You are on page 1of 14

Paper No.

1089
621.319.5 MEASUREMENTS SECTION

THE DEVELOPMENT AND DESIGN OF HIGH-VOLTAGE IMPULSE GENERATORS


By F. S. EDWARDS, M.Sc, Member, A. S. HUSBANDS, B.Sc, Associate Member,
and F. R. PERRY, M.Sc.Tech., Member.
{The paper was first received 24th October, 1949, and in revised form 2lst October, 1950. Proofs were made available to the public 30th December
1950, and the paper was read before THE INSTITUTION Wth January, and the NORTH-WESTERN CENTRE 9th January, 1951.)
SUMMARY some variation between the different specifications, all the
The paper explains the need for high-voltage impulse generators definitions result in similar waveshapes. The wave must be
which will produce the standard voltage waveshapes specified for aperiodic and reasonably free from superimposed oscillations.
impulse testing. A theoretical analysis of the impulse generator and The shape of the wave is then designated by numbers (e.g.
load circuit is given, and the development of the multi-stage generator 1/50 microsec) which give, respectively, the time during which the
for high voltages is described. The mechanism of operation of the voltage is rising to its peak value and the time during which it
multi-stage generator is analysed and is illustrated by oscillograms. subsequently declines to one-half of the peak value, both times
Illustrations of two typical generators are shown, and some of the
auxiliary equipment needed is briefly mentioned. The paper concludes being measured from the start of the wave. A typical impulse
with an example of impulse-testing technique. waveshape is shown in Fig. 1; the rising portion of the wave is
Nominal
(1) INTRODUCTION front
There is an increasing demand for impulse-voltage tests on
electrical apparatus, particularly when the latter is directly O9V -A
employed in the transmission of power. These tests are usually
designed to simulate the conditions which may arise on trans-
mission lines or associated equipment owing to the presence of 0-5V
surge voltages.1 From the knowledge already obtained of the
nature and magnitude of the surges occurring in practice, it is
possible to specify appropriate impulse-strengths for the insula- 0-1V — -
tion of the apparatus. Impulse-voltage testing is concerned with
the investigation of the electric strength of such new equipment. Time
Surge voltages have always occurred on transmission and
distribution systems, and long experience has led to the establish- Fig. 1.—Analysis of impulse-voltage waveshape.
ment of insulation levels determined on the basis of the strength
measured when power-frequency voltages are applied. Con- termed the wavefront and the declining portion is called the
sequently, the high alternating over-voltages commonly used in
the testing of electrical apparatus have partly served to test the wavetail; its duration is defined as the time taken for the
ability of the apparatus to withstand the effects of surge voltages. voltage to fall to one-half of the peak value.
It is evident that, as the knowledge of surge phenomena increases, In Britain, B.S. 923: 1940 gives the standard testing wave as
this purpose will be better served by tests which are more appro- a 1/50-microsec one, which is also the waveshape recommended
priate to the characteristics of the surges to be guarded against, by the International Electrotechnical Commission* and the
and thus it is probable that impulse-voltage tests will replace Verband Deutscher Elektrotechniker.f The waveshape recom-
some of the alternating-over-voltage tests, although, of course, mended by the American Standards Association! is a 1 • 5/40-
the latter will still be required—perhaps in a modified form—as microsec one, but the tolerance allowed by B.S. 923 is sufficient
a check on the characteristics of the apparatus when power- to include the latter value. B.S. 923 also prescribes the methods
frequency voltages are applied. of measurement of the duration of the wavefront and wave-
tail. These are designed to overcome the difficulty which
Impulse-voltage testing is now usually carried out with a arises in defining the start of the wave and the point to be taken
multi-stage generator; the paper describes the evolution of this as the wave peak on the flat portion of the crest. Referring to
type of generator and presents an analysis of the circuits Fig. 1, the "nominal" wavefront duration, tf, is therefore defined
employed. The paper is intended to collate and supplement the as being \-25t'f, where /y-is the time interval between the points
information contained in a large number of technical publica- where the voltage attains 10% and 90% of its peak value on the
tions. The auxiliary operating and measuring devices are an rising wavefront. In effect, this assumes that the nominal
essential part of impulse-generator equipment, and a description wavefront has a uniform voltage gradient equal to the average
is given of the relationship of these items to the impulse generator gradient between the two specified voltages and that the wave
itself. The technique of impulse testing has been described starts at that point where the zero-voltage axis is cut by this line
elsewhere.2 of uniform gradient. A similar method is proposed in the
(2) DEFINITIONS OF WAVESHAPE TERMS American specification, except that the suggested factor is I • 5
The shapes of the impulse voltage waves used for testing have instead of 1-25.
been standardized in various countries, and, although there is The 1/50 microsec (or equivalent) waveshape is usually
employed, but it is recognized that special occasions may arise
This is an "integrating" paper. Members are invited to submit papers in this when impulses of shorter duration are required; for this purpose
category, giving the full perspective of the developments leading to the present
practice in a particular part of one of the branches of electrical science. there are standards of 1/5 and 1/10 microsec (the latter occurs only
Mr. Edwards and Mr. Perry are at the Research Laboratory of the Metropolitan
Vickers Electrical Co., Ltd. * I.E.C. Publication 60 : 1938.
Mr. Husbands is at the Royal Technical College, Glasgow, and was formerly at t V.D.E. Specification 0450 : 1939.
the Research Laboratory of the Metropolitan Vickers Electrical Co., Ltd. t A.S.A. Specification C68.1—1942.
[155
156 EDWARDS, HUSBANDS AND PERRY: THE DEVELOPMENT AND
in the V.D.E. rules). Other special waveshapes may be required The true wavefront duration, tv i.e. the actual time taken for
for particular applications, e.g. chopped waves for transformer the voltage V to rise to its maximum value, is given by
tests, or even oscillatory waves for some purposes, but these are
necessarily specified by the requirements of the individual tests.

and the voltage efficiency, rj, of the generator is given by VmaJE,


(3) THEORETICAL ANALYSIS OF THE CIRCUIT i.e.
The simplest representation of a single-stage impulse-generator
circuit which corresponds approximately with practical condi- - a,)
tions is shown in Fig. 2, where:
Consider now the case when Rj is on the load side of R2, as
Cj = Discharge capacitance of generator. shown in Fig. 2. With the same symbols as before,
C2 = Capacitance of load.
Ly — Internal inductance of generator. X+T2ITtXf
L2 — External inductance of load and connections. or a2 = ^7S
27;
11
/?, or JRJ = Resistance for control of wavetail.
R2 — Resistance for control of wavefront. The values of /, and TJ will be determined exactly as they are
when Rj is on the generator side of R2.
R The position of Rj may sometimes be of great practical
importance. It will be immediately apparent that when Rj is
on the load side of R2, the two resistors form a potential

T I'. Jt T
c, divider which reduces the output voltage, but when Rj is on the
generator side of R2, this particular source of voltage loss is no
longer present. Fig. 3 shows how the efficiencies of the two
^"^ Alternative circuits vary with the ratio C^Cj for the particular case of a
positions of Rt 0 • 6/5-microsec wave.
Fig. 2.—Simplified circuit of impulse generator and load, showing
alternative positions of the wavetail-control resistor.

