DYNAMIC RESPONSE OF POWER TRANSFORMERS

UNDER AXIAL SHORT CIRCUIT FORCES

PART I-WINDING AND CLAMP AS INDIVIDUAL COMPONENTS
Mukund R. Patel
General Electric Company
Pittsfield,- Mass.

ABSTRACT
Three complementary approaches, analytical, numerical and The current carrying conductors of the transformer windings are
experimental, presented here provide a better understanding of the situated in the region of magnetic leakage flux and experience
dynamic behavior of power transformers under axial short circuit mechanical forces. The dramatic increase in KVA ratings of large
forces. The most general and rigorous theories for predicting the power transformers in the last 20 years, accompanied by a parallel
dynamic loads and displacements in the winding and in the clamping increase in the forces generated in the windings under short circuit
structure are developed. Both the winding and the clamp are conditions, has created an industry-wide concern of providing an
discussed individually in Part I to develop a full understanding of adequate support system to avoid mechanical damage. The present
their individual responses before discussing the combined system in day operating statistics document an increasing number of power
Part II. Both parts take a major step forward in the art of the short transformer failures under mechanical forces produced by fault
circuit strength of power transformers. currents. This trend must be reversed. It will obviously demand more
and deeper analyses of the problem.
INTRODUCTION In concentric windings, the axial component of the leakage flux
In a typical core type power transformer (Fig. 1), the windings are produces radial forces which may cause the inner winding to buckle
clamped between the top and the bottom clamping plates (rings) of like a cylindrical shell under external pressure. The radial component
insulating material. These plates are fixed against four clamping of the leakage flux generates axial forces which may produce
beams, two at the top and two at the bottom. The clamping beams excessive stresses in some parts. Since the generated (applied) forces
are held in position by a pair of tie bars on every leg. are electromagnetic in origin, they are termed here as the "electro-
magnetic" forces. They are oscillatory in nature and act on an elastic
system made up of the winding conductors, the insulation and the
clamping structure. This may cause vibrations or instability in the
system. The forces dynamically transmitted to various transformer
parts may be quite different from the applied forces depending on
2 the relationship between the natural frequencies of the structure and
the excitation frequencies.

I
I --C#
CG STATE OF THE ART AND
SCOPE OF THE PRESENT WORK
C+ The scope of the present work will be limited to the axial forces.
0
Early investigators 1, 2 approached the problem only from the static
a
point of view; It was, however, appreciated that periodic short circuit
forces were operating upon a complex elastic system. The pioneering
I L- . INmmlmvr..
-A . .i--- dynamic work was done by Tournier and his associates 3-6 in a
-I---
4

(B) WINDING framework of international cooperation. Their conclusions on this

(A) SIDE VIEW DETILS complicated problem proved to be very useful to the transformer
manufacturers. Their study, however, was limited to the winding part
1. Clamping Beams Tie-Bars only. They assumed rigid clamps which is unrealistic for practical
3. Coil Clamping Plate Cross-Bars transformers. From the dynamic response point of view, the power
5. Core Leg Yoke Laminations transformer can be divided into two major components (see Fig. 1):
C. Conductor Insulation (1) the winding and end insulation between the two clamping plates,
E. I. End-Insulation Electromagnetic and (2) the clamping structure. The present author will investigate
force the response of the total system. It will bring an additional
Figure 1 - Typical Construction complexity in this already complicated study, but the transformer
manufacturers require a complete, not a partial, understanding of the
phenomena in order to improve the transformer integrity to
withstand through faults.

The most logical way to understand the behavior of any complex

system is first to understand its components individually, and then
to use that knowledge in understanding the combined system. Part I
of the present report will, therefore, conceAtrate on the two
components, the windings and the clamps, on their individual bases.
Part II of the report will then investigate the dynamic response of the
Paper T 73 024-7, recommended and approved by the Transformers Com- combined system. The ultimate aim is to predict the dynamic forces
mittee of the IEEE Power Engineering Society for presentation at the IEEE PES
Winter Meeting, New York, N.Y., January 28-February 2, 1973. Manuscript sub- on the conductor, the internal insulation, the end insulation and the
mitted September 18, 1972; made available for printing December 14, 1972. clamping tie bars under short circuit conditions.

1558
 DYNAMIC RESPONSE OF WINDING where X =,uL,Kl = Ki!Ko and K2 = K2/KO. The latter two will be
called the relative end-spring constants.
The transformer winding is made up of a large number of
conductor turns separated by insulating material such that it can be The modified eigenvalues X n can be computed by numerical
represented by an elastic column with distributed mass and spring means, or by plotting the right hand and left hand sides of Eq. 1 and
parameters. Usually the winding ends are heavily insulated and can finding their intersections. For ready use, Xi is plotted as a function
be represented by massless springs. The appropriate mathematical of K1( and K2 in Fig. 3. In most cases X2 will be approximately
model is then an elastic column restrained between two end springs equal to 2 N1. However, as n-i _ Xn - Xn- I = 7r-
as shown in Fig. 2. The axial electromagnetic force is generated
internal to the winding. 0.1 o0 ^loo SW . o

__ i(x, t
S
F+ZTdx
dx. dx
ax
w
i

_ _ -_ ,-KRELATIVE
,,
. STIFFNESS OF-
&O

-
__3_g.0'''1&1 {1&lS

.4J
z
a

RELATIVE STIFFNESS OF BOTTOM END INSULATION.

