IEEE Transactions on Power Apparatus and Systems, Vol. PAS-98, No.6 Nov./Dec.

1979 2299

TURBINE-GENERATOR SHAFT TORQUES AND FATIGUE:

PART I - SIMULATION METHODS AND FATIGUE ANALYSIS
M.C. Jackson S.D. Umnans R.D. Dunlop S.H. Horowitz A.C. Parikh
Student Member Membear Member Senior Member Member
Massachusetts Institute of Technology
Electric Power Systems Engineering Laboratory American Electric Power Service Corporation
Cambridge, Massachusetts New York, New York
Abstract such that considerable further study is imperative.
This paper is the first of two papers concerned This set of papers describes the authors' inves-
with the effects of power system disturbances and tigation into the effects of torsional shaft oscilla-
operating practices upon turbine-generator shafts. The tions on the shaft system,with emphasis on the effects
paper presents techniques for the analysis of shaft associated with HSR. The investigation centers around
fatigue damage due to torsional oscillations. The studies of Big Sandy #2 of the American Electric Power
detailed development of a fatigue model is presented. (AEP) System and employs a physical model generator
An example illustrates the use of the techniques for and the model power system in MIT's Electric Power
calculation of the loss of life of a turbine-generator Systems Engineering Laboratory (EPSEL) as well as var-
shaft due to high speed reclosing of a close-in three ious digital and analog representation of the shaft
phase fault. system and its metallurgical properties. The emphasis
of these papers is on the evaluation of overall shaft
INTRODUCTION damage due to various disturbances. These papers at-
tempt to place the question of shaft torsional perfor-
In 1970 and again in 1971 two incidents occurred mance in the context of overall system performance and
on the West Coast that resulted in failure of a tur- to discuss issues which must be considered in a com-
bine-generator shaft. On both occasions the failures plete engineering analysis of the problem.
were attributed to torsional shaft oscillations re-
sulting from subsynchronous resonant (SSR) conditions BASIS FOR APPROACH TAKEN
that can exist in a system using series capacitor line
compensation. The problem of turbogenerator shaft damage due to
torsional oscillation is extremely complex. It can be
As an outgrowth of these incidents, various in- subdivided into three areas:
vestigators have also examined the effects on turbine-
generator shafts of system switching events such as 1. Modeling of the shaft system.
fault occurrence, fault clearing, incorrect synchron-
izing, and reclosing and line switching which general- 2. Predicting the "driving" torques which are
ly occur independent of SSR conditions. In partic- applied to the shaft system under various
ular, high speed reclosing (HSR) has often been men- power system operating conditions.
tioned as a practice deserving of special concern with
respect to possible shaft damage. 3. Estimating the amount of damage which may
result from any particular transient of
Recommendations have been made to the industry the shaft system.
that the practice of HSR be severely curtailed or eli-
minated.`-4 The recommended alternative to HSR is to Each of these is a topic which by itself will consti-
delay reclosing for a length of time (10 seconds or tute a major subject of study for many years to come.
more) sufficient for the decay of the torsional shaft Yet,each must be included in a study of turbogenerator
oscillations as excited by the initial fault. This shaft fatigue.
recommendation is made so as to avoid the possibility
of high transient torques excited by closely spaced This paper is specifically concerned with an in-
transients applied to the shaft system. Line switch- vestigation of shaft damage due to torsional oscilla-
ing produces similar effects in that resultant tran- tions initiated by events on the power system. The
sients excite torsional oscillations in the shaft sys- types of events of interest include faults, high speed
tem. reclosing, line switching and out-of-phase synchron-
izing. These papers attempt to assemble a set of tech-
Although not well documented in the literature, niques which provide answers consistent with the pre-
various criteria based upon changes in real power (AP) sent level of understanding of the various aspects of
and current (AI) as determined by conventional tran- the overall question.
sient stability studies have been proposed as means by
which one can evaluate the effect of system operations For example,this paper employs a simplified model
on shaft life. The effort to develop such simple which represents the shaft system as a set of 5 lumped
criteria is commendable. However the implications to torsional masses (the generator and turbine sections)
present operating practice and system reliability are coupled by torsional springs (the interconnecting
shafts). Although this is an approximate model, it is
adequate in view of the present state of technology
for estimating fatigue damage and cumulative loss of
shaft life. In addition, there is very little quanti-
tative information and understanding of the damping
mechanisms associated with shaft oscillations.
Thus, a heavy reliance must be placed upon engi-
neering judgment to obtain and interprettheresults of
F 78 038-2. A paper recommended and approved by the IEEE a study of this type. Certainly results of one percent
Rotating Machinery Committee of.the IEEE Power Engineering Society accuracy cannot be expected. The limitations of each
for presentation at the IEEE PES Winter Meeting, New York, NY, of the steps involved must be clearly understood in
January 29-February 3, 1978. Manuscript submitted September 1, order to properly arrive at a meaningful conclusion.
