Beruflich Dokumente
Kultur Dokumente
Transformer
J. H. McWHIRTER
MEMBER AIEE
Stresses within
Windings by Computers
of Impulse
C. D. FAHRNKOPF
MEMBER AIEE
Synopsis: A theoretical basis fordetermination of the short-time transient characteristics of transformer windings is presented. This theory was used to develop
an accurate, economical, and convenient
method of determining impulse voltage
stresses within transformer windings through
the use of modern digital and analogue computers. The method has been developed
for and applied to the shell form of construction, but the theory is equally applicable to other forms of construction.
Purpose
THE CONTINUING trend toward
higher transmission voltages, the desire to ship transformers with higher
power ratings without increasing the size
and weight, and the increased complexity of transformer designs have justified increased and more detailed consideration of the insulation of power transformers. A major portion of the core
opening of these transformers is occupied
by insulating materials. These materials
are inexpensive in themselves but they indirectly account for much of the copper
and iron required. The required insulation clearances are determined by the
voltage stresses. Since the impulse voltage stresses are the most difficult to determine and in most cases are the controlling
factor, it is well for the transformer designer to know these stresses at all parts of
the winding as accurately as possible.
History
In about 1930, design methods were
developed in the Westinghouse organization by which these impulse stresses could
be computed with fair accuracy. Basically, the method lumped the distributed
parameters into an equivalent circuit.
It had been experimentally discovered that
the voltage distribution was almost the
same whether or not the iron core was included in the measurements. For this
reason, the inductance parameters were
calculated on the basis of an air core.
Various mathematical and empirical improvements were made over the years.
One of these which reached the literature
was the work by Bellaschi and Palermo.'
Although there were errors imposed by
the basic assumption of air core induct-
FEBRUARY 1 957
J. H. STEELE
ances, these were obscured by the approximations of the simplified and often
empirical design methods which were
necessarily employed to make practical
use of the theory. A valuable tool was
developed in the form of a low-voltage
repetitive surge generator. This was
used to measure the voltage distribution
on a winding which had already been
built and it could also be used to test
models of winding arrangements. The
device has advantages in determining the
cause of a transformer failure on impulse
test and it is also useful as a function
generator in the equivalent circuit type of
analogue computer.
In 1953, one of the writers formulated
a theory which was the basis for a practical solution of the problem by means of
modern digital and analogue computers.
This theory forms the basis fpr the work
reported in this paper.
At about this same time Abetti2 reported his work on electromagneticmodels,
which represents a definite achievement
in this field and which is certainly
the most theoretically satisfying method
proposed to date. However, it was considered too expensive and it required an
excessive amount of time delay in the design as compared with the methods already in use. Because of this, and the
fact that impulse voltage distribution is
generally less of a problem in the shell
form of construction than in the core form,
these models were not applied at the Westinghouse Electric Corporation. The
method reported in this paper was developing and it promised equivalent results more quickly, conveniently, and
economically.
AsSUMPTIONS
EQUIVALENT CIRCUIT
To introduce the method by which the
problem is solved in terms of familiar
concepts, a simplified example willbeused.
Considering the schematic winding configuration in Fig. 1(A), a circuit can be
found which will be approximately
equivalent to the transformer.
Consider this as a 3-winding transformer by breaking the series connection.
The usual inductance diagram can then
be drawn on a reference turns basis. To
put these inductances on the proper turns
basis, "perfect" transformers are used in
the equivalent circuit. The terminals of
these perfect transformers are now
equivalent in inductance to the terminals
of the transformer. By reconnecting the
series connection and placing the winding
capacitances in the circuit, the complete
equivalent circuit is as shown in Fig. 1 (B).
This circuit is equivalent to the transformer under impulse conditions except
for the approximation brought about by
Paper 56-742, recommended by the AIEE Transformers Committee and Computing Devices Committee of the Communication and Electronics Division, and approved by the AIEE Committee on
Technical Operations for presentation at the AIEE
Summer and Pacific General Meeting, San Francisco, Calif., June 25-29, 1956. Manuscript submitted March 27, 1956; made available for printing May 10, 1956.
J. H. MCWHIRTER is with the Westinghouse Electric Corporation, Sharon, Pa.; and C. D. FAHRNKops and J. H. STEELE are with the Westinghouse
Electric Corporation, East Pittsburgh, Pa.