If C, is charged to a voltage E and the gap G is then closed,


the subsequent variation in the voltage, V, across C2 can be 50
analysed only by extremely laborious methods involving a
quartic differential equation. If L2 is made zero or is, in effect,
combined with Lv the equation3 is reduced to one involving
no power higher than the third: 0-25 0-5
Ratio C 2 /C x
Fig. 3.—Effect of position of wavetail-control resistor on voltage
efficiency of generator.
(a) Resistor on generator side of R2.
•(b) Resistor on load side of R2-
0) When Rj is on the load side of R2 the efficiency is very low
1 1 A 1-J
for small values of C2/Cj, but when the circuit is so arranged that
If Rj is on the generator side of R2, i.e. in the position R[, Rj is on the generator side of R2, the efficiency is highest when
the parameters are slightly different, but the equation is of the the load is zero, and the latter circuit should obviously be used
same form. Even more complicated representations have been wherever possible, especially when the resistance of R2 is not
examined, particularly by Thomason,4 but the resulting expres- very small when compared with that of Rj. •
sions are of little more than academic or mathematical interest, As a numerical example, if C2/Cj = 0-2, the voltage efficiency
especially as the stray capacitances and inductances are distri- of the generator with a 1/50-microsec wave falls from 80% to
buted throughout the circuit and no precise numerical values 77 %, if the resistor Rj is moved from the generator side to the
<^in be assigned to them. load side of R2. The effect is much greater with a 1/5 microsec
In practice, it is convenient to simplify the calculations, and wave, the efficiency being about 60% with Rj on the generator
four cases are examined in the following Sections. side of R2 and 40% with Rj on the load side of R2. Moreover,
under the latter condition the wavefront cannot be made longer
(3.1) Non-inductive Circuit than about 0-6 microsec if the wave tail is to be maintained at
If the circuit shown in Fig. 2 is considered as a non-inductive 5 microsec.
one, then L, and L2 will both be zero. Consider first the
situation when R, is on the generator side of R2, i.e. in position (3.2) Inductive Circuit Free From Oscillations
Rj'; now, if the time-constants Tt and T2 are given by CtRt and The necessary and sufficient condition for a non-oscillatory
C2R2, respectively, and the ratio C2IC} is denoted by X, it is discharge of the circuit shown in Fig. 2 (when L2 is zero) is
easily shown that the roots — ocj and — a 2 of the differential given (Goodlet)3 as:
equation for Kare:
icfi - 36)3 > #2a3 - gab + 27c)2
1+ where a, b and c are the coefficients of D2V, DV and V, respec-
tively, of eqn. (1). Such an unwieldy expression is of little
practical use, and the only simple deduction that can be made
•and V=E- (2)
is that a2 must be greater than 36 if the discharge is to be non-
oscillatory.
DESIGN OF HIGH-VOLTAGE IMPULSE GENERATORS 157
(3.3) Simplified Non-Oscillatory Inductive Circuit If the circuit is critically damped, the result is less obvious,
If the tail of the wave is long compared with its front—as it but a simple calculation gives the result:
is for a standard 1-5/40- or 1/50-microsec wave, then very little
error results from ignoring the wavetail resistance during the
calculation of the wavefront duration; the circuit shown in ~1ST
Fig. 2 then simplifies to that shown in Fig. 4, for which the
mathematical solution is very easy. (4) DEVELOPMENT OF THE MULTI-STAGE IMPULSE-
GENERATOR CIRCUIT
G L
R Until E. Marx took out his famous patent3 in 1923, no means
were known (or at least published) whereby the principle of
I o o nfiRr^—vw—| charging condensers in parallel and discharging them in series
C
T T2 could be put to practical use for the production of high-voltage
impulses.
In 1921, Grunewald6 described a single-stage circuit in which
Fig. 4.—Simplified circuit of impulse generator and load for calculation an electrostatic generator charged a condenser, which then dis-
of wavefront. charged across a spark-gap in series with the load and the con-
sequent oscillatory discharge doubled the voltage. Peek7
The critical resistance R is given by employed a transformer to supply a high alternating voltage for
charging a condenser which was then used as a single-stage
R = VC4I/O impulse generator; in 1915, he produced output voltages of
200 kV, and subsequent improvements enabled him to produce
1 1 1 2 000-kV impulses by 1923. Since the impulse waveshape could
where
not be measured, it was calculated by means whose validity was
very questionable,8 and no series damping was used. Moreover,
The voltage across C2 can then be expressed explicitly as follows: the polarity of the impulse was not readily controllable.
It was at this point that Marx produced a simple circuit which
effectively solved the fundamental problem by eliminating the
need for any individual components to operate at more than a
If, now, the inductance L is reduced to zero, but the resistance small fraction of the desired output voltage. The most con-
R remains unchanged, then vincing proof of the merit of his invention is that, after the lapse
of more than a quarter of a century, this circuit, without any
CF basic change, is still invariably used in the construction of
. . . . (6) impulse generators, although important modifications have been
made to the details.
The expressions for V in eqns. (5) and (6) both approach the
value CE/C2 as t approaches infinity, as they obviously should.
Eqns. (5) and (6) have maxima only at infinity, but if the lengths
of the wavefronts are calculated from the definitions given in Charging - i - Output
Section 2, it is found that, according to the formulae proposed voltage,E "T voltage, V
in B.S. 923: 1940 and V.D.E. Specification 0450: 1939, the
nominal wavefront is equal to 2-15CR when L is zero and to
2-1C7? when the circuit is critically damped; these two results
correspond to the conditions prescribed by eqns. (5) and (6), Fig. 5.—Marx's original voltage-doubling circuit.
respectively. On the other hand, when determined according
to the formulae given in Standard No. 4 of the American Fig. 5 shows Marx's so-called "voltage-doubling circuit,"
Institute of Electrical Engineers, the nominal wavefront duration fitted with a spark-heating condenser (shown dotted) which,
equals 3-B9CR when L is zero and 2-19CR when the circuit is although mentioned in the text of his patent specification
critically damped. appears only in two of his four illustrations. This circuit
In practice, fairly accurate results are obtained by computing, suffers from the serious defect (common to all the circuits shown
first, the value of R2 required to make the circuit shown in in the patent specification) that the charging voltage, E, is applied
Fig. 2 non-oscillatory when Rj is absent, then the value of continuously to the load across the points C and D, on which
R2 [if greater than «/(4LIC)] required to give the desired wave- the breakdown of the spark-gap merely superimposes an impulse
front, and finally the value of Rt required to give the required of the same magnitude, so that the output voltage V rises
wavetail when R t is on the generator side of R2 in the circuit suddenly from E to 2E and then decays to E.
shown in Fig. 2. The efficiency can then be approximately It is remarkable that, in an article9 published subsequent to
estimated as being ClftCl + C2) times a factor which is about the patent, Marx described a single-stage circuit (Fig. 6) and
0-95 for a 1/50 microsec wave. emphasized that there was no voltage on the load during charg-

(3.4) Crest Current on Discharge


The maximum value, /, of the current in the undamped dis-
-H-
charge of a generator into a short-circuit can readily be calcu- Output
voltage, h voltage,V
lated from the energy equation:

Fig. 6.—Marx's single-stage circuit, showing the load earthed during


i.e. I=VV(CXIL) charging.
158 EDWARDS, HUSBANDS AND PERRY: THE DEVELOPMENT AND
ing; he then went on to explain the circuit shown in Fig. 5 without point but no spark-heating condensers; Marx said that C2 and
mentioning the defect to which attention has just been drawn. C3 should have much smaller capacitances than should Cj, "as
Fig. 7 shows Marx's so-called "voltage-tripling circuit," in the capacitance of these condensers (C2 and C3) to earth has to
which the load is again subjected to the charging voltage E on be charged to a high voltage and the required energy is drawn
which the operation of the spark-gaps superimposes an impulse from Cj, which should therefore be as large as possible."
Marx's reasoning is difficult to follow; the loss of voltage caused
by charge-sharing with a condenser which is, effectively, con-
nected between the points D and F and in parallel with the load,
would surely be reduced more easily by making all the con-
densers as large as possible. Marx himself makes this very
point later in his paper, thus contradicting his earlier statement.
The numerical values quoted in the paper are:
Cl = 2 230 cm.
C , = C3 = 630 cm.
1
R = 1 megohm (approximately).
—AVWWWWN/ '
Fig. 7.—Marx's original voltage-tripling circuit. It should be noted that no series resistors are shown, that the
author takes it for granted that the discharge will be oscillatory
unless the leads are very short, and that the breakdown of one
gap would increase the voltage across the other only if there
was capacitance coupling between B and C or between D and F
(see Section 5.3). Marx evidently had trouble on this account,
for he emphasized the need to set the two gaps at exactly the
' F O B same spacing.
The circuit shown in Fig. 10 is electrically identical with
Fig. 8.—Rearrangement of Marx's original voltage-tripling circuit. Fig. 3 of Marx's specification, showing an 8-stage generator, but
it has been redrawn so that it assumes a more modern and
voltage of twice this value. The circuit of Fig. 7 is redrawn in
Fig. 8. If the point O is earthed and both spark-gaps break familiar appearance. If the point O were earthed, the left-hand
down together (although there is no reason why they should), side of the Figure would show a 4-stage circuit with a steady
the charging voltage E applied to one side of the load at D is voltage E on the load (at the point Z), on which an impulse, 3E,
augmented at this point by an impulse of E volts and at point F is superimposed when the spark-gaps break down. The right-
by an impulse of — E volts. hand side is an entirely different 4-stage circuit with the load
earthed at the point J through the resistor connected between
A little later, Marx10 again described his circuit but modified the points J and O during the charging period, an impulse voltage
it slightly as shown in Fig. 9. The new circuit has an earth of AE being applied to J when the four right-hand spark-gaps
break down.
If the point J is earthed, the spark-gap HG must be arranged
to fail first, and the voltage on Z (initially E) rises to 8E. It
should be noted that, if the point O is earthed, the spark-gaps
AO and KL are independent and the breakdown of the spark-
gap AO need not raise the voltage across the spark-gap KL.
-A/V
Fig. 11 (Fig. 4 of Marx's specification) illustrates a circuit
which is noteworthy in that an earth point is definitely shown.
1
\A/WWWWV' It is really an extension and duplication of the circuit of Fig. 7,
Fig. 9.—Marx's modified voltage-tripling circuit without spark-heating and there are two quite separate circuits. The charging voltage
condensers. is applied to the load, but if Marx had taken only the right-