1977; made available for printing November 1, 1977.
0018-9510/79/1100-2299$00.75 ( 1979 IEEE
2300

This paper attempts to provide a sufficiently com- Per Unit Torsional

plete description of the simulation and analytical Inertia % of Shaft Spring Constant of
techniques used so that the bases for the conclusions Shaft Constant Power Interconnecting
reached can be clearly seen. Since these techniques Section H(sec) Supplied Shafts (Pu/radian)
are approximate in many ways, other conclusions are High
possible and the paper is intended to encourage dis- Pressure .180 40.0
cussion in this area. The authors hope that such dis- Turbine
cussion in a public forum may lead to cooperative ef- K1 = 34.7
forts which will result in further understanding of Reheat
this problem. Turbine .251 21.0
REPRESENTATION OF GENERATOR AND SHAFT SYSTEM K2 = 42.8
1st Low
Shaft System Model Pressure .922 19.5
Turbine
The simulation technique used in this study in- K3 = 57.0
2nd Low
volves the use of the MIT-EPSEL model power system. Pressure .944 19.5
This system is described in references 5 and 6. The Turbine. K4 = 69.7
model power system consists of three model generators
which can be interconnected via transmission line Generator .837 --
models (Tr-sections) both to each other and to loads and Table 1
to an infinite bus (the laboratory 3 phase supply). Parameters of Shaft System Model (Per Unit)
The physical model generators are accurate models
capable of detailed representation of the electromech- nature of the model.
anical interactions between the generators and a power
system. They are driven by dc motors controlled by In order to complete a representation of the shaft
electronic analog representations of the steam system system appropriate for torsional oscillation studies, a
dynamics. For the purposes of this study, the shaft model for the damping mechanism must be included. There
system is represented in the form of torsional masses are two damping mechanisms which can be immediately
(representing the generator and turbine sections) con- associated with a model such as that of Fig. 1. The
nected by torsional springs (representing the inter- first of these is damping due to twisting of the inter-
connecting shaft sections) as shown in Fig. 1. This connecting shafts. Physically this mechanism is asso-
type of "lumped parameter" representation is similar ciated with the mechanical hysteresis of the shaft
to the models used in most studies of this sort which material as it undergoes cyclic stress-strain varia-
have been reported in the literature. tions.
TLp2 TGEN The second damping mechanism is associated with
the oscillation of the turbine blades in the steady-
state steam flow through the turbine. In its simplest
representation such a damping can be represented as
being proportional to the incremental speed of the tur-
bine sections as measured in the synchronous reference
frame.
The data available in the literaturel3 8 ,9 indicates
High Reheat Low Pressure Generator that damping is a function of load and is dominated by
Pressure Turbine Turbines steam-turbine blade interactions. Reference 9-presents
Turbine data which shows the measured damping time constants of
various torsional modes to range from 8 to 30 seconds
1igw'e 1 Shaft Sy&tem Modet at no load and from 2 to 10 seconds at full load. The
importance of this damping mechanism stems from the fact
The five torsional masses represent the generator, that in SSR it determines the "Q" of torsional systems
two low pressure turbine sections,a reheat turbine and and thus the degree to which torsional oscillations can
a high pressure turbine. The model has been designed be excited. In cases where torsional oscillations are
to represent AEP's Big Sandy Unit #2 which isa907 MVA excited by a system disturbance, the damping determines
unit. The model parameters as given in Table 1 are per- the rate at which resultant torsional oscillations de-
unitized to the machine base. Thus, the torsional cay and thus the degree by which cyclic shaft fatigue
spring constants of the coupling shafts are given in damage can occur.
per unit torque/radian. The parameter values are based
upon the physical dimensions of the shaft system and For purposes of these papers, no attempt was made
its material properties. to vary damping with load or to assign different values
of damping to the individual torsional modes. There
For example, for a cylindrical shaft of uniform appears to exist no analytical theory for doing this at
cross section undergoing elastic strain, the torsional the present time. The damping coefficients were ad-
spring constant, K, is given by7 (see glossary for justed so that each torsional mode decayed with a ten
definition of terms) second time constant.