General Description
1267
LV
HV
(3)
(I)
HV LV
(2 (4)
SERIES
CONNECTION
LV
LOW
VIOLTAGE
HIGH
VOLTAGE
(b)
LOW
VOLTAGE
(a)
(5)
(a)
L(3)
HV
HV
(I07
LV
Hf2)
.<
<v 9
'm
WINDING CONN.
-_PERFECT TRANS-
< FORMATION
'--EQUIVALENT
INDUCTANCE
CIRCUIT
where
p =differential operator
C=capacitance in farads
r =inverse inductance in henrys1
v =voltage, volts
i =current, amperes
m =number of unknown voltages
n =designation of node where impulse voltage is applied
node i, i.e.
il- C12. .
~~
CinPvn+rPnv//p-r12v2/p- *.--
rPmvm/P- rlnvn/p
Cln
I~x
X
-lm-ln 1
F22 .. . -r2m-2n
Ix
r2mVm/P -r2nVn/P
Solution of Problem
Pvn
- 21
pV2 I
L Cmi-Cm2. Cmm-Cmnj
.
r1 l r-2.
pvl
Cl
IM2. rmm-rmmj
L-rml-m
VI/P
Mim = O=-CmlPds-Cm2PV2-. . . +
CmmPvm - CmmP
n- rml
vil/P -
rmn
V2/P
vn/P
VM/P
This can be abbreviated to
Fig. 2 (left).
Alternate equivalent circuit of
transformer in
Fig. 1(A)
(A)
(B)
(C)
INPUT DATA
CHECKS
CALCULATION OF
LEAKAGE
FORMATION OF
INDUCTANCE
(E)
TRANSFOR14
| INVERSION
(D)
CALCULATION OF
INDUCTANCES
ELEMENTARY
CAFACITANCES
Fig. 3 (right).
Flow chart of
digital calculations
1268
TO
NODAL FORM
(G)
FORMATION OF
CAPACITANCE
MATRIX
(H)
TRANSFON
MATRIX
(F)
OF
INDUCTANCE
MATRIX
(I)
CALCULAION
OF
ANALOG
TO
COMPUTER
NODAL FORM
SETTINGS
FEBRUARY 1 957
/-
- -- -- -- -- -- -- - ---
.COIL
GROLaP
a
GO
Csk
H---------f
HF HY -i
COIL
GROUP
b
G
COIL
_K
G_
q7C Tc
CCk
/
/
/
CkO /
/
COIL /
/
/
G /
a,i
X2
aXi
X3
VO
bi
-ab1x-b2X2-a3b3X3
a3 b3
aO IC
(a) SUMMER-INVERTER
Xi
- qSatabIx1dt-Sta2b2x2dt
X2
X3
_TCa3b3c3dt-a0V0
(b) INTEGRATOR-INVERTER
Fig. 5. Two principal analogue computer elemenb and their
functions
culator. The limited memory and the
necessity for a fixed program placed a
practical restriction on the size of the
problem which could be handled by the
card-programmed calculator. The acquisition of the International Business
Machines 650 with a larger memory,
stored program, and higher speed made it
possible to refine the methods as well as to
shorten the calculation time.
The digital part of the solution may
best be explained by reference to the flow
chart of Fig. 3. In block b, the leakage
inductance coefficients between all combinations of the coil groups are calculated. In block c these are combined to
form the inductance matrix [LIJK] shown
in Appendix I. This matrix is then inverted in d and expanded by the method
shown in Appendix I. Block e is the
transformation of the inverse inductance
coefficients from a coil group basis to a
nodal basis. The mathematics of this
transformation are described in Appendix
II.
Calculation of the capacitance values,
Fig. 4, of the transformer are made in
blockf. These are then combined in block
g of Fig. 3 to form the capacitance matrix
described in Appendix I. A matrix transformation of this elementary capacitance
matrix h puts the capacitances on the
same nodal basis as the inverse inductances.