Fig. 10.—Marx's original multi-stage generator drawn for eight stages.


DESIGN OF HIGH-VOLTAGE IMPULSE GENERATORS 159
A R C Y\
+/ j—AAAA/N « AAAAA^AAAAA -

T—*^ywwvwwv^vyw\AA-i^y^^
or n^ o. T

'•OO'oO ,

I /I /I /I /I
KVVVVLVVVVMVVVVNVVVVP
Fig. 11.—Marx's original double multi-stage generator. Fig. 14.—Miner's extension of Marx's voltage-doubling circuit.
hand side of the circuit shown in Fig. 10 and connected the voltage input is connected between the points C and F, with F
charging resistors in the manner shown in Fig. 11, he would earthed, the circuit is one-half of that shown in Fig. 11; if only
have had an entirely satisfactory circuit—as shown in Fig. 12— the circuit enclosed by the points A, C, F and O is considered,
with the load earthed during the charging period and with the with the load connected between the points F and O and the
input between C and O with the point O earthed, the result is
T^wvvxA^ywvv^^^ a single-stage circuit of the type shown in Fig. 12. Finally, if the
voltage is applied between the points A and B, with B earthed,
the result is a 2-stage circuit of the type shown in Fig. 12.
It should be noted that neither series resistors nor isolating
gaps are shown in any of Marx's original drawings, but these
features appear in his book on high-voltage practice15 published
Fig. 12.—Combination of circuits shown in Figs. 10 and 11 to give in 1941; some of his diagrams are reproduced as Figs. 15-17.
new circuit.

polarity of the impulse voltage opposite to that of the charging


voltage. This latter circuit is described by Goodlet.11 Alterna-
tively, he could have added another gap and another resistor
to the top half of the circuit of Fig. 11 (as shown dotted) and
obtained another equally satisfactory circuit, with the charging '-AAAAA^-AAAA-^^
voltage and the impulse voltage of the same polarity.
Peek's circuit12 of 1929 is interesting as one which is inter- Fig. 15.—Marx's later single-stage circuit with series resistors.
mediate between those shown in Figs. 10 and 12. This circuit
has been redrawn in Fig. 13 from Peek's original diagram,

IV V VV
Fig. 16.—Marx's later voltage-doubling circuit, with isolating gaps
and the load earthed during charging.

Fig. 13.—Peek's multi-stage impulse generator.

and it will be seen to bear a close resemblance to that of Fig. 12,


particularly in that the sign of the impulse is opposite to that
of the charging voltage at the instant of breakdown of the
spark-gap AO—a point which Peek did not mention. It should Fig. 17.—Marx's later multi-stage generator, with isolating gap.
be noted that resistor R4 has to withstand the full output voltage.
Miner 13 published some comments on Marx's circuit in 1926 The circuit shown in Fig. 15 is the same as that shown dotted
and extended the circuit shown in Fig. 5 to obtain two intercon- in Fig. 11, with the important addition of a series resistor, whilst
nected circuits which have been separated and redrawn in Fig. 16 shows a rather inconvenient means of short-circuiting
Fig. 14; at a later date14 he added isolating gaps at the points the load during the charging cycle. Finally, Fig. 17 shows a
P and Q. If the point O in Fig. 14 is earthed, the circuit circuit with an isolating gap, a series resistor and a load-
becomes the original defective Marx circuit of Fig. 11 plus the discharging resistor shown as RD.
modified circuit of Fig. 12. Bouwers16 also reviewed various impulse-generator circuits
Miner also showed13 an improvement whereby one side of and introduced several new ones which took inductance into
the load was earthed; this circuit was, in effect, that shown on account and showed the use of a series resistor on the generator
the left-hand side of Fig. 10. side of the discharge resistor instead of on the load side. One
Referring back to Fig. 8, it will be seen that, if the direct- of his diagrams, shown in Fig. 18, is particularly interesting in
160 EDWARDS, HUSBANDS AND PERRY: THE DEVELOPMENT A N D
the circuit shown in Fig. 19 when a 1 /5-microsec wave is required,
but it uses the conventional connections when a 1 • 5/40-microsec
wave is used.

(5) MULTI-STAGE CIRCUITS IN PRACTICE


(5.1) Types of Generator
C.R.O. As discussed in Section 4, the original Marx circuit was
modified by the addition of an isolating gap and a discharging
resistor across the load. In the modification by Goodlet,11 the
Fig. 18.—Bouwer's multi-stage circuit, showing distributed series circuit was altered as shown in Fig. 12, the main difference
resistors. between the two circuits being that in Marx's construction the
that the series resistors are distributed throughout the generator impulse is of the same sign as the charging voltage, whilst in
(two per condenser), both to suppress internal oscillations and the modification the sign is reversed. There are also certain
to control the wavefront duration. There is also an isolating practical differences, one of which is that on discharge botb
gap, as in the circuit shown in Fig. 17, but the resistor connected sides of the first spark-gap, Gj, are raised to the charging voltage
to earth on the load side is used as a potential divider for the in the Marx connection and fall to earth potential in the modified
cathode-ray oscillograph. A loading condenser, C (shown connection.
dotted), is recommended for tests on loads of small capacitance, (5.2) Disposition of Components
"so that the standard wavefront can be obtained without (5.2.1) Wavefront-Control Resistors.
increasing \he damping resistance." The wavefront control resistors can be placed either outside the
The first reference to the use of distributed series resistors in generator, partly within and partly outside or wholly within it.
a multi-stage circuit was published in 1932, when Ackermann17 Since the inductance is distributed all round the circuit, the third
employed the circuit shown in Fig. 11 but with resistors in each method is not wholly satisfactory, because the external leads,
spark-gap shank. The advantages of such distributed resistors the load and the stray earth-capacitance form an oscillatory
have been discussed by Eisner and others, 18 . 19 and their disad- circuit which requires to be damped by an external resistor.
vantages have long been apparent to those who have computed The important practical merit of resistors distributed within
the voltage efficiency of impulse-generator circuits. the generator is that they diminish the need for an external
As the wavetail-control current flows through these resistors if resistor capable of withstanding the full voltage. Such a
they are inserted in the circuit in the manner shown by Bouwers resistor for, say, 2 000 kV will be inconveniently long and will
and Ackermann, it follows that the larger the discharge capaci- possibly occupy valuable space, so that even a small amount
tance becomes (above a certain optimum value), the lower will of internal resistance is helpful. Probably the best compromise
be the output voltage, and various authors 18 - 23 have published between conflicting requirements is to arrange for about one-
curves which show this inconvenient effect very clearly. The effect half of the total resistance to be outside the generator.
produced is essentially the same as that which results from Wavefront-control resistors may be composed of wire, liquids,
putting the resistor Rj (Fig. 2) on the load side of the series or solids in the form of pellets or rods. The use of wire is not
resistor, the consequences of which are illustrated in Fig. 3. completely satisfactory, as it is difficult to reduce the inductance
The combination of high efficiency with distributed wavefront sufficiently and the heat capacity may not be adequate if there is
resistors was achieved in 1937,20 and the circuit employed is much energy to be absorbed; liquids have a high heat-capacity
shown in Fig. 19. but are unstable and require frequent attention, whilst solid
materials—sold under trade names and of secret composition—
have improved greatly in recent years.
(5.2.2) Wavetail-Control Resistors.
Wavetail-control resistors can readily perform the functions
of the charging resistors, in which case they are distributed
throughout the generator, or of potential dividers, when they
will be external to the generator. If used as a potential divider,
the wavetail-control resistor is necessarily on the load side of
Fig. 19.—Modern multi-stage circuit showing distributed series the wave-front resistor, and causes a loss of voltage in conse-
resistors connected to give maximum efficiency. quence.
The essence of this circuit is that R3 is made large compared (5.2.3) Paralleling of Stages.
with J?j, whereas R2 is made as small as is necessary to give the A feature of most multi-stage generators is that stages can be
required length of wavetail. It will be seen that under these coupled together in pairs (or greater numbers), so that the dis-
conditions the current through R 2 does not flow through Rj charge capacitance is increased at the expense of the output
and so has no effect in reducing the initial generator-output voltage. With stages in pairs, the capacitance is quadrupled
voltage, no matter how small R2 or how large Rx may be. In a and the output voltage is halved, which is sometimes a useful
practical generator where this circuit was employed, the stage- feature when large loads may be encountered, as in cable testing.
capacitance was 0-20 fjiF, i?j was about 40 ohms and the wave- A simpler, but less effective (although usually adequate), method
tail resistance required to produce (on this occasion) a 5-microsec is to use only a fraction of the generator, so that the discharge
wavetail was about 25 ohms; R3 was made about 10 kilohms, capacitance varies inversely as the output voltage.
R2 about 25 ohms, and less than 1 % of the output voltage was
lost in Rj. With the normal connections, about two-thirds of (5.3) Magnitude of the Over-voltage on Second and Subsequent
the voltage would have been lost in Rj. This circuit is now Spark-Gaps, when First Gap Breaks Down
widely used. There are but few references in the technical literature to the
An American installation*1 employs a slight modification of reasons why the spark-gaps in a multi-stage impulse generator
DESIGN OF HIGH-VOLTAGE IMPULSE GENERATORS 161