GTrR4 The complete mathematical representation of the
K= 2Z (1) shaft system of Fig. 1 can be written (in per unit) as
in Eq. 2. The terms used in Eq. 2 are defined in the
For a turbine-generator Eq. (1) cannot be used direct- glossary. Notice that in this representation the damp-
ly since the shaft sections are not of uniform cross ing is represented as being proportional to the speed
section. However, the shaft can be considered as a of the various turbine sections measured with respect
series of uniform shafts and an effective spring con- to the synchronous reference frame. This corresponds
stant for the series combination can be calculated. to modeling the damping due to steam-turbine blade
This procedure is consistent with the lumped parameter interaction.
 2301

d 26 d
electrical side, the coupling back from the mechanical
system does not produce a significant effect on the
2HH
HP HP HP-
d2HP
4IP -~1 ( HP dRH) tHP dt (a
-o dt- response of the shaft system. This is due to the fact
that under such conditions the shaft system responds
2H dRd26 RH
RH =
RH (b in an essentially undriven fashion. In other words,
X d2
t RH '1 (6HP 6RH) i2 (6RH6LP1) -tRH dt (2b)
it is excited by the system event (such as a fault
oH dt6d initiation and clearing or a line switching operation)
dL
iP1 2 (6RH-LP1) K3 (6LP1-6LP2) dt and the ensuing torsional transient is essentially the
natural "ringing" of the torsional mass-spring system.
2 d t6d
L-2
The validity of this coupling approximation, which
L P2

=1P2 K (6LP1 6LP2) i( LP2 sgen ) 4P2 dt (2d) permits the use of the physical model generator, is
essential to the validity of using themodel power sys-
w0 dt 3

d26 tem and physical model generator to produce the elec-

enen2e trical input torque for the shaft system model. This
o
t2 (2e)
(6LP2dgent tgen
assumption has been checked via computer simulations.
Simulations with and without complete coupling show
Generator Model negligible differences in resulting shaft torques.
Other investigators have found similar results.",'12
The coupling of the torsional model to the power
system is accomplished through the generator torque, The use of the physical model generator in combina-
Tgen. This represents the electric air gap torque act- tion with an electronic analog shaft system model pro-
ing on the generator and corresponds to the real power vides a number of convenient features for these types
output of the generator. In the steady-state, this of tests. The circuit breakers open upon current zero-
torque balances the net torque due to the various tur- es in each phase, thus reproducing the interrupting
bine sections. properties of large circuit breakers. Canay13 has re-
ported this interrupting characteristic significantly
For purposes of this study, the shaft system model affects the shaft torques under these types of condi-
is implemented via analog electronics. The electro- tions. In addition, the model system can be quickly
mechanical interaction between the power system and the reconfigured so that a wide range of conditions can be
generator is simulated via the Big Sandy #2 model gen- studied. Fault location, type, and duration can be
erator5 on the MIT-EPSEL model power system. When a easily modified. The logic associated with the line
transient is initiated on the model system, the model breakers (e.g., trip and reclose times, single pole
generator terminal quantities are measured and from versus three phase operation, etc.) can be selected
these the electric torque is calculated and inputed to simply by the setting of various switches.
the shaft system analog model.
Results similar to these reported here should be
The electric torque can be calculated as1"
expected for computer simulations of a suitably de-
tailed nature. There appears to be nothing inherent
to this type of study that requires the detailed gen-
Tgen [2a(ib ic)V-3- *b(1 ia)+ lc(i-ib)] erator and system modeling capability of the model
+

power system. Rather, the authors have found it to

~~~~~~~~~(3)
3

provide a quick and efficient means to perform a large

Here, the fluxes, and currents, i, are instantaneous
i, number of tests of many types of disturbance and sys-
time quantities. The fluxes can be found by integrat- tem conditions.
ing the terminal voltages under the assumption that the
voltage drop across the armature resistance is negli- FATIGUE MODEL
gible. Under the condition of no zero sequence currents
or fluxes, such as is the case when the generator unit Description of Fatigue Process
transformer is A-connected at the armature terminals,
Eq. 3 can be reduced to The combination of the physical model generator
and the electronic analog shaft model described pre-
viously can be used to determine the torques which re-
i Vbdt ib Vadt] (4) sult in the shaft system due to disturbances in the
power system. Although these torques are themselves
of considerable interest, they are not the desired end
Equation 4, expressed in per unit, is implemented elec- result of this study. Rather, it is necessary to quan-
tronically and coupled to the electronic analog shaft titatively determine the amount of damage which the
system model. torques cause to the shaft system.
Note that this is a one way coupling. The measur- The problem thus is one of determining the fatigue
ed electric torque drives the shaft system model. How- damage which is caused by the cyclic stress variations
ever, oscillations of the torsional system do not in which the shaft material undergoes. The mechanism for
turn couple back to the electric torque. In other this fatigue damage is extremely complex. As with most
words, the oscillations of the generator rotor angle metallurgical effects, it is not the type of problem
as predicted by the shaft system torsional model are which is amenable to a rigorous theory which can lead
not translated into oscillations of the model machine to quantitative analysis.
rotor with the resultant small variations in electric
torque. Fatigue damage is the result of crack growth caus-
ed by time varying stresses applied to a material.
