 9c and 9d). It was
observed that the LC waveform was slightly deformed even at
very low voltage (5kV), and it became much more deformed
when the applied voltage is superior to 25kV. It was,
therefore, reasonable to conclude that the amplitude and the
waveform distortion were increased with the non-linearity
phenomena due to the increasing of applied voltage. The 1st
and the 3rd harmonics, although not presented here, showed a
non-linearity behavior when the applied voltage superior to
25kV.
Fig.7. 3rd/1st and 5th/1st ratios of glass insulator at 35kV
Fig. 9a. LC waveform of glass insulator at 5kV and under very heavy
pollution
Fig.8. 3rd/1st and 5th/1st ratios of silicone rubber insulator at 35kV
As the fundamental harmonic as well as the RMS of the LC

is increased with the surface conductivity, and the 3rd and 5th
harmonics are inherent in partial discharges surrounding the
band gap. The surface conductivity could lead to serious
Fig. 9b. LC waveform of glass insulator at 15kV and under very heavy
pollution
464
45
eBook for You
consequences including the partial discharge, but the latter is

considered as the direct cause of flashover. In order to
compare the acceleration of partial discharge to that of surface
conductivity, the ratio between the 3rd and the 5th harmonics of
to the 1st one (in percentage) versus the salinity was plotted in
fig 7 and 8. It was observed that below the heavy level, the
3rd/1st ratio the silicone rubber is much more sensitive to the
pollution level than that of glass insulator. But above the
heavy level, a more significant increase of the ratio was
recorded for glass insulator. Those observations are well
correlated with the level of acoustic noises due to partial
discharges that we recorded during the tests, although it is
difficult to distinguish them when observing the fig. 5 and fig.
6.
Fig. 9c. LC waveform of glass insulator at 25kV and under very heavy
pollution
correlation between the 3rd harmonic/1st harmonic ratio and the

acoustic noise, between the waveform deformation of the LC
and the applied voltage values, and between the increase of
resting time and 3rd and the 5th harmonics contents suggested
the association between the fundamental content and the
surface conductivity, and also between the high order
harmonics with the electrical discharges due to the presence of
dry bands. The results also suggested that the combination of
LC amplitudes, waveshape analysis and the moment of
measurement could provide a reliable tool on the pollution
degree of the insulator in distribution network.
Fig. 9d. LC waveform of glass insulator at 35kV and under very heavy
pollution
B. Influence of resting time

The exposed time to the environment could lead to the
evaporation of conductivity surface, and therefore lead to
wrong interpretation of the LC waveform. Fig. 10 shows the
dependence of LC parameters on the resting time (in minute)
for glass insulator. It was observed that a small deformation in
the LC waveform was recorded with the variation of resting
time at heavy pollution degree. The 5th harmonic and the THD
were slightly increased with the resting time after 60 minute,
this suggested that such a resting time would be great enough
to form dry bands on the insulator, and thus the partial
discharge could occur. The RMS amplitude and the
fundamental harmonic show a contrary trend, which could
imply the reduction in surface conductivity and the
augmentation of dry bands with the resting time. This
behavior was well correlated with that observed in the
previous paragraph, in which the high order harmonics
associated with electrical discharge were usually increased
with the decreasing of fundamental component, and so the
surface conductivity.
ACKNOWLEDGMENT
REFERENCES
[1] El-Hag A. H. Ayman, H. El-Hag, Shesha H. Jayaram and Edward A.
Cherney Fundamental and low frequency harmonics components of
leakage current as a diagnostic tool to study aging of RTV and HTV
silicone rubber in salt-fog, IEEE Trans. Dielectr. Elec. Insulation, vol.
10, pp. 655-664, August 2003.
[2] T. Suda, Frequency Characteristics of Leakage Current Waveforms of a
String of Suspension Insulators, IEEE Trans. Power Delivery, vol. 20,
pp. 481-487, January 2005.
[3] M. Fernando, Performace of Non-ceramic Insulators in Tropical
Environments, PhD thesis of Chalmers University of Technology, 1999
[4] International Electrochemical Commission, Artificial Pollution Tests on
HV Insulators to be Used on AC Systems, IEC Publications No. 507,
Second Edition, 1991
[5] International Electrochemical Commission, Guide for the Selection of
Insulators in Respect of Polluted Conditions, IEC Publications No. 815,
1979
[6] LabVIEW User Manual, National Instruments, 1998
IV. CONCLUSION
A practical system was developed in order to monitor and

analyze the leakage current in glass cap & pin and silicone
rubber insulator using in distribution network (35kV).
Depending on the surface properties, the LC showed a relative
difference in sensitivity in different degree of pollution. The
465
46
eBook for You
Fig.10. Harmonics of leakage current and THD of glass insulator as function

of resting time, at heavy condition and 35kV
Tng hp cc bi bo2009
khoa hc
giai on
2007-2012
Annual
Report
Conference
on Electrical Insulation and Dielectric Phenomena
Grounding Resistance Calculation Using FEM and

Reduced Scale Model
Thinh Pham Hong, Quan Do Van and Thang Vo Viet
Department of Power Systems-Hanoi University of Technology (HUT)
calculation was first performed in simple configuration of
grounding electrode. Reduced scale model in simulation was
used for grounding grid of 16 square mesh, with and without
ground rods. The results could be effectively used for
grounding grid design.
Abstract- The grounding grid of a substation is one of the most

important parts in an electrical system from the point of view of
the safety of the people and equipment. Depending on the nature
of the phenomena involved in the system is fault or lightning
current, the behavior of grounding system is considered under
steady state or transient point of view. For safety purpose of
grounding grid, the ground resistance is more likely calculated by
analytical or numerical method by using potential distribution
calculation along soil structure. In comparison with the analytical
method, Finite Element Method (FEM) method in calculation of
grounding resistance is more flexible in analyzing asymmetrical
geometry of the grid, as well as in case of anisotropy of soil
resistivity. However, with the increase in size and complexity of
substation grid, the FEM method could not be applied due to the
increase of divided elements. This paper presents a reduced scale
model for grounding resistance calculation using FEM method.
The results giving the effect of electrode configuration on
potential profile and grounding resistance are also presented.
The simulation of the grounding grid allows us to calculate

the grounding resistance and the potential profile above the
ground grid regardless of their shape and the geometry of the
ground electrode. Like any Finite Element Method (FEM)
based calculation, ANSYS program calculates the grounding
resistance by using one of the two methods [1]:
A. Current Flow Analysis
For an arbitrary geometry, the resistance between two
electrodes can be calculated from the voltage V and the
dissipated power P in the ground:
INTRODUCTION
In which the dissipated power P is determined by:
The grounding system of a substation is one of the most

important parts in a power system. The purpose of the
grounding system is to provide a low impedance electrical
contact between the neutral of an electrical system and earth
[1]. Depending on the nature of the phenomena involved in the
system is fault or lightning current, the behavior of grounding
system is considered under steady state or transient point of
view. For the purpose of safety, the performance of a
grounding system is evaluated by some parameters such as
ground resistance, touch voltage, step voltage, mesh voltage
[2].
The calculation method of grounding system using
analytical approaches [2, 3] in which the soil is considered as
uniform medium and the electrodes are considered as
symmetric. When the uniform soil approximation is no longer
valid and the electrodes contain irregularities, such methods
may result in unsafe or overdesigned grounding system [4].
Recently, Finite Element Method (FEM) has been used as an
excellent numerical method to calculate the grounding system
[5-7]. The main disadvantage of this method is the limited
capability of the computer in case of too large dimension of
grounding grid, specifically with the large ratio between grid
dimension and grounding electrode size. In experimental
works, the grounding system is always measured and validated
by reducing in size by the same scale factor of the physical
dimension of the grid [8-11]. But few investigations have
focused on the reduced scale model in simulation.
This work presents the FEM simulation of grounding grid
using reduced scale model in ANSYS program. The
978-1-4244-4559-2/09/$25.00 2009 IEEE

(1)
(2)
Where:
J: current density
V: electrical conductivity
E: electric field
B. Electrostatic Analysis
Another method used to calculate the grounding resistance
in FEM is to know the stored energy by the electric field in the
ground:

(3)
Where the stored energy by the electric field is given by:

(4)
The following simplifying hypotheses will be also taken into

account when calculating the grounding system:
- The soil is isotropic and uniform in each layer
- The non-linearity does not occur in the soil
- The grid behavior at power frequency is considered
in stationary regime.
- In reduced scale model, all physical dimensions of a
grounding grid are reduced by the same factor
including the conductor diameter and the buried
depth. Thus, the current injected in grounding system
is unaltered and remained 1kA in any model.
- The grounding grid is buried in a semi-infinite earth,
and in order to not distort the field inside the
278
47
eBook for You
I.
II. MODEL AND SIMULATION
calculated medium [11] we considered a surrounding

earth of the grid having a diameter equal to at least
three times the width of the grid.
300
250
200
SIMULATION RESULTS
R,
A. Ground resistance in uniform soils.

Ground resistance was first calculated in uniform soil of
400:.m in resistivity. Fig. 1 shows the potential distribution
in the soil of a horizontal rod of 2m in length. Due to
symmetry of the electrode configuration, one-fourth of the
vertical rod was necessary to simulate in 3D model. The
simulation results enabled us to observe the step voltage,
which was calculated the voltage difference between 1m apart,
along and perpendicular to the rod (Fig. 2). The step voltage
along the electrode shows a small deviation from that is
perpendicular to the electrode, and a dangerous point in step
voltage exists at 1m far from the electrode end along the
electrode. This behavior suggests that the maximum step
voltage for a horizontal rod should be calculated at a certain
point from the electrode ends. The effect of electrode length
was also examined and shown in fig. 6
150
100
50
0
10
12
14
Electrode length, m
Fig 3. Grounding resistance of horizontal rod versus the length
In order to increase the performance of the grounding

system, 3 vertical rods have been added to the horizontal
configuration to form a mixed configuration. The voltage
profile on the soil surface was shown in fig. 3. It was obvious
that maximum step voltage has the same behavior than the
previous case, but the maximum value has been reduced in
half (from |2400V to |1000V). The ground resistance has
been substantially decreased when using mixed configuration.
It was observed that 50: was the limited value in horizontal
configuration. However, the combination of vertical and
horizontal rods could reduce the grounding resistance to 8:
(fig. 9). The saturation at 4m of distance between two vertical
rods was correlated with the analytical results [1] in which the
combination may obtain the best efficiency if the distance
between two vertical rods should not exceed twice of their
length.
Fig. 1. Potential distribution in the soil for 1 horizontal rod of 0.020.02m in

size
30000
25000
20000
Perpendicular to
the rod
15000
Along the rod

Fig 4. One-fourth of mixed configuration with 3 vertical rod of 2m and 1
horizontal rod of 4m
10000
5000
12000
0
0
10
15
20
25
Distance from the electrode end, m
10000
30
8000
Fig 2. Step voltage distribution on the soil surface, along and perpendicular to
the horizontal rod
Perpendicular to the
horizontal rod
6000
Along the horizontal rod
4000
2000
0
0
10
15
20
25
30
Distance from the electrode end, m

Fig 5. Step voltage distribution on the soil surface, along and perpendicular to
the horizontal rod
279
48
eBook for You
III.
18
9000
16
8000
14
7000
20mx20m
12
6000
10mx10m
R, :10
V 5000
4000
3000
2000
1000
4mx4m
0
0
4
6
Distance between two vertical rods, m
10
20
30
40
50
X(m)
Fig 6. Ground resistance versus vertical rod spacing in mixed configuration

Fig. 8. X-axis potential profile for 16 mesh grid without ground rod
B. Reduced scale model

As the principle of the FEM is dividing the studied volume
into elements, a grounding system of a large substation,
especially in presence of vertical rods, may lead to too many
divided elements so that the computer could not solve it. This
could be one of reasons why previous works always used the
FEM to determine grounding resistance of very simple ground
electrode [7], of small grids with ground rods (the maximum
of grid dimension was 12m8m) [5], or of large grids (the
maximum of grid dimension was 100m80m) but without
ground rods [6]. In this section, the behavior of grounding
grids of 16 meshes without and with 16 ground rods installed
in the boundary junction was simulated (fig. 7). A variety of
grids with outside dimensions 20m20m, 10m10m, and
4m4m with and without ground rods, which correspond with
scale factor of 1:1, 1:2 and 1: 5 respectively, were modeled in
uniformed soil. The other parameters including buried depth,
soil resistivity and electrode size were also reduced with the
same scale factor (TABLE I and II).
X
9000
8000
7000
4mx4m
6000
10mx10m
5000
20mx20m
4000
3000
2000
10
20
30
40
50
Y(m)
Fig. 9. Y-axis potential profile for 16 mesh grid without ground rod
TABLE I
PARAMETERS AND SIMULATION RESULTS OF REDUCED SCALE MODEL IN GRID
WITHOUT GROUND RODS
Scale
Grid
Grid
Buried
factor
dimension
conductor
depth
diameter
Soil
Ground
resistivity resistance
.m
Max. step
voltage,
1251.9
cm
20x20
0.6
400
7.85
10x10
0.3
200
8.24
1266
1/5
4x4
0.4
0.12
80
8.61
1268.7
In fig. 8 and 9, potential values versus the distance from the

grid center were normalized to original grid (20m20m) with
the same scale factor. That means 1 meter in x-axis and y-axis
of 10m10m and 4m4m curves corresponds with 0.5m and
0.2m in the simulation results respectively. Also, step voltage
was calculated between 1m, 0.5m and 0.2m apart in each
model. In comparison with the original grid, the ground
resistance in 1:2 and 1:5 models did have 4.97% and 9.68% in
difference while the step voltage had only 1.13% and 1.34%.
Fig. 7. Grounding grid to be simulated
As shown in fig. 7 it is necessary to plot the surface potential

profiles along the center and diagonal lines of the grid (x and
y-axis). Typical profiles for a 16 mesh grid of three scale
factors without ground rods were plotted in fig. 8 and fig. 9.
1
1/2
280
49
eBook for You
1000
overcome by using reduced scale model. Although verification

on field would be needed, reduced scale model in simulation
could provide an inexpensive solution, and it could be
effectively used for parametric studies for grounding grid
design.
9000
8000
20mx20m
6000
10mx10m
5000
4mx4m
ACKNOWLEDGMENT
4000
3000

Center for Development and Application of Software for
Industry (DASI) at HUT is gratefully acknowledged for its
help during this study
2000
1000
0
10
20
30
40
50
X(m)
Fig. 10. X-axis potential profile for 16 mesh grid with 16 ground rods
installed in surrounding junction
REFERENCES
9000
8000
[2]
[3]
10mx10m
6000
[1]
20mx20m
7000
4mx4m
5000
4000
[4]
3000
2000
[5]
1000
0
0
10
20
30
40
50
[6]
Y( m)
Fig. 11. Y-axis potential profile for 16 mesh grid with 16 ground rods
installed in surrounding junction
[7]
TABLE II
PARAMETERS AND SIMULATION RESULTS OF REDUCED SCALE MODEL IN GRID WITH
GROUND RODS
[8]
Scale
Grid
Grid
Radius of
Buried
factor
dimension
conductor
ground
depth
diameter
rod
cm
Soil
Ground
resistivity resistance
.m
Max. step
voltage,
[9]
cm
20x20
0.6
400
7.08
1175.1
1/2
10x10
0.5
0.3
200
7.63
1181.3
1/5
4x4
0.4
0.2
0.12
80
7.90
1179
[10]
[11]
In presence of ground rods, the difference between maximum

step voltage of reduced model and the original one was much
more improved, they were 0.53% and 0.33% in 1:2 and 1:5
models respectively. However, the difference in ground
resistance was slightly higher in the previous case with 7.72%
and 11.55%.
IV.
A. P. Sakis Meliopoulos, Power System Grounding and Transients,