K-
0
0.1 10 100 soo X0
UNSTRESSED STRESSED
CONDITION CONDITION
Figure 3 - ) Vs. ( 1 9 K2)
Fig. 2 - Elastic Column Between Springs
The natural frequencies con = ,un 11k7m Rearranging this leads
The first analysis in Part I will assume a linear spring characteristic to:
for the insulation, and sufficient prestress on the winding to avoid
separation between the conductor and the insulation under short
= A / = 1 nX o (2)
n mL ng
circuit forces. This allows us to develop a linear analytical solution
with rigor and generality. It is true that the practical transformer
insulation has a nonlinear stiffness characteristic which can be where Ko = k/L, stiffness of the entire winding (excluding end
effectively handled by numerical means. Unfortunately, numerical insulations).
methods provide no insight to the solution. Besides, they are MO = mL, mass of the entire winding.
time-consuning and expensive. Therefore, any analytical solution wo = /Ko/Mo', winding's characteristic frequency.
giving even approximate answers should be welcomed. Thus, the present analysis predicts the dynamic load and
The appro-ach taken here to calculate the dynamic load and displacement at any point in the winding, and also the natural
displacement at any point in the winding is the generalized Fourier frequencies. It distinctly advances all the prior theories in that it
series of the normal modes, i.e., the standing wave approach. Unlike accounts for the asymmetricity of the actual short circuit current
all previous analyses published so far, the present analysis can be and the mechanical damping. Further, it can handle any kind of the
applied to an arbitrary space distribution of the electromagnetic electromagnetic force distribution P (x). When P (x) is expressable in
force. Further, it accounts for the mechanical damping and the a mathematical form, its Fourier coefficients, Pn, can be found from
actual time variation of a fully asymmetric short circuit current. The Eq. A-16. However, in real transformers, P(x) has the same profile as
analysis is detailed in Appendix-A, where the displacement and the the radial flux density, and it is usually difficult to express it in
/force are calculated at any point as a function of time. The analysis mathematical form. In su,ch cases Pn can be calculated by numerical
also yields an equation for calculating the natural frequencies of the means.
winding.
The above theory, like any other linear theory, breaks down in
Natural Frequencies: principle for the insulation with a nonlinear characteristic. However
when the winding is sufficiently prestressed to avoid internal
As seen in Eq. (A-18) of Appendix-A, there are infinitely many separations, use of the insulation's tangent elasticity at a given
natural frequencies of the winding. Their exact values arelj - tn4on prestress level can give reasonably good answers at about 5% cost
where t n is the damping factor for the nth mode of vibration. For compared to the numerical methods.
practical transformer t n is less than 10%. Therefore, -n represents
the natural frequencies with a good accuracy. They are given
by Ankm_7'where ,u n are the solutions of Equation A-1. Response Calculation of a Typical Winding
Table I
If Ko is the spring constant of the entire winding, then k = KoL.
Substituting this in Eq. A- 1 1, multiplying the numerator and Winding Height 75"
denominator of the right side by pAL/Ko and rearranging result in Both end-insulations 5"
the following. Distance between insulation spacer columns 7"
(KI KK2) A Radial build of the winding 3',
tan A = (1) Peripheral width of insulation spacers 1.5"
X 2 K- K-
_

_ Axial space factor, v 0.75

12
1559
Total damping per cubic inch of winding space Notice that as w-P-0, i.e., under static conditions, only 50% load
(conductor plus insulation), Lbf/in/sec. 0.03 appears at the top. The remaining half will be taken by tension in the
Damping ratio for the fundamental mode 0.075 bottom spring. The transient and the steady-state resonance occur at
Insulation's elasticity (constant), psi 100,000 different frequencies, because in a fully asymmetric electromagnetic
Fully asymmetric short circuit with R/X 0.07 force, the transient part has frequency c while the steady-state force
Fundamental mode of excitation is assumed, though not typical oscillates at frequency 2co. At resonance, the steady state force is
of all transformers. about 500% of that under the static condition, the resonance peak is
A typical large size winding whose design data are given in Table I sharp and falls below unity within t 20% from resonance. Therefore,
is chosen for the dynamic response calculations. Per spacer column, the force amplifications due to the steady state resonance can be
the winding stiffness Ko = 100,000 x (3x1.5)/(0.25x75) = 24,000 avoided by removing all the natural frequencies by + 20% from 2 w.
lbf/in. and the mass Mo 7x3x.75x75x.32/386 = 0.98 lb. This The transient resonance peak is only 175% of that under static
gives %124,000/.98' = 156.5 rad/sec. To calculate the natural conditions, mainly because the co-frequency force decays ex-
frequencies con, the both sides of Eq. (1) are plotted in Fig. 4-A ponentially and practically disappears within five cycles. Therefore,
where their intersections times w 0 are the natural frequencies. The there is only a limited build-up of resonance at this frequency.
value of the first five natural frequencies are 328, 712, 1135,1590 However, the transient resonance- peak force is slightly higher than
and 2050 rad/sec. Note that the winding is very close to the the peak force at steady state resonance since the transient
fundamental resonance. electromagnetic force peak itself is about 3.24 times the steady state
electromagnetic force peak. This suggests that it is the transient
(A) E)IN VALUrES (B) EISEN FUNCTIONS tNhORMAL MOS) resonance which should be carefully avoided. Unfortunately, it is of
wide band which makes it difficult to avoid the transient force
amplifications.
i}' DYNAMIC RESPONSE OF CLAMPS