Such a one way coupling would be totally unaccept- This crack growth is considered by some researchers to
able for studies of the phenomena of SSR. SSR is a be divided into three regimes15 -- crack generation,
property of the electromechanical coupling between the crack concentration, and growth and propagation of the
electrical system and the mechanical shaft system. terminal crack to failure. The initial generation of
Thus, this coupling must be accurately modeled to pro- a microscopic crack depends upon the microscopic mat-
perly reproduce SSR effects. However, for studies where erial properties (as determined by both the type of
torsional oscillations are excited by transients on the material as well as its manufacturing and operating
2302
history) and the nature of the applied stress. Figure 3 shows a typical stress-strain curve for a
material. Strain (£), which is the percentage elonga-
In turn, the growth and transition of these cracks tion of the material, is plotted as a function of the
to macroscopic cracks continues to depend upon material stress (a), or force per unit area, which caused the
properties, material geometry and size, applied stress, elongation. This curve resembles quite stronglythedc
as well as the initial history of the material in the magnetization curve of a magnetic material. The linear
crack generation stage. Finally, one or more of these Stress (a)
macroscopic cracks combine and grow to such a length
that total failure of the material structure occurs.
It is necessary to express engineering fatigue in
statistical terms. This is commonly done by obtaining
curves which relate the expected number of cycles to
failure at a given cyclic stress level to the stress
level.'5 16 Figure 2 shows a typical stress life (S-N)
curve for material such as steel. The fatigue life in-
creases with decreasing stress amplitude until it is
considered to become infinite at a stress level known
as the fatigue limit. Below this stress level, it is
assumed that no fatigue damage occurs to the material.
Note that at a stress level corresponding to the Elastic Plastic
ultimate strength of the material, failure occurs as Strain Strain
the material is simply loaded up to the ultimate stress Region Region
and thus the failure occurs in one half cycle (Nf=l/2).
Fatigue failure does not require the material to be (Ee ) (Ep)
stressed to its ultimate stress. In fact, most fatigue FiLguLe 3 SteAs-Sttain Cuwve
failure occurs with overall stress much less than this
level. region of the curve is known as the region of elastic
strain. The relationship between cyclic elastic
Stress Amplitude strain, se, and cyclic elastic stress, a', is given by

Amplitude of cyclic stress

£e = a'/E (5)
corresponding to failure A material which is elastically deformed will return
in nf cycles to its original shape and strength after deformation.
Values of strain outside of the linear region are
known as plastic strain. Plastic strain results in
permanent deformation of the material. Cyclic plas-
tic strain, £P, can be related to stress as'7
Fatigue Limit
p 1/n'
f
C~= E4 (=i) (6)
nf Natural logarithm of number (f
of cycles to failure
The CcFfin-Manson law'8 is assumed to relate the num-
Figu'e 2 Typicat StAes4-Li6e Cu%ve ber of cycles to failure, Nf, to the peak cyclic plas-
tic strain
Curves of this type are determined empirically for £p = c(2Nf)c (7)
each material by a large number of tests performed on
samples of the material. Each sample is cyclically Equations 5-7 can be used to express Nf as a func-
stressed at a specific stress level until it fails, tion of a', the peak cyclic stress amplitude
and the number of cylces is recorded. The nature of
the fatigue process is such that failure actually
occurs at a somewhat different number of cycles for N = 2 (U,) (8)
each stress level and thus a curve such as that of
Fig. 2 should really be described in statistical terms
in terms of a mean value and a variance. Equation 8 describes the portion of the curve of Fig.
2 for stresses larger than the fatigue limit. To
As discussed in Reference 14, the main justifica- complete the fatigue model, the stress corresponding
tion for engineering use of this type of fatigue des- to the fatigue limit must also be specified.
cription is that it has been found to work in a number
of cases. It is clear that considerable further work A complication in the application of a fatigue
into this problem is required. For example, evidence life curve such as in Fig. 2 is that it is derived
exists that fatigue properties do not scale with size from data obtained under single amplitude, cyclic
and geometry.' As a result, it is not uncommon to stress variation. In actual engineering application,
conduct fatigue tests on full size structures. a material undergoes complex stress variations which
can be thought of as the superposition of many cyclic
For a given material, an S-N curve is derived stress variations of different amplitudes. To account
from published data in the following manner. The var- for fatigue damage under these conditions, a standard
ious equations are hypothesized in form and the appro- technique is used to calculate the percentage of life
priate parameters are determined via experiment for lost due to the stress at each amplitude and then to
each material. sum them to obtain the total percentage of lost life.