Marcel Dekker, 1988
IEEE Guide for Safety in AC Substation Grounding, IEEE Std 80-2000
Mansour Loeloeiaan, R. Velazquez Dinkar and Mukhedkar, Review of
Analytical Methods for Calculating the Performance of Large
Grounding Electrodes, IEEE Trans. on PAS, Vol. PAS-104, No. 11, pp.
3134-3142, 1985.
W. Ruan, J. Ma, J. Liu, F. B. Dawalibi and R. D. Southey, Performance
of HVDC Ground Electrode in Various Soil Structures, International
Conference on Power System Technology, 2002.
J. A. Guemess and F. E. Fernando, Method for Calculating the
Grounding Resistance of Grounding Grid Using FEM, IEEE Trans. on
Power Delivery, Vol. 19, No. 2., pp. 595-600, 2004.
J. A. Gemes-Alonso, F. E. Hernando-Fernndez, F. Rodrguez-Bona,
and J. M. Ruiz-Moll, A Practical Approach for Determining the Ground
Resistance of Grounding Grids, IEEE Trans. on Power Delivery, Vol.
21, No. 3, pp. 1261-1266, 2006.
J.M. Bueno Barrachina, C.S. Caas Peuelas, S. Cataln Izquierdo and
A. Quijano Lpez, Modelization of earth electrode excited by
atmospheric discharges based on FEM, International Conference on
Renewable Energies and Power Quality, 2008
A. Puttarach, N. Chakpitak, T.Kasirawat and C. Pongsriwat,Substation
Grounding Grid Analysis with the Variation of Soil layer depth
Method, IEEE Power Tech, 2007
Chung seog Choi, Hyang Kon Kim, Hyoung Jun Gil, Woon Ki Han, Ki
Yeon Lee, The Potential Gradient of Ground Surface according to
Shapes of Mesh Grid Grounding Electrode Using Reduced Scale
Model, IEEJ Trans. on Power and Energy, Vol. 125, No. 12, pp. 11701176, 2005
Sherif Ghoneim, Holger Hirsch, Ahdab Elmorshedy, Rabah Amer,
Measurement of Earth Surface Potential Using Scale Model, 42nd
International Conference on Universities Power Engineering, 2007.
Ross Cadecott and Donald G. Kasten, Scale Model Studies of Station
Grounding Grids, IEEE Trans. on Power Apparatus and Systems, Vol.
PAS-102, No. 3, pp.558-566, 1983.
CONCLUSION
The 3D model of FEM was used for calculating ground

resistance and potential profile of different grounding grids.
The simulation offered a great flexibility in calculating a
complicated grounding system without any simplifying
assumption. The main disadvantage of FEM method when
simulating physical dimension of a grounding grid could be
281
50
eBook for You
7000
Flashover Model of Arcing Horn in Transient

Simulation
Thinh Pham and Steven Boggs
Institute of Materials Science
University of Connecticut
97 North Eagleville Road, Storrs, CT 06269-3136, USA
I.
INTRODUCTION
Service interruption of an overhead transmission line from

lightning is one of the primary concerns of electric utilities. A
direct stroke to the top of a a tower or to a phase conductor
(shielding failure) causes overvoltage on insulator string, and
flashover occurs if the overvoltage exceeds the voltage-time
withstand characteristics of the insulator. An arcing horn across
the insulator protects the insulator by flashing over before the
insulator string.
Flashover behavior of an arcing horn is one of the important parameters in lightning insulation coordination calculations
for a transmission line. For a specific probability of distribution
of lightning current (and waveform), flashover behavior of the
arcing horn is combined with other protective methods such as
line arresters and controlled footing resistance to mitigate the
effects of lightning current without a line outage. Many
flashover models for air gaps [1-6] have been proposed for use
in the study lightning-induced flashover. None of the models is
universal in the context of an arcing horn, as each model is
validated over a very limited range of voltage waveforms. In an
actual transmission line, the lighting overvoltage across the
insulator is far from a standard waveform, and even the
lightning current source varied greatly.
The voltage waveform across an insulator depends on
various factors such as the geometry of the transmission line
(length of span, height of towers, footing impedance), the
distance between the lightning stroke point and the insulator in
question, the flashover behavior of adjacent insulators, the
position of lightning current (stroke to tower top or phase
conductor), etc. In this paper, four common flashover models
of the arcing horn were analyzed for two cases, under the
978-1-4244-6301-5/10/$26.00 @2010 IEEE

II.
FLASHOVER MODELS
A. Volt-time curve
The voltage-time curve has been determined experimentally
for the specific gap of arcing horn using in 220kV transmission
line insulator (Figure 1). This curve is determined using the
standard lightning impulse waveform applied across the arcing
horn. During the simulation, the voltage across the insulator is
compared with the volt-time curve. If simulated voltage
exceeds volt-time curve, flashover occurs in the simulation (an
ideal switch is closed through the inductance of the arc. As
recommended by IEC [7], the inductance of the arc is 1H/m.
The initial time of the V-t curve must be synchronized to the
instant that the voltage reaches the position of the insulator in
question, which is differs for each tower in a transmission line.
When the simulation is performed, this time lag must be
included in the model.
4000
3500
3000
2500
2000
1500
0.0
0.5
1.0
1.5
time (s)
2.0
2.5
Figure 1. Voltage-time characteristic of the arcing horn equipped

with 220kV insulator.
B. Equal area criterion

The method for determining flashover developed by Kind
and widely recommend by IEC [7] and CIGRE [8] is based on
a voltage-time area above a threshold voltage U0. Breakdown
51
eBook for You
Keywords-Flashover, arcing horn, EMTP simulation, lightning

surge, insulation co-ordination
standard 1x70 s waveform and in an actual twin-circuit

220kV transmission line struck by various lightning
waveforms. The simulations were performed using the
ATP/EMPT transient program with the integrated simulation
language MODELS.
Voltage (kV)
AbstractThe flashover performance of an arcing horn in a

transmission line is important in lightning surge analysis. Many
flashover models have been proposed for transient simulations,
but they are only validated over a limited range of waveforms,
which is very limited relative to what may be encountered in the
field. This paper compares four flashover models of the arcing
horn of a 220kV transmission line insulator under standard
lightning impulses and as installed in a twin-circuit, 220kV
transmission line. The results provide insight into appropriate
models for simulating flashover behavior of the arcing horn in
transmission line.
is assumed to occur at a given value of area. This can be

expressed mathematically as:
provides a reasonable correlation with impulse voltage waves

of 1/50s, 2/50s, and 3/50s.
F, where u(t) is the applied voltage across
the insulator and T0 is the time from which u(t)>U0. U0 and F

are unknown parameters evaluated from volt-time curve of the
arcing horn [8]. For our model, U0=1737kV and F=0.2664V-s.
In an ATP/EMTP simulation, a switch element closes though
the inductance of the arc when this criterion is fulfilled.
Figure 2. Flashover model of arcing horn proposed by Ueda [6]
C. Leader propagation method

Based on the method developed by Motoyama [4], who
explained physical discharge of breakdown process from
experiments for long air gaps (from 1 to 3m) and short tail
lightning impulse (from 1.2 to 1.4s of rise time, and from 1 to
3s time to half value), the breakdown process is divided into
two steps:
In ATP/EMTP simulation, the first computation is performed with SW1 and SW2 open. The close times of these
switches are determined by measuring the current through the
arcing horn. By repeating this procedure from the insulator
which is the most prone to breakdown to those which are less
so until no flashover occurs, the simulation can be completed.
Step 1: Leader onset process. When average applied voltage

on the gap exceeds a constant value, the leader onset condition
is fulfilled and defined by the following equations:
A. Standard wave form aplies on the arcing horn
polarity

polarity
for positive
SIMULATION RESULTS
12
[kA]
10
8
for negative
6
4
where Ts is leader onset time, D is the gap length (m)

Step 2: Leader development process. The average value of
leader developing velocity (LAVE) is defined by the following
equation:

2
0
(f ile linear_ramp.pl4; x-v ar t) c:XX0001-
12
16
[us] 20
Figure 3. Linear ramp waveform of lightning current.
where K1 is a constant, 2.5m /(Vs) for 0xLAVE<D/4 and

0.42m2/(Vs) for D/4xLAVE< D/2, and E0=750kV/m. xLAVE is
the average value of leader developing length which is defined
as

Breakdown occurs when the leader length xLAVE>D/2 and ends
when u(t)<E0(D-2xLAVE). In an ATP/EMTP simulation, a
switch element closes though the inductance of the arc when
the breakdown condition is met.
D. Non-linear inductance model
This model originated from leader development method of
Shindo et al. [2], and was developed by Ueda et al. [6] in order
to develop a model which could be validated for both long and
short air gaps. The flashover of arcing horn is modeled by a
non-linear resistance (Rn) and non-linear inductance (Ln) in
series (Figure 2). Rn is used to hold the critical flashover
voltage irrespective of the current. SW1 closes when the leader
current starts, i.e., when it exceeds 200A. Ln is the non-linear
inductance of the arcing horn. SW2 closes to complete
flashover across the gap when the current exceeds 1000A. Lf is
inductance of the arc as previously mentioned. This model
A lightning strike is modeled as a current source in parallel

with a lightning path impedance of 400 [9]. The current
source is a linear 1/70s ramp waveform of variable amplitude
(Figure 3). The time to flashover (in second) of each model is
presented in Table 1.
TABLE I.
I(kA)
4
4.2
4.3
4.4
4.5
4.8
5
10
20
30
40
TIME TO FLASHOVER FOR THE FOUR MODELS (s)
V-t
No flashover
No flashover
No flashover
No flashover
No flashover
3.2E-6
2.9E-6
9.4E-7
5.5E-7
3.9E-7
3.1E-7
Kind
No flashover
No flashover
No flashover
No flashover
No flashover
6.3E-6
4E-6
8.7E-7
4.4E-7
2.95E-7
2.2E-7
Motoyama
No flashover
No flashover
2.9E-6
2.7E-6
2.5E-6
2.16E-6
2E-6
1.03E-6
6.7E-7
5.3E-7
4.3E-7
Ueda
No flashover
No flashover
3E-6
2.48E-6
2.41E-6
2.27E-6
2.2E-6
1.35E-6
9.25E-7
7.55E-7
6.5E-7
The only difference among the models is in the time to

breakdown. The four models can be classified in two groups in
term of breakdown behavior. The first group consists of the V-t
model and Kind model, while the second group is formed by
the Motoyama model and Ueda model. Below 4.8kA of
lightning current, breakdown cannot occur in the first group of
model but it is still observed in the second one. The latter group
52
eBook for You
III.
stops to breakdown when lightning current below 4.3kA. When

breakdown occurs in the wave front (t<10-6s) or the current
magnitude is greater than 10kA, the order of the models in
terms of increasing time to breakdown is Kind, V-t, Motoyama
and Ueda. In this case, the difference in time to breakdown is
small so that the model predictions are similar. When
breakdown occurs in the tail of the wave, the time to
breakdown of Motoyama model and Ueda model is nearly the
same.
B. Flashover performance in an actual transmission line
497m
#1
496m
560m
#2
#3
533m
#4
237m
#5
Figure 5. Voltage across the arcing horns of tower #4 when I=100kA,

no flashover is assumed.
740m
#6
#7
Figure 4. 220kV transmission line to be simulated
The simulation was performed with a lightning current

source of the same form as described in section A, which is
connected to the top of tower #4. The primarily simulation
showed that below 60kA of peak lightning stroke current,
breakdown cannot occur on any phase of the transmission line.
The simulation was therefore performed with six values of
crest magnitude: 60kA, 70kA, 80kA, 100kA, 120kA and
150kA. If one assumes no flashover in the transmission line,
the voltage distribution on struck tower (tower #4) is far from
the standard wave form (linear ramp) due to the reflection from
adjacent towers (Figure 5).
Figure 6. Voltage across the arcing horns of phase A along the

transmission line when I=100kA, no flashover is assumed
eBook for You
A twin circuit 220kV transmission line that comprises 6

phase conductors and 2 shielding wires was analyzed. In this
typical transmission line, 7 towers with 6 spans are simulated.
The spacing between towers varies from 237m to 740m (Figure
4). The frequency dependent J-Marti model was used to
simulate each span of the transmission line. As the doublecircuit vertical configuration with 2 shielding wires is used, a JMarti model of 8 conductors was considered. The tower is
represented by four lossless transmission lines [9]. Footing
impedance is a simple resistance measured at low frequency [9]
(Rf10) which can produce higher potential on the top of the
tower than the current-dependent nonlinear resistance recommended by IEEE [1]. In order to avoid the reflection from the
ends of the transmission line, the latter were connected to AC
voltage sources via multiphase matching impedances [9].
Figure 7. Voltage across the arcing horns of tower #4 when I=100kA,

flashover is assumed on phase A.
The insulator of phase A of tower 4 always suffers from the

severest stress no matter what the lightning current (Figure 6).
When I=100kA, flashover on phase A of tower 4 occurs for all
4 models, and the time to breakdown is almost the same for all
models. Thus the voltage waveform does not differ appreciably among the models.
Flashover across phase A of tower #4 diverts the lightning
current in this phase, so that the voltage across insulators of
phase B and phase C of tower #4 is decreased and distorted
(Figure 7). The potential difference between phase A of tower
#4 and phase A of other towers is inverted by the sudden
increase in potential of phase A (figure 8).
Figure 8. Voltage across the arcing horns of phase A along the

transmission line when I=100kA, flashover on arcing horn on phase
A of tower #4 is assumed.
53
In V-t and Kind models, flashover in phase A of stricken

tower eliminates the possibility of breakdown in phase B of
that tower. For those models, increased lightning current causes
flashover on phase A of adjacent towers. As tower #3 is located
closest to tower 4, flashover on phase A of tower #3 occurs
when lightning current is very high (>120kA for V-t model,
and out of the range studied in the Kind model). The number of
towers which suffer flashover as predicted by Kind is always
the less than for other models except at I=80kA (Table II).
TABLE II.
I(kA)
60
70
80
100
120
150
200
FLASHOVER POSITION IN THE TRANSMISSION LINE
V-t
No flashover
No flashover
No flashover
1A, 2A, 4A,
7A
1A, 2A, 4A,
5A, 6A, 7A
1A, 3A, 4A,
5A, 6A, 7A
Kind
No flashover
No flashover
4A
4A
Motoyama
No flashover
4A
4A, 4B
4A, 4B
Ueda
No flashover
4A
4A
4A, 4B
4A
4A, 4B
4A, 4B
4A
1A, 3A, 4A,

5A, 6A, 7A
2A, 4A, 5A,

6A
1A, 2A, 3A,

4A, 4B, 5A,
6A
1A, 2A, 3A,
4A, 4B, 5A,
6A
1A, 2A, 3A,

4A, 4B, 5A,
6A, 7A
1A, 2A, 3A,
4A, 4B, 5A,
6A, 7A
Motoyama and Ueda models produced nearly the same

result. When the lightning current is sufficiently high, flashover
takes place at phase A and phase B of stricken tower before
flashovers occur at adjacent towers. At low current (I=80kA)
Motoyama model gives a conservative prediction of the
number of phase which will flashover. With I=80kA, the leader
current through the arcing horn of phase B-tower #4 in Ueda
model is about 450A, which is well below the 1000A required
to complete flashover. However, the voltage criterion of
Motoyam is fulfilled so that flashover still occurs on phase B of
tower #4. At high crest value of lightning current (I150kA),
Ueda model provided a more conservative result than predicted
by the Motoyama model.
IV.
For a standard lightning waveform, the volt-time model and

equal-area criterion of Kind are comparable, with very small
difference in time to breakdown of these models. The
Motoyama and Ueda models, which are based on voltage and
current criteria, produced conservative predictions for low
values of lightning current. At high lightning currents or for
cases which involve flashover during the wave front, the four
models become similar.
In an actual transmission line in which a standard lightning
current waveform is applied to the top of a tower, the voltage
across the arcing horn of an insulator varies from tower to
tower and from phase to phase as a result of wave propagation
and the flashover behavior of adjacent phases. As a result, the
voltage waveform across the insulators and arcing horns is no
longer standard, and the question of which arcing horns flash
over is problematic. In our actual twin circuit 220kV transmission line, the Kind model seemed to overestimate the flashover
performance of the arcing horn, while the Motoyama and Ueda
models produce more conservative predictions.
ACKNOWLEDGMENT
necessarily reflect those of VEF. Support from the Toshiba
Surge Arrester Engineering Group is grateful acknowledged.
REFERENCES
[1]
[2]
[3]
[4]
[5]
[6]
CONCLUSION
Flashover behavior of the insulator in a transmission line

depends on various factors including the number of circuits, the
distance from stricken point to the insulator under study, etc.
This study examined flashover behavior of an arcing horn
which stands alone and as used in an actual twin circuit 220kV
transmission line using ATP/EMTP simulations. Four models
of flashover were used in order to compare their behavior in
various circumstances.
[7]
[8]
[9]
Fast Front Transient Task Force of the IEEE, Modeling guidelines for
fast front transients, IEEE Trans. Power Delivery, vol. 11, No. 1,
January 1996
T. Shindo, Y. Aoshima, I. Kishizima, T. Harada, A study of
predischarge current characteristic of long air gaps, IEEE Trans. on
Power Apparatus and Systems, vol. PAS-104, No. 11, November 1985.
A. Pigini, G. Rizzi, E. Garbagnati, A. Porrino, G. Baldo, G. Pessavento,
Performance of large air gaps under lightning overvoltages :
Experimental study and analysis of accuracy of predetermination
methods , IEEE Trans. on Power Delivery, vol. 4, no. 2, April 1999.
H. Motoyama, Experimental study and analysis of breakdown
characteristics of long air gaps with short tail lightning impulse, IEEE
Trans. on Power Delivery, vol. 11, no. 2, April 1996.
CIGRE WG 33-01, Guide to procedure for estimating the lightning
procedures the lightning performance of transmission lines, October
1991.
O. Ueda, S. Neo, T. Funabashi, T. Hagiwara and H. Wanatabe,
Flashover model for arcing horns and transmission line arresters,
International Conference on Power Systems Transients, Lisbon, 3-7
September 1995.
IEC Technical Report 60071-4, Insulation co-ordination. Part 4:
Computational guide to insulation co-ordination and modeling of
electrical networks, 2004-06.
CIGRE WG 33-02, Guidelines for representation of network elements
when calculating transients, No. 39, 1990.
A. Ametani and T. Kawamura, A method of a lightning surge analysis
recommended in Japan using EMTP, IEEE Trans. on Power Delivery,
vol. 20, no. 2, April 2005.
54
eBook for You
Thus the waveform changed greatly from the previous case

in which no flashover occurs. When the lightning current is
increased, the number phases which flash over also increases
(Table 2). Unlike the case for the standard wave, the flashover
behavior for a transmission line also depends on the span
length.

IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 27, NO. 4, OCTOBER 2012
1991
Effect of Externally Gapped Line Arrester Placement

on Insulation Coordination of a Twin-Circuit
220 kV Line
AbstractApplication of externally gapped lightning arresters
(EGLAs) to a twin circuit 220 kV transmission line was simulated
using Alternative Transients Program/Electromagnetic Transients
Program. The lightning current to the tower top as well as to
phase conductors was estimated using the electrogeometric model.
Increasing the number of phases on which EGLAs were installed
in one circuit reduces the likelihood of back ashover. Installing
EGLAs on the three phases of one circuit has two advantages,
viz., 1) eliminating double circuit outages and 2) preventing lightning-induced back ashovers for strikes to the tower top 200 kA.
However, EGLAs on one circuit cannot improve lightning performance related to the shielding failure when the unprotected
circuit is hit. EGLA duty varied little with the number of phases
on which EGLAs were installed, which indicates that the required
EGLA rating is unlikely to depend on the number of phases on
which EGLAs are installed.
Index TermsElectromagnetic Transients Program/Alternative Transients Program (EMTP)/(ATP), externally gapped line
arrester (EGLA), insulation coordination, transmission line.
I. INTRODUCTION
IGHTNING performance is one of the most important aspects of transmission line design, as most transmission
line outages are caused by lightning. Lightning can strike the
shielding wire (mid-span or tower top) to create back ashovers
or strike a phase conductor as a result of shielding failure.
Shield wires, in combination low footing resistance, improve
the lightning performance of a transmission line, and the application of line arresters provides an additional increment of protection. Such arresters come in two forms, gapless [1], [2] and
externally gapped (EGLA) [3], [9], [10]. The main advantage of
an EGLA is reduced size and weight, as clearing of the series
gap after a lightning event removes the arrester from the line so
that thermal stability of the metaloxide varistors (MOVs) after
the lightning event is not an issue. In some cases, transmission
Manuscript received September 15, 2011; accepted February 04, 2012. Date
of publication September 07, 2012; date of current version September 19, 2012.
This work was supported in part by the Vietnam Education Foundation (VEF).
Paper no. TPWRD-00783-2011.
T. Pham is with Hanoi University of Science and Technology, Hanoi, Vietnam
(e-mail: thinh.pham@ims.uconn.edu).
S. A. Boggs is with the Electrical Insulation Research Center, University of
Connecticut, Storrs, CT 06269 USA (e-mail: steven.boggs@ieee.org).
H. Suzuki and T. Imai are with the Surge Arrester Department,
Hamakawasaki Operations, Toshiba Corporation, Kawasaki-ku 210-0862,
Japan (e-mail: hironori.suzuki@toshiba.co.jp; toshiya.imai@toshiba.co.jp).
Color versions of one or more of the gures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identier 10.1109/TPWRD.2012.2205729
line arresters can replace shielding wires while improving the

lightning performance of the transmission line [2].
The lightning performance of a transmission line depends on
its conguration (voltage rating, tower height, single or twin-circuit, footing resistance, etc.), the probability density for lightning current rise time and amplitude, and the ground ash density. A utility must consider many issues when contemplating
installation of line arresters, such as the acceptable frequency of
line outages, the towers most prone to lightning strikes, and the
cost/benet of installing line arresters. Installing line arresters
is a tradeoff between the reduction in the expected line outage
rate and capital expenditure. Greatest protection can be achieved
by installing a line arrester across every insulator of each circuit. However, this solution is costly and sometimes unnecessary, as the tower conguration and footing resistance are not
uniform along the transmission line, and the grounding ash
density varies with location. In this paper, we compute the lightning performance of a twin-circuit 220 kV transmission line as
a function of EGLA placement on one circuit using EMTP simulations with the lightning current estimated from the electrogeometric model. Lightning is assumed to hit a tower top or phase
conductor of the worst case tower, that is, one whose conguration is likely to determine the lightning performance of the
entire transmission line.
II. ELECTROGEOMETRIC MODEL (EGM)
The number of lightning strokes to the shielding wires and
phase conductors is estimated using the electrogeometric model
of lightning strike distance [4]. The strike distance,
, in m,
depends on the lightning current, , in kA, according to
(1)
6.72 and
0.8 for lightning to a shielding wire
where
or phase conductor, and
6.048 and
0.8 for lightning
to earth. For a vertical stroke, the shielding wire and phase conductor have a specic exposure width,
, in meters, for each
value of lightning current. The exposure width for shielding
wires
and phase conductors
is determined as shown
in Fig. 1.
The number of strokes to shielding wire or phase conductor
per year is computed by integrating the exposure width, D(I),
over the current range times the probability of that current [8]
0885-8977/$31.00 2012 IEEE
(2)
55
eBook for You
Thinh H. Pham, Steven A. Boggs, Fellow, IEEE, Hironori Suzuki, Member, IEEE, and Toshiya Imai

1992
eBook for You
Fig. 1. Tower conguration (left) and its equivalent EGM model to determine expose width of shielding wire and phase conductors. The parameters r (I), r (I),
and r (I) are computed from (1) and are used to compute the values of D and D , which are the exposure widths which enter into ((2) and (3)). The indices c, s,
and e stand for conductor, shielding wire, and earth. Figure adapted from [8].
Fig. 3. Maximum shielding failure current of as a function of tower height

determined using EGM. The protection angle remains constant, as determined
from the tower conguration of Figure 3. Only the base section of the tower
varies with changes in tower height.
Fig. 2. Transmission line lightning stroke frequency as a function of lightning
current based on the EGM model and computed for tower height of 60 m. The
assumed ground ash density is 10 per km -year. The lightning frequency to the
shielding wire is split into two parts, which correspond to low and high values
of lightning current. Low lightning current can terminate on either the shielding
wire or phase conductors, while high lightning currents can only terminate at a
shielding wire.
current which can terminate on a shielding wire or phase conductor,

is the lightning current probability density distribution which has a lognormal form [11]
(4)
For a line protected by two shielding wires which are separated by a distance , (2) [8] becomes
(3)
where
is the ground ash density, 10 per km -year in this
study,B mn
is the
length of the line,
is maximum lightning
H thng in - i hc Bch Khoa H Ni
where
1.33 and
61.1 for
20 kA, and
0.605
and
33.3 for
20 kA.
Fig. 2 shows the frequency of strokes to shielding wires and
phase conductors per 100 km-year for a 60 m high tower as
computed using EGM. For large lightning current, the strike distance increases to the point that a strike can only terminate on
shielding wires or earth. In the present tower conguration, the
probability of shielding failure is relatively small and can occur
only for relatively low currents, that is, 28 kA for the56upper

PHAM et al.: EFFECT OF EXTERNALLY GAPPED LINE ARRESTER PLACEMENT ON INSULATION COORDINATION
1993
Fig. 4. Transmission line to be modeled. In the simulation, spans from tower 12 and 67 are included along with their matching impedances which avoid reections from the ends. The phase conductor is ACSR 330 2 (aluminum conductor steel reinforced), the shielding wire (SW) is TK-70 and OPGW-82 (optical
ground wire).
phase and 9 kA for the lower phase. As a result, the number of

strokes to shielding wires (Fig. 2) is divided into two parts, one
corresponding high current strokes, which can terminate only
on a shielding wire or tower top, and the other corresponding
to the portion of low current lightning strokes which terminate
on a shielding wire, although in this current range, a stroke can
also terminate on a phase conductor. Greater tower height results in greater exposure width which increases the probability
of very high current strokes to the tower top and shielding wires.
Increased tower height shifts all curves in Fig. 2 toward the top
right corner. Maximum shielding failure current increases linearly with tower height (Fig. 3), which suggests that computation of lightning performance should be carried out as a function of tower height, for both lightning to the top of the tower
and shielding failure. From such computations, the maximum
shielding failure current can be computed as a function of tower
height for use in simulations. For the tower conguration of the
present 220 kV transmission line (Fig. 3), a maximum shielding
failure current of 30 kA is reasonable, from which the required
EGLA rating can be estimated.
III. 220-kV TRANSMISSION LINE AND EMTP MODELS
The twin circuit 220 kV transmission line section shown in
Fig. 4 was analyzed. The transmission line section includes 6
phase conductors, 2 shielding wires, and 6 spans of 400 m each
between 7 towers. Each phase consists of two conductors bundled with a separation of 40 cm. Insulators are normally protected by arcing horns with an air gap of 2.19 m except where
replaced by an EGLA which has a series gap of 1 m. A frequency dependent J-Marti model of 8 conductors was used to
simulate each span of the transmission line. In order to avoid
reections from the ends of the model, the power sources are
connected to the ends through a resistance matrix which represents the line impedances at
Hz. As lightning can strike the
line at any time during a power frequency cycle, the power frequency voltage is included in the simulation at 0 degrees for the
upper phase, the superposition of which on the lightning strike
resultsBinmn
theH
lower
potential difference between the cross arm
thng in - i hc Bch Khoa H Ni
eBook for You
Fig. 5. Potential difference across the upper phase insulator of the stricken
tower when lightning strikes the top of tower 4 for various values of adjacent
tower footing resistance. The footing resistance of stricken tower is 20
, and
the lightning current is 100 kA with linear ramp waveform (2/70 s).
and the upper phase conductor than for the middle and lower
phases, which results in greater probability of the ashover of
the lower and middle phase insulators than the upper phase insulator. However, this difference only changes the order in which
the phases ash over but not the total number of phases which
ash in a given simulation, which is the most important parameter when implementing EGLAs in a transmission line.
The tower is represented by four lossless transmission lines,
and the footing impedance of each tower is a simple resistance
as suggested in [6]. The ashover of the EGLA series gap or
the arcing horn gap is simulated by the Motoyama model for
reasons discussed in [5], [7].
Tower footing resistance varies along a transmission line depending on the soil resistivity. When lightning strikes to the
tower top or the phase conductor, the overvoltage across an insulator depends on both the footing resistance of the stricken
tower and that of the adjacent towers. Fig. 5 shows the dependence of overvoltage across an insulator of the stricken tower
on
57

1994
TABLE I
INSULATORS WHICH FLASHOVER AS A FUNCTION OF 2/70 S LIGHTNING CURRENT AMPLITUDE FOR LIGHTNING
TO THE TOP OF TOWER 4 WITH 60 M TOWER HEIGHT, 30
FOOTING
IV. SIMULATION RESULTS

A. Towers on Which Flashover Occurs Without Arresters
In order to estimate the impact of lightning on the transmission line without the use of surge arresters, EMTP simulations
were performed on the line segment for lightning currents to the
tower top ranging from 50 kA to 300 kA and for 10 kA to 30 kA
to the upper phase conductor as a result of shielding failure. In
case of shielding failure, ashover occurs on the stricken phase
only up to 30 kA. The adjacent phases of the stricken tower and
the upper phase of the adjacent towers do not ash.
Table I shows the towers and phases on which breakdown occurs as a function of lightning current to the tower top. Lightning
to the top of tower 4 (or any other tower) causes ashover of insulators only if the lightning current is 80 kA (2/70 s). As a
result of the transmission line symmetry, an 80 kA strike has a
high probability of causing a double circuit outage. Above 90
kA, ashover occurs on two phases of the stricken tower. Most
of lightning energy is dissipated through these two phases so that
the third phase does not ash. As noted in Section III, ashover
of middle and lower phases is the result of the assumed power
frequency phase at the time of the lightning strike. For a 120 kA
lightning
strike to the tower top, ashover can occur on all three
phases of the stricken tower. Above 170 kA, ashover occurs on

at least the two adjacent towers. Above 180 kA, the contribution
of power frequency voltage to the total overvoltage is negligible
in comparison with that of lightning current. Therefore insulators of upper and middle phases of adjacent towers break down
following ashover of all three phases of the stricken tower. At
very high lightning current, insulators of all three phases of two
adjacent towers ashover.
B. Simulations in the Presence of Surge Arresters
Based on the discussion in the previous section, installation
of lightning arresters is necessary to avoid insulator ashover
for lightning strikes greater than 80 kA to tower top and 10 kA
to a phase conductor. As indicated in Section II, the maximum
shielding failure current is greater for a strike to the upper and
middle phases, and installation of a surge arrester on these
phases prevents shielding failure-induced ashovers. Since the
position of stricken tower is random, surge arresters should be
installed on the upper and middle phases of one circuit over the
entire transmission line.
Simulations without surge arresters installed indicate that installation of surge arresters on two upper phases may prevent
back ashover of the 220 kV transmission line up to 120 kA to
the tower top (Table III), for a footing resistance 30 . Above
120 kA, a surge arrester is necessary on each phase of every
tower to avoid back ashover. In this section, data are reported
for simulations with EGLAs on circuit 1. The number of EGLAs
is increased from one per tower (installed on the upper phase
only) to three per tower (installed on all three phases of circuit
1). Tables IIIV document the improvement in lightning performance for the twin circuit 220 kV transmission line as a function
of the number of EGLAs installed on one circuit.
Installing EGLAs on the upper phase of one circuit does
not improve the lightning performance of the transmission
line signicantly, as insulator ashover on the stricken tower
increases from 80 kA without an arrester to 90 kA with one
EGLA installed per tower. Above 100 kA, insulators of middle
and lower phases in second circuit (without surge arresters)
ashover (Table II). The presence of EGLAs on the upper phase
of one circuit increases the strike current at which a double
circuit ashover occurs from 80 kA without EGLAs to 120 kA
with EGLAs on the upper phase of one circuit.
58
eBook for You
the footing resistance of adjacent towers, given a 20 footing

resistance for the stricken tower and lightning current of 100
kA (2/70 s, triangular waveform). Increased footing resistance
of adjacent towers results in a moderate increase of the overvoltage in wave tail of all three phases. The overvoltage wave
front during which ashover of the EGLA gap (or arcing horn)
is likely to occur [5] does not change with the footing resistance of adjacent towers, as twice the electromagnetic propagation time between towers is comparable to the voltage rise time.
This behavior suggests that simulation with a constant footing
resistance along the transmission line is a reasonable approach
to study the lightning performance of the transmission line. In
an actual 220 kV transmission line, the footing resistance is usually less than 10 and rarely exceeds 30 except at a few locations such as hard mountain rock. For new transmission lines,
the footing resistance is usually kept around 5 . A footing resistance of 30 used in many of the present simulations is therefore reasonable for estimating the worst case lightning performance of the transmission line.