-1
i
o
- t %
4
0

tz
3
Besides windings, the other major component in the transformer
system is the clamping structure. To understand its own dynamic
behavior independent of the winding dynamics, the net axial
electromagnetic force will be assumed to be acting directly on the
-
clamps instead of the winding conductors. As mentioned before, the
-2 2
clamps were completely ignored by the previous investigators. The
first step towards analyzing this huge structure will be to develop a
Fig. 4 Eigenvalues and Eigenfunctions of Typical Winding logical mathematical model.
The solution is an infinite series. H6wever, in practical cases, the Mathematical Model:
axial distribution of the electromagnetic force can be represented
within aboui 5% accuracy in terms of the first ten normal modes. The primary purpose of this section will be to identify and
Obviously, a better accuracy can be achieved by using more terms in evaluate the important stiffnesses and masses playing roles in the
the response calculations. The first few eigenfunctions of the said dynamics of this complicated structure.
winding are plotted in Fig. 4-B to illustrate the nature of the normal
modes. Stiffnesses: Four different stiffnesses can affect the clamp's
dynamics. They are the stiffnesses of the tie bars, the clamping
Using the, analysis developed above, the dynamic force on the top beams, the tank walls and the foundation on which the transformer
end insulation was calculated for different excitation frequencies. is installed for the field operation. Out of these four, the stiffnesses
For all frequencies, a fully asymmetric short circuit was assurred of the tie bars and the clamping beams can be easily calculated. The
with the R/X ratio constant at its typical value of .07 for large power other.two are not so obvious and are discussed below.
transformers. Then, from the calculated force responses, the ratio of
the dynamic peak force to the electromagnetic peak force was
calculated for both the transient and the steady-state conditions. The Axial electromagnetic forces generated in the two windings,
They are plotted in Fig.5 as a function of 4./lwl. Though the primary and secondary, are such that the forces acting on the top and
excitation frequency of a comrmetcial power source does not vary, the bottom clamps are approximately equal and opposite. If the
this plot indicates what would happen when a transformer's natural foundation were perfectly rigid, the clamps would vibrate in the
frequency is in the neighborhood of the excitation frequency. fundamental mode with the bottom clamp fixed. On the other hand,
if the foundation were perfectly flexible, the top and bottom clamps
would vibrate in the second order mode, forming a node (stationary
1* 3.0 point) in the middle of the tie bars. However, since all large
,,,STEADY STATE RESPONSE transformers are installed on a fairly rigid reinforced foundation, the
(RIGHT SCALE) foundation will be assumed perfectly rigid for the present study. In
1 .( 2.0 other words, the bottom structure is assumed fixed relative to the
ground.
TRANSIENT RESPONSE
When the clamping beams are bolted to the tank side walls to
1.0 resist impacts during shipment, the joints may also play some part in
reducing the tie bars' short circuit burden. It is, however, com-
plicated to calculate the effective stiffness of such a joint. As a
I1 t I i I* 1 matter of fact, one has to investigate two types of tank wall stiffness,
0 0.5 1.0 1.5 2.0 2.5 3.0 one for the vertical displacement and the other for rotation. The
-/ _ vertical spring constant of a tank wall can be calculated by assuming
Fig. 5 (Dynamic/Electromagnetic) Peak Force on Top End-Insulation a specified vertical displacement of the beam-end bolted to it, and
Vs. co/wlI (t1 = 75%) then calculating the stress distribution in the wall. The appropriate