 2303
This is known as Miner's rule and can be expressed Turbogenerator shafts are composed of Ni-Cr-Mo non-
mathematically as19 magnetic high temperature steel. Reference 23 indi-
cates that this steel is in the class of ASTM A 293/2 -6
ini
(the last number 2 through 6 signifies a specific steel
= fraction of life lost (9) within this class). Reference 24 indicated that it
fi would be suitable to assume that the shaft material is
ASTM A293/4 or a close relative thereof. Table 2 lists
where various properties of this material as found in the
literature as well as a specific reference for each
n = # of cycles at ith stress level property. Values of the strain hardening coefficient
(ao) and strain-hardening exponent (n) could not be
Nf = # of cycles to failure at ith stress level explicitly found, but from values given for other low
carbon steels and because of the materials' similarity
When the fraction of life expended equals one, Miner's to 303 stainless steel, the values given in Table 2 are
rule predicts failure of the material. believed to be appropriate for ASTM A293/4.
Figure 4 shows a portion of a typical stress-time
curve which may occur for a given shaft under torsional Property Reference Value
vibration conditions. There appears to be no standard
procedure for evaluating such complex stress waveforms Young's modul us (E) 22 30xl 6psi
in terms of numbers of constant amplitude cyclic stress
reversals as required by the stress-life curve of Fig. shear modulus (G) 22 llxlO6psi
2.
tensile strength (S ) 23 110,200 psi
S t ress (C)
.1% yield strength (Y ) 23 84,100 psi
percent reduction in area (q) 25 61%
strain-hardening coefficient (a ) 26 205,000 psi
strain-hardening exponent (n) 26 .51
fatigue ductility exponent (c) 27 -. 5
Table 2
Properties of ASTM A293/4

Two additional properties can be calculated from

the properties in Table 2. From reference 28,
FiLgue 4 Typicat StAe+s-Time U6toty
The cycle counting technique chosen for this study C.= ln (11q) = .942 (10)
is quite simple. It is believed that this procedure is
in fact conservative and thus will result in pessimis- and from reference 26
tic estimates of shaft damage. This technique can be
explained with respect to Fig. 4. Each identifiable
positive stress peak (labelled A,B,C,D, etc. in Fig. 4) ar = a%(f)n = 199,000 p.s.i. (11)
is counted as one cycle of a stress equal to the value
of that peak. Thus, even though the stress is oscill- Using these parameter values, substitution in Eq. 8
ating around a mean value, ao, the peak stress amplitude yields
is measured with respect to zero (e.g. vA is the peak -(2.00/n')
stress amplitude associated with point A). In addition Nf = 1-(5.03 x 10-6 oa) (12)
stress peaks such as B and C in Fig. 4 are counted as
separate peaks of amplitude aB and ac respectively, in Finally, the value of n' must be determined. Ref-
spite of the fact that these peaks are separated by erence 24 shows a plot of number of cycles to failure
only a small stress reversal. versus stress taken from tests of a material with a
chemical composition similar to ASTM A293/4. This data
In spite of the number and degree of assumptions was obtained for bending tests upon a 9-inch diameter
involved in these descriptions of cyclic fatigue, they shaft. This data can be used to solve for n' under the
appear to represent the best available technoloqy and assumption that ASTM A293/4 has the same exponential
form the basis of published work in this area. relationship. Thus n' = .112 and the exponent in Eq.
12 is -17.86. The fatigue limit (infinite cycles to
Specific Fatigue Model failure) for the material of reference 24 is 38,000 psi
and this value will be assumed for ASTM A293/4.
In this section, the specific fatigue model deve-
loped for this study based on AEP's Big Sandy Unit #2 To complete the fatigue model for the shaft system
is developed. Except for shaft dimensions, no data it is necessary to convert the measured shaft torques
such as material properties has been made available by into material stresses. For a shaft of radius Ro,
the manufacturer of the unit. Thus, the material pro- assuming elastic torsion, the shear stress, as, varies
perties assumed have been obtained from the available linearly with radius, r,
literature but are believed to be representative of
those of the actual shaft material. (a Ro )r
aF Rma0 (13)
2304
where detail to illustrate the various facets of the test
procedure and is characteristic of the type of results
asmax = shear stress at outer radius obtained from each of the tests.
The net torque, T, is equal to the integral of a across The system configuration, shown schematically in
the shaft section Fig. 6, was simulated on the MIT-EPSEL model power sys-
tem. This is a single machine-infinite bus equivalent
T = JJ aS rdA (14) approximating the AEP system as seen from Big Sandy #2.