break down in very rapid succession, but, according to traditional Hence, the voltage, VY, across C 2 (i.e. across the gap G 2 ) is
theory on the subject, the sequence of operations is as follows: given by
Referring to Fig. 20, when the first spark-gap, G,, breaks
down, at a voltage of, say, — V volts, the potential at the point Z VY=V 1 + • • (8)

Now, if C 2 is equal to zero, VY will attain its maximum possible


value of 2V, whilst, if Cl and C 3 are both zero, VY will
equal V, its minimum possible value. From this it will be
apparent that the most favourable conditions for the operation
of the generator arise when the capacitance, C 2 , of the sphere
gap G 2 is small and the stray capacitances Cx and C 3 are large.
These deductions have been confirmed experimentally. The
setting of Gj was fixed, that of G 2 was adjusted until about
half of the failures of G , were followed by breakdown of G 2
and the ratio of the voltages corresponding to the settings of the
two gaps was then calculated. For these tests, the generator
used had a stage capacitance of 0 - 1 2 / A F . RX (Fig. 20) was
2 500 ohms for all tests, and values of 2 500 and 9 000 ohms were
D.C. used for R4, which was the same as R2 for this test. Table 1
shows the results obtained for various values of Ct and C 2 .
Fig. 20.—Diagram of multi-stage circuit showing stray capacitances. Although the theory outlined above goes a long way towards
an explanation of the observed phenomena, it is still very
imperfect and does not account for the extremely erratic
becomes zero, whereupon the potential at Y rises to + V volts, behaviour of some impulse generators when the resistances have
so that a voltage of 2V appears 1 4 - 2 2 across G 2 , which will been made very small in order to produce a very short-tailed
therefore break down if set to do so for some value between wave. A particular example encountered by the authors some
V and 2V. Similarly, voltages of W and 4V volts appear across years ago was a two-column eight-stage 1 000-kV generator of
G 3 and G 4 (provided that the previous gap has broken 0-20-/xF stage-capacitance which could not be made to work
down), and so on indefinitely. Now, although in practice it consistently at charging voltages in excess of about 60 kV (the
has been found quite feasible to operate a 50-stage impulse full charging voltage being 125 kV) when arranged to deliver a
generator, it has also been found that G 2 must usually be set wave of 5-microsec wavetail, although it was quite satisfactory
to break down at a voltage only slightly greater than that at for small charging voltages and/or when arranged to deliver a
which Gi breaks down, otherwise it does not operate; a wave with a 50-microsec tail.
little consideration will show that the theory given above is
very incomplete. When the point Y attains a potential of Table 1
+ V volts the potential at the point X does not remain at — V;
EFFECT OF THE STRAY CAPACITANCE OF A GENERATOR ON THE
in the absence of stray capacitance and if the resistance between
OVER-VOLTAGE ACROSS THE SECOND SPARK-GAP
the points X and Z is assumed to be infinite during the short
interval after the breakdown of G{, there is no reason why the
Additional capacitance Ratio of voltages across gaps G2
point X should be at any particular potential. and G]
Actually, of course, the circuit as drawn in Fig. 20 (without
the stray capacitances, which are shown dotted) is a very im- C C2 * 2 = «4 = 2. 5 k n R2 = /? 4 = 9 k n
perfect representation of the facts. The various parts of the
generator are coupled by the stray capacitances between the MM o
components, and it is these stray capacitances which control the 1-5 1-6
200 0 1-7 1-8
initial potential-distribution immediately after the breakdown 2000 0 1-8 1-9
of G,. 0 120 1-3 1-3
Two such stray capacitances, Ct and C 3 , are shown in dotted 0 2000 1-2 1-2
lines in Fig. 20, and the capacitance of the sphere-gap G 2 is 200 120 1-5
0 2000 1-4 1-3
shown as C 2 . For the purpose of the calculation of the initial 2000 120 1-9 20
potential-distribution, the resistors may be considered as open-
circuits, and it may also be assumed with inappreciable error
that the stray capacitances are negligible in comparison with the Let us reconsider Fig. 20. Hitherto the effect of the resistances
stage capacitances. has been neglected, but it is apparent that the state of affairs
Consider now what happens to the circuit when the point Z, represented by eqn. (8) is only transient, since the stray capaci-
which was previously at a potential — V volts, is earthed by the tances Cj, C 2 and C 3 immediately begin to receive or lose charge
breakdown of the gap Gv The point Y, which was initially at through the resistors, so that the over-voltage across G 2 (VY — V)
earth potential, assumes the potential + V volts, but the potential is quickly reduced to zero. Since these capacitances are normally
of the point X is fixed by the relative magnitudes of the capa- of the order of tens of micromicrofarads, the time-constants of
citances C,, C2 and C 3 and is calculated as follows: the local circuits are likely to be small; e.g. if the resistances
If the potential at the point X, after the transients have died are of 1 000 ohms each, then the time-constants will be of the
away but before the resistances have begun appreciably to affect order of 10~ 2 to 1 0 - 1 microsec, and correspondingly less for
the voltage distribution, is — Vx then lower values of resistance. In the 8-stage 1 000-kV generator
just mentioned the wavetail-control resistances were of the order
of 100 ohms, so that the time-constants were about 10~ 2
• • • (7) microsec.
VOL. 98, PART I. 12
162 EDWARDS, HUSBANDS AND PERRY: THE DEVELOPMENT AND
Let it be assumed that R2 is large (it may easily be ten times The circuit was of the form shown in Fig. 20 with twelve
RA in practice) so that R 2 can be regarded as an open-circuit stages connected normally, except that the resistor on the
(i?! plays no part at all). The shape of the potential/time curve charging side of G 4 was omitted. (Rx = 10 kilohms.)
of the point X will now be given by eqn. (9). After the initial R4 was varied from 100 to 10 000 ohms and the additional
potential-redistribution caused by the breakdown of the gap G, capacitances C (when used) were 230 /u/xF. The gap Gj was
has been effected, the subsequent events can be calculated on the set to break down at 88 kV and tests were made as follows:
assumption that R 4 is then connected between the points W (a) All gaps except Gx opened out.—The gap Gi was tripped
and Y, so that the voltage rise across the gap G2 is given by at zero time, and records were taken of the course of the voltage
(relative to earth) at points Y, W and U; the potentials at the
points Z and X differed from those at Y and W by the constant
amount of the charging voltage. The voltages across G 2 and
If (Cj + C3) is very much greater than C2, this becomes G 3 were the differences between the potentials of the points
X and Y and of the points W and U. These tests were made
VG1 — Ve- with values of R4 of 100, 1 000 and 10 000 ohms and both
so that, after a time 3i?4(C! + C3), the voltage across G 2 is with and without additional capacitances of 230 ftjuF across the
only 1-05K volts. The time-constant can be lengthened by an first three supports.
increase in (Cl + C3) or in R4, or by the introduction of in- One of the principal features of the results of these tests is
ductance into the resistor R. the extremely short duration of the over-voltage across G2,
The values Cx and C3 depend upon the physical construction which never exceeds 100%. This is shown by the full lines in
of the generator, but can be increased artificially, whilst R4 is Figs. 21 (a)-21 (c). There is an appreciable difference in the time-
fixed by the stage capacitance and the required length of wave- constant when R4 is reduced from 10 to 1 kilohm, and a much
tail. The authors are without practical experience of the third more noticeable difference when it is reduced to 100 ohms,
method, which presents certain difficulties in application. when the over-voltage persists only for about 0-1 microsec.
Besides making the observations recorded in Table 1, the It is also plain that the addition of condensers across the sup-
authors carried out oscillographic tests on another 12-stage ports has the expected effect of raising the magnitude and
generator of 0 12-/u,F stage capacitance, and some of the duration of the over-voltage, as shown by the dotted lines in
oscillograms are shown in Fig. 21. Figs. 21(a)-21(c).