1995
TABLE II
FLASHOVER OF INSULATORS AND OPERATION OF SURGE ARRESTERS AS A FUNCTION OF 2/70 S LIGHTNING CURRENT AMPLITUDE FOR LIGHTNING TO THE TOP
OF TOWER 4 WITH 60 m TOWER HEIGHT, 30
FOOTING RESISTANCE, AND EGLAS INSTALLED ON THE UPPER PHASE OF CIRCUIT 1
TABLE IV
FOOTING RESISTANCE, AND EGLAS INSTALLED ON ALL THREE PHASES OF CIRCUIT 1
Installing EGLAs on upper and middle phases of one circuit protects the insulators of both circuits from ashover up to
120 kA to the tower top. Flashover of the lower phase, which
is not protected by an EGLA, occurs only after ashover of
two phases of second, unprotected circuit for lighting currents
greater than 130 kA. Thus installation of EGLAs on the upper
and middle phases of one circuit prevents a double circuit outage
up to 130 kA. Where the probability of lightning currents above
120 kA is negligible, installation of EGLAs on only the upper
two phases of one circuit may be justied. For the transmission
line in question, the frequency of greater than 120 kA lightning
current to a shielding wire is about 7 per 100 km-year (Fig. 2),
which means 0.8 times per year for a 10 km transmission line,
which is relatively low given that most of towers on an actual
220 kV transmission line are likely to be closer to 40 m high
rather than to 60 m assumed before, with only a small fraction
of the towers near 60 m. Above 150 kA, ashover occurs across
the lower phase insulators of circuit 1 and all three phases of
circuitB
2 mn
(Table
III).
H thng in - i hc Bch Khoa H Ni
Installing EGLAs on all three phases of circuit 1 does not

eliminate insulator ashover completely. At 150 kA lightning
current to the tower top, insulators of the middle and lower
phases of circuit 2 of the stricken tower ashover (Table IV).
The advantage of installing EGLAs on all three phases compared to installing them on only two phases is that the probability of a double circuit ashover is reduced to near zero, as one
circuit is totally protected by EGLAs. To protect the line against
ashover for lightning currents above 150 kA to the shielding
wire, each phase of both circuits must be protected by an EGLA.
As mentioned in Section III, reducing the assumed footing
resistance to 5 or 10 , in combination with installation of
EGLAs, improves lightning performance of the transmission
line substantially. Fig. 6 shows the lightning current threshold
above which ashover occurs on at least one insulator, as a
function of footing resistance. At 30 footing resistance with
EGLAs on all three phases of one circuit, a 120 kA of lightning
strike does not cause insulator ashover. For a footing resistance of 5 , this current increases to 200 kA. The frequency
of
59
eBook for You
TABLE III
FOOTING RESISTANCE, AND EGLAS INSTALLED ON UPPER AND MIDDLE PHASES OF CIRCUIT 1

1996
Fig. 6. Lightning current threshold above which insulator ashover occurs as a

function of footing resistance, for four cases: without EGLAs, EGLAs installed
on the upper phase of one circuit, EGLAs installed on the upper and middle
phases of one circuit, and EGLAs installed on all three phases of one circuit for
the 60 m tower. Lightning strikes to the top of tower 4.
Fig. 7. EGLA discharge current as a function of lightning current for a different

number of phases of one circuit on which EGLAs were installed, for 60 m tower
height and 30
footing resistance. Lightning strikes to the top of tower 4.
a 200 kA lightning strike in our case is 0.01 per year-100 km,

which is very small.
The lightning performance of the transmission line, as dened by the lightning current threshold for insulator ashover,
increases gradually with the number of phases on which EGLAs
are installed, which suggests that installation of EGLAs on all
phases of a twin circuit, 220 kV transmission line may not be
necessary, except at a few towers with very high footing resistance. For relatively low footing resistance, installing EGLAs on
all phases of one circuit is sufcient to inhibit insulator ashover
from lighting to the tower top or to the shielding wire.
Increasing the number of EGLAs installed on one circuit
changes EGLA duty slightly. Fig. 7 shows the effect of the
number of EGLAs on the maximum discharge current through
an EGLA for a 60 m tower with 30 footing resistance. The
discharge current through the EGLA on the upper phase of
circuit 1 of the stricken tower is greatest for the entire range
of lightning current, from 50 kA to 300 kA. The variation
in discharge current with the number of EGLAs installed is
negligible up to 200 kA. The maximum energy absorbed by
an EGLA on circuit 1 increases with the number of EGLAs
installed on one circuit (Fig. 8), and the effect of the number
of EGLAs is more pronounced at high lightning current. Absorption energy nearly triples when the number of EGLAs is
increased from 1 to 3. However, the absorption energy is always
below 200 kJ, the rating of the lightest duty EGLA employed
in 220BkV
transmission lines.
mn H thng in - i hc Bch Khoa H Ni
Fig. 8. Worst case energy absorbed by an EGLA installed on one circuit as a

function of lightning current for a differing number of phases of one circuit on
which EGLAs were installed, for 60 m tower height and 30
footing resistance.
Lightning strikes to the top of tower 4.
Although installing EGLAs on three phases of one circuit

at every tower can protect the transmission line from back
ashover up to 150 kA, it does not provide equally complete
protection against shielding failure. A lighting strike to a
conductor not protected by an EGLA (i.e., any phase of circuit
2) results in approximately the same overvoltage as if the line
were not protected by EGLAs. For lightning to a phase conductor protected by an EGLA, the simulations indicate that up
to 30 kA, only the EGLAs of the stricken phase operate at the
stricken tower and two nearest towers (Table V). Fortunately,
60
eBook for You
TABLE V
FLASHOVER OF INSULATORS AND OPERATION OF SURGE ARRESTERS AS A FUNCTION OF 2/70 S LIGHTNING CURRENT AMPLITUDE FOR A LIGHTNING STRIKE TO
THE UPPER PHASE OF TOWER 4 WITH 60 m TOWER HEIGHT, 30
FOOTING RESISTANCE, AND EGLAS INSTALLED ON ALL THREE PHASES OF CIRCUIT 1

V. CONCLUSION
Simulations were performed for a twin circuit, 220 kV
transmission line to estimate the effectiveness of EGLAs as a
function of the number of EGLAs installed, increasing from
0 (without EGLAs) to 3 (EGLAs on all three phases of one
circuit). The EGM model was used to estimate the lightning
frequency to the tower top (or shielding wire) and to phase
conductors.
In case of lightning to a tower top, installing EGLAs on one
or two phases of one circuit improves the lightning performance
of the transmission line substantially. However a double circuit outage can still occur at very high lightning current, and/or
with high footing resistance. Installing EGLAs with appropriate
rating on all three phases of one circuit reduces the likelihood of
a double circuit outage to near zero for lightning to tower top,
but insulator ashover on the unprotected parallel circuit still
occurs for very high lightning current and/or high footing resistance. Installing EGLAs on one or more phases of the second
circuit should be considered at towers with high footing resistance ( 10 ), which is unusual for most transmission lines.
The number of EGLAs installed on one circuit has a small effect
on the maximum duty of a single EGLA (discharge current and
energy absorption) which suggests that the same rating EGLA
should be selected over the entire transmission line no matter
how many EGLAs are installed per circuit.
The lightning current and strike frequency for shielding failures (strikes to a conductor) is much lower than for strikes to the
shielding wires and tower top. However, a lightning strike to a
phase conductor can cause insulator ashover on the stricken
phase if it is not protected by an EGLA, although a double circuit outage is unlikely to occur if EGLAs are installed on each
phase of one circuit. Given the very low probability of shielding
failure, the risk of a single circuit outage may be worth taking.
REFERENCES
[1] T. Wakai, N. Itamoto, T. Sakai, and M. Ishii, Evaluation of transmission line arresters against winter lightning, IEEE Trans. Power Del.,
vol. 15, no. 2, pp. 684690, Apr. 2000.
[2] Y. Matsumoto, O. Sakuma, K. Shinjo, M. Saiki, T. Wakai, T. Sakai, H.
Nagasaka, and H. Motoyama, Measurement of lightning surges on test
transmission line equipped with arresters struck by natural and trigged
lightning, IEEE Trans. Power Del., vol. 11, no. 2, pp. 9961001, Apr.
1996.
[3] S. Furukawa, O. Usuda, T. Isozaki, and T. Irie, Development and
application of lightning arresters for transmission lines, IEEE Trans.
Power Del., vol. 4, no. 4, pp. 21212128, Oct. 1989.
[4] H. R. Amstrong and E. R. Whitehead, Field and analytical studies
of transmission line shielding, IEEE Trans. Power App. Syst., vol.
PAS-87, no. 1, pp. 270281, Jan. 1968.
[5] T. Pham, S. Boggs, T. Pham, and S. Boggs, Flashover of arcing horn

in transient simulation, presented at the IEEE Int. Symp. Elect. Insul.,
San Diego, CA, Jun. 59, 2010.
[6] A. Ametani and T. Kawamura, A method of a lightning surge analysis
recommended in japan using EMTP, IEEE Trans. Power Del., vol. 20,
no. 2, pp. 867875, Apr. 2005.
[7] H. Motoyama, Experimental study and analysis of breakdown characteristics of long air gaps with short tail lightning impulse, IEEE Trans.
Power Del., vol. 11, no. 2, pp. 972979, Apr. 1996.
[8] A. R. Hileman, Insulation Coordination for Power Systems. Boca
Raton, FL: CRC, 1999.
[9] R. Hernandez-Corona and G. Regaldo-Contreras, Performance of 230
kV transmission lines applying externally gapped type arresters, presented at the IEEE/Power Eng. Soc. Transm. Distrib. Conf. Expo., Atlanta, GA, Oct. 28Nov. 2, 2001.
[10] T. Shigeno, Experience and effectiveness of application of transmission line arresters, in Proc. IEEE/Power Eng. Soc. Transm. Distrib.
Conf. Exhibit.: Asia Pacic, Oct. 2002, pp. 610.
[11] IEEE Guide for Improving the Lightning Performance of Transmission
Lines, IEEE Standard 12431997, 1997.
Thinh H. Pham received the B.S. degree in power system engineering from
Hanoi University of Science and Technology, Hanoi, Vietnam, in 1996, the
M.Sc. degree in electrical engineering from Grenoble Institute of Technology,
Grenoble, France, in 2001, and the Ph.D. degree in electrical engineering from
University of Joseph Fourier, Grenoble, in 2005.
Since 2006, he has been a faculty member in the Department of Power Systems, Hanoi University of Science and Technology. From 2009 to 2011, he
was with the Electrical Insulation Research Center (EIRC), University of Connecticut, as a Visiting Professor and then a Postdoctoral Fellow. His research
interests focus on transient phenomena and insulation coordination in power
systems and dielectric properties of insulation subjected to high elds.
Steven A. Boggs (F92) was graduated with a B.A. degree from Reed College,
Portland, OR, in 1968, and the Ph.D. and MBA degrees from the University of
Toronto, Toronto, ON, Canada, in 1972 and 1987, respectively.
He spent 12 years with the Research Division of Ontario Hydro. He was
elected an IEEE Fellow for his contributions to understanding of SF -insulated
systems. From 1987 to 1993, he was Director of Research and Engineering at
Underground Systems, Inc. He is presently Director of the Electrical Insulation
Research Center and the Research Professor of Materials Science, Electrical
Engineering, and Physics at the University of Connecticut, Storrs, and Adjunct
Professor of Electrical Engineering at the University of Toronto.
Hironori Suzuki (M12) was born on June 29, 1959. He received the B.S. and
M.S. degrees in physics from the University of Tsukuba, Ibaraki, Japan, in 1982
and 1984, respectively.
In 1984, he joined Toshiba Corporation, Kawasaki, Japan. Since then, he
has been engaged in the development of zincoxide elements and surge arresters. Currently, he is an Assistant Secretary of the standard committee for
metaloxide surge arresters with the Institute of Electrical Engineers of Japan.
Toshiya Imai was born in Hyogo Prefecture, Kansai, Japan, in 1966. He received the B.S. degree in inorganic material engineering from Kyoto Institute
of Technology, Kyoto, Japan, in 1989.
He joined Toshiba Corporation, Kawasaki, Japan, in 1989. Since then, he has
been engaged in the development of zincoxide elements and metal-oxide surge
arresters.
61
eBook for You
the frequency of shielding failures is relatively low in this tower

conguration (i.e., half that shown Fig. 2) since one circuit is
already protected by EGLAs.
1997
European Association for the

Development of Renewable Energies, Environment
and Power Quality (EA4EPQ)
International Conference on Renewable Energies and Power Quality

(ICREPQ12)
Santiago de Compostela (Spain), 28th to 30th March, 2012
An optimization power method for Photovoltaic array by tracking Maximum

Average Current through Diode of Boost Converter
X.L. Dang1, E. Hoang1, M. Petit2, H.T. Pham3 and H. Ben Ahmed4
1
SATIE, ENS de Cachan, CNRS, UniverSud, 61, av President Wilson F-94230-Cachan, France.
Department of Energy, Suplec, Campus de Gif-sur_Yvette, 91192 GIF SUR YVETTE, France.
3
Department of Power System, Hanoi University of Technology, Hanoi, Vietnam.
4
SATIE-Brittany Branch, ENS de Cachan-Ker Lann Campus, 35170 BRUZ, France.
E-mail: xdang@satie.ens-cachan.fr
Abstract. In this paper, a method to optimize the power

delivered by the Photovoltaic array (PV array) is presented by
tracking the Maximum Average Current (MAC) passing through
the diode of DC/DC boost converter. The main difference
between the proposed method and the other Maximum Power
Point Tracking (MPPT) methods is that instead of tracking
maximum power point directly on PV array, this method enables
to search the maximum power of PV array by tracking the
maximum power output of the DC/DC converter with only one
current sensor. An hybrid power system, which includes a PV
array, a DC/DC boost converter, an ideal battery and a resistor, is
built and simulated in MATLAB/Simulink. The comparison
between analytical and simulated results has proven that this
method could be applied under various irradiation and
photovoltaic temperature conditions.
maximum power from the PV array. In the last few

decades, some methods have been proposed in [4]-[8] for
tracking the maximum output power from the PV array in
the last few decades. Once the PV array is connected to
other elements in the hybrid system such as the battery
and the load, it is required to investigate a method that
could optimize the electric power delivered by the PV
array. The voltage over the battery is assumed constant
during the calculation period.
Key words
Fig. 1. Overview circuit of the PV array system
Hybrid Power System, Photovoltaic array (PV array),

Maximum Average Current (MAC), DC/DC Boost
Converter, Maximum Power Point Tracking (MPPT).
1. Introduction
In recent years, the development on the installed capacity
and the improvement on the efficiency of the renewable
energy sources are implemented in many countries around
the world due to the exhaustion and pollution of fossil
fuels. One of the most attractive renewable energy sources
is photovoltaic array, because PV is the most direct way to
convert solar radiation into electricity and is based on the
photovoltaic effect [2]. Otherwise, there is no mechanical
moving parts, no noise, no pollution and PV panels have a
long lifetime. Moreover, the energy source is the sun
which is free, ubiquitous and inexhaustible. In some
popular appliances, it is usually used in charging battery,
hybrid vehicles, water pumping and connecting into
Distributed Energy Resources (DER) systems.
As a power source generated by solar array depending so
much on the solar radiation, temperature and array voltage,
it is essential to control the operating points to yield the
Due to the ability of keeping the voltage over the battery

(and also the output voltage of DC/DC Boost Converter)
constant during the simulation time, it is necessary to
track the maximum current passes through the diode of
DC/DC Boost Converter in order to optimize the electric
power from PV array that supplies to the battery and the
load. However, this current is not a continuous one which
depends on the duty cycle of the switch of the Boost
Converter. Hence the average current value through the
diode is considered as a parameter to optimize the power
from PV array. This is the main idea of the proposed
method which is focused on tracking the MAC through
diode.
2. Photovoltaic array model

The general current voltage (I-V) characteristic of a
solar cell is given by the following equation [2]:
I PV
qVPV I PV RS (VPV I PV RS )
1
I PH I S exp
kTC A
RSH

(1)
62
eBook for You
For an array model, the series resistance has a large impact

on the slope of the I-V curve near the open-circuit voltage
and the shunt down resistances approaches infinity which
is assumed to be open [3]. The mathematical equation
model [2] to express the relation between current and
voltage of the photovoltaic array is represented by the
following equation:
I PV
qVPV N S I PV RS N P
1
N P I PH N P I S exp
kTC A

(2)
Where NS is series number of cells of PV array and Np is

series-parallel number of PV array.
It is well known that the output current and power
characteristics of PV array are non-linear and vitally
affected by the solar radiation and the panel temperature.
Each curve in the P-V characteristics has a maximum
power point, which is the optimal operating point for the
efficient use of the PV array.
in the opposite direction known as the reverse direction,

lead to:
i D t 0
(4)
When the converter operates in steady-state, neglecting
losses in these elements of the converter and assuming
that the converter is perfectly efficient, the average
current value or dc component through the diode is given
by the equation below:
I PV 1 D
iD
From the equations (2) and (5), this average current is

defined in steady-state as:
ID
qVPV N S I PV RS N P
1 1 D
kTC A
VOUT 1 D
VIN
(7)
From the Fig.2, in steady-state, VOUT = VBatt and VIN =

VPV, and replacing these voltages into equation (7)
yields:
VPV
Fig. 2. Detailed circuit of the proposed PV array system
(6)
Otherwise, the relation between voltage output and

voltage input of the DC/DC boost converter is expressed
by the following equation:
3. The proposed MAC method