1560
stress integral will then give the force required to produce the said calculated by degenerating Milne-Thomson's15 analysis for an
displacement. The force to displacement ratio is the effective spring elliptical cylinder. That gives K.E. = 1/2 pVo2 ir a2L. Equating this to
constant. For withstanding vacuum, the tank walls are usually braced /2MhVo2, one get the plate's hydrodynamic mass Mh =pira2L. In
at about every 50 inches in both the horizontal and the vertical both of the above cases, it is interesting to observe that Mh equals
directions. The actual problem can be simplified by assuming the the mass of the fluid enclosed in a tube of radius a.
tank braces to be rigid and approximating a square into a circle. Such
a stress analysis carried out by the author, using the appropriate Airy
stress potential function 3, showed that the effective spring constant
of a typical tank wall is one order of magnitude larger than that of a
typical tie bar pair. This suggests that the walls can be assumed rigid The largest hydrodynamic mass comes from the top coil clamping
in the vertical direction relative to the tie bars. As far as rigidity in plate. Though its shape is not perfectly round, it can be ap-
rotation is concerned, qualitatively speaking, the tank wall has no proximated by a circular plate on each leg. When such a plate moves
appreciable stiffness for preventing any rotation of the beam. If there in oil, perpendicular to its plane, the kinetic energy stored in oil
were any such stiffness, its neglect would be a conservative again given by Milne-Thomson,15 is K.E. =(4/3) p a3Vc, where a is
assumption. radius of the plate. Equating this to 1/2MhVo2, one obtains the
hydrodynamic mass of the circular plate to be Mh = (8/3) p a3. It
These two facts, that the tank walls act very stiff for a vertical should be realized that this mass appears only on the faulted leg
displacement and flexible for a rotation, can be represented by where the coils cannot load the tie bars without displacing the
hinged joints (Fig. 6) between the beam ends and the tank walls. clamping plate. However, the tie bars on the other legs will be loaded
through the beams' flexural stiffness from the top without affecting
Masses: It is vital to recognize various masses attached to the the coil-clamping plates. It should also be realized that the
clamps. At a first glance, one might be inclined to consider only the hydrodynamic mass is not a dead weight. It appears only when the
dead weight of the clamping beams. But, as will be shown here, it is object moves.
not right. There are several additional masses as discussed below.
Example: To compare the "iron" and the "hydrodynamic"
The beam-yoke interface is normally under some pressure. The masses attached to the clamps, consider two solid beams, each 1"
friction existing there may make it possible for a moving beam to wide x 20" deep x 100" long per phase and a coil clamping plate of
carry some of the yoke laminations provided the yoke is free from 40" radius. Then weight of the beams' iron = 2 x 1 x 20 x 100 x .28
the legs. However, since the yoke and the leg laminations are = 1120 lbf. The hydrodynamic weight of the coil-clamping plate =
alternately lapped, the yoke is assumed fixed in the present study, (8/3) x 403 x .032 = 5480 lbf. The total effective weight of the
allowing the top beams to move relative to the fixed yoke. There clamps is then 6600 lbf. An important conclusion is that the
exists a significant friction between these two sliding parts which hydrodynamic mass is the major mass attached to the transformer
should not be ignored. clamps.
In large transformers, the clamping beams are sometimes made in Unidirectional Motion: In most transformers, the negative motion
box shape. In the short circuit time intervals, it is very likely that the of the clamping beams is prevented by collision between the massive
inside oil will move along with the beams. If so, its mass should be yoke and the cross bars connecting the top beams (Fig. 1). For
added to the clamp's mass. simplicity, this collision will be assumed to be perfectly inelastic in
Hydrodynamic Mass: Under short circuit forces, the clamping parts the sense that it absorbs all the kinetic energy in the clamps. In other
move in oil and set the surrounding oil in motion. The kinetic energy words, once the motion is obstructed, it stops completely without
stored in oil effectively increases the clamp's mass. Such a mass is going through any oscillations. Actually, there will be some local
sometimes known as the "drift mass"or the "hydrodynamic mass" and high frequency oscillations, but they will be quickly damped out by
is equal to p D where p the fluid density and D is the drag on the the yoke-beam interface -friction; so that when the next short circuit
moving object. There may be several parts moving in the same tank load cycle appears on the clamps, they will be practically in rest
of oil and may have some mutual effects. However, it will be position. This assumption will prevent any steady-state resonance
assumed that each part moves in an infinite tank of oil for calculating build-up.
its Hydrodynamic mass. The hydrodynamic mass of an object The Model: In view of the above discussion, the mathematical
moving with velocity Vo can be calculated by first calculating the model for a 3-phase 3-leg transformer clamping structure, after
fluid velocity V, and then calculating the kinetic energy of the fluid lumping small distributed masses at the tie bar locations, is shown
by integrating l2 pV2 over the entire fluid volume. Equating this to below where the fault is assumed on the middle leg.
/2MhVo2, one finds the effective hydrodynamic mass Mh, i.e., Equations of Motion:
+C To study the clamp's dynamic response, consider a hinged beam
fff hpV 2 dxdydz supported by three tie bars as in Fig. 6. For other situations, e.g.,
built-in ends, five tie bars structures or fault on a side leg, the
Mh ½V0
necessary modifications will be obvious.

TANK LENGTH
It is this mass which should be used in all dynamic calculations on
the object immersed in fluid. Three cases applicable to the
transformer clamps are discussed below.

The core yoke or the box-shape clamping beam can be approx-

imated by a cylindrical tube for simplicity. Then, as shown in
Appendix-B, its hydrodynamic mass is pir- a2L where a = radius and
L = length of the beam. Sometimes, rectangular horizontal plates are
attached to the clamping beams. When such a plate of width 2a and
length L moves perpendicular to its plane with velocity Vo in an
infinite tank of fluid, the kinetic energy stored in the fluid is Fig. 6 Mathematical Model for 3-Legged Clamp, Fault on Middle Leg
1561
 No simple method is available in the literature to analyze the Call the above K matrix to be the "total (beam + tie bars)
problem of an elastic beam supported on discrete springs. If the stiffness matrix". Then in short form, Eq. 4 can be written as MY +
spring supports were many in number and close enough in spacing, KY = P where denotes matrix. For a sinusoidal excitation at
-

they can be replaced by an equivalent continuous elastic foundation.