0 0-5 1-0 1-5 0 0-5 1-0 0 O-5 1-0 1-5


Impulse length,JJ,S Impulse length,J
() w Impulse length, JJ.S
()

0 0-5 1-0 1-5 0 0-5 1-0 1-5 0 0-5 1-0 1-5 "
Impulse length,/us Impulse length,fxs
Impulse length,us ()
(d)
Fig. 21.—Oscillograms of voltages developed across spark-gaps G2 and G3 immediately after breakdown of gaps Gi and G2,
respectively (cf. Fig. 20).
Normal generator.
Generator with extra capacitance across three supports.
In all cases, Ri = R4 = Rs.
(a) and (d) Ri = 10 kd.
(6) and (<?) Ri = 1 k£2.
( c ) a n d ( / ) Rt = 100 Q.
DESIGN OF HIGH-VOLTAGE IMPULSE GENERATORS 163
(Jb) All gaps except Gt and G 2 opened out.—The gap G 2 was knowledge, each of ten gaps broke down singly on ten successive
set at a slightly larger spacing than G ]} so that it broke down tests.
very shortly after the failure of G{ and the tests described in (a) An attempt was made to overcome the trouble by enclosing
were repeated. each gap in a large cylindrical shroud made of pressboard and
The principal features of these tests are that the voltage across greased on the inside. The immediate result was that the first
the gap G3 [Figs. 21(d)-21(f)] rises to about 2-4Kbut falls below gap broke down with great regularity and with inappreciable
2Kin less than 0-1 microsec, and that, with added capacitance deviations from the mean value, but none of the other gaps
across the supports, the over-voltage is reduced but is of rather failed at all, as the shrouds completely cut off the illumination
longer duration; as in the first test, the duration decreases as (from the spark passing across the first gap) which Meek24 and
R4 is reduced. others have since shown to be essential to a low impulse-ratio of
Tests were also made in which the over-voltages on the gaps breakdown, especially when the impulse is of very short duration.
G 2 and G 3 arising from the breakdown of the previous gap The introduction of narrow slits at the top and bottom of
(Gj or G2) were measured by the spacings of the gaps themselves each shroud permitted the irradiation from the first spark-gap
for various values of inter-stage resistance, and the results are to fall on the second one and so on, and thereafter the per-
given in Table 2. formance of the generator was usually satisfactory. It was
Table 2
found, however, that although the axes of the gaps were in one
vertical plane the slight horizontal displacement of alternate gaps
THE EFFECT OF INTER-STAGE RESISTANCE AND OF ADDED reduced the intensity of the illumination impinging on them
CAPACITANCE ON THE OVER-VOLTAGES ACROSS GAPS G 2 AND G3* from the gap below, and this sometimes resulted in irregularity
of firing. With a small constructional modification, the gaps
Ratio of voltages were arranged to be in a vertical line; the performance was then
Resistance of
inter-stage resistors
quite reliable.
Ri, R 4 and R 5 With no extra With extra
capacitance capacitance of
230 uptF (6) GENERAL ARRANGEMENT
(6.1) Requirements for an Impulse Generator
ohms
The requirements for an impulse generator are: low inductance
Between gaps G 2 and G\ (Gi firing) (for production of steep wavefronts), low height, mechanical
100 1-3 1-4 stability, ease of access and of adjustment of movable or moving
1 000 1-4 1-6 parts and ease of extension. The charging voltage must be a
10000 1-6 1-75
compromise between conflicting requirements. A low charging
Between gaps G3 and G\ (G2 firing) voltage permits the use of a small and cheap d.c. charging set
100 1-4 1-5 but requires an increased number of stages. The total cost of
1000 1-6 1-9 the condensers having a given energy-storage capacity will not
10000 20 2-2 usually vary widely with the number of stages.
The voltage selected should preferably be about 10-15% less
* Cf. Fig. 20. than the open-circuit voltage of the d.c. set if the time of
charging is to be kept down to a reasonable value. For the
In these tests, each gap was well irradiated by the spark from large generators now common in commercial practice, with an
the gap below it, which thus tended to reduce its impulse ratio output voltage of 1 000 kV or more and a discharge capacitance
(see Section 5.4). On the other hand, the very short duration of of 0-01-0-03 [AF, the most suitable charging voltage is probably
the over-voltages would increase the impulse ratio, but, whatever about 170 kV. When the discharge capacitance is only 0-002-
the inaccuracies introduced into the measurement by these two 0 008 fiF at a few hundred kilovolts, 20 or 25 kV may be used
opposing effects, the ratios given in Tables 1 and 2 are the values as the charging voltage, but with a radically different form of
which must be assumed if the setting of the gaps is based on construction for the impulse generator.
B.S. 358. The irregularity of operation of the gaps of impulse The discharge capacitance depends upon the nature and
generators has been noticed, but not explained, by various magnitude of the load. A simple calculation shows that the
10 23
observers. - energy-storage capacity required in the generator is a minimum
With the circuit shown in Fig. 19, the over-voltage across the when the discharge capacitance of the generator is the same as
second spark-gap is reduced even more than is shown by the the load capacitance, so that on discharge the initial voltage at
above calculations, as the current flowing through R t after the the generator output terminals is twice the maximum voltage
breakdown of the first gap causes a further drop in the voltage actually applied to the load and the voltage efficiency is 50%.
applied to the second one. Such a calculation, however, is almost purely academic and
ignores many important practical aspects of generator design.
(5.4) Effect of Dust and of Illumination on Rapidity and The cost of the condensers is only a part of the total cost of
Regularity of Breakdown of Gaps the equipment, so that the arrangement giving minimum con-
The presence of dust introduces a serious practical difficulty denser cost may be very uneconomical in other respects. If a
into the operation of a multi-stage generator. It has been multi-stage circuit is used, for which the charging voltage is
shown that the available over-voltage is small and of short fixed, then at least twice as many stages are needed as the output
duration, so that both the second and subsequent gaps cannot voltage nominally demands.
be set to break down at a voltage much higher than that at With a given discharge resistance, the length of the wavetail
which the first gap breaks down. On the other hand, the dis- is proportional to the generator discharge capacitance plus the
persion of the breakdown values on direct voltages in the load capacitance, and if these are equal it will double on changing
presence of excessive dust may be so great that if the difference from no-load to full-load conditions. B.S. 923: 1940 permits a
in settings is small any one of the gaps may fail irregularly at variation of ± 2 0 % from the nominal value, so that, if the
less than the intended voltage; in a case within the authors' resistances give the correct wavetail at one-half of full load, the
164 EDWARDS, HUSBANDS AND PERRY: THE DEVELOPMENT AND
wavetail will depart from this value by 33 % at full load and no to a load of 5 000//fiF (C2). The first suggestion, arising
load. At least two sets of resistors will therefore be required, from a knowledge of the largest size of condenser available,
and the length of the wavetail will be very sensitive to any was to construct a generator of 0 01 /xF (Cj) discharge capaci-
capacitance changes in the external circuit. tance with an open-circuit output voltage of 2 400kV (V).
If the open-circuit output voltage is to be double the Charge-sharing and the voltage loss on the wavefront (of a
required test voltage, the building will need to be big enough to 1/50-microsec wave) would have brought the load voltage down
afford clearances in all directions for twice the test voltage. to about the specified minimum. With the wavetail resistances
As the ratio of generator capacitance to load capacitance is adjusted to produce a 50-microsec tail with a load of 2 500 fx/xF,
increased, the required open-circuit voltage decreases and the the wavetails at zero and full load would have been just within
circuit characteristics become less and less dependent on the the tolerated limits.
external load. However, the gain is subject to the law of Unfortunately, the building was not quite high enough to
diminishing returns, so that the conflicting technical, economic house this unit, and so the stage capacitance was increased by
and constructional requirements combine to make the optimum 50% and the number of stages reduced by 15%; the discharge
capacitance ratio somewhere between 4 : 1 and 10 : 1 for a pure capacitance was thus increased by 75% and the same output
capacitance load such as a bushing or a cable. voltage was obtained on load. The energy-storage capacity
The V.D.E. rules of 1933 prescribed a minimum value of 5 : 1 of the generator was increased by 30%, but the saving in acces-
for the ratio of generator capacitance to load capacitance (plus sories due to the reduced number of stages and—much more
all stray capacitances, including that of the measuring circuit), important—saving in the necessary clearances caused by the
and recommended several times this ratio at the lower voltages, reduction both in the height of the generator and in the voltage
where the cost of the condensers would not be prohibitive. for which the clearance to the roof was required, tended to
In the 1939 revision these recommendations were omitted, offset the extra cost. Clearances to the walls were also pro-
but both versions specify a minimum discharge capacitance portionately reduced.
of 1 000 /-ijU-F, which would now seem extremely small for If, in this example, the stage capacitance had been increased
any purpose. B.S. 923: 1940 follows the old V.D.E. rules in by 100% instead of 50%, with a reduction of 20% in the number
laying down a minimum ratio of 5 : 1, but without qualification. of stages, the discharge capacitance would have been increased
The problem of the selection of the optimum ratio is not by 150% instead of 75% and the energy-storage capacity by
susceptible to exact mathematical analysis, as the sizes of the 60% instead of 30%. The cost of the second increase would
available condensers change in fairly large steps and there are have been greater than that of the first, but the additional
many imponderables (such as the value—expressed in monetary saving in height would have been only one-third of the first
terms—of an increase in available test voltage) which make saving. Actually, the first increase was sufficient and the second
precise calculations useless. Each particular case is therefore would not have been considered unless the only alternative had
best considered on its merits, but an approximate mathematical been a major building operation, such as the digging of a large
treatment does yield some interesting results. pit in unsuitable ground in order to increase the sparking
distance to the roof.
Let Cj — Discharge capacitance of generator.
C2 = Load capacitance. (6.2) Effect of the Type of Condenser on Design
V = Open-circuit output voltage of generator. Three main types of condenser are used in the construction of
Vc — Crest value of initial voltage (assumed to be non- impulse generators. The first of these consists of hollow
oscillatory) applied to load. cylinders of varnished paper with concentric metallic layers
Then (10) inserted at suitable intervals. This type, which is now rarely
used, has many disadvantages, including low working stress and
Table 3 has been compiled from eqn. (10) to show the effect inconvenient shape; it also requires an insulating supporting
of an increase in CXIC2 on the value of V. framework which is large and expensive.
Oil-impregnated condensers in metal tanks have a higher
Table 3 working stress than those just mentioned, and can be made
EFFECT OF RATIO CJC2 ON REQUIRED OPEN-CIRCUIT VOLTAGE,
weatherproof. They require an insulating framework, and have
V, OF IMPULSE GENERATOR
the minor disadvantage that the voltage for which they can be
conveniently designed is rather low, so that two or more con-
densers must be connected in series for each stage. A number
Successive Total reduction
Increase in C1/C2 reductions in V* in V* of very large generators have been built on this principle.
Oil-impregnated condensers in insulating containers have the
1-2 0-50 0-50 dielectric assembled in an insulating cylindrical container of
2-3 0-17 0-67 porcelain or varnished paper with plane metal end-plates;
3-4 008 0-75 they are usually designed to operate in the vertical position.
4-5 005 0-80 The advantage of this form of construction is that the condensers
5-6 003 0-83 in successive stages can be readily insulated from one another
6-10 007 • 0-90 by supports of the same size as the condensers, but without the
10-CXD 010 100
dielectric. Fig. 20 shows the layout as well as the electrical
• As a fraction of the crest value of the initial voltage, Vc. disposition of the components, which are arranged in two, four
or more columns according to the number of condensers con-
Clearly, the successive reductions in V for successive equal nected in parallel at each stage. A 2 000-kV generator of this
increments of C{IC2 are becoming extremely small when Cj/C2 type is shown in Figs. 22 and 23.
exceeds 5, and in practice this ratio is rarely exceeded, since In a variant of this arrangement, advantage is taken of the
C2 is the maximum load expected. impulse ratio of flashover of the supports to assemble successive
As a practical example may be mentioned the case of a stages in the form of a spiral, so that although with, say, four
generator which was to be capable of applying 1 500 kV (Vc) columns each support has to withstand an impulse voltage of
DESIGN OF HIGH-VOLTAGE IMPULSE GENERATORS 165
three times the charging voltage, it can be made less than three
times as long as one support which has to withstand the d.c.
charging voltage. There is thus some economy in height, but
the simplicity of operation of the spark-gaps is lost.
Fins are sometimes provided on varnished-paper cylinders
and sheds on porcelain containers, in order to reduce the height
required for a given charging voltage.
If the charging voltage is low the condensers may be assembled