A three-phase fault is assumed to occur on the line
between the BAKER 345 kV and TRISTATE 345 kV busses at
and thus the maximum shear stress on the shaft can be the BAKER bus as indicated in the figure. The line
expressed in terms of the torque as breakers, labeled 1 and 2, clear the fault in 3 cycles
and then high speed reclose 28.7 cycles later. The
asmax 2-r (15) relaying is such that if the reclosure occurs back
irR0 into the fault, the breakers will retrip in three
The stress in the fatigue model is axial stress, cycles. After ten seconds, the breakers will again
a, which, based upon maximum shear stress theory,29 is close automatically, supervised by a synchro -check
related to shear stress in a tensile test by relay. If the fault still remains, the breakers then
retrip and lockout.
a = 2as (16)
For this example both a successful and an unsuc-
Equation 12 can now be written in terms of shaft torque cessful HSR operation are simulated. The initial in-
and radius, where the damage is related to the maximum finite bus loading is .9 per unit power (machine base)
torque. at unity power factor before adding the 400 MW load.
The generator is operated with constant excitation.
-17.86 The reactances are given in percent on a 100 MVA base.
6 Note that the fault occurs on the weaker of the two
Nf 2 [6.40 x 106 ]
R30 (17) lines emanating from the BAKER bus and thus there re-
mains a relatively strong tie to the infinite bus when
the faulted line is cleared.
It is very useful to normalize the torque on each
shaft section to the torque associated with that shaft
under rated machine operating conditions. Performing Baker
this normalization and using in Eq. 17 the smallest 765 - kV
I
radius of each shaft section, the curves of Fig. 5 re-
sult. These curves give the number of cycles to shaft
failure for each shaft section as a function of the j4.33
normalized torque on that shaft section.
Torque

Big Sandy
-- HP-RH *2
--- RH- LP1
-- LPI -LP2
- LP2-GEN

400 MW

FUgue 6 System Con6iguW on

Figure 7a shows an oscillograph of the various

shaft torques which occur in the case of the successful
HSR. Armature current is also shown to indicate the
fault sequence. The fatigue damage evaluation is made
FigWue 5 Sha4t Fatigue Li6e Cuwve, by analysis of the torque transients by a PDP 11/40
digital computer as explained previously. Although
only the first three seconds of the transients are
Figure 5 comprises the fatigue modet which is used shown in the figure the computer monitors the torque
for this study. It is implemented on a PDP 11/40 dig- transient for 20 seconds after the initiation of a dis-
ital computer. The computer monitors the shaft torque turbance. This corresponds to twice the time constant
simulator and upon the initiation of a transient on the assumed,for the decay of torsional oscillations and is
model system, the resultant damage to the shaft system sufficiently long to insure that any further shaft
i s cal cul ated. torques are less than the fatigue limit.
TYPICAL TEST RESULTS Table 3 indicates that there was essentially no
life loss in this incident. This is due to the fact
In this section an example of a typical test is that the peak shaft torques are less than the fatigue
given. This example consists of one of the tests pre- limit except in the case of the LPl-LP2 shaft where
sented in the companion paper.30 It is presented in the peak torque is 3.05 and the fatigue limit is 2.93.
 2305

Normalized Torque
'THP-RHI

ZRH-LPI

CLPI-LP2A

'CLP2-GEN

30
TIME (cycles) 60

Figue 71a LP1-LP2 Sha4t Toxque-Succeuswuf HSR

Normalized Torque
ZHP-RH LJ\J\*\/

'[RH-LPI A A

'CPI-LP2 AA .I\

LLP2-GEN d , i

0 30 TIME (cyctes) 60

FiguAe 7b LP1-LP2 Sha4t To%que-U ucce24t5u HSR

2306
The same situation with unsuccessful HSR is shown evaluated in terms of the probability of the occurrence
in Fig. 7b. Notice that in this case the torque tran- of the initiating event.
sient before the reclosure is identical to that of Fig.
7a. However, reclosure into the fault causes further SUMMARY AND CONCLUSIONS
increases in the shaft torque to a level which is in-
terpreted by the fatigue model as having caused some This paper presents techniques for the analysis of
fatigue damage. This damage is predicted to be chiefly fatigue damage to a generator shaft due to torsional
in the LPl-LP2 shaft and is about .14% with a maximum oscillations. The central element in this analysis is
torque peak of 4.80. the development of a model for calculating the fatigue
damage in the shaft material. This fatigue model is
maximum both empirical and statistical in nature and this fact
normal i zed must be recognized in the interpretation of theresults.
shaft torque % of life lost
-successful HSR HP-RH 2.62 0 It must be recognized that shaft fatigue cannot be
RH-LPI 2.71 0 directly correlated to simple measures such as the peak
LPl-LP2 3.05 <.001 electrical or shaft torques following a disturbance.