This will simplify the analysis. However, these requirements are not
frequency w, the above matrix equation becomes co 2M +
= P. The solution of this is:
LK]
Y

met in practical transformers. In the beam vibration literature, the

method of flexibility influence coefficients is widely used, but it Y =f-w2M + ISj-Ip(5
includes no discrete spring supports. The influence coefficient The displacements Y become infinite when W2M + K =0
method will be modified and used to analyze the present problem. whose solution gives the natural frequencies. Premultiplying it by
M- 1 and rearranging,
The system of Fig. 6 can be replaced by an equivalent system of
Fig. 7 where the springs are replaced by the corresponding reduction [_w2I + M-1K] = 0 (6)
in the applied forces.
which is an eigenvalue problem. The eigenvalues of matrix M-1K are
the squares of the natural frequencies and the eigenvectors represent
the normal modes of the system. Since the actual model is nonlinear
and the short circuit force is far away from being sinusoidal, the
linear solution will not be pursued any further. The following
equations hold for the actual problem, which will be solved
numerically.

0 * X L
Fig. 7 System Equivalent to Fig. 6
MJ0+IaZK±Y
JiYi :E KJ,Yi ]xi
+ F3asP(t)
+KY3 + 2,3.
ij = 1, 2, 3. (7)
One can write the equations of deflections under static conditions
using the method of flexibility influence coefficients 12 and the Observe that the weights Mj are not considered as downward forces,
principle of linear superposition. These static equations can also be because they are predominated by a hydrodynamic mass which is not
used under dynamic conditions after replacing Pi by Pi(t) -MiYi, i= a dead weight. The extended deVogelaere method10 can be used for
1,2,3. Then, in matrix form, one gets the following. numerically integrating the above nonlinear problem. It is briefly
described in Appendix -C.
r
Y1 a11 a12 a1 3 P1(t) - MI', - k1yl Response of a Typical Clamp:
Consider a typical large size (about 500 MVA) transformer
Y2 a21 a22 a2,s P2 (t) - M2y2 - k2y2 clamping structure whose design data are given below.
Table II - Typical Clamp
Ps(t) M8y- Ml,2,3=40,000 Lbf. Tie-bar area = 4 in2/leg
Y3I a,1X aX2 a,, - - k3y,
Kl,2,3=lxlO6Lbf/in EI of beams = 6 x 1010Lbf.in2
Friction Fj=0 Beam length = 300 in.
Hinged beam ends. Distance between tie-bars=100"
where aij = flexibility influence coefficients of the beam of flexural
stiffness El, defined as the deflection at station i due to unit force at The interest centers around calculating the peak force in the
station j. For a n-legged clamp, the number of influence coefficients faulted leg tie bars under dynamic conditions. The aim is to
will be n2; however, only n(n+l)/2 will have different values, since investigate the following: (1) the frequency response, (2) the effects
aij = aji. These coefficients can be calculated by using Reference 12, of changing various parameters such as the beam stiffness, the clamp
13or ]4. mass and the tie bar area, (3) whether the clamping beam or other
Denoting the inverse of the flexibility matrix (aij) by the stiffness tie bar helps the faulted leg tie bars, (4) whether the connection
matrix (kij) and rearranging, between the beam ends and the tank wall helps.
The calculations of the eigenvalues and the eicienvectors of the
M IK matrix of the given clamps led to the following.

0M M2
Im0 0 Mo 18i Natural Frequency Normal Mode
(rad/sec)
0
[o 0 3L = 102 (.45, 1, 45)
= 158 (1, 0,-1)
C02
W3 = 207 (1,-.68, 1)
(k1 1 k1) k 12 k 13 i Y; P 1 (t1
1 Y2 1 (j~) The natural frequency of the isolated (i.e. with flexible beams)
[ k21 (Lk22+k2) k23 P2(t) faulted leg is wo -x10b/(40,000/386)= 98 rad/sec.

Lk31 k32 (k33Ak3)J[Y3 P[3(tt Numerous computer runs were made for the response com-
putations, and the results are summarized in Fig. 8.

1562
 1.5 -0.75 CONCLUSIONS
.0
.15 IAI .0
(B)
Very general and complete analyses for the winding and the
LE -I.0
E
Li 0.50 clamping structure are developed to predict their dynamic responses
on individual bases. The computations on a typical winding and a
Sll

.0
LAO typical clamp illustrate many important points about their dynamic
*0.5 0.25 ,
behavior. The most important conclusions are: (1) in practical
transformers, the dynamic forces can be completely different from
i I I I I I I 1
the applied (electromagnetically generated) forces, (2) the possi-
bilities exist in both the winding and the clamp for a force
I I I I I I

0 1 2 3 4 0 4 8 12
to e amplification due to resonance effects or for a force reduction due to
W/W/o-* BEAM STIFFNESS - 10 LBf. IN. inertial effects, and (3) oil plays a dominant part in the clamp's
dynamics by offering a large hydrodynamic mass to the system.
ACKNOWLEDGEMENTS
This paper is based on the author's Doctoral Thesis at the
Rensselaer Polytechnic Institute, Troy, N. Y. The author is grateful
to Dr. T. S. Lauber for his able guidance and critical reviews and to
Dr. E. T. B. Gross for his indispensable counsel over the past several
years.
The author extends his deep appreciation to Mr. W. J. McNutt,
Manager of Advance Development Engineering of the Power
Transformer Department of General Electric Company, for
suggesting the project. Special thanks are due to Mr. W. M. Johnson
and Dr. L. Rabins from whom he has benefited by discussions.
20 40 80 0.4 4 20 40
WEIGHT PER LEG-103 LBR TIE BAR AREA-IN2 APPENDIX A-WINDING DYNAMICS