Fig. 2 2 . - 2 000-kV 0 01-juF 12-stage impulse generator.

Fig. 24.—Small portable oil-immersed 400-kV 0-003-^F 2O-stage


impulse generator.

in one container in the manner shown in Fig. 24, the disposition


being arranged as shown in Fig. 12.
The influence of the design of the condenser on that of the
impulse generator has been discussed elsewhere.19

(7) AUXILIARY EQUIPMENT


(7.1) Earthing Switch
It is essential that the generator should be fitted with a device
to earth the condensers when the equipment is not in active
operation. The device may take the form of a suitable switch-
arm which is operated mechanically or electrically, and may be
provided with signal lamps and interlocked with safety barriers
Fig. 23.—Closer view of generator illustrated in Fig. 22. if this should be desirable. The switch arm is most conveniently
166 EDWARDS, HUSBANDS AND PERRY: THE DEVELOPMENT AND
applied to the charging point of the impulse generator; all other generator discharge. The two methods of achieving this syn-
stages will then discharge through the various inter-stage chronism are respectively designated as "controlled" and "uncon-
resistors. trolled" generator tripping. In the former method, the oscillo-
During the operation of a generator it may be necessary in an graph time-base circuit is tripped first and this causes a pulse
emergency to earth the equipment when the condensers are fully to be applied to the first gap of the generator, which is of the
charged. If this were effected with a short-circuiting switch of 3-electrode or triggered type. A short delay is introduced
the type mentioned, it would be equivalent to the sparking of between the initiation of the time-base and the application of
the lowest sphere-gap, which would cause all other gaps to the pulse to the generator gap. Thus the cathode-ray beam
spark and so produce an unwanted impulse. This can be will have already traversed a small part of the screen or film
avoided by inserting a suitable resistor, RE, in series with before the impulse reaches the deflection plates, so that the
the arm of the earthing switch, so that the time-constant, initial changes of the impulse wave are clearly recorded.
R2{Cl + C2), of the circuit enclosed by the points T, Z and X The uncontrolled method may also be used with an oscillo-
in Fig. 20, is small compared with RECE, where CE is the stray graph, but the sequence of events is reversed. The generator is
capacitance to earth between the points Z and Y (including first tripped or allowed to fire and the oscillograph time-base is
that of the gap Gx). This ensures that the voltage drop along R2 initiated by means of a pick-up aerial from the electrical dis-
(which is the voltage rise across the gap G2) is small. In practice, turbance so produced. In order to obtain the requisite delay
C,, C3 and CE are all of the same order of magnitude, so that for a complete record of the impulse wave, the voltage to be
the required condition is that RE shall be much greater than R2. recorded is conducted to the oscillograph by a delay cable
(100 m of air-insulated cable will give a delay of 0-3 microsec,
(7.2) Motor Operation of Spark-Gaps which is usually sufficient to allow the cathode-ray beam to
A considerable improvement in operating convenience is given commence its time-sweep). The delay cable and the potential
by a remotely controlled motor which varies the setting of the divider form a unit and must be considered together, as the
generator spark-gaps in conjunction with a remote indicator of presence of a long delay-cable imposes a lower limit on the
the setting. This principle should preferably be extended to all potential-divider ratio which may be used without introducing
the various controls, so that so far as possible all operations undue errors in the response of the divider. This limitation is
and adjustments may be effected from a single panel. A full largely obviated if controlled tripping is used, as the cable
description of the detailed mechanism of remote control and between the potential divider and the oscillograph is then of
indication is beyond the scope of the paper. low capacitance.
In general, it is found that with uncontrolled generator
(8) TRIPPING OF THE IMPULSE GENERATOR tripping the oscillograph may be operated with very simple
auxiliary tripping circuits, but a long cable is required having
The simplest method of operating the impulse generator is to low loss and low attenuation. Controlled tripping obviates the
allow it to "self trip." Thus, the lowest spark-gap is set to need for a long cable, but the auxiliary circuits are more complex.
flash over at a predetermined charging voltage, and the stages
are then charged until this occurs. The disadvantages of this
system are that the accuracy of the charging voltage is subject (10) IMPULSE VOLTAGE MEASUREMENTS
to the vagaries of the first spark-gap, and that the instant of (10.1) Sphere-Gap
discharge is not accurately predictable. The latter objection The sphere-gap is a recognized means of measuring the output
also applies to a mechanical closure of the first spark-gap to voltage of an impulse generator, and the conditions of its use are
cause an impulse. specified in B.S. 368: 1939, which states that the accuracy of
A simple electrical tripping circuit25-26 largely overcomes the measurement is within db 3 %, provided that the duration of the
above-mentioned disadvantages and has proved convenient in impulse voltage is not less than that of a 1/5-microsec wave.
use. It employs a 3-electrode gap as the first spark-gap and The sphere-gap is connected across the output of the generator
is so connected that the charging voltage is applied between the circuit and directly in parallel with the test apparatus. A
outer electrodes. The central electrode may be earthed normally resistor should not be used in series with the sphere-gap, as the
by means of a high resistance or may be biased at some voltage capacitance of the latter modifies the waveshape of the voltage
intermediate between those of the outer electrodes. The elec- between the spheres, and this leads to further inaccuracies in
trode system is adjusted to withstand the required charging the measurement. The impulse sparks usually produce only
voltage by a small margin, and tripping of the charged generator minor damage to the sphere-faces.
is effected by applying an unbalancing voltage to the central The sphere-gap measures the peak voltage of the impulse wave
electrode. The tripping voltage may be a short pulse, in which and probably includes the superimposed oscillations if these
case the timing of the generator tripping is accurately controlled should be present. It is possible, however, that such oscillations
by the pulse, or an alternating voltage may be applied to the would have a different quantitative effect in flashing-over or
central electrode, when tripping will occur on or near the crest breaking-down other types of air-gap and other types of insula-
of the alternating-voltage wave. By a suitably modified arrange- tion, and it is for this reason that the presence of superimposed
ment,25 the tripping may be made to occur precisely on the wave oscillations is undesirable.
crest. Alternatively, the impulse may be synchronized with any
point of a voltage wave by the use of a phase-shift circuit in (10.2) Cathode-Ray Oscillograph and Potential Divider
the tripping supply. The simple 3-electrode gap may be replaced The single-sweep cathode-ray oscillograph is the most satis-
by any other form of triggered gap, the tripping of the generator
being controlled in a similar manner. factory means of examining the shape of an impulse wave;
it may be used to obtain visual or photographic records of the
wave-trace, and the amplitude of the wave may be determined
(9) SYNCHRONIZATION OF THE OPERATION OF by direct measurement on the screen or on a record.
OSCILLOGRAPH AND GENERATOR The oscillograph is usually constructed to record amplitudes
When used with the impulse generator, the time-base of the of a few hundred volts only, and it is necessary to use a potential
oscillograph must be triggered in accurate synchronism with the divider to reduce the voltage of the impulse wave. The divider,
DESIGN OF HIGH-VOLTAGE IMPULSE GENERATORS 167
which may be of the simple capacitance or resistance type or a 120
combination of the two, has been described in principle at
various times.27-28 For most impulse testing, a simple capaci-
tance-type potential divider has proved quite satisfactory and \
-C 100 x
has the advantage of providing a basic capacitance load on the \ \ Resistor
generator. This facilitates tests on specimens having a wide
range of capacitance, by reducing the variation required in the
<
80 \?
circuit parameters to maintain a standard waveshape. o0
An oscillograph has been constructed to measure moderately
high impulse-voltages directly without a potential divider.29 This 60
instrument was provided with a pair of low-voltage deflection :?L.
plates and could be used in conjunction with a potential divider, A<vndB
-•4
which permitted it to be employed in an investigation of the 40
response of a potential divider to impulse voltages of short
duration.
(11) EXAMPLE OF TESTING TECHNIQUE 20
(11.1) Impulse Flashover Tests on a Carbon-Type Resistor
Some tests were carried out to demonstrate the suitability of