LP2-gen 2.75 0 Fatigue is a cumulative effect which is related to the
overall nature of the torque transient. Thus, the com-
plete torque transient must be analyzed in detail. In
-unsuccessful HP-RH 3.84 .003 the future, it is expected that the analysis of these
RH-LPI 4.04 .004 shaft torque transients will be refined corresponding to
LP1 -LP2 4.80 .141 improvements in cycle counting techniques and in char-
LP2-gen 3.94 <.001 acterization of material properties.
Table 3 It is clear that considerable further work in the
area of torsional fatigue of turbogenerator shafts is
*Test Results needed. Very little fatigue data for turbogenerator
shaft materials is presently available. In addition,
It is important to realize that the nature of the there appears to be a need to obtain such data from
fatigue model is such that the percentage life lost tests on structures similar in size and construction
must be considered to be a "ball park" figure. In to actual shafts. The effects of stress concentrations
fact, depending on the exact timing of the breaker op- and centrifugal stresses must be quantified. Improved
eration, it is possible to see more damage (up to sev- techniques for predicting accumulated damage must be
eral percent or higher) or no damage at all. This is a developed.
function of the exact torque distribution in the shaft
at the time of reclosure and whether or not the torques An example of an HSR breaker operation following a
caused by the unsuccessful reclosure add or subtract fault is presented and discussed. This example serves
from the existing ones. as a background for the test results which are presen-
ted in the companion paper which considers the impli-
Thus, the .14% should be interpreted as an indica- cations of torsional shaft oscillations and potential
tion that this case presents some measurable potential fatigue damage with respect to the operation of a power
for shaft damage. Even if the exact timing sequence of system.
breaker operations could be known, the nature of the
fatigue model is such that it is reasonable to inter- ACKNOWLEDGEMENTS
pret this number soley as an indication of the order of
magnitude of the loss of life percentage which could be The authors wish to acknowledge Professor Gerald
expected. L. Wilson who conceived of this project and guided its
development. We thank David Otten for the development
This example demonstrates that there is some risk of the electronic analog shaft model. The American
involved in HSR into a close-in three phase fault. Electric Power Service Corporation has supported the
This is the basis for a recommendation in reference 1, development of the MIT-EPSEL model power system and
and we assume in reference 4, that HSR be eliminated associated graduate studies and research.
entirely. However, when, for example, the fault is
single phase rather than three phase, there is no loss GLOSSARY OF TERMS
of life predicted from the simulation, as is discussed
in the companion paper. It is possible that a reason- E Young's modulus (nt/m2 )
able alternative might be to disable HSR following
three phase faults under conditions where the possi- G Shear modulus (nt/m2 )
bility of unsuccessful HSR is felt to present a signi-
ficant potential for shaft damage. H Per unitized inertia constant (sec)
This example illustrates the use of the various i Instantaneous armature current (amps)
techniques presented in this paper. The results of a
number of simulations are presented and discussed in K Torsional spring constant (nt-m/rad)
the companion paper. In general, it is found that HSR
into a close-in three phase fault is the condition that Q Shaft length (meters)
results in the greatest potential for shaft fatigue
damage.
Pbase Base power (watts)
Simulations and fatigue analyses as discussed in R0 Shaft radius (meters)
this paper can be conducted for a wide range of shaft
systems, networks, and switching conditions. However, v Instantaneous armature voltage (line-neutral)
it must be recognized that the limitations of the fat-
igue model and associated techniques must be consider- Ur Stress (nt/m2 )
ed when interpreting the results. Furthermore, the con-
cern of causing fatigue damage to a shaft must be CT Shear stress (nt/m2)
 2307
Strain 8. D. N. Walker, et al., "Results of Subsynchronous
Resonance Test At Mohave," IEEE Trans. PAS, Vol. PAS-
Magnetic flux linkages (webers) 94, No. 5, Sept./Oct. 1975, pp. 1878-1889.
Cl Damping coefficient (p.u.-sec/radian) 9. R. Quay and A.C. Schwalb, discussion of Ref. 8, IEEE
6 Angle measured with respect to synchronous
Trans. PAS, Vol. PAS-94, No. 5, Sept./Oct. 1975, pp.
1887-1888.
reference frame
10.C. Concordia, Synchronous
synchronous reference frame Company, 1951, Appendix B. Machines,
General Electric
Angular velocity of
T Torque (per unit or newton-meter) 11. R. Quay, Discussion to Ref. 20, IEEE Trans. PAS,
Vol. PAS-95, No. 1, Jan./Feb. 1976, pp. 23-24.