Most MPPT techniques attempt to track the PV voltage
that results in the maximum power point V MPP, or to find
the PV current IMPP corresponding to the maximum power
point. The proposed method searches for the maximum
average current value passing through the diode in the
DC/DC Boost Converter by changing the value of duty
cycle ( ) of switch IGBT from 0 to 1. Where T
is the switching period (equal to the inverse of the
switching frequency fS), generally lies in the range of 1
kHz to 1 MHz [1], which depends on the switching speed
of the semiconductor devices.
(5)
VBatt1 D
(8)
Introducing the equation (8) into equation (6), the current

could be expressed as following:
ID
qVBatt1 D N S I PV RS N P
1 1 D
kTC A
(9)
Thus the equation (9) gives the expression of the mean

diode current as function of the duty cycle. For the sake
of doing that, a proposed MAC model is designed by
using MATLAB/Simulink and taking into account
conducting losses in the IGBT, diode and inductor. From
Fig. 4(a) and Fig. 4(b), it is noticed that there is always
one value of the duty cycle for which the maximum of
the average current through diode is achieved.
In addition, a Tracking and Holding Maximum Value
(THMV) model is built to detect the maximum current
and the duty cycle value at this maximum point as shown
in Fig. (3).
From the Fig.2 above, when the switch IGBT is off, the
diode is connected to the inductor. Using the small-ripple
approximation, the inductor current is equal to the output
current of PV array, so the current through the diode iD(t)
is obtained:
Average Current (A)
2.9
2.8
2.7
2.6
2.5
0
0.2
0.4
0.6
0.8
1.2
1.4
1.6
1.8
1.2
1.4
1.6
1.8
time (s)
0.25
i D t i PV t
(3)
Once the IGBT is on, the inductor connects parallel with

the previous capacitor and the diode is blocked the current
Duty cycle
0.2
0.15
0.1
0.05
0
0
0.2
0.4
0.6
0.8
time (s)
Fig.3. Tracking and Holding Maximum Value
63
eBook for You
Where IPH is the light-generated current or photocurrent, IS

is the cell saturation of dark current, q is the electron
charge (q=1.610-19 C), k is the Boltzmanns constant
(k=1.3810-23 J/K), TC is the cells working temperature, A
is an ideal factor, RSH is the shunt resistance and RS is the
series resistance.
3
2
O = 0.9 (kW/m )
2.5
Tc=15oC
2.5
O = 0.6 (kW/m2)
Tc=35oC
Tc=55oC
O = 0.3 (kW/m )
<ID>(A)
1.5
1.5
0.5
0.5
0
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0
0
0.1
0.2
0.3
0.4
Duty cycle of the switch (D=ton/T)
Fig. 4(a). Average current waveform through

the diode with different solar radiations ()
0.7
0.8
0.9
modulating signal
1
Tc=15oC
0.5
Tc=35oC
Tc=55oC
0
-0.5
-1
0
Under three different atmospheric conditions with solar

radiations ( = 300W/m2; 600W/m2 and 900W/m2) and
ambiance temperatures (TC = 15oC; 35oC and 55oC). the
reference modulating signal for controlling the switch
IGBT and the average current through the diode are shown
as the simulation results in Fig. 5(a) and Fig. 5(b).
2
O = 0.9 (kW/m )
O = 0.3 (kW/m2)
O = 0.6 (kW/m2)
2
Tc=15oC
Tc=35oC
Fig. 5(b). Tracking and Holding the MAC

value through the diode with different temperatures
Additionally, the results of tracking the maximum value

are impacted by the period in Repeating Sequence block
and the averaging period in Mean Value. Besides, it
should be noted that in these figures, there are some
ripples of the average current which are unavoidable
when using the boost converter.
The instantaneous value of PV array power is shown in
Fig. 6 with = 900W/m2, it is obvious that the proposed
MAC method could enable tracking the maximum power
of PV array by tracking maximum average current
through the diode.
3
0
0
time (s)
60
40
20
0
0
-0.5
time (s)
Power (W)
1
0.5
Tc=55oC
<ID> (A)
The current from the diode is sent to a Mean Value block

in order to measure the average current through the diode.
And then, this average current is led to the THMV block to
find and then kept the maximum value during the given
period.
time (s)
0
0
The load are a resistor and a battery which is charged by

the boost chopper. The Pulse Width Modulator (PWM)
Generator block was used to control the IGBT with a
switching frequency fs=1 kHz. A Repeating Sequence
block was added with the triangular waveform to create
the value of duty cycle of switch change from 0 to 1 and
the amplitude of the waveform inside the block is set from
-1 to 1 during a given period. Furthermore, this block acts
as a reference modulating signal that is compared to the
triangular carrier waveform inside the PWM generator.
<ID> (A)
The detailed power circuit of the proposed solar array

system is shown in Fig. 7. The system consists of a nonlinear power source (PV array), a transferring power unit
(DC/DC boost converter), a storage energy unit (battery)
and a consumption energy element (a resistor). This
system has been simulated in MATLAB/Simulink to
verify the proposed MAC method.
modulating signal
0.6
Fig. 4(b). Average current waveform through

the diode with different temperatures (Tc)
4. Simulation model and results
time (s)
-1
0
time (s)
3
<ID> (A)
0.5
Duty cycle of the switch (D=ton/T)
O = 0.9 (kW/m2)
O = 0.6 (kW/m2)
2
O = 0.3 (kW/m )
0
0
Fig. 6. Tracking and Holding maximum power of PV array by

the proposed MAC method.
time (s)
Fig. 5(a). Tracking and Holding the MAC

value through the diode with different solar radiations
Table I shows that there is a little difference between the

simulation results and the MAC model results. The
difference between two models is due to the given period
in the Repeating Sequence and the Mean Value.
However, this could be reduced by increasing the given
period in these blocks. And another impacted factors are
64
eBook for You
<ID> (A)
Reference
[1]
Table I (a). Comparison results between MAC and Simulation

modesl with different solar radiations
Solar radiation
(kW/m2)
0.3
0.6
0.9
MAC
Simulation
0.93
1.91
2.87
0.88
1.82
2.74
Error
(%)
5.38
4.71
4.53
[2]
[3]
[4]
Table I (b). Comparison results between MAC and Simulation
models with different temperatures
Temperature
(oC)
15
35
55
MAC
Simulation
2.87
2.42
1.95
2.74
2.28
1.84
Error
(%)
4.53
5.79
5.64
[5]
5. Conclusion
[6]
The proposed MAC method allows optimizing the electric
power which is generated from the PV array close to the
maximum power point of the array under changing
atmospheric conditions.
In this paper, the difference between the MPPT methods
and the proposed MAC method were proven by searching
the maximum average current passing through the diode.
Furthermore, the simulation model takes into account the
conducting losses in the inductor, diode and switch IGBT.
The losses generate the fluctuation during the first few
seconds of the holding maximum value period.
[7]
[8]
R.W. Erickson, D. Maksimovic, "Fundametal of

Power Electronic", Kluwer Academic Publishers,
New York (2000), pp.13-35
T.Markvart, "Solar Electricity", JohnWilly & Sons,
Chichester, England (2000), pp.23-73
O. Wasynezuk, "Dynamic Behavior of a Class of
Photovoltaic Power Systems", IEEE Transactions on
Power Apparatus and Systems, Sept. 1983, vol.102,
no.9, pp.3031-3037
M. Veerachary, T. Senjyu, K. Uezato, "Voltagebased maximum power point tracking control of PV
system", IEEE Transactions on Aerospace and
Electronic Systems, Jan 2002, vol.38, no.1, pp.262270
Chihchiang Hua, Chihming Shen, "Study of
maximum power tracking techniques and control of
DC/DC converters for photovoltaic power system,"
Power Electronics Specialists Conference, 1998.
PESC 98 Record. 29th Annual IEEE , 17-22 May
1998, vol.1, pp.86-93
N. Femia, Petrone, G.Spagnuolo, M. Vitelli,
"Optimization of perturb and observe maximum
power point tracking method," IEEE Transactions
on Power Electronics, July 2005, vol.20, no.4, pp.
963- 973
S. Jain, V. Agarwal, "A new algorithm for rapid
tracking of approximate maximum power point in
photovoltaic systems," Power Electronics Letters,
IEEE , March 2004, vol.2, no.1, pp. 16- 19
Yeong-Chau Kuo, Tsorng-Juu Liang, Jiann-Fuh
"Novel
maximum-power-point-tracking
Chen,
controller for photovoltaic energy conversion
system,", IEEE Transactions on Industrial
Electronics, Jun 2001, vol.48, no.3, pp.594-601
Fig.7. The MAC simulation
65
eBook for You
the small ripples and the losses caused by the elements

inside boost converter.
IEEE Transactions on Dielectrics and Electrical Insulation
Vol. 16, No. 1; February 2009
Water Absorption in a Glass-Mica-Epoxy Composite

I: Influence on Electrical Properties
T. Pham Hong, O. Lesaint and P. Gonon
Grenoble Electrical Engineering Laboratory (G2E lab), CNRS and Grenoble University
CNRS, 25 Rue des Martyrs, 38042 Grenoble, BP 166, France
Index Terms Epoxy composite, water absorption, Ficks law, dielectric

spectroscopy, breakdown, diagnostic method
1 INTRODUCTION
THE influence of water uptake on dielectric properties of
epoxy-glass-mica composite materials has been previously
reported in several papers. In such materials, a silane
treatment of glass is usually performed to reduce the
degradation occurring at interfaces with the epoxy matrix. The
degradations induced by water mostly appear to be localised at
interfaces between basic components (epoxy, glass and mica),
due to the enhancement of chemical reactions [1]. Electrical
measurements show that the presence of moisture produces an
increase of permittivity and loss tangent [2, 3], growth of trees
[4, 5], and can reduce the breakdown strength [2].
Water impregnation of epoxy composites was used to study
the ageing processes induced by humidity [6]. In large
generators, epoxy composites are used for the insulation of
stator bars. In service conditions, many different degradation
processes may occur and interact [7]. Thermal, electrical and
mechanical stresses mainly contribute to the ageing of bar
insulation. In the case of water-cooled stator bars, water leaks
although very rare may also constitute a factor of premature
ageing, in addition to the causes mentioned above. Analysis of
water-permeated bars indicate local water contents in the
insulation higher than 1% close to copper strands, associated
to a large reduction of insulation life [8]. For practical
purposes, it is very important to characterize and understand
Manuscript received on 2 June 2008, in final form 17 September 2008.
the effect of water on electrical properties. The on-site

detection of water-permeated bars constitutes another
important practical issue.
This paper presents the results of investigations carried out
on insulation samples, in order to characterize water migration
processes, and degradation of dielectric properties (H, H,
breakdown field) versus water content. The influence of
temperature and electric field on H and H is also reported.
In a second paper, these data are used to calculate the field
distribution in actual stator bar geometry, and an on-site
diagnostic method of bars is derived from these calculations.
2 WATER ABSORPTION
2.1 MATERIAL AND SAMPLE PREPARATION
The epoxy composite studied here is currently used in stator
bar insulation of very large generators (some 100 MW). The
material has a laminated structure consisting of alternating
layers of glass cloth and mica, binded together by an epoxy
resin (diglycidyl ether of bisphenol A DGEBA - cured with
a polyester acid). Figure 1a shows a SEM (Secondary Electron
Microscopy) picture of the composite cross-section.
Thicknesses of the mica and glass cloth layers are about 100
m and 50 m respectively. After burning out the resin at 600
C, the residual mass corresponding to mica and glass was
measured. It accounts for about 67% of the total composite
1070-9878/09/$25.00 2009 IEEE
Authorized licensed use limited to: La Van Ut. Downloaded on February 23, 2009 at 01:30 from IEEE Xplore. Restrictions apply.
66
eBook for You
ABSTRACT
Water absorption in an epoxy composite used for insulation of stator bars in large
generators is investigated in order to study both diffusion kinetics and effects on
dielectric properties. Water absorption can be described by the superposition of two
Fickian mechanisms, the first occurring within epoxy and the other at interfaces
between epoxy and fillers. A large decrease of breakdown strength versus water uptake
is observed. It points to the large deleterious effect of water on insulation reliability.
Dielectric spectroscopy is performed under various conditions of water impregnation,
electric field and temperature. A very large increase of permittivity and losses is
recorded in the presence of water, especially at low frequency. At power frequency,
dielectric properties do not show a non-linearity when the field is increased up to
service values. At very low frequency, a decrease of losses when the field is increased is
observed. It is due to the Garton effect. Most of results show that the wet material
behaves as a low frequency dispersion system (LFD).
T. P. Hong et al.: Water Absorption in a Glass-Mica-Epoxy Composite I: Influence on Electrical Properties
mass. Glass cloth (Figure 1b) is made of woven glass fiber

yarns. Each yarn (thickness 50 m, width 500 m) is
composed of 10 m glass fibers. Woven yarns form an open
mesh structure ( 0.5 mm x 1.5 mm). Mica is constituted by
numerous flakes of irregular shapes (Figure 1c), about 10m
in thickness and several 100 m in width.
glass
mica
accuracy of weighing, dry specimens are considered to have

a water content lower than 0.01%.
After drying, samples were immersed in deionized water
with a conductivity of 1.4 PS/cm. Water absorption versus
immersion time was monitored by weighing. Samples were
periodically removed from the water, and placed for 10 mn in
a ventilated oven at 50 C in order to remove the excess water
remaining at surfaces and obtain a stable water content during
measurements. The percentage of weight change is
determined as follows:
m m0
.100(%)
m0
(1)
glass cloth
mica layer
2.2. EFFECT OF TEMPERATURE

Samples of 1mm thickness were immersed at three different
temperatures of 60, 90 and 110 C. The high ratio of diameter
(50 mm) to thickness (1 mm) of samples allows us to consider
that water enters predominantly through the flat surfaces.
Figure 2 shows the increase of weight percentage versus the
square root of time (t1/2) for three temperatures. In all samples,
saturation occurs at about 1.5%. The variation of water uptake
kinetics with temperature clearly shows that the diffusion
process is thermally activated. Saturation was observed after
100 h at 110 C, and 2 months at 60 C. At room temperature,
a negligible increase of weight was recorded during the same
duration. The amount of absorbed water at saturation (mf) is
almost constant whatever the temperature. This suggests that
no important change of water uptake mechanisms occurs up to
110 C.
The initial stages of water absorption plots are depicted in
Figure 3. The linear increase of weight versus the square root
of time suggests that water uptake follows the classical
Fickian behavior. The apparent diffusion coefficient D can be
determined as [9]:
D
c
Figure 1. a: SEM images of the composite cross-section, b: glass cloth and c:
mica layers obtained after burning the resin.
This insulation material was machined to provide diskshaped samples of 50 mm diameter with thickness from 1 to 4
mm. The faceplate of disks was parallel to glass cloth layers.
Samples were first dried at 120 C in a vacuum oven for more
than 1 week. The efficiency of the drying procedure was
monitored by weighing samples with a high precision balance
(0.1 mg resolution). The sample weight first dropped, and then
became constant. No variation was recorded during the last 3
days of the drying procedure. Taking into account the
h
S
4m f
m 2 m1
t t
2
1
(2)
where m1, m2 are water absorption values (expressed as

percentages) at times t1 and t2, and h the sample thickness
(1mm).
The variation of D versus absolute temperature T usually
follows the Arrhenius relation:
D0 exp(
A
)
kT
(3)
where D0 is the pre-exponential factor, A the activation

energy, and k the Boltzmanns constant.
Using 1.5% as common value of mf at all temperatures, D
can be determined at each temperature according to equation
(2). From the plot of these values versus 1/T (Figure 4) an
activation energy A is estimated to 0.42 eV (9.6 kcal). This
value is close to that previously recorded in pure epoxy

67
eBook for You
where, m is the sample weight at time t , and m0 is the initial

weight of the dry sample.
(10r1.5 kcal) [10]. This shows that water diffusion at the

initial stages mainly occurs in epoxy, and that the contribution
of interfaces with fillers is negligible [11].
An analytical solution of Ficks second law can be given by

[9]:
m
1.4
1.2
M(%)
1.0
0.8
0.6
60 C-measured
60 C-Fick's law
90 C-measured
90 C-Fick's law
110 C-measured
110 C-Fick's law
0.4
0.2
0.0
10
20
1/2
t(h )
30
40
Figure 2. Water absorption versus square root of time at three temperatures

(60, 90 and 110 C), with 1 mm thick samples. The lines correspond to
theoretical curves calculated with a single Fickian diffusion mode.
1.0
M(%)
0.8
0.6
0.4
0.2
0.0
60 C
90 C
110 C
4
1/2
t(h )
Figure 3. Initial stages of water absorption versus the square root of time (1
mm thick samples).
-2
D=1090exp(-4809/T)
-1
D(mm .h )
10
-3
10
-4
10
2.6
2.7
2.8
-1
1000/T(K )
2.9
3.0
Figure 4. Arrhenius plot of diffusion coefficients D corresponding to a

Fickian diffusion mode.
0.75
1 exp 7.3 Dt m
f
2
(4)
When we use the above values of D to compare the

theoretical curve (equation (4)) with experimental data, it
appears that deviations from a pure Ficks law are observed at
longer times, especially at 60 C which exhibits a large
difference from the theoretical curve (Figure 2). This
demonstrates that the complete water absorption process can
not be described by only one diffusion mechanism within
epoxy, with a single diffusion coefficient D.
To avoid this difficulty, we refer to the approach developed
for two-phase polymers [12, 13]. It considers that the
diffusion in each phase is Fickian, and defined by two
different diffusion coefficients (D1 and D2), and two different
water uptakes at saturation (m1 and m2). In our case, since
mica and glass do not absorb water, the phase 1 is constituted
by interface regions, and phase 2 by epoxy.
An example of fit of the experimental data of Figure 2 at 60
C with the superposition of two modes is presented in Figure 5,
and fitting parameters D1, D2, m1 and m2 corresponding to all
plots of Figure 2 are listed in Table 1. Parameters were
optimized in Figure 5 with the following procedure. The mass
conservation requires that m1+m2= m. As seen in Figure 2,
the fit with a single mode starts to deviate from experimental
data when m exceeds about 0.5 %. As mentioned above, it
appears that the initial diffusion stage is mainly controlled by
the epoxy resin. Therefore, it can be anticipated that m20.5 %.
This is the value used to initialize the fitting procedure. m1 can
be obtained from the values of m and m2 (m1= m- m2).
The mass conservation also requires that m1(t)+m2(t)=m(t) is
verified at any time. If this latter condition is applied at short
times, from equation (2) we obtain the relation:
m1f D1 m2 f D2
(5)
mf D
From the value of D (Figure 4) we obtain a relation

between D1 and D2. Since epoxy controls the diffusion at short
times, we arbitrarily assume that D2 is at least ten times higher
than D1. This last condition determines the initial values of D1
and D2 in the fitting procedure. Parameters optimization is
then carried out by a fitting program (Table 1).
In accordance with the considerations made by Maggana
and Pissis [12], the values of D2 corresponding to the
hydrophilic epoxy resin (polar region, containing bounded
water) are larger than D1 (non-polar region, containing free
water). On the other hand, more water can be stored at
interfaces, which explains that m1f> m2f .
Although the fitting procedure may have some influence on
the final values of parameters listed in Table 1, a fairly good
agreement between experimental results and Maggana-Pissis
(MP) model was obtained at the three temperatures (Figure 5).