carbon-composition-type resistance units for use as series 00 2 4 6 8
resistors on an impulse generator. The particular applica- Nominal wavetail duration, fj.s
tion in view called for a series resistance of about 30 ohms per
stage of the generator, with a stage capacitance of 0-23 fiF and Fig. 25.—Impulse flashover of carbon-type resistors.
a maximum charging voltage of 167 kV per stage. Thus, the Flashover voltages.
series resistor would have to withstand a maximum voltage and Starting voltages of streamers.
Resistors A and B .. . . 20 ohms, 10 in long.
energy of discharge corresponding to these values, on the assump- Resistor C 10 ohms, 8 in long.
tion that the resistance and the internal stage inductance formed
the only impedances of each stage circuit. If the inductance
were neglected, the resulting discharge wave would have a wave-
tail of about 5 microsec, and the tests described below were
conducted with wavetails of this order of length.
The carbon resistors were supplied in units, each about 8 or
10 in long and of 1-in diameter, with a nominal resistance of about
10 or 20 ohms per unit. An impulse generator having a normal
stage capacitance of 0-11/uF and no internal series resistors
was used. This was connected to operate as a 3-stage generator
with a series discharge-capacitance of 0 074 or 0-15/xF, by
connecting two or four of the normal stages, respectively, in Fig. 26.—Oscillograms showing modification of waveshape across
parallel to form single efFective stages for these tests. The test resistor A (see Fig. 25) caused by streamers as peak voltage is
resistor was then connected in series with a protective resistor raised.
directly across the output of the generator, the protective resistor Period of timing oscillation, 2 microsec.
being used to dissipate the energy of the generator if the test
sample flashed over. The test resistor was therefore subjected should be noted as the sparking became more severe at the
to a proportion of the total generator energy and output voltage, higher voltages.
and this proportion could be varied by changing the value of In Fig. 25, the full lines show the flashover voltages and the
the protective resistor. A cathode-ray oscillograph was used dotted lines the starting voltage for streamers, and it will be
in conjunction with a capacitance-type potential divider across seen that the streamers were first observed at a voltage which
the test resistor. was substantially constant irrespective of the nominal duration
It was found that the circuit inductance caused an appreciable of the wavetail. In addition, the short resistor C gave values
voltage drop and modified the waveshape of the voltage applied nearly equal to the values for the longer resistors A and B.
to the resistor. A similar effect is to be expected with any Complete flashover did not occur until the voltage across
generator, however, and tends to reduce the voltage across an A and B was increased considerably. The voltage gradient
internal series resistor. For comparative purposes, the induc- for each resistor was only about 6-8 kV (peak) per inch of
tance was neglected in the calculation of the nominal duration of length for complete flashover with the longer wavetails. These
the wavetail. low values are not uncommon with low-resistance resistors of
Impulses were applied to each resistor and the voltage was this type. The difference between A and B was perhaps due to
progressively increased until flashover occurred. It was noted the different shapes of their metallized end-caps.
that streamer discharges occurred across parts of each sample These tests demonstrated that the carbon-type resistors would
at voltages below the flashover value. The results are plotted be suitable for the series resistor of an impulse generator, pro-
in Fig. 25 with the measured peak voltages across the test vided that the voltage gradient along the resistor did not exceed
resistor as ordinates and the calculated wavetail durations as the value at which streamers started.
abscissae. The voltages were measured from oscillograms, an On account of the circuit inductance, the peak voltage which
example of which is shown in Fig. 26, which was determined would appear across such a resistor would be less than the stage
when negative impulse waves of 35, 48, 62, 67, 73 and 81-kV charging-voltage. The relationship between these two voltages
amplitude were applied to the resistor A. Surface streamers or was established approximately from the measurements made
sparks first occurred at 62 kV, and the change in waveshape during the preceding tests, and, as the circuit loops and generator
168 EDWARDS, HUSBANDS AND PERRY: THE DEVELOPMENT AND
dimensions were similar in the experimental arrangements and (9) MARX, E.: "Versuche iiber die Priifung von Isolatoren mit
in those occurring in practice, this information provided a Spannungsstossen," Elektrotechnische Zeitschrift, 1924,
further guide to the conditions under which the resistor would 45, p. 652.
operate. (10) MARX, E.: "Die Erzeugung und Messung der verschie-
denartigen Spannungen. IV Spannungsstoss," Mit-
(12) CONCLUSIONS teilungen der Hermsdorf-Schomburg-Isolatoren G.m.b.H.,
It is shown that for the standard waveshape (1/50 microsec), 1925, 17, p. 466.
the quartic equation for the impulse generator and load circuit (11) British Patent No. 344862. B. L. GOODLET,
can be simplified to a quadratic without great loss of accuracy. (12) PEEK, F. W.: "Dielectric Phenomena in High-Voltage
The evolution of the multi-stage circuit from the original ideas Engineering" (McGraw-Hill Book Co., New York, 1929),
of Marx has progressed to a point where impulse voltage p. 145.
generators can now be designed to meet any load conditions (13) MINER, D. F.: "Surge Generating Equipment," Electric
likely to occur in practice. Journal, 1926, 23, p. 596.
It is shown that stray capacitance and other parameters play (14) MINER, D. F.: "Insulation of Electrical Apparatus"
an important role in the operation of an impulse generator. (McGraw-Hill Book Co., 1941), p. 383.
The effect of dust on the performance of impulse generators (15) MARX, E.: "Hochspannungspraktikum" (Julius Springer,
is mentioned, and a method of avoiding trouble due to this cause Berlin, 1941).
is suggested, whilst the importance of adequate irradiation of (16) BOUWERS, A.: "Elektrische Hochspannungen" (Julius
the sphere-gaps is emphasized. It is also shown that the design Springer, Berlin, 1941).
of an impulse generator must be considered in conjunction with (17) ACKERMANN, O.: "New Surge Generator for Testing
that of the building. Transformers," Electric Journal, 1932, 29, p. 61.
To illustrate the versatility of the modern multi-stage generator, (18) ELSNER, R.: "Production of Standard Wave-forms in Surge
an example is given of testing in which the test object is a type Generators," Fachberichte des Verbandes Deutscher
of resistor used in impulse-generator circuits. Ingenieure, 1936, 8, p. 159.
(19) EDWARDS, F. S.: "Recent Developments in High-Voltage
Impulse Generators," C.I.G.R.fi., Paris, 1939, Paper
(13) ACKNOWLEDGMENTS No. 305.
The authors wish to thank Sir Arthur P. M. Fleming, Director (20) British Patent No. 498021. F. S. EDWARDS and G. J.
of Research and Education, and Mr. B. G. Churcher, Manager SCOLES, 1937.
of the Research Department, the Metropolitan-Vickers Electrical (21) NEWMAN, M., and KEMPPAINEN, O.: "High-Voltage Installa-
Co. Ltd., for permission to publish the paper. tion of the Precipitation-Static Project," Proceedings of
the Institute of Radio Engineers, 1946, 34, p. 247.
(22) FOUST, C. M., KUEHNI, H. P., and ROHATS, N.: "Impulse
(14) BIBLIOGRAPHY Testing Technique," General Electric Review, 1932,
(1) "Surge Phenomena," E.R.A. Report Ref. S/T35. 35, p. 358.
(2) ALLIBONE, T. E., and PERRY, F. R.: "Standardization of (23) "Surge Generators," Micafil News, December 1939, p. 21.
Impulse-Voltage Testing," Journal I.E.E., 1936,78, p. 257. (24) MEEK, J. M.: "The Influence of Irradiation on the Measure-
(3) GOODLET, B. L.: "Electromagnetic Phenomena in High- ment of Impulse Voltages with Sphere-Gaps," Journal
Voltage Testing Equipment," ibid., 1934, 74, p. 377. I.E.E., 1946, 93, Part II, p. 97.
(4) THOMASON, J. L.: "Impulse Generator Circuit Formulas," (25) EDWARDS, F. S., and SCOLES, G. J.: "A 2 000 kV Impulse
Transactions of the American I.E.E., 1934, 53, p. 169. Generator," Engineering, 1937, 144, p. 222.
(5) Deutsches Reichspatent No. 455933. E. MARX, 1923. (26) WHITE, E. L.: "A Tripping Circuit for a Multi-Stage Gene-
(6) GRt)NEWALD, F.: "Das Verhalten der Freileitungs-Isola- rator," Journal of Scientific Instruments, 1948, 25, p. 307.
toren unter der Einwirkung hochfrequenter Span- (27) BURCH, F. P.: "On Potential Dividers for Cathode-Ray
nungen," Elektrotechnische Zeitschrift, 1921, 42, p. 1377. Oscillographs," Physical Magazine, 1932, 13, p. 760.
(7) PEEK, F. W.: "The Effects of Transient Voltages on Dielec- (28) DAVIS, R.: "The Calibration of Sphere-Gaps with Impulse
trics—III," Transactions of the American I.E.E., 1923, 42, Voltages," Journal I.E.E., 1938, 82, p. 645.
p. 940. (29) NUTTALL, A. K.: "A Cathode-Ray Oscillograph for the
(8) GOODLET, B. L.: "The Testing of Porcelain Insulators," Direct Measurement of High-Voltage Transients," ibid.,
Journal I.E.E., 1929, 67, p. 1177. 1936, 78, p. 229.