Subscri pts
HP 12. A. Abolins, et al., Closure to Ref. 20, IEEE Trans.
Hi gh pressure turbine PAS, Vol. PAS-95, No. 1, Jan./Feb. 1976, pp. 24-25.
RH Reheat turbine 13. M. Canay, "Stresses in Turbogenerator Sets due to
LP1 Electrical Disturbances," Brown Boveri Review, Vol. 62,
1st low pressure turbine 1975, p. 435.
LP2 2nd low pressure turbine 14. A.M. Freudenthal, "Fatigue and Fracture Mechanics,"
Engineering Fracture Mechanics, 1973, Vol. 5, pp. 403-
gen Generator 414.
1 Shaft between HP and RH turbines 15. G.C. Dieter, Mechanical Metallurgy, second edition,
McGraw-Hill, 1976, Chapter 12.
2 Shaft between RH and LP1 turbines
16. B. I. Sandor, Fundamentals of Cyclic Stress and
3 Shaft between LP1 and LP2 turbines Strain, 1972, University of Wisconsin Press, Chapter 3.
4 Shaft between LP2 turbine and generator 17. B.I. Sandor, op. cit., page 35.
e Elastic 18. B.I. Sandor, op. cit., pp. 45-46.
p Plastic 19. B.I. Sandor, op cit., page 69.
a ,b,c Phase variables 20. A. Abolins, et al., "Effect of Clearing Short Cir-
cuits and Automatic Reclosing on Torsional Stress and
Indicates cyclic (e.g. a') Life Expenditure of Turbine Generator Shafts," IEEE
Trans. PAS, Vol. PAS-95, No. 1, Jan./Feb., 1976, pp.
Indicates per unit (e.g. T) 14-25.
REFERENCES 21. L. Ahlgren, et al., "Estimated Life Expenditure of
Turbine-Generator Shafts at Network Faults and Risk for
1. R. Quay and R.J. Placek, "Cyclic Fatigue of Turbo- Subsynchronous Resonance in the Swedish 400 kV System,"
generator Shafts," paper prepared for presentation at paper F 77 075-5 presented at the IEEE PES 1977 Winter
1976 IEEE Winter Power Meeting. Power Meeting, New York, New York.
2. P. G. Brown and R. Quay, "Transmission Line Reclos- 22.R.C. Juvinall, Engineering Considerations of Stress,
ing-Turbine Generator Duties and Stability Considera- Strain, and Strength," McGraw-Hill, NY, 1967, p. 558.
tions," paper presented at the 29th Annual Conference
for Protective Relay Engineers, Texas A&M University, 23. R.B. Ross, Metallic Materials Specification Hand-
April 12-14, 1976, College Station, Texas. book, Halstead Press, 1972, p. 499.
3. M.S. Baldwin and R. H. Daugherty, "Fault Clearing 24. F.C. Eaton, "Fatigue Tests of Large Alloy Steel
Stresses Generators," Electrical World, August 1, Shafts," Symposium on Large Fatigue Testing Machines
1977, pp. 48-49. and Their Results, ASTM STP No. 216, 1957, p. 101.
4. D.G. Ramey, et al., "Turbine-Generator Mechanical 25. F.C. Eaton, op. cit., pg. 97.
Requirements Due to Power System Interactions," paper
presented at the American Power Conference, Chicago, 26. J. Datsko, "Material Properties and Manufacturing
April, 1977. Processes," Wiley and Sons, Inc., 1966, pg. 21.
5. G. L. Wilson and S. D. Umans, "A Model Power System- 27. G. Dieter, op. cit., pg. 412-413.
Part I-Design of the Model Generator," paper C75 174-8,
presented at the IEEE PES Winter Power Meeting, New 28. ibid, pg. 339.
York, New York, January 26-31, 1975.
29. Lipson and Juvinall, Handbook of Stress and Stren-
6. S. D. Umans, J. Szajner, and G. L. Wilson, "A Model qth Design and Material Applications, MacMillian Co.,
Power System-Part II-Design of System Elements and Test 1963, pp. 94-97.
Results," paper C75 173-0, presented at the IEEE PES
Winter Power Meeting, New York, January 26-31, 1975. 30. R. Dunlop, et al., "Turbine-Generator Shaft Tor-
ques and Fatigue: Part II - Impact of System Distur-
7. J. P. Den Hartog, Strength of Materials, Dover, New bances and High Speed Reclosure," companion paper.
York, New York, 1961, Chapter II, Section 7.
For combined discussion see page 2314.