68
eBook for You
becomes weaker and is more easily damaged, a fact observed in

our experiments. In bars permeated by water with an insulation
thickness about 4 mm, local concentrations exceeding 2%
associated to material delamination were also reported [8].
1.5
60C-measured
M P model
M ode1
M ode2
1.0
M(%)
3.00
2.25
10
20
1/2
t(h )
30
40
1.50
0.75
1 mm
2 mm
4 mm
Figure 5. Superposition of two diffusion modes (MP model) to describe the

whole diffusion process at 60 C.
0.00
T(C)
60
90
110
Table 1. Fitting parameters of the MP model.

D2(mm2.h-1)
D1(mm2.h-1)
M1f(%)
M2f(%)
-4
2.10
6.55.10-3
0.9354
0.43598
6.2.10-4
1.63.10-2
1.29004
0.26013
1.55.10-3
1.8.10-1
1.44148
0.14374
2.3. INFLUENCE OF SAMPLES THICKNESS

In the above section, because of the low thickness of samples
only water diffusion normal to the material layers was
considered. However, the anisotropic structure of the material
suggests that a non negligible water migration may also occur
from the lateral face of the sample, following a direction parallel
to layers. In an isotropic material, the contribution of the lateral
face would be negligible, since its surface is much lower
compared to faceplates. In this laminated material, if the
migration of water is much easier in the direction parallel to
layers compared to the perpendicular direction, the influence of
the lateral face could be non negligible and contribute to explain
the observed deviations from a classical diffusion law.
If this hypothesis is true, the influence of water entering by
lateral faces should be more pronounced with thicker samples.
This feature can be studied by means of the Pollars method
[14], which consists to measure the water absorption of samples
having different face area to edge area ratio [14, 15].
To clarify this point, water absorption measurements were
carried out at 90 C with 1, 2 and 4 mm thick samples with the
same surface (50 mm diameter, Figure 6). However, the results
obtained were quite different from those expected. For samples
thicker than 1 mm, deviations from the diffusion process were
observed. An abnormal weight gain appears at long durations
with thick samples (after about 300 h for 2 mm and 120 h for 4
mm), and the water uptake at saturation is greater than with 1
mm. This behavior may be tentatively attributed to macroscopic
damages (microcracks, delaminations) occurring in thick
samples, where bulk water is stored up [16]. The reason why
such behavior is mainly observed in thick samples can be
explained if we assume that damages are due to the osmotic
pressure [17]. When thickness increases, the sample structure
10
20
1/2
t(h )
30
40
50
Figure 6. Water absorption versus the square root of time for 1, 2 and 4 mm
sample thickness.
As a consequence, it was not possible to conclude about

anisotropy of water migration from these experiments. Thus,
the previous model considering the superposition of two
diffusion processes in the direction perpendicular to layers
with optimized parameters was used in the numerical
simulations carried out in the part II of this study.
2.4. PREPARATION OF SAMPLES FOR DIELECTRIC
MEASUREMENTS
Dielectric measurements presented in the next sections were
carried out on samples versus water content. Disk-shaped
samples were prepared as described above, by immersion in
water for variable durations, and the water content was
monitored by weighing. In the case of samples with water
concentration below saturation, a non-homogeneous water
concentration should be present within the material, since the
diffusion process was stopped before equilibrium (water
concentration should be maximum in outer layers, and minimum
in the center). In turn, this non-homogeneity may have an
influence on measured electrical properties of samples. It was
very difficult to assess the actual influence of this phenomenon.
Several attempts were done in order to homogenize the water
distribution within samples, without changing the total mass of
absorbed water. After impregnation in the water bath, samples
were placed in a climatic chamber at high temperature (to favor
water homogenization) and high relative humidity (to avoid
drying samples during this process). The constancy of the mass
was verified by periodical weighing, but no conclusive influence
on dielectric properties was obtained from these experiments, and
this procedure was abandoned.
Thus, in the results presented below, quoted water contents
correspond to the average water content in the sample,
whereas the local concentrations may be not strictly
homogeneous within samples.

69
eBook for You
0.0
M(%)
0.5
Breakdown measurement on samples were carried out under

50 Hz ac voltage, versus water content. Samples were prepared
in the same way as previously (50 mm diameter disks), with a 1
mm thickness in order to obtain a not too high breakdown
voltage, especially with dry samples. In order to minimize the
influence of partial discharges (PDs) on breakdown
measurements, the experimental set-up described in Figure 7
was adopted. A sphere-plane electrode system was used (sphere
diameter: 2.5 cm). The main problem was to eliminate PDs
occuring at the wedge between the sphere electrode and the
sample. The best solution was found with the following
procedure. After impregnation in water, the sample was placed
between electrodes in the breakdown cell. Degassed transformer
oil was then introduced, and the test cell was placed in a
vacuum for 10mn in order to remove trapped air cavities. Then,
SF6 gas with pressure equal to 0.5 MPa was introduced above
the oil surface, in order to avoid partial discharges and
breakdown in the upper part of the test cell. A ac voltage ramp
(1 kV/s) was then applied until breakdown occurred.
Breakdown was detected by the large current spike ocurring
when the sample is short-circuited by the breakdown arc. After
breakdown had ocurred, a permanent conducting path existed
within the material, and it was impossible to applied again an
electric field, even of low intensity (1 kV/mm). Partial
discharges were also detected during the breakdown test.
have been even higher. Compared with dry samples,

breakdown voltages in the presence of water are lower and
less scattered.
In all experiments, the breakdown channel through the
sample did not follow a straight path between the sphere and
plane electrodes. In wet samples, the arc always entered at the
contact point of sphere and sample, but on the grounded plane
side it reached the electrode several mm away from the gap
axis. In dry samples, breakdown was probably triggered by
PDs in the oil wedge, and frequently enter some mm from the
contact point between sphere and sample.
Even in the worst case (1.5% water content), the short-term
breakdown strength measured here (down to 20 kV/mm)
remained higher than the nominal operating field of insulation
in stator bars (about 3 kV/mm). In more realistic conditions
(i.e. on the long term, with a larger electrode surface area, and
at higher temperature), it is clear that a breakdown strength
lower than 20 kV/mm can be expected in water-permeated
zones. However, the field redistribution occurring in the
presence of water in an actual bar geometry, presented in a
second paper, tends to counter balance the deleterious effect
of water, and explain why water-permeated bars may survive
for a rather long time.
100
eBook for You
3. BREAKDOWN MEASUREMENTS
Breakdown Strength (kV/mm)
80
60
PD inception
40
20
0
0
0.4
0.8
1.2
Water content (%)
1.6
Figure 8: Breakdown strength (ac crest value) versus water content (1 mm

thick samples).
Figure 7: Test cell for breakdown voltage measurement.
Since the total time from introduction of oil up to

breakdown is about 10 mn, we suppose that oil diffusion in
the sample is negligible (at room temperature, water diffusion
in samples is not measurable after 10 mn). During
experiments, PDs in the oil wedge were detected prior to
breakdown for voltage above 50 kV (crest voltage), i.e. only
with dry samples.
The results of breakdown measurements are plotted in
Figure 8. The breakdown voltage is drastically reduced when
large amounts of water are present. PDs in oil appears prior to
breakdown only in dry samples. One may suppose that
without PDs, the breakdown voltage of dry sample would
4. DIELECTRIC SPECTROSCOPY
4.1. EFFECT OF WATER CONTENT
Water absorption does not only reduce breakdown strength,
but also considerably modifies dielectric properties. Dielectric
responses of samples at different water contents were
measured over the frequency ranges 10-3 to 103 Hz at different
thicknesses, temperatures and electric fields.
These measurements were carried out with a dielectric
spectrometer equipped with a 30 kV amplifier, offering the
possibility to study the influence of electric field up to some
kV/mm.
Dielectric properties were measured on the same sample at
different times during the impregnation process. The following

70
procedure was adopted at each measurement: sample removal

from water/surface drying at 50 C/installation of electrodes
/dielectric measurements/removal of electrodes/impregnation in
water. This process was repeated until saturation was reached. It
was thus necessary to install and remove electrodes frequently.
Electrodes pressed on the sample surface were made either with
commercial conducting tape (a thin copper sheet with a
conducting adhesive), or with a soft conducting polymer
(silicone rubber charged with silver particles). Both systems
provided a good contact with the surface of samples (not
perfectly flat, due to the laminated structure), and give equivalent
results. The electrode system used included a guard ring on the
ground side (Figure 9). Since the faceplates of samples are
parallel to the material layers, the electrical measurements
provide the dielectric response of the material with an applied
field perpendicular to layers, such as in a real stator bar.
21] is the dominant mechanism in wet samples. In this

process, the material is assumed to include clusters. Both
inter-clusters and intra-cluster motions contribute to the final
dielectric response [20].
7
10
dry
0.05%
0.1%
0.18%
0.5%
1%
saturated
De-ionized water
10
10
10
H'
10
1/(Z)
10
10
10 -3
10
HV ELECTRODE
GUARD
ELECTRODE
10
f(Hz)
10
10
10
Figure 10: Dielectric constant of a 4 mm thick sample for different water

uptake (10 V/mm, room temperature). Data for de-ionized water come from
[22].
MEASUREMENT
ELECTRODE
10
Figure 9: Electrode arrangement for dielectric spectroscopy measurements.
10
Typical measurements of real and imaginary parts of

complex permittivity are shown in Figures 10 and 11. These
measurements were obtained at room temperature with a 4mm
sample at 10 V/mm electric field. Dielectric properties are
very sensitive to water uptake, especially at low frequency. In
dry samples, H and H remain almost constant over the whole
frequency range. Even with a very small water content, we
observe a large increase in H and H as water content
increases. At high water content (>1%) and low frequency, a
considerable increase is observed (up to 8 decades).
Many mechanisms may contribute to this increase: interfacial
phenomena [18], dc conductivity, electrode polarization [19],
low frequency dispersion (LFD, [20, 21]). Referring to water
absorption described in section 2, in which water mostly locates
at interfaces, Maxwell-Wagner interfacial polarization could
constitute a possible explanation of the variation of dielectric
properties. With this mechanism, the real and imaginary parts of
complex permittivity should vary as [21]:
H' Z v Z 2
(5)
H" Z v Z 1
10
10
10
However, as can be seen from Figure 10, H does not

follows a Z-2 variation. The only influence of dc conduction
should produce a constant H and H proportional to Z-1 , a
feature not observed in our results excepted at very low water
concentration (m=0,05%). With higher concentrations, a large
increase of H is recorded, and H varies as Z-n with n lower
than 1. These observations suggest that a LFD process [20,
-1
10
De-ionized
water
n=-0.93
dry
0.05%
0.1%
0.18%
0.5%
1%
saturated
n=-0.95
H"
n=-0,95
10
n=-0.93
n=-1
-2
10
-3
10
-2
10
-1
10
f(Hz)
10
10
10
Figure 11: Loss factor of a 4 mm thick sample for different water uptake (10
V/mm, room temperature). Data for de-ionized water come from [21].
The dielectric behavior of de-ionized water [22] was also

plotted on Figures 10 and 11 for comparison. It is interesting
to observe that at high concentrations, the slopes of H and H
versus frequency compare fairly well with water, although
absolute values of H and H remain much lower in
impregnated samples. In pure water, the huge permittivity at
low frequencies arises from space charge polarization at
electrodes, and the quasi-conduction behavior of H is due to
the free charges of high mobility [22]. In samples with a high
water concentration, these mechanisms can also contribute to
the measured behavior, in addition to the phenomena cited
above.
4.2. INFLUENCE OF THICKNESS
Dielectric responses of 1mm and 4mm samples were
investigated in order to check the effect of material

71
eBook for You
SAMPLE
-2
10
thickness. Similar overall variations were observed in all

samples. A summary of properties recorded at 50 Hz
versus water content is shown in Figures 12 and 13. The
results can be divided in two regions: at small water
content (below 0.5%) the difference of dielectric
properties at 1 and 4 mm is not significant (in the range of
scatter between samples). Above 0.5% larger values of H
and H (up to a half decade) are recorded in the 4 mm
sample. It is assumed that this behavior can be attributed
to the existence of larger water clusters, located in regions
where the material structure is broken (delaminations,
cavities).
Measurements were taken at 10 V/mm, 100 V/mm and 1

kV/mm on a 1mm thick sample, versus water content.
Figures 14 and 15 show typical dielectric responses in a dry
and wet sample.
10
H"
10
10 V/mm
100 V/mm
1 kV/mm
M=0%
-1
10
1 mm
4 mm
-2
10 -3
10
10
-1
10
f(Hz)
10
10
10
10
0.4
0.6
M(%)
0.8
10
10
M=0.1%
10 V/mm
100 V/mm
1 kV/mm
1.0
H"
0.2
10
eBook for You
H'
-2
Figure 14: Field dependence of dielectric constant versus frequency in a dry

sample (1 mm, room temperature).
10
0.0
10
Figure 12: Comparison of dielectric constant between two thicknesses: 1 and

4 mm (room temperature, 50 Hz).
2
10
-1
10 -3
10
H''
10
10
-1
10
1 mm
4 mm
-2
0.0
-2
10
-1
10
f(Hz)
10
10
10
Figure 15: Field dependence of loss factor versus frequency in a dry sample
(1 mm, room temperature).
10
10
0.2
0.4
0.6
M(%)
0.8
1.0
Figure 13: Comparison of loss factor between two thicknesses: 1 and 4 mm

(room temperature, 50 Hz).
4.3. INFLUENCE OF ELECTRIC FIELD

The quasi-conduction behavior of H suggests that a non
linearity may exist when the applied field is increased,
since conduction currents are usually non linear. If this is
the case, the sensitivity of dielectric spectroscopy could be
further enhanced at high field. It is also very interesting to
study the material in conditions close to its practical use
(fields of some kV/mm).
In the dry sample, the material shows no voltage

dependence of H and H. In the presence of water (Figure
15), the response at very low frequency shows an opposite
tendency, i.e. the decrease of loss when the field is
increased. This unexpected behavior can be attributed to the
Garton effect, mostly observed in impregnated insulation
[23]. It is due to the blocking of charges at internal
interfaces (on mica flakes, glass fibers) during an ac wave,
when the typical transit time of charges is short compared
to the ac period. This phenomenon was observed in any wet
sample whatever water content up to 1%. Its effect at the 50
Hz (or 60 Hz) power frequency is negligible. However in
test conditions (i.e. during dielectric spectroscopy at low
frequency on a bar or on a complete machine), the use of a
high voltage can lead paradoxically to a lower sensitivity to
the presence of water.
The measurements were stopped at 1% water content and
1 kV/mm in order to avoid the occurrence of breakdown.

72
Typical data obtained at power frequency (50 Hz) in a 1mm

sample are presented in Figures 16 and 17. No large effect
of the field (up to 1 kV/mm) on permittivity and losses
exist at this frequency. Additional measurements carried
out up to 5 kV/mm in a dry sample led to the same
conclusion. Thus, dielectric properties measured at low
field reasonably represent the behavior of material in
service conditions.
temperature (RT) up to 80 C. During the experiments, the

samples were placed in a climatic chamber to control the
temperature. A 65% relative humidity was kept constant
throughout measurements, in order to avoiding the drying
of wet samples at high temperature. This technique was
proven to be efficient by weighing periodically the samples.
Figures 18-23 show the typical temperature dependence of
a single sample impregnated at 0.1% and 1% water.
10
10
M=0.1%
RT
30 C
40 C
50 C
60 C
70 C
80 C
H'
H'
10
100
E(V/mm)
Figure 16: Field dependence of dielectric constant for different water content
at 50 Hz (1 mm sample, room temperature).
100
H''
10
-1
10
f(Hz)
10
10
10
Figure 18: Temperature dependence of dielectric constant versus frequency at

0.1% water content (RT: room temperature, 4 mm sample, 10 V/mm).
dry
0.05%
0.1%
0.2%
0.5%
1%
10
H"
-2
10
1000
10
10
10
10
M=0.1%
eBook for You
1
10
dry
0.05%
0.1%
0.2%
0.5%
1%
RT
30 C
40 C
50 C
60 C
70 C
80 C
0.1
10
0.01
10
-1
10
100
E(V/mm)
-2
10
-1
1000
Figure 17: Field dependence of loss factor for different water content at 50
Hz (1 mm sample, room temperature).
4.4. INFLUENCE OF TEMPERATURE

All dielectric measurements presented above were
performed at room temperature. In power generators,
because of the heat induced by the current in copper
strands, the operating temperature is typically about 60 C.
This contributes to the acceleration of water migration by
thermal activation. In addition, the temperature also affects
the dielectric response, and this factor must be considered
to assess the actual behavior of insulation.
Dielectric spectroscopy was carried out with 4 mm
samples with an applied field of 10 V/mm, from room
10
f(Hz)
10
10
10
Figure 19: Temperature dependence of loss factor versus frequency at 0.1%

water content (RT: room temperature, 4 mm sample, 10 V/mm).
The increase of H and H at high temperature suggests

that the charge exchange between clusters and inside
cluster is favored by temperature. At small water content
(0.1%, Figures 18 and 19), the plot of H shows an overall
decrease when frequency is increased. Superposed on this
decrease, it is possible to distinguish peaks (of small
amplitude and rounded shape), for instance at 10-1 Hz at 30
C. The typical peak frequency is shifted towards higher
frequency when the temperature is increased. This
phenomenon can be attributed to the response of the
polymer backbone itself. These peaks are masked by the
LFD phenomena at high water content (1%, Figures 20 and
21).