DISCUSSION BEFORE THE INSTITUTION, 11TH JANUARY, 1951


Mr. R. Davis: In principle, the surge generator is an extremely aperiodic by series resistance equal to or greater than 2V(L/C),
simple piece of equipment. It consists of a charged capacitor and the duration of the surge or the time above half-value can be
which, at an appropriate instant, discharges into an external determined by considering the circuit as a charged condenser of
network of resistance, inductance and capacitance. The output capacitance equal to the sum of the generator and load capaci-
from the generator is applied to a load, which may be an insulator tances discharging through a resistor, R, and therefore having a
string or a transformer winding, etc. Simple calculation can give time-constant of CR. The operational methods of Heaviside are
considerable guidance as to the waveshape of the voltage delivered admirably suited to the carrying out of the necessary calculations.
to the load when the latter can be represented by capacitance. The complication of exact calculations is illustrated by the authors
Thus, the natural period of the circuit consisting of a capacitor —a complication enormously increased if attempts are made to
discharging into a load capacitance is 2TT-\/LC where C is the analyse the multi-stage generator circuit, with its many degrees
capacitance of the two capacitors in series and L is the inductance of freedom, coupled with the difficulty of assigning precise values
of the whole circuit. The duration of the wavefront cannot be to many of the parameters such as stray capacitance to earth at
less than one-quarter of this period. Such a circuit is rendered different points in the circuit. Simple calculations such as those

You might also like