Fault on Middle Leg Referring to Fig. 2, let y(x, t) be the displacement of point x
Fault on Side Leg from its own rest position. Then, in a strained condition, the element
dx changes its length by (by/6x) dx, and gets strained by by/bx.
Fig.8 - Dynamic/Electromagnetic Peak Force According to Hooke's law, this develops a restraining force
on Tie-Bars of Typical Clamps F=k by/6x. Let the mass, the spring constant and the darnping per
unit length of the winding be denoted by m, k, and c, respectively,
and the applied force generated -per unit length in the winding by Po
A careful examination of the results shows the following: (1) the P(x)G(t). Ignoring the winding weight compared to the short circuit
fundamental frequency w 1 of a multi-legged clamp is approximately force, apply Newton's law of motion on a unit length and equate the
equal to that of the isolated faulted leg, wo. (2) in the case of a applied force with the sum of the inertial reaction, elastic restraint
mid-leg fault, the end-joints have negligible effect on the dynamic and damping resistance to get the equation of motion:
load on the tie bars, Fd-tb. However, for a side leg fault it is
definitely better to have rigid joints. For example, Fd-tb with a free
end is about 60% higher than the hinged ends anrd 150%-higher than
the built-in ends. This suggests that the beam ends should always be mat + C ay k a - po p(x) G(t) (A-1 )
connected as rigidly as possible, to the tank walls. When this is done,
a fault on the middle leg is more severe. Otherwise, a side leg fault
will be more severe. This is, of course, intuitively obvious. The tank The requirement of the force continuations at the boundary points
walls provide a significant help as an additional spring in the case of a X=O and L gives the first two of the following four conditions, while
side leg fault, but not so much for a middle leg fault. (3) the flexural the last two come from the assumption of the initial rest condition.
stiffness of the clamping beams does help in reducing the load on
faulted leg tie bars. One can reduce the tie bar loading by increasing
this stiffness up to a certain limit, beyond which the system's natural (A-2)
frequencies may approach the exciting frequencies providing no K1 y(O,t) -
ka t)* 0
extra support. It looks that the stiffness values governed by other
considerations are about right for the optimum benefit from this K2 y(L,t) + kaC(Lst)
2lx,o s ,ay(xso) a 0 (A-3)
side. (4) Attaching as much mass as possible to the clamps is ax

advantageous. This brings the natural frequencies down and causes a aay(x,O)a
considerable force to be absorbed in inertia instead of appearing on y(x,) at (A-4 & 5)
the tie bars. For the given case, if the tie bar material were good for
80 000 psi, the dynamic considerations lead to the design area of 1.2
inZ (Fig. 8-D) instead of 4 sq. in. dictated by the static consider- Solution:
ations. This is a substantial reduction. (5) Large tie bar area will Assume a solution in the form of separated variables, i.e., y(x,t) =
usually cause high force amplification by bringing the natural 0 (x) T (t). Substituting this into equation (A-1) with the right hand
frequencies close to the excitation frequency. However, it reduces side zero for the time being, dividing throughout by 0 (x) T(t) and
the stresses because the increase in the force is smaller than the rearranging result in
corresponding increase in the area. This amounts to that doubling the
area will not reduce the stress to one-half and vice-versa. This fact
signifies the importance of the dynamic study on the clamps. (A-6)
k T kT ' - 'X, say.

1563
 To meet the boundary conditions A-2 and 3, X must be negative L
(X > 0 will lead to a trivial solution). Therefore, letX =-A2. Then Eq. Unt f Y (x,t) 9 Cx) dx (A-1 5)
A-6 becomes eb=-YL 2 0. To find the eigenvalues and eigenfunctions 0 f
(normal modes) of the given problem by Sturn-Liouville theory, the
following "regular eigenvalue problem with unmixed boundary Also, expand P(x) in another series
conditions" corresponding to the original problem is set up.
P (x) pPn n(x) where
01, _P2*
n-1
a
(A-7)
L
K 0(O) - ko'(0) a 0 (A-8) pn a
fi 0
P(x) n (x) dx (A-1 6)
K29 (L) + kb' (L) - 0 (A-9)
Now, to solve the original nonhonogeneous problem, multiply Eq.
The general solution of Eq. A-7 is c =Acosux+ Bsinuxwhere A and B A-i by On(x) and integrate between o to L to get
are arbitrary constants. Boundary conditions (A-8) implies that
B=Kj A/pk and (A-9) implies that
n n (t)n
m U (t) + c Uin (t) + kip n Un - P0 G(t)
l
(A-i? )
A [cos pL(KI+K2) + sin PL( kp2 kW)] -O
For a transformer with a low.resistance to leakage inductance
ratio, a, -the worst (fully asymmetric) short circuit current is
For a non-trivial solution A* 0, therefore the term in the approximately given by I(t)= 10 (eat.-coswt) where w is the angular
rectangular bracket must be zero. After rarranging, this becomes frequency of the voltage wave and t is timne. The electromagnetic
force is proportional to the current squared. Therefore, G(t) can be
tan iL
K11 +
K22 (A-11) written as