FEBRUARY 1957
Fig. 6. Electronic
Associates 16-31R
analogue computer
showing pre-patch
panel
Finally, in block i, the inverse inductance and capacitance matrices are converted to a form suitable for the analogue
computer. This involves a time-scale
change of 105 and division of the rows of
the inductance and capacitance matrices
by the respective diagonal term of the
capacitance matrix. Each row of the inverse inductance marix is then divided by
1, 10, or 100, as necessary, to make all of
the terms less than ten. These divisors
are printed and are the b and c coefficients in the analogue computer setup.
In effect, this portion of the digital computer program calculates the analogue
computer settings.
Checks on the program and on the
computer operation may be made by repeating all of the calculations in reverse
order. The matrix inversion and some of
the other calculations may be checked for
accuracy. Some redundancy is included
ANALOGUE COMPUTER
The basic equations are solved by
means of an electronic analogue computer; i.e., the type employing high-gain
d-c amplifiers in conjunction with passive
elements to form integrators and summers. This type of computer is covered
by Korn and Korn4 and, for purposes of
this paper, it is sufficient to indicate the
function of the computer elements in
Fig. 5. The analogue computer used is
an Electronic Assoeiates, Inc., model 1631 R which has been expanded to include
33 amplifiers and modified by the addition of 60 single-pole double-throw
switches, fifteen 5-position selector
1269
(a1,rl,'v1- a.T12'v!
..
aj,r15'v5-a16rj6'v16)bjci =0
00z
z~~~~Ic.
ox:
(fZ
COL.
COL.
COL.
COL.
COL.
COL.
panel is shown in Fig. 6. All of the components which were added to the basic
computer are terminated at the patch
panel so that they may conveniently be
used on any problem other than the one
for which they were especially designed.
Following is an explanation of Fig. 7.
1. Pot bank 1 and switch bank 2 are used
to set the magnitudes of the inductance
matrix. 2. Switch bank 1 is used for setting the signs of the inductance matrix.
3. Pot bank 2 is used to set the capacitance matrix values. 4. The selector
2
3
4
5
6
INDUCTANCE
COEFFICIENTS
OFF
3
4
a-0
-JOI
C,)
1270
(a,Fiti'v1 -aj,r12'V,.
a 1,r,5'v5 -a,6r716'v6) bic1 = 0
0IWO
0-
ing
FEBRUARY 1957
_r~swvo
TIME- MICROSECONOS
TIME- MICROSECONDS
a
49
0I
0.
41
IL
0
se
l.I
U.
0
101q
NODE-9
>
-20'
or
60
30
40
50
TIME- MICROSECONDS
20
10
70
80
90
gn
a
:;
-%
-i
NO1DE-8
IL)
--
--
U. 0
7'
-- - ------ -
.1
if
--- -
.1
--
N-...
0 >
OR
TIME- MICROSECONDS
-20
10
20
iL
60
TIME - MICROSECONDS
30
40
50
70
80
90
TIME- MICROSECONDS
TIME- MICROSECONDS
I)
0
w
a
4
!Li
41
IL.
0
100
0
49
410
0.
49
U*.
0
k-
TIME - MICROSECONDS
This is equal to
ft
-NODE10
FERUARY 1957
20
_I__
TRANSFORME R hlEASUIRFMIkENS-
20
.+
alr,,r5'v,;,p +0 161'i6'VI6/p
__
---
__
60
50
30
40
TIME- MICROSECONDS
__
70
__
80
1271
nections.
2. This information is punched onto cards
and the cards are fed into an International
Business Machines card transceiver which
reproduces them at the location of the digital computer.
3. The data cards together with the program cards are fed into the computer. The
computations are then made, taking about
50 minutes' time.
4. The digital answers are punched onto
cards which are returned to the transformer
plant via transceiver.
5. The digital answers are printed onto a
form which is designed to facilitate the settings of the analogue computer.
6. The potentiometers and switches are
set up from the digital report sheet and the
voltages recorded on a 6-channel recorder.
This step requires about 30 minutes for one
solution. Subsequent solutions for impulsing another line terminal, with the same
unknown voltages and transformer configuration, canl be obtained in an additional
ten minutes.
7. The designer then compares the voltages with his insulation clearances and
makes any necessary changes in his design.
Experimental Results
transformers.