73

10
M=1%
4
10
10
10
10
10
10
H"
H'
10
RT
30 C
40 C
50 C
60 C
70 C
80 C
10
-1
10
-2
10
-1
10
f(Hz)
10
10
10
H''
Figure 20: Temperature dependence of dielectric constant versus frequency at

1% water content (RT: room temperature, 4 mm sample, 10 V/mm).
10
10
10
10
10
10
M=1%
RT
30 C
40 C
50 C
60 C
70 C
80 C
10
-1
10
f(Hz)
10
10
10
H'
Figure 21: Temperature dependence of loss factor versus frequency at 1%

water content (RT: room temperature, 4 mm sample, 10 V/mm).
10
10
10
dry
0.1%
0.2%
0.5%
1%
saturated
30
40
50
T(C)
60
70
30
40
50
T(C)
60
70
80
Figure 23: Temperature dependence of loss factor at power frequency (4 mm

sample, 10 V/mm).
5. CONCLUSION
10 -2
10
-2
10
80
Figure 22: Temperature dependence of dielectric constant at power frequency

(4 mm sample, 10 V/mm).
At power frequency, the permittivity shows a slight

increase when temperature is increased (Figure 22). Although
the effect of temperature is not negligible, it remains much
lower than the increase induced by water uptake. Losses are
more sensitive to temperature as can be seen in Figure 23.
They are multiplied by about 10 when the temperature rises
from ambient to 80 C. This phenomenon points to the
combined effect of water and temperature.
The water absorption in a composite dielectric material leads to

severe degradation of dielectric properties. Such materials are able
to absorb large quantities of water, exceeding 1%. The complexity
of their structure (several basic components, anisotropic structure)
induces complex water migration processes and dielectric response.
Both types of measurements are rather difficult to interpret.
Water uptake measurements carried out in accelerated
conditions reveal that water diffusion can be described by the
superposition of two classical Fickian modes, occurring at
interfaces and within epoxy. In addition, the apparition of structural
damages in thick samples also contribute to increase the water
penetration in the material. On the long term, chemical reactions
between water and the epoxy matrix (not investigated here), will
also contribute to destroy the material structure.
Water absorption induces a large decrease of the short-term
breakdown strength at power frequency. Although breakdown
mechanisms were not investigated in much details here, two basic
mechanisms can be invoked in the case of wet materials: the
thermal breakdown process and the influence of field
rearrangement. The first one is quite obvious, considering the very
high losses measured in wet samples. The positive coupling
between temperature and losses (Figure 23) can lead to the creation
of unstable hot spots, possibly leading to the boiling of water
clusters, until final destruction of the material. The presence of
large amounts of water mostly localized at interfaces should also
induce a modification of the field distribution within the material.
Because of the high permittivity of water (H|80, and even much
higher at low frequency), the field will be locally enhanced in the
surrounding materials of lower permittivity and conductivity
(epoxy resin, glass, mica). In conditions of a real stator bar, field
calculations carried out with the data obtained here [24] show that
the bar will survive for some time to water invasion, but the long
term reliability will be clearly degraded.
Dielectric spectroscopy shows a large sensitivity of dielectric
properties to water uptake. The wet material behaves mostly as a
LFD system, and this suggests that it is composed by a cluster
structure according to the theory of Hill and Dissado. At the power
frequency, measurements at high field do not show any non-

74
eBook for You
10
dry
0.1%
0.2%
0.5%
1%
saturated
linearity. However, a Garton effect was observed at low frequency.

These results confirm that dielectric spectroscopy, especially at low
field and low frequency, constitutes an efficient and nondestructive method of water detection.
All data obtained here are used in the next paper [24] to calculate
the field distribution in an actual stator bar geometry in the
presence of a water leak.
ACKNOWLEDGMENTS
The authors wish to acknowledge Electricit de France
company (EDF) for its financial support to this study. One
author (P.H. Thinh) was supported by a grant from the French
Ministry of Education, which is also greatly acknowledged.
The authors wish to thank warmly Mr. H. Debruyne and A.
Petit (EDF) for their technical assistance and helpful advices,
and L. Saragossi for the building of experimental systems.
REFERENCES
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
C. Schrijver, A. Herden and H. C. Kner, A chemical approach to the

dielectric aging of fibre reinforced polymer insulators, European Trans.
Electr. Power, Vol. 7, pp.99-105, 1997.
H. C. Krner and M. Ieda, Technical sspect of interfacial phenomena in
solid insulating systems, IEEE Intern. Conf. Properties and Applications
of Dielectric Materials (ICPADM), pp.592-597, 1991.
A. Fukuda, H. Mitsui, Y. Inoue and K. Goto, The influence of water
absorption on dielectric properties of cycloaliphatic epoxy resin, IEEE
Intern. Conf. Properties and Applications of Dielectric Materials
(ICPADM), pp.58-61, 1997.
M. Nagao, K. Oda, K. Nishioka, Y. Muramoto and N. Hozumi, Effect of
moisture on treeing phenomena in epoxy resin with filler under AC
voltage, IEEE Conf. Electr. Insul. Dielectr. Phenomena (CEIDP), pp.951954, 2002.
M. Abder- razzaq, Abder-Razzaq, M. Auckland, D.W. Varlow and B.R
Effect of water absorption on the growth of trees in composite insulation,
IEEE Conf. Electr. Insul. Dielectr. Phenomena (CEIDP), pp. 766-769,
1996.
V. Pouills, Caractrisation dilectrique de matriaux composites isolants:
Contribution la modlisation de leur vieillissement, Ph.D. thesis of
Universit Paul Sabatier (in French), 1995.
K. Kimura, Progress of insulation ageing and diagnostics of high voltage
rotating machine windings in Japan, IEEE Electr. Insul. Mag., Vol.9, No.
3, pp.13-21, 1993.
H. Ito, S. Sekito, T. Taguchi, H. Shimada and M. Tari, Wet-heat
deterioration and residual life estimation technology on water cooled stator
coils of turbine generator, EPRI, IEEE&CIGRE, Rotating Electric
Machinary Colloquium, Lake Buena Vista, Finland, pp.1-5, 1999.
C. G. Springer, Environmental effects on composite materials, Technomic
Publication, 1981.
L. L. Marsh, D.C. Van Hart and S. M. Kotkiewicz, Moisture solubility and
diffusion in epoxy and epoxy glass composites, IBM J. Res. Develop Vol.
28, pp.18-25, 1984.
P. Gonon, S. Bourdelais, O. Lesaint, T. Pham Hong, P. Guuinic and H.
Debruyne, Effect of Hydrothermal Aging on the Dielectric Properties of
Epoxy Composites, IEEE 7th Intern. Conf. Properties and Applications of
Dielectric Materials (ICPADM), Nagoya Japan, pp. 936-939, 2003.
C. Maggana and P. Pissis, Water sorption and diffusion studies in an
epoxy resin system, J. Polymer Sci. : Part B : Polymer Phys., Vol. 37,
pp.1165-1182, 1999.
P. M. Jacobs and F. R. Jones, "Diffusion of moisture into two-phase
polymers", J. Materials Sci., Vol. 24, pp. 2331-2336, 1989.
A. Pollard, R. Baggott, G. H. Wostenholm and B. Yates, Influence of
hydrostatic pressure on the moisture absorption of glass fibre-reinforced
polyester, J. Materials Sci., Vol. 24, pp.1665-1669, 1989
[15] T. Pham Hong, Caractrisation et modlisation du comportement

dilectrique dun matriau composite soumis un vieillissement
hydrothermique, Ph.D. thesis of Universit Joseph Fourrier (in French),
2005
[16] A. Chateauminois, Comportemen viscolastique et tenue en fatigue de
composites verre/epoxy - Influence du vieillissement hygrothermique, Ph.D.
thesis of Universit Lyon I (in French), 1991.
[17] C. J. Jones, The environmental fatigue of reinforced plastic Composites,
Vol 14, p.288, 1983.
[18] H. Janssen, J. M. Seifert and H.C. Kner, Interfacial phenomena in
composite high voltage insulation, IEEE Trans. Dielectr. Electr. Insul.,
Vol. 6, pp.651-659,1999.
[19] V. Adamec and J. H. Calderwood, Electrode polarisation in polymerics
dielectrics, IEEE Trans. Electr. Insul., Vol. 24, pp.205-213, 1989.
[20] L. A. Dissado and R. M. Hill, Anomalous low frequency dispersion. Near
direct current conductivity in disordered low-dimensional materials, J.
Chem. Soc., Faraday Trans.II, Vol. 80, pp.291-319, 1984.
[21] A. K. Jonscher, Dielectric relaxation in solids, Chelsea Dielectric Press,
1983.
[22] D. G. Frood and T. J. Gallagher, Space charge dielectric properties of
water and aqueous electrolytes, J. Molecular Liquids, Vol. 69, pp. 183200, 1996.
[23] C. G. Garton , Dielectric Loss in Thin Films of Insulating Liquids, J. IEE,
Vol. 88, p.103-121, 1941.
[24] T. P. Hong, P. Gonon and O. Lesaint, Water Absorption in a Glass/Mica /
Epoxy Composite. II: Field Distribution and Diagnostic in a Stator Bar
Geometry, IEEE Trans. Dielectr. Electr. Insul., Vol. 16, pp.11-16, 2009.
Thinh Pham Hong was born in 1978 in Hanam,

Vietnam. He received the Engineer degree in power
system engineering from Hanoi University of
Technology (HUT), Vietnam, in 2001, the M.Sc.
degree in electrical engineering from Grenoble
National Institute of Polytechnic (INPG), France, in
2002 and the Ph.D. degree from the University Joseph
Fourier of Grenoble, France, in 2005. His current
interest focuses on the water ingress in composite materials, dielectric
properties of composite materials subjected to hydrothermal ageing,
diagnostic methods of insulation and high voltage engineering in power
system. He is currently a lecturer at the HUT.
Olivier Lesaint was born in 1960 in France. He
obtained
his
Engineer
degree
(specialty:
semiconductor physics) in 1982, and the Ph.D. degree
(energetic physics) at the Institute Polytechnique de
Grenoble in 1987. He is currently Directeur de
Recherches (full time researcher) at the Centre
National de la Recherche Scientifique (CNRS), and
Deputy Director of the Grenoble Electrical
Engineering Laboratory (G2E lab) in Grenoble. His
main fields of interest are the dielectric properties of insulating materials for
high voltage systems (power transformers, rotating machines), prebreakdown
phenomena in liquids, and diagnostics methods of insulation.
Patrice Gonon was born in 1964. He received the
Ph.D. degree in physics (specialty Microelectronics)
in 1993 from the University of Grenoble, France.
From 1993 to 1994 he worked as a Postdoctoral
Fellow at the Fraunhofer Institute in Germany, then
from 1994 to 1996 as a Research Fellow at The
University of Melbourne in Australia. In 1997 he
joined the Laboratory for Electrostatics and
Dielectric Materials (LEMD) at the University of
Grenoble where he is currently serving as an
Associate Professor of Electrical Engineering. His research interests focus on
the dielectric properties of solid-state insulators, including epoxy insulation,
low-k and high-k materials for electrical engineering.

75
eBook for You
10
11
Water Absorption in a Glass-mica-epoxy Composite

II: Field Distribution and Diagnostic
in a Stator Bar Geometry
T. Pham Hong, P. Gonon and O. Lesaint
Grenoble Electrical Engineering Laboratory (G2E lab), CNRS and Grenoble University
CNRS, 25 Rue des Martyrs, 38042 Grenoble, BP 166, France
ABSTRACT
eBook for You
In this paper, the water concentration profile within the insulation of a stator bar in the
presence of a water leak is calculated by resolving the Ficks equation. The
temperature strongly activates diffusion processes, but the shapes of concentration
profiles remain similar. From the measured dielectric properties versus water content,
an electrical model is developed in order to calculate the field and potential distribution
within the insulation. The field is reduced in wet zones due to the high local permittivity
and conductivity compared to dry zones. This effect is favorable since the breakdown
field is lowered in wet zones. On the other hand, the field in dry zones is reinforced, but
the value attained remains lower than the short-term breakdown strength. This
explains why stator bars in the presence of waters leaks may survive for long times, in
spite of high local water concentrations in excess of 1%. The high sensitivity of
dielectric properties at low frequency to the presence of water suggests non-destructive
methods to detect wet bars in a real power generator.
Index Terms Epoxy composite, water absorption, rotating machine, dielectric
spectroscopy, diagnostic method.
1 INTRODUCTION
IN a previous paper [1], water uptake in glass-mica-epoxy
samples was studied, together with its influence on dielectric
properties.
In this paper, water absorption is calculated in conditions
close those existing in water-cooled bars, using data obtained
in [1]. Water concentration profiles are calculated for two
temperatures: at room temperature (corresponding to
conditions usually found during testing of generators), and at
60 C (representative of typical operating conditions). The
change of local dielectric properties versus water content are
known from the preceding study. When the water
concentration profile and local dielectric properties are
known, it is possible to calculate the field and potential
distribution by developing an electrical equivalent model of
the insulation. From these calculations, field distributions in
various conditions of temperature and frequency can be
obtained. A diagnostic method for the on-site detection of wet
bars is then proposed.
Manuscript received on 2 June 2008, in final form 17 September 2008.
2 WATER CONCENTRATION PROFILE

CALCULATION
In water-cooled stator bars, water leaks although very rare
may occur from copper strands into the insulation, as
illustrated in Figure 1.
Dry hydrogen
insulation
Copper strands
and cooling water
x(mm)
Figure 1. Cross section of a bar.
Water absorption measurements described in [1] allow us to

understand the mechanism of water diffusion in the material,
but not the distribution of water concentration in the case of a
water leak, since conditions are rather different in both cases.
In order to determine the concentration profile in an actual bar
geometry, the following hypothesis are taken into account:
1070-9878/09/$25.00 2009 IEEE
76
T. P. Hong et al.: Water Absorption in a Glass-mica-epoxy Composite II: Field Distribution and Diagnostic in a Stator Bar Geometry
1. The real thickness of groundwall insulation is about

4mm. With such thickness, structural damages and abnormal
high water saturation values were observed in samples
immersed in water. In the case of water leaking from inside
the bar, the diffusion takes place from only one side, and the
effect of osmotic pressure within the material should be lower.
Therefore, for the purpose of simplicity, we will consider that
diffusion process follows a Ficks law identical to that found
with 1mm thick samples.
2. Once the water starts to leak from copper strands, we
make the hypothesis that the presence of water at this place
becomes permanent. Thus, water concentration in the material
at this side of insulation is at saturation. On the other side, the
insulation is dried out in permanence by the dry hydrogen
circulating in the machine, and the water concentration at this
place will be considered to be maintained at zero.
3. The material is dry before the apparition of a leak, and
the water concentration is zero in the material at t=0.
4. For simplicity, we will consider that the Fickien diffusion
occurs along a single dimension x across the material.
Since water concentration (C) is proportional to the
percentage of absorbed mass (m), we can replace the former
by the latter in Ficks second law.
Thus, Ficks equation may be written as:
wm
wt
w2m
wx 2
Figure 2. Water concentration profile in the insulation, calculated at room

temperature (25 C).
(1)
The initial and boundary conditions corresponding to the

hypothesis presented above are:
m(0,t)=0; m(l,t)=0; m(x,0)=mf;
(insulation thickness).
This leads to a solution [2, 3]:
m( x , t )
0<x<l
and
l=4mm
f 2m
mf
nSx
n 2S2 t
f
x mf
sin
exp D 2 , (2)
l
l
n 1 nS
l
Since the diffusion process in the material can be described

by the superposition of two different Fickian processes, the
whole process is the sum of m1(x,t) and m2(x,t) with the
values of mf and D for each phase determined in Part I.
At room temperature (RT), the knowledge of water
concentration is needed. Since the time required to attain the
saturation is very long (several years for a composite material
of some millimeter [4]), we must extrapolate the coefficients
(mf and D) for each phase at RT from Arrhenius plots
presented in part I [3]. On the other hand, the coefficients in
service condition (60 C) are available in [1].
The change of water concentration profiles at different
instants for the two temperatures are shown in Figures 2 and
3. At equilibrium, the concentration profile becomes linear. It
appears that equilibrium is attained after 120 000 h (|15
years) at room temperature (RT), and after about 20 000 h (|2
years) in service conditions at 60 C. These results show the
considerable effect of temperature on the diffusion process.
The shapes of water concentration profiles are comparable,
although time scales are very different.
Figure 3. Water concentration profile in the insulation, calculated at 60 C.
3 FITTING OF DIELECTRIC PROPERTIES

VERSUS WATER CONTENT
In [1], the sample dielectric properties were determined
versus water content up to saturation. By modeling the
material as a RC parallel equivalent circuit, values of R and C
at a fixed frequency can be determined versus water content.
From these values, equivalent values of dielectric constant and
resistivity can be determined.
The relation between water content and the equivalent
dielectric constant at power frequency (50 Hz), under a field
of 10 V/mm at RT and 60 C is depicted in Figure 4. The lines
correspond to the fit of experimental data by a polynomial
function. On the other hand, the equivalent resistivity at the
same temperatures can be approximated by a quasipercolation relation, as previously described in [5]:
U
U0
(3)
x x c t
where 0 is a factor related to the resistivity of the

conducting phase, xc is the percolation threshold and t is the
critical exponent.