(ik Ak) G (t) - 1 + e -2wt -2e

- -'Cos
t at +
1
CosZw
2wt

Taking the Laplace Transformers of the above equation, substituting

Solving the above transcendental equation leads to the eigen- it into that of Eq. A- 17, and simplifying result in the following
values Mn, n = 1, 2, ...The nth order eigenfunction, after substituting
for B, becomes

n m L2s(s2 s w2 )
0 (x) A- [cos pn x + kmn sin nu~x
nn I
p x (A-12)

1 2 (s+,)
where An is still arbitrary. To normalize 'n(x) over the interval 2 2 2 2 2
(o,L), write (s+2ac)Cs *2&nWnS+Wn) (I(s+4) +w )(s +2n,ns+w;)2
L
I o2(
n x)
dx - 1
o + 2 2 2 ~ 2)
This leads to the normalizing constants 2(s +4w ) (s +2CnFnS+Wn

The inverse Laplace Transform of the above can be obtained by the

A L[ + ...) J+d [i-( .)jsin 2u L theory of residues. 16,17 The final reslt is as follows:
-h

2 [1-cos2unL]J (A-1 3) u(t)

n
-
N ~2wn
I0 O°n +t vT [l

-~~~ Wn -2at
Any function, including the solution, satisfying Dirichlet's conditions
can be expanded in a generalized Fourier series of the eigenfunctions
4 n(x), i.e.,
sin(/1-2n 'wn tenl'J1n 1
n n
in( ink ' C +[
nn
2 cc
Y(x,t) -
n-I
U (t) n(x) 4 (A-1 )
-C w t
we nnn
where Un(t) are the Fourier coefficients (can be function of time) / (-CD n
2( A% w2
+
given by

1564
2e n
n
wnt
1B Bzn2 )
r I(B n2n3 B +B nl B)
n4n CosJi~w
consider a circular cylinder of radius a moving with constant velocity
VO in an infinite tank of fluid at rest. Assume that the potential flow
conditions exist.
+ (BnlBn3-Bn2Bn4) sin JTnnnt] p

2 t [B5 cos wt + B sin wtJ

nS n6
l Figg.9 - Cylinder Moving
in Inviscid Fluid
I
~.~/ 'c T 7 [os (2wt-O03

-F; nn
w t
*
.
*
. . . . ... . .
'
e
sin (/jTni t-en42JJ (A-18) The required fluid velocity can be calculated by considering a
nO steady cylindrical object in a uniform stream of velocity -Vo and
then superimposing +VO on the entire system. For the former part,
where various constants are: the appropriate stream function is16
w -VO sinO r ra|
En l tan [/ ; e n2 ' tan-l
n The e and r-components of the required fluid velocity at any
point P (r,9) are:

0
n3 =tan
10wz -4w
- Ve (r,6) = - a r - Vo sine - sine
VorTi

n4 nl- tan
[4wz+2&zwzw'J
~~nfnl
and V r (r,) ra3e o
Ir ± Vcose VO or - cosO
-
The kinetic energy stored in the fluid, per unit length of the
cylinder, is
Bni
nl =' 2&2W2+W2+a2-w2
nfn &
-2aw n3 n
'( 2 2-irV2 ) d 1 22
* 2P (Vae Vr r de dr iwra V
- a 0
B
n2
2w -) T

Equating this to ½/MhVo2, one finds the effective mass of the

n nf n

cylinder Mh=pr.a2, which also happens to be the fluid mass

nB4 n
Bns a 4 2+w2_Xl2_m
n 2< n9 displaced by the cylinder. This confirms with a special case of
B * wFA -IZT_ Milne-Thomson's15 much more complicated analysis for an elliptical
cylinder rotating about its axis and translating at any angle.
Bn6 a2wtn wn2w (A-i19)

Thus, the displacement Y(x,t) is completely expressed by equation APPENDIX-C

(A-14) where Un(t) is given by (A-18) and On (x) by (A-12). The
compressive force at any point in the winding is given by METHOD OF NUMERICAL INTEGRATION
The system of ordinary differential equations of motion derived
for the clamp's dynamics can be rearranged in the following form.
aY(xIt3 (A-20) F n(tY.i) 1,2,..N
Yn + Cn Yn i,n a