Future Developments
Since this method of determining impulse voltages is mathematical, it holds
promise for future development. At present, it is felt that economically and for
flexibility, the analogue computer is best
suited for solving the integrodifferential
equations. However, later it may be
desirable to solve these equations with
the same digital computer that calculates
the equation coefficients. It then may
become possible to eliminate the human
being from the design loop. That is, the
machine will compare the calculated
stresses with the strengths and make the
appropriate changes. The machine will
automatically go through this process
several times until the optimum design is
obtained. This is harder to do than to
talk about but in view of what has already been done along this line,' it appears
quite feasible.
Although this development has been
SPECIAL FEATURES
A number of special modifications to
the problem are available with a small
amount of patch-panel rewiring. These
are as follows:
carried out for the shell form of construction, the theory is also applicable to
the core form, where the benefits would be
even greater because of the inherently
greater problems involved.
Conclusions
1272
Appendix 1. Derivation of
Currents for Elementary Coil
Groups
The windings in any transformer, for purposes of this analysis, can be broken down
into the elements shown in Fig. 4. Each of
these elements comprises a coil group, a
through or series capacitance, capacitances
from both ends to other windings or to the
core, and capacitances from both ends to
ground. The first steps in writing the equations for the complete transformer are:
fIkbPiOb. LkkPik
The self-inductances and mutual inductances can be expressed in terms of the turns
and an inductance coefficient. That is
Laa
Mab
Laa'Na2
=
tions gives
...
Nji,)
FEBRUARY 1957
LIJ" =-LI'-2M,,'+LJJ'
where L1j3" is the leakage inductance between coils I and J. The equations then
become
Va/Na-kl/Nk=LaakNaPia+Loak,NbPib.**
LajkNjPij
Vb/Nb-Vk/Nk =LbakNaPia +LbbkNbPib * * * +
LbjkNjPi,
~t
It
Caa+Csa+CGa
-Csa
-
Cab+CsaCGa"
Cab
0
0
0
V2
0 0
V3
0-1-1_ V4
[ T1 {VN}
SiN}
= [FNI {VN}
References
WORKS, P. L.
raafrab'.. rakt' Va
Pik
=L0-1
_0 0
0-1
where
Pia
lVC )
Capacitance Currents
O
O
(rkk/Nk2) Vt
Pir,
Va
(rbk/NbNak)Vt
VI
s
{Vc I
iak
ikbI
* *
11.
VC
'1k
Ii
V5
Vk/Nk)
I/NJ -Vk/NN)
NbPib = rba(Va/Na -Vk/Nk) +rbb(vb/Nb Vk/Nk)... + rbj (vI/Nj -Vk/Nk)
(I'ak/NaNk)vk
pib =(rb /N NOva + (rbb/Nb2) Vb + . . +
Fig.
l'b
iaI
. . . + raj(V
'b
Capacitance Currents
Again using the assumption that the summation of ampere-turns is zero, the equations can be expanded to the Kth order as
follows:
l0a
Vb
134-38.
6. TE:NSOR ANALYSIS OF NETWORKS (book), G.
Kron. John Wiley & Sons, Inc., New York, N. Y.,
1939, chapt. IV.
Discussion
FEBRUARY 1957
1273
still not a simple and straightforward calculation, with it the effect of variation of coil
constants and connections can be much
more readily determined. The method can
be classed as a design tool rather than a test
tool, the first design tool available to transformer designers which permits them to design for optimum strength and economy.
It is no longer necessary to assume a factor
of ignorance in making transformers, and
large reductions in size, weight, and costs of
transformers, and improvement in efficiencies should ultimately result.
E. C. Wentz (Westinghouse Electric Corporation, Sharon, Pa.): This paper presents what is undoubtedly the best method
so far devised for determination of impulse
stresses within transformer windings. The
problem of determining impulse voltage distribution can, in general, be approached in
any of three ways, as follows:
1274
purposes.
The desirability of the new method has
long been appreciated but the essential difficulties of the problem have prevented a
solution until computer means, as described
in the paper, were devised to eliminate the
great amount of numerical work required for
a solution. This new method not only permits accurate determination of performance
of the design but permits design of the winding to effect optimum performance under
certain conditions. Of course, it is true
that the optimum design can be obtained by
either full-scale or model building, if enough
models or enough transformers are built and
tested. However, the over-all result is uncertain and will usually require an unreasonably long time and an unreasonably
large expense. Although determination of
the optimum design by the new method is
FEBRUARY 1 957