77
eBook for You
12
13
with:
Hequivalent
Zi
10
Z total
10
0,0
0,5
1,0
M(%)
1,5
2,0
2,5
Uequivalent(:.mm)
Figure 4. The relationship between water content and the equivalent

permittivity (fit by a polynomial function).
10
14
10
13
10
12
10
11
10
10
10
10
60C
25C
10
0,0
0,5
1,0
M(%)
1,5
2,0
(6)
Zi
i 1
where Zi is the impedance of each element and Ztotal is the

total impedance of the insulation. The equation (5) can be
written in the form:
'l
1
(7)
Zi
A 1
2
ZH 0 H i
Ui2
60C
25C
(5)
1
2
ZC i
R i2
2,5
Figure 5. The relationship between water content and equivalent resistivity

(fit by a percolation law).
4 FIELD CALCULATIONS
In a dry bar, the field within the insulation is constant
across its thickness. Once the water starts to leak into the
insulation, the water concentration distribution from wet to
dry sides is very inhomogeneous, as showed in Figures 1 and
2. The local dielectric properties depend upon water
concentration as illustrated in Figures 4 and 5. As a result, the
field distribution is strongly modified within the insulation.
To calculate the field distribution across the insulation of a
wet bar, we consider an equivalent electrical model in which
the insulation is divided into n discrete RC elements in series.
The Figure 6 shows the electrical scheme of the insulation
where V0 is the nominal voltage of the bar (copper strand
side) and Vn is the voltage of ground side (=0).
where A is a surface area

This formula allows us to calculate the potential of every
element in the insulation, from the dielectric properties of the
material.
From the potential calculated above, the field distribution
within each element is given by:
Vi Vi 1
E i gradV
(8)
'l
Since the applied field has no significant influence on
dielectric properties [1], it is reasonable to use dielectric
properties measured at low field (10 V/mm) to describe the
behavior of insulation.
We take V0=13 kV (crest value) as nominal voltage of the
generator, the field distribution is calculated for three instants
of water diffusion, corresponding to the concentration profiles
presented in Figure 2.
Figure 7 shows the potential distribution within the
insulation at these moments, at 60 C. At dry condition, the
potential decreases linearly (constant field). When water
penetrates the material, the potential applied on the second
half of insulation (between 2 and 4mm) increases with time up
to about 75% of applied voltage when water concentration
reaches a steady state at t=20 000 h.
dry
Figure 6. Electrical equivalent model.
The potential of the element i is calculated as follows:

V0 i
Vi
. Z j
Z total j 1
Figure 7. Potential distribution across the insulation (60 C, 13 kV, 50 Hz).
(4)
The corresponding field distribution is presented in figure

8. When water penetrates into the insulation, the field is

78
eBook for You
10
14
reduced at inner layers (copper strand side) and is reinforced

at outer layers (ground side). At steady state of water
diffusion, the maximum field may reach 10 kV/mm, i.e.
three times the field in dry condition (3.2 kV/mm). The field
arrangement occurs in a favorable way, and tends to
counter-balance the deleterious effect of water: the field is
reduced in wet areas with degraded breakdown field. The
maximum field in dry zones remains lower than the shortterm breakdown field for dry samples [1], and the field in
degraded wet areas is very low (down to 1 kV/mm). This
explains why wet bars may survive for a long time in spite
of high local water concentrations up to 2% [7]. However, it
is clear that the long term reliability will be degraded, in a
hardly predictable way.
maximum field enhancement (a factor of 2). These

conditions are relevant of test conditions, when a machine
is stopped and cooled down to room temperature to
perform measurements.
5 EFFECT OF FREQUENCY
Figure 10 shows the potential distribution at steady
state of diffusion (t=20 000 h at 60 C), corresponding to
6 different frequencies with an average field of 10 V/mm.
At this field, the data are available for all water contents
up to saturation [1]. The lower the frequency, the higher
the field distortion. When the frequency is less than 0.01
Hz, more than 90% of applied voltage is applied on 1/10
of the sample thickness close to outer layers.
Figure 8. Field distribution across the insulation (60 C, 13 kV, 50 Hz).

Figure 10. Effect of frequency on potential distribution (60 C, t=20 000 h, 10
V/mm).
dry
Figure 9. Field distribution across the insulation at room temperature (25 C,

13 kV, 50 Hz).
Figure 11. Field distribution across the insulation (60 C, 13 kV).
As can be seen in Figure 4 and 5, decreasing the

temperature lowers resistivity and permittivity. These
results at room temperature have been used to calculate
the effect of temperature on field distribution. The latter
was calculated for the same moments as in Figures 7 and
8, with dielectric properties at RT (Figure 9). The same
trends are observed in this case, with a slightly lower
In Figure 11, the field at steady state of water diffusion is

calculated for two frequencies: 50 Hz and 0.001 Hz. It is
found that the field enhancement is very high (about 40) if a
low frequency high voltage is applied. The maximum field
reaches 120 kV/mm, which is higher than the short-term ac
breakdown field of dry samples (about 80 kV/mm [1]). These

79
eBook for You
dry
conditions are found when a high dc voltage is applied on wet

bars. In this case, the risk of breakdown is very high. This
conclusion agrees with practical observations of abnormally
low dc breakdown voltages of wet bars.
One advantage of dielectric spectroscopy for water
detection is the high sensitivity at low frequency, measured on
material samples [1]. However, when water concentration is
non homogenous within samples, the field distribution is
distorted and the field is lowered in wet areas. This should
logically induce a lower detection sensitivity, but this point
has not been investigated in detail.
6 PRACTICAL DIAGNOSTIC METHODS

Dielectric spectroscopy measurements on the whole stator
winding can provide data about the average state of the
machine, but cannot provide information about the
localization of wet bars. Once water leak problems are
suspected, a basic method consists to disconnect bars within
the machine and test them separately to find wet bars. The
large distortion of the field distribution within wet bars,
especially at low frequency (Figure 10), suggests a nondestructive detection method of wet bars. A technology
sometimes used to grade the field at extremities of bars
consists to enclose internal floating electrodes within
insulation (Figure 12). This principle is identical to that used
in high voltage capacitor bushings. If the electrode is located
in the middle of insulation, its potential should be 50% of
applied voltage in a dry bar, and much higher in a wet bar (up
to 100% at very low frequency, see Figure 10).
voltmeter
probe
Conducting
grounded
paint
7 CONCLUSION
Calculated water concentration profiles within the insulation at
room and service temperature in the presence of a water leak show
the same tendency, although water diffusion is considerably slower
at room temperature. The non-homogeneity of water concentration
leads to variable dielectric properties versus position. As a
consequence, the field distribution within the material is strongly
modified.
We have developed an equivalent electrical model by
considering the insulation as a network of RC parallel circuits in
series, in order to calculate the potential and field distribution. The
field distribution in the presence of water becomes frequencydependent. At 50 Hz, the field is enhanced in the dry zones of the
material, while it is reduced in wet zones. This arrangement is
favorable since the field is reduced in wet areas where the
breakdown field is reduced. At very low frequency, this tendency is
accentuated and the risk of breakdown in dry zones becomes high.
Based on these calculations, non-destructive methods to detect
wet bars in a real machine are proposed. These methods have been
tested successfully on full scale bars. Finally, calculations show that
the risk of breakdown is high when low frequency or dc high
voltage is applied on a wet bar.
ACKNOWLEDGMENTS
The authors wish to acknowledge Electricit de France company
(EDF) for its financial support to this study. One author (P.H.
Thinh) was supported by a grant from the French Ministry of
Education, which is also gratefully acknowledged. The authors
wish to thank warmly Mr. H. Debruyne and A. Petit (EDF) for
their technical assistance and helpful advices, and L. Saragossi for
the building of experimental systems.
Internal floating
electrode
REFERENCES
[1]
Copper strands
LF Generator
[2]
[3]
Figure 12. Principle of detection of wet bars with an electrostatic voltmeter.

[4]
The induced voltage on the internal electrode can be

measured with a non-contact electrostatic voltmeter. A similar
method was recently proposed in [6].
This non-destructive method has been tested successfully
on full scale bars, with a low frequency applied voltage (30
Hz). In the case of wet bars, the potential of the internal
electrode can be has high as 90% of the applied voltage,
instead of 50% in dry bars.
Another method, which does not require the presence of an
internal floating electrode, consists to use a field probe, such
as described in [8]. When a voltage of variable frequency is
applied to the copper strands, the measured field at the outer
surface of the bar changes versus frequency in the case of a
wet bar, while it remains constant with a dry bar.
15
[5]
[6]
[7]
[8]
T. P. Hong, O. Lesaint and P. Gonon, Water absorption in a

glass/mica/epoxy composite. I: Influence of Electrical Properties, paper
submitted to IEEE Trans. Dielectr. and Electr. Insul.
H. S. Carslaw and J. C. Jaeger J. C., Conduction of Heat in Solids, Oxford
Clarendon Press, 1959
T. P. Hong, Caractrisation et modlisation du comportement dilectrique
dun matriau composite soumis un vieillissement hydrothermique,
Ph.D. thesis of Universit Joseph Fourrier (in French), 2005.
T. Tanaka, K. Naito and J. Kitagawa, A Basis Study on Outdoor Insulators
of Organic Materials, IEEE Trans. Dielectr. Electr. Insul., Vol.13, pp.184193, 1978.
P. Gonon, T. P. Hong, O. Lesaint, S. Bourdelais and H. Debruyne,
Influence of high levels of water absorption on the resistivity and
dielectric permittivity of epoxy composites, Polymer Testing, Vol. 24, pp.
799-804, 2005.
Y. Inoue, H. Hasegawa, S. Sekito, H. Ogawa and H. Shimada, Advance
methods for detecting wet bars in water cooled stator windings of turbine
generators, EPRI- Intern. Conf. Electric Generator Predictive Maintenance
and Refurbishment, Florida, USA, pp. 88-103, 2003.
H. Ito, S. Sekito, T. Taguchi, H. Shimada and M. Tari, Wet-heat
deterioration and residual life estimation technology on water cooled
stator coils of turbine generator, EPRI, IEEE & CIGRE, Rotating
Electric Machinary Colloquium, Lake Buena Vista, Finland, pp.1-5,
1999.
A. Petit and O. Lesaint, A simple method to electrically determine the
limits of conductive/ non conductive parts of a stator bar, IEEE Int.
Symposium Electr. Insul., (ISEI), Toronto, pp.14-16, 2006.

80
eBook for You
16
Thinh Pham Hong was born in 1978 in Hanam,

Vietnam. He received the Engineer degree in power
system engineering from Hanoi University of
Technology (HUT), Vietnam, in 2001, the M.Sc.
degree in electrical engineering from Grenoble
National Institute of Polytechnic (INPG), France in
2002 and the Ph.D. degree of the University Joseph
Fourier of Grenoble, France in 2005. His current
interest focuses on the water ingress in composite
materials, dielectric properties of composite materials subjected to
hydrothermal ageing, diagnostic methods of insulation and high voltage
engineering in power system. He is currently a lecturer at the HUT.
Patrice Gonon is born in 1964. He received the

Ph.D. degree in Physics (specialty Microelectronics)
in 1993 from the University of Grenoble, France.
From 1993 to 1994 he worked as a Postdoctoral
Fellow at the Fraunhofer Institute in Germany, then
from 1994 to 1996 as a Research Fellow at The
University of Melbourne in Australia. In 1997 he
joined the Laboratory for Electrostatics and Dielectric
Materials (LEMD) at the University of Grenoble
where he is currently serving as an Associate
Professor of Electrical Engineering. His research interests focus on the
dielectric properties of solid-state insulators, including epoxy insulation, lowk and high-k materials for electrical engineering.
eBook for You
Olivier Lesaint is born in 1960 in France. He

obtained
his
Engineer
degree
(specialty:
semiconductor physics) in 1982, and the Ph.D. degree
(Energetic Physics) at the Institute Polytechnique de
Grenoble in 1987. He is currently Directeur de
Recherches (full time researcher) at the Centre
National de la Recherche Scientifique (CNRS), and
Deputy Director of the Grenoble Electrical
Engineering Laboratory (G2E lab) in Grenoble. His
main fields of interest are the dielectric properties of
insulating materials for high voltage systems (power transformers, rotating
machines), prebreakdown phenomena in liquids, and diagnostics methods of
insulation.

81

Các bài báo khoa học về Cao áp và vật liệu điện

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Các bài báo khoa học về Cao áp và vật liệu điện

Uploaded by

Copyright:

Available Formats

eBook for You

Tng hp cc bi bo khoa hc giai on 2007-2012

eBook for You

B mn H thng in - i hc Bch Khoa H Ni

Tng hp cc bi bo khoa hc giai on 2007-2012

NGHIN CU CNG NGH PHN TCH CC PHN T C TNH

Nguyn nh Thng, inh Quc Tr

eBook for You

The separation of elements with different conductivities is widely applied in

khong ly t m Cm ha (Cm xuyn,

Bng 1. c tnh ca mt s ht khong cht c phn tch

Trng lng ring, g/cm3

B mn H thng in - i hc Bch Khoa H Ni

Tng hp cc bi bo khoa hc giai on 2007-2012

ni. Vic la chn m hnh thit b kiu

1.2 Kch thc quy i ca cc ht

tnh ton v m phng ngi ta thng

 M hnh thit b ny hin cn cha

quy v hnh cu, hnh elip hoc bn elip

 Thit b c nghin cu c nhiu u

[2,3]. Bng thit b chuyn dng v

phng php quy hnh dng cc ht v

dng ti Vit Nam: c cng sut v

dng hnh cu c th tnh c bn knh

hiu sut phn tch cao, d dng lp

tng ng ca cc ht theo cng thc

t vn hnh hiu chnh, tiu hao t

 Vic s dng m hnh ny cho php

eBook for You

hnh dng rt a dng, thun tin cho

vn hnh thc t, nhng vi u im

rt ra kt lun s b: bn knh tng

Hnh 1. S ca m hnh thit b phn

vt liu ca Trng i hc Bch khoa H

tch kiu mng nghing

B mn H thng in - i hc Bch Khoa H Ni

Tng hp cc bi bo khoa hc giai on 2007-2012

1 - Phu cha nguyn liu; 2 - Mng

Tin hnh nhiu th nghim vi vic

thay i tr s gc nghing ca mng ta

di (hnh tr); 5 - Khay hng sn phm;

cng d dng nhn thy thay i ca s

thun tin cho vic phn tch cc

phn b lng ht trong cc khay thu hi

cc khay thu hi ph thuc vo kch thc

v t trng ring, nhng hiu sut phn

in cao p tnh in cng nh mt s nh

eBook for You

nhm hon thin m hnh thit b.

Hnh 2. Phn b khi lng cc ht trong

b. Trng hp c in trng: (in cc

ch chu tc ng ca trng lc, lc ny

pha trn c cp in p cc tnh m,

c tnh theo cng thc (2) nh sau:

cn in cc pha di cc tnh dng).

in trng s xut hin gia cc

Trong : m - khi lng ca ht; rtd - bn

trng hp u, cc ht chuyn ng theo

knh tng ng ca ht; J - t trng

mng nghing sau chuyn ng trong

ring ca ht; g - gia tc ri t do vi tr

s g = 9,8m/s2; D - gc nghing ca mng.

B mn H thng in - i hc Bch Khoa H Ni

Tng hp cc bi bo khoa hc giai on 2007-2012

M hnh thit b ny hin cn cha

Thit b c nghin cu c nhiu u

Vic s dng m hnh ny cho php

Mu tinh qung Zircon nhn c cha