associated with initial conditions

Cos V X-4A sin p n x
1
- -k E
n-l
Un(t)An k n n
Yn(0) - fn and n (0) *g

APPENDIX-B- where Fn(t, Yi) are arbitrary functions of their arguments and Cn are
non-negative constants.
HYDRODYNAMIC MASS OF A CYLINDER
MOVING IN OTHERWISE STEADY FLUID The most suitable and economical method for solving the present
problem is the extended deVogelaere method. It requires lesser
The method of calculating the hydrodynamic mass of any object storage and computing time compared to other methods, is self--
moving in a fluid was outlined in Eq. 3. To illustrate the principle, starting and has the error of the order of h5 (h=time increment). It is
1565
fully described by Ful0 and gives the following set of formulas to be
used cyclically to predict the solution at tm+l =(m+l)h when the REFERENCES
solution is known at tm=mh where h is a small time interval. The 1) W. Knaack, The Mechanical Stressing of Transformer Windings
calculation makes use of an intermediate step at tm+1/2=(m+l/2)h. Upon Short Circuit, CIGRE Report No. 135, 1956.
2) E. T. Norris, Mechanical Strength of Power Transformers in
Y
n,m+½
=y +.-hY + --.h
n,m 2 n,m 24
4F -Fn
n - Service, Proc. IEE, Vol. 104A, p. 289, 1957.
( 3) Y. Tournier, G. Ebersohl, A. Ciniero, S. Yakov, A. B. Madin
and J. D. Whitaker, A Study of the Dynamic Behavior of
-~ ~~ ~~~~~W Cn(4Ynp Yn'm-hJ Transformer Windings under Short Circuit Conditions, CIGRE
Report No. 143 and 143-a, 1962.
( 4) Y. Tournier, M. Richard, A. Ciniero, S. Yakov, A. B. Madin and
n,m½h = Af,FY
4+Cnh ,1 4 (F n,r0+Fn,m+h
-a -Cn n,m J. D. Whitaker, A Study of the Dynamic Behavior of
Transformer Windings under Short Circuit Conditions, CIGRE
Report 134 and 134-a, 1964.
n,a1 Ynm+h Yn,m+6h [Fn m+2Fflm.½ ( 5) A. B. Madin and J. D. Whitaker, The Dynamic Behavior of a
Transformer Winding under Axial Short Circuit Forces, Proc.
IEE, Vol. 110, pp. 535-550, March 1963.
- cn (in ,m+2Yn,m+½) ( 6) G. B. Watts, A Mathematical Treatment of the Dynamic
Behavior of a Power Transformer Winding under Axial Short
Circuit Forces, Ibid., p. 551.
n,m+l
6--- h L n,m + 6Ft
6-+C [4
L n,il
oM +4F n,+
41n,D'h½ Fnn,m
( 7) K. Kurita, T. Kuriyama, K. Hiraishi, S. Kusumoto, S. Shida and
Y. Hori, Mechanical Strength of Transformer Windings under
WCn (4Yn ,+ Yn,) Short Circuit Conditions, Trans. IEEE, Vol. PAS-88, pp.
222-230, March 1969.
8) W. J. McNutt, W. M. Johnson, R. A. Nelson and R. E. Ayers,
dY (mh) Power Transformer Short Circuit Strength--Requirements, De-
where Y n,m Yn (mh); Y n1,m sign and Demonstration, Trans. IEEE, Vol. PAS-89, pp.
J i 1955-69, Nov./Dec. 1970.
9) K. Hiraishi, Y. Hori and S. Shida, Mechanical Strength of
and Fn F n(mh'Yi m) Transformer Windings under Short Circuit Conditions, Trans.
IEEE, Vol. PAS-90, pp. 2381-90, Sept./Oct. 1971.
To start the integration at t = 0, one needs to know Yn,-.1 and Yn,-%
which can be found from the equations (10) C. C. Fu, A Method for the Numerical Integration of the
Equations of Motion Arising from a Finite-Element Analysis,
Trans. ASME, Journal of Applied Mechanics, Vol. 37E, pp.
Y fn -2 1h g +-1 h~~~2(F 599-605, Sept. 1970.
n-½h n 8 (FnO Cng )
(11) M. Waters, The Short Circuit Strength of Power Transformers,
MacDonald, London, 1966 (good source book for
bibliography).
n,-h
4-Cnh [(4+C h) gn - h(F + F )]
(12) W. T. Thomson, Vibration Theory and Application,
Prentice-Hall, Inc., 1965.
Since the system under study is initially at rest, (13) J. Prescott, Applied Elasticity, Dover Publications, 1961.
(14) S. P. Timoshenko, Strength of Materials, Pt. I & II, 3rd.
fn n °0 and Yn
f Yn.. Edition, D. Van Nostrand Co., 1955.
(15) L. M. Milne-Thomson, Theoretical Hydrodynamics, 3rd. Edi-
tion, MacMillan Co., N.Y. 1955.
BRIEF FLOW CHART (16) I. S. Sokolnikoff and R. M. Redheffer, Mathematics of Physics
and Modern Engineering, 2nd. Edition, McGraw-Hill Book Co.,
1966.
Input Data. Calculate necessary constants
and initiate main computations at t=O. (17) F. E. Nixon, Handbook of Laplace Transformation,
Prentice-Hall Inc., 1960.
Let tm+ =(m+l )h . Calculate Y] anfid meet (18) M. R. Patel, Dynamic Response of Power Transformer Windings
various constraints. CalculateFn|. and Clamps under Axial Short Circuit Forces, D. Eng. Thesis,
Rensselaer Polytechnic Institute, Troy, N.Y., 1972.
Y ,mand Yn,.l Meet vrio$us constraints.
(19) R. B. Steel, W. M. Johnson, J. J. Narbus, M. R. Patel and R. A.
Nelson, Dynamic Measurements in Power Transformers under
n. ml~-nm l Short Circuit Conditions, CIGRE Report No. 12-01, 1972.

t
Is (time?
>shrt ircitoff For Combined Discussion, see pp. 1576.
-Tyes~~~\
1566